cs.RO / 1 / 2609.31760
What Stops Recursive Self-Improvement in Robotics? Lessons from 123 Rounds of Agentic Skill Discovery
是什么阻碍了机器人领域的递归自我改进？来自123轮自主技能发现的经验教训
Abstract
Can a robot improve itself the way coding agents now improve software? We built an agentic system to find out. It watches a robot fail, works out which capability is missing, writes new skills or finds and installs external models, tests every change in simulation, and repeats, with no human writing robot code. We ran it for 123 improvement rounds on household manipulation tasks. This report describes what we learned. The good news is that the agent can discover capabilities on its own: noticing that its targets were out of view, it asked for an active-viewing model, debugged it, and deployed a working search skill. The bad news is that its improvements did not add up. Changes kept passing their tests, yet the target task, putting condiments on the top shelf of a fridge, never succeeded. We found that the agent was rarely the bottleneck. Three things around it were. First, chained perception modules do not understand relations. Segmenters such as SAM 3 find shelves but not "the top shelf", so the agent filled the gap with ever more geometric rules that never converged, when what it needed was a different kind of model. Second, skill chains lock learning onto the first step. Long tasks mostly fail early, so evidence and fixes pile up there, and later skills are rarely reached, tested, or improved. Third, what the agent learns is decided by the harness. The agent optimized exactly what the evaluator measured, including where it was wrong, and weak tests and misleading memory turned activity into a standstill. We distill these lessons into concrete recommendations for building robot systems that improve themselves, each paired with an experiment that could prove it wrong.
Chinese Translation
机器人能否像编码代理一样自我改进软件？我们构建了一个自主系统来探究这个问题。该系统观察机器人失败，分析缺失的能力，编写新技能或寻找并安装外部模型，在模拟环境中测试每一次变更，并不断重复这一过程，无需人工编写机器人代码。我们在家庭操作任务上进行了123轮改进。本文报告了我们的发现。好消息是，该代理能够自主发现能力：当它注意到目标超出视野时，它请求一个主动视图模型，调试并部署了一个有效的搜索技能。坏消息是，它的改进并没有累积成效。尽管变更通过了测试，但目标任务——将调味品放置在冰箱的顶层——始终未能成功。我们发现，代理本身很少是瓶颈。围绕它的三个因素才是。首先，链式感知模块无法理解关系。像SAM 3这样的分割器能够找到架子，但无法识别“顶层架子”，因此代理用越来越多的几何规则填补这一空白，但这些规则从未收敛，实际上它需要的是一种不同类型的模型。其次，技能链将学习锁定在第一步。较长的任务大多在早期失败，因此证据和修复措施在此堆积，而后续技能很少被达到、测试或改进。第三，代理所学的内容由其框架决定。代理优化了评估者所测量的内容，包括其错误之处，而薄弱的测试和误导性的记忆使得活动陷入停滞。我们将这些经验教训提炼为具体的建议，以构建能够自我改进的机器人系统，并为每个建议配对一个可以证明其错误的实验。
cs.RO / 2 / 2609.31770
Robot Manipulation with GPT-6-Astra: Body Knowledge, Experience Reuse, Emergent Skills, and Sim2Real Transfer
基于GPT-6-Astra的机器人操控：身体知识、经验重用、新兴技能与仿真到现实的转移
Abstract
General-purpose multimodal agents can write robot-control programs, but repeated exploration and model-mediated action selection can make execution slow. We study how external body knowledge, successful experience, and executable skills improve an XLeRobot controlled by GPT-6-Astra in a simulated and a physical elevator-button task. In 30 fixed-start simulation trials, complete robot geometry and camera information reduce mean completion time by 57.4% relative to a baseline with only the common control interface and no prior experience; images with synchronized action and state records reduce it by 68.6% without additional body assets. In nine paired comparisons (18 trials) at starts displaced by 10-100 cm, experience recorded at the original start reduces mean time by 58-63% relative to no experience, demonstrating generalization to the tested new starting positions. During experience experiments, GPT-6-Astra spontaneously generates a short visual-feedback program. Researcher-refactored versions reduce mean local-task time by 29-31% in 27 simulation trials. Finally, 12 real-robot trials using operator-confirmed button contact demonstrate sim2real reuse: at a shared nominal start, simulation XML assets and simulation experience reduce mean time by 53.0% and 49.9%, respectively; real experience also transfers to two new starts. These results suggest a practical way to build general-purpose manipulation experiments around GPT-6-Astra: supply machine-readable body descriptions and synchronized demonstrations, and turn useful agent-generated feedback routines into reusable skills, while the agent adapts actions from current images. We release all task prompts, trial-level experimental data, and acquired skill implementations at https://github.com/hesd10/astra-robot-sim2real.
Chinese Translation
通用多模态智能体能够编写机器人控制程序，但重复探索和模型介导的动作选择可能导致执行速度缓慢。我们研究了外部身体知识、成功经验和可执行技能如何改善由GPT-6-Astra控制的XLeRobot在模拟和实际电梯按钮任务中的表现。在30次固定起始的模拟试验中，完整的机器人几何形状和摄像头信息使得平均完成时间相较于仅有通用控制接口且没有先前经验的基线减少了57.4%；与同步动作和状态记录的图像结合使用时，平均完成时间减少了68.6%，而无需额外的身体资产。在九组配对比较（18次试验）中，当起始位置偏移10-100厘米时，记录于原始起始位置的经验使得平均时间相较于没有经验减少了58-63%，展示了对测试的新起始位置的泛化。在经验实验中，GPT-6-Astra自发生成了一段简短的视觉反馈程序。经过研究人员重构的版本在27次模拟试验中使得平均局部任务时间减少了29-31%。最后，12次使用操作员确认按钮接触的真实机器人试验展示了仿真到现实的重用：在共享的名义起始位置，仿真XML资产和仿真经验分别使平均时间减少了53.0%和49.9%；真实经验也可以转移到两个新的起始位置。这些结果表明了一种围绕GPT-6-Astra构建通用操控实验的实用方法：提供机器可读的身体描述和同步演示，并将有用的智能体生成反馈程序转化为可重用的技能，同时智能体从当前图像中调整动作。我们在https://github.com/hesd10/astra-robot-sim2real发布了所有任务提示、试验级实验数据和获得的技能实现。
cs.RO / 3 / 2609.31771
Reduced Cartesian Kinetostatics for Tendon-Driven Continuum Robots: Residual-Stabilized Full-Shape Propagation
减缩的笛卡尔运动静力学用于肌腱驱动的连续机器人：残余稳定的全形状传播
Abstract
Many planning and control tasks for tendon-driven continuum robots (TDCRs) require the complete Cartesian backbone geometry. We present a reduced Cartesian framework for planar, axially compressible TDCRs that propagates equilibrium configurations along prescribed tendon-force and tendon-displacement trajectories. The backbone is represented by two global position fields. Following exact variation, a Taylor-Galerkin reduction condenses prescribed spatial properties and distributed loads into offline moment vectors, yielding analytic reduced residuals and Jacobians without online spatial quadrature or numerical differentiation. Analytical differentiation and residual correction yield first-order rate systems requiring one fixed-dimensional linear solve per rate evaluation after initial equilibrium alignment on a regular branch. Across four simulated cases covering variable tendon routing, nonuniform geometry, axial compression, and their combined effects, the propagated Cartesian shapes and distributed strains closely match pointwise geometrically variable-strain (GVS) equilibrium solutions. Residual correction suppresses propagation drift across the tested step sizes while adding only about 0.98% to the mean update time of uncorrected Euler. The proposed method requires 0.508 ms per update on average, approximately 11 times faster than pointwise GVS solves. Displacement-driven experiments yield a maximum normalized mean backbone position error of 1.02% and a maximum end-effector position error of 0.28%. These results support efficient and accurate Cartesian full-shape prediction along prescribed actuation paths.
Chinese Translation
许多肌腱驱动的连续机器人（TDCRs）的规划和控制任务需要完整的笛卡尔主干几何形状。我们提出了一种针对平面、轴向可压缩的TDCRs的减缩笛卡尔框架，该框架沿着预定的肌腱力和肌腱位移轨迹传播平衡配置。主干由两个全局位置场表示。经过精确变分后，泰勒-加勒金（Taylor-Galerkin）减缩将预定的空间属性和分布载荷浓缩为离线力矩向量，从而在不进行在线空间积分或数值微分的情况下，产生解析的减缩残余和雅可比矩阵。解析微分和残余校正产生一阶速率系统，在初始平衡对齐后，每次速率评估只需进行一次固定维度的线性求解。在涵盖可变肌腱布置、不均匀几何形状、轴向压缩及其组合效应的四个模拟案例中，传播的笛卡尔形状和分布应变与逐点几何可变应变（GVS）平衡解紧密匹配。残余校正抑制了在测试步长下的传播漂移，同时仅增加了约0.98%的未校正欧拉（Euler）平均更新时间。所提出的方法平均每次更新需0.508毫秒，速度约为逐点GVS求解的11倍。位移驱动实验的最大归一化主干位置误差为1.02%，最大末端执行器位置误差为0.28%。这些结果支持沿着预定驱动路径进行高效且准确的笛卡尔全形状预测。
cs.RO / 4 / 2609.31773
Timed Rule-Based Supervision of an End-to-End Autonomous Parking Policy
基于时间的端到端自主停车策略的规则监督
Abstract
We study whether a manually specified runtime supervisor can correct recurring failures of an existing end-to-end parking policy in a fixed CARLA parking lot. The vision-based Transformer architecture is inherited from Yang et al.; our contribution is a timed, rule-based Parametric Safety Shield (PSS) applied to its control outputs. The PSS uses hand-calibrated speed, position, and duration thresholds to intervene in observed failure modes, including boundary exits, delayed braking, and stalled or oscillatory control. In the reported closed-loop evaluation, 16 held-out target slots and six initial poses are each evaluated in four rounds (384 attempts per configuration). Target success increases from 327/384 (85.16%) for the retrained policy to 375/384 (97.66%) with the PSS; mean position and orientation errors among successful attempts are 0.21m and 0.33 degrees. These results show an improvement within this simulator setup. The repeated attempts share one map, vehicle, and sensor configuration, and the PSS uses simulator world coordinates; thus the results do not establish generalization to other lots or real vehicles, or a formal safety guarantee.
Chinese Translation
我们研究了手动指定的运行时监督是否能够纠正现有端到端停车策略在固定CARLA停车场中出现的重复性故障。基于视觉的Transformer架构继承自Yang等人的研究；我们的贡献是将基于时间的规则参数安全保护盾（Parametric Safety Shield, PSS）应用于其控制输出。PSS使用手动校准的速度、位置和持续时间阈值来干预观察到的故障模式，包括边界退出、延迟刹车以及失控或振荡控制。在报告的闭环评估中，16个保留目标车位和六个初始姿态在四轮中各自进行了评估（每个配置384次尝试）。目标成功率从重新训练的策略的327/384（85.16%）提高到使用PSS的375/384（97.66%）；成功尝试的平均位置和方向误差分别为0.21米和0.33度。这些结果表明在该模拟器设置中有所改善。重复尝试共享一个地图、车辆和传感器配置，PSS使用模拟器世界坐标；因此，结果并未建立对其他停车场或真实车辆的泛化，或提供正式的安全保证。
cs.RO / 5 / 2609.31803
Path Planning with Motion Primitives in Dynamic Environments: SIPP on Lattices
动态环境中基于运动原语的路径规划：在格子上的安全间隔路径规划（SIPP）
Abstract
Autonomous navigation in dynamic environments is a critical challenge, particularly when spaces are shared with other mobile agents whose future trajectories are known. While traditional grid-based planners efficiently find collision-free paths, their reliance on stop-and-turn mechanics over $2^k$-connected grids produces piecewise-linear trajectories that are kinodynamically highly sub-optimal for differentially constrained robots. In this paper, we present an adaptation of Safe Interval Path Planning (SIPP) that operates on state lattices, utilizing precomputed, kinodynamically smooth motion primitives. To efficiently handle dynamic environments, we rasterize the spatiotemporal swept volumes of moving obstacles directly onto the grid, treating grid cells as atomic units of space, whose resolution is typically dictated by inherent localization noise. We perform a comprehensive comparative analysis between our lattice-based approach and $2^k$-connected grid planners across diverse topological environments. Our evaluation considers a broad spectrum of performance metrics, including planning time, path angularity, cumulative heading change (angle-over-length), and bending energy. The results demonstrate that while the expanded state space of lattice-based search increases computational overhead, it yields trajectories with significantly superior kinodynamic properties. Specifically, our method achieves a reachability comparable to highly connected grids while ensuring smooth, continuous, and physically executable paths ready for real-world deployment.
Chinese Translation
在动态环境中进行自主导航是一个关键挑战，尤其是在与其他移动代理共享空间且其未来轨迹已知的情况下。虽然传统的基于网格的规划器能够有效地找到无碰撞路径，但它们对$2^k$-连通网格的停顿和转向机制的依赖，导致生成的路径是分段线性的，这对于具有差分约束的机器人来说在运动动力学上是高度次优的。在本文中，我们提出了一种在状态格子上运行的安全间隔路径规划（SIPP）的改编，利用预计算的、运动动力学平滑的运动原语。为了有效处理动态环境，我们将移动障碍物的时空扫掠体直接栅格化到网格上，将网格单元视为空间的原子单位，其分辨率通常由固有的定位噪声决定。我们对基于格子的方法与$2^k$-连通网格规划器在多种拓扑环境中的表现进行了全面的比较分析。我们的评估考虑了广泛的性能指标，包括规划时间、路径角度、累积航向变化（角度与长度之比）和弯曲能量。结果表明，尽管基于格子的搜索扩展了状态空间并增加了计算开销，但它产生的轨迹在运动动力学特性上显著优越。具体而言，我们的方法实现了与高度连通网格相当的可达性，同时确保了平滑、连续且物理可执行的路径，准备好用于实际部署。
cs.RO / 6 / 2609.31840
Humanoid Badminton: Learning Dynamic Racket Skills from Limited Human Motion Data
类人羽毛球：从有限的人类运动数据中学习动态球拍技能
Abstract
High-speed racket sports provide a demanding testbed for humanoid robots, requiring time-critical decisions, precise striking, and dynamic whole-body coordination. In badminton, fast-changing shuttle trajectories require timely contact decisions, while successful returns demand precise racket pose and velocity within a brief contact window and across a broad three-dimensional striking workspace. Human motion data provide valuable priors for such athletic skills, but usable badminton references are limited and imperfect. Direct tracking provides insufficient executable variation for diverse shuttle conditions, while purely task-driven optimization may produce unnatural motion. To address these challenges, we present a three-stage hierarchical reinforcement learning framework for dynamic humanoid badminton. First, task-randomized motion augmentation expands sparse annotated hitting events into executable target-conditioned stroke variations, forming a continuous latent skill space. Second, a high-level planner outputs continuous latent skill codes to compose these skills online according to the observed shuttle state. Third, a context-conditioned adversarial regularizer encourages more natural planner-level skill usage while preserving return performance. When deployed on a real humanoid robot, our system achieves sustained multi-skill rallies with human players, including forehand, backhand, and highly dynamic jump returns. This is the first real-world humanoid racket-sport system to demonstrate multi-skill human--robot rallies including highly dynamic jump returns.
Chinese Translation
高速球拍运动为类人机器人提供了一个严苛的测试平台，要求进行时间关键的决策、精确的击打和动态的全身协调。在羽毛球运动中，快速变化的羽毛球轨迹要求及时的接触决策，而成功的回球则需要在短暂的接触窗口内以及广泛的三维击打工作空间中保持精确的球拍姿态和速度。人类运动数据为这些运动技能提供了宝贵的先验知识，但可用的羽毛球参考资料有限且不完美。直接跟踪提供的可执行变体不足以应对多样的羽毛球条件，而纯粹以任务为驱动的优化可能会产生不自然的运动。为了解决这些挑战，我们提出了一种三阶段的层次强化学习框架，用于动态类人羽毛球。首先，任务随机化的运动增强将稀疏标注的击球事件扩展为可执行的目标条件击球变体，形成一个连续的潜在技能空间。其次，高层规划器根据观察到的羽毛球状态输出连续的潜在技能代码，以在线组合这些技能。第三，基于上下文的对抗正则化器鼓励更自然的规划者级技能使用，同时保持回球表现。当在真实的类人机器人上部署时，我们的系统能够与人类玩家进行持续的多技能对打，包括正手、反手和高度动态的跳跃回球。这是第一个在真实世界中展示多技能人机对打的类人球拍运动系统，包括高度动态的跳跃回球。
cs.RO / 7 / 2609.31855
PHIRL: Aligning Learned Rewards with Task Progress for Inverse Reinforcement Learning
PHIRL：将学习奖励与任务进展对齐的逆强化学习
Abstract
Human demonstrations provide dense policy-level information but sometimes lack local precision. Human feedback presents accurate local critiques, but offers sparse evaluations rather than direct policy guidance. We propose Progress-Heuristicized Inverse Reinforcement Learning (PHIRL), a data-efficient framework that learns robust reward functions by jointly leveraging demonstrations and feedback. Specifically, we use progress, a feedback modality that describes cumulative task completion. PHIRL iteratively infers a reward function from demonstrations via inverse reinforcement learning, calculates the learned rewards over the progress-annotated demonstrations, and aligns the rewards with progress annotations over four dimensions. We evaluate PHIRL on real and simulated robot tasks, with additional exploration using a fine-tuned vision-language model to provide progress feedback. Results demonstrate that PHIRL significantly outperforms the baselines, achieving substantially higher environmental return rewards and task success with only twenty percent of demonstrations annotated. Analysis of reward-hacking scenarios demonstrates that PHIRL learned reward functions are reliable against exploitation.
Chinese Translation
人类示范提供了密集的策略级信息，但有时缺乏局部精确性。人类反馈提供了准确的局部评估，但提供的是稀疏的评价，而非直接的策略指导。我们提出了进展启发式逆强化学习（Progress-Heuristicized Inverse Reinforcement Learning，PHIRL），这是一个数据高效的框架，通过联合利用示范和反馈来学习稳健的奖励函数。具体而言，我们使用进展这一反馈模式，描述累积的任务完成情况。PHIRL通过逆强化学习从示范中迭代推断奖励函数，计算在进展标注的示范上的学习奖励，并在四个维度上将奖励与进展标注对齐。我们在真实和模拟的机器人任务上评估了PHIRL，并通过微调的视觉-语言模型提供进展反馈进行额外探索。结果表明，PHIRL显著优于基线，获得了更高的环境回报奖励和任务成功率，仅用二十%的标注示范。对奖励操控场景的分析表明，PHIRL学习的奖励函数在抵御利用方面是可靠的。
cs.RO / 8 / 2609.31880
Efficient Bezier Velocity Optimization for Free-Floating Space Manipulators
自由浮动空间操纵器的高效贝塞尔速度优化
Abstract
We present FAVOR (Free-floating Arm Velocity Optimization with Recursive Sensitivities), a planner for collision-free reaching, tracking, and prescribed-time pre-grasp interception on an unactuated spacecraft. It optimizes Bezier joint-velocity curves with linear velocity, acceleration, and continuity constraints. Decision dimension is independent of rollout resolution. Analytical recursive sensitivities provide task and clearance gradients through the coupled base-arm motion. Parallel evaluation, caching, and incremental collision discovery reduce computation. With a seven-DoF arm and five simulated spacecraft models, FAVOR achieves 99.8% point-to-point success with 1.779 s mean computation, versus 62.9% and 50.270 s for an IK-initialized position-spline baseline. Means include failures and timeouts. FAVOR completes 30 of 36 tracking cases, versus 15 for single-step QP, and all 36 interception instances, versus 22 for the spline baseline. A controlled ablation shows that finite differences increase mean planning time 4.7-fold.
Chinese Translation
我们提出了 FAVOR（自由浮动臂速度优化与递归灵敏度），这是一个用于在无驱动航天器上实现无碰撞的到达、跟踪和预定时间的抓取拦截的规划器。它在满足线性速度、加速度和连续性约束的情况下优化贝塞尔关节速度曲线。决策维度与展开分辨率无关。分析性的递归灵敏度通过耦合的基座-臂运动提供任务和间隙梯度。并行评估、缓存和增量碰撞发现减少了计算量。在一个七自由度的臂和五个模拟航天器模型的测试中，FAVOR 实现了 99.8% 的点对点成功率，平均计算时间为 1.779 秒，而 IK 初始化的位置样条基线的成功率为 62.9%，平均计算时间为 50.270 秒。平均值包括失败和超时。FAVOR 完成了 36 个跟踪案例中的 30 个，而单步 QP 仅完成了 15 个，所有 36 个拦截实例均成功，而样条基线仅完成了 22 个。控制消融实验表明，有限差分使平均规划时间增加了 4.7 倍。
cs.RO / 9 / 2609.31885
Differentiable Dynamics for Autonomous Micro-Mobility Navigation
用于自主微移动导航的可微分动力学
Abstract
Autonomous micro-mobility vehicles (MMVs) such as wheelchairs, scooters, and bicycles have the potential to improve mobility access and support safe low-speed transportation in pedestrian-shared spaces. Achieving MMV autonomy will require realistic, predictable MMV motion. However, many existing autonomous vehicle stacks rely on simplified kinematic models that fail to capture key MMV characteristics such as tire slip, friction, and wheel layouts, limiting realism and gradient-based optimization. In this paper, we explore differentiable formulations of dynamics models for autonomous micro-mobility systems. We first construct DiffKBM, a differentiable version of the kinematic bicycle model (KBM). Then, we introduce DiffGM3, a differentiable formulation of the General Micro-Mobility Model (GM3), a unified tire-based dynamics formulation for micro-mobility vehicles that supports a wide range of MMV configurations. DiffKBM and DiffGM3 enable end-to-end differentiable optimization through MMV dynamics, making them suitable for integration into differentiable autonomy stacks. We evaluate these dynamics models in both open-loop and closed-loop settings: (1) open-loop trajectory matching, where DiffKBM and DiffGM3 are integrated as a dynamics layer within DiffStack and optimized to reproduce real-world MMV trajectories, and (2) closed-loop autonomous navigation, where DiffKBM and DiffGM3 are paired with a differentiable MPC controller in CrowdNav pedestrian scenarios. In the open-loop setting, DiffGM3 outperforms DiffKBM in reproducing trajectories with improvements in ADE and NLL across bicycle, scooter, and motorcycle modes, and reductions in planning loss for bicycle and motorcycle trajectories. We also find that, in closed-loop settings, DiffGM3 improves on DiffKBM's CrowdNav performance by producing 55\% fewer collisions and a 75\% lower discomfort frequency for the bicycle mode.
Chinese Translation
自主微移动车辆（MMVs），如轮椅、电动滑板车和自行车，有潜力改善出行便利性，并支持在行人共享空间中的安全低速运输。实现MMV的自主性需要现实且可预测的MMV运动。然而，许多现有的自主车辆系统依赖于简化的运动学模型，这些模型未能捕捉到MMV的关键特性，如轮胎滑移、摩擦和轮布局，从而限制了现实性和基于梯度的优化。本文探讨了自主微移动系统的可微分动力学模型的形式化。我们首先构建了DiffKBM，这是运动学自行车模型（KBM）的可微分版本。然后，我们引入了DiffGM3，这是通用微移动模型（GM3）的可微分形式，该模型为微移动车辆提供了一种统一的基于轮胎的动力学形式，支持广泛的MMV配置。DiffKBM和DiffGM3通过MMV动力学实现了端到端的可微分优化，使其适合集成到可微分自主系统中。我们在开放环路和闭环设置中评估这些动力学模型：（1）开放环路轨迹匹配，其中DiffKBM和DiffGM3作为DiffStack中的动力学层集成，并优化以重现真实世界的MMV轨迹；（2）闭环自主导航，其中DiffKBM和DiffGM3与可微分模型预测控制器（MPC）配对，在CrowdNav行人场景中进行测试。在开放环路设置中，DiffGM3在重现轨迹方面优于DiffKBM，在自行车、电动滑板车和摩托车模式下的平均距离误差（ADE）和负对数似然（NLL）均有所改善，并且在自行车和摩托车轨迹的规划损失上有所减少。我们还发现，在闭环设置中，DiffGM3在CrowdNav性能上优于DiffKBM，摩托车模式下的碰撞数量减少了55\%，不适频率降低了75\%。
cs.RO / 10 / 2609.31901
Game-Theoretic Control with Constrained Potential Surgery
带约束的潜在手术的博弈论控制
Abstract
Constrained general-sum dynamic games are a popular formulation for highly interactive multi-agent planning problems. In recent years, Generalized Nash Equilibrium (GNE) solvers have achieved real-time performance for small dynamic games. However, solution speed still remains a bottleneck, and controlling even a small number of agents (e.g., more than four) in a dynamic task remains elusive. In addition, Newton solvers that focus on the first-order conditions are vulnerable to non-Nash saddle points, limiting the usefulness of the provided solution. In this work, we propose a fast and versatile interior point solver for constrained dynamic games, along with a computationally efficient second-order correction that increases the probability of converging to a local GNE solution in constrained dynamic games. The performance of the proposed method is evaluated on numerical benchmarks and a physical experiment involving scaled race cars.
Chinese Translation
带约束的广义和动态博弈是高度互动的多智能体规划问题的一种流行表述。近年来，广义纳什均衡（Generalized Nash Equilibrium, GNE）求解器在小规模动态博弈中实现了实时性能。然而，求解速度仍然是一个瓶颈，控制即使是少量智能体（例如，超过四个）在动态任务中仍然难以实现。此外，专注于一阶条件的牛顿求解器易受到非纳什鞍点的影响，这限制了所提供解决方案的实用性。在本研究中，我们提出了一种快速且多功能的内点求解器，用于带约束的动态博弈，并结合一种计算效率高的二阶修正，增加了在带约束动态博弈中收敛到局部GNE解的概率。我们在数值基准测试和涉及缩放赛车的物理实验中评估了所提方法的性能。
cs.RO / 11 / 2609.31904
GT-VLA: Target-Conditioned Trace Guidance for Generalizable Robotic Manipulation
GT-VLA：针对目标条件的轨迹引导以实现可泛化的机器人操作
Abstract
Vision-Language-Action (VLA) models have shown strong performance on robotic manipulation, but they often struggle to generalize to unseen tasks, configurations, and long-horizon settings. A key challenge is that VLAs overfit to training scenes and fail to follow novel language instructions. Off-the-shelf vision-language models (VLMs) often provide stronger generalization, but cannot directly control robot actions. To combine the common sense of VLMs with VLA control, we propose Guided Trace VLA (GT-VLA), a steerable framework that accepts guidance from an external generalist VLM through trace-conditioned action generation. GT-VLA uses a generalist model to identify semantic guidance for the current skill, converts this guidance into a 2D visual trace, and conditions its action policy on the resulting trace-rendered observation. This design separates semantic target acquisition, trace generation, and low-level action execution, allowing high-level guidance to propagate to robot actions. GT-VLA uses a Mixture-of-Experts architecture with skill-specific trace and action modules for robust execution. We evaluate GT-VLA on LIBERO and a physical robot platform, showing improved generalization over recent VLA baselines in both settings. The code and additional supplemental materials are available on our project website at https://ivaniz.github.io/gt-vla/.
Chinese Translation
视觉-语言-动作（VLA）模型在机器人操作方面表现出色，但它们通常难以泛化到未见过的任务、配置和长时间跨度的设置。一个关键挑战是VLA模型对训练场景的过拟合，导致无法遵循新颖的语言指令。现成的视觉-语言模型（VLM）通常提供更强的泛化能力，但无法直接控制机器人动作。为了将VLM的常识与VLA控制相结合，我们提出了引导轨迹VLA（GT-VLA），这是一个可引导的框架，通过轨迹条件的动作生成接受来自外部通用VLM的指导。GT-VLA使用通用模型识别当前技能的语义指导，将该指导转换为二维视觉轨迹，并根据生成的轨迹渲染观察条件其动作策略。该设计将语义目标获取、轨迹生成和低级动作执行分开，允许高层指导传播到机器人动作。GT-VLA采用混合专家架构，具有特定技能的轨迹和动作模块，以实现稳健的执行。我们在LIBERO和物理机器人平台上评估GT-VLA，显示出在这两种设置下相较于近期VLA基线的改进泛化能力。代码和其他补充材料可在我们的项目网站 https://ivaniz.github.io/gt-vla/ 上获取。
cs.RO / 12 / 2609.31905
End-to-end QP-based policies: A unified perspective on robust control and robot learning
基于端到端QP的策略：对鲁棒控制与机器人学习的统一视角
Abstract
We present an end-to-end QP-based policy framework that enables systematic policy construction with minimal domain-specific design, while preserving the transparency and interpretability of model-based control. The proposed policy representation supports both domain-randomized model-based auto-tuning, where policy parameters are optimized over distributions of disturbances and model variations, and black-box policy construction, where the problem is formulated in terms of a (possibly) unknown model without requiring explicit notions of states, inputs, or the underlying system dynamics. We establish connections to existing policy representations and control paradigms, including robust control, multilayer perceptrons, and robot learning, and interpret our end-to-end QP policies as a common abstraction of these approaches. We validate the resulting framework in both simulation and hardware, demonstrating a broad range of applications spanning robustness, automatic policy tuning, and control under unknown system dynamics.
Chinese Translation
我们提出了一种基于端到端QP的策略框架，该框架能够在最小化领域特定设计的同时，系统地构建策略，并保持基于模型控制的透明性和可解释性。所提出的策略表示支持领域随机化的基于模型的自动调优，其中策略参数在干扰和模型变化的分布上进行优化，以及黑箱策略构建，其中问题以（可能）未知模型的形式进行表述，而不需要明确的状态、输入或基础系统动态的概念。我们建立了与现有策略表示和控制范式的联系，包括鲁棒控制、多层感知器和机器人学习，并将我们的端到端QP策略解释为这些方法的共同抽象。我们在仿真和硬件中验证了所得到的框架，展示了涵盖鲁棒性、自动策略调优和在未知系统动态下控制的广泛应用。
cs.RO / 13 / 2609.31924
HapticWorld: an Interactive World Simulator with Real-time Torque Feedback
HapticWorld：一个具有实时扭矩反馈的互动世界模拟器
Abstract
Contact-rich manipulation depends on force sensing that is hard to infer from visual signals alone, both for collecting demonstrations and for training policies. Force-annotated data, however, remains hard to obtain at scale: real-robot collection ties every demonstration to physical hardware, physics simulators report contact forces that deviate systematically from real measurements, and learned world simulators, though scalable and realistic, are vision-only, so operators feel nothing during data collection and the data carries no force/torque (F/T) labels. We present HapticWorld, an interactive world simulator that predicts joint torque together with observations and renders it back to the operator in real time, closing the haptic loop between a human and a learned world model. Across three contact-rich tasks, torque feedback raises data collection throughput by 1.6 times on average. Policies trained on HapticWorld-generated demonstrations succeed in 54/60 real-world trials, approaching the 56/60 upper bound of real-world data, and far exceeding the 19/60 success rate of the vision-only baseline. Moreover, the success rates measured inside HapticWorld closely match real-world evaluation, demonstrating that HapticWorld can serve as a stand-alone F/T-conditioned policy evaluation platform.
Chinese Translation
依赖于接触丰富的操作需要力传感，这种传感仅通过视觉信号难以推断，无论是在收集演示还是训练策略方面。然而，带有力注释的数据在规模上仍然难以获得：真实机器人收集将每个演示与物理硬件绑定，物理模拟器报告的接触力与真实测量系统性偏离，而学习的世界模拟器虽然可扩展且逼真，但仅限于视觉，因此在数据收集过程中操作员无法感知任何触觉，且数据不携带力/扭矩（F/T）标签。我们提出了HapticWorld，一个互动世界模拟器，它预测关节扭矩并与观察结果一起实时呈现给操作员，从而闭合人类与学习的世界模型之间的触觉循环。在三个接触丰富的任务中，扭矩反馈使数据收集的吞吐量平均提高了1.6倍。在HapticWorld生成的演示上训练的策略在60次真实世界试验中成功54次，接近真实世界数据的56/60上限，远超仅依赖视觉的基线的19/60成功率。此外，在HapticWorld内部测量的成功率与真实世界评估紧密匹配，证明HapticWorld可以作为一个独立的F/T条件策略评估平台。
cs.RO / 14 / 2609.31929
Inertia-Corrected Newton Method For Generalized Nash Equilibria in Dynamic Games with Optimality Verification
具有最优性验证的动态博弈中广义纳什均衡的惯性校正牛顿法
Abstract
Newton methods efficiently find Generalized Nash Equilibria (GNE) in dynamic games by solving for the KKT necessary conditions. These methods are fast and can support multi-agent Model Predictive Control (MPC) for highly dynamic robots. However, a small KKT residual alone does not certify that the returned solution satisfies the second-order sufficient conditions for a local GNE. In this paper, we propose an efficient numerical method to verify the second-order sufficient conditions (SOSC) for a local GNE. We connect the inertia of the agent KKT matrix with the positive definiteness of the reduced Hessian of the cost function, projected onto the null space of the constraints. Furthermore, we introduce an inertia-corrected update step that improves convergence to local GNEs by destabilizing strict saddle points with weak cross-agent coupling. Our main contribution is a fast Newton solver for Constrained Dynamic Games that provides efficient optimality checking. Through numerical benchmarks, we demonstrate the proposed solver's runtime and convergence performance in practical multi-agent planning problems. We also validate the solver's real-time capabilities in physical experiments using a platform of miniature autonomous race cars.
Chinese Translation
牛顿方法通过求解KKT必要条件有效地在动态博弈中寻找广义纳什均衡（GNE）。这些方法速度快，能够支持高动态机器人的多智能体模型预测控制（MPC）。然而，仅仅较小的KKT残差并不能证明返回的解满足局部GNE的二阶充分条件。在本文中，我们提出了一种高效的数值方法来验证局部GNE的二阶充分条件（SOSC）。我们将智能体KKT矩阵的惯性与约束条件的零空间上成本函数的约化Hessian的正定性联系起来。此外，我们引入了一种惯性校正的更新步骤，通过破坏弱跨智能体耦合的严格鞍点来改善收敛到局部GNE。我们的主要贡献是为约束动态博弈提供一个快速的牛顿求解器，能够高效地进行最优性检查。通过数值基准测试，我们展示了所提求解器在实际多智能体规划问题中的运行时间和收敛性能。我们还在使用微型自主赛车的平台上进行的物理实验中验证了求解器的实时能力。
cs.RO / 15 / 2609.31941
Low Cost Eye Tracking for Vision Screening
低成本眼动追踪用于视力筛查
Abstract
This engineering study assesses gaze-assisted data acquisition during visual screening with a Gaze Quest implementation via webcam and GC308 near infrared camera with an Orlosky eye tracking pipeline. Ten subjects performed under both acquisition conditions. Accuracy of Tumbling E orientation, reported application logMAR value,and response latency were determined from keyboard responses and hence are behavioral outputs as opposed toeye tracker screening results. Frame-to-frame change in theangle between successive normalized gaze vectors (28.2degrees) served as the GC308 eye tracker derived diagnostic with an effective sampling rate of roughly 8 FPS. No screen target calibration or gaze to screen translation was done on the GC308 eye tracker output data for the current analysis and hence, these data represent acquisition stability and not gaze accuracy in relation to the target. The webcam pipeline relied on 25-point screen calibration. Because the display resolution and pixel pitch could not be retained in the records of the study, the reported application logMAR values cannot be viewed as measures of visual acuity in any way.
Chinese Translation
本工程研究评估了在视觉筛查过程中通过网络摄像头和GC308近红外相机实施的Gaze Quest眼动辅助数据采集。十名受试者在两种采集条件下进行测试。通过键盘响应确定了翻转字母E的准确性、报告的应用logMAR值以及响应延迟，因此这些结果是行为输出，而非眼动仪的筛查结果。连续标准化注视向量之间的帧间变化角度（28.2度）作为GC308眼动仪得出的诊断，具有大约8帧每秒的有效采样率。在当前分析中，GC308眼动仪输出数据未进行屏幕目标校准或注视到屏幕的转换，因此这些数据代表了采集的稳定性，而非与目标相关的注视准确性。网络摄像头管道依赖于25点屏幕校准。由于显示分辨率和像素间距无法在研究记录中保留，报告的应用logMAR值不能被视为视觉敏锐度的任何衡量标准。
cs.RO / 16 / 2609.32064
Grasp2Twist: Learning Bimanual Dexterous Jar Opening by Reinforcement Learning
Grasp2Twist：通过强化学习学习双手灵巧开罐
Abstract
This paper presents Grasp2Twist, a bimanual dexterous manipulation system that learns to grasp and twist open jar lids using reinforcement learning. Learning this task raises three challenges: learning a unified policy for a multi-stage task, sustaining lid twisting, and sim-to-real transfer. To address the first challenge, we introduce a continuous enclosure measure to guide grasp formation and a binary enclosure indicator to guide the grasp-to-twist transition for unified policy learning. We derive both from the geometric relationship between the object center and the convex hull formed by the hand's palm and fingertips. Kinematic constraints limit how far the hand can rotate the lid with fixed contacts, so sustained twisting requires finger contact reconfiguration. We use a three-stage curriculum to facilitate exploration of these contact changes and also improve robustness for sim-to-real transfer. With our approach, the learned policy demonstrates finger gaiting, reconfiguring hand-object contacts to sustain lid rotation. It transfers zero-shot to the physical system and achieves an 88% task success rate across six household containers, including peanut-butter, vitamin, and instant-coffee jars. Ablations further validate the roles of the geometric enclosure in grasp formation and the curriculum in contact-reconfiguration exploration.
Chinese Translation
本文介绍了Grasp2Twist，一个双手灵巧操作系统，该系统通过强化学习学习抓取和扭开罐子盖。学习这一任务面临三个挑战：为多阶段任务学习统一策略、维持盖子的扭动以及模拟到现实的转移。为了解决第一个挑战，我们引入了一种连续包围度量来指导抓取形成，以及一种二元包围指示器来指导抓取到扭动的过渡，以实现统一策略学习。两者均源于物体中心与手掌和指尖形成的凸包之间的几何关系。运动学约束限制了手在固定接触下旋转盖子的距离，因此维持扭动需要重新配置手指接触。我们使用三阶段课程来促进对这些接触变化的探索，并提高模拟到现实转移的鲁棒性。通过我们的方法，学习到的策略展示了手指的步态，重新配置手-物体接触以维持盖子的旋转。该策略在物理系统中实现了零-shot转移，并在六种家庭容器（包括花生酱、维生素和速溶咖啡罐）中达到了88%的任务成功率。消融实验进一步验证了几何包围在抓取形成中的作用以及课程在接触重新配置探索中的重要性。
cs.RO / 17 / 2609.32069
Find Something You Can't Do: Agentic Real-World Reinforcement Learning for Self-Improving VLA Models
寻找无法完成的任务：自我改进的代理式真实世界强化学习框架用于 VLA 模型
Abstract
Vision--language--action (VLA) models provide strong priors for robotic manipulation but are typically deployed as frozen policies, unable to improve from their own failures. Real-world reinforcement learning (RL) offers a path to continued improvement, yet manual environment resets and task-success supervision hinder autonomous learning. We introduce \textbf{FIND}, an agentic real-world RL framework that closes the loop between scene understanding, weakness-aware practice, self-evaluation, and policy improvement in a persistent workspace. FIND reframes autonomous practice as a scene-conditioned, performance-aware task-selection problem: instead of restoring a predefined scene after each rollout, it uses the resulting scene to determine what to practice next. A vision--language agent identifies feasible tasks from a predefined library, prioritizes those with lower recent success rates, and evaluates outcomes using paired pre- and post-execution observations. We instantiate FIND with a frozen $\pi_{0.5}$ VLA and residual off-policy RL. Across eight real-world manipulation tasks, the independent human-assessed success rate improves from $55\%$ to $71.9\%$. A representative run completes 456 autonomous episodes within 6 hours of interaction, requiring 30 scene-recovery interventions and no human-provided reward labels during online learning. Ablations and systematic evaluations further examine key design choices, agent evaluation accuracy, and human intervention requirements. Our website is made publicly available at: FIND.github.io.
Chinese Translation
视觉-语言-动作 (VLA) 模型为机器人操作提供了强有力的先验知识，但通常作为固定策略部署，无法从自身的失败中改进。真实世界的强化学习 (RL) 提供了持续改进的途径，但手动环境重置和任务成功监督阻碍了自主学习。我们提出了 extbf{FIND}，一个代理式真实世界强化学习框架，它在持久工作空间中闭合了场景理解、弱点意识实践、自我评估和策略改进之间的循环。FIND 将自主实践重新定义为一个场景条件下的、性能意识的任务选择问题：它不是在每次回合后恢复预定义的场景，而是利用结果场景来确定接下来要实践的内容。一个视觉-语言代理从预定义库中识别可行任务，优先选择近期成功率较低的任务，并使用配对的执行前后观察结果进行评估。我们使用固定的 $ extpi_{0.5}$ VLA 和残差离线策略强化学习实现了 FIND。在八个真实世界的操作任务中，独立的人类评估成功率从 $55\%$ 提升至 $71.9\\%$。一个代表性的实验在 6 小时的交互中完成了 456 个自主回合，需进行 30 次场景恢复干预，并且在在线学习期间没有人类提供的奖励标签。消融实验和系统评估进一步考察了关键设计选择、代理评估准确性和人类干预需求。我们的网站已公开发布，地址为：FIND.github.io。
cs.RO / 18 / 2609.32079
Fiber-Normalized Manipulability and Determinant Proxies: Intrinsic Redundancy Optimization Across and Within Task Fibers
光纤归一化可操作性与行列式代理：任务光纤内外的内在冗余优化
Abstract
This work establishes that determinant-based manipulability is an exact objective for fixed-task redundancy optimization, despite its dependence on task coordinates and the choice of task-space metric used for volume measurement. On every regular task fiber, the determinant proxy, its representation in any task chart, and every metric-completed manipulability differ only by positive constants. They consequently induce the same complete ordering, constrained extrema, gradient directions, critical points, and local optimality classifications. For comparisons and trajectory optimization across task fibers, this work introduces fiber-normalized manipulability: the capability attained at an internal state divided by the best capability available on the same fiber. The resulting dimensionless scalar is invariant under coordinate changes on the internal-state and task manifolds and independent of the task-space metric. Its associated loss provides an intrinsic objective for physically admissible cross-fiber trajectories, including problems with prescribed or free task evolution and temporal coupling. Planar-manipulator and redundant aerodynamic-allocation examples demonstrate fixed-fiber equivalence, task-dependent cross-fiber differences, and invariant fiber-normalized trajectory optimization.
Chinese Translation
本研究确立了基于行列式的可操作性是固定任务冗余优化的确切目标，尽管它依赖于任务坐标和用于体积测量的任务空间度量的选择。在每个常规任务光纤上，行列式代理、其在任何任务图上的表示以及每个度量完备的可操作性仅相差正的常数。因此，它们引导相同的完全排序、约束极值、梯度方向、临界点和局部最优分类。为了在任务光纤之间进行比较和轨迹优化，本研究引入了光纤归一化可操作性：在内部状态下获得的能力除以在同一光纤上可用的最佳能力。由此产生的无量纲标量在内部状态和任务流形上的坐标变化下是不变的，并且独立于任务空间度量。其相关损失为物理上可接受的跨光纤轨迹提供了一个内在目标，包括具有规定或自由任务演变和时间耦合的问题。平面操纵器和冗余气动分配的示例展示了固定光纤等价性、任务依赖的跨光纤差异以及不变的光纤归一化轨迹优化。
cs.RO / 19 / 2609.32099
Active 3D weaves for load-bearing and damage-resilient locomotion
用于承载和抗损伤运动的主动三维织物
Abstract
The craft of weaving, where different materials are interlaced, has tremendous potential for creating active and functional systems for use in soft robots, prosthetics, wearables, exoskeletons, and more. These textile-like systems are flexible and safe for human-machine interaction; however, this inherent flexibility limits their ability to carry loads, which is essential for many robotic functions. In this work, we introduce a general framework for integrating active materials into three-dimensional (3D) woven shells to create robotic structures that combine high axial stiffness for load bearing, low bending stiffness for efficient actuation, and system-level resilience for damage tolerance. These woven robots can be modularly assembled from 'woven corners', a fundamental unit of 3D woven structures. We use eigenvalue calculations to identify load bearing and actuation mechanisms of the 3D woven structures, and use that information to make five different robots capable of locomotion. We demonstrate that these 3D woven robots can locomote carrying loads 70 times their self-weight, and can maintain repeatable performance even after being subjected to extreme compression. This work is a pathway toward the design, manufacturing, and simulation of future 3D woven robotic systems where load bearing, high stiffness, active functional deformation, locomotion, and system-level resilience are all needed.
Chinese Translation
编织工艺通过交织不同材料，具有为软机器人、假肢、可穿戴设备、外骨骼等创造主动和功能性系统的巨大潜力。这些类纺织系统灵活且安全，适合人机交互；然而，这种固有的灵活性限制了它们承载负荷的能力，而这对于许多机器人功能至关重要。在本研究中，我们提出了一个将主动材料集成到三维（3D）织物壳体中的通用框架，以创建结合高轴向刚度以承载负荷、低弯曲刚度以实现高效驱动以及系统级韧性以抵御损伤的机器人结构。这些织物机器人可以从“织物角”（3D织物结构的基本单元）模块化组装而成。我们使用特征值计算来识别3D织物结构的承载和驱动机制，并利用这些信息制造出五种不同的能够运动的机器人。我们证明这些3D织物机器人能够在承载其自重70倍的负荷下运动，并且即使在经历极端压缩后仍能保持可重复的性能。本研究为未来3D织物机器人系统的设计、制造和仿真提供了一条途径，其中承载、较高刚度、主动功能变形、运动和系统级韧性都是必需的。
cs.RO / 20 / 2609.32129
Residual Denoising Enables Sample-Efficient Multi-Agent Coordination on Demand
残差去噪促进按需的样本高效多智能体协调
Abstract
Pretrained robot policies offer strong manipulation skills but are typically limited to single-agent settings, where a robot acts in isolation. In this work, we study how to adapt pretrained single-agent diffusion policies to multi-agent settings using minimal collaborative data, co-optimizing for two key objectives: high coordination performance and single-agent skill retention. To this end, we introduce ALTER, an adaptation method for coordination on demand: the adapted policy coordinates with other robots when deployed in a team while remaining capable of acting independently when operating alone. Execution is decentralized: each robot acts only on its own visual observations, without explicit inter-agent communication. Our method trains a coordination head that predicts a residual denoiser to transform single-agent behavior into coordinated multi-agent behavior when necessary while also preserving single-agent capabilities. To preserve single-agent capabilities, we augment a small number of collaborative demonstrations with self-distilled data generated by the base policy during training of the residual denoiser. In simulation,ALTER achieves higher coordination success over our baselines while retaining much higher source-skill retention. In our hardware experiments, we find similar trends where ALTER better co-optimizes for coordination success and single-agent skill retention than the baselines
Chinese Translation
预训练的机器人策略提供了强大的操作技能，但通常仅限于单智能体环境，其中机器人独立行动。在本研究中，我们探讨如何利用最少的协作数据将预训练的单智能体扩散策略适应于多智能体环境，同时共同优化两个关键目标：高协调性能和单智能体技能保留。为此，我们提出了ALTER，一种按需协调的适应方法：适应后的策略在团队中部署时与其他机器人协调，同时在单独操作时仍然能够独立行动。执行是去中心化的：每个机器人仅根据自己的视觉观察进行行动，而无需明确的智能体间通信。我们的方法训练了一个协调头，预测残差去噪器，在必要时将单智能体行为转变为协调的多智能体行为，同时保留单智能体能力。为了保留单智能体能力，我们用在残差去噪器训练期间由基础策略生成的自蒸馏数据增强少量的协作示范。在仿真中，ALTER在协调成功率上优于我们的基线，同时保留了更高的源技能保留。在我们的硬件实验中，我们发现类似的趋势，即ALTER在协调成功率和单智能体技能保留方面的共同优化效果优于基线。
cs.RO / 21 / 2609.32154
GAUGE: Planner-Conditioned Active Calibration of Opaque Quadruped Velocity Interfaces
GAUGE：基于规划者条件的非透明四足动物速度接口的主动校准
Abstract
In this paper, we present a Goal-Aware Uncertainty-Guided Exploration (GAUGE) framework for planner-conditioned active calibration of opaque quadruped velocity interfaces. Commercial quadrupeds commonly expose planar-velocity commands, but the underlying locomotion controller remains inaccessible and can produce systematic discrepancies between commanded and realized motion. A navigation planner typically uses a structured subset of the command envelope. GAUGE maintains a Bayesian command-to-motion model and selects authorized trials according to their expected reduction of posterior epistemic uncertainty under the planner-induced command distribution. The resulting posterior supports validation-based stopping, bounded inverse compensation, and task-relevant recalibration after detected interface shifts. In three controlled response families, GAUGE reaches the joint criterion for task-facing accuracy and uncertainty with fewer trials than passive, D-optimal, and task-agnostic alternatives. Across six held-out Isaac Sim navigation maps, it meets the declared noninferiority margins against dense calibration. Code is available at https://github.com/EurekaZang/CalibAgent.
Chinese Translation
本文提出了一种目标感知不确定性引导探索（GAUGE）框架，用于基于规划者条件的非透明四足动物速度接口的主动校准。商业四足机器人通常暴露平面速度指令，但其底层运动控制器是不可访问的，可能导致指令与实际运动之间的系统性差异。导航规划器通常使用指令包络的结构化子集。GAUGE 维护一个贝叶斯指令到运动模型，并根据其在规划者引导的指令分布下对后验知识不确定性预期减少的程度选择授权试验。由此产生的后验支持基于验证的停止、有限逆补偿和在检测到接口变化后的任务相关重新校准。在三种受控响应类别中，GAUGE 以比被动、D-最优和任务无关的替代方案更少的试验次数达到了面向任务的准确性和不确定性的联合标准。在六个保留的 Isaac Sim 导航地图上，它满足了与密集校准的非劣性边界。代码可在 https://github.com/EurekaZang/CalibAgent 获取。
cs.RO / 22 / 2609.32155
RecastVLA: From Past Interaction to Future Control with Adaptive Policy States
RecastVLA：从过去的交互到未来的控制与自适应策略状态
Abstract
Sequential manipulation requires a robot to track what has already happened, even when the current scene no longer reveals it. Policies with explicit history representations make past interactions available as context for current decisions. We ask how action generation itself can form a persistent state for subsequent control. Building on action-side test-time training, RecastVLA maintains an adaptive policy state within a flow-matching vision-language-action policy. The state is represented by shared fast weights and remains fixed throughout action generation. Depth-specific interfaces read the same state, while features across depths and flow evaluations jointly define one update for the next policy call. Subsequent action losses train the initialization, interfaces, and update rule by differentiating through earlier state transitions. At deployment, updates use the policy's own action-generation features without expert action labels. Across LIBERO, RoboTwin, RoboDojo, and twelve real-robot tasks, RecastVLA improves mean success over a matched policy trained without test-time training, including 10.68 percentage points on RoboTwin Clean-to-Clean. In controlled RoboTwin comparisons, retaining state improves success, and the shared design exceeds independently trained layer-local TTT by 2.58 points.
Chinese Translation
顺序操作要求机器人跟踪已经发生的事件，即使当前场景不再显示这些信息。具有明确历史表示的策略使得过去的交互可以作为当前决策的上下文。我们探讨如何将动作生成本身形成一个持久状态，以便于后续控制。在动作侧的测试时训练基础上，RecastVLA 在一个流匹配的视觉-语言-动作策略中维持一个自适应策略状态。该状态由共享的快速权重表示，并在整个动作生成过程中保持不变。深度特定接口读取相同的状态，而跨深度和流评估的特征共同定义了下一个策略调用的一个更新。后续的动作损失通过对早期状态转变的微分训练初始化、接口和更新规则。在部署时，更新使用策略自身的动作生成特征，而不依赖专家动作标签。在 LIBERO、RoboTwin、RoboDojo 和十二个真实机器人任务中，RecastVLA 在没有测试时训练的匹配策略上提高了平均成功率，包括在 RoboTwin Clean-to-Clean 上提高了 10.68 个百分点。在受控的 RoboTwin 比较中，保持状态提高了成功率，而共享设计的表现超过了独立训练的层局部 TTT，提升了 2.58 个百分点。
cs.RO / 23 / 2609.32156
AquaBEV-Nav: Learned BEV Occupancy for Underwater Navigation and Exploration
AquaBEV-Nav：用于水下导航和探索的学习型鸟瞰视图占用预测
Abstract
Safe underwater exploration requires a robot to understand where surrounding structures are located and which regions are available for motion. Existing vision-based underwater exploration systems commonly obtain this information indirectly by estimating monocular depth, unprojecting the geometry into 3D space, and accumulating it into a 2D bird's-eye-view occupancy map. This reliance on intermediate depth estimation is particularly problematic underwater, where scattering and wavelength-dependent attenuation degrade visual cues and limit the reliability of monocular depth estimates. We introduce AquaBEV-Nav, an underwater exploration framework that bypasses explicit monocular depth estimation through direct bird's-eye-view occupancy prediction. Built upon the CORAL hierarchical exploration framework, AquaBEV-Nav replaces its depth-based perception front end with AquaBEV. Given a single RGB frame, AquaBEV maps visual features into a learned polar representation, performs causal reasoning along the range dimension, and reconstructs local Cartesian occupancy without relying on intermediate depth prediction. The resulting occupancy map is accumulated into CORAL's persistent spatial memory, providing spatial context for VLM-based high-level planning and collision constraints for dynamics-aware local trajectory generation. Across ten simulated reef environments and six occupancy backbones evaluated under a single protocol, AquaBEV-Nav reaches 37.48 structure IoU and 53.2 target IoU, 88.95% closed-loop coverage with zero collisions.
Chinese Translation
安全的水下探索要求机器人理解周围结构的位置以及可供运动的区域。现有的基于视觉的水下探索系统通常通过估计单目深度、将几何形状反投影到三维空间并累积成二维鸟瞰视图占用图来间接获取这些信息。这种对中间深度估计的依赖在水下尤其成问题，因为散射和波长依赖的衰减会降低视觉线索的质量，从而限制单目深度估计的可靠性。我们提出了AquaBEV-Nav，一个水下探索框架，通过直接的鸟瞰视图占用预测来绕过显式的单目深度估计。AquaBEV-Nav建立在CORAL分层探索框架之上，替换了其基于深度的感知前端，采用AquaBEV。给定一帧单一的RGB图像，AquaBEV将视觉特征映射到学习的极坐标表示中，沿范围维度进行因果推理，并在不依赖中间深度预测的情况下重建局部笛卡尔占用。生成的占用图被累积到CORAL的持久空间记忆中，为基于视觉语言模型（VLM）的高层规划提供空间上下文，并为动态感知的局部轨迹生成提供碰撞约束。在十个模拟的珊瑚礁环境和六个占用骨干网络的单一协议评估中，AquaBEV-Nav达到了37.48的结构交并比（IoU）和53.2的目标交并比（IoU），实现了88.95%的闭环覆盖率且没有发生碰撞。
cs.RO / 24 / 2609.32158
FutureRay: Control-Aligned Future Range for Agile Quadruped Navigation
FutureRay：面向灵活四足导航的控制对齐未来范围
Abstract
Moving obstacles can block a previously clear route while a quadruped robot executes a motion command. We investigate whether predicting changing clearance improves navigation when motion selection accounts for the robot footprint and the time needed to react and brake. We present FutureRay, which predicts ranges across viewing directions and future times, together with encounter risk, from depth-derived range history and observable robot motion. Training emphasizes near-term clearance and penalizes errors that overstate available space. A local planner queries the same forecast for candidate headings and combines it with current observations to check clearance around the robot footprint. Model-based reaction--braking limits guide speed selection, and the resulting velocity commands are passed to a fixed locomotion policy. In paired evaluations on 60 static and dynamic simulation scenes, FutureRay achieves 93.3% completion, compared with 75.0% for current-range persistence and 80.0% for Cartesian Kalman rollout, with perception, planning, and locomotion held fixed. FutureRay also records fewer collisions than both baselines. Qualitative trials on a physical quadruped show avoidance initiated while an obstacle is approaching the route, followed by renewed goal progress. These results show that joint range and encounter-risk prediction can improve obstacle avoidance without retraining the locomotion policy.
Chinese Translation
移动障碍物可能会在四足机器人执行运动指令时阻挡先前清晰的路线。我们研究了在运动选择考虑机器人的足迹以及反应和制动所需时间的情况下，预测变化的间隙是否能改善导航。我们提出了FutureRay，它从深度衍生的范围历史和可观察的机器人运动中预测不同视角和未来时间的范围，以及相遇风险。训练强调近期间隙，并惩罚高估可用空间的错误。局部规划器查询相同的预测以获取候选航向，并将其与当前观察结合，以检查机器人足迹周围的间隙。基于模型的反应-制动限制指导速度选择，生成的速度指令传递给固定的运动策略。在60个静态和动态仿真场景的配对评估中，FutureRay的完成率为93.3%，而当前范围持久性为75.0%，笛卡尔卡尔曼展开为80.0%，感知、规划和运动保持不变。FutureRay的碰撞次数也低于这两个基线。在物理四足机器人上的定性试验显示，在障碍物接近路线时启动避障，随后重新向目标推进。这些结果表明，联合范围和相遇风险预测可以在不重新训练运动策略的情况下改善避障能力。
cs.RO / 25 / 2609.32236
RoboFFT: Finetuning generative robot policy via online reinforcement learning with forward process
RoboFFT：通过在线强化学习与前向过程微调生成机器人策略
Abstract
Generative models, such as diffusion and flow-based models, have shown strong promise for robot policy learning by capturing complex and multimodal action distributions from demonstrations. However, policies trained solely with imitation learning often suffer from imperfect demonstrations and distributional shifts, while further improvement typically requires additional expert data. Reinforcement learning offers a natural solution through environment interaction, but effectively finetuning generative robot policies remains challenging due to the intractability of likelihood estimation. In this work, we propose RoboFFT, a forward-process reinforcement learning framework for finetuning generative robot policies, which applies forward noising to sampled actions and uses the weighted score / flow matching loss to construct a surrogate policy ratio for PPO-style updates. We evaluate RoboFFT with popular generative robot policies on representative simulation benchmarks, including long-horizon planning and sparse reward settings. Extensive experiments and analysis demonstrate that RoboFFT consistently improves performance while achieving better stability and training efficiency. We further integrate RoboFFT into a real world RL framework and demonstrate its effectiveness in real world tasks. Project website: https://student-of-holmes.github.io/RoboFFT/.
Chinese Translation
生成模型，如扩散模型和基于流的模型，在机器人策略学习中展现出强大的潜力，能够从示范中捕捉复杂和多模态的动作分布。然而，仅通过模仿学习训练的策略往往受到不完美示范和分布转移的影响，而进一步改进通常需要额外的专家数据。强化学习通过环境交互提供了一个自然的解决方案，但由于似然估计的不可处理性，有效微调生成机器人策略仍然具有挑战性。在这项工作中，我们提出了RoboFFT，一个用于微调生成机器人策略的前向过程强化学习框架，该框架对采样动作应用前向噪声，并使用加权得分/流匹配损失构建PPO风格更新的替代策略比率。我们在代表性模拟基准上评估了RoboFFT，涵盖了长时间规划和稀疏奖励设置，包括流行的生成机器人策略。大量实验和分析表明，RoboFFT在提高性能的同时，能够实现更好的稳定性和训练效率。我们进一步将RoboFFT集成到现实世界的强化学习框架中，并展示其在现实任务中的有效性。项目网站：https://student-of-holmes.github.io/RoboFFT/
cs.RO / 26 / 2609.32239
Federated Subspace Guided Vision-Language-Action Policy Distillation for Non-IID Multi-Robot Manipulation
用于非独立同分布多机器人操作的联邦子空间引导视觉-语言-动作策略蒸馏
Abstract
Federated learning offers a natural way for multiple robots to jointly improve manipulation policies without requiring centralized access to training demonstrations. However, non-IID task and environment distributions can induce representation drift and mutually incompatible robot-policy updates, making naive parameter aggregation destructive. We present FedDRMan, a federated subspace-guided distillation framework for heterogeneous robot manipulation. At each communication round, the server model provides a frozen teacher for local behavior cloning, while low-rank multimodal subspace and action-distribution distillation preserve globally useful representation geometry and policy behavior. To address heterogeneous aggregation, FedDRMan groups clients by update compatibility and maintains a persistent model for each cluster. The server then spectrally rebalances each compatible aggregate to mitigate attenuation of weaker task-relevant robot-policy update directions. Extensive experiments on LIBERO across diverse non-IID settings, heterogeneity levels, client participation variation, together with ablations and aggregation analyses, show that FedDRMan substantially improves knowledge transfer and consistently outperforms strong federated baselines achieving a peak mean success rate of 80.7%, 11.6 percentage points above the strongest evaluated federated baseline.
Chinese Translation
联邦学习为多个机器人共同改善操作策略提供了一种自然的方法，而无需集中访问训练示例。然而，非独立同分布的任务和环境分布可能导致表示漂移和相互不兼容的机器人策略更新，使得简单的参数聚合变得具有破坏性。我们提出了 FedDRMan，一种用于异构机器人操作的联邦子空间引导蒸馏框架。在每个通信回合中，服务器模型提供一个冻结的教师模型用于本地行为克隆，而低秩多模态子空间和动作分布蒸馏则保留了全球有用的表示几何和策略行为。为了解决异构聚合问题，FedDRMan 按照更新兼容性对客户端进行分组，并为每个集群维护一个持久模型。然后，服务器对每个兼容聚合进行谱重平衡，以减轻较弱任务相关机器人策略更新方向的衰减。在 LIBERO 上进行的广泛实验涵盖了多种非独立同分布设置、异构程度、客户端参与变化，以及消融实验和聚合分析，结果表明 FedDRMan 显著改善了知识转移，并始终优于强大的联邦基线，达到了 80.7% 的峰值平均成功率，比评估的最强联邦基线高出 11.6 个百分点。
cs.RO / 27 / 2609.32253
DS-VLA: A Dendritic-inspired Vision-Language-Action Model for Robust Action Control
DS-VLA：一种基于树突灵感的视觉-语言-动作模型用于稳健的动作控制
Abstract
Vision-language-action (VLA) models have achieved strong performance in language-conditioned manipulation, yet success under nominal evaluation does not necessarily translate into robust closed-loop behavior when executed actions are transiently corrupted. We introduce DS-VLA, a dendritic-inspired action architecture that incorporates dendritic spiking dynamics into VLA control to address this limitation. Specifically, to enable modularized feature processing and temporal information integration, DS-VLA equips action neurons with multiple sparsely connected dendritic branches, each featuring heterogeneous, learned decay factors. Furthermore, to suppress unreliable state updates while preserving task-relevant historical information, we introduce a neuron-wise inhibitory gate that adaptively regulates the admission of new multimodal evidence into dendritic states prior to somatic dynamics. We evaluate DS-VLA on all four LIBERO suites under both nominal rollouts and a unified closed-loop action-perturbation protocol. DS-VLA achieves a 91.6\% average nominal success rate and an 87.35\% average perturbed success rate, retaining 95.4\% of its nominal performance. Under the same reported perturbation setting, OpenVLA-OFT, FAST, $\pi_0$, and GR00T achieve 39.45\%, 23.90\%, 28.55\%, and 30.75\%, respectively. A controlled ablation isolates the contribution of neuron-wise shared inhibition, while analyses of neural dynamics and post-perturbation trajectories associate robust performance with selective evidence suppression and effective behavioral recovery. Together, these results demonstrate that integrating brain-inspired computational mechanisms offers a promising architectural prior for robust embodied intelligence beyond merely scaling vision-language backbones or generative action decoders.
Chinese Translation
视觉-语言-动作（VLA）模型在语言条件下的操作中表现出色，但在名义评估下的成功并不一定能转化为在执行动作时出现瞬时干扰的稳健闭环行为。我们提出了DS-VLA，这是一种基于树突灵感的动作架构，将树突脉冲动态融入VLA控制中，以解决这一局限性。具体而言，为了实现模块化特征处理和时间信息整合，DS-VLA为动作神经元配备了多个稀疏连接的树突分支，每个分支具有异质的、学习得到的衰减因子。此外，为了在保留与任务相关的历史信息的同时抑制不可靠的状态更新，我们引入了一种神经元级的抑制门，能够自适应地调节新多模态证据进入树突状态的过程，优先于体细胞动态。我们在所有四个LIBERO套件上评估了DS-VLA，包括名义回放和统一的闭环动作干扰协议。DS-VLA实现了91.6%的平均名义成功率和87.35%的平均干扰成功率，保留了95.4%的名义性能。在相同的干扰设置下，OpenVLA-OFT、FAST、$ ext{π}_0$和GR00T的成功率分别为39.45%、23.90%、28.55%和30.75%。通过控制性消融实验，我们隔离了神经元级共享抑制的贡献，同时对神经动态和干扰后轨迹的分析将稳健性能与选择性证据抑制和有效行为恢复关联起来。综合来看，这些结果表明，整合受脑启发的计算机制为超越单纯扩展视觉-语言骨干或生成动作解码器的稳健体现智能提供了有前景的架构先验。
cs.RO / 28 / 2609.32292
Affordance-Conditioned Decision Making: Bridging the Semantic-Spatial Gap in Zero-Shot Cross-Floor Vision-and-Language Navigation
基于可供性条件的决策制定：弥合零-shot跨楼层视觉与语言导航中的语义-空间差距
Abstract
Vision-and-language navigation increasingly relies on general-purpose semantic planners, yet translating correct high-level intent into reliable physical execution remains difficult in spatially constrained transitions. Reaching a staircase, doorway, or narrow passage does not ensure traversal; the agent must identify an executable affordance pose and recover from accumulated action errors. We propose PACE (Preference-refined Affordance-Conditioned Execution), a supervised local execution module that augments frozen zero-shot semantic planners for reliable cross-floor navigation. PACE grounds transition-related semantics into a long-horizon, agent-centric traversable affordance pose and conditions short-horizon action generation on this spatial target, thereby aligning semantic goals with physical execution. We further post-train PACE through failure-aware preference refinement using rollout-derived pairs that contrast normal or recovery behaviors with deviation-amplifying behaviors, thereby improving closed-loop correction. We integrate PACE into six open-source zero-shot VLN navigators and demonstrate consistent improvements on the cross-floor subsets of R2R-CE and RxR-CE, increasing the average success rate from 16.35% to 27.65% and from 4.76% to 12.06%, respectively. Real-world experiments further demonstrate PACE's applicability in unseen environments, highlighting the potential of traversable affordances to bridge semantic intent and reliable embodied behavior.
Chinese Translation
视觉与语言导航日益依赖于通用语义规划器，但在空间受限的转换中，将正确的高层意图转化为可靠的物理执行仍然困难。到达楼梯、门口或狭窄通道并不确保能够通过；代理必须识别可执行的可供性姿态，并从累积的行动错误中恢复。我们提出了PACE（Preference-refined Affordance-Conditioned Execution），这是一个监督的局部执行模块，增强了冻结的零-shot语义规划器，以实现可靠的跨楼层导航。PACE将与转换相关的语义基础建立在一个长时间跨度、以代理为中心的可遍历可供性姿态上，并将短时间跨度的行动生成条件于这一空间目标，从而将语义目标与物理执行对齐。我们进一步通过使用对比正常或恢复行为与偏差放大行为的回滚派生对进行失败感知的偏好细化，后训练PACE，从而改善闭环校正。我们将PACE集成到六个开源的零-shot视觉语言导航器中，并在R2R-CE和RxR-CE的跨楼层子集上展示了一致的改进，将平均成功率分别从16.35%提高到27.65%和从4.76%提高到12.06%。现实世界的实验进一步证明了PACE在未见环境中的适用性，突显了可遍历可供性在弥合语义意图与可靠体现行为之间的潜力。
cs.RO / 29 / 2609.32313
MemTransfer: Benchmarking Memory Beyond Matched Experience in Embodied Decision-Making
MemTransfer：在具身决策中超越匹配经验的记忆基准测试
Abstract
Memory lets an embodied agent reuse past experience, yet retaining useful information does not ensure that the agent can apply it when conditions change. We present MemTransfer, a benchmark comparing six memory representations, a working-memory baseline and five representations of past experience, under a shared frozen vision-language-model policy. It comprises 100 navigation cases across ten task types in a simulated warehouse, with expert demonstrations supplying the history. Three comparisons vary the starting pose, route availability, and amount and task relevance of history. With one demonstration per task, Full-context and Episodic memory reach 95.3% and 100.0% success at the original demonstration start, but lose 48-49 percentage points at a new test start. Summary changes little between these two test starts, yet with four demonstrations per task it retains a smaller fraction of its unchanged-route success after blocking (39.3%) than Working memory (44.8%) or the two trajectory memories (56-58%). At the new test start, increasing from one to four relevant demonstrations raises Episodic success by 14.3 percentage points, while the other evaluated representations gain no more than 1.3 percentage points. Replacing half of the relevant histories with other-task experience lowers success for both trajectory memories. These results show that robustness to one kind of mismatch does not imply robustness to another, motivating evaluation of both stored information and its use at decision time.
Chinese Translation
记忆使具身智能体能够重用过去的经验，但保留有用信息并不确保智能体在条件变化时能够应用这些信息。我们提出了MemTransfer，这是一个基准测试，比较六种记忆表示，包括一个工作记忆基线和五种过去经验的表示，在一个共享的冻结视觉-语言模型策略下进行。该基准测试涵盖了在模拟仓库中进行的100个导航案例，涉及十种任务类型，专家演示提供了历史数据。三种比较方式改变了起始姿态、路径可用性以及历史的数量和任务相关性。在每个任务中有一个演示的情况下，完整上下文（Full-context）和情节记忆（Episodic memory）在原始演示起始时分别达到95.3%和100.0%的成功率，但在新的测试起始时失去48-49个百分点。这两个测试起始之间的摘要变化不大，但在每个任务有四个演示的情况下，摘要在阻断后保留的未变路径成功率（39.3%）小于工作记忆（Working memory，44.8%）或两个轨迹记忆（Trajectory memories，56-58%）。在新的测试起始时，从一个相关演示增加到四个相关演示使情节记忆的成功率提高了14.3个百分点，而其他评估的表示最多只增加1.3个百分点。用其他任务经验替换一半的相关历史会降低两个轨迹记忆的成功率。这些结果表明，对一种不匹配类型的鲁棒性并不意味着对另一种的鲁棒性，促使我们在决策时评估存储信息及其使用。
cs.RO / 30 / 2609.32336
WSM-Aware HRI: An IoT-Enhanced Framework for Early Detection and Norm-Guided Repair of Failures with LLM Guidance
WSM感知的人机交互：一种基于物联网的框架，用于早期检测和规范引导的故障修复，辅以大型语言模型指导
Abstract
Human-robot interaction (HRI) failures remain a major barrier to deploying robots in real-world environments. Prior work often treats failures as isolated technical faults or focuses on post-hoc recovery behaviors. In practice, many breakdowns arise because humans and robots operate under inconsistent assumptions about the current world state. We propose WSM-Aware HRI, an IoT-enhanced modular framework that unifies diverse HRI breakdowns as World-State Mismatches (WSMs) between a human's instruction-implied assumptions and a robot's grounded world model built from multimodal perception and digital augmentation. A Large Language Model (LLM) is used to make implicit assumptions explicit, map them to a small set of mismatch types, and specify the evidence needed for verification against the robot's world state. WSM-Aware HRI shifts failure handling from execution-time recovery to proactive mismatch detection during intention formation, enabling interventions guided by safety, norm compliance, and multi-user coordination with transparent explanations. We evaluate mismatch identification in ten everyday cases spanning both visual and latent-state mismatches. The system can accurately produce the expected output results, and ablations show that reliable identification depends on appropriate grounding representations and verification-oriented refinement. These results indicate that treating interaction breakdowns as explicit world-state mismatches enables earlier detection of impending failures and offers a principled mechanism for integrating external evidence and social constraints into human-robot interaction.
Chinese Translation
人机交互（HRI）故障仍然是将机器人部署到现实环境中的主要障碍。以往的研究通常将故障视为孤立的技术故障，或关注于事后恢复行为。实际上，许多故障的发生是因为人类和机器人在当前世界状态的假设上存在不一致。我们提出了WSM感知的人机交互（WSM-Aware HRI），这是一个增强物联网的模块化框架，将多样的人机交互故障统一为人类指令隐含假设与机器人基于多模态感知和数字增强构建的实际世界模型之间的世界状态不匹配（World-State Mismatches, WSMs）。利用大型语言模型（Large Language Model, LLM）将隐含假设显性化，将其映射到一小组不匹配类型，并指定验证所需的证据，以便与机器人的世界状态进行对比。WSM感知的人机交互将故障处理从执行时的恢复转变为在意图形成过程中主动检测不匹配，从而实现基于安全性、规范遵循和多用户协调的干预，并提供透明的解释。我们在十个日常案例中评估了不匹配识别，这些案例涵盖了视觉和潜在状态的不匹配。该系统能够准确生成预期的输出结果，消融实验表明，可靠的识别依赖于适当的基础表示和以验证为导向的细化。这些结果表明，将交互故障视为显式的世界状态不匹配能够更早地检测到即将发生的故障，并为将外部证据和社会约束整合到人机交互中提供了一个有原则的机制。
cs.RO / 31 / 2609.32346
Human Motion Prediction for Human-Robot Collaboration
人类运动预测在人与机器人协作中的应用
Abstract
This paper compares three human motion prediction methods - GMM-based clustering, TransFusion, and Graph-Mixer - evaluating their effectiveness in human-robot collaboration. These predictions are valuable for improving robot motion planning by anticipating human movements and enhancing safety and efficiency in dynamic environments.
Chinese Translation
本文比较了三种人类运动预测方法——基于高斯混合模型（GMM）的聚类、TransFusion和Graph-Mixer，评估它们在人与机器人协作中的有效性。这些预测对于通过预判人类运动来改善机器人运动规划，提升动态环境中的安全性和效率具有重要价值。
cs.RO / 32 / 2609.32354
Proactive Motion Planning for Human-Robot Cooperation
人机协作的主动运动规划
Abstract
This abstract addresses the incorporation of human motion prediction into proactive and dynamic human-aware motion planning, with the goal of enabling safe collaboration between humans and robots. A deep learning, graph-based model is used to forecast human motion and is integrated into a planning framework. This framework employs a static roadmap along with a time-variant A* algorithm to modify the trajectory of a UR5e manipulator. This method greatly improves human-robot interaction and enables proactive collision avoidance by combining precise motion forecasts with adaptive trajectory planning.
Chinese Translation
本摘要探讨了将人类运动预测纳入主动和动态人机感知运动规划的过程，旨在实现人类与机器人之间的安全协作。采用深度学习的图模型来预测人类运动，并将其集成到一个规划框架中。该框架利用静态路线图和时间变化的 A* 算法来调整 UR5e 机械臂的轨迹。此方法通过将精确的运动预测与自适应轨迹规划相结合，显著改善了人机交互，并实现了主动避碰。
cs.RO / 33 / 2609.32416
RE-0: Verified Recursive Improvement of Embodied Code-as-Policy Agents through Local On-Policy Distillation
RE-0：通过局部在线策略蒸馏验证的具身代码作为策略代理的递归改进
Abstract
Code-as-Policy agents accomplish long-horizon embodied tasks by generating and executing code, yet continually improving them with teachers that are stronger but not globally reliable remains a key challenge. Existing distillation methods typically treat the teacher's complete behavior as the supervision target and thus misassign training credit on states where the teacher fails. We propose RE-0, a recursively verified policy improvement framework: rather than assuming that the teacher globally outperforms the student, RE-0 requests local corrections from the teacher on the student's own failure histories and checks in the environment whether each correction is genuinely beneficial; verified corrections yield immediate improvement. Building on this, we propose RE-OPD, which turns verified interventions into supervision for on-policy distillation. Only counterfactually verified teacher interventions provide distribution-level supervision, weighted by their measured local benefit, and the improvement they induce is projected back into the standalone student, so both where supervision is applied and how much credit the teacher receives co-evolve with the student policy. We further prove that the student's per-round gain is lower-bounded by its verified intervention gain up to verification and projection error terms. Experiments across multiple Code-as-Policy embodied tasks show that RE-0 improves both teacher-assisted execution and the standalone student, and generalizes to novel robots and scenes.
Chinese Translation
代码作为策略的代理通过生成和执行代码来完成长期的具身任务，但如何在不具备全球可靠性的情况下，利用更强的教师不断改进这些代理仍然是一个关键挑战。现有的蒸馏方法通常将教师的完整行为视为监督目标，因此在教师失败的状态上错误地分配了训练信用。我们提出了RE-0，一个递归验证的策略改进框架：RE-0并不假设教师在全局上优于学生，而是要求教师对学生自身的失败历史进行局部修正，并在环境中检查每个修正是否确实有益；经过验证的修正会带来即时的改进。在此基础上，我们提出了RE-OPD，它将经过验证的干预转化为在线策略蒸馏的监督。只有经过反事实验证的教师干预才能提供分布级别的监督，且根据其测量的局部收益加权，而它们所引发的改进则被投射回独立的学生，因此监督的应用位置和教师获得的信用量与学生策略共同演化。我们进一步证明，学生每轮的收益下限由其经过验证的干预收益决定，直到验证和投影误差项。针对多个代码作为策略的具身任务的实验表明，RE-0不仅改善了教师辅助执行，还提升了独立学生的表现，并且能够推广到新型机器人和场景。
cs.RO / 34 / 2609.32453
DRAM: Delta-rule Recurrent Associative Memory for Robot Manipulation Policies
DRAM：用于机器人操作策略的增量规则递归关联记忆
Abstract
Robotic manipulation is inherently history-dependent, yet most pretrained robotic policies condition on only the current observation or a short temporal window. Equipping such policies with long-term memory remains challenging: existing approaches either feed the backbone multi-frame observation windows, which substantially increase inference cost, or rely on pre-defined semantic features, which limit task generality and may also require the retraining of the backbone to adapt to the memory. We introduce DRAM (Delta-rule Recurrent Associative Memory), a plug-and-play memory module that can be attached to a wide range of pretrained robotic policies, endowing them with long-horizon memory without architectural modification or backbone retraining, requiring only task-specific post-training of the memory module and action expert. DRAM maintains a fixed-size associative memory using gated delta-rule linear attention, with a modified update that incorporates all tokens within each frame in parallel. An architecture-agnostic readout integrates historical context into action prediction across different policy architectures. Experiments show that DRAM consistently improves frozen pretrained policies over short-context baselines and alternative compact memory designs, validating its effectiveness as a fixed-size, post-hoc memory module trained with the backbone frozen.
Chinese Translation
机器人操作本质上依赖于历史信息，然而大多数预训练的机器人策略仅基于当前观察或短时间窗口进行条件判断。为这些策略配备长期记忆仍然具有挑战性：现有方法要么提供多帧观察窗口给主干网络，这会显著增加推理成本，要么依赖于预定义的语义特征，这限制了任务的通用性，并可能需要对主干网络进行重新训练以适应记忆。我们提出了DRAM（增量规则递归关联记忆），这是一种即插即用的记忆模块，可以附加到广泛的预训练机器人策略上，使其具备长时间记忆，而无需对架构进行修改或对主干网络进行重新训练，仅需对记忆模块和动作专家进行任务特定的后训练。DRAM使用门控增量规则线性注意力维护固定大小的关联记忆，并通过修改的更新机制并行整合每帧中的所有标记。架构无关的读出机制将历史上下文整合到不同策略架构的动作预测中。实验表明，DRAM在短上下文基线和其他紧凑记忆设计上始终改善了冻结的预训练策略，验证了其作为固定大小、后置记忆模块的有效性，且在主干网络冻结的情况下进行训练。
cs.RO / 35 / 2609.32471
GlowTact: Simple and Compact Vision-Based Tactile Sensing with High Sensitivity and Spatial Resolution
GlowTact：一种简单紧凑的基于视觉的触觉传感器，具有高灵敏度和空间分辨率
Abstract
Vision-based tactile sensors (VBTS) provide rich contact information for robotic manipulation, but existing designs can be hard to simplify and adapt to the size and constraints of humanoid fingertips. We introduce \textbf{GlowTact}, a pressure-responsive vision-based tactile sensing mechanism that directly visualizes contact pressure. \textbf{GlowTact} requires only single-color, non-directional illumination, and the raw tactile image directly represents the pressure distribution without explicit geometry reconstruction. This simple sensing principle enables compact, customizable tactile sensors while preserving high sensitivity and rich spatial detail. We demonstrate gram-scale contact detection, accurate normal-force estimation, and reconstruction of fine contact geometry, including M1 screw threads. These results establish \textbf{GlowTact} as a practical new sensing technology for compact humanoid fingertips, combining a durable nitrile membrane and simple optical design with sensitive and information-rich tactile perception.
Chinese Translation
基于视觉的触觉传感器（VBTS）为机器人操作提供丰富的接触信息，但现有设计往往难以简化并适应类人手指的尺寸和限制。我们提出了 extbf{GlowTact}，一种压力响应的基于视觉的触觉传感机制，能够直接可视化接触压力。 extbf{GlowTact}仅需单色、非定向照明，且原始触觉图像直接表示压力分布，无需显式的几何重建。这一简单的传感原理使得紧凑、可定制的触觉传感器成为可能，同时保持高灵敏度和丰富的空间细节。我们展示了克级接触检测、准确的法向力估计，以及细微接触几何的重建，包括M1螺纹。这些结果确立了 extbf{GlowTact}作为一种实用的新型传感技术，适用于紧凑的类人手指，结合了耐用的氯丁橡胶膜和简单的光学设计，以及敏感且信息丰富的触觉感知。
cs.RO / 36 / 2609.32576
Assisting for Open-Ended Tasks: Goal-Oriented Shared Autonomy as a Particle Filter
开放式任务的辅助：基于粒子滤波的目标导向共享自主性
Abstract
A common approach for shared autonomy blends human inputs with autonomous assistance based on the human's likely goal. However, most existing approaches assume that a static set of possible goals is known a priori, which limits the use of such methods in unstructured assistive settings. We instead investigate how to enable shared autonomy with open-ended and dynamically changing goals. We formulate goal-oriented shared autonomy as a particle filter in which particles represent candidate human goals. Unlike conventional approaches with a fixed goal set, our transition model dynamically proposes new candidate goals as the interaction evolves, and human actions update the belief over these goals in real time. We instantiate this framework with foundation models (e.g., vision grounding and large language models) that propose context-relevant semantic goals, generate goal-conditioned assistance from low-level skill primitives, and refine those skills from human corrections. We assess our approach through a user study where 12 participants perform a variety of tabletop manipulation tasks with our method and state-of-the-art shared autonomy baselines. The results show that our particle filter-based approach reduces the amount of time users spend teleoperating the system and improves user satisfaction. User study videos: https://youtu.be/Ii26XuRqm9c
Chinese Translation
共享自主性的常见方法将人类输入与基于人类可能目标的自主辅助相结合。然而，大多数现有方法假设已知一组静态的可能目标，这限制了这些方法在非结构化辅助环境中的应用。相反，我们研究如何实现具有开放式和动态变化目标的共享自主性。我们将目标导向的共享自主性形式化为粒子滤波，其中粒子代表候选的人类目标。与传统的固定目标集方法不同，我们的转移模型在交互演变时动态提出新的候选目标，而人类的行动实时更新对这些目标的信念。我们使用基础模型（例如，视觉定位和大型语言模型）实例化这一框架，这些模型提出与上下文相关的语义目标，从低级技能原语生成目标条件的辅助，并根据人类的修正来完善这些技能。我们通过一项用户研究评估我们的方法，12名参与者使用我们的方法和最先进的共享自主性基线执行各种桌面操作任务。结果表明，我们基于粒子滤波的方法减少了用户远程操作系统的时间，并提高了用户满意度。用户研究视频：https://youtu.be/Ii26XuRqm9c
cs.RO / 37 / 2609.32591
Think Fast, Plan Selectively: Adaptive Deliberation for Efficient Data-Driven MPC
快速思考，选择性规划：高效数据驱动模型预测控制的自适应深思
Abstract
Data-driven model predictive control (MPC) combines learned world models with online trajectory optimization, achieving strong performance in continuous control. However, the per-step cost of sampling and evaluating hundreds of candidate trajectories restricts deployment to control frequencies well below what real-time robotics demands. Motivated by the dual-process theory of human cognition, which distinguishes between fast, intuitive processing (System 1) and slower, deliberative reasoning (System 2), we ask whether every decision requires the same degree of computational deliberation. We propose Fast-TD-MPC, a lightweight framework that adaptively routes between fast policy execution and test-time planning, reserving costly deliberation for states where it is most needed. Fast-TD-MPC delivers competitive task performance across 103 continuous control tasks while achieving up to ~4x faster inference. Under external disturbances, Fast-TD-MPC selectively falls back to planning, maintaining robustness comparable to the original planner.
Chinese Translation
数据驱动的模型预测控制（MPC）将学习到的世界模型与在线轨迹优化相结合，在连续控制中实现了强大的性能。然而，采样和评估数百条候选轨迹的每步成本限制了其在控制频率上应用于实时机器人所需的水平。受到人类认知的双重过程理论的启发，该理论区分了快速、直觉的处理（系统1）和较慢、深思的推理（系统2），我们提出了一个问题：每个决策是否都需要相同程度的计算深思。我们提出了Fast-TD-MPC，这是一种轻量级框架，能够自适应地在快速策略执行和测试时规划之间切换，将高成本的深思保留给最需要的状态。Fast-TD-MPC在103个连续控制任务中提供了具有竞争力的任务性能，同时实现了约4倍的推理速度提升。在外部干扰下，Fast-TD-MPC选择性地回退到规划，保持了与原始规划器相当的鲁棒性。
cs.RO / 38 / 2609.32595
RECAST: Recasting Vision-Language Semantics into an Actionable Cost Map for Robot Navigation
RECAST：将视觉-语言语义重塑为可操作的成本地图以实现机器人导航
Abstract
Safe and robust robot navigation across diverse environments requires a high-level understanding of complex scenes and the ability to carry it into stable motion. Recent works tackle this with learning-based models trained at scale and with approaches built on vision-language models (VLMs). However, learning-based models break down outside their training distribution, while VLM-based approaches bring that understanding but rarely ground it in the scene or align the action with it. To address these limitations, we present RECAST, a robot navigation framework that combines the reasoning of a VLM with the spatial grounding of vision foundation models to build an Actionable Cost map. Given the robot's front view and the user's instruction, we first decompose the scene with the VLM, judging which surfaces are traversable, which objects pose a risk, which heading to prefer, and which gaps are passable. Vision foundation models then ground these surfaces and objects in the image, and all four judgments are spatially recast into one compact cost map. This map both conditions the trajectory decoders and scores their proposals to select the one to execute. As the VLM's answers trail the live scene, both steps draw on cost maps from two points in time: the pivot frame the VLM judged, which carries all four judgments, and the current frame, whose terrain and collision costs are rebuilt from the current image. RECAST improves success over the strongest prior method by 13.3 points in simulation and 31.4 points on a real quadruped, and reduces the collision rate relative to it by 9.6 and 14.3 points, reaching the lowest collision rate among all methods. The project page is available at https://recast-nav.github.io/
Chinese Translation
在多样化环境中安全且稳健的机器人导航需要对复杂场景的高层次理解，并能够将其转化为稳定的运动。近期的研究通过大规模训练的基于学习的模型和建立在视觉-语言模型（VLMs）上的方法来解决这一问题。然而，基于学习的模型在其训练分布之外会失效，而基于VLM的方法虽然提供了理解，但很少将其与场景相结合或使动作与之对齐。为了解决这些局限性，我们提出了RECAST，这是一种结合了VLM推理和视觉基础模型空间定位的机器人导航框架，以构建可操作的成本地图。在给定机器人前视图和用户指令的情况下，我们首先利用VLM对场景进行分解，判断哪些表面是可通行的，哪些物体存在风险，优先选择哪个方向，以及哪些间隙是可通过的。然后，视觉基础模型将这些表面和物体在图像中进行定位，所有四个判断被空间重塑为一个紧凑的成本地图。该地图既为轨迹解码器提供条件，又对其提案进行评分，以选择要执行的方案。随着VLM的回答跟随实时场景，这两个步骤都依赖于两个时间点的成本地图：VLM判断的关键帧，包含所有四个判断，以及当前帧，其地形和碰撞成本是从当前图像重建的。RECAST在模拟中比最强的先前方法提高了13.3个百分点，在真实四足机器人上提高了31.4个百分点，并相对减少了9.6和14.3个百分点的碰撞率，达到了所有方法中最低的碰撞率。项目页面可访问 https://recast-nav.github.io/
cs.RO / 39 / 2609.32626
World SLAM Model: Joint World Modeling for SLAM and Navigation
世界SLAM模型：用于SLAM和导航的联合世界建模
Abstract
We introduce World SLAM Model (WSM), a unified framework that brings the SLAM paradigm directly into downstream navigation. Rather than treating SLAM merely as an upstream module that provides poses, maps or tokens, WSM adopts its core mechanisms, including incremental state updates with persistent memory and backend refinement of accumulated errors, to maintain a consistent world state during interaction. Given the current observation and a navigation goal, WSM predicts future visual states and jointly estimates their camera motion and dense geometry, grounding visual prediction in an evolving spatial world state. This spatial state is continuously updated as new observations arrive and provides the basis for action generation and closed-loop navigation. WSM is trained end-to-end with a joint navigation--SLAM objective, enabling downstream navigation to benefit directly from SLAM-style state maintenance and refinement while preserving accurate geometric estimation. Experiments demonstrate improved navigation performance together with strong SLAM accuracy, highlighting the potential of SLAM as an intrinsic mechanism for long-horizon world modeling and embodied interaction.
Chinese Translation
我们介绍了世界SLAM模型（World SLAM Model, WSM），这是一个统一框架，将SLAM范式直接引入下游导航。WSM并不是将SLAM仅仅视为一个提供位姿、地图或标记的上游模块，而是采用其核心机制，包括具有持久记忆的增量状态更新和累积误差的后端优化，以在交互过程中保持一致的世界状态。在给定当前观察和导航目标的情况下，WSM预测未来的视觉状态，并联合估计相机运动和稠密几何，将视觉预测基于不断演变的空间世界状态。随着新观察的到来，这一空间状态会不断更新，并为动作生成和闭环导航提供基础。WSM通过联合导航- SLAM目标进行端到端训练，使下游导航能够直接受益于SLAM风格的状态维护和优化，同时保持准确的几何估计。实验表明，导航性能得到了改善，同时SLAM的准确性也很强，突显了SLAM作为长期世界建模和具身交互的内在机制的潜力。
cs.RO / 40 / 2609.32634
PF-RL: Progress Field Reinforcement Learning via Goal-Conditioned Value Geometry for Vision-Language-Action Models
PF-RL：通过目标条件值几何的进展领域强化学习用于视觉-语言-动作模型
Abstract
Reinforcement Fine-Tuning~(RFT) has emerged as a promising paradigm for improving Vision-Language-Action~(VLA) policies, yet sparse task-level outcomes provide limited credit for intermediate transitions, especially in long-horizon manipulation. A natural approach is to model intermediate task progress and use it as dense feedback for policy improvement. Despite their architectural differences, existing progress-aware methods commonly formulate task progress as an explicit scalar prediction, providing limited structure for modeling how intermediate observations relate to the task goal, which may hinder effective transition-level credit assignment. We introduce Progress Field Reinforcement Learning (PF-RL), which learns a structured goal-conditioned progress representation over pretrained VLA features and converts it into dense credit for policy optimization. A lightweight shared Progress Field head maps current and goal representations into a compact progress space, where geometric distance induces goal-conditioned value, while complementary temporal and goal-structure objectives shape the learned geometry. Transition-level value changes naturally yield dense progress advantages, enabling fine-grained credit assignment for both offline policy improvement and online reinforcement fine-tuning. Extensive experiments on LIBERO, RoboTwin2.0, and real-world bimanual manipulation tasks show that PF-RL consistently improves policy performance over strong supervised fine-tuning, reinforcement fine-tuning, and progress-aware baselines.
Chinese Translation
强化微调（Reinforcement Fine-Tuning, RFT）已成为改善视觉-语言-动作（Vision-Language-Action, VLA）策略的有前景的范式，但稀疏的任务级结果对中间过渡提供的信用有限，尤其是在长时间范围的操作中。一种自然的方法是建模中间任务进展，并将其作为策略改进的密集反馈。尽管现有的进展感知方法在架构上存在差异，但通常将任务进展公式化为明确的标量预测，这为建模中间观察与任务目标之间的关系提供了有限的结构，这可能会阻碍有效的过渡级信用分配。我们提出了进展领域强化学习（Progress Field Reinforcement Learning, PF-RL），它学习基于预训练的VLA特征的结构化目标条件进展表示，并将其转换为策略优化的密集信用。一个轻量级共享的进展领域头将当前和目标表示映射到一个紧凑的进展空间，其中几何距离引入目标条件值，而互补的时间和目标结构目标则塑造学习到的几何结构。过渡级值的变化自然产生密集的进展优势，使得离线策略改进和在线强化微调的细粒度信用分配成为可能。在LIBERO、RoboTwin2.0和真实双手操作任务上的广泛实验表明，PF-RL在强监督微调、强化微调和进展感知基线之上始终改善策略性能。
cs.RO / 41 / 2609.32698
SEES: A Self-Evolving Embodied System via Failure-Guided VLA Policy Adaptation
SEES：一种通过失败引导的VLA策略自我演化的具身系统
Abstract
Recent vision-language-action (VLA) policies demonstrate promising generalization across diverse short-horizon tasks. However, they remain unreliable on long-horizon tasks, partly because the large-scale training data is biased toward single-stage manipulation tasks that are cheaper to demonstrate. A single weak atomic skill can cause failures across multiple multi-stage tasks. To address such failures, existing methods often require experts to identify the bottleneck and provide additional demonstrations, making the improvement costly and potentially impractical after deployment. To this end, we present a Self-Evolving Embodied System (SEES) that learns from failures and improves the VLA policy without additional expert demonstrations. SEES decomposes long-horizon tasks into atomic tasks and routes them to corresponding family policies. Each family consists of related atomic skills that share one VLA adapter. During execution, the system automatically monitors atomic-task outcomes to identify the most frequently failing atomic skills as the current bottlenecks. To overcome these bottlenecks, SEES constructs tailored RL tasks in simulation by restoring previously encountered states and generating task-specific success criteria with an LLM. Online RL updates the shared family adapters to promote positive transfer among related atomic skills and cumulative improvement across evolution rounds. Extensive experiments show that SEES can be integrated with different VLA backbones to progressively improve their long-horizon performance. We also observe continued improvement on unseen tasks, providing evidence of transfer beyond the evolution settings.
Chinese Translation
近期的视觉-语言-动作（VLA）策略在多样化的短期任务中展现出良好的泛化能力。然而，它们在长期任务中仍然不可靠，部分原因是大规模训练数据偏向于更便宜的单阶段操作任务。一项单一的弱原子技能可能导致多个多阶段任务的失败。为了解决这些失败，现有方法通常需要专家识别瓶颈并提供额外的演示，这使得改进成本高昂且在部署后可能不切实际。为此，我们提出了一种自我演化的具身系统（SEES），该系统能够从失败中学习，并在没有额外专家演示的情况下改进VLA策略。SEES将长期任务分解为原子任务，并将其路由到相应的家族策略。每个家族由共享一个VLA适配器的相关原子技能组成。在执行过程中，系统自动监控原子任务的结果，以识别当前瓶颈中最常失败的原子技能。为了克服这些瓶颈，SEES通过恢复先前遇到的状态并利用大型语言模型（LLM）生成特定任务的成功标准，在模拟中构建量身定制的强化学习（RL）任务。在线RL更新共享的家族适配器，以促进相关原子技能之间的积极迁移，并在演化轮次中实现累积改进。大量实验表明，SEES可以与不同的VLA骨干网络集成，以逐步提升其长期表现。我们还观察到在未见任务上的持续改进，提供了超越演化设置的迁移证据。
cs.RO / 42 / 2609.32745
MORPH: Self-Organising Multi-Robot Task Allocation via Neuroplasticity-Inspired Adaptive Topology
MORPH：通过神经可塑性启发的自组织多机器人任务分配
Abstract
Multi-robot task allocation (MRTA) in dynamic environments faces a fundamental tension: effective coordination requires learned structure, but that structure must adapt when conditions change. Existing methods resolve this by assuming prior task knowledge, a utility function, a cost matrix, or a trained policy making them brittle when deployed without such knowledge or when task distributions shift. We present MORPH(Multi-agent Online Rewiring through Plasticity-guided Hierarchy), a training-free MRTA framework where global allocation quality emerges from 4 local plasticity rules (synaptic, homeostatic, structural, and metaplasticity) applied to a directed pairwise preference matrix updated from runtime co-occurrence and task-completion feedback. MORPH requires no task model, no bid computation, and no offline training; response decisions use learned AGV-to-Picker preferences rather than a fixed proximity rule. Within the Gerkey-Mataric MRTA taxonomy, MORPH is the first method in the single-task, single-robot, instantaneous-assignment class to learn directed pairwise allocation preferences online. Evaluated on the TA-RWARE warehouse benchmark (8-24 agents, 4 maps, 800 steps per episode, 5 seeds), MORPH achieves 110% of all-to-all throughput at N=24 while using only 21% of possible coordination links as an efficiency advantage that grows monotonically with fleet size. Under spatial task distribution shift, MORPH degrades 3x less than proximity-based methods while its learned preferences remain uncorrelated with Manhattan distance. Systematic ablation confirms all four plasticity rules contribute measurably. Two allocation properties emerge without programming: cross-type preference dominance and progressive preference sparsification, mirroring the developmental refinement of biological neural circuits. Learned preferences are driven by task co-occurrence history, not spatial proximity.
Chinese Translation
动态环境中的多机器人任务分配（MRTA）面临着一个基本的矛盾：有效的协调需要学习到的结构，但当条件变化时，该结构必须适应。现有方法通过假设先前的任务知识、效用函数、成本矩阵或训练好的策略来解决这个问题，这使得它们在没有这些知识或任务分布变化时变得脆弱。我们提出了MORPH（多智能体在线重连通过可塑性引导的层次结构），这是一个无训练的MRTA框架，其中全球分配质量源于应用于从运行时共现和任务完成反馈中更新的有向成对偏好矩阵的四个局部可塑性规则（突触性、稳态性、结构性和元可塑性）。MORPH不需要任务模型、不需要竞标计算，也不需要离线训练；响应决策使用学习到的AGV与拣货员的偏好，而不是固定的接近规则。在Gerkey-Mataric MRTA分类法中，MORPH是单任务、单机器人、即时分配类别中第一个能够在线学习有向成对分配偏好的方法。在TA-RWARE仓库基准测试（8-24个代理，4个地图，每个回合800步，5个种子）中评估时，MORPH在N=24时实现了110%的全对全吞吐量，同时仅使用21%的可能协调链接，这一效率优势随着车队规模单调增长。在空间任务分布变化下，MORPH的性能下降比基于接近的方法低3倍，而其学习到的偏好与曼哈顿距离保持不相关。系统性消融实验确认所有四个可塑性规则都有显著贡献。没有编程就出现了两种分配特性：跨类型偏好主导性和渐进偏好稀疏化，反映了生物神经电路的发育精炼。学习到的偏好是由任务共现历史驱动的，而不是空间接近性。
cs.RO / 43 / 2609.32762
An Empirical Study on What Matters for Viewpoint-Generalizable Policies in Visual Imitation Learning
关于视觉模仿学习中视角可泛化策略的重要性实证研究
Abstract
Visual imitation learning is a promising approach to training robot manipulation policies capable of completing a wide variety of tasks. However, policies today remain brittle to viewpoint perturbations, making deployment in diverse environments a challenge. We present a controlled empirical study of which design choices allow visuomotor policies to generalize across viewpoints. We find that viewpoint generalization improves when dense visual tokens are retained and the action head participates in geometric reasoning. On a suite of simulated tasks that span a wide range of camera poses, we show that these design choices yield a policy that remains performant across viewpoints. As a practical consequence, a policy trained with these design choices also transfers zero-shot from simulation to the real world under random camera configurations.
Chinese Translation
视觉模仿学习是一种有前景的方法，用于训练能够完成多种任务的机器人操作策略。然而，目前的策略对视角扰动仍然脆弱，这使得在多样化环境中的部署面临挑战。我们进行了一项受控的实证研究，探讨哪些设计选择能够使视觉运动策略在不同视角间实现泛化。我们的研究发现，当保留密集的视觉标记并且动作头参与几何推理时，视角泛化效果得以改善。在一系列涵盖广泛相机姿态的模拟任务中，我们展示了这些设计选择能够产生在不同视角下仍然表现良好的策略。作为实际结果，采用这些设计选择训练的策略在随机相机配置下也能够实现从模拟到现实世界的零样本迁移。
cs.RO / 44 / 2609.32767
CLAP: Closed-Loop Alignment with Pressure for Precise Suction Manipulation
CLAP：基于压力的闭环对齐以实现精确的吸力操控
Abstract
Stacking and palletising demand precise placement: error left in one layer is inherited by the next, and a flat pad offers no feature to funnel a wrong pose into the right one. Top-down suction suits such dense arrangements, and suction has already been brought into vision-language-action (VLA) policies. What that work does not report, however, is a policy conditioned on a measured vacuum signal, or one that uses it to abandon an action already under way. Vision does not settle the question here, because at the moment it matters the cup and the face it holds occlude each other. We present CLAP, which makes the attachment state observable through a pressure module tapped into the vacuum line. The decoded reading replaces the suction command in the policy's proprioception, is fused with the visual features, and terminates the open-loop execution window so that the policy re-infers from a fresh observation. For data, we record goal-state disassembly on the physical robot and reverse the joint-state sequence offline, without a simulation replay. Targeted phase demonstrations, 8.3% of the training frames, cover the suction transitions and the configurations an interrupted grasp leaves behind. On a real Unitree Z1, one multi-task checkpoint reaches 96.67% average success in both colour settings, 16.67 and 10.00 points above the strongest baseline, its monochromatic four-block successes averaging 15.92 mm of error. Four ablation settings fall 5.00 to 13.33 points short. We will release code and trained weights.
Chinese Translation
堆叠和打包需要精确的放置：一层中留下的误差会被继承到下一层，而平坦的垫子没有特征可以将错误的姿态引导到正确的姿态。自上而下的吸力适合这种密集的排列，吸力已经被引入到视觉-语言-动作（VLA）策略中。然而，之前的工作并未报告基于测量的真空信号的策略，或是利用该信号放弃正在进行的动作。视觉在此并不能解决问题，因为在关键时刻，杯子和它所持有的面会相互遮挡。我们提出了CLAP，它通过连接到真空管道的压力模块使得附着状态可观察。解码后的读数替代了策略的本体感知中的吸力命令，与视觉特征融合，并终止开放循环执行窗口，以便策略从新的观察中重新推断。为了获取数据，我们在物理机器人上记录目标状态的拆解，并离线逆转关节状态序列，而无需模拟重放。针对特定阶段的演示，占训练帧的8.3%，涵盖了吸力过渡和中断抓取留下的配置。在真实的Unitree Z1上，一个多任务检查点在两种颜色设置下的平均成功率达到96.67%，比最强基线高出16.67和10.00个百分点，其单色四块的成功率平均误差为15.92毫米。四个消融设置的表现低于5.00到13.33个百分点。我们将发布代码和训练权重。
cs.RO / 45 / 2609.32779
Copper-Policy: Focus on the Representation for Robust Robot Manipulation
铜政策：聚焦于鲁棒机器人操作的表征
Abstract
World Action Models (WAMs) acquire behavioral priors by modeling future scene evolution, but predicting detailed futures in pixel or latent space incurs substantial cost. Recent evidence that co-training gains persist without test-time generation raises a question: what must a WAM learn to improve control? We introduce Copper-Policy, which learns a compact World representation with the policy rather than relying on a predefined target space. Through temporal joint-embedding prediction, it predicts future observation embeddings conditioned on task intention without reconstructing pixels. This prediction and action decoding shape the representation jointly, while the policy retains access to current-frame spatial detail for execution. Representation analyses show that the learned features better separate task-driven change from perturbations and provide complementary information for control. Compact prediction targets reduce training tokens per sample, enabling a 2B-parameter model trained in 9.67 hours on 8$\times$ RTX 5090 GPUs and 6$\times$ faster than Fast-WAM on matched A100 GPUs. Copper-Policy outperforms every compared method without embodied pretraining on RoboTwin and several embodied-pretrained VLAs on LIBERO-Plus (80.85%). On three challenging real-robot tasks, it performs comparably to $\pi_{0.5}$ and attains a higher average score. Together, these results show that Copper-Policy combines strong control performance with efficient training.
Chinese Translation
世界行动模型（World Action Models, WAMs）通过建模未来场景演变来获取行为先验，但在像素或潜在空间中预测详细的未来会产生相当大的成本。最近的证据表明，联合训练的收益在没有测试时生成的情况下依然存在，这引发了一个问题：WAM必须学习什么以改善控制？我们引入了铜政策（Copper-Policy），它通过策略学习紧凑的世界表征，而不是依赖于预定义的目标空间。通过时间联合嵌入预测，它在不重建像素的情况下，基于任务意图预测未来观察嵌入。这种预测和动作解码共同塑造表征，同时策略保留对当前帧空间细节的访问以进行执行。表征分析表明，学习到的特征更好地区分任务驱动的变化与扰动，并为控制提供互补信息。紧凑的预测目标减少了每个样本的训练标记，使得一个具有20亿参数的模型在8× RTX 5090 GPU上训练9.67小时，并且在匹配的A100 GPU上比Fast-WAM快6倍。铜政策在RoboTwin上超越了所有比较方法，并在LIBERO-Plus上超过了多个经过实体预训练的视觉语言模型（VLA），达到了80.85%的表现。在三个具有挑战性的真实机器人任务中，它的表现与π_{0.5}相当，并获得了更高的平均分。综合这些结果表明，铜政策将强大的控制性能与高效的训练相结合。
cs.RO / 46 / 2609.32783
CollisionGAT: Controller-Agnostic One-Step Collision Screening for Multi-Agent Motion
CollisionGAT：一种控制器无关的多智能体运动一步碰撞筛查方法
Abstract
Before a team of robots moves, each proposed step must be checked for collisions with other robots and with obstacles. We present CollisionGAT, a graph-attention network that reads the current and proposed states of moving agents together with locally relevant stationary obstacles and returns one collision-risk score per moving agent. Any controller can use these scores to accept, repair, replan, or postpone a proposed step. We mount CollisionGAT on a continuous path-following controller and on GATeD, an obstacle-blind D* Lite planner that uses typed vetoes to update its planning graphs. Exact geometric checks supply the training labels and independently audit every executed step.
Chinese Translation
在一组机器人移动之前，必须检查每个提议步骤是否与其他机器人和障碍物发生碰撞。我们提出了CollisionGAT，一种图注意力网络，它读取移动智能体的当前状态和提议状态，以及局部相关的静态障碍物，并为每个移动智能体返回一个碰撞风险评分。任何控制器都可以使用这些评分来接受、修复、重新规划或推迟提议的步骤。我们将CollisionGAT应用于一个连续路径跟踪控制器和GATeD，一个使用类型否决来更新其规划图的障碍盲D* Lite规划器。精确的几何检查提供训练标签，并独立审核每个执行步骤。
cs.RO / 47 / 2609.32806
Towards Kinematic Actionable Infeasibility Detection in Motion Planning
朝着运动规划中的运动学可操作性不可行性检测
Abstract
Motion planning in robotics requires not only computing collision-free paths but also certifying infeasibility when no such path exists. Complete methods are limited to low-dimensional spaces, while sampling-based planners scale efficiently but cannot provide finite-time infeasibility certificates, leaving this problem largely unresolved in high-dimensional spaces. In this letter, we present a geometry-driven framework for certifying infeasibility through an explicit resolution-dependent analysis of configuration space topology. Leveraging signed distance field representations, the proposed method traces separating manifolds induced by obstacle boundaries directly in configuration space, enabling both detection of infeasibility and identification of the specific geometric cause. To address computational challenges, we develop a parallel frontier-expansion algorithm that exploits GPU acceleration for efficient simplicial reconstruction in high-dimensional spaces. We validate the approach on 4-DOF and 5-DOF robot scenarios, certifying infeasibility within seconds for 4-DOF cases and under four minutes for 5-DOF cases. We further discuss avenues for improving scalability to higher-dimensional spaces.
Chinese Translation
机器人运动规划不仅需要计算无碰撞路径，还需要在不存在此类路径时证明不可行性。完整的方法仅限于低维空间，而基于采样的规划器在扩展性方面表现良好，但无法提供有限时间内的不可行性证明，这使得在高维空间中这一问题仍然未得到有效解决。在本文中，我们提出了一种基于几何的框架，通过对配置空间拓扑的显式分辨率依赖分析来证明不可行性。利用有符号距离场表示，所提出的方法直接在配置空间中追踪由障碍边界引起的分离流形，从而实现不可行性的检测和特定几何原因的识别。为了解决计算挑战，我们开发了一种并行前沿扩展算法，利用GPU加速在高维空间中进行高效的单纯形重建。我们在4自由度（4-DOF）和5自由度（5-DOF）机器人场景中验证了该方法，在4-DOF情况下能够在几秒内证明不可行性，而在5-DOF情况下则在四分钟内完成。此外，我们进一步讨论了提高向更高维空间扩展的可行性途径。
cs.RO / 48 / 2609.32837
Scanning While Imagining: A Scene-Graph World Model for Robotic Ultrasound Navigation
边扫描边想象：用于机器人超声导航的场景图世界模型
Abstract
Ultrasound (US) acquisition depends on the operator's ability to interpret anatomy and anticipate how the view will change with probe motion. Many robotic US navigation methods select actions without explicitly predicting these anatomical changes. We propose SonoGraph-WM, an action- and goal-conditioned world model for anticipatory probe navigation. The model represents anatomy as scene graphs (SGs), capturing visible structures, their geometry, and spatial relationships without synthesizing US images. Given a history of SGs and probe poses, a unified Transformer jointly predicts future SGs and poses. A receding-horizon planner recursively imagines candidate trajectories, selects the shortest predicted path reaching a goal graph, and follows it over a short execution horizon before replanning from new observations. To reduce reliance on tracked and anatomically annotated US sequences, we generate aligned SG--pose training data from computed tomography (CT) label maps along surface-constrained probe trajectories. On four held-out CT cases, spatial relation F1 remains above 93% over 20 prediction steps, and closed-loop navigation achieves 77.50% and 75.00% success for the gallbladder and pancreas, respectively, using annotation-derived SGs. In robot--phantom navigation experiments with label-map-derived SGs, the planner reached the target view in 73.7% of trials. These findings support CT-supervised anatomical world modeling for probe planning and highlight the importance of frequent observation updates for reliable navigation. Project Page: https://noseefood.github.io/us-sonograph-wm/
Chinese Translation
超声（US）采集依赖于操作员对解剖结构的理解能力以及对探头运动如何改变视图的预判。许多机器人超声导航方法在没有明确预测这些解剖变化的情况下选择行动。我们提出了SonoGraph-WM，这是一种基于动作和目标条件的世界模型，用于预期探头导航。该模型将解剖结构表示为场景图（SGs），捕捉可见结构、其几何形状和空间关系，而无需合成超声图像。给定一系列场景图和探头姿态，统一的Transformer共同预测未来的场景图和姿态。一个递归的前视规划器想象候选轨迹，选择到达目标图的最短预测路径，并在短期执行范围内跟随该路径，然后根据新的观察结果重新规划。为了减少对跟踪和解剖标注超声序列的依赖，我们从计算机断层扫描（CT）标签图中生成与表面约束探头轨迹对齐的场景图-姿态训练数据。在四个保留的CT案例中，空间关系F1在20个预测步骤中保持在93%以上，闭环导航在使用标注派生的场景图时，胆囊和胰腺的成功率分别达到了77.50%和75.00%。在使用标签图派生的场景图进行的机器人-幻影导航实验中，规划器在73.7%的试验中达到了目标视图。这些发现支持基于CT监督的解剖世界建模用于探头规划，并强调了频繁观察更新对可靠导航的重要性。项目页面：https://noseefood.github.io/us-sonograph-wm/
cs.RO / 49 / 2609.32855
FINE: Future-Informed Navigation Encoding for Data-Efficient Vision-Language Navigation
FINE：面向未来的导航编码以实现数据高效的视觉-语言导航
Abstract
Adapting vision-language navigation (VLN) policies to new environments is expensive because every additional route and instruction requires an embodied demonstration. Yet standard observation-to-action training uses only a small fraction of the information already contained in each trajectory. In particular, future observations reveal the instruction-relevant landmarks that the agent will encounter, including what they look like and how they are arranged in 3D. We introduce FINE, a Future-Informed Navigation Encoding framework that extracts this latent supervision from existing demonstrations. FINE equips a VLN backbone with two complementary auxiliary representations. First, explicit landmark tokens follow the ordered landmarks specified by the instruction and are trained to predict future landmark regions in both semantic 2D patch-feature space and viewpoint-dependent 3D geometric feature space. Second, an implicit future token learns to distinguish the landmark state that is actually reached from plausible same-scene counterfactual futures generated by a video world model. On R2R-CE and RxR-CE val-unseen, FINE improves InternVLA-N1 by 2.6 and 4.5 success-rate points, respectively, at full training data. More importantly, as demonstrations become limited, the benefit grows: at a 70% demonstration budget, FINE improves success rate by 6.8 points, recovering roughly one-third of the performance lost by reducing the training demonstrations. Project page is available at https://finevln.github.io/.
Chinese Translation
将视觉-语言导航（VLN）策略适应新环境是昂贵的，因为每条额外的路线和指令都需要一个具体的演示。然而，标准的观察到行动训练仅使用了每条轨迹中已有信息的一小部分。特别是，未来的观察揭示了代理将遇到的与指令相关的地标，包括它们的外观以及它们在三维空间中的排列。我们引入了FINE，一个面向未来的导航编码框架，从现有演示中提取这种潜在的监督信息。FINE为VLN主干网络配备了两个互补的辅助表示。首先，显式地标令牌遵循指令中指定的有序地标，并被训练以预测语义二维补丁特征空间和视点依赖的三维几何特征空间中的未来地标区域。其次，隐式未来令牌学习区分实际到达的地标状态与由视频世界模型生成的可行同场景反事实未来。在R2R-CE和RxR-CE val-unseen数据集上，FINE分别将InternVLA-N1的成功率提高了2.6和4.5个百分点，使用完整的训练数据。更重要的是，随着演示的限制，收益增长：在70%的演示预算下，FINE将成功率提高了6.8个百分点，恢复了大约三分之一因减少训练演示而损失的性能。项目页面可访问 https://finevln.github.io/。
cs.RO / 50 / 2609.32862
RoboFoundry: System-as-Policy Evolution for Self-Learning Embodied Agents
RoboFoundry：自学习具身智能体的系统作为策略演化
Abstract
A foundation model should not act in isolation as an embodied agent. Yet, existing methods often optimize individual components of the agent stack, such as memory, context, skills, or action interfaces, rather than treating the supporting system itself as a unified policy. Moreover, interaction alone does not yield self-improvement unless execution experience is converted into persistent, validated system changes. We therefore propose RoboFoundry, the first embodied agentic framework that formulates this process as Self-Evolving System-as-Policy. RoboFoundry diagnoses capability gaps in decision-making and memory management, converts execution traces into validated task-specific system updates, and promotes recurring improvements to the general system. Evolution operates over two complementary surfaces: a context system that manages active internal context and persistent file-system memory, and a hierarchical skill system that organizes atomic skills, reusable compositions, and failure-conditioned recovery. A shared semantic interface separates embodiment-invariant decisions from embodiment-specific execution, allowing evolved system capabilities to transfer across heterogeneous robots. On EmbodiedBench, RoboFoundry achieves state-of-the-art performance, notably improving GPT-5.5 by 27.8%. It also brings Qwen3.7-Plus to near parity with GPT-5.5 (70.3% vs. 72.7%), showing consistent gains from system-as-policy evolution across foundation models. For long-horizon memory, RoboFoundry outperforms all baselines on RoboMemArena by at least 39.0%, even against methods assisted by external foundation models. On LIBERO-PRO, it further outperforms Cap-Agent0 by 243.8%-679.7% across all perturbation types. In real-world deployments, RoboFoundry demonstrates zero-shot transfer and online evolution across robots and tasks, highlighting its potential for fully autonomous embodied agents.
Chinese Translation
基础模型不应作为具身智能体孤立行动。然而，现有方法往往优化智能体堆栈的单个组件，如记忆、上下文、技能或动作接口，而不是将支持系统本身视为统一的策略。此外，仅仅进行交互并不会带来自我改进，除非执行经验被转化为持久的、经过验证的系统变更。因此，我们提出了RoboFoundry，这是第一个将这一过程表述为自我演化的系统作为策略的具身智能体框架。RoboFoundry诊断决策和记忆管理中的能力差距，将执行轨迹转化为经过验证的任务特定系统更新，并促进对整体系统的持续改进。演化在两个互补的层面上进行：一个管理活动内部上下文和持久文件系统记忆的上下文系统，以及一个组织原子技能、可重用组合和失败条件恢复的层次技能系统。一个共享的语义接口将具身不变的决策与具身特定的执行分开，允许演化的系统能力在异构机器人之间转移。在EmbodiedBench上，RoboFoundry实现了最先进的性能，显著提高了GPT-5.5的表现，提升幅度达到27.8%。它还使Qwen3.7-Plus与GPT-5.5几乎持平（70.3%对72.7%），显示出系统作为策略演化在基础模型中的一致性收益。对于长时间记忆，RoboFoundry在RoboMemArena上的表现超过所有基线，至少提高了39.0%，即便是对外部基础模型辅助的方法。在LIBERO-PRO上，它在所有扰动类型中进一步超越了Cap-Agent0，提升幅度达到243.8%-679.7%。在实际部署中，RoboFoundry展示了跨机器人和任务的零样本迁移和在线演化，突显了其作为完全自主具身智能体的潜力。
cs.RO / 51 / 2609.32935
Communication-Aware Heterogeneous Graph Learning for Decentralized Multi-Human Multi-Robot Task Allocation
面向通信的异构图学习用于去中心化多人多机器人任务分配
Abstract
Multi-human multi-robot (MH-MR) teams combine robotic autonomy with human expertise, but effective task allocation requires coordinating scarce, dynamically available human support with distributed robot execution. Limited robot-robot and human-robot communication further complicates this coupling by delaying information exchange and supervisory intervention. We introduce CommHG, a communication-aware heterogeneous graph learning framework for decentralized MH-MR task allocation. CommHG represents coupled human-robot-task interactions through local graphs conditioned on information availability and age. Learned communication actions enable robots to decide when and which operator to query, and when to share information with peers. Allocation and communication are jointly optimized through cooperative multi-agent reinforcement learning, allowing the team to acquire useful information while managing limited communication and supervisory resources. We also introduce a benchmark integrating heterogeneous humans and robots, dynamic tasks and operational states, and constrained robot-robot and bidirectional human-robot links. Experiments across heterogeneous teams with up to 16 robots, 6 humans, and 112 tasks show that CommHG improves timely weighted mission completion as coordination scale increases, outperforming the strongest baseline in the Large scenario.
Chinese Translation
多人人多机器人（MH-MR）团队将机器人自主性与人类专业知识相结合，但有效的任务分配需要协调稀缺且动态可用的人类支持与分布式机器人执行。有限的机器人间和人机间通信进一步复杂化了这种耦合，延迟了信息交换和监督干预。我们提出了CommHG，一个面向通信的异构图学习框架，用于去中心化的MH-MR任务分配。CommHG通过基于信息可用性和时效性的局部图表示耦合的人机任务交互。学习到的通信动作使机器人能够决定何时以及向哪个操作员询问，以及何时与同伴共享信息。通过合作多智能体强化学习，任务分配和通信被共同优化，使团队能够在管理有限的通信和监督资源的同时获取有用信息。我们还引入了一个基准，整合了异构的人类和机器人、动态任务和操作状态，以及受限的机器人间和双向人机链接。在最多16个机器人、6个人类和112个任务的异构团队实验中，结果表明CommHG在协调规模增加时提高了及时加权任务完成率，在大型场景中超越了最强基线。
cs.RO / 52 / 2609.32958
Learning Geometry-Aware Virtual Fixtures From Sparse Demonstrations
从稀疏示范中学习几何感知的虚拟夹具
Abstract
In many teleoperation applications, collecting a large number of demonstrations as required for traditional probabilistic learning from demonstration (LfD) approaches may not be feasible. To still give operators the ability to intuitively create trajectories as Virtual Fixtures (VFs), we propose to leverage a motion prior in the learning process. Particularly, by using Linear Quadratic Tracking (LQT), users are able to define guiding trajectories from the demonstration of just a few via points. To account for orientation guidance, we further reformulate classical LQT on Riemannian manifolds, introducing an additional geometric prior. Through a probabilistic interpretation of the LQT solution, we derive a covariance estimate at each trajectory point which we use to modulate the stiffness of the resulting fixture, resulting in strong guidance around the via points and softer guidance when far away. The covariance information is also used to define a validity region of the fixture, allowing the operator to leave its influence area and conduct unmodeled tasks. We evaluate the proposed Riemannian LQT formulation in a set of toy examples and the full framework on a cutting task requiring high precision in a minimally invasive surgery setting on the da Vinci Research Kit (dVRK).
Chinese Translation
在许多远程操作应用中，收集大量示范数据以满足传统概率学习示范（LfD）方法的需求往往不可行。为了仍然赋予操作者直观创建轨迹作为虚拟夹具（Virtual Fixtures, VFs）的能力，我们提出在学习过程中利用运动先验。具体而言，通过使用线性二次跟踪（Linear Quadratic Tracking, LQT），用户能够仅通过少量关键点的示范定义引导轨迹。为了考虑姿态引导，我们进一步在黎曼流形（Riemannian manifolds）上重新构造经典的LQT，引入额外的几何先验。通过对LQT解的概率解释，我们推导出每个轨迹点的协方差估计，用于调节生成夹具的刚度，实现关键点附近的强引导及远离时的柔性引导。协方差信息还用于定义夹具的有效区域，允许操作者离开其影响范围执行未建模任务。我们在一组示例任务中评估了所提出的黎曼LQT方法，并在da Vinci Research Kit（dVRK）上针对需要高精度的微创手术切割任务验证了完整框架的有效性。
cs.RO / 53 / 2609.33000
TriDrive: Joint Driver, Vehicle, and Road Modeling for Forecasting and Driver Monitoring
TriDrive：联合驾驶员、车辆和道路建模用于预测和驾驶员监控
Abstract
Predicting how drivers, vehicles, and road scenes interact and evolve together is central to driver monitoring. Prior work models in-cabin activity or traffic-conditioned driver motion in isolation, motivating joint driver, vehicle, and road modeling with real-time on-vehicle evaluation. We introduce TriDrive, to our knowledge the first unified framework that jointly forecasts driver kinematics, vehicle dynamics, and road demands through an automation-conditioned transition model. Modality-specific encoders (an anchored kinematic representation of the driver, causal CAN-bus dynamics, and frozen V-JEPA 2 road latents with structured road margins) are connected by directed residual connections through which driver and road context refine vehicle forecasts. We evaluate TriDrive on three downstream tasks. On the public AIDE benchmark, its kinematic encoder recipe sets a new full-set state of the art (SOTA) among published baselines (48.05 versus 71.47 All-MPJPE). On 197.2 hours of naturalistic BATON subset, directed connections and road margins raise assistance-engaged PR-AUC by 0.084 for steering onset and 0.286 for time-to-collision drops. For real-time use, we distill the road encoders and run TriDrive on a comma four with an external 8 GB GPU, where a lightweight current-state warning probe updates at 5 Hz with 177 ms p95 latency while the joint model forecasts concurrently. The probe is above an openpilot-based baseline on human-labeled manual-driving warnings (AUROC 0.725 versus 0.563), and in a paired on-road study 14 drivers rate its warnings as more appropriate (+1.79) and timely (+2.67) than those of openpilot's driver-monitoring system.
Chinese Translation
预测驾驶员、车辆和道路场景如何相互作用并共同演变是驾驶员监控的核心。以往的研究通常孤立地建模车内活动或交通条件下的驾驶员运动，这促使我们提出联合驾驶员、车辆和道路建模，并进行实时车载评估。我们介绍了TriDrive，这是我们所知的第一个统一框架，通过自动化条件下的转移模型联合预测驾驶员运动学、车辆动力学和道路需求。特定模态的编码器（驾驶员的锚定运动学表示、因果CAN总线动力学以及带有结构化道路边界的冻结V-JEPA 2道路潜变量）通过有向残差连接相连，驾驶员和道路上下文通过这些连接来优化车辆预测。我们在三个下游任务上评估了TriDrive。在公共AIDE基准测试中，其运动学编码器的配方在已发布的基线中设定了新的全集状态的最佳（SOTA）记录（48.05对比71.47 All-MPJPE）。在197.2小时的自然驾驶BATON子集中，有向连接和道路边界使得转向开始的协助参与PR-AUC提高了0.084，碰撞时间下降的PR-AUC提高了0.286。为了实现实时使用，我们提炼了道路编码器，并在配备外部8 GB GPU的comma four上运行TriDrive，其中轻量级的当前状态警告探针以5 Hz的频率更新，具有177毫秒的p95延迟，同时联合模型同时进行预测。该探针在人工标注的手动驾驶警告上优于基于openpilot的基线（AUROC 0.725对比0.563），在一项配对的道路研究中，14名驾驶员评价其警告比openpilot的驾驶员监控系统更合适（+1.79）和及时（+2.67）。
cs.RO / 54 / 2609.33007
CAPEX: Efficiently Distilling Foundation Model Behavior into Deployable Robot Policies through Experience-Adaptive Reasoning
CAPEX：通过经验自适应推理高效提炼基础模型行为为可部署机器人策略
Abstract
Robot learning has largely relied on human-teleoperated demonstrations to acquire effective learnable behaviors. However, human-operated data collection processes can be unintuitive, difficult to scale, and inherently asynchronous. We explore an alternative: distilling physical behavior from general-purpose multimodal foundation models into deployable robot policies by using the foundation model itself as an autonomous demonstrator. While sufficiently capable models can generate successful zero-shot manipulation trajectories, repeatedly invoking them during physical execution is slow and expensive, limiting their utility as scalable data generators. As a solution, we introduce CAPEX, an experience-conditioned demonstration collection framework that uses execution experience from previous attempts to adapt how frequently the foundation model must observe, reason, and replan. We evaluate across RoboCasa tasks and on physical Franka and bimanual YAM-arm platforms, measuring task success, model calls, token usage, collection time, and cost. We further train Diffusion Policy and ACT on matched sets of human-teleoperated and foundation-model-generated demonstrations to evaluate the downstream learning value of autonomously collected data. We find that CAPEX increases the number of successful demonstrations by 4.3x while reducing the cost per successful demonstration by 80%. Policies trained on CAPEX-generated data approach the performance of those trained on matched human demonstrations; with longer training, this gap largely closes for policies trained from scratch. These results suggest that foundation models can serve as scalable sources of reusable robot experience. Project page: https://capex-paper.github.io/
Chinese Translation
机器人学习在很大程度上依赖于人类遥控演示来获取有效的可学习行为。然而，人类操作的数据收集过程往往不直观、难以扩展，并且本质上是异步的。我们探索了一种替代方案：通过将基础模型本身作为自主演示者，从通用多模态基础模型中提炼物理行为为可部署的机器人策略。虽然足够强大的模型可以生成成功的零-shot操作轨迹，但在物理执行过程中反复调用这些模型的速度较慢且成本高昂，限制了它们作为可扩展数据生成器的实用性。为此，我们提出了CAPEX，一个经验条件演示收集框架，利用之前尝试的执行经验来调整基础模型观察、推理和重新规划的频率。我们在RoboCasa任务以及物理Franka和双手YAM-arm平台上进行评估，测量任务成功率、模型调用次数、令牌使用量、收集时间和成本。我们进一步在匹配的人类遥控和基础模型生成的演示集上训练Diffusion Policy和ACT，以评估自主收集数据的下游学习价值。我们发现CAPEX将成功演示的数量提高了4.3倍，同时将每个成功演示的成本降低了80%。基于CAPEX生成数据训练的策略接近于基于匹配人类演示训练的策略性能；随着训练时间的延长，这一差距在从零开始训练的策略中基本缩小。这些结果表明，基础模型可以作为可扩展的可重用机器人经验来源。项目页面：https://capex-paper.github.io/
cs.RO / 55 / 2609.33049
Is Online Interaction Necessary for Recovery? A Minimalist Approach to Robust Planning via Perturbation
在线互动对恢复是否必要？通过扰动实现稳健规划的简约方法
Abstract
Behavior cloning (BC) is vulnerable to covariate shift during closed-loop execution, where small prediction or execution errors can drive the robot toward states poorly covered by the demonstration data. We focus on action-sequence planning, where a policy predicts a finite-horizon sequence of actions as a reference trajectory for robot execution. Existing approaches to covariate shift often rely on collecting additional corrective demonstrations, requiring further environment interaction and access to an expert or reference policy. We propose Perturbation-Augmented Recovery Supervision (PARS), a simple training approach for improving recovery using only existing demonstrations. PARS perturbs the robot's proprioceptive state and anchors only the later portion of the predicted trajectory to the original demonstration, leaving the earlier portion free to generate corrective motion and recover toward the demonstrated behavior within the planning horizon. Unlike conventional input-noise augmentation, which preserves the original supervision target over the entire prediction horizon, PARS explicitly provides trajectory-level supervision for recovery from perturbed states. PARS requires neither additional environment interaction nor expert queries and can be instantiated across different action-sequence policy classes with only minor modifications to their BC objectives. Experiments on 51 RLBench manipulation tasks with flow-based, transformer-based, and diffusion-based policies demonstrate that PARS improves robustness to covariate shift during closed-loop execution.
Chinese Translation
行为克隆（BC）在闭环执行过程中容易受到协变量偏移的影响，其中小的预测或执行误差可能会使机器人朝向演示数据覆盖不充分的状态。我们关注于动作序列规划，其中策略预测有限时间范围内的动作序列作为机器人执行的参考轨迹。现有的协变量偏移处理方法通常依赖于收集额外的纠正演示，这需要进一步的环境交互以及对专家或参考策略的访问。我们提出了扰动增强恢复监督（PARS），这是一种仅利用现有演示来改善恢复的简单训练方法。PARS扰动机器人的本体状态，并仅将预测轨迹的后半部分锚定到原始演示上，使得前半部分可以自由生成纠正运动，并在规划范围内恢复到演示行为。与传统的输入噪声增强方法不同，后者在整个预测范围内保持原始监督目标不变，PARS明确提供了从扰动状态恢复的轨迹级监督。PARS既不需要额外的环境交互，也不需要专家查询，并且可以在不同的动作序列策略类别中实例化，只需对其BC目标进行少量修改。在51个RLBench操作任务上进行的实验表明，PARS在闭环执行过程中提高了对协变量偏移的稳健性。
cs.RO / 56 / 2609.33053
SwingRL: Adaptive Observation Reinforcement Learning with World-Model Prediction for Cable-Suspended Hoisting Control
SwingRL：基于世界模型预测的自适应观察强化学习用于电缆悬挂起重控制
Abstract
Cable-suspended hoisting is widely used to move heavy or bulky payloads that cannot be handled conveniently by rigid pick-and-place systems, for example in crane-assisted construction. Robotic hoisting using flexible cables is challenging because payload motion is underactuated, external disturbances vary, and delayed or lost visual observations can make the perceived payload state stale at control execution. These effects are particularly critical during precise insertion of a suspended payload's sockets onto rebar pins, which is a very common task in construction environments. We present SwingRL, a residual reinforcement-learning (RL) framework that combines an age-aware world model, a classical anti-swing prior, and a recurrent residual policy to address two coupled problems: stale feedback and uncertain dynamics. The world model propagates the newest received payload observation to the current control step using the executed commands, providing a time-aligned state estimate under delayed and lossy sensing. The prior supplies nominal tracking and swing damping. The residual policy learns bounded corrections to the prior rather than the complete control law, compensating for system-parameter variation, external disturbances, and remaining state-estimation errors. We evaluate SwingRL against classical and learning-based baselines across a cumulative difficulty ladder covering system-parameter variation, wind disturbance, degraded sensing, and strong gusts. Under the most difficult setting, SwingRL achieves 69.5% strict and 77.3% broad success, exceeding all baselines by at least 60 percentage points, respectively. World-model ablations support the role of time-aligned state estimation in maintaining insertion success as observation loss increases. Finally, without real-robot fine-tuning, SwingRL achieves 90% success on the physical rig.
Chinese Translation
电缆悬挂起重广泛用于移动重型或笨重的货物，这些货物无法通过刚性抓取系统方便地处理，例如在起重机辅助的建筑中。使用柔性电缆的机器人起重面临挑战，因为货物运动是欠驱动的，外部干扰变化多端，延迟或丢失的视觉观察可能导致在控制执行时感知的货物状态过时。这些影响在将悬挂货物的插座精确插入到钢筋销上时尤为关键，这在建筑环境中是一项非常常见的任务。我们提出了SwingRL，一个残差强化学习（RL）框架，结合了一个考虑时间的世界模型、一个经典的抗摆动先验和一个递归残差策略，以解决两个耦合问题：过时反馈和不确定动态。世界模型利用执行的命令将最新接收到的货物观察传播到当前控制步骤，在延迟和丢失感知的情况下提供时间对齐的状态估计。先验提供了名义跟踪和摆动阻尼。残差策略学习对先验的有界修正，而不是完整的控制律，以补偿系统参数变化、外部干扰和剩余的状态估计误差。我们在涵盖系统参数变化、风干扰、感知退化和强阵风的累积难度梯度上评估SwingRL与经典和基于学习的基线的表现。在最困难的设置下，SwingRL实现了69.5%的严格成功率和77.3%的广泛成功率，分别超过所有基线至少60个百分点。世界模型的消融实验支持了在观察丢失增加时，时间对齐状态估计在维持插入成功中的作用。最后，在没有真实机器人微调的情况下，SwingRL在物理装置上实现了90%的成功率。
cs.RO / 57 / 2609.33095
REALM: A Coarse-to-Fine Generative Framework for Embodied Reactive Listening
REALM：一种粗到细的生成框架用于具身反应性倾听
Abstract
Generating responsive listener facial motion is an important task for embodied conversational AI. Two modeling challenges are central: accounting for the timing of speaker cues while maintaining continuity with the listener's ongoing motion, and capturing locally variable facial events alongside the overall motion trajectory. Listener responses may follow preceding cues with a temporal lag, while brief expressions and blinks introduce variation that is difficult to predict deterministically. These challenges motivate a framework that combines history-aware temporal alignment with stochastic expression refinement. We propose REALM (Reactive Embodied Audio-driven Listening Model), a coarse-to-fine framework for audio-driven reactive listening. A Reactive Gated Speaker-Listener Fusion module combines listener motion history with speaker audio through a delay-centered attention prior and adaptive gating. A coarse decoder predicts a base motion trajectory, which is augmented by audio-conditioned stochastic residuals in the expression subspace while retaining the coarse pose parameters. Evaluations on ViCo and L2L show improvements over the evaluated baselines across multiple motion-quality metrics. Additional analyses examine delay sensitivity, gate behavior, and blink dynamics. Finally, deployment on an Ameca humanoid robot and a perceptual user study demonstrate the applicability of the generated behavior to physical embodiment. Code: https://github.com/lipzh5/REALM Demo: https://youtu.be/Tf5mpd5S8VQ
Chinese Translation
生成响应性倾听者面部动作是具身对话人工智能中的一项重要任务。两个建模挑战是核心：在保持倾听者持续动作的连贯性的同时，考虑说话者提示的时机，以及捕捉局部变化的面部事件与整体运动轨迹。倾听者的反应可能会在时间上滞后于前面的提示，而短暂的表情和眨眼则引入了难以确定性预测的变化。这些挑战促使我们提出一个框架，结合历史感知的时间对齐与随机表情细化。我们提出了REALM（反应性具身音频驱动倾听模型），这是一个用于音频驱动反应性倾听的粗到细框架。反应性门控说话者-倾听者融合模块通过以延迟为中心的注意力先验和自适应门控，将倾听者的运动历史与说话者音频结合起来。粗略解码器预测基础运动轨迹，该轨迹在保持粗略姿态参数的同时，通过音频条件的随机残差在表情子空间中进行增强。在ViCo和L2L上的评估显示，在多个运动质量指标上，相较于评估的基线有显著改善。附加分析考察了延迟敏感性、门控行为和眨眼动态。最后，在Ameca类人机器人上的部署以及感知用户研究展示了生成行为在物理具身中的适用性。代码：https://github.com/lipzh5/REALM 演示：https://youtu.be/Tf5mpd5S8VQ
cs.RO / 58 / 2609.33096
Humanoids for Robot-Assisted Surgery: Bimanual Base Placement and Tool-Mount Optimization via Capability Maps
用于机器人辅助外科手术的人形机器人：通过能力图进行双手基础放置和工具安装优化
Abstract
Rapid advances in humanoid robotics have motivated growing interest in the application of humanoids for healthcare and clinical tasks. However, it remains unclear how close contemporary humanoids are to meeting the kinematic demands of robot-assisted laparoscopic surgery. In this work, we address the question of optimal robot positioning through a quantitative analysis of workspace and robot setup configurations. We present a capability-map-based robot setup framework that optimizes humanoid base placement and tool mounting orientation to maximize bimanual humanoid reachability while accounting for tool-tip kinematics and remote-center-of-motion (RCM) constraints. We evaluate three humanoid platforms spanning different body dimensions and kinematic redundancy on workspace reachability for three representative general surgery procedures: cholecystectomy, inguinal hernia repair, and sleeve gastrectomy. The proposed joint optimization of base placement and tool mounting consistently outperforms base-only optimization and heuristic baselines. For cholecystectomy and inguinal hernia repair, which are characterized by relatively small and minimally overlapping workspaces, humanoid reachability approached 90%. For the larger, overlapping multi-port arm workspace of sleeve gastrectomy, humanoids yield substantially lower coverage. These results quantify the near-term promise of humanoids for selected laparoscopic procedures and clarify key limitations that must be addressed for broader deployment.
Chinese Translation
人形机器人技术的快速进展激发了对人形机器人在医疗和临床任务中应用的日益关注。然而，目前尚不清楚当代人形机器人在满足机器人辅助腹腔镜手术的运动学要求方面有多接近。在本研究中，我们通过对工作空间和机器人设置配置的定量分析，解决了最佳机器人定位的问题。我们提出了一种基于能力图的机器人设置框架，该框架优化人形机器人的基础放置和工具安装方向，以最大化双手人形机器人的可达性，同时考虑工具尖端的运动学和远程运动中心（RCM）约束。我们评估了三种不同体型和运动学冗余的人形机器人平台在三种代表性普通外科手术（胆囊切除术、腹股沟疝修补术和袖状胃切除术）中的工作空间可达性。所提出的基础放置和工具安装的联合优化始终优于仅优化基础和启发式基线。在胆囊切除术和腹股沟疝修补术中，由于工作空间相对较小且重叠较少，人形机器人的可达性接近90%。而在袖状胃切除术的较大重叠多端臂工作空间中，人形机器人的覆盖率显著较低。这些结果量化了人形机器人在特定腹腔镜手术中的短期应用前景，并阐明了必须解决的关键局限性，以便更广泛的部署。
cs.RO / 59 / 2609.33101
Evolving Dexterous Robots from Scratch
从零开始进化灵巧机器人
Abstract
Little is known about how to manually design agents capable of dexterous manipulation. Some design principles have been inferred from close examination of how animals manipulate objects, but these structures and behaviors have so far resisted biomimicry and may not be optimal for artificial machines. Here we evolve freeform robots to pick up, hold, rotate, and use diverse objects. Unlike other approaches to optimizing robot hands, we do not presuppose the presence, articulation, or geometry of any part of the body. Although familiar prehensile forms such as tails, beaks, paws and claws may emerge spontaneously under certain conditions--and while such conditions could be of interest to evolutionary biologists--de novo manipulator design can also reveal whole new solutions, overlooked or unknown structures which may be better suited for the task at hand. We use contrastive learning to create a highly searchable genetic embedding of design space, an autoregressive developmental model to decode designs, evolutionary strategies to find good designs, and reinforcement learning to train each evolved design. Winning designs were automatically converted into a manufacturable blueprint, printed, assembled and tested in the real world in a zero-shot manner. The results represent the state-of-the-art in evolutionary robotics in terms of performance, diversity and complexity.
Chinese Translation
关于如何手动设计能够进行灵巧操作的智能体知之甚少。一些设计原则是通过仔细观察动物如何操控物体而推断出来的，但这些结构和行为迄今为止仍然抵制生物仿生，可能并不适合人工机器。在这里，我们进化出自由形态的机器人，以拾取、持有、旋转和使用各种物体。与其他优化机器人手的方法不同，我们并不预设身体任何部分的存在、关节或几何形状。尽管在某些条件下，熟悉的抓握形式如尾巴、喙、爪子和爪钩可能会自发出现——而这些条件可能对进化生物学家有趣——但全新操控器设计也可以揭示出完全新的解决方案，未被注意或未知的结构，这些结构可能更适合当前的任务。我们使用对比学习创建一个高度可搜索的设计空间遗传嵌入，采用自回归发展模型解码设计，利用进化策略寻找良好设计，并通过强化学习训练每个进化出的设计。获胜设计被自动转换为可制造的蓝图，进行打印、组装并在现实世界中以零样本方式进行测试。结果在性能、多样性和复杂性方面代表了进化机器人技术的最先进水平。
cs.RO / 60 / 2609.33104
VPTwin: Real-Sim-Real Video Prediction for Robotic Manipulation Planning
VPTwin：用于机器人操作规划的真实-模拟-真实视频预测
Abstract
While action-conditioned video prediction provides an intuitive world model for robotics, purely data-driven predictors often suffer from compounding errors and physically implausible hallucinations in long-horizon rollouts, severely undermining downstream action planning. We propose VPTwin, a Real-Sim-Real video prediction framework that anchors real-world future prediction using real-synchronized simulation twins. For a target manipulation task, a VLM reconstructs an executable digital twin from a real demonstration episode. To accommodate the ill-posed estimation of unobserved physical properties, Isaac Sim simulates multiple forward dynamic rollouts across randomized physical configurations under candidate action trajectories. Using these rollouts as in-context references, VPTwin harmonizes both domains, using simulation dynamics to enforce physical plausibility while capturing unmodeled contact interactions from real video. Furthermore, we establish a predictive planning loop using VPTwin to visually verify VLM-proposed actions and guide reliable real-world execution. Evaluations show substantial reductions in physical hallucinations during video prediction and marked improvements in manipulation planning performance.
Chinese Translation
尽管基于动作的视频预测为机器人提供了直观的世界模型，但纯数据驱动的预测器在长时间跨度的预测中常常遭受累积误差和物理上不合理的幻觉，严重削弱了下游的动作规划。我们提出了VPTwin，一个真实-模拟-真实视频预测框架，通过真实同步的模拟双胞胎来锚定现实世界的未来预测。对于目标操作任务，VLM（可执行数字双胞胎）从真实演示片段重建。为了适应未观测物理属性的病态估计，Isaac Sim在候选动作轨迹下对随机物理配置进行多次前向动态预测。使用这些预测作为上下文参考，VPTwin协调两个领域，利用模拟动态来强制执行物理合理性，同时捕捉来自真实视频的未建模接触交互。此外，我们使用VPTwin建立了一个预测规划循环，以视觉验证VLM提出的动作并指导可靠的现实世界执行。评估结果显示，在视频预测过程中物理幻觉显著减少，操作规划性能显著提高。
cs.RO / 61 / 2609.33138
Multi-Modal Non-Prehensile Estimation of Physical Parameters via Press-and-Pull Tipping
通过按压和拉动倾斜的多模态非抓取物理参数估计
Abstract
Recovering physical properties of unknown objects through non-prehensile interaction is challenging because no single manipulation primitive reveals all relevant parameters. Planar pushing couples mass and friction, while conventional tipping cannot recover friction and may fail entirely when low-friction or curved-base objects slide or rotate instead of tipping. We introduce a multi-modal estimation framework that combines a sliding interaction with a press-and-pull tipping primitive to recover object mass, center of mass height, and surface friction. The press-and-pull interaction increases the object-table sliding threshold and stabilizes the pivot, enabling controlled tipping without any prior geometric object model. Wrist force/torque sensing, RGB-D perception, and robot proprioception are fused to estimate the physical parameters from the two complementary interaction modes. Experiments on an ABB IRB120 across four objects with varied geometry, mass, center of mass, and friction achieve low relative error, while successfully operating on curved-base objects that fail under conventional forward tipping. The results demonstrate that complementary non-prehensile interactions can recover a compact set of physical parameters without grasping, a prior object model, or learned interaction dynamics.
Chinese Translation
通过非抓取交互恢复未知物体的物理属性具有挑战性，因为没有单一的操作原语能够揭示所有相关参数。平面推力将质量和摩擦力耦合在一起，而传统的倾斜方法无法恢复摩擦力，并且在低摩擦或曲面基底物体滑动或旋转而不是倾斜时可能完全失效。我们提出了一种多模态估计框架，结合滑动交互与按压和拉动倾斜原语，以恢复物体的质量、质心高度和表面摩擦力。按压和拉动交互提高了物体与桌面之间的滑动阈值，并稳定了支点，使得在没有任何先前几何物体模型的情况下能够进行受控倾斜。通过融合腕部力/扭矩传感、RGB-D 感知和机器人本体感知，从这两种互补的交互模式中估计物理参数。在对四种几何形状、质量、质心和摩擦力各异的物体进行的 ABB IRB120 实验中，实现了较低的相对误差，同时成功操作在传统前倾方法下失效的曲面基底物体。结果表明，互补的非抓取交互能够在没有抓取、先前物体模型或学习的交互动态的情况下恢复一组紧凑的物理参数。
cs.RO / 62 / 2609.33145
Beyond State-as-Action: Exploiting Command-State Discrepancy for Robot Imitation Learning
超越状态作为动作：利用指令-状态差异进行机器人模仿学习
Abstract
Constructing action targets from measured robot motion is an established approach in imitation learning. Under interaction constraints, however, command-state discrepancy may reflect control demands that motion alone does not capture. We investigate when this information matters and how to exploit it. Across three real-robot tasks, task and phase analyses reveal larger supervision gaps under constrained interaction, while selective command retention provides evidence of locally useful command information. Building on these findings, we propose Command-State Discrepancy Weighting (CSDW), which accounts for robot response times and combines subsequent progress, persistent unmet demand, and demand changes into continuous weights for command supervision. The method requires no task-phase annotations or changes to policy architecture or inference. CSDW improves over uniform command supervision on constrained tasks, while methods perform similarly in the less constrained task.
Chinese Translation
从测量的机器人运动中构建动作目标是模仿学习中的一种成熟方法。然而，在交互约束下，指令-状态差异可能反映出运动本身无法捕捉的控制需求。我们研究了这些信息何时重要以及如何利用它。在三个真实机器人任务中，任务和阶段分析揭示了在受限交互下更大的监督差距，而选择性保留指令提供了局部有用指令信息的证据。在这些发现的基础上，我们提出了指令-状态差异加权（Command-State Discrepancy Weighting, CSDW），该方法考虑了机器人响应时间，并将后续进展、持续未满足的需求和需求变化结合为指令监督的连续权重。该方法不需要任务阶段注释或对策略架构或推理的更改。在受限任务中，CSDW相较于均匀指令监督表现更佳，而在不太受限的任务中，各方法表现相似。
cs.RO / 63 / 2609.33157
TimelyDAgger: Timing-Aware Expert Querying for VLA Policy Improvement
TimelyDAgger：基于时序的专家查询以改善VLA策略
Abstract
DAgger improves robot policies by aggregating expert supervision from states visited during policy execution. Robot-gated DAgger automates expert queries, allowing the robot to decide when to request expert takeover. While existing gates emphasize detecting the need for assistance, takeover timing also shapes the content of these demonstrations and their value for policy learning. We propose TimelyDAgger, combining Bridge-PCA monitoring of internal vision-language-action (VLA) features with Feedback-guided Threshold Adaptation based on expert behavior to improve takeover timing. We introduce an evaluation framework linking failure detection, takeover timing, and policy improvement, including Target-Aligned Supervision Ratio (TASR) for assessing supervision quality without retraining. Experiments show that takeover timing affects policy learning, with TimelyDAgger achieving competitive failure detection and higher post-training success in most evaluated settings under matched expert-action budgets.
Chinese Translation
DAgger通过聚合在策略执行过程中访问的状态下的专家监督来改善机器人策略。机器人门控DAgger自动化专家查询，使机器人能够决定何时请求专家接管。虽然现有的门控机制强调检测对帮助的需求，但接管时机也影响这些演示的内容及其对策略学习的价值。我们提出了TimelyDAgger，结合了对内部视觉-语言-动作（VLA）特征的Bridge-PCA监控与基于专家行为的反馈引导阈值适应，以改善接管时机。我们引入了一个评估框架，将失败检测、接管时机和策略改进联系起来，包括目标对齐监督比率（TASR），用于在不重新训练的情况下评估监督质量。实验表明，接管时机影响策略学习，TimelyDAgger在大多数评估设置下，在匹配专家动作预算的情况下，取得了竞争性的失败检测和更高的训练后成功率。
cs.RO / 64 / 2609.33165
Beyond Tasks: A Vision for Reproducing an Animal-like Behavioral Substrate Using Modern Robot Learning Techniques
超越任务：利用现代机器人学习技术重现类动物行为基础的愿景
Abstract
Recent advances in robot learning have produced increasingly capable embodied agents. Yet comparatively less attention has been given to a more basic form of competence that animals exhibit continuously: the ability to remain situated, responsive, and behaviorally coherent as physical, environmental, and social demands change over time. We propose the ethological behavioral substrate as a conceptual lens for studying this form of competence in artificial agents. Rather than treating these behaviors that animals exhibit as a set of isolated skills, we argue that their continual coordination under competing demands constitutes an important and underexplored target for modern robot learning. We further propose robotic animal companions as a useful research setting for studying sustained interaction and adaptation in human-centered environments. Such systems provide an opportunity to investigate how social behavior, memory, and continual learning develop over long periods of interaction. This perspective motivates further investigation of how such persistent behavioral competence may complement higher-level capabilities in embodied agents.
Chinese Translation
近年来，机器人学习的进展使得具身智能体的能力不断增强。然而，相较之下，动物持续展现的一种更基本的能力——在物理、环境和社会需求随时间变化时保持定位、响应和行为一致性——却受到的关注较少。我们提出将伦理学行为基础作为研究人工智能体这一能力的概念视角。我们认为，与其将动物展现的这些行为视为一组孤立的技能，不如将它们在竞争需求下的持续协调视为现代机器人学习的重要且未被充分探索的目标。此外，我们进一步提出将机器人动物伴侣作为研究人本环境中持续互动和适应的有用研究场景。这类系统为研究社会行为、记忆和持续学习在长期互动中的发展提供了机会。这一视角激励我们进一步探讨这种持久的行为能力如何与具身智能体的更高层次能力相辅相成。
cs.RO / 65 / 2609.33172
Dynamic Manipulation with World-Action Models via Counterfactual Planning
通过反事实规划实现动态操控与世界-动作模型
Abstract
World-Action models (WAMs) trained on static demonstrations often fail to manipulate moving targets even when they possess the required manipulation skills. We attribute this failure to target-response collapse: as execution advances, the policy becomes increasingly biased toward the learned continuation of its ongoing behavior and less responsive to target relocation. To bridge the gap between what the model has learned and what it can generate from the current context, we formulate dynamic manipulation as counterfactual planning by decoupling the context used for plan generation from the physical state used for execution. Our framework, Dynamic Predictive Planning (DPP), first uses the WAM's predictive rollout to estimate when an interaction is expected to occur, and combines this timing estimate with observed target motion to predict the target's future interaction position. DPP then constructs a counterfactual observation that places this predicted target position in a familiar robot context, allowing the model to invoke an existing manipulation skill rather than generate a recovery behavior from an unfamiliar robot-target configuration. The resulting plan is connected to the robot's actual state during execution. DPP enables real-time dynamic manipulation on a single consumer GPU without additional training on dynamic data. Experiments in simulation and on a real robot demonstrate consistent improvements across diverse target motions, with simulation performance surpassing all evaluated baselines, including methods additionally trained on dynamic data. Project page: https://methoder00.github.io/DPP/
Chinese Translation
基于静态演示训练的世界-动作模型（WAMs）在操控移动目标时往往表现不佳，即使它们具备所需的操控技能。我们将这种失败归因于目标响应崩溃：随着执行的推进，策略越来越倾向于学习到的持续行为，而对目标位置变化的响应则减少。为了弥合模型所学与其在当前上下文中能够生成之间的差距，我们将动态操控形式化为反事实规划，通过将用于计划生成的上下文与用于执行的物理状态解耦。我们的框架，动态预测规划（Dynamic Predictive Planning, DPP），首先利用WAM的预测回放来估计何时预期会发生交互，并将这一时间估计与观察到的目标运动结合，以预测目标未来的交互位置。DPP随后构建一个反事实观察，将这一预测的目标位置置于熟悉的机器人上下文中，从而使模型能够调用现有的操控技能，而不是从不熟悉的机器人-目标配置中生成恢复行为。最终的计划在执行过程中与机器人的实际状态相连接。DPP使得在单个消费级GPU上实现实时动态操控成为可能，无需对动态数据进行额外训练。模拟和真实机器人上的实验表明，在多种目标运动下均实现了一致的性能提升，模拟性能超越了所有评估的基线，包括那些额外在动态数据上训练的方法。项目页面：https://methoder00.github.io/DPP/
cs.RO / 66 / 2609.33174
Multi-Terrain Mastery: A Comprehensive Controller for Bipedal Locomotion
多地形掌控：一种综合的双足运动控制器
Abstract
Advancing bipedal robots to navigate diverse terrains remains a significant challenge in robotics. Traditional locomotion controllers excel on specific surfaces but struggle across varied environments, limiting their practical applications. Given the unpredictable nature of real-world environments, a single controller capable of handling multiple terrains is ideal, eliminating the need for multiple specialized controllers. We propose a multi-terrain controller to enhance the versatility and robustness of bipedal locomotion. Building on previous work with a stance ankle motor for stability on inclined and rough surfaces, this paper extends capabilities to steep wet uneven grassy slopes, and compliant terrains such as sand, gravel, rocks, and constrained terrains like staircases. To address the unique demands of these terrains, we introduce a new impact map that is essential for maintaining performance and robustness against unseen terrains. We also discuss in detail the control structure for real-time deployment on the robot. We validate our controller on the 20 degree-of-freedom Cassie bipedal robot.
Chinese Translation
推动双足机器人在多样地形中导航仍然是机器人技术中的一个重大挑战。传统的运动控制器在特定表面上表现优异，但在多变环境中则显得力不从心，限制了它们的实际应用。考虑到现实环境的不可预测性，能够处理多种地形的单一控制器是理想选择，这样可以消除对多个专用控制器的需求。我们提出了一种多地形控制器，以增强双足运动的多样性和鲁棒性。在之前的工作基础上，我们使用了一个用于在倾斜和粗糙表面上保持稳定的支撑踝关节电机，本文将其能力扩展到陡峭湿滑的不平整草坡，以及沙子、砾石、岩石等柔性地形和楼梯等受限地形。为了应对这些地形的独特需求，我们引入了一种新的冲击图，这对于维持性能和对未知地形的鲁棒性至关重要。我们还详细讨论了在机器人上实时部署的控制结构。我们在具有20个自由度的Cassie双足机器人上验证了我们的控制器。
cs.RO / 67 / 2609.33177
DeltaWAM: Change-Centric Visual Foresight via Delta Tokens for an Efficient World-Action Model
DeltaWAM：通过增量令牌实现高效世界行动模型的变化中心视觉预见
Abstract
World-Action Models (WAMs) offer visual foresight for robotic manipulation, but pixel-space models repeatedly reconstruct entire future scenes, incurring high computational cost and spatio-temporal redundancy. In physical manipulation, consecutive frames often share most of their visual context; the changes between them are what an action policy needs to anticipate. We introduce DeltaWAM, a change-centric WAM that makes a compact delta token the unit of future prediction. Each token is a single vector encoding changes between consecutive dense DINO feature maps. DeltaWAM builds on DeltaWorld, a latent world model pretrained on large-scale videos, to autoregressively predict one delta token per future frame. A flow-matching action expert then conditions on the predicted transitions and current DINO features, which serve as spatial anchors, to generate action chunks. Trained for 256 GPU hours on two H100 GPUs, DeltaWAM has 0.725B parameters and achieves 92.8% average success on LIBERO. It also shows robust generalization under procedural perturbations on LIBERO-Pro. Inference takes 142.1 ms per action chunk with 3.86 GB peak memory.
Chinese Translation
世界行动模型（WAMs）为机器人操作提供视觉预见，但像素空间模型反复重建整个未来场景，导致高计算成本和时空冗余。在物理操作中，连续帧通常共享大部分视觉上下文；它们之间的变化正是动作策略需要预见的内容。我们提出了DeltaWAM，一种以变化为中心的WAM，将紧凑的增量令牌作为未来预测的单位。每个令牌是一个单一向量，编码连续的密集DINO特征图之间的变化。DeltaWAM基于DeltaWorld，这是一个在大规模视频上预训练的潜在世界模型，能够自回归地预测每个未来帧的一个增量令牌。然后，一个流匹配动作专家根据预测的过渡和当前的DINO特征（作为空间锚点）生成动作块。在两台H100 GPU上训练了256个GPU小时，DeltaWAM具有7.25亿参数，并在LIBERO上实现了92.8%的平均成功率。它在LIBERO-Pro上也显示出在程序扰动下的强健泛化。推理每个动作块耗时142.1毫秒，峰值内存为3.86 GB。
cs.RO / 68 / 2609.33197
TAO-DA: Towards Autonomous Operation--A Dual-Arm Vision-Language-Action Model for Coordinated Manipulation
TAO-DA：迈向自主操作——一种用于协调操作的双臂视觉-语言-动作模型
Abstract
Vision-Language-Action (VLA) models provide a unified framework for grounding high-level semantic information into low-level robot actions, enabling scalable robotic manipulation across diverse tasks. However, existing VLA models lack explicit mechanisms to disentangle the states and intents of the two arms, leading to unintended cross-arm interference that degrades task execution success. To address this issue, we propose a symmetric Dual-Arm Expert (DAE) architecture built upon a shared Vision-Language Model (VLM) backbone with decoupled, arm-specific expert towers. Expert selection is carried out through a two-stage dual-arm intent routing scheme, in which experts are routed either by explicit language instructions in the first stage or by implicit visual semantics in the second stage. Moreover, we introduce a lightweight task progress prediction module that leverages cross-attention between the pre-chunk temporal features and semantic representations of proprioceptive and visual observations to accurately estimate frame-wise task completion progress. This module facilitates task progress synchronization to support coordinated scheduling for collaborative multi-robot tasks. Experimental results demonstrate the effectiveness of our model in dual-arm intent routing and the disentanglement of cross-arm interference, and further provide preliminary evidence of emergent skill generalization from single- to dual-arm tasks (as well as the reverse), together with cross-arm motion-domain skill transfer.
Chinese Translation
视觉-语言-动作（VLA）模型提供了一个统一的框架，将高层次的语义信息与低层次的机器人动作相结合，从而实现多样任务的可扩展机器人操作。然而，现有的VLA模型缺乏明确的机制来解耦双臂的状态和意图，导致意外的跨臂干扰，从而降低任务执行的成功率。为了解决这一问题，我们提出了一种对称的双臂专家（DAE）架构，该架构基于共享的视觉-语言模型（VLM）主干，并配备解耦的、臂特定的专家塔。专家选择通过一个两阶段的双臂意图路由方案进行，在第一阶段通过明确的语言指令路由专家，或在第二阶段通过隐含的视觉语义进行路由。此外，我们引入了一个轻量级的任务进度预测模块，该模块利用预切块时间特征与本体感知和视觉观察的语义表示之间的交叉注意力，准确估计逐帧的任务完成进度。该模块促进了任务进度的同步，以支持协作多机器人任务的协调调度。实验结果表明，我们的模型在双臂意图路由和跨臂干扰解耦方面的有效性，并进一步提供了从单臂到双臂任务（以及反向）的技能泛化的初步证据，以及跨臂运动领域技能的转移。
cs.RO / 69 / 2609.33235
Learning with Object-centric Representations of Tactile Interactive Perception for Robot Manipulation
基于物体中心表示的触觉交互感知学习用于机器人操作
Abstract
Implicit object properties that are difficult to directly infer from vision, such as material, container contents, or softness, can be revealed through tactile sensing and exploratory interactions. However, because tactile signals are transient and sparse, extracting informative tactile events and effectively incorporating them into robotic manipulation remains a challenge. In this work, we present an object-centric context-aware manipulation framework that learns task-agnostic object representations through tactile exploration. A token learner autonomously selects representative tactile segments from long-horizon exploration, while contrastive alignment with descriptive text embeddings enables a latent space that captures multiple physical object properties. These learned representations are then used as semantic context to guide object-centric manipulation policies and adapt strategies based on object properties. Experiments show that the learned representations achieve 93% and 84% property estimation accuracy on seen and unseen objects. Evaluated on three tasks involving visually ambiguous objects, i.e. multi-object rearrangement, pouring, and box opening, the proposed framework improves both target selection and property-dependent manipulation adaptation, raising task success, aggregated over all evaluation trials, from 41% to 92% on seen objects and from 19% to 67% on unseen objects over baseline policies without object-context conditioning. Videos and additional results are available at https://xinyiyxyx.github.io/tactile-object-centric/.
Chinese Translation
一些隐含的物体属性，如材料、容器内容或柔软度，难以通过视觉直接推断，但可以通过触觉感知和探索性交互揭示。然而，由于触觉信号是瞬时和稀疏的，提取有信息量的触觉事件并有效地将其纳入机器人操作仍然是一个挑战。在本研究中，我们提出了一种物体中心的上下文感知操作框架，通过触觉探索学习任务无关的物体表示。一个标记学习器自主选择来自长时间探索的代表性触觉片段，同时与描述性文本嵌入的对比对齐使得潜在空间能够捕捉多个物理物体属性。这些学习到的表示随后被用作语义上下文，以指导物体中心的操作策略，并根据物体属性调整策略。实验表明，学习到的表示在已见和未见物体上的属性估计准确率分别达到93%和84%。在涉及视觉模糊物体的三项任务（即多物体重排、倒水和开箱）中，所提出的框架改善了目标选择和基于属性的操作适应性，将所有评估试验的任务成功率从41%提升至92%（在已见物体上），并将未见物体的成功率从19%提升至67%，相较于没有物体上下文条件的基线策略。视频和更多结果可在 https://xinyiyxyx.github.io/tactile-object-centric/ 获取。
cs.RO / 70 / 2609.33237
SurgFlow: 3D Object-Centric Contact Flow for Surgical Robot Manipulation
SurgFlow：用于外科机器人操作的3D物体中心接触流
Abstract
Paired video-action demonstrations enable autonomous surgical behavior, but such data is scarce: robots perform roughly 1% of surgeries, while video-only data is abundant. Learning 3D object flow offers an embodiment-agnostic way to utilize video data, but flow alone specifies how an object should move, not where and when the tool should engage it, a distinction that is critical in surgery. We introduce SurgFlow, a framework that learns 3D Object-Centric Contact Flow from stereo surgical video without action labels. For each object point, it predicts a future 3D trajectory and contact scores. We extract targets via 3D tracking and tool-object proximity, train a flow matching generator to predict them, and use predicted contact to trigger grasp and release while optimizing end effector motion from flow. On the da Vinci Research Kit (dVRK), SurgFlow succeeds in 37 of 39 stage evaluations across tissue retraction, bimanual reveal, needle pickup, and handover, outperforming baselines trained on equal data with or without action labels. Zero-shot transfer to a humanoid-based laparoscopic robot achieves 85% and 70% average success under similar and novel camera viewpoints, respectively.
Chinese Translation
配对的视频-动作演示能够实现自主外科行为，但此类数据稀缺：机器人仅执行大约1%的手术，而仅有视频数据则相对丰富。学习3D物体流提供了一种与具体实现无关的方式来利用视频数据，但流本身仅指定了物体应如何移动，而未说明工具应何时何地与其接触，这一区别在外科手术中至关重要。我们提出了SurgFlow，一个从立体外科视频中学习3D物体中心接触流的框架，无需动作标签。对于每个物体点，它预测未来的3D轨迹和接触评分。我们通过3D跟踪和工具-物体接近度提取目标，训练一个流匹配生成器来预测它们，并利用预测的接触来触发抓取和释放，同时优化末端执行器的运动。基于达芬奇研究工具包（dVRK），SurgFlow在组织回缩、双手揭示、针头拾取和交接等39个阶段评估中成功完成了37个，超越了在相同数据上训练的基线模型，无论是否有动作标签。在向基于人形的腹腔镜机器人进行零样本迁移时，在相似和新颖的摄像机视角下分别实现了85%和70%的平均成功率。
cs.RO / 71 / 2609.33256
ActionGround: Training-Free Runtime Refinement of Frozen VLA Policies
ActionGround：无训练的冻结 VLA 策略运行时优化
Abstract
Vision-Language-Action (VLA) models map visual observations and language instructions directly to robot actions, but they do not explicitly represent the phase structure of manipulation tasks or the rigid-body dynamics governing execution. We present ActionGround, a neuro-symbolic, training-free runtime layer that wraps a frozen VLA policy without retraining, fine-tuning, or weight access, adding less than 1 ms of overhead per control step. A symbolic phase-aware finite-state machine identifies the manipulation phase (approach, grasp, transport, or place) and applies a phase-specific rule-based correction. In parallel, an always-on, inertia-weighted Euler-Lagrange term incorporates the robot's equations of motion into each control step, while its dynamics residual is logged as a consistency diagnostic rather than used as a gate. We evaluate ActionGround across OpenVLA, OpenVLA-OFT, Force-VLA, and Generalist-VLA on ten LIBERO-Spatial pick-and-place tasks using a 7-DoF Franka Panda. With fixed parameters across tasks and backbones, ActionGround improves success rate by up to 6 percentage points and stability by up to 19.3 percentage points, while improving trajectory efficiency by up to 15%. In a separate Robosuite noise sweep, ActionGround provides approximately a 10x improvement in trajectory-jerk robustness under injected action noise. In a matched-seed Robosuite simulation companion to a real Agilex Piper trial, simulated baseline success increases from 35% to 95%. The physical-hardware experiment is presented as a qualitative deployment demonstration; quantitative per-trial success on the real arm is left for future work. Our evaluation is limited to rigid-object pick-and-place manipulation.
Chinese Translation
视觉-语言-动作（VLA）模型将视觉观察和语言指令直接映射到机器人动作，但它们并未明确表示操作任务的阶段结构或支配执行的刚体动力学。我们提出了 ActionGround，这是一种神经符号的、无训练的运行时层，能够在不进行再训练、微调或权重访问的情况下包装冻结的 VLA 策略，每个控制步骤的额外开销不足 1 毫秒。一个符号的、阶段感知的有限状态机识别操作阶段（接近、抓取、运输或放置），并应用阶段特定的基于规则的修正。同时，一个始终开启的、惯性加权的欧拉-拉格朗日项将机器人的运动方程纳入每个控制步骤，而其动力学残差被记录为一致性诊断，而非作为门控。我们在十个 LIBERO-Spatial 拾取和放置任务上，使用 7 自由度的 Franka Panda 评估 ActionGround，涵盖 OpenVLA、OpenVLA-OFT、Force-VLA 和 Generalist-VLA。通过在任务和基础架构中固定参数，ActionGround 的成功率提高了最多 6 个百分点，稳定性提高了最多 19.3 个百分点，同时轨迹效率提高了最多 15%。在一次独立的 Robosuite 噪声测试中，ActionGround 在注入动作噪声下提供了约 10 倍的轨迹抖动鲁棒性提升。在与真实的 Agilex Piper 试验相匹配的 Robosuite 模拟中，模拟基线成功率从 35% 提升至 95%。物理硬件实验作为定性部署演示呈现；真实机械臂的定量每次试验成功率留待未来工作。我们的评估仅限于刚性物体的拾取和放置操作。
cs.RO / 72 / 2609.33258
PORTER: Edge-Cloud Residency for Persistent 3D Scene Graph Memory
PORTER：用于持久3D场景图记忆的边缘-云驻留
Abstract
Recent task-driven and just-in-time 3D Scene Graph (3DSG) methods reduce per-task representations by constructing or activating only task-relevant information. Yet sparse per-task working sets do not bound onboard memory usage over a robot's lifetime: as tasks change, payloads accumulated for earlier tasks may become irrelevant to the current task but can be useful again in future tasks. Over repeated task switches and expanding environments, retaining such reusable payloads causes local memory to grow, whereas discarding them entirely can lead to costly repeated construction of the same payloads later. We introduce PORTER, which decouples persistence from residency: lightweight anchors remain in the limited memory of the edge robot while heavy object payloads migrate between the edge and the cloud. Relevance alone is insufficient for deciding residency because multiple relevant payloads may provide redundant information. We therefore decompose each task into functional requirements and introduce Irreplaceable Support Erasure (ISE), which measures the loss in requirement coverage caused by offloading. ISE discounts replaceable support and penalizes losses more strongly when the remaining coverage of a requirement is weak. PORTER constructs a budget-aware local working set by repeatedly offloading the payload with the smallest marginal ISE per byte. Experiments on JITOMA-Bench evaluate PORTER across four 3DSG builders. Under progressive compression, pooled relative mR@3 remains at 100% through 91% payload-byte offloading.
Chinese Translation
近期的任务驱动和即时3D场景图（3DSG）方法通过仅构建或激活与任务相关的信息来减少每个任务的表示。然而，稀疏的每任务工作集并不能限制机器人生命周期内的板载内存使用：随着任务的变化，早期任务累积的负载可能对当前任务变得无关紧要，但在未来任务中可能再次有用。在重复的任务切换和不断扩展的环境中，保留这些可重用的负载会导致本地内存增长，而完全丢弃它们则可能导致后续重复构建相同负载的高昂成本。我们提出了PORTER，它将持久性与驻留解耦：轻量级锚点保留在边缘机器人的有限内存中，而重型对象负载在边缘与云之间迁移。仅凭相关性不足以决定驻留，因为多个相关负载可能提供冗余信息。因此，我们将每个任务分解为功能需求，并引入不可替代支持消除（ISE），该方法衡量因卸载而导致的需求覆盖损失。ISE会折扣可替代支持，并在需求的剩余覆盖较弱时对损失给予更强的惩罚。PORTER通过反复卸载每字节边际ISE最小的负载来构建一个预算感知的本地工作集。在JITOMA-Bench上的实验评估了PORTER在四个3DSG构建器中的表现。在逐步压缩下，汇总相对mR@3在91%的负载字节卸载过程中保持在100%。
cs.RO / 73 / 2609.33269
Q-WAM: 4-Bit Quantization of World Action Models with Action-Subspace Protection
Q-WAM：具有动作子空间保护的世界动作模型的4位量化
Abstract
World Action Models (WAMs) jointly generate video and robot actions through iterative diffusion and perform strongly in robotic manipulation. However, their prohibitive compute and memory costs pose substantial deployment challenges. Post-training quantization (PTQ) can reduce these costs, but existing PTQ methods such as smoothing and rotation are insufficient to maintain the precision of action generation. To overcome this limitation, we propose Q-WAM, a new 4-bit weight-activation quantization for WAMs that preserves the actions the model generates. Specifically, we introduce the \textit{Action Observability Gramian (AOG)}, which measures how much rounding errors in each weighted combination of a layer's input channels change the final action through all denoising steps. We also develop Action-Subspace Protection (ASP), which keeps the few most action-sensitive channel combinations in a tiny 16-bit low-rank branch and quantizes the complementary weights and activations to 4 bits, both as dense matrix multiplications that run efficiently on GPUs. Finally, to preserve action quality with minimal overhead, we identify the experts that matter most for the generated action by aggregating the AOG-derived action mass across the layers of each expert and apply ASP only to those experts. We evaluate Q-WAM on three WAMs, both in simulation and in real-world deployment. On the RoboTwin 2.0 benchmark, it reaches 89.6--93.0\% average success rate, within 1.1 percentage points of the 16-bit models, while reducing the memory of the quantized blocks by 3.1--3.4$\times$. Our method outperforms the strongest baseline, SVDQuant, by 2.5--8.7 percentage points. On a Unitree G1 humanoid and a bimanual UR3 robot, it improves success over SVDQuant by 12.8-17.6 percentage points.
Chinese Translation
世界动作模型（WAMs）通过迭代扩散共同生成视频和机器人动作，在机器人操控中表现出色。然而，它们高昂的计算和内存成本带来了显著的部署挑战。后训练量化（PTQ）可以降低这些成本，但现有的PTQ方法，如平滑和旋转，无法有效保持动作生成的精度。为克服这一限制，我们提出了Q-WAM，一种新的4位权重-激活量化方法，旨在保留模型生成的动作。具体而言，我们引入了 extit{动作可观察性Gramian（AOG）}，它测量每个加权组合的输入通道中的舍入误差如何通过所有去噪步骤改变最终动作。我们还开发了动作子空间保护（ASP），该方法将少数对动作敏感的通道组合保留在一个小的16位低秩分支中，并将其余的权重和激活量化为4位，均以密集矩阵乘法的形式高效地在GPU上运行。最后，为了在最小开销下保持动作质量，我们通过聚合AOG导出的动作质量，识别出对生成动作最重要的专家，并仅对这些专家应用ASP。我们在三个WAMs上评估了Q-WAM，包括模拟和实际部署。在RoboTwin 2.0基准测试中，它达到了89.6%到93.0%的平均成功率，距离16位模型仅有1.1个百分点，同时将量化块的内存减少了3.1到3.4倍。我们的方法在性能上超越了最强基线SVDQuant，提升了2.5到8.7个百分点。在Unitree G1人形机器人和双手UR3机器人上，它的成功率较SVDQuant提高了12.8到17.6个百分点。
cs.RO / 74 / 2609.33299
AquaWAM: A Dynamics-aware World Action Model for Underwater Embodied Agents
AquaWAM：一种动态感知的水下具身智能世界动作模型
Abstract
World Action Models (WAMs) are becoming increasingly important and useful for embodied intelligence, as they enable robots to anticipate the consequences of candidate actions before interacting with the physical environment. However, underwater robots are usually subject to passive dynamics, such as inertia, buoyancy, hydrodynamic drag, and persistent drift, which can continue to affect the vehicle even after an action is completed. Existing WAMs, which primarily predict action-conditioned visual observations, are not explicitly designed to capture such passive motion dynamics. In this paper, we present AquaWAM, the first World Action Model designed for underwater embodied agents. Instead of predicting future images, AquaWAM models both action-conditioned and passive physical dynamics, including the thruster dead band, the inertial glide that outlasts each command, and ambient currents. Specifically, it senses through the DVL, IMU, pressure sensor and joint encoders, while cameras supply only semantics for understanding goals and target pose. By modeling compact navigation states rather than high-dimensional visual observations, AquaWAM substantially reduces the model size and computational cost compared with conventional WAMs. Experimentally, AquaWAM achieves a 72.6% task success rate across 20 underwater tasks on the USIM benchmark, outperforming existing methods while making action decisions 2.7x faster than U0 on an NVIDIA Jetson AGX Orin. Our model also remains effective when some onboard sensor measurements are unavailable. For example, without DVL velocity measurements, our method still achieves a 61.6% success rate, compared with 39.4% for U0.
Chinese Translation
世界动作模型（WAMs）在具身智能中变得越来越重要和实用，因为它们使机器人能够在与物理环境互动之前预测候选动作的后果。然而，水下机器人通常受到被动动态的影响，例如惯性、浮力、水动力阻力和持续漂移，这些因素在动作完成后仍会影响车辆。现有的WAMs主要预测基于动作的视觉观测，并未明确设计用于捕捉这种被动运动动态。在本文中，我们提出了AquaWAM，这是第一个为水下具身智能体设计的世界动作模型。AquaWAM不仅预测未来图像，还建模基于动作的物理动态和被动物理动态，包括推进器死区、超出每个指令的惯性滑行和环境水流。具体而言，它通过多波束声纳（DVL）、惯性测量单元（IMU）、压力传感器和关节编码器进行感知，而相机仅提供理解目标和目标姿态的语义信息。通过建模紧凑的导航状态而非高维视觉观测，AquaWAM显著减少了模型大小和计算成本，与传统的WAMs相比。在实验中，AquaWAM在USIM基准测试中在20个水下任务上实现了72.6%的任务成功率，超越了现有方法，并且在NVIDIA Jetson AGX Orin上做出动作决策的速度比U0快2.7倍。我们的模型在某些机载传感器测量不可用时仍然有效。例如，在没有DVL速度测量的情况下，我们的方法仍然实现了61.6%的成功率，而U0的成功率为39.4%。
cs.RO / 75 / 2609.33310
CompliantWBC: Whole-Body Compliance for Heavy Humanoids via Force Latent Estimation and Residual Impedance Targets
CompliantWBC：通过力潜在估计和残余阻抗目标实现重型人形机器人全身合规控制
Abstract
Whole-body compliant control is essential for deploying heavy humanoids under high payload in human-centric environments. Most prior force-aware learning-based pipelines focus on end-effector resistance, per-link upper-body springs, or end-effector stiffness modulation, leaving arbitrary-site perturbations on heavy platforms with lower-body engagement largely unaddressed. We close this gap with CompliantWBC comprising: (1) A base policy trained with RL to maximize compliance-fidelity reward, guided by a multi-site whole-body impedance reference controller, extending classical Cartesian impedance to any controlled link; (2) A bounded residual policy that edits the per-link impedance equilibrium over a frozen base, correcting the coarse but structured wrench estimate supplied by a force encoder co-trained behind a gradient barrier; (3) A Phong-weighted force-origin sampler with an axis-decoupled pelvis anchor induces lower-body-inclusive compliance curriculum training via two interpretable parameters. We evaluate CompliantWBC in simulation against both compliant and stiff baselines, achieving best compliant fidelity of 2.58cm deviation from analytical solutions, and demonstrate it on a real heavy humanoid across static/dynamic force reaction, board wiping, squat under payload, and cooperative payload transport. Project website: https://dotandung.github.io/compliantwbc/
Chinese Translation
全身合规控制对于在以人为中心的环境中部署重型人形机器人以承载高负载至关重要。大多数先前的基于力感知的学习管道集中于末端执行器的阻抗、每个关节的上半身弹簧或末端执行器的刚度调节，导致在重型平台上与下半身互动的任意位置扰动问题未得到充分解决。我们通过CompliantWBC填补了这一空白，具体包括：(1) 一个基于强化学习（RL）训练的基础策略，以最大化合规性奖励，受多点全身阻抗参考控制器的指导，将经典的笛卡尔阻抗扩展到任何受控关节；(2) 一个有界的残余策略，在冻结的基础上编辑每个关节的阻抗平衡，修正由在梯度屏障后共同训练的力编码器提供的粗略但结构化的扭矩估计；(3) 一个Phong加权的力源采样器，结合轴解耦的骨盆锚点，通过两个可解释的参数引导下半身合规性课程训练。我们在模拟中评估了CompliantWBC，相较于合规和刚性基线，达到了与解析解相差2.58厘米的最佳合规性，并在真实重型人形机器人上展示了其在静态/动态力反应、擦板、负载下深蹲和协作负载运输等任务中的表现。项目网站：https://dotandung.github.io/compliantwbc/
cs.RO / 76 / 2609.33311
SocialHumanoid: Towards Expressive Humanoid Behavior via One-Step Co-Speech Motion Generation
SocialHumanoid：通过一步共语运动生成实现富有表现力的人形行为
Abstract
Humanoid robots are increasingly expected to serve as embodied social agents that communicate naturally with humans through face-to-face interaction. During such communication, humanoid robots require body behaviors that are synchronized with speech, affectively expressive, and suitable for real-time execution. However, existing co-speech methods are primarily developed for digital humans and lack joint support for affective control and low-latency continuous generation on physical embodiments. To bridge this gap, we present SocialHumanoid, a system for expressive humanoid behavior via one-step co-speech motion generation. Given response speech and a specified affective condition, SocialHumanoid generates each full-body motion window in a single forward pass and connects successive windows through motion-history conditioning. The generated human motion is further converted online into embodiment-compatible robot references and tracked by a whole-body controller for physical execution. To provide explicit supervision for affective body expression, we further introduce AffectMoCap, a 4-hour dataset captured from two professional actors, containing synchronized speech, body motion, fine-grained hand motion, and emotion annotations. On BEAT2, SocialHumanoid achieves the best FGD among the compared generation methods, competitive speech-motion synchrony, and approximately $6\times$ faster inference than GestureLSM under the same protocol. Perceptual evaluations further show that training with AffectMoCap improves affect recognition from generated body motion, while real-robot experiments demonstrate continuous affect-conditioned behavior and stable long-horizon execution. Our project page is https://rex0191.github.io/SocialHumanoid/.
Chinese Translation
人形机器人越来越被期望作为具身的社交代理，通过面对面的互动与人类自然交流。在这样的交流中，人形机器人需要与语言同步的身体行为，这些行为应具有情感表达能力，并适合实时执行。然而，现有的共语方法主要是为数字人类开发的，缺乏对情感控制和物理实体上低延迟连续生成的联合支持。为了解决这一问题，我们提出了SocialHumanoid，一个通过一步共语运动生成实现富有表现力的人形行为的系统。在给定响应语音和指定情感条件的情况下，SocialHumanoid在一次前向传递中生成每个全身运动窗口，并通过运动历史条件连接连续的窗口。生成的人类运动进一步在线转换为适合具身化的机器人参考，并由全身控制器跟踪以实现物理执行。为了提供情感身体表达的明确监督，我们进一步引入了AffectMoCap，这是一个由两位专业演员捕获的4小时数据集，包含同步的语音、身体运动、细粒度手部运动和情感注释。在BEAT2上，SocialHumanoid在比较的生成方法中实现了最佳的FGD，具有竞争力的语音-运动同步性，并且在相同协议下的推理速度约为GestureLSM的6倍。感知评估进一步表明，使用AffectMoCap进行训练提高了从生成的身体运动中识别情感的能力，而真实机器人实验则展示了连续的情感条件行为和稳定的长时间执行。我们的项目页面是 https://rex0191.github.io/SocialHumanoid/.
cs.RO / 77 / 2609.33354
Traceable Human-to-Humanoid Sign Language Benchmarking
可追溯的人类与类人机器人手语基准测试
Abstract
Sign data collection is costly, and teleoperation scales poorly, motivating reuse of large video corpora. Humanoid signing requires converting video-derived human motion into robot trajectories while preserving linguistic motion cues. Errors from fitting, human-motion repair, retargeting, robot geometry repair, and control are hard to separate from the final trajectory alone. We introduce HumanoidCSL-20K, a dataset and benchmark of 20,648 sentence-level Chinese Sign Language sequences, each with four aligned versions: the source, the repaired human motion, the direct robot reference, and the geometry-repaired robot reference. Observation-supported local human-motion repair, full-robot geometry repair, and cross-representation provenance make each transformation traceable. Paired evaluations measure human-motion continuity and content preservation, robot-reference feasibility, and physical execution. A sign-specific kinematic-reference protocol scores handshape, location, palm orientation, and inter-hand relation over the full planned motion. Full-corpus results show fewer abnormal arm / hand steps and less inter-hand and hand-body penetration after repair. Control experiments separate reference learnability from curriculum effects, while component scores expose remaining execution errors.
Chinese Translation
手语数据收集成本高昂，远程操作的扩展性较差，这促使我们重用大型视频语料库。类人机器人手语需要将视频中获取的人类动作转换为机器人轨迹，同时保留语言运动线索。由于拟合、人体运动修复、重定向、机器人几何修复和控制等过程中的错误难以仅通过最终轨迹来分离，我们引入了HumanoidCSL-20K，这是一个包含20,648个句子级中文手语序列的数据集和基准测试，每个序列都有四个对齐版本：源序列、修复的人类动作、直接的机器人参考和几何修复的机器人参考。观察支持的局部人类运动修复、全机器人几何修复和跨表示来源使每个转换过程可追溯。配对评估测量人类运动的连续性和内容保留、机器人参考的可行性以及物理执行的效果。特定手语的运动学参考协议对手形、位置、掌心方向和双手关系在整个计划运动中进行评分。完整语料库的结果显示，修复后异常的手臂/手步态减少，双手和手与身体之间的穿透现象减轻。对照实验将参考学习能力与课程效应分开，而各组件的评分则揭示了剩余的执行错误。
cs.RO / 78 / 2609.33378
Recursive Harness Distillation across Agents for Robot Manipulation
跨代理的递归工具蒸馏用于机器人操控
Abstract
A central goal in robotics is to enable manipulation across changing tasks and environments. Vision-language-action (VLA) models provide broad manipulation capabilities but can struggle when execution requires diagnosing failures and adapting behavior. Strong agents can discover effective interventions through interaction with these policies. We propose Recursive Harness Distillation to accumulate this experience as reusable guidance across agents. A strong agent distills its experience into a playbook for a light agent, then recursively refines the playbook using the light agent's execution feedback. The resulting playbook enables agents to reuse accumulated intervention knowledge in new task instances without updating model parameters. In real-world manipulation, the harness improves success from 37.3% to 64.0%. On SimplerEnv Bridge, the light agent with the playbook achieves 66.7% success, compared with 41.7% for the GR00T-only baseline, and outperforms the strong agent without a playbook. The same playbook also benefits the strong agent, which reaches 79.2% success. These results demonstrate the feasibility of harness distillation for robotics: intervention experience can be accumulated, refined through execution, and reused across agents to improve manipulation.
Chinese Translation
机器人技术的一个核心目标是实现对不断变化的任务和环境的操控。视觉-语言-动作（VLA）模型提供了广泛的操控能力，但在执行过程中需要诊断失败和调整行为时可能会遇到困难。强大的代理可以通过与这些策略的互动发现有效的干预措施。我们提出了递归工具蒸馏（Recursive Harness Distillation），以将这种经验积累为跨代理的可重用指导。强代理将其经验蒸馏为轻代理的操作手册，然后利用轻代理的执行反馈递归地完善该操作手册。最终生成的操作手册使代理能够在新的任务实例中重用积累的干预知识，而无需更新模型参数。在实际操控中，该工具将成功率从37.3%提升至64.0%。在SimplerEnv Bridge上，使用该操作手册的轻代理成功率达到66.7%，而仅使用GR00T的基线成功率为41.7%，并且在没有操作手册的情况下，强代理的表现也不如使用操作手册的轻代理。相同的操作手册也使强代理受益，成功率达到79.2%。这些结果证明了工具蒸馏在机器人技术中的可行性：干预经验可以被积累、通过执行进行精炼，并在代理之间重用，以提高操控能力。
cs.RO / 79 / 2609.33423
Large Language Models for Model-Based Robot Design
基于模型的机器人设计中的大型语言模型
Abstract
Large Language Models (LLMs) can contribute useful engineering knowledge to robot design, but directly generated designs may rely on implicit assumptions and provide no guarantees of feasibility or optimality. These assumptions are critical because different reasonable modeling choices can materially change which designs are predicted to be feasible or optimal. We therefore present a framework that uses LLMs to construct explicit engineering models containing physical relationships, compatibility constraints, and objectives, allowing these modeling choices to be inspected and revised before formal optimization. The model can then be updated with additional engineering, manufacturer, or system-specific information before formal multi-objective optimization provides feasibility and Pareto-optimality guarantees with respect to the finalized model and specified design space. We evaluate the framework on quadcopter and line-following robot component-selection problems. Across 30 direct LLM design trials, none could be verified as feasible under the corresponding finalized model. Comparisons with an independently developed expert model and successive stages of model refinement further showed that changes in modeling assumptions substantially altered the predicted feasible and Pareto-optimal design sets. Together, these results show that using LLMs to construct explicit engineering models makes the underlying design choices available for inspection and revision before those assumptions determine the optimized designs. Explicit modeling therefore provides an interface for combining LLM-generated engineering knowledge, system-specific information, and formal design optimization.
Chinese Translation
大型语言模型（LLMs）可以为机器人设计提供有用的工程知识，但直接生成的设计可能依赖于隐含假设，并且无法保证其可行性或最优性。这些假设至关重要，因为不同的合理建模选择可能会实质性地改变预测的可行或最优设计。因此，我们提出了一个框架，利用LLMs构建包含物理关系、兼容性约束和目标的明确工程模型，使这些建模选择在正式优化之前能够被检查和修订。然后，可以在正式的多目标优化之前，使用额外的工程、制造商或系统特定信息更新模型，从而提供关于最终模型和指定设计空间的可行性和帕累托最优性保证。我们在四旋翼和循迹机器人组件选择问题上评估了该框架。在30次直接的LLM设计试验中，没有一个能够在相应的最终模型下被验证为可行。与独立开发的专家模型及模型细化的各个阶段的比较进一步表明，建模假设的变化实质性地改变了预测的可行和帕累托最优设计集。这些结果表明，使用LLMs构建明确的工程模型使得潜在的设计选择在假设决定优化设计之前可以被检查和修订。因此，明确建模提供了一个接口，用于结合LLM生成的工程知识、系统特定信息和正式设计优化。
cs.RO / 80 / 2609.33464
VIDEAS: Distilling Explicit Action Semantics from Demonstration Videos for World Models via Prior-Guided Simulation
VIDEAS：通过先验引导模拟从演示视频中提取显式动作语义以构建世界模型
Abstract
World models learn internal representations of environment dynamics to predict future states, enabling agents to optimize action plans without physical interactions. However, developing world models that genuinely internalize underlying causal physical laws to explicitly reason about action preconditions and subsequent state transitions remains an open challenge. In this paper, we propose VIDEAS, a data distillation framework that transforms continuous physical dynamics from operational videos into explicit action semantics for foundation models. Specifically, it deconstructs visual demonstrations into discrete action trajectories and utilizes advanced vision-language models (VLMs) to extract structured knowledge encapsulating action preconditions and effects. To ensure physical consistency, we introduce a prior-guided trajectory simulation mechanism grounded within a text-based environment to rigorously validate the extracted knowledge. Notably, we incorporate negative trajectories to enrich knowledge completeness and enhance data diversity to mitigate cognitive bias. Furthermore, we present VIDEAS-WM, an 8B/9B-parameter suite of language-based world models trained on 34K high-quality samples derived from AgiBot-World dataset. Extensive experiments demonstrate that VIDEAS-WM establishes state-of-the-art performance in high-level embodied action semantic reasoning, exhibiting profound physical understanding and robust generalization across unseen scenarios.
Chinese Translation
世界模型学习环境动态的内部表征，以预测未来状态，使得智能体能够在没有物理交互的情况下优化行动计划。然而，开发能够真正内化潜在因果物理法则的世界模型，以显式推理动作前提和随后的状态转变，仍然是一个未解决的挑战。本文提出了VIDEAS，一种数据蒸馏框架，将操作视频中的连续物理动态转化为基础模型的显式动作语义。具体而言，它将视觉演示解构为离散的动作轨迹，并利用先进的视觉-语言模型（VLMs）提取封装动作前提和效果的结构化知识。为了确保物理一致性，我们引入了一种基于文本环境的先验引导轨迹模拟机制，以严格验证提取的知识。值得注意的是，我们引入了负轨迹，以丰富知识的完整性并增强数据的多样性，从而减轻认知偏差。此外，我们还提出了VIDEAS-WM，一个基于语言的世界模型套件，包含8B/9B参数，训练于来自AgiBot-World数据集的34K高质量样本。大量实验表明，VIDEAS-WM在高层次的具身动作语义推理中建立了最先进的性能，展现出深刻的物理理解能力和在未见场景中的强健泛化能力。
cs.RO / 81 / 2609.33484
AMBIT: Anticipatory Multimodal Body Recruitment for Bimanual Tracking on a Humanoid
AMBIT：用于类人机器人双手追踪的预期多模态身体招募
Abstract
A humanoid with 5-DoF arms cannot track generic bimanual end-effector trajectories with its arms alone; pelvis and waist motion must be recruited, but which motion, and when, is not uniquely determined. On a Unitree R1 in fixed double support, the set of dynamically valid recruitment strategies (pelvis pose and waist trajectories) for a task is a diverse continuous manifold, and a deterministic regressor trained on it mode-averages into strategies valid only 35% of the time, against 52% for a conditional variational autoencoder (CVAE) and 82% for the best of 16 CVAE samples. We introduce AMBIT: the CVAE proposes strategies from a preview of the commanded trajectory, a non-learned selector filters, ranks and verifies them, and a receding-horizon loop commits to one with hysteresis. The committed strategy is the reference of the same whole-body differential-IK QP a reactive tracker runs, which keeps authority over residual error. On 160 held-out episodes that admit a valid strategy, in full MuJoCo dynamics under a torque controller, AMBIT reaches 85% success at a 3 cm/15 deg tolerance against 74% for the tracker (disjoint confidence intervals) and recruits the body before the arms saturate in 48% of episodes against 35%. Because diversity is preserved, constraints unknown at training time are enforced by selection alone: under five zero-shot shifts AMBIT beats the warm-started tracker on every shift and matches a test-time re-optimisation baseline 17x more expensive. On a Unitree G1, with hyperparameters unchanged, the protocol reproduces the structure of the valid set and widens the gap over the tracker to 0.85 against 0.53. Five selected strategies execute on the externally supported physical R1, distinct in pelvis excursion and tracking the planned end-effector motion to a median of 11 mm by encoder forward kinematics, which establishes kinematic realisability, not balance.
Chinese Translation
一具具有5自由度手臂的类人机器人无法仅依靠手臂追踪通用的双手末端执行器轨迹；必须招募骨盆和腰部运动，但具体招募哪种运动以及何时招募并没有唯一的确定性。在固定双支撑状态下的Unitree R1上，针对某一任务的动态有效招募策略（骨盆姿态和腰部轨迹）的集合是一个多样的连续流形，而在其上训练的确定性回归器仅在35%的时间内平均出有效策略，相比之下，条件变分自编码器（CVAE）的有效率为52%，而16个CVAE样本中最好的效果则达到了82%。我们提出了AMBIT：CVAE根据指令轨迹的预览提出策略，一个非学习的选择器对其进行过滤、排序和验证，并通过一个递归视野循环承诺一个带滞后的策略。所承诺的策略是同一全身微分逆运动学（IK）QP的参考，反应式追踪器在此基础上运行，保持对残余误差的控制。在160个允许有效策略的保留回合中，在扭矩控制器下的完整MuJoCo动态中，AMBIT在3 cm/15度的容差下达到了85%的成功率，而追踪器的成功率为74%（置信区间不重叠），并且在48%的回合中在手臂饱和之前招募身体，而追踪器为35%。由于保留了多样性，训练时未知的约束仅通过选择得以执行：在五次零样本转移中，AMBIT在每次转移中都超越了预热启动的追踪器，并且与测试时重新优化的基准相匹配，后者的成本高出17倍。在Unitree G1上，超参数不变，该协议再现了有效集合的结构，并将追踪器的差距扩大至0.85，而追踪器为0.53。五个选定策略在外部支持的物理R1上执行，在骨盆运动和追踪计划末端执行器运动方面表现出明显的差异，通过编码器前向运动学追踪到中位数为11毫米，建立了运动学可实现性，而非平衡性。
cs.RO / 82 / 2609.33522
AI-Driven Collaborative Assembly Line Inspection: System Integration and Deployment Challenges
基于人工智能的协作装配线检测：系统集成与部署挑战
Abstract
Manual visual inspection on assembly lines is a persistent manufacturing bottleneck: operator fatigue over extended shifts lowers defect-detection rates. This paper presents the design, integration, and field deployment of an AI-assisted collaborative inspection cell at the Silverline kitchen-appliance factory, developed within the AI-PRISM project. The cell couples a Universal Robots UR 10e cobot carrying a machine-vision defect-detection pipeline with a Comau Racer-5 cobot for functional tests, coordinated through ROS 2 Humble on an Ubuntu 22.04 LTS server. Multi-modal data (Basler camera imagery, TIA microphone acoustics, and SPS electrical-safety measurements) are logged locally and visualised in real time with Grafana. We report the practical deployment challenges (close-proximity safety, AI robustness under glare and reflections, ROS 2 namespace collisions across two cobots, and operating-system and dependency issues) together with the engineering solutions adopted, and structure the integration through a four-level Human-Robot Interaction analysis. The deployed cell cuts per-unit quality-check time from 82 s to 61 s (about 25%), raises final-control resource efficiency from 0.75 to 0.88, reduces operator visual-inspection viewing time by 82%, and significantly lowers operator mental demand (p = 0.005, NASA-TLX).
Chinese Translation
手动视觉检测在装配线中是一个持续的制造瓶颈：长时间的操作员疲劳降低了缺陷检测率。本文介绍了在AI-PRISM项目中开发的Silverline厨房电器工厂的AI辅助协作检测单元的设计、集成和现场部署。该单元结合了一台携带机器视觉缺陷检测管道的Universal Robots UR 10e 协作机器人与一台用于功能测试的Comau Racer-5 协作机器人，通过在Ubuntu 22.04 LTS服务器上的ROS 2 Humble进行协调。多模态数据（Basler相机图像、TIA麦克风声学和SPS电气安全测量）被本地记录并通过Grafana实时可视化。我们报告了实际部署中的挑战（近距离安全、在眩光和反射下的AI鲁棒性、两个协作机器人之间的ROS 2命名空间冲突，以及操作系统和依赖性问题），并介绍了所采用的工程解决方案，同时通过四级人机交互分析结构化集成。部署的单元将每单位质量检查时间从82秒缩短至61秒（约25%），将最终控制资源效率从0.75提高至0.88，减少操作员视觉检测的观看时间82%，并显著降低操作员的心理负担（p = 0.005，NASA-TLX）。
cs.RO / 83 / 2609.33542
A Disk-Shaped Magnetoelastic Torque Sensor for Robotic Joints Using Permanent Magnetization
一种用于机器人关节的盘形磁弹性扭矩传感器，采用永久磁化技术
Abstract
Direct torque sensing is a growing need in the robotics community to enable precise control and interactions where torque estimation from motor current is not sufficient. This paper presents a novel disk-shaped magnetoelastic torque sensor with a compact axial envelope of about 1 cm, suitable for integration in robotic joints. A four-magnetometer architecture is used to measure the field modulated by the stress affecting a narrow magnetized region while rejecting the effects of parasitic cross forces. A custom-designed magnetic shield enhances the torque sensitivity while reducing the external stray fields by a factor 7x. The device measures the torque with an accuracy of 1.34 %FS relative to the 50 Nm full scale (FS). The paper details the development of the sensor through the mechanical design, the magnetization procedure, and the experimental validation. The results demonstrate the potential of the proposed sensor for robotic applications.
Chinese Translation
直接扭矩传感的需求在机器人领域日益增长，以实现精确控制和交互，而仅依靠电机电流估算扭矩已不足够。本文提出了一种新型的盘形磁弹性扭矩传感器，其紧凑的轴向外形约为1厘米，适合集成于机器人关节中。该传感器采用四个磁力计架构，以测量受应力影响的狭窄磁化区域所调制的磁场，同时抑制寄生交叉力的影响。定制设计的磁屏蔽增强了扭矩灵敏度，同时将外部杂散场降低了7倍。该设备在50 Nm满量程（FS）下，扭矩测量精度达到1.34 %FS。本文详细介绍了传感器的开发过程，包括机械设计、磁化过程和实验验证。结果表明，所提出的传感器在机器人应用中的潜力。
cs.RO / 84 / 2609.33546
Steer2Grasp: Inference-Time Embodiment-Aware Steering for Diverse Physically Feasible Grasp Diffusion
Steer2Grasp：针对多样化物理可行抓取扩散的推理时刻体现感知引导
Abstract
Current grasp diffusion models provide rich priors for generation, yet their object-centric approach can violate the kinematic and collision constraints imposed by the embodiment and the environment. Existing embodiment-aware methods primarily perform local corrections around generated grasps through gradient guidance or optimization, making it difficult to recover from fundamentally infeasible modes. We present Steer2Grasp, a training-free, embodiment-agnostic framework for inference-time grasp steering that adapts a frozen Cartesian grasp diffusion model using deployment-specific rewards. Through Feynman-Kac (FK) inspired particle reweighting and resampling, the method reallocates population mass from infeasible to high-reward grasp modes, enabling population-level mode transitions without modifying the pretrained diffusion model or requiring differentiable constraints. The framework enables a unified treatment for single and dual arm grasping through reachability and collision aware rewards, followed by gradient free gripper level local refinement. Across diverse objects, robot embodiments, and constrained environments, our method substantially improves feasible grasp generation while maintaining proximity to the underlying grasp prior.
Chinese Translation
当前的抓取扩散模型为生成提供了丰富的先验信息，但其以物体为中心的方法可能会违反由体现和环境施加的运动学和碰撞约束。现有的体现感知方法主要通过梯度引导或优化在生成的抓取周围进行局部修正，这使得从根本上不可行的模式中恢复变得困难。我们提出了Steer2Grasp，这是一个无训练、与体现无关的推理时刻抓取引导框架，通过使用特定于部署的奖励来调整一个冻结的笛卡尔抓取扩散模型。通过受到费曼-卡克（Feynman-Kac）启发的粒子重加权和重采样，该方法将种群质量从不可行的抓取模式重新分配到高奖励的抓取模式，从而实现种群级模式转换，而无需修改预训练的扩散模型或要求可微分约束。该框架通过可达性和碰撞感知奖励为单臂和双臂抓取提供了统一的处理，随后进行无梯度的夹持器级局部优化。在多样化的物体、机器人体现和受限环境中，我们的方法显著提高了可行抓取的生成，同时保持了与基础抓取先验的接近性。
cs.RO / 85 / 2609.33551
FoLD: Force-Informed Learning for Dexterous Articulated Object Manipulation
FoLD：基于力的信息学习用于灵巧关节物体操控
Abstract
Transferring human demonstrations to dexterous robots remains challenging because differences in hand morphology and contact dynamics often cause retargeted motions to fail at producing the intended object behavior. We present \textbf{FoLD}, a framework for learning dexterous manipulation of articulated objects through explicit force guidance. FoLD compute compensatory force fields from human demonstrations together with the robot's current interaction state, yielding a force prior that promotes the demonstrated object motion. This force prior informs a residual policy that adapts retargeted hand motions to the contact requirements of the task. We evaluate FoLD on a public benchmark for articulated object manipulation, where it consistently outperforms state-of-the-art baselines across tasks and embodiments. We further validate FoLD on real dexterous robot platforms, demonstrating successful transfer of human manipulation skills to robot execution. Here is the link of our project page: https://gghgghgghgg.github.io/FoLD-project-page/.
Chinese Translation
将人类示范转移到灵巧机器人上仍然具有挑战性，因为手部形态和接触动态的差异往往导致重新定向的动作无法产生预期的物体行为。我们提出了 extbf{FoLD}，一个通过明确的力指导学习关节物体灵巧操控的框架。FoLD从人类示范和机器人的当前交互状态中计算补偿力场，从而产生促进示范物体运动的力先验。这一力先验为残差策略提供信息，使其能够将重新定向的手部动作适应任务的接触要求。我们在一个公共的关节物体操控基准上评估了FoLD，结果显示其在各项任务和实现上始终优于最先进的基线。此外，我们还在真实的灵巧机器人平台上验证了FoLD，成功展示了人类操控技能向机器人执行的转移。我们的项目页面链接如下：https://gghgghgghgg.github.io/FoLD-project-page/.
cs.RO / 86 / 2609.33575
SLIP-VLA: Single-Step Latent Imagination for Policy Learning in Vision-Language-Action Models
SLIP-VLA：用于视觉-语言-动作模型的单步潜在想象政策学习
Abstract
Vision-Language-Action models are increasingly effective for robotic manipulation, yet most predict actions directly from current observations without explicitly modeling future scene evolution. Recent methods introduce future prediction to improve action generation, but dense future modeling often requires expensive iterative denoising, while one-step alternatives can underperform their multi-step counterparts. To reconcile efficient future modeling with strong action performance, we present SLIP-VLA, a policy learning framework that equips VLA models with a Single-Step Latent Imagination for future-aware action prediction. SLIP-VLA obtains temporally dense future latent representations with a single denoising update, and we improve the perceptual sufficiency of these representations by aligning intermediate latents with future geometric and semantic features. We further improve their control sufficiency through action-conditioned latent world modeling and inverse dynamics modeling, explicitly coupling latent transitions with robot actions. SLIP-VLA achieves state-of-the-art performance across diverse simulation benchmarks and real-world manipulation tasks, while its single-step latent imagination takes only 12 ms.
Chinese Translation
视觉-语言-动作模型在机器人操作中越来越有效，但大多数模型直接从当前观察中预测动作，而没有明确建模未来场景的演变。最近的方法引入了未来预测以改善动作生成，但密集的未来建模通常需要昂贵的迭代去噪，而一步替代方案的性能往往不及多步方案。为了调和高效的未来建模与强大的动作性能，我们提出了SLIP-VLA，这是一种政策学习框架，为VLA模型配备了单步潜在想象，以实现未来感知的动作预测。SLIP-VLA通过单次去噪更新获得时间上密集的未来潜在表示，并通过将中间潜在表示与未来几何和语义特征对齐来提高这些表示的感知充分性。我们进一步通过动作条件的潜在世界建模和逆向动力学建模来提高其控制充分性，明确将潜在转变与机器人动作耦合。SLIP-VLA在各种仿真基准和现实世界操作任务中实现了最先进的性能，而其单步潜在想象仅需12毫秒。
cs.RO / 87 / 2609.33595
Beyond One-Step Accuracy: State-Affine Latent Transition for Reliable Visual Planning
超越一步预测准确性：状态仿射潜在转移用于可靠的视觉规划
Abstract
Joint-embedding world models enable visual planning by learning action-conditioned dynamics in latent space. Yet they are commonly trained for one-step prediction on encoded states, while planning recursively applies the learned transition to its own predictions. One-step accuracy therefore does not capture how prediction errors propagate under recursive rollout. We decompose multi-step rollout error into the errors introduced at individual steps and their propagation through subsequent transitions. We show that state-affine dynamics are precisely the differentiable transitions with state-independent Jacobians, eliminating the nonlinear propagation residual and making the error propagation operators depend only on the action sequence. Guided by this result, we introduce SALT (State-Affine Latent Transition), an action-conditioned state-affine dynamics model in which the action modulates both the state transformation and the additive update. We train SALT through recursive multi-step rollout supervision, feeding each predicted latent state back into the transition so that training matches how the model is used during planning. Across four visual planning environments, SALT exhibits $1.48$--$2.19\times$ higher one-step prediction error than the matched LeWM baseline, yet improves closed-loop success in every environment by $10.0$ percentage points on average. On OGBench-Cube, the fraction of episodes that fail with a sharp rise in model-predicted cost after execution decreases from $23.3%$ to $2.0%$.
Chinese Translation
联合嵌入世界模型通过在潜在空间中学习基于动作的动态来实现视觉规划。然而，它们通常是在编码状态上进行一步预测的训练，而规划则递归地将学习到的转移应用于自身的预测。因此，一步准确性并不能捕捉到预测误差在递归展开下的传播。我们将多步展开误差分解为在各个步骤引入的误差及其通过后续转移的传播。我们表明，状态仿射动态正是具有状态无关雅可比矩阵的可微转移，消除了非线性传播残差，使得误差传播算子仅依赖于动作序列。在这一结果的指导下，我们引入了SALT（状态仿射潜在转移），这是一种基于动作的状态仿射动态模型，其中动作调节状态变换和附加更新。我们通过递归多步展开监督训练SALT，将每个预测的潜在状态反馈到转移中，以使训练与模型在规划中的使用方式相匹配。在四个视觉规划环境中，SALT的单步预测误差比匹配的LeWM基线高出1.48到2.19倍，但在每个环境中平均提高了10.0个百分点的闭环成功率。在OGBench-Cube上，执行后模型预测成本急剧上升而导致失败的剧集比例从23.3%降至2.0%。
cs.RO / 88 / 2609.33647
InfraVLA: Extending Vision-Language-Action Navigation with Infrastructure Cameras
InfraVLA：利用基础设施摄像头扩展视觉-语言-动作导航
Abstract
Many indoor environments in which robots operate, such as warehouses, offices, and hospitals, already have cameras installed. They observe parts of the building that the robot cannot see from where it stands, yet navigation policies, including recent vision-language-action (VLA) models, do not use them. We propose InfraVLA, an end-to-end method that adapts a pretrained navigation VLA to such static infrastructure views: a closed-circuit television (CCTV) encoder turns each external view into tokens of the input sequence. Because the views matter only at rare decision points, fine-tuning alone did not make the policy use them in our experiments; we therefore train in two stages, on demonstrations with upsampled counterfactual data and then on recovery data. We evaluate on two simulated warehouse tasks, finding an object named in the instruction and rerouting around blocked aisles, where the deciding information is often visible only to the infrastructure cameras. Tested in distribution, InfraVLA reached a success rate of 100% on both, against 34.0% and 73.6% for a baseline without CCTV input. On out-of-distribution test sets it reached 88.2% and 88.9%. On a real quadruped fine-tuned with under 10 minutes of demonstrations, the policy reached 83.3% against 29.2% for the on-board-only baseline.
Chinese Translation
许多机器人操作的室内环境，如仓库、办公室和医院，已经安装了摄像头。这些摄像头观察到机器人在其当前位置无法看到的建筑部分，但导航策略，包括最近的视觉-语言-动作（VLA）模型，并未利用这些信息。我们提出了InfraVLA，一种端到端的方法，将预训练的导航VLA适配于这些静态基础设施视图：闭路电视（CCTV）编码器将每个外部视图转换为输入序列的标记。由于这些视图仅在少数决策点上重要，仅进行微调在我们的实验中并未使策略使用它们；因此，我们采用两阶段训练，首先在上采样的反事实数据上进行演示，然后在恢复数据上进行训练。我们在两个模拟仓库任务上进行评估，找到指令中提到的物体并绕过被阻塞的过道，而这些关键信息通常仅对基础设施摄像头可见。在分布测试中，InfraVLA在这两个任务上均达到了100%的成功率，而没有CCTV输入的基线分别为34.0%和73.6%。在分布外测试集上，成功率为88.2%和88.9%。在一台经过不到10分钟演示微调的真实四足机器人上，该策略达到了83.3%的成功率，而仅依赖机载系统的基线为29.2%。
cs.RO / 89 / 2609.33653
Demonstration-Free Success-Probability Reward Learning for Generalist Robot Policies
无示范成功概率奖励学习的通用机器人策略
Abstract
Reinforcement learning (RL) enables generalist robot policies to improve through trial-and-error interaction, yet its effectiveness is fundamentally constrained by sparse task rewards. Existing general-purpose reward models typically alleviate this issue by learning task progress from expert demonstrations, but introduce a distribution mismatch with the mixed-quality rollouts encountered during policy optimization, making their estimates unreliable on suboptimal and failed behaviors from which the policy must learn. In this work, we introduce a demonstration-free reward learning paradigm where dense reward feedback can be learned directly from sparse task outcomes and policy experience. We theoretically show that terminal task outcomes implicitly define dense success-probability feedback at intermediate timesteps, which can be recursively learned through bootstrapping. Based on this insight, we introduce eVTA$_0$, which learns success probabilities from mixed-quality policy rollouts through temporal-difference-style bootstrapping, without expert demonstrations or intermediate annotations. We further introduce RL with Evolving Rewards (RLER), a closed-loop framework that adapts eVTA$_0$ using newly collected rollouts as the policy evolves. Experiments show that eVTA$_0$ provides more informative rewards than state-of-the-art reward models and achieves the best average policy performance across all LIBERO task suites under the same RL training budget, improving success rates by 5.4%-13.8% over the initial policy. In real-world manipulation, RLER further improves overall success rates by 20%-26%, with 35%-36% gains under out-of-distribution conditions. These results demonstrate the effectiveness of demonstration-free reward learning and adapting rewards as the policy evolves. Project webpage: https://duowuyms.github.io/evta0.
Chinese Translation
强化学习（RL）使得通用机器人策略能够通过试错交互进行改进，但其有效性在根本上受到稀疏任务奖励的限制。现有的通用奖励模型通常通过学习专家示范中的任务进展来缓解这一问题，但在策略优化过程中引入了与混合质量的回放之间的分布不匹配，使得它们在次优和失败行为上的估计不可靠，而这些行为是策略必须学习的。在本研究中，我们提出了一种无示范奖励学习范式，其中可以直接从稀疏任务结果和策略经验中学习到密集的奖励反馈。我们理论上证明，终端任务结果隐含地定义了中间时间步的密集成功概率反馈，这可以通过自举方法递归学习。基于这一见解，我们引入了 eVTA$_0$，该方法通过时间差分风格的自举从混合质量的策略回放中学习成功概率，无需专家示范或中间注释。我们进一步引入了带有演变奖励的强化学习（RLER），这是一个闭环框架，随着策略的演变，使用新收集的回放来适应 eVTA$_0$。实验表明，eVTA$_0$提供比最先进奖励模型更具信息性的奖励，并在相同的 RL 训练预算下，在所有 LIBERO 任务套件中实现最佳的平均策略性能，成功率提高了 5.4%-13.8%。在真实世界的操作中，RLER 进一步将整体成功率提高了 20%-26%，在分布外条件下的增益为 35%-36%。这些结果证明了无示范奖励学习的有效性以及在策略演变过程中适应奖励的能力。项目网页：https://duowuyms.github.io/evta0.
cs.RO / 90 / 2609.33681
Observability-Informed Optimal Sensor Placement for Soft Robots
基于可观测性优化的软机器人传感器最佳布置
Abstract
This paper presents the application and experimental evaluation of a systematic method for optimal sensor placement in soft robots. Existing methods either lack generalizability across different soft robot morphologies or do not account for system dynamics. The applied method uses convex optimization to find the optimal sensor configuration that maximizes an observability Gramian-based metric. The framework is experimentally evaluated using position and strain measurements on a soft continuum arm. Kalman filter state estimates using optimal sensor placements yield lower reconstruction error than a baseline across all sinusoidal input trials, with improvements on the order of millimeters. This case study shows that linear control theory tools can guide optimal sensor placement in soft robots, suggesting an interpretable approach to sensor placement that may extend to other morphologies.
Chinese Translation
本文提出了一种系统性方法在软机器人中进行最佳传感器布置的应用和实验评估。现有方法要么缺乏在不同软机器人形态上的普适性，要么未考虑系统动态。所应用的方法利用凸优化来寻找最大化基于可观测性Gramian度量的最佳传感器配置。该框架通过对软连续臂的位置信息和应变测量进行实验评估。使用最佳传感器布置的卡尔曼滤波状态估计在所有正弦输入试验中产生的重构误差低于基线，改进量级达到毫米级。该案例研究表明，线性控制理论工具可以指导软机器人的最佳传感器布置，暗示了一种可解释的传感器布置方法，可能扩展到其他形态。
cs.RO / 91 / 2609.33737
MomWorld: Momentum-Aware Latent World Model for Long-Horizon Autonomous Driving
MomWorld：面向长时间自主驾驶的动量感知潜在世界模型
Abstract
Long-horizon planning enables autonomous vehicles to anticipate scene evolution and potential risks, supporting safe and stable decisions in complex interactions. However, existing methods struggle to propagate motion trends from observed history into the future. Long rollouts based on a single latent state may further attenuate useful dynamics, retain stale motion patterns, and disrupt reliable near-term plans. We introduce MomWorld, a momentum-aware latent world model for long-horizon planning. MomWorld extracts scene motion trends from historical-to-current observations and propagates latent momentum into future horizons, jointly predicting future configuration and momentum states. A learnable momentum persistence mechanism preserves stable trends, scene-conditioned momentum updates adapt future dynamics, and a scene-adaptive reset gate suppresses stale momentum under abrupt changes. We further propose MoFlow, a momentum-conditioned flow-matching module that refines a base trajectory to align with the predicted future scene evolution in only a few integration steps, with a horizon-aware residual fusion that preserves near-term planning stability while permitting stronger long-range corrections. Extensive experiments on NAVSIM, nuScenes and Bench2Drive demonstrate that MomWorld improves long-horizon planning consistency and reduces the average collision rate by 12.2% relative to MomAD over a 6-second planning horizon.
Chinese Translation
长时间规划使自主车辆能够预测场景演变和潜在风险，从而在复杂交互中支持安全和稳定的决策。然而，现有方法在将观察历史中的运动趋势传播到未来方面存在困难。基于单一潜在状态的长时间展开可能进一步削弱有用的动态，保留过时的运动模式，并干扰可靠的短期计划。我们提出了MomWorld，一种面向长时间规划的动量感知潜在世界模型。MomWorld从历史到当前的观察中提取场景运动趋势，并将潜在动量传播到未来的时间范围，联合预测未来的配置和动量状态。可学习的动量持续机制保持稳定趋势，场景条件下的动量更新适应未来动态，而场景自适应重置门在突变情况下抑制过时动量。我们进一步提出了MoFlow，一种动量条件流匹配模块，它在仅需少量整合步骤的情况下，细化基础轨迹以与预测的未来场景演变对齐，并通过时间感知的残差融合保持短期规划的稳定性，同时允许更强的长距离修正。在NAVSIM、nuScenes和Bench2Drive上的大量实验表明，MomWorld提高了长时间规划的一致性，并相对于MomAD在6秒的规划范围内将平均碰撞率降低了12.2%。
cs.RO / 92 / 2609.33748
AnyStep-WAM: Budget-Aligned Distillation and Adaptive Inference for World Action Models
AnyStep-WAM：预算对齐的蒸馏与自适应推理用于世界行动模型
Abstract
World-action models (WAMs) couple predictive visual modeling with action generation, typically relying on iterative denoising with a fixed denoising steps. However, manipulation tasks contain actions chunks with varying sensitivity to generation errors: critical actions require precision, while less sensitive actions allow faster generation with fewer denoising steps. Here we introduce AnyStep World Action Model, a general framework for tunable-budget prediction and scene-dependent computation allocation. Our budget-aligned teacher-trajectory distillation trains interval-conditioned flow maps using explicit frozen-teacher transitions and shared low-rank adapters, supporting action generation from one-step prediction to multi-step refinement. Building on this capability, a lightweight risk-benefit scheduler predicts teacher-curvature-based difficulty and budget-specific student-teacher fidelity from a single one-step preview, selecting the smallest budget predicted to satisfy risk-adaptive fidelity requirements. We evaluate our framework on three widely used WAMs Motus, FastWAM, and LingBotVA using RoboTwin 2.0. Our method reduces average denoising steps by 60.2%, 49.8%, and 85.28%, respectively, while maintaining baseline task success rates. In particular, our AnyStep training substantially improves model performance under a one-step denoising budget, increasing task success rates by 7.07%, 12.08%, and 8.94% on Motus, FastWAM, and LingBotVA, respectively. Experiments on six real-world manipulation tasks further validate its effectiveness.
Chinese Translation
世界行动模型（WAMs）将预测视觉建模与动作生成结合起来，通常依赖于固定去噪步骤的迭代去噪。然而，操控任务包含对生成错误敏感性不同的动作片段：关键动作需要精确性，而不太敏感的动作则允许以更少的去噪步骤更快地生成。在此，我们引入了AnyStep世界行动模型，这是一个可调预算预测和场景依赖计算分配的一般框架。我们的预算对齐教师轨迹蒸馏通过显式的冻结教师过渡和共享低秩适配器训练区间条件流图，支持从一步预测到多步细化的动作生成。在此能力的基础上，一个轻量级风险-收益调度器从单一步骤预览中预测基于教师曲率的难度和预算特定的学生-教师保真度，选择预测满足风险适应性保真度要求的最小预算。我们在三种广泛使用的WAMs Motus、FastWAM和LingBotVA上使用RoboTwin 2.0评估我们的框架。我们的方法分别将平均去噪步骤减少了60.2%、49.8%和85.28%，同时保持基线任务成功率。特别是，我们的AnyStep训练在一步去噪预算下显著提高了模型性能，分别将Motus、FastWAM和LingBotVA的任务成功率提高了7.07%、12.08%和8.94%。在六个真实世界操控任务上的实验进一步验证了其有效性。
cs.RO / 93 / 2609.33765
Principal Steering Subspaces for Online Adaptation of Frozen Generative Robot Policies
用于冷冻生成机器人策略在线适应的主导引导子空间
Abstract
Generative robot policies provide expressive behavior priors, but updating a large diffusion or flow-matching model through online interaction is costly. Latent-space reinforcement learning avoids updating the pretrained generator by controlling its initial sampling noise, yet high-dimensional noise can have strongly anisotropic effects on decoded actions. We introduce Principal Steering Subspaces (PSS), a forward-query interface that constructs a fixed low-dimensional control basis from finite-difference decoder responses. Soft Actor-Critic controls the leading response directions, while the orthogonal complement is independently resampled from the Gaussian prior at each query. On three RoboMimic tasks with diffusion and flow-matching policies, response spectra reveal substantial concentration. Across five matched task-generator pairs, the training curves indicate that PSS generally converges faster and exhibits more stable late-training behavior than full-latent control, while achieving stronger final performance overall. Controlled Diffusion-Square ablations further show that leading-response directions outperform random and least-responsive subspaces of equal dimension. We further integrate PSS with a frozen, closed-source 3B-parameter vision-language-action (VLA) policy in a humanoid learning system with synchronous transition collection, reset-time optimization, and latency-aware asynchronous deployment. In an exploratory screwdriver-placement evaluation, success is observed in 2/10 trials for the frozen VLA policy and 6/10 after SAC+PSS adaptation. These results support decoder-response geometry as a practical basis for online adaptation of frozen generative robot policies.
Chinese Translation
生成机器人策略提供了丰富的行为先验，但通过在线交互更新大型扩散或流匹配模型的成本较高。潜在空间强化学习通过控制预训练生成器的初始采样噪声来避免更新，但高维噪声可能对解码动作产生强烈的各向异性影响。我们引入了主导引导子空间（Principal Steering Subspaces, PSS），这是一种前向查询接口，通过有限差分解码器响应构建固定的低维控制基。软演员评论家（Soft Actor-Critic）控制主要响应方向，而正交补空间则在每次查询时从高斯先验中独立重采样。在三个使用扩散和流匹配策略的RoboMimic任务中，响应谱显示出显著的集中趋势。在五对匹配的任务-生成器中，训练曲线表明PSS通常收敛更快，并且在后期训练中表现出更稳定的行为，同时整体上实现了更强的最终性能。受控扩散-平方消融实验进一步表明，主要响应方向优于相同维度的随机和响应最小的子空间。我们还将PSS与一个冷冻的、闭源的3B参数视觉-语言-动作（Vision-Language-Action, VLA）策略集成到一个具有同步过渡收集、重置时间优化和延迟感知异步部署的人形学习系统中。在一次探索性的螺丝刀放置评估中，冷冻VLA策略在10次试验中成功了2次，而在SAC+PSS适应后成功了6次。这些结果支持解码器响应几何作为冷冻生成机器人策略在线适应的实用基础。
cs.RO / 94 / 2609.33807
CodeActionBench: Evaluating Agentic Code-as-Policy for Embodied Manipulation
CodeActionBench：评估具身操作的代理代码作为政策
Abstract
How well can general-purpose multimodal models turn visual understanding and reasoning into embodied manipulation via executable code? We introduce CodeActionBench, a benchmark of 25 manipulation tasks that evaluates this capability through agentic Code-as-Policy. Without task-specific fine-tuning, demonstrations, external specialist perception or grasp modules, privileged scene state, or predefined task policies, agents should select visual evidence, form task-relevant 3D estimates, construct manipulation targets, and iteratively execute and revise their policies. A shared robot API provides RGB observations, calibrated geometric operations, robot feedback, and bounded motion, leaving task-dependent decisions to the evaluated agent. Fixed task instances, resource budgets, and a hidden physical-outcome verifier support controlled comparisons across models and harness configurations. Extensive evaluations across nine configurations and 675 attempts achieve success rates ranging from 2.7% to 73.3%. The strongest configuration, GPT-6 Astra with Codex CLI, solves 22 of 25 tasks at least once in three attempts, demonstrating the best performance while still leaving substantial room for improvement. Trajectory analyses reveal difficulties in spatial alignment, object retention, and completion judgment, including task failures despite successfully completed motions. CodeActionBench provides a controlled testbed for measuring how general-purpose models translate their capabilities into manipulation behavior and for examining typical failure scenarios in that process.
Chinese Translation
通用多模态模型在可执行代码的支持下，能够将视觉理解和推理转化为具身操作的能力如何？我们引入了CodeActionBench，这是一个包含25个操作任务的基准，旨在通过代理代码作为政策评估这一能力。在没有特定任务的微调、演示、外部专业感知或抓取模块、特权场景状态或预定义任务政策的情况下，代理应选择视觉证据，形成与任务相关的3D估计，构建操作目标，并迭代执行和修订其政策。共享的机器人API提供RGB观测、校准几何操作、机器人反馈和有限运动，将任务依赖的决策留给被评估的代理。固定任务实例、资源预算和隐藏的物理结果验证器支持跨模型和配置的受控比较。通过九种配置和675次尝试的广泛评估，成功率范围从2.7%到73.3%。最强的配置，GPT-6 Astra与Codex CLI，至少在三次尝试中解决了25个任务中的22个，展现了最佳性能，但仍有相当大的改进空间。轨迹分析揭示了在空间对齐、物体保持和完成判断方面的困难，包括尽管成功完成动作但仍出现任务失败的情况。CodeActionBench为测量通用模型如何将其能力转化为操作行为提供了一个受控测试平台，并用于检查该过程中的典型失败场景。
cs.RO / 95 / 2609.33832
Achieve What You Imagined: Learning to Align Actions with Visual Plans
实现你所想象的：学习将行动与视觉计划对齐
Abstract
World-action models can jointly predict future visual observations and robot actions. However, discrepancies may exist between their visual predictions and the consequences implied by generated actions. We observe that WAMs can often generate visually plausible task-completion outcomes before producing action sequences that reliably achieve them. Consequently, we treat the WAM-generated visual prediction as a goal-conditioned visual proposal rather than a directly executable plan. We use a frozen action-conditioned world model to predict action-conditioned consequences and construct feedback based on consistency between the two future predictions and alignment with the terminal goal. Leveraging this feedback, we employ Flow Policy Optimization (FPO) to optimize the action head of the WAM. This framework avoids online robot interaction and additional training of task-specific reward models. Across four real-world UR5 manipulation tasks, our method increases the mean success rate from 43.4% to 75.1%, compared with 61.4% for $\pi_{0.5}$. These results show that cross-model prediction discrepancy can provide useful feedback for improving robot policies under the evaluated manipulation tasks. Website: https://imagine-to-achieve.github.io/
Chinese Translation
世界行动模型（World-action models）可以共同预测未来的视觉观察和机器人行动。然而，它们的视觉预测与生成的行动所暗示的后果之间可能存在差异。我们观察到，世界行动模型（WAMs）通常能够在生成可靠的行动序列之前，产生视觉上合理的任务完成结果。因此，我们将WAM生成的视觉预测视为一种目标条件的视觉提案，而不是直接可执行的计划。我们使用一个冻结的行动条件世界模型来预测行动条件的后果，并基于两个未来预测之间的一致性以及与最终目标的对齐构建反馈。利用这一反馈，我们采用流政策优化（Flow Policy Optimization, FPO）来优化WAM的行动头。该框架避免了在线机器人交互和额外的任务特定奖励模型训练。在四个真实世界的UR5操作任务中，我们的方法将平均成功率从43.4%提高到75.1%，而$ ext{π}_{0.5}$的成功率为61.4%。这些结果表明，跨模型预测差异可以为改善机器人在评估操作任务中的策略提供有用的反馈。网站：https://imagine-to-achieve.github.io/
cs.RO / 96 / 2609.33836
DeltaSeek: Toward Active Perception in Evolving Construction Environments
DeltaSeek：朝向动态建筑环境中的主动感知
Abstract
Construction environments evolve continuously, causing large geometric changes that degrade static mapping and registration performance. This necessitates active perception, where robots deliberately select sensing configurations to resolve the environment's current state. We present DeltaSeek, an initial framework toward active perception in evolving built environments. While our broader objective is a system that reasons about where, how, and when to observe, this paper addresses a critical prerequisite: how a robot's sensing embodiment constrains the observations it can acquire. We formalize an embodiment's permissible observation set and evaluate with a Husky A300 equipped with a UR5e on an IFC-derived benchmark under chassis-mounted and wrist-mounted RGB-D configurations, scoring observations by geometric visibility and effort by drivable distance. In a room-scale scene with eight controlled changes spanning four observability conditions, exhaustive evaluation over 240 permissible base poses and five arm postures shows that two changes admit no chassis viewpoint whatsoever, while the wrist camera resolves both. For changes observed by both embodiments, the median base travel is $6.0$~m for the wrist camera and $15.2$~m for the chassis camera. These results distinguish sensing limitations from acquisition costs, clarifying whether an observation is impossible or simply requires more travel.
Chinese Translation
建筑环境持续演变，导致大规模几何变化，从而降低静态映射和配准性能。这就需要主动感知，机器人有意选择传感配置以解析环境的当前状态。我们提出了DeltaSeek，这是一个朝向动态建筑环境中主动感知的初步框架。虽然我们的更广泛目标是构建一个能够推理何时、如何以及在哪里观察的系统，但本文关注一个关键前提：机器人的感知体现如何限制其能够获取的观察。我们形式化了一个体现的可允许观察集，并在一个配备UR5e的Husky A300上进行评估，基于底盘安装和腕部安装的RGB-D配置，在一个基于IFC的基准测试中，通过几何可见性和可行距离对观察进行评分。在一个包含八个受控变化的房间规模场景中，跨越四种可观察性条件，对240个可允许的基础姿态和五种手臂姿势进行了全面评估，结果显示有两个变化根本不允许底盘视角，而腕部相机则能够解析这两者。对于两个体现都能观察到的变化，腕部相机的中位基础移动距离为6.0米，而底盘相机为15.2米。这些结果区分了感知限制与获取成本，澄清了一个观察是否不可能，或仅仅是需要更多的移动。
cs.RO / 97 / 2609.33872
Robot-GST: geometry-aware spatial-temporal robot policy representation and evaluation
Robot-GST：几何感知的时空机器人策略表示与评估
Abstract
Robotic manipulation policies are advancing rapidly with increasing reliance on vision-language models for end-to-end decision making. However, reliable deployment remains challenging because many policies lack explicit mechanisms for predicting task outcomes and evaluating whether generated actions will achieve desired final states, causing execution errors to accumulate during long-horizon manipulation. We present Robot-GST, a geometry-aware spatio-temporal behaviour representation and evaluation framework that constructs a Gaussian-SAM robotic environment for real-to-sim policy verification and improves the reliability of real-world manipulation deployment. Our approach constructs a high-fidelity robotic environment from RGB-D observations using 3D Gaussian Splatting and SAM3D, enabling ``simulation and evaluation before acting''. It integrates visual observations and language instructions with spatio-temporal reasoning for long-horizon task planning using large vision-language models. To bridge high-level planning and real-world execution, we introduce Gaussian-aware final-state estimation through geometric sampling and state-based trajectory planning. Before execution, candidate action sequences are simulated and evaluated in the Gaussian-SAM environment to filter infeasible behaviours. We validate our approach on representative manipulation tasks involving rigid, soft, and deformable objects, including cube placing, toy packing, and duck rearrangement, demonstrating that geometry-aware spatio-temporal reasoning and state-aware execution improve manipulation reliability across different object categories. Our results suggest that combining geometry-aware reconstruction with high-quality rendering and simulation provides a scalable approach for evaluating robotic manipulation behaviours. Website: https://robot-gst.github.io
Chinese Translation
机器人操控策略正在迅速发展，越来越依赖视觉-语言模型进行端到端决策。然而，由于许多策略缺乏明确的机制来预测任务结果并评估生成的动作是否能够实现期望的最终状态，导致在长时间操作过程中执行错误的累积，从而使得可靠部署面临挑战。我们提出了Robot-GST，一个几何感知的时空行为表示与评估框架，构建了一个Gaussian-SAM机器人环境用于真实到仿真的策略验证，并提高了现实世界操控部署的可靠性。我们的方法利用RGB-D观测通过3D Gaussian Splatting和SAM3D构建高保真机器人环境，实现“在行动前进行仿真与评估”。它将视觉观测和语言指令与时空推理相结合，使用大型视觉-语言模型进行长时间任务规划。为了弥合高层规划与现实执行之间的差距，我们引入了通过几何采样和基于状态的轨迹规划的高斯感知最终状态估计。在执行之前，候选动作序列在Gaussian-SAM环境中进行仿真和评估，以过滤不可行的行为。我们在涉及刚性、柔性和可变形物体的代表性操控任务上验证了我们的方法，包括立方体放置、玩具打包和鸭子重新排列，证明了几何感知的时空推理和状态感知执行在不同物体类别中提高了操控的可靠性。我们的结果表明，将几何感知重建与高质量渲染和仿真相结合提供了一种可扩展的方法来评估机器人操控行为。网站：https://robot-gst.github.io
cs.RO / 98 / 2609.33882
DexTaG: Tactile-as-Guidance in Reinforcement Learning for Dexterous Manipulation
DexTaG：在强化学习中利用触觉指导灵巧操作
Abstract
Glove-based motion capture is emerging as a scalable approach to collecting dexterous-hand demonstration data. However, due to the kinematic gap between the human and robot hand, the recorded human motions cannot be executed directly on the robot, especially for contact-rich tool-use tasks involving in-hand reorientation. Prior work bridges this gap in simulation through reinforcement learning (RL) or trajectory optimization, but the human contact pattern is hard to preserve under such formulations, often producing unnatural manipulation and unstable functional grasps. These methods also train a separate policy or solve a separate optimization for each reference trajectory, which is inefficient. To solve these problems, we propose DexTaG, a tactile-guided RL framework for dexterous manipulation. During training, tactile signals captured by the glove guide policy search toward the measured human contact pattern, reducing reliance on precise reference geometry for contact supervision. To improve efficiency, we train a single generalizable retargeter jointly on all training trajectories of the same object. The retargeter is further distilled into a tactile-free student controller conditioned on the target object trajectory for real-world deployment. On marker-pen and hammer manipulation tasks, DexTaG learns natural, contact-rich behaviors that baselines with distance-based contact heuristics fail to learn, generalizes to held-out trajectories of the same object and task, and outperforms single-trajectory baselines on OakInk2.
Chinese Translation
基于手套的运动捕捉正逐渐成为收集灵巧手演示数据的一种可扩展方法。然而，由于人类手与机器人手之间的运动学差距，记录的人类动作无法直接在机器人上执行，特别是在涉及手内重新定向的接触丰富的工具使用任务中。之前的研究通过强化学习（RL）或轨迹优化在模拟中弥补了这一差距，但在这种框架下很难保持人类的接触模式，常常导致不自然的操作和不稳定的功能性抓握。这些方法还为每个参考轨迹训练一个单独的策略或解决一个单独的优化问题，效率低下。为了解决这些问题，我们提出了DexTaG，一种用于灵巧操作的触觉指导强化学习框架。在训练过程中，手套捕捉的触觉信号引导策略搜索朝向测量的人类接触模式，从而减少对接触监督的精确参考几何形状的依赖。为了提高效率，我们在同一物体的所有训练轨迹上联合训练一个单一的可泛化重定向器。重定向器进一步被提炼为一个无触觉的学生控制器，条件是目标物体轨迹，以便于在现实世界中部署。在标记笔和锤子操作任务中，DexTaG学习到了自然的、接触丰富的行为，而基于距离的接触启发式基线未能学习这些行为，能够推广到相同物体和任务的保留轨迹，并在OakInk2上超越单轨迹基线。
cs.RO / 99 / 2609.33892
Residual Learning-Based Control of Vehicle Platoons with $\ell_2$ Stability Guarantees via Recurrent Equilibrium Networks
基于残差学习的车辆编队控制方法：通过递归平衡网络实现 $ ext{l}_2$ 稳定性保证
Abstract
This paper proposes a residual learning-based control framework for heterogeneous vehicle platoons subject to parametric uncertainty and external disturbances. A nominal controller designed via Linear Matrix Inequalities (LMIs), along with disturbance-observer compensation, is enhanced by a Recurrent Equilibrium Network (REN) trained offline using stored trajectories and nominal-model prediction errors. The REN is constrained to satisfy a prescribed $\ell_2$-gain bound, enabling sufficient small-gain conditions for local closed-loop stability and disturbance string stability. Experiments demonstrate reduced spacing and velocity errors relative to the nominal controller.
Chinese Translation
本文提出了一种基于残差学习的控制框架，适用于受参数不确定性和外部干扰影响的异构车辆编队。通过线性矩阵不等式（Linear Matrix Inequalities, LMIs）设计的名义控制器，结合干扰观测器补偿，利用离线训练的递归平衡网络（Recurrent Equilibrium Network, REN）进行增强，该网络使用存储的轨迹和名义模型预测误差进行训练。REN被约束以满足规定的 $ ext{l}_2$ 增益界限，从而实现局部闭环稳定性和干扰串稳定性的充分小增益条件。实验结果表明，相较于名义控制器，所提方法在间距和速度误差方面有显著降低。
cs.RO / 100 / 2609.33931
ArticulateArena: A Metric for Articulated Kinematics
ArticulateArena：一种用于关节运动学的度量
Abstract
Modern methods reconstruct or generate simulation-ready articulated objects, predicting not only their geometry but also how their parts are connected and allowed to move. Evaluating the geometry is straightforward, but evaluating the predicted articulation is not, because articulation specifies a motion rather than a shape, and there is no agreed distance between two motions. More specifically, existing protocols score joint type, axis direction, origin, and motion limits separately, although these parameters jointly describe a single physical motion, and the same motion can be written as different parameter values. As a result, a joint can score maximally wrong against an equivalent encoding of itself, and several component errors are ill-conditioned or undefined exactly where predictions become accurate. We propose ArticulateArena, a representation-invariant counterpart of Chamfer distance for articulation that compares the motions one-DOF joints induce rather than the parameters that encode them. It represents each joint by the unordered pair of its Lie-algebra endpoint twists, and we prove that the resulting quotient distance is a metric. It unifies fixed, revolute, prismatic, and helical joints, brings continuous joints into the same score through a compactification, and reads as the RMS motion of the moving part in meters when weighted by its mass distribution. A motion-aware tree edit distance lifts the metric to full kinematic trees, pricing structural errors such as spurious or missing joints in the same motion units as joint errors, and for a fixed inner product it remains a metric on trees up to relabeling. Alongside the metric we release ArticulateArena-20K, a new suite of 19,977 articulated objects with verified kinematics, and we re-evaluate published reconstruction methods on it under the new metric. Project page: https://heyumeng.com/ArticulateArena-web/
Chinese Translation
现代方法重建或生成适合模拟的关节对象，不仅预测其几何形状，还预测其各部分如何连接及允许的运动。评估几何形状是直接的，但评估预测的关节运动则并不简单，因为关节运动指定的是一种运动而非形状，并且没有公认的两种运动之间的距离。更具体地说，现有的协议分别对关节类型、轴方向、原点和运动限制进行评分，尽管这些参数共同描述了一种单一的物理运动，而同一种运动可以用不同的参数值表示。因此，一个关节可能在与其自身的等效编码对比时得分极低，并且在预测变得准确的地方，多个组件错误的条件不良或未定义。我们提出了ArticulateArena，这是一个与Chamfer距离相对应的表示不变的关节运动度量，它比较的是一自由度（one-DOF）关节所引发的运动，而不是编码它们的参数。它通过无序对其李代数端点扭转来表示每个关节，并且我们证明了由此产生的商距离是一个度量。它统一了固定关节、旋转关节、滑动关节和螺旋关节，通过紧致化将连续关节纳入同一评分，并在按其质量分布加权时以米为单位表示移动部分的均方根运动。一个运动感知的树编辑距离将该度量提升至完整的运动学树，以相同的运动单位对结构错误（如虚假或缺失的关节）和关节错误进行定价，并且在固定内积下，它在重新标记的树上仍然保持为一个度量。除了该度量外，我们还发布了ArticulateArena-20K，这是一个包含19,977个经过验证的关节对象的新套件，并在新的度量下重新评估了已发布的重建方法。项目页面：https://heyumeng.com/ArticulateArena-web/
cs.RO / 101 / 2609.33939
EpiTransfer: Sparse, Training-Free Long-Range Depth Estimation from Temporal Monocular Aerial Frames
EpiTransfer：基于时间单目航空帧的稀疏、无训练的长距离深度估计
Abstract
Reliable 3D spatial understanding is essential for autonomous navigation, obstacle avoidance, and scene reconstruction. While state-of-the-art learned depth estimation techniques achieve high accuracy in-distribution, they often generalize poorly to novel viewpoints and altitudes. This paper presents a geometrically derived, training-free depth estimation method using epipolar transfer with only two monocular images and camera pose estimates. By leveraging camera motion to synthesize a virtual stereo pair with a freely chosen baseline, our approach transforms temporal correspondence into a stereo triangulation task while mitigating geometric degeneracies inherent to direct two-view triangulation. Validated across outdoor drone flights (to a maximum range of approximately 90\,m) and indoor OptiTrack environments against LiDAR ground truth, the method achieves an indoor AbsRel of 0.092 and $\delta < 1.25$ of 0.940, comparable to direct triangulation (AbsRel 0.073) while retaining valid depth over a larger fraction of challenging scenes, and substantially outperforms off-the-shelf learning-based baselines such as ZoeDepth (AbsRel 0.225) and Depth Anything V2 (AbsRel 0.570), which are not trained or fine-tuned for this domain, with no training data required.
Chinese Translation
可靠的三维空间理解对于自主导航、障碍物避免和场景重建至关重要。尽管最先进的学习深度估计技术在分布内实现了高精度，但它们在新视角和高度下的泛化能力往往较差。本文提出了一种几何推导的、无训练的深度估计方法，利用仅有的两幅单目图像和相机姿态估计进行视差传递。通过利用相机运动合成一个具有自由选择基线的虚拟立体对，我们的方法将时间对应关系转化为立体三角测量任务，同时减轻了直接双视图三角测量固有的几何退化。在户外无人机飞行（最大范围约90米）和室内OptiTrack环境中与LiDAR地面真实值进行验证，该方法在室内的绝对相对误差（AbsRel）为0.092，$ ext{δ} < 1.25$为0.940， comparable to 直接三角测量（AbsRel 0.073），同时在更大比例的挑战性场景中保持有效深度，并显著优于现成的基于学习的基线，如ZoeDepth（AbsRel 0.225）和Depth Anything V2（AbsRel 0.570），这些方法并未针对该领域进行训练或微调，且无需训练数据。
cs.RO / 102 / 2609.33944
ReSync: Re-Aligning the Two Clocks of Asynchronous World-Action Models
ReSync：重新对齐异步世界-动作模型的两个时钟
Abstract
Jointly generating future video and actions has become a standard recipe for world-action models, and the strongest systems denoise the two streams on separate schedules: actions are decoded in few steps so control stays fast, while the video stream runs longer to keep the predicted future sharp. The design is deliberate, but it leaves the two streams on different clocks, and an action can become executable while the future that should justify it is still largely unresolved. We formalize this as a two-clock view of asynchronous inference and introduce the commitment-evidence gap, a quantity read directly from a model's own sampling schedule rather than measured by search. The gap is predictive: as it widens, candidate utility becomes harder to identify and extra candidate sampling buys less, while advancing the world stream buys more, and the two cross. Spending more world computation is therefore not simply better. The useful interval is closed at both ends, and both ends can be read off the schedule before any rollout. ReSync places the computation inside it: hold the action state, advance only the world within the supported window, then resume native denoising. No parameters change and no candidates are compared. On a frozen paired RoboCasa panel this improves success by 4.48 points, while an equal-compute control that waits without advancing the world does not move, and the same rule transfers to a second benchmark and a second backbone without retuning.
Chinese Translation
联合生成未来视频和动作已成为世界-动作模型的标准方法，而最强的系统在不同的时间表上对两个流进行去噪：动作在少数步骤中解码，以保持控制的快速性，而视频流运行更长时间以保持预测的未来清晰。这个设计是经过深思熟虑的，但它使得两个流处于不同的时钟上，一个动作可以在其应有的未来仍然大部分未解决时变得可执行。我们将此形式化为异步推理的双时钟视角，并引入承诺-证据差距（commitment-evidence gap），这一量度直接从模型自身的采样时间表中读取，而不是通过搜索来测量。该差距具有预测性：随着差距的扩大，候选效用变得更难识别，额外的候选采样带来的收益减少，而推进世界流则带来更多收益，二者相交。因此，花费更多的世界计算并不简单地意味着更好。有效区间在两端都被封闭，且在任何展开之前，两端都可以从时间表中读取。ReSync 将计算置于其中：保持动作状态，仅在支持的窗口内推进世界，然后恢复本地去噪。没有参数变化，也没有候选进行比较。在一个冻结的配对 RoboCasa 面板上，这提高了成功率 4.48 分，而一个没有推进世界的等计算控制则没有变化，且同样的规则可以转移到第二个基准和第二个主干网络，而无需重新调优。
cs.RO / 103 / 2609.33951
Integrity Detection and Characterization of Malicious Injections in RAVEN II
RAVEN II中恶意注入的完整性检测与特征化
Abstract
The increasing adoption of robotic systems in surgery, together with the expanding range of procedures they can support and the growing level of autonomy they provide, has substantially increased the complexity of surgical robots. As these systems integrate more sensors, controllers, communication interfaces, and model-driven control components, their attack surface continues to expand. A compromise of the integrity of a surgical robot can therefore cause unintended robot behavior and potentially threaten patient safety. In this paper, we characterize the detection boundary of malicious injections on RAVEN II using a public dataset that pairs the platform's telemetry with external high-resolution encoder ground truth. We identify three injection points spanning the command and observation paths and evaluate three injection patterns with increasing temporal dispersion. To capture different detection behaviors, we perform detection at two timescales: the window scale and the session scale. Rather than reporting detection rates at an arbitrarily chosen threshold, we quantify, for each injection point and injection pattern, the smallest end-effector deviation that can be resolved while maintaining an alarm rate acceptable for surgical operation. Our results show that detectability is strongly influenced by how the injected deviation is distributed over time. An abrupt step can be detected at deviations well below the 1 mm clinical tolerance, whereas the same overall deviation spread across a window or a session can remain hidden from single-window statistics. The open source code can be found at http://github.com/RAVENIIROS/RAVENIIIntegrity.
Chinese Translation
随着手术中机器人系统的日益普及，以及它们能够支持的手术范围的扩大和提供的自主性水平的提高，手术机器人的复杂性显著增加。随着这些系统集成了更多的传感器、控制器、通信接口和基于模型的控制组件，它们的攻击面也在不断扩大。因此，手术机器人完整性的妥协可能导致意外的机器人行为，并可能威胁到患者安全。本文利用一个公共数据集，该数据集将平台的遥测与外部高分辨率编码器的真实值相结合，表征了RAVEN II上恶意注入的检测边界。我们识别了跨越命令和观察路径的三个注入点，并评估了三种具有递增时间分散的注入模式。为了捕捉不同的检测行为，我们在两个时间尺度上进行检测：窗口尺度和会话尺度。我们不仅报告在任意选择的阈值下的检测率，还量化了每个注入点和注入模式下，在保持手术操作可接受的报警率的同时，能够解决的最小末端执行器偏差。我们的结果表明，检测能力受到注入偏差随时间分布方式的强烈影响。突变步骤可以在远低于1毫米临床容忍度的偏差下被检测到，而同样的总体偏差如果分布在一个窗口或一个会话中，则可能在单窗口统计中保持隐藏。开源代码可以在http://github.com/RAVENIIROS/RAVENIIIntegrity找到。
cs.RO / 104 / 2609.33973
FINGR: Learning Dexterous Hand Control for Real-World Rubik's Cube Solving
FINGR：用于现实世界魔方求解的灵巧手控制学习
Abstract
Manipulating a Rubik's Cube with a single dexterous hand is a challenging test of sustained, contact-rich control: the hand must execute successive layer turns while keeping the cube secure. Each turn requires some fingers to support the cube while others push a moving layer, release contact, and reset for the next move. To learn this coordination, we introduce FINGR (Future-supervised Interaction Network with Geometric Representations), a policy that combines finger-relative geometry with future interaction prediction. A shared point encoder expresses the cube relative to each fingertip and aggregates its points without depending on cubie indexing. Learned future tokens share the observation encoder and receive supervision for contact-force changes, layer-turn progress, and finger joint displacement at multiple time scales. The resulting representation conditions a flow policy that directly generates finger actions. On a real dexterous hand, our policy achieves 99.0% success over 300 turn attempts, compared with 79.7% for the base flow policy. Integrated with grasping and table-assisted regrasping, the policy solves all ten scrambled $2\times2\times2$ cubes in a mean complete-system time of approximately 137 seconds. The project website is available at https://www.lyt0112.com/projects/FINGR
Chinese Translation
用单只灵巧手操控魔方是一项对持续接触控制的挑战性测试：手必须在保持魔方安全的同时执行连续的层转动。每次转动需要一些手指支撑魔方，而其他手指则推动移动的层，释放接触并为下一步重置。为了学习这种协调性，我们提出了FINGR（未来监督交互网络与几何表示），这是一种将手指相对几何与未来交互预测相结合的策略。共享的点编码器将魔方相对于每个手指尖进行表达，并在不依赖魔方块索引的情况下聚合其点。学习到的未来标记共享观察编码器，并在多个时间尺度上接受接触力变化、层转动进度和手指关节位移的监督。最终的表示条件生成一个流策略，直接生成手指动作。在真实的灵巧手上，我们的策略在300次转动尝试中实现了99.0%的成功率，而基础流策略的成功率为79.7%。与抓取和桌面辅助重抓相结合，该策略在平均完整系统时间约为137秒内解决了所有十个打乱的$2 imes2 imes2$魔方。项目网站可访问 https://www.lyt0112.com/projects/FINGR
cs.RO / 105 / 2609.33982
Test-Time Spatial Reasoning for Robot Manipulation Using Generative Real-to-Sim
基于生成真实到仿真的机器人操作测试时空间推理
Abstract
Spatial reasoning is fundamental to general robot intelligence, as it enables robots to complete long-horizon tasks involving multi-object interaction. We introduce Simify, a training-free, test-time framework that performs explicit spatial reasoning via massively parallel physics simulation. From a single RGB-D image of a scene, Simify reconstructs simulation-ready assets leveraging 3D generative models and vision-language models. Then given a task specified by a reward function (e.g., build the tallest tower), Simify launches thousands of parallel rollouts in simulation and performs an evolutionary search to optimize object arrangements, typically converging within seconds. We conduct quantitative experiments on real-robot hardware to demonstrate the ability of our framework to execute complex object rearrangement tasks end-to-end with previously unseen objects. Results show that our framework outperforms prior work on foundation models for spatial reasoning by effectively exploiting large-scale parallel simulation during inference, and also highlight the importance of complete and accurate geometry for successful sim-to-real transfer.
Chinese Translation
空间推理是通用机器人智能的基础，因为它使机器人能够完成涉及多物体交互的长期任务。我们介绍了Simify，这是一种无训练、测试时框架，通过大规模并行物理仿真执行显式空间推理。Simify从场景的单个RGB-D图像中重建可用于仿真的资产，利用3D生成模型和视觉-语言模型。然后，给定由奖励函数指定的任务（例如，建造最高的塔），Simify在仿真中启动数千个并行回合，并执行进化搜索以优化物体排列，通常在几秒钟内收敛。我们在真实机器人硬件上进行了定量实验，以展示我们的框架在端到端执行复杂物体重排任务时的能力，且任务中包含先前未见的物体。结果表明，我们的框架在空间推理的基础模型上优于之前的工作，通过有效利用推理过程中的大规模并行仿真，此外还强调了完整和准确的几何形状对于成功的仿真到现实转移的重要性。
cs.RO / 106 / 2609.34006
TacGooseBumps (TacGB): Retrofitting Normal-Only Tactile Sensors with Shear Encoding for Learning Contact-Rich Manipulation
TacGooseBumps (TacGB)：为学习接触丰富的操作改造仅具法向感知的触觉传感器以实现剪切编码
Abstract
Contact-rich policies often fail because distinct physical states look alike yet require different actions. Cameras may not reveal whether a connector is aligned or fully seated, while many normal-only tactile sensors can miss the tangential interactions perpendicular to the grasping direction that distinguish these states. We ask whether a learning policy needs calibrated shear measurements, or only a repeatable observation that separates shear-dependent contact states. We introduce TacGooseBumps (TacGB), a passive domed film that mechanically encodes tangential loading as pattern changes in an existing sensor's pressure map. Tangential loading tilts each dome and redistributes pressure across its footprint; an end-to-end policy consumes the resulting maps without added electronics, force reconstruction, or taxel-level dome alignment. Across four imitation-learning tasks and two data-collection pipelines, TacGB improves goal attainment, efficiency, and contact quality: insertion success increases by up to 36 percentage points, and successful insertions are completed faster, while fragile-object placement becomes gentler and drawing becomes more continuous and straight. Signal, stage-wise, failure-mode, and trajectory analyses link these gains to contact regimes in which task-relevant tangential interactions are poorly resolved by vision and normal pressure alone. Together, these results show that shear need not be measured metrically to benefit robot learning; it can instead be mechanically encoded without changing the underlying tactile sensor or the policy's pressure-map input format.
Chinese Translation
接触丰富的策略常常失败，因为不同的物理状态看起来相似但需要不同的动作。相机可能无法揭示连接器是否对齐或完全就位，而许多仅具法向感知的触觉传感器可能会错过与抓取方向垂直的切向交互，这些交互能够区分这些状态。我们探讨学习策略是否需要经过校准的剪切测量，或者仅需要可重复的观察来区分依赖剪切的接触状态。我们介绍了TacGooseBumps (TacGB)，这是一种被动的圆顶薄膜，它通过在现有传感器的压力图中机械编码切向载荷的模式变化。切向载荷使每个圆顶倾斜，并在其接触面上重新分配压力；端到端的策略可以直接使用生成的图像，而无需额外的电子设备、力重建或税点级圆顶对齐。在四个模仿学习任务和两个数据收集管道中，TacGB提高了目标达成率、效率和接触质量：插入成功率提高了多达36个百分点，成功插入的速度更快，而脆弱物体的放置变得更加温和，绘图变得更加连续和直线。信号、阶段性、失败模式和轨迹分析将这些收益与任务相关的切向交互的接触状态联系起来，这些状态仅通过视觉和法向压力无法很好地解析。综合来看，这些结果表明，剪切不必以度量方式进行测量即可促进机器人学习；它可以通过机械编码的方式实现，而无需改变基础的触觉传感器或策略的压力图输入格式。
cs.RO / 107 / 2609.34010
ZeroBot: Learning from Scratch in Minutes with Generative Real2Sim
ZeroBot：在几分钟内通过生成式真实到仿真学习从零开始
Abstract
We present ZeroBot, a real2sim framework for learning a robot manipulation task from scratch in minutes under challenging conditions: zero human demonstrations, zero policy pre-training, and zero known object models. Given only a single view of an object and a goal pose for that object, ZeroBot uses image-to-3D generative models to obtain a complete object mesh, which is used in simulation for large-scale parallel reinforcement learning. To accelerate training, we introduce an action space which leverages the generated geometry and learned value function to sample states involving robot-object contact. When evaluated on real-world tasks including grasping, pushing, articulated object interaction, and multi-stage manipulation, ZeroBot achieves an 87% success rate with an average training time of 119 seconds. These results show the value of using image-to-3D models in a real2sim framework for rapid, autonomous robot learning.
Chinese Translation
我们提出了ZeroBot，一个真实到仿真（real2sim）框架，用于在具有挑战性的条件下从零开始学习机器人操控任务：零人类示范、零策略预训练和零已知物体模型。仅给定物体的单一视图和该物体的目标姿态，ZeroBot利用图像到三维（image-to-3D）生成模型获取完整的物体网格，该网格用于大规模并行强化学习的仿真。为了加速训练，我们引入了一种动作空间，该空间利用生成的几何形状和学习到的价值函数来采样涉及机器人与物体接触的状态。在对包括抓取、推动、关节物体交互和多阶段操控等实际任务进行评估时，ZeroBot实现了87%的成功率，平均训练时间为119秒。这些结果表明，在真实到仿真框架中使用图像到三维模型对于快速、自动的机器人学习具有重要价值。
cs.RO / 108 / 2609.34018
Estimate, Don't Imitate: Reusing Differentiable State-Based Policies for Visuomotor Control
估计，而非模仿：重用可微分状态基础策略进行视觉运动控制
Abstract
Simulation-trained manipulation policies can exploit privileged state information to learn effective contact-rich behaviours, but deployment requires acting from partial observations such as noisy camera images. A common solution is teacher-student distillation, in which a visuomotor policy is trained to reproduce the actions of the privileged expert. This requires the student to jointly infer the task-relevant state and relearn the expert's action mapping that is already available. An alternative is to reuse the state-based expert and learn only a perceptual interface that reconstructs its missing state inputs. However, minimising the state estimate error alone does not necessarily minimise the downstream control error induced by these estimates. To bridge this gap, we train a visual state estimator using both direct state supervision and an action-consistency loss backpropagated through the frozen, differentiable expert. A scheduled objective first establishes a physically meaningful state estimate and progressively emphasises errors that affect the expert's actions. Across five goal-conditioned manipulation tasks, retaining the expert consistently outperforms direct pixel-to-action imitation from the same expert demonstration corpus. We further demonstrate sim-to-real transfer on a physical Panda robot, achieving 76% success without retraining the underlying expert.
Chinese Translation
通过仿真训练的操控策略可以利用特权状态信息学习有效的接触丰富行为，但在实际应用中需要从部分观测中进行操作，例如嘈杂的摄像头图像。一个常见的解决方案是教师-学生蒸馏，其中视觉运动策略被训练以重现特权专家的动作。这要求学生共同推断任务相关的状态，并重新学习已经可用的专家动作映射。另一种选择是重用基于状态的专家，仅学习重建其缺失状态输入的感知接口。然而，仅仅最小化状态估计误差并不一定能最小化这些估计所引起的下游控制误差。为了解决这一问题，我们使用直接状态监督和通过冻结的可微分专家反向传播的动作一致性损失来训练视觉状态估计器。一个调度目标首先建立一个物理上有意义的状态估计，并逐步强调影响专家动作的误差。在五个目标条件的操控任务中，保留专家的表现始终优于直接从同一专家演示语料库进行的像素到动作的模仿。我们进一步在物理Panda机器人上展示了仿真到现实的转移，实现了76%的成功率，而无需重新训练基础专家。
cs.RO / 109 / 2609.34061
Quantile Head for Vision-Language-Action Models
量子头用于视觉-语言-动作模型
Abstract
Vision-Language-Action (VLA) models integrate pretrained Vision-Language Models (VLMs) with action heads for robot control. Common action heads have distinct limitations: point regression provides only a point estimate of the action distribution, while standard flow-matching samplers require costly iterative sampling. To address these limitations, we unify regression and flow matching under a shared objective and extend it to derive a quantile objective. This quantile objective guides the design of our Quantile Head, which predicts a median and positive gaps to form ordered marginal action quantiles in one forward pass. These quantiles support multiple sampling strategies without retraining and are jointly supervised to train the default median policy. Our local analysis of this joint supervision shows that, with calibrated nearby quantiles, fixed gaps, and matched correction speed, direct median updates have lower variance than under median-only supervision. Experiments show that this jointly supervised median policy achieves the highest average success rates among the compared methods on LIBERO, LIBERO-Plus, LIBERO-Pro, and two real-robot tasks, together with the shortest mean episode time among matched LIBERO baselines; code is available at https://github.com/xwangrs/Quantile-Head-for-VLA.
Chinese Translation
视觉-语言-动作（VLA）模型将预训练的视觉-语言模型（VLMs）与动作头结合用于机器人控制。常见的动作头存在明显的局限性：点回归仅提供动作分布的点估计，而标准流匹配采样器则需要昂贵的迭代采样。为了解决这些局限性，我们在共享目标下统一了回归和流匹配，并扩展以推导出量子目标。该量子目标指导我们量子头的设计，量子头在一次前向传递中预测中位数和正间隔，以形成有序的边际动作量子。这些量子支持多种采样策略而无需重新训练，并共同监督以训练默认的中位数策略。我们对这种联合监督的局部分析表明，在经过校准的邻近量子、固定间隔和匹配的修正速度下，直接的中位数更新的方差低于仅在中位数监督下的情况。实验表明，这种联合监督的中位数策略在LIBERO、LIBERO-Plus、LIBERO-Pro和两个真实机器人任务中实现了最高的平均成功率，并且在匹配的LIBERO基线中具有最短的平均回合时间；代码可在 https://github.com/xwangrs/Quantile-Head-for-VLA 获取。
cs.RO / 110 / 2609.34085
AD-E2E-JEPA: A Joint-Embedding Predictive Architecture For End-to-End Autonomous Driving
AD-E2E-JEPA：一种用于端到端自主驾驶的联合嵌入预测架构
Abstract
Autonomous driving requires \textit{world models} that can understand the physical world, reason and plan, and operate safely. In this paper, we first systematically evaluate existing action-conditioned joint-embedding predictive architecture (JEPA) world models, including LeWM, DINO-WM, and JEPA-WM for end-to-end autonomous driving (E2EAD). To isolate world-model quality from policy learning, we employ a goal-conditioned zero-shot planning setting that evaluates these models using ground-truth future observations as goals, without training any driving policy. We find that existing JEPA-based world models are either accurate for driving but computationally expensive, or computationally efficient but insufficient for planning. To address this trade-off, we propose \textbf{AD-E2E-JEPA}, which introduces a SIGReg-regularized learnable projector applied to projected patch embeddings. The projector reduces the number of planning patches by $16\times$ and the embedding dimension by $4\times$, achieving a $100\times$ inference speedup while retaining planning performance, with a 0.8-second runtime for an 8-frame rollout over 256 candidate trajectories. \textit{Without} training any driving policy, the world model itself reaches the goals located 20 meters away on average within the displacement of respectively 4.0/2.8 meters, using world-model rollouts over trajectory vocabularies of respectively 256/8,192 candidates. On the NAVSIMv2 benchmark, it achieves 67.3/72.9 EPDMS with multiplicative safety metrics and 84.1/86.5 EPDMS$^{\dagger}$ without them in goal-conditioned zero-shot planning. Experiments further show that the self-supervised pretrained projector improves downstream imitation learning performance from 80.2 to 85.4 EPDMS. The source code is available at https://github.com/HaoranZhuExplorer/AD-E2E-JEPA
Chinese Translation
自主驾驶需要能够理解物理世界、进行推理和规划并安全操作的 extit{世界模型}。在本文中，我们首先系统性地评估现有的基于动作条件的联合嵌入预测架构（JEPA）世界模型，包括LeWM、DINO-WM和JEPA-WM，以用于端到端自主驾驶（E2EAD）。为了将世界模型的质量与策略学习隔离，我们采用了一种目标条件的零样本规划设置，该设置使用真实未来观测作为目标来评估这些模型，而不训练任何驾驶策略。我们发现现有的基于JEPA的世界模型要么在驾驶上准确但计算成本高，要么计算效率高但规划能力不足。为了解决这一权衡问题，我们提出了 extbf{AD-E2E-JEPA}，该模型引入了一种应用于投影补丁嵌入的SIGReg正则化可学习投影器。该投影器将规划补丁的数量减少了$16 imes$，将嵌入维度减少了$4 imes$，在保持规划性能的同时，实现了$100 imes$的推理加速，对于256条候选轨迹的8帧展开，其运行时间为0.8秒。 extit{在不}训练任何驾驶策略的情况下，世界模型本身在平均位移分别为4.0/2.8米的情况下，能够达到距离20米的目标，使用的世界模型展开分别为256/8,192条轨迹词汇。在NAVSIMv2基准测试中，它在目标条件的零样本规划中实现了67.3/72.9的EPDMS，使用乘法安全指标时为84.1/86.5 EPDMS$^{ extdagger}$。实验进一步表明，自监督预训练的投影器将下游模仿学习性能从80.2提升至85.4 EPDMS。源代码可在https://github.com/HaoranZhuExplorer/AD-E2E-JEPA获取。
cs.RO / 111 / 2609.34145
Beyond Retrieval Relevance: Scene-Grounded Risk Entailment for Vision-Language Driving
超越检索相关性：基于场景的风险推理在视觉-语言驾驶中的应用
Abstract
Retrieval-augmented generation (RAG) gives vision--language driving systems access to external safety knowledge, yet a retrieved risk rule may be relevant without applying to the current scene. A vision--language model (VLM) receiving such knowledge must ground objects, bind entities across time, and verify relations before deciding how to act, leaving the support for risk conclusions implicit. We address this relevance--applicability gap with a Driving-Risk Knowledge Graph (DRKG) and Semantic Web Rule Language (SWRL) reasoning stage before VLM decision-making. Structured perception instantiates scene facts, from which SWRL rules derive events and directed risk relations when their antecedents are jointly satisfied. Recognized events, bound risk relations, and semantic descriptions of activated rules form compact evidence that conditions the VLM and diffusion planner. In matched comparisons on nuReasoning, our method improved the nuReasoning planning score (NPS) by 1.30 points and the non-at-fault collision score (NC) by 2.76 points over the relevance retrieval-based baseline. These gains indicate that scene-applicable risk evidence improves safety-weighted planning relative to semantically retrieved risk knowledge.
Chinese Translation
检索增强生成（RAG）为视觉-语言驾驶系统提供了外部安全知识的访问，但检索到的风险规则可能是相关的，却不一定适用于当前场景。接收此类知识的视觉-语言模型（VLM）必须对物体进行定位，跨时间绑定实体，并验证关系，然后才能决定如何行动，这使得风险结论的支持变得隐含。我们通过驾驶风险知识图（DRKG）和语义网规则语言（SWRL）推理阶段来解决这种相关性与适用性之间的差距，进而为VLM的决策提供支持。结构化感知实例化场景事实，从中SWRL规则在其前提共同满足时推导出事件和有向风险关系。被识别的事件、绑定的风险关系以及激活规则的语义描述形成了紧凑的证据，条件化VLM和扩散规划器。在nuReasoning的匹配比较中，我们的方法将nuReasoning规划分数（NPS）提高了1.30分，非过失碰撞分数（NC）提高了2.76分，相较于基于相关性检索的基线。这些提升表明，适用于场景的风险证据相较于语义检索的风险知识改善了安全加权规划。
cs.RO / 112 / 2609.34163
Reliability-Aware Sparse Route Memory for Round-Trip Vision-Language Navigation
面向往返视觉-语言导航的可靠性意识稀疏路线记忆
Abstract
Vision-language navigation (VLN) is typically evaluated as a one-way task, although deployed robots may need to return after reaching a goal. We study continuous round-trip VLN and diagnose failures in directional observability, deviation recovery, and termination stability. We propose a reliability-aware sparse route memory that records the executed Outbound trajectory as ordered geometric anchors and queries them in reverse through a structured hint, action-level arbitration, and terminal verification. On 50 reverse-paired episodes using NaVILA and a simulated Unitree Go2, language-only Return succeeds in 22.0% of episodes, while our online system reaches 55.1%. With exact route information, the same interfaces achieve 86.0%, showing that effective Return requires both accurate information and consistent action on that information. The remaining online gap arises mainly from geometric evidence that is too unreliable to authorise intervention. These results distinguish information quality, behavioural consistency, and online reliability as separate limits in long-horizon navigation.
Chinese Translation
视觉-语言导航（VLN）通常被评估为单向任务，尽管部署的机器人在到达目标后可能需要返回。我们研究了连续的往返 VLN，并诊断了方向可观测性、偏差恢复和终止稳定性方面的失败。我们提出了一种可靠性意识的稀疏路线记忆，它将执行的外出轨迹记录为有序的几何锚点，并通过结构化提示、动作级仲裁和终端验证以反向查询这些锚点。在使用 NaVILA 和模拟 Unitree Go2 的 50 个反向配对情节中，仅使用语言的返回成功率为 22.0%，而我们的在线系统达到了 55.1%。在获取准确的路线信息后，相同的接口成功率达到了 86.0%，这表明有效的返回需要准确的信息和对该信息的一致行动。剩余的在线差距主要源于几何证据过于不可靠，无法授权干预。这些结果将信息质量、行为一致性和在线可靠性区分为长时间导航中的不同限制。
cs.RO / 113 / 2609.34170
RAVEL: Asynchronous Rolling Inference for Flow-Based Vision-Language-Action Models
RAVEL：基于流的视觉-语言-动作模型的异步滚动推理
Abstract
Flow-based vision-language-action (VLA) models are highly effective for generalist robot manipulation, yet their reliance on computationally expensive VLM encoding and multi-step iterative action generation imposes a significant latency bottleneck. The resulting inference latency makes it difficult for robots to respond quickly, especially in dynamic environments. We address this limitation with RAVEL (Rolling Asynchronous VLA Enabling Low-Latency Control), an asynchronous inference framework that addresses the computational bottlenecks of both the VLM backbone and the action expert. To reduce the delay from multi-step action denoising, RAVEL allows near-term actions to be executed after a single denoising step by carrying partially denoised future actions forward in a rolling buffer. To avoid blocking on slow VLM encoding, RAVEL decouples VLM encoding from rolling action generation, allowing the action expert to operate continuously using the latest available VLM context, while a lightweight Fast Observation Pathway (FOP) directly conditions the action expert on current observations. Across simulated and real-world manipulation tasks, RAVEL consistently achieves substantially lower response latency while maintaining the task capability of the underlying VLA, enabling high-frequency and responsive closed-loop control.
Chinese Translation
基于流的视觉-语言-动作（VLA）模型在通用机器人操作中表现出色，但其对计算资源密集型的VLM编码和多步骤迭代动作生成的依赖导致了显著的延迟瓶颈。这种推理延迟使得机器人在动态环境中难以快速响应。我们通过RAVEL（滚动异步VLA以实现低延迟控制）来解决这一限制，RAVEL是一个异步推理框架，旨在解决VLM主干和动作专家的计算瓶颈。为了减少多步骤动作去噪带来的延迟，RAVEL允许在经过单次去噪步骤后执行近期动作，通过在滚动缓冲区中转发部分去噪的未来动作。为了避免在慢速VLM编码上造成阻塞，RAVEL将VLM编码与滚动动作生成解耦，使得动作专家能够持续使用最新的VLM上下文进行操作，同时轻量级的快速观察通道（Fast Observation Pathway, FOP）直接根据当前观察条件化动作专家。在模拟和真实世界的操作任务中，RAVEL始终实现了显著更低的响应延迟，同时保持了基础VLA的任务能力，从而实现高频率和响应迅速的闭环控制。
cs.RO / 114 / 2609.34175
FailPatch: Failure Residual Patching for Vision-Language-Action Models
FailPatch：用于视觉-语言-动作模型的失败残差修补
Abstract
Vision-Language-Action (VLA) policies are typically adapted using successful demonstrations, which provide direct action supervision but rarely cover failure-prone states. Deployment failures expose these states, yet lack the corrective actions needed for conventional supervised learning. We propose FailPatch, a failure-driven residual patching framework that decouples action supervision from execution-reliability supervision. Successful demonstrations ground how the policy should act, while deployment trajectories indicate when its behavior becomes unreliable. We further observe that action hidden representations exhibit clear linear separability between reliable and failure-associated states while directly conditioning action generation. Building on these insights, FailPatch introduces a Null-gated Residual Expert Bank into the action hidden space of a frozen VLA policy. A unified Preserve--Redirect--Trust objective retains the original policy in reliable states, selects residual experts in failure-associated states and redirects representations from failure regions toward success-associated regions under bounded intervention. With only 0.52% trainable parameters, FailPatch improves success rates by 11.0 percentage points on four long-horizon RoboTwin tasks under clean evaluation, 9.5 percentage points under clean-to-random generalization, and 16.7 percentage points over the baseline across three real-world tasks. Project and code: https://github.com/yupeng-2003/FailPatch.
Chinese Translation
视觉-语言-动作（VLA）策略通常通过成功示范进行适应，这些示范提供了直接的动作监督，但很少涵盖易失败的状态。部署失败暴露了这些状态，但缺乏传统监督学习所需的纠正措施。我们提出了FailPatch，一种以失败为驱动的残差修补框架，旨在将动作监督与执行可靠性监督解耦。成功的示范为策略的行为提供了基础，而部署轨迹则指示其行为何时变得不可靠。我们进一步观察到，动作的隐藏表示在可靠状态与与失败相关的状态之间表现出明显的线性可分性，同时直接影响动作生成。基于这些见解，FailPatch在冻结的VLA策略的动作隐藏空间中引入了一个空门控残差专家库。统一的保留-重定向-信任目标在可靠状态下保留原始策略，在与失败相关的状态下选择残差专家，并在有限干预下将来自失败区域的表示重定向到成功相关区域。仅用0.52%的可训练参数，FailPatch在四个长时间跨度的RoboTwin任务中，在干净评估下提高了11.0个百分点的成功率，在干净到随机的泛化下提高了9.5个百分点，并在三个现实世界任务中较基线提高了16.7个百分点。项目和代码：https://github.com/yupeng-2003/FailPatch。
cs.RO / 115 / 2609.34182
Unified Visual-Tactile-Action Modeling from Human Demonstrations for Dexterous Manipulation
基于人类示范的统一视觉-触觉-动作建模用于灵巧操作
Abstract
Dexterous manipulation requires tactile feedback.However, robot tactile demonstrations are difficult to scale,because dexterous-hand teleoperation provides limited tactile feedback to the operator. In contrast, human demonstrations offer a substantially more scalable source of diverse tactile interactions. Motivated by a simple premise: hands can change, but the underlying physics of interaction does not. We leverage human tactile data to improve dexterous manipulation policies. Specifically, we first build a tactile motion-capture system that synchronously records images, tactile signals, and hand motions. Using this system, we construct the UVTA dataset spanning five contact-rich tasks, with 1,000 human demonstrations covering diverse interaction patterns and 150 robot demonstrations per task. To transfer the underlying physics of human interaction to robot control, we propose a Unified Visual-Tactile-Action Model that maps both embodiments into aligned tactile and action representations and jointly predicts future action and tactile trajectories. The joint objective enables human demonstrations to supervise contact-aware representation learning, while only robot actions are executed during deployment. In real-robot evaluations across five tasks, our method achieves an average success rate of 70%, outperforming the strongest visual-tactile baseline, which achieves 29%, and an architecture ablation, which achieves 42%. Performance improves consistently with additional human demonstrations and exhibits no saturation at 1,000 demonstrations per task, validating the effectiveness of scalable human tactile data for dexterous manipulation. Project page is available at https://uni-vta.github.io/.
Chinese Translation
灵巧操作需要触觉反馈。然而，机器人触觉示范难以扩展，因为灵巧手的遥操作为操作员提供的触觉反馈有限。相比之下，人类示范提供了一个更具可扩展性的多样化触觉交互源。我们的动机基于一个简单的前提：手可以改变，但交互的基本物理原理不会改变。我们利用人类触觉数据来改善灵巧操作策略。具体而言，我们首先构建了一个触觉运动捕捉系统，该系统同步记录图像、触觉信号和手部运动。利用该系统，我们构建了UVTA数据集，涵盖五个接触丰富的任务，包含1,000个涵盖多样化交互模式的人类示范和每个任务150个机器人示范。为了将人类交互的基本物理原理转移到机器人控制中，我们提出了一个统一的视觉-触觉-动作模型，该模型将两种体现映射到对齐的触觉和动作表示，并共同预测未来的动作和触觉轨迹。联合目标使人类示范能够监督接触感知表示学习，而在部署过程中仅执行机器人动作。在五个任务的真实机器人评估中，我们的方法实现了70%的平均成功率，超越了最强的视觉-触觉基线（29%）和架构消融（42%）。随着额外人类示范的增加，性能持续改善，并且在每个任务的1,000个示范中没有饱和，验证了可扩展的人类触觉数据在灵巧操作中的有效性。项目页面可访问 https://uni-vta.github.io/。
cs.RO / 116 / 2609.34199
WB-WAM: Heterogeneous Body-Hand Pre-training for Humanoid Loco-Manipulation
WB-WAM：用于类人运动操控的异构身体-手预训练
Abstract
Humanoid loco-manipulation demands coordinated body and hand behavior, while conventional robot pre-training data provide limited coverage of such whole-body motion. We present WB-WAM, a World Action Model that incorporates explicit whole-body action supervision into generative video pre-training. A shared physical action space integrates body, root, and dexterous hand annotations from heterogeneous sources, enabling joint video and action learning from 1880.2 hours of partially annotated video and motion data. The resulting priors are refined through PICO mid-training and adapted to robot tasks with auxiliary forward kinematics supervision. We construct WB-Datasets to support these stages with retargeted egocentric human demonstrations and robot trajectories, allowing task-aligned human motion to supplement limited robot data. Evaluations in simulation demonstrate strong whole-body task performance with 81.9% in HumanoidArena, while real-world experiments further validate WB-WAM with 84.0% mean success across five tasks. Moreover, task-aligned PICO mid-training improves downstream task performance while reducing the need for real-robot demonstrations. These results support heterogeneous whole-body pre-training and human motion transfer as a practical route to data-efficient humanoid loco-manipulation.
Chinese Translation
类人运动操控要求身体和手的协调行为，而传统的机器人预训练数据对这种全身运动的覆盖有限。我们提出了WB-WAM，一个世界动作模型（World Action Model），将显式的全身动作监督融入生成视频的预训练中。一个共享的物理动作空间整合了来自异构来源的身体、根部和灵巧手的注释，使得从1880.2小时的部分注释视频和运动数据中进行联合视频和动作学习成为可能。通过PICO中期训练对生成的先验进行精炼，并通过辅助的正向运动学监督将其适应于机器人任务。我们构建了WB-Datasets，以支持这些阶段，提供重新定向的自我中心人类演示和机器人轨迹，使得任务对齐的人类运动能够补充有限的机器人数据。模拟评估表明，在HumanoidArena中，WB-WAM在全身任务表现上达到了81.9%的优异成绩，而现实世界实验进一步验证了WB-WAM，在五个任务中平均成功率达到84.0%。此外，任务对齐的PICO中期训练在提高下游任务性能的同时减少了对真实机器人演示的需求。这些结果支持异构全身预训练和人类运动迁移作为实现数据高效类人运动操控的实用途径。
cs.RO / 117 / 2609.34210
RLE-Bench: A Qualifying Exam for Coding Agents as Robot Learning Engineers
RLE-Bench：作为机器人学习工程师的编码代理资格考试
Abstract
Coding agents are beginning to move beyond purely digital tasks to tackle physical-world challenges, particularly in robotics. Existing robotics benchmarks, however, primarily focus on the performance of individual artifacts, such as policies or controllers, offering limited coverage of coding agents' broader engineering capabilities. Real-world robotics extends beyond control: agents must build, integrate, diagnose, and improve heterogeneous artifacts under resource constraints and reason from multimodal feedback. To evaluate these broader capabilities, we introduce RLE-Bench, a benchmark of robot-learning tasks spanning four representative robotics development workflows: interactive control, policy learning, perception and estimation, and mechanical design. We use diverse task-specific metrics to evaluate the artifacts submitted by the coding agents, from the success rate the agents achieved to the policy agents trained, the harness agent built, and the mechanical structures the agent designed. We aggregate these metrics into an overall RLE Index and report workflow-specific capability profiles, enabling systematic comparison of coding agents' capabilities across multiple capability dimensions. Beyond performance ranks, we also conduct in-depth case studies examining agent behavior on representative tasks, highlighting both current capabilities and limitations, and pointing to the opportunities robotics tasks have to offer for future agent training.
Chinese Translation
编码代理开始超越纯数字任务，面临物理世界的挑战，特别是在机器人领域。然而，现有的机器人基准主要关注单个工件的性能，如策略或控制器，提供的覆盖面有限，无法全面反映编码代理的更广泛工程能力。现实世界的机器人技术不仅仅是控制：代理必须在资源约束下构建、集成、诊断和改进异构工件，并从多模态反馈中进行推理。为了评估这些更广泛的能力，我们引入了RLE-Bench，这是一个涵盖四个代表性机器人开发工作流程的机器人学习任务基准：交互控制、策略学习、感知与估计，以及机械设计。我们使用多样的任务特定指标来评估编码代理提交的工件，从代理所达到的成功率到训练的策略、构建的装置以及设计的机械结构。我们将这些指标汇总为一个整体的RLE指数，并报告特定工作流程的能力概况，从而实现对编码代理在多个能力维度上的系统比较。除了性能排名外，我们还进行深入的案例研究，考察代理在代表性任务上的行为，突出当前的能力和局限性，并指出机器人任务为未来代理训练提供的机会。
cs.RO / 118 / 2609.34220
mmHRI: Towards Privacy-Preserving Human-Robot Interaction with Millimeter-Wave Radar
mmHRI：朝着隐私保护的人机交互与毫米波雷达的结合
Abstract
Assistive robots increasingly operate in many human-centered environments and perform various human-robot interaction (HRI) tasks, such as object delivery. However, most existing HRI systems rely on RGB cameras that continuously observe humans to respond to non-verbal commands, such as hand gestures. This raises privacy concerns in privacy- critical environments, such as hospital wards or restaurants, where direct camera observation of humans is restricted. To develop privacy-preserving HRI, we leverage millimeter-wave (mmWave) radar, which can sense human motion through privacy barriers without identifiable imagery. We propose mmHRI, the first multi-modal robot manipulation framework that achieves mmWave radar-guided privacy-preserving HRI. mmHRI introduces two key designs to mitigate the sparsity and temporal inconsistency of radar data in cluttered robot manipulation environments. First, we propose a dual-stream architecture that jointly learns from unfiltered raw radar tensors and radar point clouds to estimate both human actions and 3D poses. To mitigate signal inconsistency, mmHRI further incorporates a memory-based state-space model (MSSM) that retains historical radar features to reduce abrupt changes in pose/action. These estimated human states are then converted into structured textual robot instructions, which control a vision-language-action (VLA) policy for closed-loop robot manipulation and human-aware reactions. Our evaluation covers human action recognition and closed-loop delivery and retrieval. In the privacy-preserving curtain setting, mmHRI achieves 85.09% action-recognition accuracy, outperforming existing radar-based alternatives. Robot trials further demonstrate successful delivery and retrieval under visual occlusion, with stable task performance across unseen subjects, clutter configurations, and environments.
Chinese Translation
辅助机器人越来越多地在许多人为中心的环境中操作，并执行各种人机交互（HRI）任务，例如物体递送。然而，大多数现有的人机交互系统依赖于RGB摄像头，持续观察人类以响应非语言指令，如手势。这在隐私至关重要的环境中引发了隐私担忧，例如医院病房或餐馆，在这些地方对人类的直接摄像头观察受到限制。为了开发隐私保护的人机交互，我们利用毫米波（mmWave）雷达，该雷达可以通过隐私屏障感知人类运动，而无需可识别的图像。我们提出了mmHRI，这是第一个实现毫米波雷达引导的隐私保护人机交互的多模态机器人操作框架。mmHRI引入了两个关键设计，以减轻杂乱机器人操作环境中雷达数据的稀疏性和时间不一致性。首先，我们提出了一种双流架构，该架构共同学习未经过滤的原始雷达张量和雷达点云，以估计人类的动作和三维姿态。为了减轻信号不一致性，mmHRI进一步结合了一种基于记忆的状态空间模型（MSSM），该模型保留历史雷达特征，以减少姿态/动作的突变。这些估计的人类状态随后被转换为结构化的文本机器人指令，控制视觉-语言-动作（VLA）策略，以实现闭环机器人操作和人类感知反应。我们的评估涵盖人类动作识别和闭环递送与检索。在隐私保护的帷幕设置中，mmHRI实现了85.09%的动作识别准确率，优于现有的基于雷达的替代方案。机器人试验进一步证明了在视觉遮挡下成功的递送和检索，且在未见过的对象、杂乱配置和环境中保持稳定的任务表现。
cs.RO / 119 / 2609.34222
Proprioceptive Force Estimation for Quadruped Locomotion and Human-Robot Interaction
四足动物运动与人机交互中的本体感觉力估计
Abstract
Payload forces must be accommodated during locomotion, while leash forces can specify desired motion. We investigate whether a shared three-dimensional force estimate in newtons, inferred from proprioceptive history under sustained loading, can support both tasks. An estimator and locomotion policy are jointly trained with supervised force and velocity outputs and learned latent context. The estimated force conditions locomotion and additionally generates planar-velocity and yaw-rate commands for leash guidance through an analytical map. In sustained-force simulation sweeps, temporal means of componentwise force root mean square error range from 1.44 to 2.83\,N. Compared with a domain-randomized baseline, the framework reduces velocity-tracking and base-orientation error scores by 21.6\% and 46.5\%, respectively, and increases mean survival from 68.29\% to 94.60\% in separate sustained-force tests. Unitree Go1 experiments demonstrate stationary vertical and horizontal force estimation, locomotion with an 8.5\,kg payload whose weight exceeds the 70\,N training force limit, and leash guidance using the same force-estimation interface.
Chinese Translation
在运动过程中，必须适应负载力，而牵引力则可以指定期望的运动。我们研究了在持续加载下，从本体感觉历史推断出的以牛顿为单位的共享三维力估计是否能够支持这两项任务。一个估计器和运动政策通过监督的力和速度输出以及学习的潜在上下文进行联合训练。估计的力条件控制运动，并通过分析映射生成平面速度和偏航率命令以指导牵引。在持续力模拟测试中，分量力均方根误差的时间均值范围为1.44到2.83 N。与领域随机化基线相比，该框架分别减少了21.6%和46.5%的速度跟踪和基座方向误差得分，并在单独的持续力测试中将平均生存率从68.29%提高到94.60%。Unitree Go1实验展示了静态垂直和水平力估计，承载超过70 N训练力限制的8.5 kg负载的运动，以及使用相同力估计接口的牵引指导。
cs.RO / 120 / 2609.34233
GAE: General Action Expert for Real-Time Humanoid Teleoperation
GAE：实时类人机器人遥操作的通用动作专家
Abstract
Humanoid avatars extend human physical presence beyond the body, enabling people to participate in social, service, and labor activities through remotely operated robots. This requires teleoperation systems capable of realizing diverse and dynamic whole-body behaviors while maintaining responsive human-robot synchronization. We present General Action Expert(GAE), a unified learning framework for general-purpose, low-latency humanoid whole-body teleoperation. To cover diverse human behaviors, GAE builds a large-scale human motion dataset from heterogeneous sources, including videos, animations, and motion capture, followed by standardization and augmentation. GAE then addresses the noise and embodiment mismatch in human motions with a two-stage training paradigm: a privileged generator policy first tracks human motion references in simulation and rolls out feasible humanoid trajectories; a deployable executor policy then learns to track these generated trajectories under curriculum domain randomization. For responsive human-robot synchronization, GAE introduces a latency-conditioned anticipation mechanism that adaptively compensates for end-to-end delay during real-time teleoperation. Simulation and real-world experiments on Unitree G1 and Westlake O1 robots demonstrate that GAE enables humanoids to smoothly mirror diverse, agile, and expressive human behaviors. Project website: https://wangyf0928.github.io/gae-wlrobotics/
Chinese Translation
类人化身扩展了人类的物理存在，使人们能够通过远程操作的机器人参与社交、服务和劳动活动。这需要能够实现多样化和动态的全身行为，同时保持人机同步响应的遥操作系统。我们提出了通用动作专家（General Action Expert，GAE），这是一个用于通用、低延迟类人全身遥操作的统一学习框架。为了覆盖多样的人类行为，GAE 从异构来源构建了一个大规模的人类运动数据集，包括视频、动画和动作捕捉，随后进行了标准化和增强。GAE 通过两阶段训练范式解决了人类运动中的噪声和体现不匹配问题：特权生成策略首先在仿真中跟踪人类运动参考，并推出可行的类人轨迹；可部署执行策略随后在课程领域随机化下学习跟踪这些生成的轨迹。为了实现响应式的人机同步，GAE 引入了一种延迟条件的预期机制，能够自适应地补偿实时遥操作中的端到端延迟。在 Unitree G1 和 Westlake O1 机器人上的仿真和现实世界实验表明，GAE 使类人机器人能够平滑地模仿多样、灵活和富有表现力的人类行为。项目网站：https://wangyf0928.github.io/gae-wlrobotics/
cs.RO / 121 / 2609.34250
WAM-OPD: Sharpening World Action Models via On-Policy Distillation
WAM-OPD：通过在线蒸馏提升世界动作模型的性能
Abstract
Pretrained world action models (WAMs) provide generalist capabilities across diverse robotic manipulation tasks, yet improving target-task performance to an expert level without degrading pretrained skills remains challenging. We explore on-policy distillation (OPD) for WAMs and introduce WAM-OPD. WAM-OPD inherits the advantage of OPD methods that transfer task-specific teacher knowledge under the student's own induced distribution, rather than directly fitting the student to a narrow task-specific data distribution. However, in closed-loop manipulation, the observation histories change as the student policy evolves, requiring fresh environment rollouts to remain on-policy. Applying OPD to WAMs entails repeated data collection, which is costly even in simulation and often impractical on real robots. To avoid repeated environment rollouts during distillation, we introduce prefix-weighted trajectory replay (PWTR). PWTR uses a fixed trajectory pool composed primarily of initial-student rollouts, supplemented with task-specific teacher rollouts to broaden trajectory coverage. For each trajectory replayed from this pool, PWTR conditions the current policy on successive stored histories to generate fresh denoising paths, along which the task-specific teacher provides supervision. Although these denoising paths are refreshed as the policy evolves, the replayed environment trajectories remain fixed. PWTR therefore reweights per-decision distillation losses using proxy importance weights derived from path scores accumulated over the trajectory prefix preceding each decision to mitigate the resulting shift in the history distribution. Simulated and real-world experiments demonstrate task adaptation without additional environment interaction during distillation. In both settings, WAM-OPD improves target-task performance while retaining near-initial performance on tasks excluded from adaptation.
Chinese Translation
预训练的世界动作模型（WAMs）在多样化的机器人操作任务中提供了通用能力，但在不降低预训练技能的情况下将目标任务性能提升至专家水平仍然具有挑战性。我们探索了针对WAM的在线蒸馏（OPD），并引入了WAM-OPD。WAM-OPD继承了OPD方法的优势，该方法在学生自身诱导的分布下转移任务特定的教师知识，而不是直接将学生拟合到狭窄的任务特定数据分布中。然而，在闭环操作中，观察历史会随着学生策略的演变而变化，这要求进行新的环境回合以保持在线策略。在WAM上应用OPD需要重复的数据收集，即使在模拟中也成本高昂，且在真实机器人上往往不切实际。为了避免在蒸馏过程中重复环境回合，我们引入了前缀加权轨迹重放（PWTR）。PWTR使用一个主要由初始学生回合组成的固定轨迹池，并辅以任务特定的教师回合以扩大轨迹覆盖范围。对于从该池中重放的每个轨迹，PWTR将当前策略条件化于连续存储的历史，以生成新的去噪路径，沿着这些路径，任务特定的教师提供监督。尽管这些去噪路径会随着策略的演变而更新，但重放的环境轨迹保持不变。因此，PWTR使用从每个决策前的轨迹前缀累积的路径得分导出的代理重要性权重，重新加权每个决策的蒸馏损失，以减轻历史分布的变化。模拟和真实世界的实验表明，在蒸馏过程中无需额外的环境交互即可实现任务适应。在这两种设置中，WAM-OPD在保持接近初始性能的同时改善了目标任务的表现，而这些任务并未被纳入适应过程。
cs.RO / 122 / 2609.34256
UMR: Universal Manipulation Representation
UMR：通用操作表示
Abstract
General-purpose embodied manipulation hinges on a unified action representation that generalizes across embodiments and scales readily. Yet existing policies rely on embodiment-specific action spaces, making cross-embodiment demonstrations difficult to leverage at scale and limiting transfer to new embodiments and spatial variations. To this end, we introduce Universal Manipulation Representation (UMR), a unified action representation that enables zero-shot skill transfer from human demonstrations to heterogeneous robots. UMR decomposes manipulation into two functionally distinct yet geometrically linked components: embodiment-agnostic World Flow, which describes task-relevant object motion in the world frame, and Ego Trajectory, which represents end-effector motion relative to the current pose. We instantiate UMR as World--Ego Point VLA (WEPVLA), a compact 0.5B-parameter policy that learns in the unified geometric action space through a dual-stream Point Action Adapter and a unified Point Action Expert, with an $SE(3)$ conjugation coupling the two components. To improve data efficiency, we complement UMR with a Data-Efficient Strategy (DES) that diversifies object configurations through stage-aware point-cloud editing while preserving demonstrated contact geometry. In simulation, WEPVLA achieves average success rates of 97.5\% on LIBERO and 85.7\% on the 10-task RLBench benchmark. In real-world experiments, a single policy trained on human demonstrations augmented by DES transfers zero-shot to diverse deployment conditions. With about 10 minutes of collected human demonstrations per task and no robot demonstrations, it achieves 91.7\% average success across six evaluation settings, compared with 60.8\% for HumanEgo. Code and additional materials are available at https://umr-wepvla.github.io/.
Chinese Translation
通用的具身操作依赖于一个统一的动作表示，该表示能够跨具身体和规模进行泛化。然而，现有策略依赖于特定于具身体的动作空间，使得跨具身体的演示难以大规模利用，并限制了向新具身体和空间变体的迁移。为此，我们提出了通用操作表示（Universal Manipulation Representation, UMR），这是一种统一的动作表示，能够实现从人类演示到异构机器人之间的零-shot技能迁移。UMR将操作分解为两个功能上独立但在几何上相关的组件：与具身体无关的世界流（World Flow），描述了在世界坐标系中与任务相关的物体运动，以及自我轨迹（Ego Trajectory），表示相对于当前姿态的末端执行器运动。我们将UMR实例化为世界-自我点VLA（World--Ego Point VLA, WEPVLA），这是一个紧凑的0.5B参数策略，通过双流点动作适配器和统一的点动作专家在统一的几何动作空间中进行学习，并通过$SE(3)$共轭将两个组件耦合。为了提高数据效率，我们将UMR与数据高效策略（Data-Efficient Strategy, DES）相结合，通过阶段感知的点云编辑多样化物体配置，同时保持演示的接触几何。在仿真中，WEPVLA在LIBERO上实现了97.5\%的平均成功率，在10任务RLBench基准上实现了85.7\\%。在真实世界实验中，单一策略在由DES增强的人类演示上训练，能够零-shot迁移到多样的部署条件。每个任务收集约10分钟的人类演示且没有机器人演示，它在六个评估设置中实现了91.7\\%的平均成功率，而HumanEgo为60.8\\%。代码和其他材料可在https://umr-wepvla.github.io/获取。
cs.RO / 123 / 2609.34261
RoboICL: Embodied In-Context Learning with GPT-6 Astra
RoboICL：基于GPT-6 Astra的具身上下文学习
Abstract
General-purpose vision-language models offer a promising way to zero-shot robot control: \gptastra{} excels at open-ended and language- or image-conditioned manipulation but remains substantially weaker on high-precision and long-horizon tasks. We introduce \emph{RoboICL}, an in-context robot-control framework that narrows these gaps without robot-specific parameter updates or a learned VLA. RoboICL separates \emph{demonstration context}, which provides recorded examples when available, from \emph{interaction memory}, which accumulates the model's own actions and observed outcomes. Both use a shared observation--action--receipt--observation grammar. To preserve experience across task stages, RoboICL combines sampled demonstration blocks with bounded anchored memory. Fixed anchors keep earlier rollout interactions available for in-context learning, while the latest interaction supports immediate error correction. Across 30 RoboDojo tasks, using zero shot for Open and one demonstration elsewhere, RoboICL improves on official zero-shot \gptastra{} by 20--27 progress-score points in every category. It leads the leaderboard baselines on Memory and Open, achieves comparable performance to the strongest Precision baseline, and remains competitive on Long-Horizon. Its 30-task Overall score is 50.64, versus 33.68 for the strongest baseline. On a separate ten-task subset, RoboICL scores 60.60, within 2.00 points of the $\pi_{0.5}$ + \gptastra{} hybrid approach. On three real-robot tasks, mean progress rises from 14.45 at zero shot to 63.33 at one shot and 78.89 at three shots. On two development tasks, optional Jev-gated action reuse reduces \gptastra{} calls by 33--48\%. Code is available at \href{https://github.com/Mosi-AI/RoboICL}{https://github.com/Mosi-AI/RoboICL}.
Chinese Translation
通用视觉-语言模型为零-shot机器人控制提供了一种有前景的方法： exttt{gptastra}在开放式和语言或图像条件下的操作中表现出色，但在高精度和长时间任务上仍然显著较弱。我们提出了 extit{RoboICL}，一个上下文机器人控制框架，它在不进行机器人特定参数更新或学习的视觉-语言-动作（VLA）模型的情况下缩小了这些差距。RoboICL将 extit{演示上下文}（在可用时提供记录示例）与 extit{交互记忆}（累积模型自身的动作和观察结果）分开。两者都使用共享的观察-动作-反馈-观察语法。为了在任务阶段之间保留经验，RoboICL结合了采样的演示块和有界的锚定记忆。固定的锚点使早期的回放交互可用于上下文学习，而最新的交互支持即时错误修正。在30个RoboDojo任务中，使用零-shot进行开放任务，其他地方使用一个演示，RoboICL在每个类别中比官方零-shot的 exttt{gptastra}提高了20到27个进展分数点。它在记忆和开放任务的排行榜基准中领先，达到与最强精度基准相当的性能，并在长时间任务中保持竞争力。其30任务的总体得分为50.64，而最强基准为33.68。在一个单独的十任务子集中，RoboICL得分为60.60，距离$ extpi_{0.5}$ + exttt{gptastra}混合方法仅有2.00分。在三个真实机器人任务中，平均进展从零-shot的14.45上升到一-shot的63.33和三-shot的78.89。在两个开发任务中，可选的Jev门控动作重用将 exttt{gptastra}的调用减少了33%至48%。代码可在 exttt{https://github.com/Mosi-AI/RoboICL}获取。
cs.RO / 124 / 2609.34268
SAGE: Symbolic Action-Gating and Editing for LLM Task Planners
SAGE：用于大型语言模型任务规划的符号动作门控与编辑
Abstract
Large language models (LLMs) are now the default cognitive core of embodied household agents, yet the plans they emit are rarely checked against a grounded model of the environment before execution, and the task-success they report is often measured on benchmarks so saturated that no method can be separated from another. We present SAGE (Symbolic Action-Gating and Editing), a single-LLM planner built from two lightweight mechanisms: a domain-agnostic symbolic gate (~250 lines of Python, zero tokens, $O(|\pi|)$) that blocks precondition-violating actions with typed reasons as a runtime safety monitor, and a local edit that regenerates only the failed sub-goal's suffix, keeping completed and untouched work intact; a hybrid seed+live memory store supports cold-start coverage. We evaluate under a leak-free protocol (leave-one-out retrieval) over five open-weight models and a 75-task AI2-THOR benchmark. On the standard benchmark goal-completeness saturates (52% of instances trivially solved) and SAGE ties strong hierarchical baselines. On a harder, method-agnostic multi-goal composition, SAGE's completeness lead re-emerges large (+0.06 to +0.23 across four models). Under injected mid-execution failures, SAGE recovers as reliably as whole-plan replanners at 2.4-3.3x fewer LLM calls. As a verify-before-execute gate, the symbolic monitor blocks unsafe actions before actuation and raises simulator-reported step-success for every planner tested (up to +0.11), a signal the verifier never sees (non-circular). Because the gate calls no model (0.008 ms/plan), it is a safety layer that runs essentially free on the edge: SAGE planning reproduces its quality on a Jetson AGX Orin, where small-model verification helps most. We release the benchmark, the leak-free protocol, the recovery and safety-gate harnesses, and a verifier-portability study (auto-induced on ALFWorld, 0.89 held-out).
Chinese Translation
大型语言模型（LLMs）现在是具身家居代理的默认认知核心，但它们生成的计划在执行前很少与环境的基础模型进行核对，而它们报告的任务成功率往往是在高度饱和的基准上测量的，以至于没有任何方法能够与其他方法区分开。我们提出了SAGE（Symbolic Action-Gating and Editing），这是一个基于两个轻量级机制构建的单LLM规划器：一个领域无关的符号门（约250行Python代码，零令牌，$O(| ext{π}|)$），作为运行时安全监控器，阻止违反前置条件的动作，并提供类型化的理由；一个局部编辑机制仅重新生成失败子目标的后缀，保持已完成和未触及的工作不变；一个混合的种子+实时内存存储支持冷启动覆盖。我们在五个开放权重模型和75任务的AI2-THOR基准上，在无泄漏协议（留一法检索）下进行评估。在标准基准上，目标完成率饱和（52%的实例被轻松解决），而SAGE与强大的层次基线持平。在一个更困难的、方法无关的多目标组合中，SAGE的完整性领先再次显现出显著提升（在四个模型中增加了+0.06到+0.23）。在注入的中途执行失败下，SAGE以2.4-3.3倍更少的LLM调用，恢复的可靠性与整体计划重规划器相当。作为一个执行前验证门，符号监控器在动作执行前阻止不安全的行为，并提高了每个测试规划器的模拟器报告的步骤成功率（最高增加+0.11），这是验证者从未看到的信号（非循环）。由于该门不调用任何模型（每个计划0.008毫秒），它是一个几乎在边缘上免费运行的安全层：SAGE规划在Jetson AGX Orin上重现其质量，其中小模型验证帮助最大。我们发布了基准、无泄漏协议、恢复和安全门的工具，以及一个验证者可移植性研究（在ALFWorld上自动引发，0.89的保留）。
cs.RO / 125 / 2609.34270
Bayesian Active Learning for Intent Disambiguation in Interactive Robot Planning
用于交互式机器人规划中的意图消歧的贝叶斯主动学习
Abstract
Interactive robot planning requires robots to infer and execute human intentions from natural language instructions that are often ambiguous, incomplete, or underspecified. Although large language models (LLMs) provide a powerful interface for clarification, relying on the generative model to drive an multi-turn conversation can introduce systematic failures. We propose a Bayesian framework that treats clarification as an active learning problem over grounded Signal Temporal Logic (STL) task specifications. Our method uses LLMs to initialize candidate formal specifications and translate informative contrasts into natural-language clarification questions, while Bayesian optimization maintains uncertainty estimation over user intent and selects queries that maximize information gain. After convergence, the inferred STL specification is passed to a formal planner to synthesize a verifiable robot trajectory. Across four simulated and real-world task domains, our approach generally achieves higher task satisfaction and requires fewer clarification rounds than LLM baselines, while helping smaller models close the performance gap against larger reasoning models.
Chinese Translation
交互式机器人规划要求机器人从自然语言指令中推断和执行人类意图，而这些指令通常模糊、不完整或不明确。尽管大型语言模型（LLMs）提供了强大的澄清接口，但依赖生成模型驱动多轮对话可能会引入系统性失败。我们提出了一种贝叶斯框架，将澄清视为基于基础信号时序逻辑（Signal Temporal Logic, STL）任务规范的主动学习问题。我们的方法利用LLMs初始化候选的形式规范，并将信息对比转化为自然语言澄清问题，同时贝叶斯优化保持对用户意图的不确定性估计，并选择最大化信息增益的查询。在收敛后，推断出的STL规范被传递给形式规划器，以合成可验证的机器人轨迹。在四个模拟和真实世界的任务领域中，我们的方法通常实现了更高的任务满意度，并且所需的澄清轮次比LLM基线更少，同时帮助较小的模型缩小与较大推理模型之间的性能差距。
cs.RO / 126 / 2609.34276
NavHarness: Towards Lifelong Embodied Navigation
NavHarness：迈向终身具身导航
Abstract
Frontier models can now perform well on individual embodied navigation tasks through multi-round multimodal reasoning with simple tools. Across successive tasks, however, an agent must also rely on an evolving map and earlier search records, both of which may be incomplete or conflict with new observations. We present NavHarness, a training-free embodied harness towards lifelong navigation that makes memory processing part of the navigation loop. During navigation, its multi-round agentic session draws on maps, task records, and house knowledge, checking them against observations and recording corrections to guide its actions. NavHarness preserves this experience across fresh conversations for new tasks or recovery attempts, while outcome verification and run-end summaries support its later reuse. On GOAT-Bench, NavHarness improves s-SR over context-only independent sessions by 18.6 points with Astra and 22.6 with Opus 5. Using SLAM-estimated poses, NavHarness with GPT-6 Astra achieves state-of-the-art task success of 83.7 s-SR with 36.9 e-SR on GOAT-Bench and 85.9 s-SR on IR2R-CE. To understand these gains, we examine how experience is carried between sessions and find that structured recovery handovers outperform length-matched summaries. In extended deployments across houses, consolidation improves navigation beyond retaining maps and task records, with case studies showing how agents use earlier experience to interpret new goals, investigate unresolved questions, and resume failed searches. We suggest that progress towards lifelong navigation depends on how successive reasoning sessions build on prior experience, alongside improvements in single-task capability.
Chinese Translation
前沿模型现在能够通过多轮多模态推理和简单工具在个体具身导航任务中表现良好。然而，在连续任务中，智能体还必须依赖不断演变的地图和早期搜索记录，而这两者可能不完整或与新的观察结果相冲突。我们提出了NavHarness，这是一种无需训练的具身工具，旨在实现终身导航，使记忆处理成为导航循环的一部分。在导航过程中，其多轮智能会话利用地图、任务记录和背景知识，检查这些信息与观察结果的一致性，并记录修正以指导其行动。NavHarness在新的任务或恢复尝试中保留这些经验，同时结果验证和运行结束摘要支持其后续重用。在GOAT-Bench上，NavHarness在仅依赖上下文的独立会话中，使用Astra提高了18.6分的s-SR，使用Opus 5提高了22.6分。利用SLAM估计的姿态，NavHarness与GPT-6 Astra结合在GOAT-Bench上实现了83.7的s-SR和36.9的e-SR的最新任务成功率，在IR2R-CE上实现了85.9的s-SR。为了理解这些提升，我们研究了经验如何在会话之间传递，发现结构化的恢复交接优于长度匹配的摘要。在跨房屋的扩展部署中，整合不仅改善了导航，还超越了保留地图和任务记录的能力，案例研究显示智能体如何利用早期经验来解释新目标、调查未解决的问题并恢复失败的搜索。我们建议，迈向终身导航的进展取决于连续推理会话如何在先前经验的基础上构建，以及单任务能力的提升。
cs.RO / 127 / 2609.34297
TLC-DiT: Task-Aligned Local Visual Conditioning for Robust Multitask Robot Manipulation
TLC-DiT：用于稳健多任务机器人操作的任务对齐局部视觉条件
Abstract
Language-conditioned robot policies have made clear progress in multitask manipulation, but task-relevant local visual evidence usually stays hidden inside a visual backbone or attention layers. This leaves the policy difficult to inspect and fragile under visual change, two symptoms of a missing explicit, task-aligned local visual channel. We present TLC-DiT, a plug-in extension of the Multitask Diffusion Transformer (DiT) policy that adds explicit task-guided local visual feature maps without changing the diffusion objective or the action-generation process. For each camera view, frozen DINOv2 patch features are modulated by the CLIP task embedding through FiLM and refined by a lightweight CoordConv CNN adapter into smooth spatial maps, which are concatenated with the original global image, language, joint-state, and timestep conditions. On LIBERO, TLC-DiT reaches a 93.5% average success rate, compared with 86.5% for Multitask DiT and 79.25% for SmolVLA. On LIBERO-plus, the total success rate improves from 54.07% to 57.24%, with larger gains under camera, background, and sensor-noise changes. In real-world bimanual tasks, TLC-DiT raises Teabag Putting completion from 44% to 89% while maintaining comparable Match Box Opening performance. Feature-map visualizations confirm that the model attends to task-relevant regions across views and perturbations, providing a direct way to inspect the visual evidence.
Chinese Translation
基于语言的机器人策略在多任务操作中取得了显著进展，但与任务相关的局部视觉证据通常隐藏在视觉主干或注意力层中。这使得策略难以检查，并且在视觉变化下表现脆弱，这两个问题表明缺乏明确的、与任务对齐的局部视觉通道。我们提出了TLC-DiT，这是多任务扩散变换器（Multitask Diffusion Transformer，DiT）策略的一个插件扩展，它在不改变扩散目标或动作生成过程的情况下，添加了明确的任务引导局部视觉特征图。对于每个相机视图，冻结的DINOv2补丁特征通过FiLM与CLIP任务嵌入进行调制，并通过轻量级的CoordConv CNN适配器精炼为平滑的空间图，这些图与原始的全局图像、语言、关节状态和时间步条件进行连接。在LIBERO数据集上，TLC-DiT的平均成功率达到93.5%，而多任务DiT为86.5%，SmolVLA为79.25%。在LIBERO-plus上，总成功率从54.07%提高到57.24%，在相机、背景和传感器噪声变化下取得了更大的提升。在真实世界的双手任务中，TLC-DiT将茶包放置的完成率从44%提高到89%，同时保持了可比的配对盒打开性能。特征图的可视化确认模型在不同视图和扰动中关注与任务相关的区域，提供了一种直接检查视觉证据的方法。
cs.RO / 128 / 2609.34300
When World Models Lie: Adaptive Safety Analysis Under Wrong Imaginations
当世界模型失真：错误想象下的自适应安全分析
Abstract
World models offer a powerful substrate for safety reasoning in high-dimensional robotic systems, but they are also fallible: their predictions can be biased, miscalibrated, or confidently wrong. This creates a central challenge for latent-space safety filters, which often learn Hamilton-Jacobi safety value functions on the dynamics of a world model. If the world model is incorrect, the resulting value function can inherit its errors and produce overconfident safety estimates. Existing latent safety filters often rely on auxiliary signals such as ensemble disagreement or value-target consistency residuals for adaptation, but these signals can remain small even when the world model's predictions deviate from observations. We propose an adaptive latent safety filter that calibrates safety reasoning using directly observed world-model error. Our method uses Adaptive Conformal Inference to construct online uncertainty sets from discrepancies between predicted and observation-inferred latent states, then evaluates safety pessimistically by minimizing the learned value function over these sets. This allows the filter to remain minimally conservative when the world model is accurate, while becoming more cautious when observations reveal model mismatch. We provide a finite-time coverage guarantee for the adaptive uncertainty radius. Through simulation and hardware experiments, we show that our method significantly reduces failures relative to state-of-the-art latent safety filters while preserving task completion.
Chinese Translation
世界模型为高维机器人系统中的安全推理提供了强大的基础，但它们也可能出错：其预测可能存在偏差、校准不当或自信地错误。这为潜在空间安全过滤器带来了一个核心挑战，这些过滤器通常在世界模型的动态上学习哈密顿-雅可比安全价值函数。如果世界模型不正确，生成的价值函数可能会继承其错误并产生过于自信的安全估计。现有的潜在安全过滤器通常依赖于辅助信号，如集成不一致性或价值目标一致性残差进行适应，但即使在世界模型的预测与观察偏离时，这些信号也可能保持较小。我们提出了一种自适应潜在安全过滤器，通过直接观察的世界模型误差来校准安全推理。我们的方法使用自适应保形推断（Adaptive Conformal Inference）从预测与观察推断的潜在状态之间的差异构建在线不确定性集，然后通过在这些集合上最小化学习到的价值函数来悲观地评估安全性。这使得过滤器在世界模型准确时保持最小保守性，而在观察揭示模型不匹配时变得更加谨慎。我们为自适应不确定性半径提供了有限时间覆盖保证。通过仿真和硬件实验，我们展示了我们的方法在保持任务完成的同时，显著减少了相对于最先进的潜在安全过滤器的失败率。
cs.RO / 129 / 2609.34356
Predictive Semantic Safety: From Visual Physical Reasoning to Safety-Critical Control
预测语义安全：从视觉物理推理到安全关键控制
Abstract
Physical interactions can create future hazards that are not apparent from the robot's current geometric surroundings. We present a framework termed Predictive Semantic Safety (PSS), which connects visual physical reasoning to backup-based safety filtering. A vision-language model (VLM) predicts physical events and their timing or directly predicts object displacements. An explicit motion model converts event hypotheses into object trajectories. Split conformal prediction calibrates position errors jointly across specified objects, observation times, and future times; geometric shape bounds convert the resulting position regions into predicted object occupancy. PSS evaluates a prescribed backup maneuver against this occupancy and derives input-affine constraints for minimally modifying the nominal input while preserving backup feasibility under the robot dynamics and input limits. MuJoCo experiments with a Unitree Go1 consider falling fixtures, impact-driven support loss, and contact propagation. PSS achieves a safe episode rate of 99.3%, compared with 43.3% for a Backup Control Barrier Function baseline that only uses current obstacle geometry.
Chinese Translation
物理交互可能会产生未来的危险，这些危险在机器人当前的几何环境中并不明显。我们提出了一种框架，称为预测语义安全（Predictive Semantic Safety, PSS），它将视觉物理推理与基于备份的安全过滤相连接。视觉-语言模型（Vision-Language Model, VLM）预测物理事件及其时机，或直接预测物体位移。显式运动模型将事件假设转换为物体轨迹。分裂共形预测（Split Conformal Prediction）在指定的物体、观察时间和未来时间上联合校准位置误差；几何形状边界将结果位置区域转换为预测的物体占用。PSS评估规定的备份机动与该占用情况，并推导出输入仿射约束，以最小修改名义输入的方式保持备份在机器人动力学和输入限制下的可行性。MuJoCo实验使用Unitree Go1考虑了跌落装置、冲击驱动的支撑丧失和接触传播。与仅使用当前障碍几何的备份控制屏障函数基线相比，PSS实现了99.3%的安全事件率，而基线的安全事件率为43.3%。
cs.RO / 130 / 2609.34362
FutureDuet: Decoupling Observation Access from Future Supervision in World Action Models
FutureDuet：在世界动作模型中将观察访问与未来监督解耦
Abstract
World Action Models (WAMs) augment robot action generation with future visual supervision. Existing WAMs commonly fuse main and wrist observations into one visual stream and train both with the same future-video objective, despite their different visual dynamics. A stable main camera reveals scene-level task evolution, whereas wrist cameras move with the end effector, mixing local interaction changes with viewpoint shifts and self-occlusion. These contrasting predictive demands suggest that the two views may benefit from different future objectives. We introduce FutureDuet, which retains both views for control, while allowing each visual stream to receive a different future objective. For the main view, future RGB models task evolution, while interaction masks and robot skeletons focus supervision on task objects and robot motion. For the wrist stream, future latent prediction models short-horizon interaction changes without requiring pixel-level reconstruction. ActionDiT jointly reads the resulting Task State and Interaction State, combining scene-level progress with close-range interaction evidence. All auxiliary prediction modules are training-only, adding no inference overhead. FutureDuet achieves 94.2% clean and 94.1% randomized success on RoboTwin50 and 99.2% average success on LIBERO. The improvements are most pronounced on six RoboTwin50 tasks that require precise interaction, averaging gains of 9.2% and 12.8% over Fast-WAM in clean and randomized settings. Controlled studies further show complementary gains from separating the wrist pathway and designing future supervision separately for the two views.
Chinese Translation
世界动作模型（WAMs）通过未来视觉监督增强机器人动作生成。现有的WAMs通常将主摄像头和腕部摄像头的观察融合为一个视觉流，并用相同的未来视频目标进行训练，尽管它们具有不同的视觉动态。稳定的主摄像头揭示了场景级任务演变，而腕部摄像头则随末端执行器移动，混合了局部交互变化与视角变化和自遮挡。这些对比的预测需求表明，这两种视角可能受益于不同的未来目标。我们提出了FutureDuet，它保留了两种视角用于控制，同时允许每个视觉流接收不同的未来目标。对于主视角，未来RGB模型关注任务演变，而交互掩码和机器人骨架则将监督重点放在任务对象和机器人运动上。对于腕部流，未来潜在预测模型在不需要像素级重建的情况下建模短期交互变化。ActionDiT共同读取生成的任务状态和交互状态，将场景级进展与近距离交互证据相结合。所有辅助预测模块仅用于训练，不增加推理开销。FutureDuet在RoboTwin50上实现了94.2%的干净成功率和94.1%的随机成功率，在LIBERO上实现了99.2%的平均成功率。改进在六个需要精确交互的RoboTwin50任务中最为明显，在干净和随机设置下平均提高了9.2%和12.8%相较于Fast-WAM。控制研究进一步表明，分离腕部路径和为两种视角单独设计未来监督带来了互补的收益。
cs.RO / 131 / 2609.34384
RoboIRGBench: Benchmarking Implicit Referential Grounding in Vision-Language-Action Models
RoboIRGBench：在视觉-语言-动作模型中基准测试隐式指称基础
Abstract
Vision-Language-Action (VLA) models have shown strong capabilities in robotic manipulation, yet existing benchmarks typically assume that task-relevant information is explicitly specified in the instruction. In practice, however, humans frequently refer to objects, quantities, and relations implicitly, requiring robots to recover the intended target from linguistic and perceptual context. We study this capability as Implicit Referential Grounding (IRG) and introduce RoboIRG-Bench, a manipulation benchmark designed to systematically evaluate it. Built upon RoboMME, RoboIRG-Bench contains 40 variants derived from 11 tasks and covers four challenges, including direct, reasoning-mediated, spatial, and contextual referential grounding. As IRG often requires retaining and retrieving previously established context, we evaluate representative VLAs spanning different memory mechanisms. Our evaluation reveals a noticeable referential robustness gap. Models that perform well under explicit instructions can degrade sharply when the same task-relevant information must be recovered from context. Reasoning-mediated and spatial references are particularly challenging, while models using external VLMs show greater robustness but still exhibit significant failures. Moreover, replacing the external VLM with a stronger model does not eliminate these gaps. We further validate these findings on a Franka Research 3 robot arm, where the gap persists under real-world manipulation and manifests as both incorrect referent grounding and downstream execution failures. These results establish IRG as a distinct and underexplored capability for reliable robotic instruction following and highlight the need for VLAs that can robustly integrate language, perception, reasoning, and action.
Chinese Translation
视觉-语言-动作（VLA）模型在机器人操作中展现了强大的能力，但现有基准通常假设任务相关信息在指令中是明确指定的。然而，在实践中，人类经常隐式地提及物体、数量和关系，这要求机器人从语言和感知上下文中恢复预期的目标。我们将这种能力研究为隐式指称基础（IRG），并引入RoboIRG-Bench，这是一个旨在系统评估该能力的操作基准。RoboIRG-Bench建立在RoboMME基础上，包含从11个任务衍生的40个变体，并涵盖四个挑战，包括直接、推理介导、空间和上下文指称基础。由于IRG通常需要保留和检索先前建立的上下文，我们评估了跨不同记忆机制的代表性VLA。我们的评估揭示了明显的指称鲁棒性差距。在明确指令下表现良好的模型在必须从上下文中恢复相同任务相关信息时，性能可能急剧下降。推理介导和空间引用尤其具有挑战性，而使用外部视觉语言模型（VLM）的模型显示出更大的鲁棒性，但仍然表现出显著的失败。此外，用更强的模型替换外部VLM并不能消除这些差距。我们进一步在Franka Research 3机器人手臂上验证了这些发现，在实际操作中，这一差距依然存在，并表现为不正确的指称基础和后续执行失败。这些结果确立了IRG作为可靠机器人指令跟随的一种独特且未被充分探索的能力，并强调了需要能够稳健整合语言、感知、推理和行动的VLA。
cs.RO / 132 / 2609.34400
Contact-Aware Impedance Controller for Robot-Assisted Ultrasound Imaging
面向接触感知的阻抗控制器用于机器人辅助超声成像
Abstract
Safe robot-assisted ultrasound imaging requires a reliable controller able to detect and localize probe--tissue interaction. In this paper, we present a B-mode ultrasound image-based contact perception method and a contact-aware impedance controller for robotic ultrasound imaging. The proposed method detects acoustic contact independently of force measurements, enabling contact-conditioned force/torque taring to reduce residual wrench bias. During contact, the method continuously estimates the effective contact location along the curved probe surface and uses it to update the controller interaction frame, enabling visual servoing of the physical probe--tissue contact point during imaging. Experiments on an agar phantom demonstrated a contact-localization RMSE of $\mathbf{1.46 \pm 0.14}$~mm over probe roll angles from $\mathbf{-15^\circ}$ to $\mathbf{15^\circ}$. During static rolling, the proposed controller maintained task-space tracking accuracy comparable to a conventional fixed-frame impedance controller while reducing the maximum compressive interaction force from $\mathbf{31.56}$~N to $\mathbf{20.09}$~N, corresponding to a $\mathbf{36.3\%}$ reduction. These results demonstrate the potential of ultrasound images as direct contact feedback for safe and accurate robot-assisted ultrasound imaging.
Chinese Translation
安全的机器人辅助超声成像需要一个可靠的控制器，能够检测和定位探头与组织的相互作用。本文提出了一种基于B模式超声图像的接触感知方法和一个面向接触感知的阻抗控制器，用于机器人超声成像。所提方法独立于力测量检测声学接触，使得接触条件下的力/扭矩归零成为可能，从而减少残余扭矩偏差。在接触过程中，该方法持续估计沿着曲面探头的有效接触位置，并利用该位置更新控制器的交互框架，从而在成像过程中实现物理探头与组织接触点的视觉伺服。对琼脂幻影的实验表明，在探头滚转角度从-15°到15°范围内，接触定位的均方根误差为1.46 ± 0.14 mm。在静态滚动过程中，所提控制器保持了与传统固定框架阻抗控制器相当的任务空间跟踪精度，同时将最大压缩相互作用力从31.56 N降低至20.09 N，减少幅度为36.3%。这些结果表明超声图像作为直接接触反馈在安全和准确的机器人辅助超声成像中的潜力。
cs.RO / 133 / 2609.34412
From Language to Task Maps: Compiling Semantic Relations While Preserving Task-Relevant Freedom
从语言到任务地图：在保持任务相关自由的同时编译语义关系
Abstract
Natural-language manipulation instructions specify qualitative relations, whereas continuous controllers require state-evaluable task quantities, differentials, and completion conditions. Because a qualitative relation generally leaves part of the relative configuration unspecified, expanding it into a complete pose can introduce unintended constraints. We present a typed semantic-to-geometric interface in which language specifies entities, relations, and phases, while each relation indexes a registered specification of its task-relevant distinctions and preserved freedoms. A robot-side compiler grounds these specifications, constructs relation-specific task maps and consistent differentials using conformal geometric algebra, and composes the resulting policies through RMPflow. To evaluate the division of responsibility between the language model and the compiler, we compared a Semantic Topology interface with one that additionally requires relation-specific geometric specifications over 60 instructions. Both produced correct shared semantic content in 41/60 cases, but critical errors under their respective interface requirements occurred in 19/60 and 58/60 cases. Across 64 grounded evaluations spanning eight geometric relation forms, the task maps preserved registered null directions and responded to relation-relevant perturbations; analytic directional derivatives agreed with finite differences, and Jacobian ranks matched the registered dimensions. In three closed-loop ablations using a simulated Franka Emika Panda in MuJoCo, fixing a relation-preserved coordinate increased median terminal progress error by 20.24--71.00~mm while the retained relation errors remained within their evaluation bounds. These results support compiling relation-visible geometry and preserved freedom together into composable continuous objectives.
Chinese Translation
自然语言操作指令指定了定性关系，而连续控制器则需要可评估状态的任务量、微分和完成条件。由于定性关系通常会留下部分相对配置未指定，将其扩展为完整的姿态可能会引入意想不到的约束。我们提出了一种类型化的语义-几何接口，其中语言指定实体、关系和阶段，而每个关系则索引其任务相关的区分和保留自由的注册规范。机器人端编译器将这些规范落实，利用共形几何代数构建关系特定的任务地图和一致的微分，并通过RMPflow组合生成的策略。为了评估语言模型与编译器之间的责任分配，我们比较了一个语义拓扑接口与一个额外要求关系特定几何规范的接口，涵盖了60条指令。两者在41/60的案例中产生了正确的共享语义内容，但在各自接口要求下分别出现了19/60和58/60的关键错误。在跨越八种几何关系形式的64次基础评估中，任务地图保留了注册的零方向，并对与关系相关的扰动作出响应；解析方向导数与有限差分一致，雅可比矩阵的秩与注册维度相匹配。在使用MuJoCo模拟Franka Emika Panda进行的三次闭环消融实验中，固定一个保留关系的坐标使得中位终端进展误差增加了20.24--71.00毫米，而保留的关系误差仍在其评估范围内。这些结果支持将可见的关系几何和保留自由编译为可组合的连续目标。
cs.RO / 134 / 2609.34414
From World Models to World Action Models: Rethinking Next-State Prediction
从世界模型到世界行动模型：重新思考下一状态预测
Abstract
Predicting the next state is a core paradigm of World Models for modeling physical dynamics, emphasizing prediction fidelity. As World Models evolve into World-Action Models (WAMs), existing methods still fix the next state before training as RGB, a single latent feature, or a static combination of predefined targets, thereby constraining action learning to the inductive biases preserved by a particular representation. To address this limitation, we propose CF-WAM, a dynamic next-state prediction framework that samples visual, semantic, geometric, and interaction projections of the same future, standardizes them into a common video form, and supervises a unified WAM across these projections. The action-relevant constraints exposed by these projections accumulate across training steps, forcing WAM to capture the underlying state-transition structure that supports multiple projections of the same action-conditioned future. This dynamic mechanism also provides a natural cross-embodiment dynamics reference frame for Human and Robot learning. By jointly learning across different next-state parameterizations, heterogeneous Human and Robot experience can bypass appearance differences and directly contribute to shared state-transition learning, improving cross-embodiment generalization. Experiments show that CF-WAM improves both training efficiency and final control performance, while translating Human experience effectively into policy gains. CF-WAM achieves state-of-the-art performance on RoboCasa-GR1 with an average success rate of 82.50%, while reaching 82.65% on LIBERO-Plus and up to 84.00% in real-world evaluations.
Chinese Translation
预测下一状态是世界模型（World Models）用于建模物理动态的核心范式，强调预测的准确性。随着世界模型演变为世界行动模型（World-Action Models, WAMs），现有方法仍然在训练之前将下一状态固定为RGB、单一潜在特征或预定义目标的静态组合，从而限制了行动学习受到特定表示所保留的归纳偏见的约束。为了解决这一限制，我们提出了CF-WAM，一种动态下一状态预测框架，它对同一未来的视觉、语义、几何和交互投影进行采样，将其标准化为共同的视频形式，并在这些投影上监督一个统一的WAM。这些投影所暴露的与行动相关的约束在训练步骤中不断累积，迫使WAM捕捉支持同一行动条件未来的潜在状态转移结构。这种动态机制还为人类和机器人学习提供了自然的跨体现动态参考框架。通过跨不同的下一状态参数化进行联合学习，异构的人类和机器人经验可以绕过外观差异，直接为共享的状态转移学习做出贡献，从而提高跨体现的泛化能力。实验表明，CF-WAM提高了训练效率和最终控制性能，同时有效地将人类经验转化为策略收益。CF-WAM在RoboCasa-GR1上实现了82.50%的平均成功率，达到LIBERO-Plus的82.65%，并在真实世界评估中高达84.00%的表现。
cs.RO / 135 / 2609.34484
ARS: Agentic Reward System for Robot Learning
ARS：用于机器人学习的自主奖励系统
Abstract
Progress reward modeling is the problem of estimating how a robot's behavior changes task progress over time. Reliable estimation requires distinguishing meaningful state changes from failed attempts and task-irrelevant actions. We introduce the Agentic Reward System (ARS), an inference framework for progress reward modeling with general-purpose vision-language models (VLMs), without additional reward-model training. Given an offline trajectory and a task instruction, ARS uses adaptive visual inspection for both event proposal and verification. A subagent proposes a task-relevant event timeline, which a primary agent verifies and revises before estimating per-frame progress. ARS can incorporate optional terminal outcome labels and visual references to inform its judgments. It can also audit progress estimates from external reward models. We evaluate ARS with a 27B VLM on a controlled semantic-mismatch benchmark and downstream policy learning in simulation and on a real robot. The benchmark reveals that several evaluated reward baselines assign spurious progress to wrong-object manipulation even in simple pick-and-place scenes. ARS better suppresses these errors and outperforms these baselines in simulation policy learning. We further demonstrate that ARS supports long-horizon policy learning from mixed-quality offline experience on real-robot multi-screw fastening in a full-scale laboratory replica of an industrial washing-machine assembly line. These results suggest that structured inference and verification can improve the usefulness of general-purpose VLMs for robot reward modeling. Code is at https://github.com/midea-ai/ars
Chinese Translation
进展奖励建模是估计机器人行为如何随时间变化任务进展的问题。可靠的估计需要区分有意义的状态变化与失败尝试和与任务无关的动作。我们提出了自主奖励系统（ARS），这是一个用于进展奖励建模的推理框架，利用通用视觉-语言模型（VLMs），无需额外的奖励模型训练。在给定离线轨迹和任务指令的情况下，ARS使用自适应视觉检查进行事件提议和验证。一个子代理提出与任务相关的事件时间线，主代理在估计每帧进展之前进行验证和修订。ARS可以结合可选的终端结果标签和视觉参考来指导其判断。它还可以审计来自外部奖励模型的进展估计。我们在一个受控的语义不匹配基准测试和仿真及真实机器人上的下游策略学习中评估了ARS，使用了27B VLM。基准测试揭示了几个评估的奖励基线即使在简单的拾取与放置场景中也会错误地将虚假的进展分配给错误物体的操作。ARS更好地抑制了这些错误，并在仿真策略学习中优于这些基线。我们进一步证明ARS支持从混合质量的离线经验中进行长时间跨度的策略学习，在一个工业洗衣机装配线的全尺寸实验室复制品上进行多螺丝固定。这些结果表明，结构化推理和验证可以提高通用VLM在机器人奖励建模中的有效性。代码可在 https://github.com/midea-ai/ars 获取。
cs.RO / 136 / 2609.34485
Trajectory-Safe Orienteering for Human-Robot Shared Environments
人机共享环境中的轨迹安全定向问题
Abstract
Orienteering problem (OP) has wide real-world applications and also great potential in human-robot collaboration. However, existing approaches struggle to simultaneously ensure safe and feasible trajectories while achieving high-quality task execution in shared workspaces. To this end, this work studies the OP with time windows and variable profits (OPTWVP). A two-stage DEcoupled discrete-Continuous Optimization with Service-time-guided Trajectory (DeCoST) approach is proposed to effectively solve OPTWVP in shared spaces. Meanwhile, the safety-aware time windows of nodes and the discretized workspace are introduced to ensure collision-free trajectories between the end effector and the human. Preliminary results validate the effectiveness of DeCoST in generating collision-free trajectory plans while preserving the quality of orienteering tasks.
Chinese Translation
定向问题（Orienteering Problem, OP）在现实世界中有广泛的应用，并且在人机协作中具有巨大的潜力。然而，现有的方法在确保安全且可行的轨迹的同时，难以实现高质量的任务执行，尤其是在共享工作空间中。为此，本文研究了带时间窗口和可变利润的定向问题（Orienteering Problem with Time Windows and Variable Profits, OPTWVP）。我们提出了一种两阶段的解耦离散-连续优化方法（DEcoupled discrete-Continuous Optimization with Service-time-guided Trajectory, DeCoST），以有效解决共享空间中的OPTWVP。同时，引入了节点的安全感知时间窗口和离散化工作空间，以确保末端执行器与人类之间的无碰撞轨迹。初步结果验证了DeCoST在生成无碰撞轨迹规划方面的有效性，同时保持了定向任务的质量。
cs.RO / 137 / 2609.34486
Model-Informed Safe Reinforcement Learning for Bipedal Locomotion via Step-to-Step Prediction
基于模型的安全强化学习在双足运动中的逐步预测
Abstract
Humanoid robots promise versatile mobility in cluttered, human-centric environments, but real deployment demands principled safety. Classical model-based gait generators yield interpretable motions but often lack the robustness and adaptability of modern reinforcement learning (RL) based approaches. We propose a model-informed reinforcement learning framework anchored to the analytical Angular Momentum Linear Inverted Pendulum (ALIP) template. We provide a step-to-step safety certificate for ALIP stepping via a discrete exponential control barrier function (DECBF) and use it as (i) a training-time shaping signal and (ii) a runtime action filter that minimally adjusts swing-foot placement to satisfy template-level constraints. Full-order safety is evaluated empirically on the Digit humanoid in MuJoCo with a whole-body controller stack. Compared to an unconstrained baseline, our approach reduces safety-violation events in the reported external-disturbance trial, while larger lateral-velocity transients reveal a safety-tracking tradeoff.
Chinese Translation
类人机器人在杂乱的人类中心环境中展现出多样的移动能力，但实际部署要求遵循原则性的安全性。经典的基于模型的步态生成器能够产生可解释的运动，但往往缺乏现代强化学习（RL）方法的鲁棒性和适应性。我们提出了一种基于模型的强化学习框架，该框架以分析性的角动量线性倒立摆（Angular Momentum Linear Inverted Pendulum, ALIP）模板为基础。我们通过离散指数控制障碍函数（Discrete Exponential Control Barrier Function, DECBF）为ALIP步态提供逐步安全证书，并将其作为（i）训练时的塑形信号和（ii）运行时的动作过滤器，最小调整摆动脚的位置以满足模板级约束。在MuJoCo中，我们在Digit类人机器人上通过全身控制器堆栈对全阶安全性进行了实证评估。与无约束基线相比，我们的方法在报告的外部干扰试验中减少了安全违规事件，而更大的横向速度瞬态则揭示了安全跟踪的权衡。
cs.RO / 138 / 2609.34498
Robot-Assisted Deployment and Maintenance of Inflatable Modules for Lunar Habitation: A Field Demonstration
机器人辅助的可充气模块在月球居住环境中的部署与维护：现场演示
Abstract
Long-term human habitation and in-situ development on the Moon open a new era of space utilization. In this context, robots are a key technology for facilitating the construction of future human outposts. Toward the deployment and establishment of human habitation modules on the lunar surface, we propose a combined system consisting of inflatable modules and a modular, reconfigurable robotic system. This paper presents a report demonstrating various robot-assisted task executions using real hardware, namely the modular and reconfigurable robot MoonBot and the inflatable module HIDAS, to enhance the reliability of their deployment and maintenance. The demonstrated tasks include robotic inspection during inflation, module position alignment, final safety locking, and three-dimensional mapping for post-deployment maintenance. All demonstrations were conducted either in a laboratory environment or at a lunar analogue test site. Finally, lessons learned are discussed to provide essential insights for this robotic application to future lunar habitation.
Chinese Translation
长期人类在月球的居住和原位开发开启了空间利用的新纪元。在这一背景下，机器人技术成为促进未来人类前哨建设的关键技术。为了在月球表面部署和建立人类居住模块，我们提出了一种由可充气模块和模块化、可重构的机器人系统组成的综合系统。本文展示了一项报告，演示了使用真实硬件（即模块化和可重构机器人MoonBot以及可充气模块HIDAS）进行的各种机器人辅助任务执行，以增强其部署和维护的可靠性。演示的任务包括充气过程中的机器人检查、模块位置对齐、最终安全锁定以及后期维护的三维映射。所有演示均在实验室环境或月球类测试场地进行。最后，讨论了所获得的经验教训，以为这一机器人应用于未来月球居住提供重要见解。
cs.RO / 139 / 2609.34512
MonoEgo: Monocular Metric Egocentric Demonstration Capture with Passive Wrist Constellations and Sparse Workstation Anchors
MonoEgo：基于被动腕部星座和稀疏工作站锚点的单目度量自我中心演示捕获
Abstract
Image-aligned metric demonstrations often require dedicated tracking hardware and synchronization across devices. We present MonoEgo, a capture system that replaces active wrist instrumentation with offline monocular reconstruction. One 90-FPS global-shutter camera observes calibrated passive wrist constellations, sparse workstation anchors, and the scene on a shared image clock. MonoTag SLAM combines marker corners with ORB geometry and uses visual evidence to reject ambiguous planar-marker poses. Its metric Atlas supports interval scale re-anchoring, verified map merging, and retrospective localization of earlier frames supported by the final map. Camera and wrist-constellation outputs retain validity and map provenance, and unsupported motion is left missing. Experiments show metric tracking beyond continuous anchor visibility, reconnection of supported map components, and recovery of some missing camera poses. Comparisons against a multisensor camera reference and separate stationary-constellation tests characterize trajectory agreement and precision while revealing incomplete coverage and residual geometric uncertainty. The results indicate that passive fixtures and offline reconstruction can reduce capture-side requirements. Dynamic accuracy, deployment, and downstream policy benefits require further study.
Chinese Translation
图像对齐的度量演示通常需要专用的跟踪硬件和设备间的同步。我们提出了MonoEgo，这是一种捕获系统，它用离线单目重建替代了主动腕部仪器。一台90帧每秒的全局快门相机在共享图像时钟上观察经过校准的被动腕部星座、稀疏的工作站锚点和场景。MonoTag SLAM结合了标记角点与ORB几何，并利用视觉证据拒绝模糊的平面标记姿态。其度量地图支持区间尺度重新锚定、经过验证的地图合并，以及基于最终地图的早期帧的回溯定位。相机和腕部星座的输出保持有效性和地图来源，而不支持的运动则被标记为缺失。实验表明，度量跟踪超越了连续锚点可见性，重新连接了支持的地图组件，并恢复了一些缺失的相机姿态。与多传感器相机参考和单独静态星座测试的比较表征了轨迹一致性和精度，同时揭示了不完整覆盖和残余几何不确定性。结果表明，被动装置和离线重建可以减少捕获端的要求。动态精度、部署和下游策略的好处需要进一步研究。
cs.RO / 140 / 2609.34550
Gaze Prompts: Temporally Dense Human Attention for Vision-Language-Action Fine-Tuning
注视提示：用于视觉-语言-动作微调的时间密集型人类注意力
Abstract
Vision-Language-Action (VLA) fine-tuning pairs images with actions at every step, yet typically provides only a task-level language instruction, leaving moment-to-moment visual relevance implicit. We introduce \emph{eye-tracker-supervised gaze prompting}, which uses gaze recorded during VR teleoperation to provide frame-level visual guidance for VLA fine-tuning. During training, recorded gaze locations are rendered as crosshairs on the robot's head-camera images. At deployment, a lightweight predictor estimates gaze locations from recent images and the instruction, supplying the same type of visual prompt without an eye tracker or changes to the policy architecture. Instantiated with $\pi_0$, gaze prompting increases mean success from $26.3\%$ to $56.0\%$ across six real-world bimanual manipulation tasks, with gains also observed when a single policy is trained on all six tasks. We release \textsc{GazeMani}, a dataset of $1{,}200$ teleoperated trajectories with synchronized gaze.
Chinese Translation
视觉-语言-动作（VLA）微调在每一步将图像与动作配对，但通常仅提供任务级的语言指令，使得瞬时的视觉相关性隐含在其中。我们提出了 extit{眼动追踪监督的注视提示}，该方法利用在虚拟现实远程操作中记录的注视数据，为VLA微调提供帧级的视觉指导。在训练过程中，记录的注视位置被渲染为机器人头部摄像头图像上的十字准星。在部署阶段，一个轻量级预测器根据近期图像和指令估计注视位置，提供相同类型的视觉提示，而无需眼动追踪器或对策略架构进行更改。通过使用$ extpi_0$实例化，注视提示使六个真实世界双手操作任务的平均成功率从$26.3\%$提高到$56.0\\%$，在对所有六个任务训练单一策略时也观察到了增益。我们发布了 extsc{GazeMani}，一个包含$1{,}200$条同步注视的远程操作轨迹的数据集。
cs.RO / 141 / 2609.34554
Where Memory Belongs: Ledger, an Object Ledger for Memory-Augmented VLAs
记忆的归属：Ledger，一种用于记忆增强的对象账本
Abstract
Memory is essential for long-horizon, partially observed robotic manipulation: a robot must remember which object was placed in a drawer, whose cup it moved, or how many action cycles have elapsed. Recent vision-language-action (VLA) models embed memory directly inside the policy, but benchmarks show no single in-policy mechanism covers all spatio-temporal dimensions, trailing oracle methods by a wide margin. We argue that memory type dictates where memory should reside: short-term perceptual memory (repetition, timing, retracing) belongs inside the policy, while long-term object memory (persistent spatial state, containment, event history) belongs outside as an explicit, readable record. We present Ledger, a harness that realizes this split over a single fine-tuned $\pi_{0.5}$ policy by pairing an in-policy frame-sampling memory with an external spatio-temporal object memory, the ledger, built from a SAM3 tracker and a VLM captioner of the demonstration and read by an LLM planner that decides at step boundaries. On RoboMME, Ledger reaches the highest four-suite average among the evaluated methods, 64.3% (vs. 45.9% for the strongest prior method under identical evaluation), leading object reference (60.7% vs. 40.3%) and object permanence (86.7% vs. 56.2%) using a single set of weights. Choosing the memory source at runtime, from the instruction and the record, removes the need for a task-level router.
Chinese Translation
记忆对于长时间跨度、部分观察的机器人操控至关重要：机器人必须记住哪个物体被放置在抽屉中、它移动了哪个杯子，或者经过了多少个动作周期。最近的视觉-语言-动作（VLA）模型将记忆直接嵌入到策略中，但基准测试显示，没有任何单一的策略内机制能够覆盖所有时空维度，且在很大程度上落后于神谕方法。我们认为，记忆类型决定了记忆应当存放的位置：短期感知记忆（重复、时机、回溯）应当属于策略内部，而长期物体记忆（持久空间状态、包含、事件历史）应当作为明确且可读的记录存在于外部。我们提出了Ledger，这是一种在单一微调的 $ ext{π}_{0.5}$ 策略上实现这种分离的工具，通过将策略内的帧采样记忆与外部时空物体记忆（账本）配对而构建，该账本由SAM3跟踪器和VLM描述生成器构成，并由在步骤边界做决策的LLM规划器读取。在RoboMME上，Ledger在评估的方法中达到了最高的四套平均值，64.3%（相比之下，最强的先前方法在相同评估下为45.9%），在物体引用（60.7%对比40.3%）和物体持久性（86.7%对比56.2%）方面表现突出，且使用的是同一组权重。在运行时根据指令和记录选择记忆源，消除了对任务级路由器的需求。
cs.RO / 142 / 2609.34594
Aerial GRIPPER: A Gradient-based Real-time Inverse-game Predictor and Planner
空中抓取器：一种基于梯度的实时逆博弈预测器和规划器
Abstract
Accurate capture of non-cooperative targets is critical. In an attempt to tackle this intractable challenge, an aerial gripper system integrated with a Gradient-based Real-time Inverse-game Predictor and PlannER (GRIPPER) framework is proposed. The interaction is formulated as a general-sum pursuit-evasion game under incomplete information. Specifically, underlying cost parameters of the target are inferred online, and the open-loop Nash equilibrium (OLNE) strategy is iteratively refined within a receding-horizon loop. To ensure high-frequency execution, a computationally friendly gradient-based inverse-game solver is developed. Without explicit computation of the Hessian inverse, the optimized solution is updated (> 50 Hz) based on implicit differentiation and fast Hessian-vector products. Meanwhile, an anti-disturbance controller is developed to overcome disturbances of uncertain payload and gripper actuation, enabling precise tracking of the planned trajectory and accurate grasping of the target. Simulations and real-world experiments illustrate the superior computational efficiency and task performance of GRIPPER. The task of capturing and delivering a non-cooperative target is accomplished, highlighting the robustness, adaptability, and real-time performance of the framework in highly adversarial scenarios.
Chinese Translation
准确捕捉非合作目标至关重要。为了解决这一棘手的挑战，提出了一种集成了基于梯度的实时逆博弈预测器和规划器（GRIPPER）框架的空中抓取系统。该交互被形式化为在不完全信息下的一般和追逐-逃避博弈。具体而言，目标的潜在成本参数在线推断，并在递归地平滑视界循环中迭代优化开放式纳什均衡（OLNE）策略。为了确保高频执行，开发了一种计算友好的基于梯度的逆博弈求解器。在不显式计算海森矩阵逆的情况下，优化解基于隐式微分和快速海森-向量乘积进行更新（> 50 Hz）。同时，开发了一种抗干扰控制器，以克服不确定负载和抓取器驱动的干扰，从而实现对规划轨迹的精确跟踪和对目标的准确抓取。仿真和实际实验展示了GRIPPER在计算效率和任务性能方面的优越性。成功完成了捕捉和交付非合作目标的任务，突显了该框架在高度对抗性场景中的鲁棒性、适应性和实时性能。
cs.RO / 143 / 2609.34608
Efficient World Action Model Inference with Adaptive Intermediate States
具有自适应中间状态的高效世界动作模型推断
Abstract
World Action Models (WAMs) enable future-aware control by jointly modeling actions and environment dynamics. However, iterative diffusion or flow inference incurs substantial denoising latency. Prior inference state offers a natural opportunity for acceleration, yet changing planning contexts, observations, and intermediate representations can quickly render retained state stale. Preserving useful computation therefore requires adapting inference state rather than reusing it as-is. To this end, we present $\mathrm{WAM}{\scriptstyle\mathrm{ACHINE}}$, a training-free framework that accelerates WAM inference by preserving and adapting inference state for efficient and accurate continuation as the control loop evolves. Across closed-loop replans, Trajectory Remapping remaps replan state from the preceding replan to initialize the next replan, reducing redundant trajectory generation. Across denoising steps, Observation Rebinding performs anticipatory inference during action execution and rebinds retained denoising state to the real observation for continuation when consistency checks pass, reducing latency exposed to the control loop. Across Transformer layers, Residual Rescaling selectively rescales retained layer state and refreshes it through full computation of the middle layers when probe checks fail, reducing repeated Transformer computation. Evaluations of three representative WAM architectures on LIBERO and RoboTwin 2.0 show that $\mathrm{WAM}{\scriptstyle\mathrm{ACHINE}}$ achieves 1.47-3.05$\times$ speedups in observation-to-action latency and 2.23-3.27$\times$ speedups in GPU inference time per replan, while preserving 96.69-99.54% of native WAM task success.
Chinese Translation
世界动作模型（World Action Models，WAMs）通过联合建模动作和环境动态，实现未来感知控制。然而，迭代扩散或流推断会导致显著的去噪延迟。先前的推断状态为加速提供了自然的机会，但变化的规划上下文、观察和中间表示会迅速使保留的状态变得过时。因此，保留有用的计算需要调整推断状态，而不是直接重用。为此，我们提出了$ ext{WAM}_{ ext{ACHINE}}$，这是一个无训练的框架，通过保留和调整推断状态来加速WAM推断，以便在控制循环演变时实现高效和准确的延续。在闭环重新规划中，轨迹重映射（Trajectory Remapping）将前一个重新规划的状态重映射，以初始化下一个重新规划，从而减少冗余的轨迹生成。在去噪步骤中，观察重绑定（Observation Rebinding）在动作执行期间进行预期推断，并在一致性检查通过时将保留的去噪状态重新绑定到真实观察，以便在延续时减少暴露于控制循环的延迟。在Transformer层中，残差重缩放（Residual Rescaling）在探测检查失败时选择性地重新缩放保留的层状态，并通过对中间层的完全计算来刷新它，从而减少重复的Transformer计算。在LIBERO和RoboTwin 2.0上对三种代表性WAM架构的评估表明，$ ext{WAM}_{ ext{ACHINE}}$在观察到动作的延迟上实现了1.47-3.05倍的加速，在每次重新规划的GPU推断时间上实现了2.23-3.27倍的加速，同时保留了96.69-99.54%的原生WAM任务成功率。
cs.RO / 144 / 2609.34666
On the Numerical Reliability of Differentiable Physics-Based Optimization for Robotic Material Manipulation
可微物理基础优化在机器人材料操作中的数值可靠性研究
Abstract
Differentiable physics is increasingly used in robotic material manipulation for system identification, trajectory or skill optimization, demonstration generation, and robot or end-effector design. These applications depend on gradients propagated through long, contact-rich simulation rollouts. We study the numerical reliability of those gradients using two Material Point Method (MPM) system-identification benchmarks derived from elastoplastic and granular manipulation. The benchmarks provide controlled cases for three effects that also arise in broader differentiable physics-based optimization. GPU many-to-one sums whose order depends on thread scheduling changed long-horizon gradients and reversed the sign of one parameter gradient relative to a deterministic reference. Finite-difference checks became less reliable for longer rollouts because repeated-run loss variation grew much faster than the loss change produced by the tested parameter perturbations. Observation and loss definitions changed optimization behaviour and the solution preferred by an independent metric. These results motivate reproducible accumulation, finite-difference validation that compares perturbation-induced loss changes with repeated-run variation, and explicit reporting of objective construction when differentiable simulation is used for robotic optimization.
Chinese Translation
可微物理在机器人材料操作中越来越多地应用于系统识别、轨迹或技能优化、演示生成以及机器人或末端执行器设计。这些应用依赖于通过长时间、接触丰富的仿真展开传播的梯度。我们研究了这些梯度的数值可靠性，使用了两个基于材料点法（Material Point Method, MPM）的系统识别基准，这些基准源于弹塑性和颗粒操作。这些基准提供了受控案例，以研究在更广泛的可微物理基础优化中出现的三种效应。GPU 多对一求和的顺序依赖于线程调度，导致长时间范围的梯度发生变化，并且相对于确定性参考，一个参数的梯度符号发生了反转。有限差分检查在长时间展开中变得不那么可靠，因为重复运行的损失变化比测试参数扰动产生的损失变化增长得更快。观察和损失定义改变了优化行为以及独立指标所偏好的解决方案。这些结果促使我们进行可重复的累积、有限差分验证，比较扰动引起的损失变化与重复运行的变化，并在使用可微仿真进行机器人优化时明确报告目标构建。
cs.RO / 145 / 2609.34674
HOI-Retarget: Contact-Centric Retargeting for Human-Object Interaction
HOI-Retarget：以接触为中心的人-物交互重定向
Abstract
Learning from demonstration (LfD) has enabled humanoid robots to acquire diverse whole-body skills, but extending this paradigm to human-object interaction (HOI) is limited by the availability of robot-compatible interaction references. We present HOI-Retarget, a contact-centric retargeting method that transfers HOI onto a humanoid robot for large-scale motion-data generation. Its windowed trajectory optimization uses every labeled contact as a target in the object frame, balancing body tracking, foot support and smoothness under the robot's kinematic limits. The method can augment a single demonstration across object sizes, absorb contacts reconstructed from monocular video, and extend to several robots manipulating one object. We publicly release the code and the retargeted motion dataset.
Chinese Translation
从示范学习（LfD）中，类人机器人能够获得多样化的全身技能，但将这一范式扩展到人-物交互（HOI）受到机器人兼容的交互参考可用性的限制。我们提出了HOI-Retarget，这是一种以接触为中心的重定向方法，旨在将HOI转移到类人机器人上，以生成大规模的运动数据。其窗口化轨迹优化将每个标记的接触点作为目标在物体框架中使用，平衡身体跟踪、脚部支撑和在机器人运动学限制下的平滑性。该方法可以在不同物体尺寸之间增强单一示范，吸收从单目视频重建的接触，并扩展到多个机器人共同操控一个物体。我们公开发布了代码和重定向的运动数据集。
cs.RO / 146 / 2609.34684
Natural State-Prediction Accuracy can Hide Weak Controlled Responsiveness in VLA Readouts
自然状态预测准确性可能掩盖 VLA 读出中的弱控制响应性
Abstract
Accurately decoding object states from the internal representations of vision-language-action (VLA) models does not establish that the predictions respond faithfully to changes in the target physical state. In natural observations, object state, robot configuration, occlusion, and task progress vary together, allowing contextual cues to contribute to prediction. In this paper, we introduce an evaluation framework that separates prediction accuracy, target-state responsiveness, and context stability using physically validated observations that cross target coordinates with robot contexts. We demonstrate that high natural-trajectory accuracy can coexist with weak controlled target-state responsiveness in fixed representation-readout pairs. Comparisons and interventions involving representations, readouts, and training data show that the three properties provide distinct diagnostic information. Furthermore, adding responsiveness and context sensitivity to a failure predictor based on initial state error and physical variables reduces policy-failure prediction error on new initializations relative to the specified baseline while same-observation controlled MAE is also informative. These findings motivate evaluating target-state responsiveness and context stability alongside natural prediction accuracy, and examining their relationship to actual policy behavior and task outcomes.
Chinese Translation
从视觉-语言-行动（VLA）模型的内部表征中准确解码物体状态并不能证明预测对目标物理状态变化的响应是忠实的。在自然观察中，物体状态、机器人配置、遮挡和任务进展是共同变化的，这使得上下文线索能够对预测产生影响。本文引入了一个评估框架，该框架使用物理验证的观察结果，将预测准确性、目标状态响应性和上下文稳定性分开，这些观察结果跨越目标坐标与机器人上下文。我们证明，高自然轨迹准确性可以与固定表征-读出对中的弱控制目标状态响应性共存。涉及表征、读出和训练数据的比较和干预表明，这三种属性提供了不同的诊断信息。此外，在基于初始状态误差和物理变量的失败预测器中添加响应性和上下文敏感性，相对于指定的基线，能够减少新初始化下的策略失败预测误差，同时相同观察下的控制均方误差（MAE）也提供了有价值的信息。这些发现促使我们在评估自然预测准确性的同时，也关注目标状态响应性和上下文稳定性，并考察它们与实际策略行为和任务结果的关系。
cs.RO / 147 / 2609.34695
Sufficiency of Zeroth-Order Reward Shaping for Policy Gradient in Stabilization Control
零阶奖励塑形在稳定控制中的策略梯度充分性
Abstract
Reward shaping is fundamental to modern robotic control with deep reinforcement learning (RL), yet practitioners still rely heavily on heuristic principles borrowed from classical optimal control and trajectory optimization. Existing methods rarely distinguish reward terms that are intrinsic to the control objective from numerical regularizers, leading to brittle hyperparameter tuning. To determine which quantities a reward must contain, we study the stabilization control problem with a focus on zeroth-order (configuration) and first-order (velocity) information. We theoretically and empirically demonstrate that policy gradient methods can successfully solve stabilization tasks without first-order reward terms, adding such terms can instead introduce severe sensitivity as their scale grows. Conversely, our findings confirm that reward functions must be zeroth-order complete over goal-relevant coordinates, while the first-order state remains necessary in the policy observation under our low-dissipation assumptions. Overall, these results provide actionable and principled guidance for reward design in robotic RL.
Chinese Translation
奖励塑形是现代机器人控制与深度强化学习（RL）中的基础，但从业者仍然在很大程度上依赖于借鉴自经典最优控制和轨迹优化的启发式原则。现有方法很少区分与控制目标内在相关的奖励项和数值正则化项，导致超参数调优不稳定。为了确定奖励必须包含哪些量，我们研究了稳定控制问题，重点关注零阶（配置）和一阶（速度）信息。我们理论上和实证上证明，策略梯度方法可以成功解决稳定任务，而无需一阶奖励项，添加此类项反而可能在其规模增大时引入严重的敏感性。相反，我们的研究结果确认，奖励函数必须在与目标相关的坐标上是零阶完整的，而在我们的低耗散假设下，一阶状态仍然在策略观察中是必要的。总体而言，这些结果为机器人强化学习中的奖励设计提供了可操作且有原则的指导。
cs.RO / 148 / 2609.34702
MarsLab: A Martian Rover Simulator for Planetary Rover Autonomous Navigation
MarsLab：一个用于行星探测车自主导航的火星探测车模拟器
Abstract
Future Mars missions will require rover autonomy that can operate across unstructured terrain, changing illumination, atmospheric dust, and limited communication. Simulation is a practical way to study these conditions before deployment, but existing Mars-relevant resources differ in scope, including mission-oriented simulators, fixed analog datasets, task-specific environments, and open robotics interfaces. In this context, we present MarsLab, an open-source, ROS2-native Mars rover simulator for autonomy and navigation algorithm development. MarsLab combines HiRISE-derived and procedural terrain with customizable rock, crater, solar-illumination, and atmospheric-dust settings, and runs a Perseverance-class rover model in NVIDIA Isaac Sim. The runtime publishes RGB, depth, RGB-D point clouds, LiDAR, IMU, wheel odometry, and Ground Truth (GT) pose data through standard ROS2 topics. We demonstrate MarsLab with Simultaneous Localization and Mapping (SLAM) benchmarks across sensing modalities, dust levels, scene geometry, and route length, and with Visual Place Recognition (VPR) benchmarks over repeated Mars Base traversals under illumination and dust changes. The results illustrate how controlled scene variation and shared GT trajectories can be used to compare trajectory-level estimation and image-level place recognition within the same simulator. Our Project Page: https://kimhoyun-robotair.github.io/MarsLab/.
Chinese Translation
未来的火星任务将需要能够在非结构化地形、变化的光照、气候尘埃和有限通信条件下操作的探测车自主性。模拟是一种在部署前研究这些条件的实用方法，但现有的与火星相关的资源在范围上各不相同，包括以任务为导向的模拟器、固定的类比数据集、特定任务的环境和开放的机器人接口。在此背景下，我们提出了MarsLab，一个开源的、基于ROS2的火星探测车模拟器，用于自主性和导航算法的开发。MarsLab结合了基于HiRISE的和程序生成的地形，具有可定制的岩石、陨石坑、太阳光照和气候尘埃设置，并在NVIDIA Isaac Sim中运行一个与“毅力号”（Perseverance）相当的探测车模型。运行时通过标准的ROS2主题发布RGB、深度、RGB-D点云、激光雷达（LiDAR）、惯性测量单元（IMU）、轮子里程计和真实位姿（Ground Truth, GT）数据。我们通过不同的传感模式、尘埃水平、场景几何和路线长度的同时定位与地图构建（SLAM）基准测试，以及在光照和尘埃变化下重复火星基地穿越的视觉位置识别（VPR）基准测试来演示MarsLab。结果展示了如何利用受控场景变化和共享的GT轨迹在同一模拟器内比较轨迹级估计和图像级位置识别。我们的项目页面：https://kimhoyun-robotair.github.io/MarsLab/
cs.RO / 149 / 2609.34707
SOR-Nav: Search or Relocate? Context-Gated Exploration and Cross-Region Relocation for Object Navigation
SOR-Nav：搜索还是重新定位？上下文门控探索与跨区域重新定位用于物体导航
Abstract
Object navigation requires an embodied agent to find an object in an unseen environment under partial observability and a limited motion budget. Existing methods primarily optimize where the robot should go next by ranking candidate destinations. In contrast to these methods, we present SOR-Nav, a hierarchical navigation system that explicitly arbitrates between continuing to explore the current context and abandoning it for a more promising reachable region. First, an autonomous semantic exploration system is built that accumulates persistent 3D object clusters and organizes reachable frontiers into a cluster decision graph to provide an efficient search abstraction. Then, SOR-Nav uses a context-gated LLM-driven object-search supervisor to evaluate the suitability of the current search context and decide whether to continue exploration or perform cross-region relocation to another reachable frontier cluster. Across the complete, unfiltered validation sets of HM3D-v1, HM3D-v2, and MP3D, SOR-Nav achieves the strongest reported Success Rate (SR) and Success weighted by Path Length (SPL) on all three benchmarks. On MP3D in particular, it more than doubles the previous best SPL from 18.1\% to 38.5\% while increasing SR from 50.7\% to 61.8\%. Nested HM3D-v2 ablations validate the proposed decision structure, while a continuous three-target physical deployment demonstrates persistent ObjectNav operation in real-world scenarios.
Chinese Translation
物体导航要求具身代理在部分可观测和有限运动预算的情况下，在未知环境中找到物体。现有方法主要通过对候选目的地进行排名来优化机器人下一步的去向。与这些方法不同，我们提出了SOR-Nav，一个分层导航系统，明确调解是继续探索当前上下文还是放弃它以寻找更有前景的可达区域。首先，构建了一个自主语义探索系统，该系统积累持久的3D物体聚类，并将可达前沿组织成聚类决策图，以提供高效的搜索抽象。然后，SOR-Nav使用上下文门控的基于LLM（大语言模型）的物体搜索监督器来评估当前搜索上下文的适宜性，并决定是继续探索还是进行跨区域重新定位到另一个可达的前沿聚类。在HM3D-v1、HM3D-v2和MP3D的完整未过滤验证集上，SOR-Nav在所有三个基准测试中实现了报告的最高成功率（Success Rate, SR）和按路径长度加权的成功率（Success weighted by Path Length, SPL）。特别是在MP3D上，它将之前最佳的SPL从18.1%提高到38.5%，同时将SR从50.7%提高到61.8%。嵌套的HM3D-v2消融实验验证了所提出的决策结构，而连续的三目标物理部署展示了在真实场景中持续的ObjectNav操作。
cs.RO / 150 / 2609.34719
ExcavaTwin: Training-Free Geometry-Guided Semantic Elevation Mapping for Autonomous Excavation
ExcavaTwin：无训练的几何引导语义高程映射用于自主挖掘
Abstract
Autonomous excavation requires a spatial representation that jointly captures terrain geometry and task-relevant semantics. Existing excavation mapping is largely elevation-centric, while generic semantic models remain unstable in unstructured outdoor scenes. We present ExcavaTwin, a pure-vision geometry-guided semantic elevation mapping framework without excavation-specific training. Given multi-view RGB images, the framework: 1) reconstructs scene geometry and semantic observations using frozen vision models; 2) derives terrain and non-terrain geometric support; 3) performs geometry-constrained multi-view semantic fusion to suppress implausible predictions and recover incomplete observations; and 4) projects the fused state into a task-oriented semantic elevation map. Experiments on public datasets and real excavation scenes demonstrate reliable geometric and semantic perception. In real excavation, the system achieved an average update interval of approximately 1.4 s and a mean elevation error of 12.74cm in dynamically modified regions. Larger errors mainly occur during rapid terrain changes and transient visual disturbances caused by machine motion.
Chinese Translation
自主挖掘需要一种空间表示，能够共同捕捉地形几何和与任务相关的语义。现有的挖掘映射主要以高程为中心，而通用语义模型在非结构化的户外场景中仍然不稳定。我们提出了ExcavaTwin，一种纯视觉的几何引导语义高程映射框架，无需针对挖掘的特定训练。该框架基于多视角RGB图像：1）使用冻结的视觉模型重建场景几何和语义观测；2）推导地形和非地形的几何支持；3）执行几何约束的多视角语义融合，以抑制不合理的预测并恢复不完整的观测；4）将融合状态投影到面向任务的语义高程图中。在公共数据集和真实挖掘场景上的实验表明，该系统在几何和语义感知方面表现可靠。在真实挖掘中，该系统实现了约1.4秒的平均更新间隔和12.74厘米的动态修改区域的平均高程误差。较大的误差主要发生在快速地形变化和由机器运动引起的瞬态视觉干扰期间。
cs.RO / 151 / 2609.34724
DexWeave: Learning Dexterous Humanoid Loco-Manipulation from Human Demonstrations
DexWeave：从人类示范中学习灵巧的人形机器人运动操控
Abstract
Learning dexterous humanoid loco-manipulation from human demonstrations requires transferring not only human motion, but also the coordinated interaction structure underlying the demonstrated behavior. This is challenging because embodiment differences distort the coupling among body motion, wrist placement, finger articulation, and object interaction, while kinematically accurate references may still be difficult to realize under robot dynamics. We present DexWeave, a unified framework that connects interaction-consistent motion retargeting with anatomy-aware whole-body policy learning. DexWeave first employs a two-stage retargeting procedure that initializes body and hand motions with specialized solvers and subsequently performs coupled refinement over the upper-body interaction chain while preserving lower-body support. The resulting references are tracked by an anatomy-aware Transformer policy that represents anatomical regions as structured tokens and uses directed masked attention to model their dependencies, with object information selectively conditioning the upper-body pathway for dexterous interaction. The policy jointly outputs body and dexterous-hand actions and is trained directly with reinforcement learning, without pretrained tracking policies, teacher-student distillation, or subsequent residual refinement. DexWeave improves retargeting fidelity and interaction consistency while achieving higher manipulation performance and faster policy convergence than MLP baselines. We further deploy the learned policies on a physical Unitree G1 humanoid equipped with Inspire dexterous hands, demonstrating dexterous whole-body loco-manipulation in the real world. See our project page (https://dexweave.github.io) for videos.
Chinese Translation
从人类示范中学习灵巧的人形机器人运动操控不仅需要转移人类的运动，还需要传递示范行为背后的协调交互结构。这一过程具有挑战性，因为身体的体现差异会扭曲身体运动、手腕位置、手指关节和物体交互之间的耦合关系，而在机器人动力学下实现运动学准确的参考仍然可能困难。我们提出了DexWeave，这是一个统一框架，将交互一致的运动重定向与关注解剖结构的全身策略学习相连接。DexWeave首先采用两阶段的重定向程序，利用专门的求解器初始化身体和手部运动，随后在保持下肢支撑的同时，对上半身交互链进行耦合优化。生成的参考运动由一个关注解剖结构的Transformer策略跟踪，该策略将解剖区域表示为结构化的标记，并使用定向掩蔽注意力来建模它们之间的依赖关系，物体信息选择性地调节上半身路径以实现灵巧交互。该策略共同输出身体和灵巧手部的动作，并通过强化学习直接进行训练，无需预训练的跟踪策略、师生蒸馏或后续的残差优化。DexWeave在提高重定向精度和交互一致性的同时，达到了比多层感知器基线更高的操控性能和更快的策略收敛速度。我们进一步将学习到的策略部署在配备Inspire灵巧手的物理Unitree G1人形机器人上，展示了在现实世界中的灵巧全身运动操控。请访问我们的项目页面（https://dexweave.github.io）观看视频。
cs.RO / 152 / 2609.34730
Action Sequence Transfer via LLMs for Heterogeneous Environments
通过大型语言模型在异构环境中进行动作序列转移
Abstract
We present an action sequence transfer system that adaptively transfers user action sequences across different target spaces. Given an input action sequence from a source space and scene graph representations of both the source and target environments, our system predicts a corresponding action sequence in the target space by adapting to the spatial and object constraints of the new environment. To achieve this, we leverage multi-level representations of user activity to generalize actions at varying levels of abstraction. To demonstrate our system, we collect a new scene graph-based dataset derived from the Ego4D GoalStep dataset for evaluation. Results indicate that our system can generate valid action sequences even between spaces with drastically different object configurations.
Chinese Translation
我们提出了一种动作序列转移系统，该系统能够自适应地在不同目标空间之间转移用户的动作序列。给定来自源空间的输入动作序列以及源环境和目标环境的场景图表示，我们的系统通过适应新环境的空间和物体约束，预测目标空间中相应的动作序列。为实现这一目标，我们利用用户活动的多层次表示，以在不同抽象层次上对动作进行概括。为了展示我们的系统，我们收集了一个新的基于场景图的数据集，该数据集源自Ego4D GoalStep数据集以进行评估。结果表明，我们的系统能够在物体配置截然不同的空间之间生成有效的动作序列。
cs.RO / 153 / 2609.34743
Simulation for Planetary Robotic Perception and Autonomy: A Concise Survey of Recent Capabilities and Gaps
行星机器人感知与自主性的仿真：近期能力与不足的简要调查
Abstract
Planetary robotics is an important enabler of scientific exploration in environments where direct human-in-the-loop operation is costly, hazardous, or infeasible. However, developing and validating planetary robotic systems remains difficult because representative field testing is expensive, limited, and often unrepeatable under mission-relevant conditions. In this setting, simulation serves as a central tool for perception and autonomy research, synthetic data generation, system integration, and pre-deployment evaluation. Despite its importance, the literature on planetary robotics simulation remains dispersed across different simulation engines, implementations, and application settings. This paper surveys simulation works for planetary robotic perception and autonomy across four practical axes: Openness and Availability, Scenario and Platform Coverage, Sensor and Perception Support, and Environmental and Operational Realism. The surveyed simulation works report visual or physical fidelity and support perception-oriented workflows. They also indicate uneven public availability, rover-centered coverage, partial support for specialized sensing modalities, and uneven reporting of operational constraints such as onboard computation, energy, and communication restrictions.
Chinese Translation
行星机器人技术是科学探索的重要推动力，尤其是在直接人类操作成本高昂、危险或不可行的环境中。然而，开发和验证行星机器人系统仍然面临挑战，因为在与任务相关的条件下，代表性的现场测试既昂贵又有限，且往往不可重复。在这种背景下，仿真作为感知与自主研究、合成数据生成、系统集成和部署前评估的核心工具发挥着重要作用。尽管其重要性不言而喻，但关于行星机器人仿真的文献仍然分散在不同的仿真引擎、实现方式和应用场景中。本文对行星机器人感知与自主性的仿真工作进行了调查，涵盖四个实际维度：开放性与可用性、场景与平台覆盖、传感器与感知支持，以及环境与操作现实性。被调查的仿真工作报告了视觉或物理的逼真度，并支持以感知为导向的工作流程。它们还表明公共可用性不均、以探测车为中心的覆盖、对专业传感模式的部分支持，以及对操作约束（如机载计算、能源和通信限制）的不均匀报告。
cs.RO / 154 / 2609.34782
CoHuB: A Simulation Benchmark for Multi-Humanoid Collaboration
CoHuB：多类人形协作的仿真基准
Abstract
Many physical tasks in human environments require collaboration, from assisting a partner to jointly manipulating an object. Yet, existing humanoid benchmarks largely focus on single-humanoid skills and lack evaluation of multi-humanoid collaboration under egocentric visual observations. We introduce CoHuB (Collaborative Multi-Humanoid Benchmark), a simulation benchmark for multi-humanoid collaboration under egocentric visual observations. CoHuB provides 10 tasks, eight with two humanoids and two with three humanoids, spanning diverse collaboration patterns. We also provide synchronized demonstrations collected through a multi-operator VR teleoperation pipeline, in which each operator controls one humanoid from its egocentric view. Experiments with representative visuomotor policies reveal substantial challenges across different forms of coordinated perception and control. CoHuB provides a foundation for developing and evaluating multi-humanoid collaboration policies.
Chinese Translation
许多在人工环境中的物理任务需要协作，从协助伙伴到共同操控物体。然而，现有的人形基准主要集中于单一人形技能，缺乏对在自我中心视觉观察下的多类人形协作的评估。我们提出了CoHuB（协作多类人形基准），这是一个针对在自我中心视觉观察下的多类人形协作的仿真基准。CoHuB提供了10个任务，其中8个涉及两个类人形，2个涉及三个类人形，涵盖了多样的协作模式。我们还提供了通过多操作员虚拟现实远程操作管道收集的同步演示，每个操作员从其自我中心视角控制一个类人形。对具有代表性的视觉运动策略的实验揭示了在不同形式的协调感知和控制中存在的重大挑战。CoHuB为开发和评估多类人形协作策略提供了基础。
cs.RO / 155 / 2609.34823
AGRO-SUVIDE: Agentic Robotics for Surgical Viscoelastic Debridement
AGRO-SUVIDE：用于外科粘弹性清创的自主机器人技术
Abstract
Augmented dexterity has the potential to reduce the fatigue experienced by surgeons during repetitive surgical tasks. In this paper, we propose the first AGentic RObotics framework for SUrgical VIscoelastic DEbridement (AGRO-SUVIDE), the repeated removal of small fragments attached to a viscoelastic substrate. Leveraging the self-improving and coding capability of agents, AGRO-SUVIDE adopts a modular framework. Specifically, the demonstration analysis module automatically identifies recurring skills from a single expert demonstration, using both visual and kinematic information. The construction module then builds each skill, either as a procedural model-based skill the agent codes against a scaffolded library or as a model-free policy-based skill. At runtime, the monitoring module composes the skills into a loop-style graph sized to the number of fragments it observes, then verifies pre- and post-conditions of each skill to decide whether to advance or retry. We evaluate AGRO-SUVIDE through 340 physical trials on the da Vinci Research Kit (dVRK). AGRO-SUVIDE achieves an average single-fragment removal success rate of 85%, completing consecutive three-fragment removal at 60% and at 95% with one human intervention. It further generalizes to unseen five-fragment scenarios with an average success rate of 80% for single-fragment removal. Project page: https://surgical-robotics.github.io/AGRO-SUVIDE/
Chinese Translation
增强的灵活性有潜力减少外科医生在重复外科任务中经历的疲劳。在本文中，我们提出了第一个用于外科粘弹性清创的自主机器人框架（AGRO-SUVIDE），即反复去除附着在粘弹性基材上的小碎片。AGRO-SUVIDE 利用代理的自我改进和编码能力，采用模块化框架。具体而言，演示分析模块自动识别来自单一专家演示的重复技能，使用视觉和运动学信息。构建模块则将每个技能构建为代理根据支架库编码的程序模型技能或无模型的基于策略的技能。在运行时，监控模块将技能组合成一个循环样式的图，大小与其观察到的碎片数量相对应，然后验证每个技能的前置和后置条件，以决定是继续还是重试。我们通过在达芬奇研究工具包（dVRK）上进行340次物理试验来评估 AGRO-SUVIDE。AGRO-SUVIDE 实现了85%的单碎片去除成功率，在连续去除三个碎片时成功率为60%，在一次人类干预下达到95%。它进一步推广到未见过的五碎片场景，单碎片去除的平均成功率为80%。项目页面：https://surgical-robotics.github.io/AGRO-SUVIDE/
cs.RO / 156 / 2609.34911
Don't Throw Away the Tail: Action Upcycling for Policy Acceleration
不要丢弃尾部：政策加速的行动再利用
Abstract
Modern robot policies predict a chunk of future actions from a single observation, execute only a prefix, and discard the rest before replanning. Choosing the length of this prefix, the execution horizon, poses a trade-off between reactivity and efficiency. A short horizon keeps the policy reactive to the environment, but requires frequent policy calls. Recent test-time methods adaptively select the horizon for each chunk, but they either read model internals, where the signal must be chosen for each architecture, or draw extra samples, which adds cost. We propose *Action Upcycling*, a training-free algorithm that reuses actions the policy would otherwise discard, without accessing model internals or drawing extra samples. We find that discarded actions stay close to their replanned versions as long as the action velocity remains smooth. Action Upcycling therefore extends the execution horizon up to the point where the velocity begins to fluctuate. Extensive experiments on simulated and real-world manipulation tasks show that Action Upcycling reduces policy calls by 1.2--1.7$\times$ with no loss in success rate, across multiple Vision-Language-Action Models (VLAs) and even a World Action Model (WAM). It applies to any chunked policy at negligible cost and is orthogonal to other policy acceleration methods such as few-step sampling and streaming action decoding, opening a new axis for policy acceleration.
Chinese Translation
现代机器人政策从单一观察中预测未来一系列动作，仅执行前缀并在重新规划之前丢弃其余部分。选择这个前缀的长度，即执行视野，存在反应性与效率之间的权衡。较短的视野使政策对环境保持反应性，但需要频繁调用政策。最近的测试时方法自适应地为每个动作块选择视野，但它们要么读取模型内部信息（信号必须为每种架构选择），要么需要额外采样，这增加了成本。我们提出了*行动再利用*（Action Upcycling），这是一种无训练的算法，能够重复使用政策本会丢弃的动作，而无需访问模型内部或进行额外采样。我们发现，只要动作速度保持平稳，被丢弃的动作与其重新规划版本之间的距离保持较近。因此，行动再利用将执行视野延长到速度开始波动的点。针对模拟和真实世界操作任务的大量实验表明，行动再利用在多个视觉-语言-动作模型（Vision-Language-Action Models, VLA）甚至一个世界动作模型（World Action Model, WAM）中，将政策调用减少了1.2到1.7倍，而成功率没有下降。它适用于任何分块政策，成本微乎其微，并且与其他政策加速方法（如少步采样和流式动作解码）是正交的，为政策加速开辟了新的方向。
cs.RO / 157 / 2609.34968
RoboFL: Federated Expert Assembly for World Action Models
RoboFL：用于世界行动模型的联邦专家组装
Abstract
Vision-language-action and world-action models are increasingly popular, yet remain bottlenecked by physical interaction data that is scarce, institutionally siloed, and task-heterogeneous. A natural federated solution is to let each client adapt a shared foundation model through parameter-efficient fine-tuning, avoiding the exchange of full-model updates. However, federating these adapters is nontrivial, as naive aggregation can entangle incompatible updates, while incorporating MoE-style routing into federated aggregation may dilute specialization and destabilize expert selection. We present RoboFL, which instantiates MoSAIC (Mixture of Slotted Adapters) for federated world-action learning. MoSAIC directly installs locally trained LoRA adapters as the expert branches of a server MoE. Server-side routers learn token assignments over these prior-informed branches while jointly refining routing and expert parameters. Foresight-to-Action Routing Distillation (FARD) aligns routing across the model's three paths, while Path-Consensus Expert Aggregation (PCEA) converts complete expert updates into a compact global adapter for personalized redistribution. Experiments on RoboTwin 2.0, RLBench, and a real-world Franka robot arm show the superiority of RoboFL with structured expert assembly, as it outperforms centralized PEFT InternVLA-A1 by 12.23% on the Franka arm, while reducing per-round client communication by up to 86.81% relative to MoE-based federated VLA baselines.
Chinese Translation
视觉-语言-行动和世界-行动模型越来越受到欢迎，但仍然受到稀缺的物理交互数据的瓶颈，这些数据在机构间被孤立且任务异质。一个自然的联邦解决方案是让每个客户端通过参数高效的微调来适应共享的基础模型，从而避免交换完整模型更新。然而，联邦这些适配器并非易事，因为简单的聚合可能会纠缠不兼容的更新，而将 MoE（混合专家）风格的路由纳入联邦聚合可能会稀释专业化并使专家选择不稳定。我们提出了 RoboFL，它实现了 MoSAIC（混合插槽适配器）用于联邦世界-行动学习。MoSAIC 直接将本地训练的 LoRA 适配器安装为服务器 MoE 的专家分支。服务器端路由器在这些先前信息丰富的分支上学习令牌分配，同时共同优化路由和专家参数。前瞻性-行动路由蒸馏（FARD）在模型的三条路径上对齐路由，而路径共识专家聚合（PCEA）将完整的专家更新转换为紧凑的全局适配器，以便个性化重新分配。在 RoboTwin 2.0、RLBench 和真实世界的 Franka 机器人手臂上的实验表明，RoboFL 在结构化专家组装方面具有优越性，相较于集中式 PEFT InternVLA-A1 在 Franka 手臂上提高了 12.23%，同时相较于基于 MoE 的联邦 VLA 基线减少了每轮客户端通信高达 86.81%。
cs.RO / 158 / 2609.34969
NavJev: Efficient Vision-Language Navigation via Action-Centric Visual Compression and Discriminative Action-Semantic Memory
NavJev：通过以动作为中心的视觉压缩和区分性动作语义记忆实现高效的视觉-语言导航
Abstract
Recent zero-shot Vision-and-Language Navigation (VLN) methods increasingly rely on multimodal large language models (MLLMs) to reason over visual observations, navigation instructions, and candidate actions. Although effective, repeatedly invoking autoregressive multimodal reasoning at every navigation step introduces substantial inference latency, limiting the responsiveness of embodied agents. We propose NavJev, an efficient VLN framework that reformulates online navigation from repeated multimodal generation into compact visual compression followed by lightweight typed action selection. Specifically, Action-Centric Visual Compression (ACVC) integrates waypoint geometry, BLIP captions, and RAM semantic tags into compact representations of candidate actions, while Discriminative Action-Semantic Memory (DASM) filters shared semantics and maintains discriminative action-specific evidence across navigation steps. Based on these representations, Jev directly performs structured probabilistic decisions over the available action set. Experiments on R2R-CE show that NavJev achieves 27.0% SR and 22.4% SPL with only 0.65 s per navigation step, while substantially reducing inference latency and cost compared with MLLM-based VLN methods. The project page is available at https://kai-sheng-caesar.github.io/NavJev/.
Chinese Translation
近期的零-shot视觉与语言导航（VLN）方法越来越依赖多模态大型语言模型（MLLMs）对视觉观察、导航指令和候选动作进行推理。尽管有效，但在每个导航步骤中反复调用自回归多模态推理会引入显著的推理延迟，限制了具身智能体的响应能力。我们提出了NavJev，一个高效的VLN框架，它将在线导航从重复的多模态生成重新构建为紧凑的视觉压缩，随后进行轻量级的类型化动作选择。具体而言，动作中心视觉压缩（Action-Centric Visual Compression, ACVC）将航点几何、BLIP字幕和RAM语义标签整合为候选动作的紧凑表示，而区分性动作语义记忆（Discriminative Action-Semantic Memory, DASM）则过滤共享语义，并在导航步骤中保持区分性动作特定证据。基于这些表示，Jev直接对可用动作集进行结构化的概率决策。在R2R-CE上的实验表明，NavJev在每个导航步骤仅需0.65秒的情况下实现了27.0%的成功率（SR）和22.4%的成功路径长度（SPL），同时显著降低了推理延迟和成本，相较于基于MLLM的VLN方法。项目页面可访问：https://kai-sheng-caesar.github.io/NavJev/。
cs.RO / 159 / 2609.35000
Graph-Based Simultaneous Path and Foothold Planning for Multi-Limbed Intra-Vehicular Robots in Space Stations
基于图的多肢体空间站内机器人同时路径和支撑点规划
Abstract
Robot-aided operations in space stations are essential for reducing the workload of astronauts and improving the efficiency of on-orbit activities. Multi-limbed intra-vehicular robots (MLIVRs) equipped with grappling end-effectors have emerged as a promising solution, as they can securely grasp pre-existing interfaces, such as handrails and seat tracks, thereby enabling stable locomotion and forceful manipulation in microgravity environments. Since graspable locations on these interfaces are spatially limited and discretely distributed, motion planning for MLIVRs must be addressed jointly with foothold planning. This paper presents a simultaneous path and foothold planning framework based on graph theory for MLIVRs. The proposed method efficiently searches for feasible stance sequences for a multi-limbed robot while satisfying manipulability constraints. The effectiveness of the proposed framework is validated through simulations in a 3D model of the International Space Station (ISS) cabin, demonstrating its capability to generate feasible and efficient locomotion plans in realistic intra-vehicular environments.
Chinese Translation
在空间站中，机器人辅助操作对于减轻宇航员的工作负担和提高在轨活动的效率至关重要。配备抓取末端执行器的多肢体空间内机器人（MLIVRs）作为一种有前景的解决方案，能够安全地抓取预先存在的接口，如扶手和座椅轨道，从而在微重力环境中实现稳定的运动和强有力的操作。由于这些接口上的可抓取位置在空间上有限且离散分布，因此MLIVRs的运动规划必须与支撑点规划共同解决。本文提出了一种基于图论的MLIVRs同时路径和支撑点规划框架。所提出的方法高效地搜索多肢体机器人的可行站立序列，同时满足可操作性约束。通过在国际空间站（ISS）舱室的三维模型中进行模拟，验证了所提出框架的有效性，展示了其在现实的空间内环境中生成可行且高效的运动规划的能力。
cs.RO / 160 / 2609.35003
Learning to Act under Visual Interruptions with Vision-Language-Action Models
在视觉中断下学习行动的视觉-语言-行动模型
Abstract
Vision-language-action (VLA) models have demonstrated strong capabilities in robotic manipulation, but they are typically developed and evaluated with all camera streams available throughout task execution. When a camera stops delivering frames during task execution, the policy must continue acting without access to subsequent observations from the missing view. Despite its practical importance, how such interruptions affect closed-loop manipulation remains insufficiently understood. To investigate this problem, we introduce MAIL-Bench, a benchmark that evaluates visual interruptions with VLA models. By interrupting different cameras at multiple stages of each policy's successful reference trajectory, MAIL-Bench measures how well policies retain their capabilities when visual inputs become unavailable. Building on this benchmark, we propose MINT, which first trains VLA policies to remain functional under missing visual inputs. At inference time, MINT selectively supplements missing observations using optical-flow extrapolation or an action-conditioned world model, and withdraws predicted views when they become unreliable. Experiments on $\pi_{0.5}$ and GR00T N1.5 show that MINT significantly improves task success under camera loss over the original models. Experiments on AgiBot G2 further demonstrate the real-robot deployment under camera loss. The benchmark is available at https://minglejiang.github.io/Mail-Bench/
Chinese Translation
视觉-语言-行动（VLA）模型在机器人操作中表现出强大的能力，但通常是在任务执行过程中始终可用所有摄像头流的情况下进行开发和评估。当摄像头在任务执行过程中停止提供帧时，策略必须在没有来自缺失视角的后续观察的情况下继续行动。尽管这一问题在实际应用中具有重要意义，但这种中断对闭环操作的影响仍然理解不足。为了解决这一问题，我们引入了MAIL-Bench，这是一个评估VLA模型在视觉中断情况下表现的基准。通过在每个策略成功参考轨迹的多个阶段中中断不同的摄像头，MAIL-Bench测量在视觉输入不可用时，策略保持其能力的程度。在此基准的基础上，我们提出了MINT，该方法首先训练VLA策略在缺失视觉输入的情况下保持功能。在推理时，MINT使用光流外推或基于动作的世界模型选择性地补充缺失的观察，并在预测视图变得不可靠时撤回它们。在$ ext{π}_{0.5}$和GR00T N1.5上的实验表明，MINT在摄像头丢失情况下显著提高了任务成功率。对AgiBot G2的实验进一步展示了在摄像头丢失情况下的真实机器人部署。该基准可在https://minglejiang.github.io/Mail-Bench/获取。
cs.RO / 161 / 2609.35039
Do Not Cut When Uncertain: Rejectable and Calibrated Decision Heads for VLA Policies in Robotic Harvesting
在不确定时不要裁剪：用于机器人采摘的可拒绝和校准决策头的 VLA 策略
Abstract
Vision-Language-Action (VLA) policies trained with behavior cloning or flow matching are optimized to output an action trajectory, but they cannot express "I don't know" or "I should not act." In robotic harvesting, occlusion makes single-frame decisions fundamentally ambiguous: identical pixels can correspond either to a cuttable stem or to no stem at all. Existing VLAs are forced to commit, leading to high-confidence errors with irreversible consequences. We argue that the failure mode of a VLA is determined not by backbone scale but by its output interface. We propose Rejectable and Calibrated Decision Heads (RCDH), a typed, rejectable, and calibrated output interface that can be attached to a frozen VLA backbone without retraining or new features. RCDH introduces (i) a decision schema with explicit rejection and ordered, conditional decomposition, and (ii) a calibration procedure for risk-aware abstention. We evaluate RCDH on a robotic harvesting platform with controllable leaf occlusion, comparing generative, enumerated, calibrated, and rejectable interfaces. We show that replacing only the output head restores out-of-distribution usability under occlusion while preserving in-distribution performance. We further test whether the ordering of the rejection space is critical. Our results suggest that the right to refuse, rather than a larger model, is the missing interface for reliable manipulation under uncertainty.
Chinese Translation
使用行为克隆或流匹配训练的视觉-语言-动作（VLA）策略被优化为输出一个动作轨迹，但它们无法表达“我不知道”或“我不应该行动”。在机器人采摘中，遮挡使得单帧决策在本质上模糊：相同的像素可以对应于可裁剪的茎或根本没有茎。现有的 VLA 被迫做出承诺，导致高置信度的错误并产生不可逆的后果。我们认为，VLA 的失效模式并不是由主干网络的规模决定，而是由其输出接口决定。我们提出了可拒绝和校准决策头（Rejectable and Calibrated Decision Heads, RCDH），这是一种类型化的、可拒绝的、校准的输出接口，可以在不重新训练或添加新特征的情况下附加到冻结的 VLA 主干上。RCDH 引入了（i）具有明确拒绝和有序、条件分解的决策框架，以及（ii）用于风险意识弃权的校准程序。我们在一个具有可控叶片遮挡的机器人采摘平台上评估 RCDH，比较生成、枚举、校准和可拒绝的接口。我们展示了仅替换输出头可以在遮挡下恢复分布外的可用性，同时保持在分布内的性能。我们进一步测试拒绝空间的排序是否至关重要。我们的结果表明，拒绝的权利，而不是更大的模型，是在不确定性下可靠操作所缺失的接口。
cs.RO / 162 / 2609.35047
EMPIRIC: Experiment-Driven Learning of Residual World Models for Robot Planning
EMPIRIC：基于实验驱动的机器人规划残余世界模型学习
Abstract
A robot should be able to learn through experiments how unfamiliar objects behave and interact, then plan with that knowledge. It need not start from scratch: physics engines supply knowledge of motion and contact, but can omit entire mechanisms, such as glue curing, water heating, or wind. We present EMPIRIC, an agent that learns a residual world model: a physics engine extended with code for the missing mechanisms. The learned programs can introduce new forces, constraints, and hidden state, and Bayesian inference estimates their parameters and states from noisy observations. The resulting model lets the agent predict the outcomes of actions, choose informative experiments, and revise its hypotheses when predictions fail. Across five simulated domains, EMPIRIC learns interpretable, reusable models, and solves more tasks with fewer environment interactions than all three baselines. On a physical robot, it learns wind forces and domino masses to solve a manipulation task. Website and code: https://yichao-liang.github.io/empiric
Chinese Translation
机器人应能够通过实验学习不熟悉物体的行为和相互作用，然后利用这些知识进行规划。它无需从头开始：物理引擎提供运动和接触的知识，但可能会遗漏整个机制，例如胶水固化、水加热或风力。我们提出了EMPIRIC，一个学习残余世界模型的智能体：一个扩展了缺失机制代码的物理引擎。所学习的程序可以引入新的力、约束和隐藏状态，贝叶斯推断根据噪声观察估计它们的参数和状态。最终模型使得智能体能够预测动作的结果、选择信息丰富的实验，并在预测失败时修正其假设。在五个模拟领域中，EMPIRIC学习了可解释、可重用的模型，并在与所有三个基线相比时，以更少的环境交互解决了更多任务。在一个物理机器人上，它学习了风力和多米诺骨牌的质量，以解决一个操作任务。网站和代码：https://yichao-liang.github.io/empiric
cs.RO / 163 / 2609.35094
QuadHand: A Compact Quadrotor Aerial Manipulator with MRC-SDF-Based Whole-Body Motion Planning
QuadHand：一种基于MRC-SDF的紧凑型四旋翼空中操控器的全身运动规划
Abstract
Uncrewed aerial manipulators (UAMs) integrate robotic arms with aerial platforms for three-dimensional physical interaction. However, enlarging the workspace increases arm-induced disturbances, while existing geometric representations face a trade-off between geometric fidelity and computational efficiency in close-proximity interaction. This paper presents QuadHand, a compact quadrotor aerial manipulator with a 3-DoF arm, gripper, and battery-assisted passive CoG compensation module to reduce dominant arm-induced disturbances. We further propose MRC-SDF, a Multi-articulated Robot-Centric Signed Distance Field that preserves fine geometric detail with tractable computation, and a spatiotemporal whole-body trajectory optimization framework that jointly optimizes the quadrotor and manipulator for safe and executable trajectory generation. Simulations and real-world experiments demonstrate safe and executable aerial manipulation in complex environments.
Chinese Translation
无人机操控器（UAM）将机械臂与空中平台结合，实现三维物理交互。然而，扩大工作空间会增加机械臂引起的干扰，而现有的几何表示在近距离交互中面临几何保真度与计算效率之间的权衡。本文提出了QuadHand，一种紧凑型四旋翼空中操控器，配备3自由度的机械臂、夹持器和电池辅助的被动重心补偿模块，以减少主要的机械臂引起的干扰。我们进一步提出了MRC-SDF（多关节机器人中心签名距离场），该方法在可处理的计算下保留了细致的几何细节，并提出了一种时空全身轨迹优化框架，联合优化四旋翼和操控器，以生成安全且可执行的轨迹。仿真和现实世界实验表明，在复杂环境中实现了安全且可执行的空中操控。
cs.RO / 164 / 2609.35200
ReCAT: Remember, Count, and Time: Structured Recurrent Memory for Robot Manipulation
ReCAT：记忆、计数与时间：用于机器人操作的结构化递归记忆
Abstract
Memory-dependent manipulation requires robots to make decisions using information that is no longer available to their current sensors, such as recalling an earlier visual cue, tracking task progress, counting repeated events, or estimating elapsed time. We present ReCAT, a language-conditioned policy with structured recurrent memory. An instruction-conditioned encoder forms features from the current observation. A recurrent memory integrates the observation stream through Mamba-2 layers and one causal attention layer. A flow-matching Transformer decoder reads the current and the historical representation through separate cross-attention in every block. ReCAT reaches 95.3\% average success on LIBERO and 62.4\% on RMBench, with the best or tied-best result on six of nine tasks. On three real-robot tasks probing spatial recall, event counting, and interval timing, the best ReCAT variant reaches 66.7\% average success, against 8.3\% for the strongest short-history baseline. Controlled comparisons within ReCAT show that the observation encoder and every-block memory conditioning are needed for this performance. They also show that update rules developed for efficient sequence modeling behave differently as robot memory: additive updates have the highest observed success on counting and timing, and delta-rule updates on spatial recall. Project website is at https://intuitive-robots.github.io/ReCAT
Chinese Translation
依赖记忆的操作要求机器人利用当前传感器无法获得的信息做出决策，例如回忆早期的视觉线索、跟踪任务进展、计数重复事件或估计经过的时间。我们提出了ReCAT，一种具有结构化递归记忆的语言条件策略。指令条件编码器从当前观察中形成特征。递归记忆通过Mamba-2层和一个因果注意力层整合观察流。流匹配Transformer解码器通过每个块中的独立交叉注意力读取当前和历史表示。ReCAT在LIBERO上达到95.3%的平均成功率，在RMBench上达到62.4%，在九个任务中的六个任务上取得最佳或并列最佳结果。在三个真实机器人任务中，探测空间回忆、事件计数和时间间隔，最佳的ReCAT变体达到66.7%的平均成功率，而最强的短历史基线仅为8.3%。在ReCAT内部的控制比较显示，观察编码器和每个块的记忆条件对于这一性能是必需的。它们还表明，为高效序列建模开发的更新规则在机器人记忆中表现不同：加法更新在计数和计时方面观察到的成功率最高，而增量规则更新在空间回忆方面表现最佳。项目网站为 https://intuitive-robots.github.io/ReCAT
cs.RO / 165 / 2609.35231
Zero-Shot Reactive Obstacle Avoidance for Generative Robot Policies
生成机器人策略的零-shot反应式障碍物避让
Abstract
We propose NUDGE (Nudge Update via Differentiable GEometry), a training-free obstacle-avoidance procedure that can be incorporated in any robot policy based on diffusion or flow matching, including diffusion policies and vision-language-action models. Our work injects gradients from a signed distance field, a function returning each point's distance to the nearest obstacle, into the policy at inference time to steer it away from obstacles. It supports any common action parameterization, from absolute or relative joint poses to end-effector poses, through a differentiable joint-trajectory decoder. Experiments show that NUDGE preserves the policy's task distribution and runs reactively in real time.
Chinese Translation
我们提出了NUDGE（通过可微几何进行的推力更新），这是一种无训练的障碍物避让程序，可以融入任何基于扩散或流匹配的机器人策略中，包括扩散策略和视觉-语言-动作模型。我们的工作在推理时将来自有符号距离场的梯度注入到策略中，该函数返回每个点到最近障碍物的距离，以引导其远离障碍物。它支持任何常见的动作参数化，从绝对或相对关节姿态到末端执行器姿态，通过可微的关节轨迹解码器实现。实验表明，NUDGE保持了策略的任务分布，并能够实时反应。
cs.RO / 166 / 2609.35249
Spatial Grafting: Grounding 3D Features for Flow-Matching Robot Policies
空间嫁接：为流匹配机器人策略奠定三维特征基础
Abstract
Pretrained robot manipulation policies such as vision-language-action models (VLAs) or world-action models (WAMs) leave interaction-relevant metric geometry implicit. Recent breakthroughs in spatial reconstruction can supply the necessary geometry reliably, but their features describe local shape without stating where it lies with respect to the robot. How best to deliver these features to a pretrained policy remains unresolved. We propose Spatial Grafting, a versatile, lightweight spatial module that binds frozen reconstruction features to metric, robot-relative geometry. Spatial Grafting constructs metric-grounded spatial tokens and injects them into the flow-matching action expert through cross-attention, without modifying the host's perceptual pathway, so the host retains the full benefit of its pretraining. We evaluate it more broadly than any geometry-aware policy we compare against: one graft architecture, with no per-host redesign, on two VLAs and two WAMs, across four simulation benchmarks that span short-horizon manipulation, visual robustness, clutter and long-horizon mobile manipulation, and on three real-robot platforms with single- and dual-arm configurations. On RoboTwin 2.0, a dual-arm manipulation benchmark, the graft improves every host across VLAs and WAMs. Grafted $\pi_{0.5}$ gains 11.3% and 15.6% on clean and randomized scenes, reaching 94.0% and 92.4%, above the strongest published 3D-conditioned policy, WAM4D (93.8% and 89.9%). The margin widens as the horizon lengthens: on tasks from BEHAVIOR-1K, a dual-arm mobile manipulation challenge scored by average task progress, it surpasses the 2025 challenge winner on five of six tasks,by up to 0.47 Q-score, and exceeds a map-conditioned spatial policy on average across the three tasks both report.
Chinese Translation
预训练的机器人操作策略，如视觉-语言-动作模型（VLAs）或世界-动作模型（WAMs），将与交互相关的度量几何隐含在内。最近在空间重建方面的突破可以可靠地提供所需的几何信息，但其特征描述的是局部形状，而未说明其相对于机器人的位置。如何将这些特征有效地传递给预训练策略仍然没有解决方案。我们提出了空间嫁接（Spatial Grafting），这是一种多功能、轻量级的空间模块，将冻结的重建特征绑定到度量的、相对于机器人的几何信息上。空间嫁接构建了度量基础的空间标记，并通过交叉注意力将其注入到流匹配动作专家中，而不修改主机的感知路径，从而使主机保留其预训练的全部优势。我们在比较的几何感知策略中，对其进行了更广泛的评估：在两个VLAs和两个WAMs上，采用一种嫁接架构，无需对每个主机进行重新设计，涵盖了四个模拟基准，包括短期操作、视觉鲁棒性、杂乱环境和长期移动操作，以及在三个具有单臂和双臂配置的真实机器人平台上。在双臂操作基准RoboTwin 2.0上，嫁接提高了每个主机在VLAs和WAMs上的表现。嫁接后的$ ext{π}_{0.5}$在干净和随机场景上分别提高了11.3%和15.6%，达到了94.0%和92.4%，超过了最强的已发布3D条件策略WAM4D（93.8%和89.9%）。随着任务时间的延长，差距进一步扩大：在BEHAVIOR-1K的任务中，这是一项通过平均任务进展评分的双臂移动操作挑战，它在六个任务中的五个任务上超越了2025年挑战的获胜者，最高提高了0.47 Q分数，并在三个任务上平均超过了基于地图的空间策略。
cs.RO / 167 / 2609.35264
Repair Before You Fuse: Frozen-Host Adaptation for Corrupted-but-Present Sensors
修复再融合：针对损坏但存在传感器的冻结主机适应
Abstract
Camera-LiDAR detectors can continue to consume unreliable features even when both sensors remain present, synchronized, and calibrated. We introduce \emph{Boundary Feature Repair} (BFR), a frozen-host adaptation framework that learns task-supervised residual corrections at modality interfaces the detector already consumes. BFR-C repairs each camera feature level read by fusion, whereas BFR-L aligns host-conditioned LiDAR candidates to a selected boundary and routes site-wise innovations relative to the frozen anchor. Their jointly trained composition is BFR-CL. Zero-initialized per-channel scales make every variant an exact detector-level identity before optimization; only the repair modules train, while the encoders, fusion consumer, router, detection head, and host normalization statistics remain fixed. At inference, BFR requires neither clean references, corruption metadata, temporal history, nor online updates. Across the complete 20-corruption, five-severity KITTI-C grid, BFR-C reduces RCE from $14.07$ to $11.92$ on MVX-Net and from $14.29$ to $11.00$ on Focals Conv-F relative to their reproduced frozen baselines. On the latter host, BFR-L raises AP$_{\mathrm{cor}}$ from $73.65$ to $74.48$, while BFR-CL reaches $77.01$ AP$_{\mathrm{cor}}$ and $10.46$ RCE with $86.02$ clean AP. On nuScenes-R, BFR-CL raises the reproduced MoME baseline's mAP robustness ratio from $80.1$ to $81.4$. These results establish boundary repair as a targeted retrofit for corrupted-but-present sensing without retraining the deployed detector.
Chinese Translation
相机-激光雷达检测器即使在两个传感器都存在、同步且已校准的情况下，仍然可以继续使用不可靠的特征。我们提出了 extit{边界特征修复}（Boundary Feature Repair, BFR）框架，这是一种冻结主机适应框架，能够在检测器已使用的模态接口上学习任务监督的残差修正。BFR-C修复每个由融合读取的相机特征级别，而BFR-L则将主机条件的激光雷达候选对齐到选定的边界，并根据冻结的锚点进行现场创新的路由。它们的联合训练组合为BFR-CL。每个变体在优化前都是零初始化的逐通道尺度，使其成为精确的检测器级别身份；只有修复模块进行训练，而编码器、融合消费者、路由器、检测头和主机归一化统计保持固定。在推理时，BFR不需要干净的参考、损坏元数据、时间历史或在线更新。在完整的20种损坏、五种严重程度的KITTI-C网格上，BFR-C将MVX-Net的RCE从$14.07$降低到$11.92$，将Focals Conv-F的RCE从$14.29$降低到$11.00$，相对于其再现的冻结基线。在后者主机上，BFR-L将AP$_{ ext{cor}}$从$73.65$提升至$74.48$，而BFR-CL则达到了$77.01$的AP$_{ ext{cor}}$和$10.46$的RCE，且干净AP为$86.02$。在nuScenes-R上，BFR-CL将再现的MoME基线的mAP鲁棒性比率从$80.1$提升至$81.4$。这些结果确立了边界修复作为一种针对损坏但存在的传感的有针对性的改造，而无需重新训练已部署的检测器。
cs.RO / 168 / 2609.35267
GuardPIBT: Counterfactually Gated Neural Guidance for Ultra-Large-Scale 3D Multi-Agent Path Finding
GuardPIBT：用于超大规模3D多智能体路径规划的反事实门控神经引导
Abstract
Large-scale 3D multi-agent path finding becomes increasingly difficult under dense traffic. Priority Inheritance with Backtracking (PIBT) scales well, but its one-step goal-directed ordering may become insufficient under dense interactions and large-scale congestion. We present GuardPIBT, which augments rather than replaces the PIBT executor: neural predictions only propose residual reorderings of PIBT's native candidates, while final actions remain determined by PIBT. First, local graph attention models nearby interactions, while global source--goal transport features provide population-level coordination context for candidate reordering. Second, a counterfactual group gate filters reorderings whose closed-loop effects may degrade coordination. Third, for ultra-large populations, population-adaptive grouping preserves decision granularity, asynchronous cached inference amortizes neural computation, and selective repair resolves long-tail agents. PIBT retains validity checking, priority inheritance, and backtracking throughout. Experiments with up to 100,000 agents demonstrate reliable completion across 2D and 3D environments, including all three 100,000-agent warehouse runs with zero audited graph violations. The project website is available at {\color{magenta}\texttt{https://guardpibt.github.io/GuardPIBT/}}.
Chinese Translation
在密集交通下，大规模3D多智能体路径规划变得越来越困难。带回溯的优先继承（Priority Inheritance with Backtracking，PIBT）具有良好的扩展性，但其一步目标导向的排序在密集交互和大规模拥堵下可能显得不足。我们提出了GuardPIBT，它增强而非替代PIBT执行器：神经网络预测仅提出PIBT原生候选项的剩余重排序，而最终动作仍由PIBT决定。首先，局部图注意力模型处理附近的交互，而全局源-目标传输特征为候选重排序提供了群体级协调上下文。其次，反事实组门控过滤那些可能降低协调的重排序。第三，对于超大规模群体，适应群体的分组保持决策粒度，异步缓存推理摊销神经计算，选择性修复解决长尾智能体问题。PIBT在整个过程中保留有效性检查、优先继承和回溯。对多达100,000个智能体的实验表明，在2D和3D环境中可靠完成，包括所有三个100,000智能体的仓库运行，且未出现任何审计图违规。项目网站可访问于{ exttt{https://guardpibt.github.io/GuardPIBT/}}。
cs.RO / 169 / 2609.35318
DexAgent: An Agentic Human2Sim2Robot Framework for Dexterous Manipulation with Self-Evolving Tool Library
DexAgent：一个用于灵巧操作的自主人类-模拟-机器人框架，配备自我演化工具库
Abstract
Human videos offer a scalable source of demonstrations for dexterous robot manipulation. However, existing human-to-simulation-to-robot (Human2Sim2Robot) pipelines rely on predefined procedures that struggle to accommodate diverse object properties and interactions, particularly those involving articulated and deformable objects. We introduce DexAgent, an agentic Human2Sim2Robot framework that converts a single egocentric human video and a task prompt into physically grounded robot trajectories for policy training. It operates through four stages: semantic understanding of human videos, property-based simulation reconstruction, robot trajectory optimization, and robot data generation. At each stage, DexAgent adapts its approach to the task and object properties by selecting suitable skills from its tool library or developing new ones when needed. Property-specific verifiers assess stage outcomes for physical validity and task-specific requirements and provide feedback for refinement, preventing error propagation through the workflow. This adaptive, verification-guided process allows DexAgent to process diverse objects and long-horizon tasks. In the final stage, DexAgent varies object and robot states in simulation to generate diverse robot trajectories from a single human video, then retextures the rendered observations to facilitate sim-to-real transfer. Newly developed skills and verifiers are retained in its tool library, making it self-evolving to accumulate reusable capabilities. This reduces processing time as DexAgent encounters more human videos. Across eleven real-world tasks, policies trained with DexAgent-generated data achieve a 3.5x higher success rate than competing baselines. Project website: https://dexagent123.github.io/.
Chinese Translation
人类视频为灵巧机器人操作提供了可扩展的演示来源。然而，现有的人类-模拟-机器人（Human2Sim2Robot）管道依赖于预定义的程序，难以适应多样的物体属性和交互，尤其是涉及关节和可变形物体的情况。我们提出了DexAgent，一个自主的人类-模拟-机器人框架，它将单个自我中心的人类视频和任务提示转换为用于策略训练的物理基础机器人轨迹。该框架通过四个阶段运作：对人类视频的语义理解、基于属性的模拟重建、机器人轨迹优化和机器人数据生成。在每个阶段，DexAgent根据任务和物体属性调整其方法，通过从工具库中选择合适的技能或在需要时开发新技能来实现。属性特定的验证器评估每个阶段的结果，以确保物理有效性和任务特定要求，并提供反馈以进行改进，从而防止错误在工作流程中传播。这一自适应、验证引导的过程使DexAgent能够处理多样化的物体和长时间跨度的任务。在最后阶段，DexAgent在模拟中改变物体和机器人状态，以从单个的人类视频生成多样的机器人轨迹，然后重新纹理化渲染的观察结果，以促进模拟到现实的转移。新开发的技能和验证器被保留在其工具库中，使其具备自我演化的能力，以积累可重用的能力。这在DexAgent处理更多人类视频时减少了处理时间。在十一项真实世界任务中，使用DexAgent生成的数据训练的策略成功率比竞争基线高出3.5倍。项目网站：https://dexagent123.github.io/
cs.RO / 170 / 2609.35375
From Pixel to Poses: Object-centric Tool Manipulation Learning from Human Demonstrations
从像素到姿态：基于对象的工具操作学习来自人类示范
Abstract
Scaling up robotic manipulation is primarily bottlenecked by the scarcity of real-world robot data. While recent approaches leverage human video demonstrations to mitigate this shortage, they remain computationally expensive and still rely on paired human-robot data for domain alignment. Although current state-of-the-arts excel at long-horizon tasks, they struggle with the delicate and precise control required for complex tool manipulation. To overcome these limitations, we introduce P2P-T, from Pixel to Poses for Tool Manipulation, a data-efficient, object-centric framework that learns tool use directly from human demonstrations. P2P-T bridges the cognitive and physical execution gap through a two-stage approach. First, pretraining an object-centric world model to extract stable pose priors; second, integrating these priors into an efficient, pose-aware low-level policy. By utilizing a robust automated data processing pipeline powered by modern foundation models, P2P-T completely bypasses the need for human-robot aligned data. This reduces overall training overhead drastically. With minimal per-task fine-tuning, our framework achieves a 73% improvement over the previous state of the art in execution performance on complex, real-world tool manipulation tasks that currently remain out of reach for standard large-scale pretrained models.
Chinese Translation
扩展机器人操作的主要瓶颈在于现实世界机器人数据的稀缺。尽管最近的方法利用人类视频示范来缓解这一短缺，但它们仍然计算成本高，并且依赖于配对的人类-机器人数据进行领域对齐。尽管当前的最先进技术在长时间任务中表现出色，但在复杂工具操作所需的精细和精准控制方面仍然面临挑战。为了克服这些限制，我们提出了P2P-T（从像素到姿态的工具操作），这是一种数据高效的、以对象为中心的框架，能够直接从人类示范中学习工具使用。P2P-T通过两阶段的方法弥合了认知和物理执行之间的差距。首先，预训练一个以对象为中心的世界模型，以提取稳定的姿态先验；其次，将这些先验整合到一个高效的、姿态感知的低级策略中。通过利用由现代基础模型驱动的强大自动化数据处理管道，P2P-T完全绕过了对人类-机器人对齐数据的需求。这大大减少了整体训练开销。在每个任务上进行最小的微调后，我们的框架在复杂的现实世界工具操作任务的执行性能上实现了比之前的最先进技术提高73%的成绩，而这些任务目前仍然超出标准大规模预训练模型的能力范围。
cs.RO / 171 / 2609.35415
Adaptive Safety Filtering for Frozen ACC Policies via Conformal Residual Calibration
通过符合残差校准的自适应安全过滤器用于冻结自适应巡航控制策略
Abstract
Frozen adaptive cruise control (ACC) policies can violate constraints when deployment dynamics differ from their training conditions. We propose residual-aware conformal action filtering (RACF), which calibrates residuals of a fixed nominal predictor and converts their quantile into an operating margin for finite-model action projection. Completed transitions update margins and candidate selection without retraining the policy. In a registered comparison over 2,400 controller-trial units, Adaptive RACF achieves 94.3% episode safety, improving by 19.9 percentage points over the evaluated nominal CBF-QP baseline while reducing projection frequency from 8.11% to 6.63%. A controlled study isolates a 4.54-point improvement from residual-margin injection. In a separate matched-hardware evaluation, Adaptive reduces mean amortized rollout time by 21.2% relative to Robust CBF-QP, with 161/180 versus 170/180 safe episodes. We characterize conditions linking one-step residual coverage to constraint satisfaction and quantify the observed safety-computation trade-offs.
Chinese Translation
冻结的自适应巡航控制（ACC）策略在部署动态与其训练条件不同时可能会违反约束。我们提出了一种残差感知的符合动作过滤（RACF），该方法校准固定名义预测器的残差，并将其分位数转换为有限模型动作投影的操作边际。完成的过渡在不重新训练策略的情况下更新边际和候选选择。在对2400个控制器试验单元的注册比较中，自适应RACF实现了94.3%的情节安全性，相比评估的名义CBF-QP基线提高了19.9个百分点，同时将投影频率从8.11%降低到6.63%。一项受控研究将残差边际注入带来的改进隔离为4.54点。在一项单独的匹配硬件评估中，自适应方法相对于稳健CBF-QP减少了21.2%的平均摊销展开时间，安全情节数量为161/180对比170/180。我们描述了将一步残差覆盖与约束满足联系起来的条件，并量化了观察到的安全性与计算之间的权衡。
cs.RO / 172 / 2609.35431
Memory in the Sky: Low-Altitude Question Answering with Multi-Agent Memory Aggregation
天空中的记忆：基于多智能体记忆聚合的低空问答
Abstract
This paper studies low-altitude question answering (LAQA), in which distributed unmanned aerial vehicle (UAV) memories are aggregated at a ground server to answer questions about observations over a long horizon. Unlike conventional resource allocation based on sensing, communication, control, or computation metrics, LAQA requires an explicit measure of memory value. We propose a generative adversarial exam (GAE) that uses forward simulation to evaluate memory retrieval and exam scores to quantify memory quality. This enables the downstream QA value of candidate memories to be measured and optimized without accessing the internal mechanisms of the black-box captioning, retrieval, and reasoning pipeline. Building on this metric, we develop a memory-centric (MemCen) framework that jointly selects UAVs and allocates transmit power to maximize memory quality under communication constraints. In the noise-limited regime, we derive a QoM-aware capped water-filling law that explicitly connects task utility with physical-layer power allocation. We further develop penalty successive optimization (PSO) and learning to memorize (L2M) solvers. MemCen achieves QA accuracies of 92.4% and 84.0% in CARLA Town04 and Town05 under static and dynamic communication conditions, respectively. In real-world experiments, MemCen achieves 88.5% QA accuracy on the panoramic multi-agent system (PMAS) benchmark. Finally, UAV-to-robot-dog demonstrations further validate the practical utility of the acquired memories for environmental understanding and navigation.
Chinese Translation
本文研究了低空问答（LAQA），在该过程中，分布式无人机（UAV）记忆在地面服务器上聚合，以回答关于长期观察的问题。与基于传感、通信、控制或计算指标的传统资源分配不同，LAQA 需要明确的记忆价值度量。我们提出了一种生成对抗考试（GAE），利用前向模拟来评估记忆检索，并通过考试分数量化记忆质量。这使得候选记忆的下游问答（QA）价值可以在不访问黑箱字幕、检索和推理管道的内部机制的情况下进行测量和优化。在此度量的基础上，我们开发了一个以记忆为中心的框架（MemCen），该框架共同选择无人机并分配传输功率，以在通信约束下最大化记忆质量。在噪声限制的情况下，我们推导出了一种关注质量（QoM）的限制水填充法则，明确将任务效用与物理层功率分配联系起来。我们进一步开发了惩罚性连续优化（PSO）和学习记忆（L2M）求解器。MemCen 在 CARLA Town04 和 Town05 的静态和动态通信条件下分别实现了 92.4% 和 84.0% 的问答准确率。在真实世界实验中，MemCen 在全景多智能体系统（PMAS）基准上实现了 88.5% 的问答准确率。最后，无人机与机器人狗的演示进一步验证了所获取记忆在环境理解和导航中的实际效用。
cs.RO / 173 / 2609.35432
Self-Evolving Coding Agents: From Digital Programs to Physical-World Intelligence
自我进化编码代理：从数字程序到物理世界智能
Abstract
Vision-language-action (VLA) and world-action (WAM) models map observations and instructions directly to robot actions. This directness ties a policy to training: minor layout or viewpoint changes cause failure, and instructions generalize poorly. The root cause lies in representation: task requirements, conditions, progress, and failure recovery are implicitly encoded in action sequences, making them difficult to inspect or revise. Digital coding agents offer a precedent: LLMs call tools, verify results, and revise from feedback as executable code. The same working pattern of explicit state, manageable execution, and revisable procedures underlies generalization and long-horizon execution in the physical world, letting physical experience return as reusable programs, memory, or evidence. We propose Physical Coding, representing task state and execution as code. Code as World records objects, relations, constraints, and progress; Code as Policy organizes planning, verification, recovery, and execution. We build HexaAnything, which calls perception, planning, and control tools, including VLA/WAM policies, and makes in-the-loop decisions from external feedback. Verified traces become data and memory, enabling evolution from tools and Harness to model weights, architectures, and ultimately hardware and task design. On RoboCasa365, HexaAnything improves Composite-Unseen and overall success over XR-1 VLA, and its Harness-trained HexaModel beats the base on every split, indicating code traces internalize physical execution. On PhyBench and a dual-arm AgileX robot, the agent autonomously completes physics experiments and most tabletop tasks, often faster than published results. We observe data, model, and tool self-evolution; future work targets weight internalization, autonomous redesign of architectures, languages, representations, and tasks, and deployment in manufacturing and science.
Chinese Translation
视觉-语言-行动（VLA）和世界-行动（WAM）模型将观察和指令直接映射到机器人动作。这种直接性将策略与训练紧密联系：轻微的布局或视角变化会导致失败，而指令的泛化能力较差。根本原因在于表示：任务要求、条件、进展和失败恢复隐含地编码在动作序列中，使其难以检查或修订。数字编码代理提供了一个先例：大型语言模型（LLMs）调用工具、验证结果，并根据反馈修订可执行代码。明确的状态、可管理的执行和可修订的程序的相同工作模式支撑着物理世界中的泛化和长时间执行，使得物理经验可以作为可重用的程序、记忆或证据返回。我们提出了物理编码，将任务状态和执行表示为代码。代码作为世界记录对象、关系、约束和进展；代码作为策略组织规划、验证、恢复和执行。我们构建了HexaAnything，它调用感知、规划和控制工具，包括VLA/WAM策略，并根据外部反馈做出循环内决策。经过验证的轨迹成为数据和记忆，使得从工具和Harness演变到模型权重、架构，最终到硬件和任务设计成为可能。在RoboCasa365上，HexaAnything在Composite-Unseen和整体成功率上优于XR-1 VLA，其Harness训练的HexaModel在每个分割上都超越了基础模型，表明代码轨迹内化了物理执行。在PhyBench和双臂AgileX机器人上，代理自主完成物理实验和大多数桌面任务，往往比已发布的结果更快。我们观察到数据、模型和工具的自我进化；未来的工作将针对权重内化、架构、语言、表示和任务的自主重设计，以及在制造和科学中的部署。
cs.RO / 174 / 2609.35439
Revision, Not Restart: Revisable Visual Plans for Closed-Loop World-Action Models
修订，而非重启：可修订的视觉计划用于闭环世界-行动模型
Abstract
World-action models use predicted visual futures to condition robot actions, yet execution feedback can invalidate parts of a prediction while leaving its task structure useful. We propose Revisable Temporal Planning (RTP), which maintains the visual future as a persistent action condition and revises it after feedback. Its central mechanism is a learned revision bridge: it resumes an intermediate state saved during visual generation and adapts its continuation to current observations. Visual and action supervision connect this revision to subsequent control. Time-aware history supplies observed evidence, and an adaptive policy selects retention, bridge revision, or fresh replanning from new noise before decoding the next action. On RoboMME and RMBench, RTP achieves task-averaged success rates of 48.6% and 84.8%, respectively. Matched comparisons support learned continuation; estimated checkpoint-source and action-prefix effects are positive but less precisely resolved. These results connect feedback-driven visual-plan revision to closed-loop task performance. Project Page: https://PLACEHOLDER.github.io/RTP/
Chinese Translation
世界-行动模型利用预测的视觉未来来调节机器人动作，但执行反馈可能会使预测的部分失效，同时保留其任务结构的有效性。我们提出了可修订的时间规划（Revisable Temporal Planning, RTP），该方法将视觉未来作为持久的动作条件，并在反馈后进行修订。其核心机制是一个学习的修订桥接：它恢复在视觉生成过程中保存的中间状态，并将其延续适应当前的观察。视觉和动作监督将这一修订与后续控制连接起来。时间感知历史提供观察到的证据，而自适应策略则在解码下一个动作之前，从新噪声中选择保留、桥接修订或全新规划。在RoboMME和RMBench上，RTP分别实现了任务平均成功率48.6%和84.8%。匹配比较支持学习的延续；估计的检查点源和动作前缀效应为正，但精确度较低。这些结果将基于反馈的视觉计划修订与闭环任务表现联系起来。项目页面：https://PLACEHOLDER.github.io/RTP/
cs.RO / 175 / 2609.35450
Uni-VLaT: Whole-Body Tactile Adaptation of VLA Policies for Humanoid Loco-Manipulation
Uni-VLaT：类人机器人运动操控的全身触觉适应VLA策略
Abstract
Physical contact often determines how a humanoid should respond during loco-manipulation, yet vision and proprioception alone are often insufficient to characterize physical interaction, especially when the contact region is occluded. Unlike sparse force or torque measurements at predefined regions, distributed tactile sensing preserves spatially resolved contact patterns across the robot body. We therefore study how to integrate such whole-body tactile information into vision-language-action (VLA) policies for contact-rich control. Our approach, Uni-VLaT, introduces a tactile pathway whose latent state is trained not only for action generation, but also to predict future tactile, proprioceptive, and visual representations. This predictive objective builds a tactile-anchored multimodal context, encouraging a more structured understanding of the physical world. We evaluate Uni-VLaT on five real-robot tasks covering tactile-triggered locomotion, sustained physical interaction, human-robot contact, and loco-manipulation. Uni-VLaT achieves a 75% average success rate, outperforming a baseline without tactile input by 43 points and a tactile-input baseline without predictive supervision by 7 points. Across two pretrained VLA backbones, our method improves Table Sweeping by 30 points on both backbones and Back-Tap Walking by 85-90 points. Ablations further show that contextualized tactile prediction and absolute future targets are critical to performance. These results indicate that predictive tactile learning provides an effective route for extending pretrained VLA policies to whole-body physical interaction.
Chinese Translation
物理接触通常决定了类人在运动操控过程中应如何响应，然而仅靠视觉和本体感觉往往不足以表征物理交互，尤其是在接触区域被遮挡时。与在预定义区域进行稀疏的力或扭矩测量不同，分布式触觉传感能够保留机器人身体上空间分辨的接触模式。因此，我们研究如何将这种全身触觉信息整合到视觉-语言-动作（VLA）策略中，以实现丰富接触的控制。我们的方法Uni-VLaT引入了一条触觉通道，其潜在状态不仅用于动作生成，还用于预测未来的触觉、本体感觉和视觉表征。这个预测目标构建了一个以触觉为锚的多模态上下文，促进对物理世界的更结构化理解。我们在五个真实机器人任务上评估了Uni-VLaT，这些任务涵盖了触觉触发的运动、持续的物理交互、人机接触和运动操控。Uni-VLaT实现了75%的平均成功率，超越了没有触觉输入的基线43个百分点，以及没有预测监督的触觉输入基线7个百分点。在两个预训练的VLA骨干网络上，我们的方法使桌面清扫任务的表现提升了30个百分点，后续行走任务提升了85-90个百分点。消融实验进一步表明，上下文化的触觉预测和绝对未来目标对性能至关重要。这些结果表明，预测性触觉学习为将预训练的VLA策略扩展到全身物理交互提供了一条有效的途径。
cs.RO / 176 / 2609.35469
Rethinking Causal Action Tokenization with Conditional Annealing in Flow Matching
通过条件退火重新思考流匹配中的因果动作标记化
Abstract
Autoregressive Vision-Language-Action (VLA) models offer a scalable path to robot learning, yet existing action tokenizers treat tokenization as a compression problem, producing representations that are semantically misaligned with the autoregressive backbone. We propose CATok, a causal action tokenizer that reframes tokenization as a causally structured generative process. CATok introduces a conditional annealing mechanism that extracts action tokens by progressively annealing a flow-matching process: each token is conditioned on all preceding tokens and encodes the residual reconstruction signal at a specific noise level, establishing a coarse-to-fine causal token space whose generative semantics are structurally aligned with autoregressive modeling. A token-conditioned flow-matching decoder built on Multimodal Diffusion Transformer (MMDiT) reconstructs continuous action chunks from these discrete tokens with the precision of hybrid diffusion-head architectures. This discrete bottleneck enforces knowledge insulation by design, cleanly separating high-level semantic reasoning from low-level motor execution without requiring explicit attention masking. Extensive evaluations across three simulation benchmarks and real-world robotic manipulation tasks demonstrate that CATok consistently surpasses existing tokenization methods in both reconstruction fidelity-compression tradeoff and inference efficiency, while improving VLA task success rate and training efficiency, establishing a high-performance, scalable foundation for purely autoregressive VLA systems.
Chinese Translation
自回归视觉-语言-动作（VLA）模型为机器人学习提供了一条可扩展的路径，但现有的动作标记器将标记化视为压缩问题，产生的表示在语义上与自回归主干不一致。我们提出了CATok，一种因果动作标记器，它将标记化重新构想为一种因果结构生成过程。CATok引入了一种条件退火机制，通过逐步退火流匹配过程来提取动作标记：每个标记都以所有前置标记为条件，并在特定噪声水平上编码残余重建信号，从而建立一个粗到细的因果标记空间，其生成语义在结构上与自回归建模对齐。基于多模态扩散变换器（Multimodal Diffusion Transformer, MMDiT）构建的标记条件流匹配解码器，从这些离散标记中重建连续的动作块，具有混合扩散头架构的精确性。这个离散瓶颈通过设计强制知识隔离，干净地将高层语义推理与低层运动执行分离，而无需显式的注意力屏蔽。在三个仿真基准和真实世界机器人操作任务中的广泛评估表明，CATok在重建保真度-压缩权衡和推理效率方面始终超越现有的标记化方法，同时提高了VLA任务的成功率和训练效率，为纯自回归VLA系统建立了高性能、可扩展的基础。
cs.RO / 177 / 2609.35476
CoBrush: A Hierarchical Planning Framework for Human-Robot Co-Painting
CoBrush：一种用于人机协同绘画的分层规划框架
Abstract
Embodied co-painting requires a robot to repeatedly update a shared physical canvas while human intent evolves over interaction. Existing reference-driven painters or reactive assistants are typically optimized for single-shot rendering or sketch completion, limiting their ability to sustain coherent multi-round collaboration or to construct complex, content-rich scenes over time. We present CoBrush, a hierarchical framework that formulates multi-round co-painting as a coordinated semantic, spatial, and execution process. By separating high-level intent inference from spatial grounding and stroke-level control, the system supports progressive scene development on real acrylic canvases. We evaluate the framework through real human-robot painting sessions, stress tests, and user studies. Compared to single-turn baselines, our approach achieves stronger semantic alignment, more stable spatial progression, and higher perceived plausibility of robot actions. These results demonstrate that structured multi-stage reasoning improves the coherence and robustness of interactive painting and supports the progressive development of content-rich physical artworks.
Chinese Translation
具身协同绘画要求机器人在与人类互动的过程中，持续更新共享的物理画布，同时人类的意图也在不断演变。现有的参考驱动型绘画工具或反应式助手通常针对单次渲染或草图完成进行了优化，这限制了它们在持续一致的多轮协作或随着时间构建复杂、内容丰富场景的能力。我们提出了CoBrush，一个将多轮协同绘画形式化为协调的语义、空间和执行过程的分层框架。通过将高层意图推断与空间基础和笔画级控制分离，该系统支持在真实的丙烯画布上逐步发展场景。我们通过真实的人机绘画会话、压力测试和用户研究对该框架进行了评估。与单轮基线相比，我们的方法实现了更强的语义对齐、更稳定的空间进展以及更高的机器人行为的可信度。这些结果表明，结构化的多阶段推理提高了互动绘画的连贯性和鲁棒性，并支持内容丰富的物理艺术作品的逐步发展。
cs.RO / 178 / 2609.35479
Robot Tool Design from Scratch via Behavior-Aware Hierarchical Optimization
通过行为感知的分层优化从零开始设计机器人工具
Abstract
The ability to design a tool for a task marks a level of intelligence beyond merely understanding, selecting, or using one. Existing methods for robotic tool design typically optimize a tool's continuous shape and action within a structure that is prescribed or generated beforehand, so the structure itself stays outside the physical optimization loop. We study task-driven tool design from scratch, where tool structure, shape, and action are all derived from the desired physical outcome. Here we show that the three elements can be designed jointly by HOT, a hierarchical optimization whose upper level searches over discrete tool structures with BASS, while lower-level physical optimization evaluates their task behavior and returns milestone progress as behavioral evidence for the search, ultimately providing jointly optimized shape and action. On four tool-use tasks with distinct physical functions, HOT discovers functional structures after evaluating only a small fraction of search spaces containing up to 56 million structures, and the subsequent refinement of their geometry lowers the task loss on all tasks while preserving success, through deformations that are functionally interpretable. Once 3D printed, the tools accomplish all tasks on a real robot with the actions found in simulation. Designing tools from required physical effects, rather than a catalog of known tools, is a step toward the open-ended tool making seen in humans and animals.
Chinese Translation
设计工具以完成特定任务的能力标志着一种超越单纯理解、选择或使用工具的智能水平。现有的机器人工具设计方法通常在预先规定或生成的结构框架内优化工具的连续形状和动作，因此结构本身并不在物理优化循环中。我们研究从零开始的任务驱动工具设计，其中工具的结构、形状和动作均源于所需的物理结果。在这里，我们展示了这三个元素可以通过HOT（分层优化）共同设计，其上层在离散工具结构上进行搜索，使用BASS（行为自适应搜索策略），而下层物理优化则评估其任务行为并返回里程碑进展作为搜索的行为证据，最终提供共同优化的形状和动作。在四个具有不同物理功能的工具使用任务中，HOT在仅评估了包含多达5600万个结构的搜索空间的一小部分后发现了功能性结构，随后对其几何形状的细化降低了所有任务的任务损失，同时保持了成功率，通过功能上可解释的变形。一旦3D打印，这些工具在真实机器人上完成了在仿真中找到的所有任务。从所需的物理效果出发设计工具，而不是依赖于已知工具的目录，这是向人类和动物所见的开放式工具制造迈出的重要一步。
cs.RO / 179 / 2609.35482
ForVis: An In-Field Dataset and Benchmark for VIO Using Under-Canopy UAV Flights in Forests
ForVis：用于森林中无人机飞行的视觉惯性同时定位与地图构建（VIO）的现场数据集和基准测试
Abstract
Visual-inertial Simultaneous Localization and Mapping (VI-SLAM) for UAVs remains difficult to evaluate in real forest environments, where motion, illumination changes, repetitive vegetation, and vibration can all affect estimation. We present ForVis, an in-field dataset and benchmark for evaluating VI-SLAM during UAV flight in forest environments. The dataset contains twelve flights across open meadow, above-canopy, and under-canopy conditions in each environment. In total, it provides 563.8s of flight over 1096.8m of trajectory, recorded simultaneously with an Intel RealSense D435i and an OAK-D Pro Wide together with inertial and flight-controller data. We benchmark seven open-source VI-SLAM systems over 504 runs. The results show that sensor choice has a larger effect on trajectory error than the spread between algorithms: all seven methods achieve lower median error on the OAK-D Pro than on the D435i. ForVis is intended to support evaluation of speed, accuracy and robustness for VI-SLAM in challenging forest flight.
Chinese Translation
在真实森林环境中评估无人机的视觉惯性同时定位与地图构建（VI-SLAM）仍然具有挑战性，因为运动、光照变化、重复的植被和振动都会影响估计。我们提出了ForVis，这是一个用于评估森林环境中无人机飞行时VI-SLAM的现场数据集和基准测试。该数据集包含在每种环境中开放草地、树冠上方和树冠下方条件下的十二次飞行。总共提供了563.8秒的飞行数据，覆盖1096.8米的轨迹，同时记录了Intel RealSense D435i和OAK-D Pro Wide的图像数据，以及惯性和飞行控制器数据。我们对七个开源VI-SLAM系统进行了504次运行的基准测试。结果表明，传感器选择对轨迹误差的影响大于算法之间的差异：所有七种方法在OAK-D Pro上的中位误差均低于在D435i上的表现。ForVis旨在支持在具有挑战性的森林飞行环境中对VI-SLAM的速度、准确性和鲁棒性进行评估。
cs.RO / 180 / 2609.35493
Terrain-Aware Autonomous Planetary Exploration for Exteroceptive-Proprioceptive Mapping with Quadruped Scouts
基于地形感知的自主行星探索：四足侦察机器人在外部感知-本体感知映射中的应用
Abstract
Autonomous planetary exploration requires robots to navigate unknown, uneven terrain while assessing risk, traversability, and energetic cost. Quadruped scouts are well suited for this task because they can traverse irregular surfaces and gather mobility-relevant information during locomotion. This paper presents a terrain-aware exploration framework that combines exteroceptive and proprioceptive mapping for a quadruped robot in lunar-like environments. An onboard RGB-D camera builds robot-centered elevation maps, estimates geometric traversability, and derives navigation costs for autonomous planning. In parallel, proprioceptive measurements provide interaction-aware terrain cues that complement geometry-based assessment. Local maps are incrementally registered into a global multi-layer representation, which is used by an exploration module to select targets in unexplored regions of interest. The targets are reached by an autonomous navigation system that guides collision-aware motion using the available map and cost layers. Simulation results on NVIDIA Isaac Sim show autonomous exploration, map expansion, and spatial association between terrain geometry and robot-terrain interaction. Subsequent navigation using this information exhibits lower average Cost of Transport (CoT) than initial exploration.
Chinese Translation
自主行星探索要求机器人在未知且不平坦的地形中导航，同时评估风险、可通行性和能量成本。四足侦察机器人非常适合这一任务，因为它们能够在不规则表面上行走并在运动过程中收集与移动相关的信息。本文提出了一种地形感知探索框架，结合了外部感知和本体感知映射，适用于类月环境中的四足机器人。机载RGB-D相机构建以机器人为中心的高程图，估计几何可通行性，并推导出自主规划的导航成本。同时，本体感知测量提供了与交互相关的地形线索，以补充基于几何的评估。局部地图逐步注册到全球多层表示中，供探索模块选择未探索的感兴趣区域的目标。目标通过自主导航系统到达，该系统利用可用的地图和成本层引导碰撞感知运动。在NVIDIA Isaac Sim上的仿真结果显示了自主探索、地图扩展以及地形几何与机器人-地形交互之间的空间关联。后续使用这些信息的导航显示出比初始探索更低的平均运输成本（Cost of Transport, CoT）。
cs.RO / 181 / 2609.35516
Inspection-SPARS: Task-Oriented Sparse Roadmaps for Inspection Planning
Inspection-SPARS：面向任务的稀疏路线图用于检查规划
Abstract
Inspection planning seeks a minimum-length collision-free robot tour that observes a given set of points of interest (POIs). Sampling-based methods reduce this continuous problem to a graph inspection planning (GIP) problem over a discrete roadmap, which is then solved using combinatorial solvers. Dense roadmaps capture diverse inspection viewpoints and motion shortcuts, and thus admit higher-quality solutions, but they induce large combinatorial search spaces on which state-of-the-art GIP solvers struggle to find good solutions within practical time budgets. Roadmap sparsification---restructuring a dense roadmap into a compact representation that preserves connectivity and path lengths---can alleviate this burden. However, existing sparsification approaches are either agnostic to the underlying inspection task, or strive to ensure coverage of the POIs without accounting for the quality of the resulting inspection plan. We present Inspection-SPARS, which is, to our knowledge, the first inspection-roadmap sparsifier with POI coverage and path-quality guarantees relative to the dense roadmap. To this end, we generalize the SPARS framework, a popular task-agnostic sparsifier, from purely geometric criteria to task-oriented ones, introducing an inspection-aware vertex admission mechanism that treats POI coverage as a first-class sparsification criterion alongside connectivity and path quality. Experiments in realistic 3D environments show that Inspection-SPARS reduces vertex and edge counts by 4-8x while preserving coverage, allowing the GIP solver to compute tours up to 25% shorter than with the dense roadmap or state-of-the-art inspection roadmap. More broadly, Inspection-SPARS shows that sparsification can be made task-aware without sacrificing guarantees on solution quality.
Chinese Translation
检查规划旨在寻找一条最短的无碰撞机器人巡检路径，以观察给定的一组兴趣点（POIs）。基于采样的方法将这一连续问题简化为离散路线图上的图形检查规划（GIP）问题，然后使用组合求解器进行求解。密集路线图捕捉了多样的检查视角和运动捷径，因此能够提供更高质量的解决方案，但它们会引入大型组合搜索空间，使得最先进的 GIP 求解器在实际时间预算内难以找到良好的解决方案。路线图稀疏化——将密集路线图重构为保留连通性和路径长度的紧凑表示——可以减轻这一负担。然而，现有的稀疏化方法要么对基础检查任务无感，要么努力确保覆盖 POIs，而不考虑最终检查计划的质量。我们提出了 Inspection-SPARS，据我们所知，这是第一个相对于密集路线图具有 POI 覆盖和路径质量保证的检查路线图稀疏化器。为此，我们将流行的任务无关稀疏化框架 SPARS 从纯几何标准推广到面向任务的标准，引入了一种检查感知的顶点接纳机制，将 POI 覆盖视为与连通性和路径质量并列的首要稀疏化标准。在现实的 3D 环境中的实验表明，Inspection-SPARS 将顶点和边的数量减少了 4-8 倍，同时保持覆盖，使得 GIP 求解器能够计算出比密集路线图或最先进的检查路线图短多达 25% 的巡检路径。更广泛地说，Inspection-SPARS 表明稀疏化可以在不牺牲解决方案质量保证的情况下变得面向任务。
cs.RO / 182 / 2609.35570
EdgeVLN: Runtime-Aware Deployment Ready Quantized Vision Language Navigation Model
EdgeVLN：运行时感知的量化视觉语言导航模型的部署准备
Abstract
Vision-language navigation (VLN) models perform well but target compute-rich platforms, limiting deployment on memory- and power-constrained robotic edge devices. Compression alone does not establish whether a VLN model fits the memory, latency, and energy budgets of an edge platform while preserving navigation behavior. We introduce EdgeVLN, a runtime-aware, deployment-ready quantized VLN model that closes this gap. EdgeVLN combines a quantized StreamVLN model with Latent Trajectory Termination Extractor (LATTE), a lightweight causal transformer that improves real-time stopping by predicting a Stop Action verifier rank. Both execute through our llama.cpp VLN driver, which reconstructs streaming context and prunes memory tokens on-board. We characterize a pretrained StreamVLN backbone across weight quantization from 8 to 2 bits and multiple inference runtimes to identify a feasible operating point. LATTE reuses backbone hidden states within the budget freed by quantization, requiring neither a second vision encoder nor an additional backbone forward pass. We evaluate six backbone precisions and seven candidate stop heads on BF16 and IQ4 NL across all 1,839 R2R VLN-CE val-unseen episodes. We measure success rate (SR) in simulation and latency, energy, and resident memory on an NVIDIA Jetson Orin NX 16 GB. LATTE achieves our highest SR, 58.02 percent on the deployed 4-bit model, exceeding the BF16 baseline with only 0.013 s additional latency per navigation step. Four-bit formats achieve nearly identical SR, but step energy varies 36.8 times by execution path. Only IQ4 NL under our VLN driver fits the board, using 11.35 GB resident memory while running 20.8 times faster and using 13.3 times less energy than storage-streamed BF16. INT2 collapses. Runtime selection, memory-token pruning, and quantization are essential for efficient edge deployment.
Chinese Translation
视觉语言导航（VLN）模型表现良好，但主要针对计算资源丰富的平台，这限制了其在内存和功耗受限的机器人边缘设备上的部署。仅仅进行压缩并不能确定一个 VLN 模型是否适合边缘平台的内存、延迟和能量预算，同时保持导航行为。我们提出了 EdgeVLN，这是一种运行时感知的、准备部署的量化 VLN 模型，旨在填补这一空白。EdgeVLN 将量化的 StreamVLN 模型与潜在轨迹终止提取器（Latent Trajectory Termination Extractor, LATTE）相结合，LATTE 是一种轻量级的因果变换器，通过预测停止动作验证器的排名来改善实时停止。两者通过我们的 llama.cpp VLN 驱动程序执行，该驱动程序重建流媒体上下文并在板载上修剪内存标记。我们对预训练的 StreamVLN 主干进行了特征描述，涵盖了从 8 位到 2 位的权重量化和多种推理运行时，以识别可行的操作点。LATTE 在量化释放的预算内重用主干隐藏状态，无需第二个视觉编码器或额外的主干前向传递。我们在 BF16 和 IQ4 NL 上评估了六种主干精度和七种候选停止头，涵盖了所有 1,839 个 R2R VLN-CE 验证集未见的剧集。我们测量了模拟中的成功率（SR），以及在 NVIDIA Jetson Orin NX 16 GB 上的延迟、能量和驻留内存。LATTE 在部署的 4 位模型上实现了最高的 SR，达到 58.02%，超越了 BF16 基线，仅增加了每个导航步骤 0.013 秒的延迟。四位格式的 SR 几乎相同，但步骤能量因执行路径而变化 36.8 倍。只有 IQ4 NL 在我们的 VLN 驱动程序下适合板载，使用 11.35 GB 的驻留内存，同时运行速度比存储流式 BF16 快 20.8 倍，能量消耗减少 13.3 倍。INT2 崩溃。运行时选择、内存标记修剪和量化对于高效的边缘部署至关重要。
cs.RO / 183 / 2609.35575
F4R: Failure-Driven Recognition, Reconstruction, Refinement, and Redeployment for Continual Robot Self-Improvement
F4R：基于失败的识别、重建、优化与再部署以实现持续的机器人自我改进
Abstract
The real-world performance of current vision-language-action models is fundamentally constrained by the limited coverage of expert demonstrations and their insufficient understanding of physical interactions. A common remedy is to collect additional real-world demonstrations of newly encountered failures. However, this process is costly, inefficient, potentially unsafe, and difficult to scale. To address this challenge, we propose Failure for Rising (F4R), a failure-driven real-to-sim-to-real closed-loop learning framework that converts real-world failures into targeted policy improvement. F4R first uses an agent to automatically identify and diagnose failures from rollouts. It reconstructs each failure as an interactive, object-centric table-top environment that preserves the task-relevant spatial and physical conditions. The policy is then refined through failure-conditioned sim-real co-training followed by targeted reinforcement learning in the reconstructed environments. The improved policy is subsequently redeployed, while newly observed failures are continuously fed back into the next reconstruction and learning cycle. Real-world evaluations on four manipulation tasks show that F4R achieves 93.75% In-Distribution and 90.0% Out-of-Distribution (OOD) success, outperforming the budget-matched Targeted BC baseline by 18.75 percentage points under OOD conditions without collecting additional real-world corrective demonstrations.
Chinese Translation
当前视觉-语言-动作模型在现实世界中的表现受到专家示范覆盖范围有限和对物理交互理解不足的根本限制。一个常见的解决方案是收集新遇到的失败的额外现实世界示范。然而，这一过程成本高、效率低、潜在不安全且难以扩展。为了解决这一挑战，我们提出了基于失败的提升框架（F4R），这是一个基于失败驱动的真实-模拟-真实闭环学习框架，将现实世界的失败转化为针对性的策略改进。F4R首先利用代理自动识别和诊断从回放中产生的失败。它将每个失败重建为一个交互式的、以对象为中心的桌面环境，保留与任务相关的空间和物理条件。然后，通过基于失败的模拟-真实共同训练和在重建环境中的针对性强化学习来优化策略。改进后的策略随后被再部署，同时新观察到的失败不断反馈到下一个重建和学习周期。对四个操作任务的现实世界评估表明，F4R在分布内成功率达到93.75%，在分布外（OOD）成功率达到90.0%，在OOD条件下超越了预算匹配的目标BC基线18.75个百分点，而无需收集额外的现实世界纠正示范。
cs.RO / 184 / 2609.35619
CollisionSplatting: Collision-Aware Motion Planning in 3DGS Scenes with Image-Conditioned Objectives and Adjustable Conservatism
CollisionSplatting：在具有图像条件目标和可调保守性的3DGS场景中进行碰撞感知的运动规划
Abstract
Incorporating dense visual information into motion planning remains challenging, as geometric planners rely on abstracted scene representations that discard visual richness, while learned visual models often lack geometric interpretability and computational efficiency. This paper introduces CollisionSplatting, a simple, modular, GPU-accelerated, probability-inspired distance metric with tunable conservatism that operates directly on standard 3D Gaussian Splatting (3DGS) scenes. When combined with learned image-conditioned reward functions, this metric enables joint geometric and visual planning by unifying collision-aware costs with image-space objectives. We integrate the metric into GPU-accelerated Model Predictive Path Integral (MPPI) and Rapidly-Exploring Random Tree (RRT) planners, and show on-par or better collision-classification performance compared to representative baselines while achieving substantially higher collision-checking throughput and significantly lower VRAM usage. Finally, we demonstrate the effectiveness of our metric in real-world vision-guided navigation and manipulation tasks, highlighting 3DGS as a practical bridge between rich perception and real-time motion planning.
Chinese Translation
将密集视觉信息纳入运动规划仍然具有挑战性，因为几何规划器依赖于抽象的场景表示，这些表示丢弃了视觉丰富性，而学习的视觉模型往往缺乏几何可解释性和计算效率。本文介绍了CollisionSplatting，这是一种简单、模块化、基于概率的距离度量，具有可调的保守性，能够直接在标准的3D高斯点云（3DGS）场景中操作。当与学习的图像条件奖励函数结合时，该度量通过将碰撞感知成本与图像空间目标统一起来，从而实现了几何与视觉的联合规划。我们将该度量集成到GPU加速的模型预测路径积分（MPPI）和快速探索随机树（RRT）规划器中，并展示了与代表性基线相比，碰撞分类性能持平或更优，同时实现了显著更高的碰撞检测吞吐量和显著更低的显存使用。最后，我们展示了该度量在现实世界视觉引导的导航和操作任务中的有效性，强调了3DGS作为丰富感知与实时运动规划之间的实用桥梁。
cs.RO / 185 / 2609.35651
Denoising Multi-Robot Trajectories
去噪多机器人轨迹
Abstract
Multi-robot trajectory planning is a fundamental problem in multi-robot coordination but remains computationally challenging due to its nonconvex, multimodal, and high-dimensional nature. This work builds upon D4orm, a dynamics-aware diffusion-denoising framework, and develops a family of planning architectures for diverse operational requirements. Unlike conventional numerical optimization methods, D4orm employs sampling-based optimization to generate solution trajectories through massively parallel sampling, leveraging modern computing architectures such as GPUs. Its diffusion-denoising structure iteratively optimizes \textit{deformations} to candidate control trajectories, providing an efficient and versatile paradigm for generating kinodynamically feasible and conflict-free trajectories. Using D4orm as the building block for advanced planners, we present a decoupled planner for improved scalability, an online receding-horizon planner with feedback control, and a distributed planner for resource-constrained settings. Evaluations with differential-drive and holonomic robots in 2D and 3D environments demonstrate that D4orm-based approaches find high-quality solutions faster and more reliably than other sampling-based optimization methods, such as MPPI, as well as a learned diffusion-model-based method. We further demonstrate zero-shot deployment on ten real quadrotors with obstacles, large-scale deconfliction with 100 simulated robots, and fully onboard distributed `lifelong' operation with six ground robots. Overall, these results establish diffusion denoising as a scalable and reliable framework for multi-robot coordination. Code and video: https://github.com/proroklab/d4orm
Chinese Translation
多机器人轨迹规划是多机器人协调中的一个基本问题，但由于其非凸性、多模态性和高维特性，仍然在计算上具有挑战性。本研究基于 D4orm（动态感知扩散去噪框架），开发了一系列适用于不同操作需求的规划架构。与传统的数值优化方法不同，D4orm 采用基于采样的优化，通过大规模并行采样生成解决方案轨迹，利用现代计算架构（如 GPU）。其扩散去噪结构通过迭代优化候选控制轨迹的 extit{变形}，为生成运动动力学可行且无冲突的轨迹提供了一种高效且多功能的范式。以 D4orm 作为高级规划器的构建块，我们提出了一种解耦规划器以提高可扩展性、一种带反馈控制的在线递归视野规划器，以及一种适用于资源受限环境的分布式规划器。在 2D 和 3D 环境中对差分驱动和全向机器人进行的评估表明，基于 D4orm 的方法比其他基于采样的优化方法（如 MPPI）以及基于学习的扩散模型方法更快、更可靠地找到高质量解决方案。我们进一步展示了在十架真实四旋翼上零样本部署、对 100 个模拟机器人进行大规模去冲突，以及与六个地面机器人进行完全自主的分布式“终身”操作。总体而言，这些结果确立了扩散去噪作为多机器人协调的可扩展且可靠的框架。代码和视频： https://github.com/proroklab/d4orm
cs.RO / 186 / 2609.35652
MM-ABC: Towards Generalist Mobile Manipulation via Seeing, Coordinating and Imagining
MM-ABC：通过视觉、协调和想象实现通用移动操控
Abstract
Mobile manipulation extends robot interaction beyond a fixed kinematic workspace by making the reachable region itself controllable. This flexibility introduces two central challenges: spatially grounded perception under continuous ego-motion and coordinated control of heterogeneous arm and base actions. Existing approaches strengthen geometry through explicit 3D representations or predictive world models, and often decouple mobility and manipulation into separate action streams. We argue that effective mobile manipulation requires not only decoupling, but also representations that support efficient cross-stream collaboration. We present MM-ABC, a foundation model built around Seeing, Coordinating, and Imagining Arm-Base Collaboration. MM-ABC combines sparse multi-level VLM features for spatial perception; a training-only future branch that uses world imagination and geometric intent as extra supervision, strengthening perception and manipulation-intent prediction and improving the overall learning signal; and MM-APT, which coordinates separate manipulation and mobility streams through masked joint attention and clean-action x-prediction. In controlled ablations, replacing clean-action prediction with velocity prediction lowers success on RoboCasa365 composite-seen tasks from 32.8% to 29.2%, and removing future supervision or multilevel conditioning causes larger drops. We pretrain MM-ABC on 5,000+ hours of heterogeneous robot data spanning 400K+ episodes, 12 datasets, and 17 embodiments. Experiments cover EBench, RoboCasa365, ManiSkill-HAB, LIBERO, LIBERO-Plus, and real-world mobile manipulation. MM-ABC achieves 44.71% success on EBench, 61.2% on RoboCasa365, 99.1% on LIBERO, 82.8% on LIBERO-Plus without perturbation training, and 83% mean success on five real-world tasks.
Chinese Translation
移动操控通过使可达区域本身可控，扩展了机器人与环境的交互，超越了固定的运动学工作空间。这种灵活性带来了两个核心挑战：在持续自我运动下的空间感知和异构臂部与底座动作的协调控制。现有的方法通过显式的三维表示或预测世界模型来增强几何特征，通常将移动性和操控分解为独立的动作流。我们认为，有效的移动操控不仅需要分解，还需要支持高效跨流协作的表示。我们提出了MM-ABC，一个围绕视觉、协调和想象臂部-底座协作构建的基础模型。MM-ABC结合了稀疏的多层次视觉语言模型（VLM）特征用于空间感知；一个仅用于训练的未来分支，利用世界想象和几何意图作为额外的监督，增强感知和操控意图预测，并改善整体学习信号；以及MM-APT，通过掩蔽的联合注意力和干净动作的交叉预测来协调独立的操控和移动流。在控制性消融实验中，将干净动作预测替换为速度预测使RoboCasa365复合场景任务的成功率从32.8%降低到29.2%，而去除未来监督或多层次条件会导致更大的下降。我们在超过5000小时的异构机器人数据上预训练MM-ABC，涵盖了超过40万次的实验，12个数据集和17种实现。实验涵盖了EBench、RoboCasa365、ManiSkill-HAB、LIBERO、LIBERO-Plus以及真实世界的移动操控。MM-ABC在EBench上取得了44.71%的成功率，在RoboCasa365上为61.2%，在LIBERO上为99.1%，在LIBERO-Plus上为82.8%（未经过扰动训练），在五个真实世界任务上平均成功率为83%。
cs.RO / 187 / 2609.35690
Agent Priors-guided Policy Learning
基于代理先验的策略学习
Abstract
Robots that learn from a few demonstrations often require two forms of generalization. Compositional generalization recombines skills to solve new tasks, and skill generalization lets the learned policy behind each skill work in new situations. The two depend on each other, yet information is lost between composition and the skills it calls. Where a skill works is determined by the structure its policy is trained with, while composition sees the skill only through a separate description, such as a name, an instruction, or a symbolic operator, that omits this structure. Our key idea is to use each policy's structural prior as part of the interface between composition and the skill. A structural prior states what a behavior depends on, for example that a grasp depends only on the gripper's pose relative to the object. Built into training, it shapes where the policy generalizes; stated in language, it tells composition where the policy applies. We instantiate this idea in Agent Priors-guided Policy Learning (APPL). A construction agent segments complete demonstrations into reusable skills, proposes several structural priors for each skill, and trains and verifies one policy per prior. A runtime agent then selects among these prior-specific policies and composes them toward new task goals using their interfaces. Across MetaWorld and long-horizon ManiSkill tasks, APPL improves out-of-distribution skill generalization and enables previously unseen skill compositions; ablating the interface information substantially reduces performance. These results support the use of training-time structural assumptions as a bridge between skill learning and skill composition.
Chinese Translation
从少量示例中学习的机器人通常需要两种形式的泛化。组合泛化重新组合技能以解决新任务，而技能泛化则使得每个技能背后的学习策略能够在新情境中发挥作用。这两者相互依赖，但在组合与其调用的技能之间会丢失信息。技能的有效性取决于其策略训练所用的结构，而组合仅通过单独的描述（如名称、指令或符号运算符）来理解技能，这种描述忽略了该结构。我们的关键思想是将每个策略的结构先验作为组合与技能之间接口的一部分。结构先验声明了行为所依赖的内容，例如抓取仅依赖于夹持器相对于物体的姿态。将其内置于训练中，它塑造了策略的泛化范围；用语言表述时，它告诉组合策略的适用范围。我们在基于代理先验的策略学习（Agent Priors-guided Policy Learning, APPL）中实现了这一思想。构建代理将完整的示例分割为可重用的技能，为每个技能提出多个结构先验，并为每个先验训练和验证一个策略。运行时代理随后在这些特定于先验的策略中进行选择，并利用它们的接口组合以实现新的任务目标。在MetaWorld和长时间跨度的ManiSkill任务中，APPL提高了分布外技能泛化能力，并实现了之前未见的技能组合；去除接口信息显著降低了性能。这些结果支持在训练期间使用结构假设作为技能学习与技能组合之间的桥梁。
cs.RO / 188 / 2609.35700
LQR-ArUco Fusion: Robust Hierarchical Control for Navigation and Asymmetric Manipulation in Two-Wheeled Robots
LQR-ArUco 融合：用于双轮机器人导航和不对称操作的鲁棒层次控制
Abstract
We propose a hierarchical control framework to address severe dynamic instabilities and navigational drift that arise when a two-wheeled inverted pendulum (TWIP) robot attempts asymmetric object manipulation. While two-wheeled platforms are highly manoeuvrable, their constant balancing adjustments make onboard odometry highly unreliable for precise navigation. Furthermore, the addition of a side-mounted robotic arm introduces unactuated lateral roll moments when a payload is lifted, a challenge heavily compounded on uneven terrain. To solve these coupled problems, our architecture divides the workload. An offboard vision system tracks overhead ArUco markers to provide high-latency global waypoint navigation, bypassing odometry drift. Simultaneously, a low-latency onboard control loop rejects active physical disturbances using inertial and encoder data. In our physical experiments, this dual-loop approach enabled the custom-built robot to navigate accurately, reject transient impacts from speed bumps, adapt to a dynamic seesaw ramp, and carry a payload securely without falling over its narrow wheelbase.
Chinese Translation
我们提出了一种层次控制框架，以应对双轮倒立摆（TWIP）机器人在进行不对称物体操作时出现的严重动态不稳定性和导航漂移。尽管双轮平台具有高度的机动性，但其持续的平衡调整使得车载里程计在精确导航时极不可靠。此外，侧装机械臂的加入在提升负载时会引入未激活的横向滚转力矩，这在不平坦地形上尤为复杂。为了解决这些耦合问题，我们的架构将工作负载进行了划分。一个离线视觉系统跟踪头顶的 ArUco 标记，以提供高延迟的全局航点导航，从而绕过里程计漂移。同时，一个低延迟的车载控制回路利用惯性和编码器数据拒绝主动的物理干扰。在我们的物理实验中，这种双回路方法使定制机器人能够准确导航，抵御来自减速带的瞬态冲击，适应动态的跷跷板坡道，并在不翻倒其狭窄轮距的情况下安全地携带负载。
cs.RO / 189 / 2609.35709
Humanoid Loco-Manipulation With Discrete VLA Model
基于离散VLA模型的人形机器人运动操控
Abstract
Vision-language-action (VLA) models using discrete action tokens have proven effective for controling robotic arms on manipulation tasks. For a humanoid, however, the whole-body action space -- legs, torso, arms, and hands -- is far higher-dimensional and heterogeneous, raising tokenization, training, and real-time inference challenges that the previous VLA models do not address. We present Holo-M, to our knowledge the first discrete VLA model for humanoid loco-manipulation that intrinsically exploits the language model by extending its vocabulary with action tokens. In this model, we devise a unified action tokenizer that decomposes the humanoid action space into four body-part-specific tokenizers -- end-effector, body, hand, and kinematics -- enabling training across drastically different embodiments and data sources, including humanoid teleoperation, ego-centric human video, and simulation. By extending the language model's vocabulary with these action tokens, we avoid the knowledge-insulation problem inherent to the models that use separate continuous action experts. To meet real-time control requirements, we decode each body part's action tokens through grouped discrete diffusion decoding, rather than using autoregression on the action tokens. We have conducted extensive experiments on the SIMPLE humanoid loco-manipulation benchmark, in which Holo-M achieves the highest success rates in both the generalist and specialist evaluations, leading the second best by significant margins. We will release all the code and model weights.
Chinese Translation
使用离散动作标记的视觉-语言-动作（VLA）模型在控制机器人手臂进行操控任务方面已被证明是有效的。然而，对于人形机器人而言，整个身体的动作空间——包括腿部、躯干、手臂和手——的维度更高且异质性更强，这带来了标记化、训练和实时推理的挑战，而之前的VLA模型并未解决这些问题。我们提出了Holo-M，至今为止我们所知的第一个用于人形机器人运动操控的离散VLA模型，该模型通过扩展其词汇表以包含动作标记，内在地利用了语言模型。在该模型中，我们设计了一个统一的动作标记器，将人形机器人的动作空间分解为四个特定于身体部位的标记器——末端执行器、身体、手和运动学——使得在截然不同的体现和数据源（包括人形机器人遥操作、自我中心的人类视频和仿真）之间进行训练成为可能。通过用这些动作标记扩展语言模型的词汇表，我们避免了使用单独的连续动作专家所固有的知识隔离问题。为了满足实时控制的要求，我们通过分组离散扩散解码来解码每个身体部位的动作标记，而不是对动作标记使用自回归方法。我们在SIMPLE人形机器人运动操控基准上进行了广泛的实验，其中Holo-M在通用和专业评估中均取得了最高的成功率，显著领先于第二名。我们将发布所有代码和模型权重。
cs.RO / 190 / 2609.35717
RoboCompiler: Graph-Native Compilation of Closed-Chain Robots for Consistent Modeling, Control, and Simulation
RoboCompiler：闭环机器人图形原生编译的一致建模、控制与仿真
Abstract
Robots with kinematic loops, coupled actuators, and changing contacts require consistent models of configuration, motion, force, and dynamics. Yet these interfaces are often reconstructed separately for control and simulation, making closure and actuation consistency difficult to maintain. This paper presents RoboCompiler, a graph-native framework that compiles a canonical mechanism graph into a shared mechanical interface. From bodies, joints, frames, inertias, and actuator ports, it constructs closure paths and analytic residual Jacobians, then assembles feasible configurations through rank-checked continuation and correction. A tangent lift maps independent velocities to full robot and task motion, while paired actuator-port maps preserve virtual work. A constraint-curvature correction extends the reduction to accelerations and projected rigid-body dynamics, including floating-base and support modes. Cycle-local evaluation, generated Jacobians, and dependency-aware reuse enable localized updates when closure inputs change. We evaluate physical loops and task-induced constraints on a Komatsu excavator, Unitree Go2, Franka Panda, Kangaroo, and a six-UPS Stewart platform. High-precision constrained-dynamics and independent Pinocchio checks confirm mechanical consistency; MuJoCo and Isaac Sim/PhysX executions demonstrate task performance and model reuse under native contact. For Kangaroo, compilation reduces residual-and-Jacobian evaluation time by 96.7% and closed-loop rollout wall time by 66.8%, with dynamics and control held fixed.
Chinese Translation
具有运动环、耦合执行器和变化接触的机器人需要一致的配置、运动、力和动力学模型。然而，这些接口通常在控制和仿真中分别重构，使得闭合和执行一致性难以维持。本文提出了RoboCompiler，一个图形原生框架，将规范机制图编译成共享机械接口。它从物体、关节、框架、惯性和执行器端口构建闭合路径和解析残余雅可比矩阵，然后通过秩检查的延续和修正组装可行配置。切线提升将独立速度映射到完整的机器人和任务运动，而配对执行器端口映射则保持虚拟功。约束曲率修正将简化扩展到加速度和投影刚体动力学，包括浮动基座和支撑模式。循环局部评估、生成的雅可比矩阵和依赖感知重用使得在闭合输入变化时能够进行局部更新。我们在小松挖掘机、Unitree Go2、Franka Panda、Kangaroo和六个UPS斯图尔特平台上评估物理环和任务诱导约束。高精度约束动力学和独立的Pinocchio检查确认了机械一致性；MuJoCo和Isaac Sim/PhysX的执行展示了在原生接触下的任务性能和模型重用。对于Kangaroo，编译将残余和雅可比评估时间减少了96.7%，闭环展开的墙面时间减少了66.8%，同时保持了动力学和控制的固定。
cs.RO / 191 / 2609.35761
DexRoam: Learning Mobile Bimanual Dexterous Manipulation from Egocentric Whole-Body Human Demonstrations
DexRoam：从自我中心的全身人类示范中学习移动双手灵巧操作
Abstract
Mobile bimanual dexterous manipulation requires continuous coordination of locomotion, whole-body motion, and finger-level dexterity within a single trajectory, creating a severe robot demonstration bottleneck. Egocentric human demonstrations offer a scalable alternative, but prior approaches ease the transfer by simplifying human motion, discarding exactly the fine-grained, coupled structure such tasks depend on. We present DexRoam, a complete system for learning mobile bimanual dexterous manipulation from human demonstrations, in which whole-body motion remains continuous and coupled throughout the human-to-robot transfer process. To enable scalable collection of whole-body human manipulation demonstrations, we develop a tracker-free capture system using only a consumer VR headset and a head-mounted stereo camera, without external cameras or motion trackers. We then perform three explicit alignment stages---embodiment, action-semantic, and temporal---to map captured motion into the robot action space, preserving fine-grained whole-body motion and allowing human and robot demonstrations to be jointly learned by standard VLA policies. Real-world experiments with different VLA backbones show that human demonstrations consistently improve policy learning across training paradigms, raising average success from 29% to 56% on GR00T N1.7 and from 32% to 57% on pi0.5, while matching robot-only training with half the robot demonstrations. Ablations confirm that each alignment stage is necessary. These results highlight the potential of human demonstrations for scalable whole-body mobile manipulation with preserved fine-grained motion structure.
Chinese Translation
移动双手灵巧操作需要在单一路径中持续协调移动、全身运动和手指级别的灵巧性，这造成了机器人示范的严重瓶颈。自我中心的人类示范提供了一种可扩展的替代方案，但之前的方法通过简化人类运动来缓解转移，恰恰忽略了这些任务所依赖的细粒度耦合结构。我们提出了DexRoam，一个完整的系统，用于从人类示范中学习移动双手灵巧操作，在此过程中，全身运动在从人类到机器人的转移过程中保持连续和耦合。为了实现可扩展的人类全身操作示范的收集，我们开发了一种无追踪器的捕捉系统，仅使用消费级虚拟现实（VR）头戴设备和头戴式立体相机，无需外部相机或运动追踪器。然后，我们执行三个明确的对齐阶段——体现、动作语义和时间——将捕获的运动映射到机器人动作空间，保留细粒度的全身运动，并允许人类和机器人示范通过标准的变分学习算法（VLA）策略共同学习。与不同的VLA骨干网络进行的现实世界实验表明，人类示范在训练范式中始终提高了策略学习，将GR00T N1.7上的平均成功率从29%提升至56%，将pi0.5上的成功率从32%提升至57%，同时与仅使用一半机器人示范的机器人单独训练相匹配。消融实验确认每个对齐阶段都是必要的。这些结果突显了人类示范在保留细粒度运动结构的可扩展全身移动操作中的潜力。
cs.CV / 1 / 2609.31651
PalmLeaf-VQA: A Multi-Script Visual Question Answering Benchmark for Historical Palm-Leaf Manuscript Understanding Across Diverse Regions
PalmLeaf-VQA：一个针对历史棕榈叶手稿理解的多脚本视觉问答基准，涵盖多样化地区
Abstract
Historical manuscripts remain largely absent from modern vision-language benchmarks, leaving open how well multimodal large language models (MLLMs) handle culturally diverse, degraded, and non-Latin document images. We introduce \textbf{PalmLeaf-VQA}, a multi-script visual question answering benchmark for historical palm-leaf manuscript understanding across South and Southeast Asian traditions. PalmLeaf-VQA contains \textbf{923 curated manuscript images} and \textbf{7,384 question--answer pairs} from eight collection groups: Balinese, Grantha, Jathakam, Kambaramayanam, Kannada, Khmer, Sundanese, and Tamil. Unlike recognition-oriented resources, the benchmark targets manuscript-aware visual reasoning over preservation-relevant cues, including physical condition, line structure, material and coating, binding holes, margins, symbols, drawings, and localized visual artifacts. We evaluate recent proprietary and open-weight MLLMs under open-answer and constrained-answer prompting and provide fine-grained analysis across collections, question categories, and task types. The strongest evaluated model reaches only \textbf{58.00\% exact-match accuracy} on the held-out test split, revealing substantial limitations in current MLLMs for rare-script, degraded-layout, and preservation-oriented document understanding. PalmLeaf-VQA provides a standardized benchmark for advancing culturally grounded and layout-aware multimodal document analysis.
Chinese Translation
历史手稿在现代视觉-语言基准中仍然缺乏，尚不清楚多模态大型语言模型（MLLMs）在处理文化多样性、退化及非拉丁文档图像方面的表现。我们提出了 extbf{PalmLeaf-VQA}，这是一个针对南亚和东南亚传统的历史棕榈叶手稿理解的多脚本视觉问答基准。PalmLeaf-VQA包含 extbf{923幅精心挑选的手稿图像}和 extbf{7,384对问答}，来自八个收藏组：巴厘岛文、Grantha文、Jathakam文、Kambaramayanam文、卡纳达文、高棉文、巽他文和泰米尔文。与以识别为导向的资源不同，该基准旨在针对手稿相关的视觉推理，关注与保存相关的线索，包括物理状态、行结构、材料和涂层、装订孔、边距、符号、图画以及局部视觉伪影。我们在开放回答和限制回答提示下评估了最近的专有和开放权重的MLLM，并提供了跨收藏、问题类别和任务类型的细致分析。评估的最强模型在保留的测试集上仅达到 extbf{58.00\%的精确匹配准确率}，揭示了当前MLLM在稀有脚本、退化布局和以保存为导向的文档理解方面的重大局限性。PalmLeaf-VQA为推动文化根植和布局意识的多模态文档分析提供了一个标准化的基准。
cs.CV / 2 / 2609.31654
Temporal-Attention Head Specialization During Video Diffusion Training
视频扩散训练中的时间注意力头专业化
Abstract
Video diffusion transformers depend on temporal attention to coordinate information across frames, yet nearly everything known about this mechanism comes from analyzing trained models, so when and where temporal-attention structure forms during training remains poorly characterized. Population averages can also hide it, since a few specializing heads and a diffusing majority cancel in the mean. We therefore conduct a checkpoint-resolved census of every temporal-attention head across nine Open-Sora STDiT training runs spanning three model scales (306M to 1.03B parameters), scoring each head with an entropy-normalized measure of cross-frame attention concentration (CFAC) under a preregistered change-point and effect-size selection rule. The census reveals the sparse picture that averages obscure. Aggregate CFAC is flat or decreasing in every run, while a small minority of heads, roughly 4--13% in full-grid runs, develops pronounced concentration. Across seeds, the reproducible signal is positional but block-level. Selected heads repeatedly arise in the first temporal block, whereas individual head coordinates do not reproduce once block membership is accounted for. Among the analyzed 760M selected heads, attention maps converge to a small repertoire of local frame-routing motifs, self-frame diagonals and adjacent-frame bands, even when the responsible coordinates differ across runs. Correlation and ablation analyses do not establish a causal link to generated video quality, and we bound our claims accordingly. Beyond this STDiT family, the study contributes a transferable methodology. Checkpoint-resolved, per-head analysis under fixed selection rules can expose sparse temporal organization in other factorized video diffusion transformers and, with adapted routing metrics, in joint spatio-temporal architectures.
Chinese Translation
视频扩散变换器依赖于时间注意力来协调帧间信息，然而关于这一机制的已知信息几乎全部来自对训练模型的分析，因此在训练过程中时间注意力结构何时何地形成仍然缺乏明确的描述。总体平均值可能掩盖这一点，因为少数专业化的头部与大多数扩散的头部在平均值上相互抵消。因此，我们对九个 Open-Sora STDiT 训练运行中的每个时间注意力头进行了检查点分辨的普查，涵盖三个模型规模（306M 到 1.03B 参数），并根据预注册的变更点和效应大小选择规则，使用熵归一化的跨帧注意力集中度（CFAC）对每个头部进行评分。普查揭示了平均值所掩盖的稀疏图景。在每次运行中，聚合 CFAC 要么持平要么下降，而少数头部（在全网格运行中约占 4% 至 13%）则表现出明显的集中性。在不同种子下，可重复的信号是位置性的，但在块级别上。选定的头部在第一个时间块中反复出现，而一旦考虑到块成员资格，单个头部坐标则不再重现。在分析的 760M 个选定头部中，注意力图趋向于一小组局部帧路由模式、自帧对角线和相邻帧带，即使负责的坐标在不同运行中有所不同。相关性和消融分析未能建立与生成视频质量之间的因果关系，因此我们相应地限制了我们的主张。超越这一 STDiT 家族，本研究贡献了一种可转移的方法论。在固定选择规则下的检查点分辨、逐头分析可以揭示其他分解的视频扩散变换器中的稀疏时间组织，并且通过调整路由度量，可以在联合时空架构中实现。
cs.CV / 3 / 2609.31655
One Evaluation, Any Operating Point: Hypernetwork-Amortized MeanFlow for 3D MRI Reconstruction
一次评估，任意操作点：超网络摊销的MeanFlow用于3D MRI重建
Abstract
Generative priors reconstruct accelerated 3D MRI well but pay heavily at deployment: tens of network evaluations per volume, and protocol-specific hyperparameter tuning. A third hidden cost is the scanner's fixed sampling pattern. We treat the whole operating point as an input. A 3D MeanFlow patch network (a one-step flow model) is fine-tuned end-to-end through a warm-started, five-iteration differentiable conjugate-gradient projection. A small hypernetwork maps the operating point (data-consistency weight, acceleration, and the Cartesian sampling pattern itself) to the network's per-channel modulation. Three findings follow. (i) Learning the acquisition is worth more than any other operating point: on clinical knee data, the learned mask gains up to +2.34 dB over the protocol's variable-density mask. This gain requires the solver: with a feed-forward reconstructor the same learned mask hurts at 4x (-1.7 dB), but with the data-consistency projection it adds +4.6 dB. (ii) One evaluation is highly effective: it beats a 20-step patch-diffusion prior by up to +3.1 dB on brain and +2.9 dB on knee. Three to five evaluations extend the front to +6 dB while using a quarter of the prior's network calls. (iii) Fully sampled targets are optional: trained self-supervised on a split of acquired samples, the reconstructor matches its supervised twin at 4x on real data. Finally, we report what failed and why: subject-adaptive acquisition from measured energy, combining self-supervision with learned acquisition, and amortising the data-consistency weight.
Chinese Translation
生成性先验能够很好地重建加速的3D MRI，但在部署时代价沉重：每个体积需要数十次网络评估，并且需要针对特定协议进行超参数调优。第三个隐性成本是扫描仪的固定采样模式。我们将整个操作点视为输入。一个3D MeanFlow补丁网络（一个一步流模型）通过预热启动的五次迭代可微共轭梯度投影进行端到端的微调。一个小型超网络将操作点（数据一致性权重、加速和笛卡尔采样模式本身）映射到网络的每通道调制。以下是三个发现。(i) 学习采集比任何其他操作点更有价值：在临床膝关节数据上，学习的掩膜比协议的可变密度掩膜提高了最高+2.34 dB。这个增益需要求解器：使用前馈重建器时，相同的学习掩膜在4倍下会造成-1.7 dB的损失，但使用数据一致性投影则增加了+4.6 dB。(ii) 一次评估非常有效：在脑部和膝关节上，它比20步补丁扩散先验高出最多+3.1 dB和+2.9 dB。三到五次评估将增益扩展至+6 dB，同时使用了先验网络调用的四分之一。(iii) 完全采样的目标是可选的：在获取样本的分割上进行自我监督训练，重建器在真实数据上与其监督双胞胎在4倍下相匹配。最后，我们报告了失败的原因及其原因：从测量能量中进行主题自适应采集，将自我监督与学习的采集相结合，以及摊销数据一致性权重。
cs.CV / 4 / 2609.31657
Enhancing Foundation Models for Imbalanced SAR Ship Classification via Targeted Oversampling
通过针对性过采样增强不平衡SAR船舶分类的基础模型
Abstract
Remote-sensing foundation models offer strong representations for SAR imagery, but their behavior under severe long-tail class imbalance is still not well characterized. We benchmark DOFA and SAR-JEPA on the imbalanced OpenSARShip dataset and compare them with ImageNet-pretrained baselines under a fixed, training-efficient protocol that keeps the backbone frozen. To mitigate imbalance without fine-tuning, we apply four oversampling methods in embedding space exclusively to minority classes and train a lightweight classifier head on the augmented embeddings. Across both foundation models, oversampling improves Macro-F1 and test accuracy relative to their respective baselines, with the largest Macro-F1 gains observed for DOFA using ADASYN (34.39 to 38.56) and for SAR-JEPA using SVM-SMOTE (25.89 to 32.30). We also report class-wise behavior, showing that aggregate improvements can coexist with persistent failures on specific rare classes. Code for embedding extraction and reproducible multi-seed evaluation is provided to support rapid experimentation on free-tier hardware.
Chinese Translation
遥感基础模型为SAR图像提供了强大的表征能力，但在严重的长尾类别不平衡下，它们的表现仍未得到充分表征。我们在不平衡的OpenSARShip数据集上对DOFA和SAR-JEPA进行了基准测试，并在固定的、训练高效的协议下将它们与ImageNet预训练的基线进行了比较，该协议保持了主干网络的冻结。为了在不进行微调的情况下缓解不平衡，我们在嵌入空间中专门对少数类应用了四种过采样方法，并在增强的嵌入上训练了一个轻量级分类器头。在这两种基础模型中，过采样相对于各自的基线提高了Macro-F1和测试准确率，其中DOFA使用ADASYN的Macro-F1增益最大（从34.39提高到38.56），而SAR-JEPA使用SVM-SMOTE的增益为（从25.89提高到32.30）。我们还报告了类别级别的表现，显示出整体改进可以与特定稀有类别的持续失败共存。提供了嵌入提取和可重复多种子评估的代码，以支持在免费硬件上的快速实验。
cs.CV / 5 / 2609.31658
Cross-Dataset Transfer and Unknown-Class Detection in Imbalanced SAR Ship Classification
不平衡合成孔径雷达船舶分类中的跨数据集迁移与未知类别检测
Abstract
Ship classification from Synthetic Aperture Radar (SAR) imagery is a critical computer vision task, yet the robustness of models under deployment shifts remains unclear. While models are often trained on one dataset and deployed on another, we lack a comprehensive understanding of their cross-dataset generalization. To address this, we evaluate six pretrained models on two SAR ship datasets in three settings: in-domain classification, cross-dataset transfer, and unknown-class detection. For unknown detection, we hold out all classes one at a time. In-domain, SARDet100K gives the best balanced accuracy on both datasets (73.4\% on FUSARShip and 53.2\% on OpenSARShip). In cross-dataset transfer, we observe strong failures: some models show moderate accuracy but near-chance balanced accuracy (for example, 64.5\% accuracy but 33.3\% balanced accuracy for OpenSARShip to FUSARShip). In unknown detection, performance depends on the held-out class and dataset, while MC-dropout variance is often close to random. These findings show that cross-dataset generalization in SAR remains limited and that task-specific uncertainty scores are often more informative than MC-dropout variance for held-out-class detection, although their relative ranking depends on the dataset and held-out class.
Chinese Translation
从合成孔径雷达（SAR）图像中进行船舶分类是一项关键的计算机视觉任务，但模型在部署转变下的鲁棒性仍不明确。虽然模型通常是在一个数据集上训练并在另一个数据集上部署，但我们对它们的跨数据集泛化缺乏全面的理解。为了解决这个问题，我们在两个SAR船舶数据集上评估了六个预训练模型，设置了三种场景：领域内分类、跨数据集迁移和未知类别检测。在未知检测中，我们逐一保留所有类别。在领域内，SARDet100K在两个数据集上都表现出最佳的平衡准确率（FUSARShip上为73.4%，OpenSARShip上为53.2%）。在跨数据集迁移中，我们观察到强烈的失败：一些模型显示出适度的准确率，但平衡准确率接近随机（例如，从OpenSARShip到FUSARShip的准确率为64.5%，但平衡准确率为33.3%）。在未知检测中，性能依赖于保留的类别和数据集，而MC-dropout方差通常接近随机。这些发现表明，SAR中的跨数据集泛化仍然有限，并且特定任务的不确定性评分通常比MC-dropout方差在保留类别检测中更具信息性，尽管它们的相对排名依赖于数据集和保留类别。
cs.CV / 6 / 2609.31661
ForensicZoom: Adaptive Visual Inspection with Multimodal LLMs for Industrial-Grade Face Forgery Detection
ForensicZoom：基于多模态大语言模型的工业级人脸伪造检测的自适应视觉检查
Abstract
Reliable face forgery detection is critical to the security of online identity verification systems, where missed attacks compromise security and excessive false positives disrupt legitimate users. Specialized forensic detectors achieve strong detection performance but provide limited interpretability, while multimodal large language models (MLLMs) offer strong semantic understanding and interpretable reasoning yet remain substantially weaker for face forgery detection. We argue that a key limitation lies in how visual evidence is acquired: subtle forensic artifacts may be poorly represented at standard resolution, while uniformly processing all cases at higher resolution is computationally inefficient. We therefore introduce ForensicZoom, an industrial-grade MLLM framework for adaptive visual inspection. ForensicZoom first equips a general-purpose MLLM with forensic-aware visual representations and aligns the language model with these features. Its central mechanism, NEED_ZOOM, enables the model to autonomously request magnified views of suspicious regions when the initial evidence is insufficient, turning fixed-pass classification into adaptive multi-round forensic reasoning. The zoom behavior is learned through reward shaping that balances detection accuracy with unnecessary visual inspection, concentrating additional computation on difficult cases. A final attribution optimization stage improves natural-language forensic reports while preserving detection performance. On large-scale industrial identity verification data, ForensicZoom achieves over 97% TPR at 0.1% FPR, substantially outperforming both specialized detectors and existing MLLM-based methods while producing actionable forensic attributions. These results demonstrate that ForensicZoom can provide an effective path toward accurate, interpretable, and scalable MLLM-based face forgery detection.
Chinese Translation
可靠的人脸伪造检测对于在线身份验证系统的安全至关重要，漏检攻击会危及安全，而过多的误报则会干扰合法用户。专门的法医检测器在检测性能上表现出色，但其可解释性有限；而多模态大语言模型（MLLM）则提供了强大的语义理解和可解释推理，但在面部伪造检测方面仍显著较弱。我们认为，关键的限制在于视觉证据的获取方式：微妙的法医伪影在标准分辨率下可能表现不佳，而在更高分辨率下统一处理所有案例则计算效率低下。因此，我们提出了ForensicZoom，一个用于自适应视觉检查的工业级MLLM框架。ForensicZoom首先为通用MLLM配备法医感知的视觉表示，并将语言模型与这些特征对齐。其核心机制NEED_ZOOM使模型能够在初始证据不足时自主请求可疑区域的放大视图，将固定的分类过程转变为自适应的多轮法医推理。缩放行为通过奖励塑造进行学习，以平衡检测准确性与不必要的视觉检查，将额外的计算集中在困难案例上。最后的归因优化阶段在保持检测性能的同时改善自然语言法医报告。在大规模工业身份验证数据上，ForensicZoom在0.1%假阳性率下实现了超过97%的真正率，显著优于专门检测器和现有的基于MLLM的方法，同时生成可操作的法医归因。这些结果表明，ForensicZoom能够为准确、可解释和可扩展的基于MLLM的人脸伪造检测提供有效路径。
cs.CV / 7 / 2609.31662
Learning Steadily: Accumulating Relative Point Margin Scores for Face Image Quality Assessment
稳步学习：面部图像质量评估的相对点边际分数累积
Abstract
Face Image Quality Assessment determines the suitability of captured face images for automated face recognition (FR), a critical capability for reliable biometric systems. Existing state-of-the-art FR-integrated FIQA methods suffer from temporal instability: as the feature space evolves during training, single-epoch quality estimates fluctuate, creating a moving target that undermines reliable quality prediction. We introduce CARPM-FIQA, a stabilization strategy for FR-integrated FIQA that accumulates relative point margin measurements, the ratio between intra-class compactness and inter-class separation, across the entire training trajectory rather than relying on single-epoch estimates. This cumulative averaging approach provides theoretically grounded advantages: reduced variance in quality estimates, improved mean squared error, and enhanced ranking stability with convergence guarantees as training progresses. Through controlled experiments on the SynFIQA dataset with labeled quality groups, we demonstrate that cumulative averaging achieves superior discriminative ability, and ablation studies across different training configurations confirm consistent improvements. Evaluated against twelve FIQA methods on eight challenging benchmarks with four FR models at two FMR thresholds, CARPM-FIQA places 4th (CARPM-FIQA(L)) and 6th (CARPM-FIQA(S)) of 17 compared methods by pAUC-EDC and AUC-EDC averaged across FR models and, after per-benchmark normalization, across benchmarks, staying within a few percent of the best method's normalized average for every FR model, providing a principled solution to training instability while maintaining the performance benefits of FR integration. More broadly, our work demonstrates that temporal aggregation strategies can stabilize training objectives in deep learning systems where target values inherently fluctuate due to evolving feature representations.
Chinese Translation
面部图像质量评估（FIQA）确定捕获的面部图像是否适合自动面部识别（FR），这是可靠生物识别系统的重要能力。现有的最先进的集成FR的FIQA方法存在时间不稳定性：随着特征空间在训练过程中演变，单个周期的质量估计会波动，形成一个移动目标，从而削弱可靠的质量预测。我们提出了CARPM-FIQA，这是一种针对集成FR的FIQA的稳定化策略，它在整个训练过程中累积相对点边际测量，即类内紧凑性与类间分离度的比率，而不是依赖单个周期的估计。这种累积平均的方法提供了理论基础的优势：降低质量估计的方差，改善均方误差，并随着训练的进行增强排名的稳定性和收敛保证。通过在带有标记质量组的SynFIQA数据集上进行的控制实验，我们证明了累积平均能够实现更优的区分能力，而在不同训练配置下的消融研究也确认了一致的改进。在对比十二种FIQA方法和八个具有挑战性的基准测试，以及在两个FMR阈值下的四个FR模型的评估中，CARPM-FIQA在17种对比方法中，基于pAUC-EDC和AUC-EDC的平均结果，分别位列第4（CARPM-FIQA(L)）和第6（CARPM-FIQA(S)），在每个FR模型的标准化平均值上与最佳方法的标准化平均值相差仅几个百分点，为训练不稳定性提供了一个有原则的解决方案，同时保持了FR集成的性能优势。更广泛地说，我们的工作表明，时间聚合策略可以在深度学习系统中稳定训练目标，因为目标值由于特征表示的演变而固有波动。
cs.CV / 8 / 2609.31665
Language-Augmented Video Action Anticipation: Design Fundamentals, Benchmarks, and Open Challenges
语言增强的视频动作预测：设计基础、基准测试与开放挑战
Abstract
Action anticipation predicts future human actions from partial video under incomplete context and temporal uncertainty. Recent systems introduce large language models (LLMs), vision-language models (VLMs), or language-derived semantics at different stages, but reported gains are difficult to interpret when task formulation, visual pretraining, supervision, decoder design, and evaluation code change simultaneously. The central contribution of this review is an evidence-aware design map that crosses task regime with the point at which language-derived information intervenes. We characterise task regimes along six axes. These axes organise the literature into five broad task families: single-action, sequence, object-interaction, cross-view, and planning-oriented settings. C1-C3 locate interventions in context construction, goal/intention modelling, and future decoding, while C4 is treated as an adjacent, emerging grounding/executability extension. Unlike a generic processing pipeline, the map links each intervention to an appropriate counterfactual, failure diagnosis, and permissible evidence claim. Supporting contributions include a protocol-level audit of Ego4D-LTA and EPIC-KITCHENS-100, a multidimensional evidence profile, and the Backbone-Aware Comparison and Ablation Protocol (BCAP). The unresolved EK-100 record is treated as a reporting-comparability case study and is not used as a leaderboard. Evidence for LLM benefits, goal ambiguity, and horizon effects is therefore formulated as testable hypotheses requiring matched validation, not as causal conclusions. The accompanying package contains the coded evidence, source locators, protocol metadata, and versioned catalogue used in the review.
Chinese Translation
动作预测是在不完整的上下文和时间不确定性下，从部分视频中预测未来人类动作的过程。近期的系统在不同阶段引入了大型语言模型（LLMs）、视觉-语言模型（VLMs）或基于语言的语义，但当任务表述、视觉预训练、监督、解码器设计和评估代码同时变化时，报告的收益难以解释。本综述的核心贡献是一个证据感知的设计图，交叉了任务范畴与语言衍生信息介入的时点。我们沿六个轴线对任务范畴进行了特征化。这些轴线将文献组织为五个广泛的任务家族：单一动作、序列、物体交互、跨视角和面向规划的设置。C1-C3定位了在上下文构建、目标/意图建模和未来解码中的介入，而C4被视为一个相邻的新兴基础/可执行性扩展。与通用处理流程不同，该图将每个介入与适当的反事实、失败诊断和允许的证据声明联系起来。支持性贡献包括对Ego4D-LTA和EPIC-KITCHENS-100的协议级审计、多维证据概况，以及主干感知比较与消融协议（BCAP）。未解决的EK-100记录被视为一个报告可比性案例研究，而不作为排行榜使用。因此，LLM的益处、目标模糊性和视野效应的证据被表述为需要匹配验证的可测试假设，而非因果结论。随附的包包含了综述中使用的编码证据、源定位器、协议元数据和版本目录。
cs.CV / 9 / 2609.31668
Query-aligned video frame selection for long video understanding
针对长视频理解的查询对齐视频帧选择
Abstract
Multimodal large language models (MLLMs) process multimodal inputs by converting text, images, and videos into token sequences that are subsequently processed by a backbone language model. While MLLMs have achieved excellent performance in understanding the content of individual images, video understanding remains significantly more difficult because videos contain large number of video frames. MLLMs typically process only a subset of these frames, usually ranging from 8 to 64. MLLMs usually sample frames uniformly, regardless of their relevance to the question being answered. To address this limitation, several training-free, model-agnostic methods for selecting question-relevant frames have recently been proposed. In this work, we introduce a frame-selection method designed specifically for multiple-choice questions. We extend the query text by appending semantic cues derived from the answer choices and employ a direct query-frame alignment scoring mechanism. To the best of our knowledge, our method is the first to directly utilize answer choices as inference-time cues for selecting frames relevant to answering a question. The method first constructs a compact candidate pool by subsampling video frames at a fixed rate. The frames are then scored according to their maximum cosine similarity across all question-answer pairs to identify the most relevant frames for a given query. This approach preserves a fixed token budget while improving the relevance of the visual evidence provided to the downstream MLLM. We evaluate the effectiveness of our frame-selection method on the MLVU, Video-MME, and LongVideoBench benchmarks using three MLLMs: LLaVA-Mini, Qwen2-VL, and LLaVA-Video. Experimental results demonstrate that answer-aware frame selection generally outperforms uniform sampling and existing training-free frame-selection methods under the same frame budget.
Chinese Translation
多模态大型语言模型（MLLMs）通过将文本、图像和视频转换为标记序列来处理多模态输入，这些序列随后由主干语言模型进行处理。尽管MLLMs在理解单个图像内容方面取得了优异的表现，但视频理解仍然显著更具挑战性，因为视频包含大量视频帧。MLLMs通常仅处理这些帧的一个子集，通常范围在8到64帧之间。MLLMs通常以均匀的方式采样帧，而不考虑它们与所回答问题的相关性。为了解决这一限制，最近提出了几种无训练、模型无关的问题相关帧选择方法。在本研究中，我们提出了一种专门为多项选择题设计的帧选择方法。我们通过附加来自答案选项的语义线索来扩展查询文本，并采用直接的查询-帧对齐评分机制。根据我们的最佳知识，我们的方法是首个直接利用答案选项作为推理时线索来选择与回答问题相关的帧的方法。该方法首先通过以固定速率对视频帧进行子采样来构建一个紧凑的候选池。然后，根据所有问题-答案对之间的最大余弦相似度对帧进行评分，以识别与给定查询最相关的帧。这种方法在提高下游MLLM提供的视觉证据相关性的同时，保持了固定的标记预算。我们在MLVU、Video-MME和LongVideoBench基准上使用三种MLLM（LLaVA-Mini、Qwen2-VL和LLaVA-Video）评估了我们帧选择方法的有效性。实验结果表明，在相同的帧预算下，基于答案的帧选择通常优于均匀采样和现有的无训练帧选择方法。
cs.CV / 10 / 2609.31670
One-Step Is Optimal: Unconditional Rectified Flows are Noise2Noise Denoisers, and Multi-Step Integration Provably Hurts---A Benchmark and Task-Based Detectability Study on Low-Dose CT
一步法是最优的：无条件整流流是Noise2Noise去噪器，多步积分显著降低性能——低剂量CT的基准和基于任务的可检测性研究
Abstract
Iterative and generative denoisers are increasingly used under the assumption that multi-step refinement outperforms a single regression pass. We show the opposite for \emph{label-free} denoising. An \emph{unconditional} rectified flow trained on two noisy observations of the same signal, as in Noise2Noise, has a minimiser whose one-step readout is exactly the MMSE denoiser without requiring clean targets. In contrast, multi-step integration provably departs from the MMSE solution because the flow terminates at the noisy data distribution rather than the clean-signal distribution. This departure is exact in a tractable Gaussian model and is confirmed experimentally: one-step flow matches a direct regressor, whereas multi-step Euler integration progressively reduces fidelity. Counterintuitively, the degradation increases with training quality, as a better velocity field more faithfully transports samples toward the noisy terminal law. The key ingredient is therefore the decorrelated \emph{pairing}, not the flow machinery: a one-step regressor trained on matched noisy pairs gives our best label-free result ($+1.99$,dB). We evaluate these findings on \textbf{CTDenoiser}, a controlled low-dose CT benchmark spanning five architectures and supervised, similarity-based, blind-spot, and per-image methods. Among label-free approaches, only correlated-noise-aware Noise2Sim improves over the noisy baseline, while Noise2Void is flat-to-negative because CT noise violates its pixel-independence assumption. Finally, although supervised denoisers gain approximately $4$,dB PSNR, a channelized Hotelling observer shows reduced low-contrast lesion detectability, revealing clinically relevant degradation missed by PSNR and SSIM.
Chinese Translation
迭代和生成去噪器越来越多地被使用，假设多步细化优于单次回归。然而，我们展示了在 extit{无标签}去噪中相反的情况。一个在两个相同信号的噪声观测值上训练的 extit{无条件}整流流，如Noise2Noise中所示，其最小化器的一步输出正是MMSE去噪器，而无需干净的目标。相比之下，多步积分显著偏离MMSE解，因为流在噪声数据分布处终止，而不是在干净信号分布处。这种偏离在一个可处理的高斯模型中是准确的，并且通过实验得到了确认：一步流与直接回归器匹配，而多步欧拉积分逐渐降低了保真度。反直觉的是，随着训练质量的提高，降级现象也随之增加，因为更好的速度场更忠实地将样本传输到噪声终端分布。因此，关键因素是去相关的 extit{配对}，而不是流的机制：在匹配的噪声对上训练的一步回归器给出了我们最佳的无标签结果（+1.99 dB）。我们在 extbf{CTDenoiser}上评估这些发现，这是一个涵盖五种架构的受控低剂量CT基准，包括监督、基于相似性、盲点和逐图像方法。在无标签方法中，只有考虑相关噪声的Noise2Sim在噪声基线之上有所改善，而Noise2Void则表现平平或负面，因为CT噪声违反了其像素独立性假设。最后，尽管监督去噪器的PSNR大约提高了4 dB，但通道化Hotelling观察者显示低对比度病灶的可检测性降低，揭示了PSNR和SSIM未能捕捉到的临床相关降级。
cs.CV / 11 / 2609.31671
Unsupervised spiking feature learning for event-based pedestrian crossing detection: approaching supervised accuracy without labelled training data
无监督脉冲特征学习用于基于事件的行人过马路检测：在没有标注训练数据的情况下接近监督精度
Abstract
Event cameras are well suited to pedestrian crossing detection, and spiking neural networks (SNNs) can process their output natively, but current SNN detectors are trained with supervised backpropagation and therefore require costly frame-level crossing labels. We investigate crossing detection with no labels in feature learning and report the first unsupervised results on the recent DVS-PedX pedestrian-crossing benchmark. A single spiking layer trained with winner-take-all spike-timing-dependent plasticity learns a dictionary from unlabelled event-frame patches; frames are encoded by cosine similarity to the learned filters with spatial pooling and read out by a linear classifier, the only supervised component. On the 24,454-frame test split the method attains 90.3% accuracy and an area under the receiver operating characteristic curve (AUROC) of 0.936, compared with 92.0% and 0.943 for a supervised spiking network trained end-to-end on the same frames; under adverse weather the AUROC is 0.913. The result is insensitive to the choice of plasticity rule but depends strongly on the readout protocol: with the classical neuron-assignment readout the same network attains only 0.70 AUROC. On the benchmark's real converted portion, a readout refit lifts performance from chance (0.53) to 0.67 AUROC, within the published supervised range. These findings indicate that the accuracy cost of removing labels from feature learning is small on this benchmark, and that reported weaknesses of unsupervised spiking networks may be attributable to the readout protocol rather than to the learning rule.
Chinese Translation
事件相机非常适合行人过马路检测，而脉冲神经网络（SNN）能够原生处理其输出，但目前的 SNN 检测器是通过监督反向传播训练的，因此需要昂贵的帧级过马路标签。我们研究了在无标签的特征学习中进行过马路检测，并报告了在最近的 DVS-PedX 行人过马路基准测试中首次获得的无监督结果。一个使用赢家通吃的脉冲时序依赖可塑性训练的单层脉冲网络从无标签的事件帧补丁中学习字典；帧通过与学习到的滤波器的余弦相似度进行编码，并通过线性分类器读取，这是唯一的监督组件。在 24,454 帧的测试集上，该方法达到了 90.3% 的准确率和 0.936 的接收者操作特征曲线下面积（AUROC），而在同一帧上经过端到端训练的监督脉冲网络的准确率为 92.0%，AUROC 为 0.943；在恶劣天气条件下，AUROC 为 0.913。结果对可塑性规则的选择不敏感，但对读取协议高度依赖：使用经典的神经元分配读取时，同一网络仅达到 0.70 AUROC。在基准测试的真实转换部分，重新调整读取将性能从偶然（0.53）提升至 0.67 AUROC，处于已发布的监督范围内。这些发现表明，在该基准测试中，从特征学习中去除标签的准确性成本较小，并且无监督脉冲网络的报告弱点可能是由于读取协议而非学习规则造成的。
cs.CV / 12 / 2609.31677
A Comparative Transfer-Learning Study of CNN Backbones for Partial Face Recognition on the SoF Dataset
针对SoF数据集部分人脸识别的卷积神经网络主干的比较迁移学习研究
Abstract
Face recognition is widely deployed in surveillance, access control, and forensic workflows, yet accuracy degrades sharply once the face is occluded by accessories, foreground objects, or the frame edge. Because most faces in the wild are partial, robust partial face recognition (PFR) remains open. This paper compares three pretrained convolutional backbones, ResNet-50, VGG-16, and FaceNet, fine-tuned for PFR by transfer learning under identical preprocessing, splitting, and optimization protocols on the Specs-on-Faces (SoF) dataset. All three arms use a common 160x160 input and a frozen backbone with a trainable head under a fixed epoch budget and no per-backbone hyperparameter search. The FaceNet configuration, denoted PFN (Partial FaceNet), substantially outperforms the other two, reaching 97.4% test accuracy with macro-averaged 87.04% precision, 84.61% recall, and 84.17% F1 over the 112 identity classes, the highest accuracy and recall reported on SoF.
Chinese Translation
人脸识别广泛应用于监控、访问控制和法医工作流程，但一旦人脸被配饰、前景物体或框架边缘遮挡，识别准确率会急剧下降。由于自然环境中大多数人脸是部分可见的，因此稳健的部分人脸识别（PFR）仍然是一个未解决的问题。本文比较了三种预训练的卷积主干网络：ResNet-50、VGG-16和FaceNet，这些网络通过迁移学习在Specs-on-Faces（SoF）数据集上进行了针对PFR的微调，采用相同的预处理、划分和优化协议。所有三种模型均使用统一的160x160输入，冻结主干网络，并在固定的训练周期预算下使用可训练的头部，而不进行每个主干网络的超参数搜索。FaceNet配置（标记为PFN，即部分FaceNet）显著优于其他两者，在112个身份类别上达到了97.4%的测试准确率，以及宏平均的87.04%精确率、84.61%召回率和84.17% F1分数，创造了SoF数据集上报告的最高准确率和召回率。
cs.CV / 13 / 2609.31679
Toward AI-Assisted Poultry Coccidiosis Diagnosis: Evaluating Gemini and BiomedParse on Eimeria Microscopy Images
朝着人工智能辅助的家禽球虫病诊断：评估 Gemini 和 BiomedParse 在艾美耳虫显微镜图像上的表现
Abstract
Coccidiosis caused by Eimeria parasites is a major economic burden in poultry production, and effective control depends on accurate species-level diagnosis. This study evaluates whether a general-purpose multimodal large language model can support such diagnosis. Google Gemini was assessed on 4,225 mi- croscopy images covering the seven fowl-infecting Eimeria species under two prompting conditions, one without candidate labels and one with a predefined class list, and was further tested for pathology-report generation, while BiomedParse was examined for parasite segmentation. Without candidate labels, the model produced broad and taxonomically inconsistent outputs. With candidate labels, overall accuracy reached only 14.9%, with a strong bias toward E. tenella at 74% and no correct classifications for E. acervulina, E. mitis and E. praecox. Generated treatment reports were coherent but unverified, and segmentation was only partial. Current multimodal models are therefore not yet reliable for standalone Eimeria diagnosis without domain-specific fine- tuning and expert validation.
Chinese Translation
由艾美耳虫寄生虫引起的球虫病是家禽生产中的主要经济负担，有效控制依赖于准确的物种级诊断。本研究评估了一种通用多模态大型语言模型是否能够支持此类诊断。Google Gemini 在 4,225 张显微镜图像上进行了评估，这些图像涵盖了七种感染家禽的艾美耳虫物种，测试在两种提示条件下进行：一种是没有候选标签，另一种是有预定义的类别列表，并进一步测试了病理报告生成，而 BiomedParse 则用于寄生虫分割。在没有候选标签的情况下，该模型产生了广泛且在分类上不一致的输出。在有候选标签的情况下，整体准确率仅达到 14.9%，且对 E. tenella 的偏倚强烈，达到 74%，而对 E. acervulina、E. mitis 和 E. praecox 则没有正确分类。生成的治疗报告虽然连贯但未经验证，分割结果也仅为部分。因此，目前的多模态模型在没有领域特定的微调和专家验证的情况下，尚不可靠用于独立的艾美耳虫诊断。
cs.CV / 14 / 2609.31681
Devanagari Handwritten Character Recognition Using TrOCR: A Transformer-Based Model with Real-Time Web Deployment
基于 TrOCR 的德瓦那加里手写字符识别：一种具有实时网页部署的变换器模型
Abstract
Devnagari is a one of the ancient language of the Indian subcontinent consisting of 36 vowels, 14 consonants and 10 numerals. The accurate recognition of handwritten Devnagari characters is challenging due to high complexity of Devnagari scripts. This paper presents a method to fine tune the pre trained TrOCR model to accurately recognize Devanagari handwritten characters. The Methodology consists of a preprocessing mechanism where input images are standardized into RGB format, tokenized in batches and integrated with Hugging Face Dataset. The pre-trained microsoft/trocr-base-handwritten model is fine-tuned over a dataset of nearly 5000 character images that are uniformly partitioned in the ratio 8:1:1 for training, evaluation and testing. Further optimization is done is the training process through mixed precision training, gradient checkpointing, and early stopping mechanism. The model achieves a character error rate (CER) of 3.95% and a character-level accuracy of 96.05%, outperforming the previous CNN based models. A scalable web application is developed using Vite.js, Golang, and FastAPI which practically deploys the OCR model and serves character recognition task with a latency less than 5 seconds per request. This study demonstrates the use of TrOCR model to build a scalable handwritten Devnagari character recognition system and also a foundation to future research on Devnagari Script Recognition using Transformers.
Chinese Translation
德瓦那加里是印度次大陆的一种古老语言，包含 36 个元音、14 个辅音和 10 个数字。由于德瓦那加里书写的复杂性，准确识别手写德瓦那加里字符具有挑战性。本文提出了一种方法，通过微调预训练的 TrOCR 模型，准确识别德瓦那加里手写字符。该方法论包括一个预处理机制，将输入图像标准化为 RGB 格式，按批次进行标记，并与 Hugging Face 数据集集成。预训练的 microsoft/trocr-base-handwritten 模型在近 5000 张字符图像的数据集上进行了微调，这些图像按 8:1:1 的比例均匀划分为训练、评估和测试集。在训练过程中，通过混合精度训练、梯度检查点和早停机制进一步优化模型。该模型实现了 3.95% 的字符错误率 (CER) 和 96.05% 的字符级准确率，优于之前基于 CNN 的模型。使用 Vite.js、Golang 和 FastAPI 开发了一个可扩展的网页应用程序，实际部署了 OCR 模型，并以每个请求少于 5 秒的延迟完成字符识别任务。本研究展示了使用 TrOCR 模型构建可扩展的德瓦那加里手写字符识别系统的可能性，并为未来基于变换器的德瓦那加里脚本识别研究奠定了基础。
cs.CV / 15 / 2609.31682
Towards Transparent Diagnostics: Investigating Architectural Trade-offs and Explainability in Malaria Detection
迈向透明的诊断：研究疟疾检测中的架构权衡与可解释性
Abstract
More than 80 countries have reported malaria cases with 610 thousand deaths and are projected to increase. Identifying malaria early and accurately helps save lives. The effective way to diagnose malaria is through microscopic methods that are labor intensive and require experts with special equipment. Deep learning (DL) has shown promising results in medical diagnosis. Here, we explored various DL models: ResNet18, MobileNetV2, EfficientNet-B2, VGG19 and also proposed a model for detecting malaria presence using blood smears taken from the NIH Malaria dataset. Our experiment shows that MobileNetV2 achieved 96.35% accuracy with the smallest model size (8.49 MB) and fastest inference (1.35 ms). The proposed model achieved 97.67% accuracy, 0.9756 AUC with longest inference time (13.17 ms). The larger architecture outputs a larger model size with moderate accuracy. Upon further pruning, the proposed model gained a slight improvement in accuracy and inference time. The GRAD-CAM, SHAP and LIME shade explainable AI (XAI) insights of the model.
Chinese Translation
超过80个国家报告了疟疾病例，导致61万人死亡，并预计这一数字将继续增加。早期准确识别疟疾有助于挽救生命。有效的疟疾诊断方法是通过显微镜技术，这种方法劳动密集且需要具备特殊设备的专家。深度学习（Deep Learning, DL）在医学诊断中显示出良好的前景。在本研究中，我们探索了多种深度学习模型：ResNet18、MobileNetV2、EfficientNet-B2、VGG19，并提出了一种基于美国国立卫生研究院（NIH）疟疾数据集的血涂片检测疟疾存在的模型。实验结果表明，MobileNetV2以最小的模型大小（8.49 MB）和最快的推理时间（1.35 ms）达到了96.35%的准确率。所提模型的准确率为97.67%，AUC为0.9756，但推理时间最长（13.17 ms）。较大的架构输出了更大的模型大小，但准确率适中。经过进一步修剪，所提模型在准确率和推理时间上略有改善。GRAD-CAM、SHAP和LIME提供了该模型的可解释人工智能（Explainable AI, XAI）洞察。
cs.CV / 16 / 2609.31690
Adapting Vision-Language Models for Human-Readable XAI in Industrial Object Detection
为工业物体检测适应人类可读的可解释人工智能（XAI）视觉-语言模型
Abstract
Explainable Artificial Intelligence (XAI) solutions are essential for building trust in AI technologies and their integration in real manufacturing lines. However, most existing methods are tailored to technical experts, limiting their accessibility to diverse user groups such as blue-collar workers in manufacturing lines who use AI for quality control. In this work, we introduce an XAI interface for object detection in industrial manufacturing based on a fine-tuned vision-language model, designed to generate intuitive explanations for non-expert users. We benchmark existing vision-language models and demonstrate that out-of-the-box models often fall short in delivering clear, context-relevant explanations for non-expert users. To address this, we fine-tune a vision-language model and integrate it into our interface, enabling contextualized, accessible explanations for non-expert users. We demonstrate improvements in explanation clarity, instruction adherence, image groundedness, and contextual awareness over GPT 4o-mini on proprietary and public robotics dataset. This approach advances the accessibility and usability of AI explanations, making them more intuitive and applicable in manufacturing domain.
Chinese Translation
可解释人工智能（XAI）解决方案对于建立对人工智能技术的信任以及其在实际制造线中的整合至关重要。然而，现有的大多数方法都是针对技术专家量身定制的，限制了其对多样化用户群体的可访问性，例如在制造线中使用人工智能进行质量控制的蓝领工人。在本研究中，我们基于经过微调的视觉-语言模型，提出了一种用于工业制造物体检测的XAI接口，旨在为非专家用户生成直观的解释。我们对现有的视觉-语言模型进行了基准测试，并证明现成的模型在为非专家用户提供清晰、与上下文相关的解释方面往往不足。为了解决这一问题，我们对视觉-语言模型进行了微调，并将其集成到我们的接口中，使非专家用户能够获得上下文化、可访问的解释。我们在专有和公共机器人数据集上展示了在解释清晰度、指令遵循、图像基础性和上下文意识方面相较于GPT 4o-mini的改进。这种方法提高了人工智能解释的可访问性和可用性，使其在制造领域更加直观和适用。
cs.CV / 17 / 2609.31692
Architecture-aware Robustness Evaluation of Explainable Deep Learning for Breast Cancer Diagnosis
考虑架构的可解释深度学习在乳腺癌诊断中的鲁棒性评估
Abstract
Explainable Artificial Intelligence (XAI) has become essential in medical image analysis to ensure transparency of deep learning (DL)-based diagnostic systems. However, selecting appropriate XAI techniques for breast cancer recognition remains largely ad hoc, with limited systematic evaluation across different DL architectures. This study presents a systematic architecture-aware evaluation protocol to assess the effectiveness of nine widely used XAI techniques across four categories of DL models: very deep, lightweight, transformer-based and hybrid neural networks. The evaluation is conducted on a breast ultrasound dataset comprising 780 images using clinically aligned spatial metrics, including Pointing Game, Intersection over Union and Mean Coverage, to quantify agreement between generated explanations and expert-annotated lesion regions. Results indicate that explanation quality is primarily influenced by the interaction between model architecture and XAI method, rather than any single technique consistently outperforming others. Hybrid architectures produce more spatially coherent explanations, while lightweight and transformer-based models exhibit greater variability across methods. The findings show that no single technique generalises across architectures and evaluation criteria, emphasising the need for joint selection of DL models and XAI techniques. Explainability depends on both model design and explanation strategy and should not be considered independently. This work provides a structured evaluation protocol and practical guidance for selecting XAI techniques in breast cancer diagnosis, supporting more transparent clinical decision-support systems. \textcolor{blue}{Code is publicly available at https://github.com/Nishan-Charlie/Explainable-AI}
Chinese Translation
可解释人工智能（XAI）在医学图像分析中变得至关重要，以确保基于深度学习（DL）的诊断系统的透明性。然而，针对乳腺癌识别选择合适的XAI技术仍然主要是临时性的，且在不同的DL架构之间的系统评估有限。本研究提出了一种系统的考虑架构的评估协议，以评估九种广泛使用的XAI技术在四类DL模型（非常深度、轻量级、基于变换器和混合神经网络）中的有效性。评估是在一个包含780幅图像的乳腺超声数据集上进行的，使用临床对齐的空间度量，包括指向游戏（Pointing Game）、交并比（Intersection over Union）和平均覆盖率（Mean Coverage），以量化生成的解释与专家标注的病变区域之间的一致性。结果表明，解释质量主要受到模型架构与XAI方法之间相互作用的影响，而不是任何单一技术始终优于其他技术。混合架构产生了更具空间一致性的解释，而轻量级和基于变换器的模型在方法之间表现出更大的变异性。研究结果表明，没有单一技术能够在所有架构和评估标准中普遍适用，强调了DL模型与XAI技术的联合选择的必要性。可解释性依赖于模型设计和解释策略，不能独立考虑。本研究提供了一种结构化的评估协议和选择乳腺癌诊断中XAI技术的实用指导，支持更透明的临床决策支持系统。代码可在 https://github.com/Nishan-Charlie/Explainable-AI 上公开获取。
cs.CV / 18 / 2609.31693
Disentangle and Drop: Robust Universal Removal of Image Watermarks via Reconstructive Grayscale Residual Decomposition
解耦与丢弃：通过重构灰度残差分解实现图像水印的鲁棒性通用去除
Abstract
Invisible image watermarks are commonly evaluated against benign postprocessing operations such as compression, resizing, blur, and color changes. These tests leave out a different threat: a learned remover that preserves semantic image content while discarding residual evidence that carries the payload. We propose Disentangle and Drop (DnD), an attack that is agnostic to the watermark method and treats watermark removal as a representation routing problem. DnD decomposes a watermarked image into a semantic grayscale carrier and an auxiliary residual branch, and then suppresses the residual branch to reduce watermark evidence. The model is trained with latent spectral perturbations and low-strength diffusion exposure so that the drop operation remains stable under adaptive reconstruction. Experiments on seven representative watermark families show that one shared operating setting gives competitive removal with high visual fidelity. Operating scans and ablations separate usable attacks from image-damaging settings: stronger noise or diffusion can raise removal scores by damaging the image, while the practical regime comes from dropping the residual latent. These results argue for evaluating watermark robustness against learned removal at the representation level, not only against conventional image edits.
Chinese Translation
隐形图像水印通常在压缩、调整大小、模糊和颜色变化等良性后处理操作下进行评估。这些测试忽略了另一种威胁：一种能够在保留语义图像内容的同时丢弃携带有效载荷的残余证据的学习去除器。我们提出了解耦与丢弃（Disentangle and Drop, DnD），这是一种对水印方法无关的攻击，将水印去除视为一种表示路由问题。DnD将带水印的图像分解为语义灰度载体和辅助残差分支，然后抑制残差分支以减少水印证据。该模型通过潜在光谱扰动和低强度扩散暴露进行训练，以确保在自适应重构下丢弃操作保持稳定。在七个代表性水印系列上的实验表明，单一共享操作设置可以实现具有高视觉保真度的竞争性去除。操作扫描和消融实验将可用攻击与损害图像的设置区分开来：更强的噪声或扩散可以通过损害图像来提高去除分数，而实用的机制则来自于丢弃残余潜在。这些结果支持在表示层面评估水印鲁棒性，以应对学习去除，而不仅仅是针对传统图像编辑。
cs.CV / 19 / 2609.31694
RemTraceNet: Few-Shot Forensic Detection of Invisible Watermark Attacks
RemTraceNet：少样本法对隐形水印攻击的取证检测
Abstract
Removing an invisible watermark and concealing the forensic evidence are distinct objectives: successfully disrupting the embedded watermark does not imply that the removal process is forensically undetectable. When verification fails, removal traces can provide complementary evidence for provenance and ownership verification, whereas their absence leaves the cause of the failure ambiguous. Existing methods are typically evaluated by watermark suppression and perceptual quality, while forensic stealth is rarely considered. We therefore study watermark-attack-specific few-shot forensics: for each known pipeline, a specialist can separate its outputs from paired clean and unattacked watermarked controls. Separate Attack-vs-Clean and Attack-vs-Watermarked evaluations prevent watermark-presence shortcuts. Image-aligned and prompt-matched controls are used for post-hoc and generator-integrated schemes, respectively. In this work, we introduce RemTraceNet, which fuses constrained residuals, local relations, FFT/Haar statistics, and block-DCT evidence at native resolution. Across 23 removal pipelines and 10 watermark configurations, we evaluate native 256 x 256 and 512 x 512 inputs. With 100 attacked training images per pipeline, the three-seed TPR@1%FPR, macro-averaged over attacks and watermark configurations, ranges from 82.75% to 88.17% across resolutions and control types. Under the condition of same labels and protocol, RemTraceNet outperforms retrained SRNet, ZhuNet, and SiaStegNet baselines by 10.80--15.68 percentage points. Extensive experimental results show that erasing a watermark and erasing evidence of its removal are distinct challenges, and that removal traces remain learnable under limited supervision.
Chinese Translation
去除隐形水印和隐藏取证证据是两个不同的目标：成功破坏嵌入的水印并不意味着去除过程在取证上是不可检测的。当验证失败时，去除痕迹可以为来源和所有权验证提供补充证据，而其缺失则使失败的原因变得模糊。现有方法通常通过水印抑制和感知质量进行评估，而取证隐蔽性则很少被考虑。因此，我们研究了针对水印攻击的少样本取证：对于每个已知的处理流程，专家可以将其输出与配对的干净和未攻击的水印控制进行区分。单独的攻击与干净和攻击与水印的评估防止了水印存在的捷径。图像对齐和提示匹配的控制分别用于事后和生成器集成方案。在本研究中，我们引入了RemTraceNet，它在原生分辨率下融合了受限残差、局部关系、FFT/Haar统计和块DCT证据。在23个去除处理流程和10种水印配置下，我们评估了原生256 x 256和512 x 512的输入。在每个处理流程中使用100张被攻击的训练图像，三种种子的TPR@1%FPR在不同分辨率和控制类型下，针对攻击和水印配置的宏平均范围为82.75%至88.17%。在相同标签和协议的条件下，RemTraceNet的表现优于重新训练的SRNet、ZhuNet和SiaStegNet基线，提升幅度为10.80至15.68个百分点。大量实验结果表明，去除水印和去除其去除证据是两个不同的挑战，并且去除痕迹在有限监督下仍然是可学习的。
cs.CV / 20 / 2609.31697
Video Captioning in Low-Light Conditions through Efficient Uncertainty-Aware Caption Correction
在低光照条件下通过高效的不确定性感知字幕修正进行视频字幕生成
Abstract
Low-light conditions can significantly degrade the ability of vision-language models (VLMs) to accurately describe human actions in videos. In this work, I propose an efficient uncertainty-aware representation correction framework for improving captions generated by VideoChat2 under real-world low-light conditions. Instead of fine-tuning the underlying VLM, the proposed framework introduces a lightweight sparse Gaussian process-based error estimation module between the projection layer and the language model to correct the intermediate representation. The correction module learns to estimate the residual between the original projected representation and a verified target representation, which is then adaptively scaled using a newly formulated uncertainty-aware coefficient and added to the original representation. To further improve residual estimation, I introduce a partitioned combined-kernel design. The correction model is trained separately using only 44 samples from the ARID dataset and requires only a small additional computational overhead during inference. The effectiveness of the proposed correction is evaluated through quantitative residual prediction and qualitative analysis of the generated captions. Although VideoChat2 is used in my experiments, the proposed framework is designed to be applicable to other compatible VLM architectures. \textbf{Code Availability: The implementation accompanying this work is publicly available at}:\href{https://github.com/areferezaee/Low-rank-SVGP-NP-update}{https://github.com/areferezaee/Low-rank-SVGP-NP-update}
Chinese Translation
低光照条件会显著降低视觉-语言模型（VLMs）准确描述视频中人类动作的能力。在本研究中，我提出了一种高效的不确定性感知表示修正框架，以改善在现实世界低光照条件下由VideoChat2生成的字幕。该框架并不对底层VLM进行微调，而是在投影层和语言模型之间引入了一个基于稀疏高斯过程的轻量级误差估计模块，以修正中间表示。修正模块学习估计原始投影表示与经过验证的目标表示之间的残差，然后使用新制定的不确定性感知系数进行自适应缩放，并添加到原始表示中。为了进一步提高残差估计的准确性，我引入了分区组合核设计。修正模型仅使用ARID数据集中的44个样本进行单独训练，并且在推理过程中仅需少量额外的计算开销。通过定量残差预测和生成字幕的定性分析评估所提修正的有效性。尽管我的实验中使用了VideoChat2，但所提框架旨在适用于其他兼容的VLM架构。\textbf{代码可用性：与本工作相关的实现可在以下链接公开获取：}: exttt{https://github.com/areferezaee/Low-rank-SVGP-NP-update}
cs.CV / 21 / 2609.31698
RPA: Residual Patch-Token Adapter for Image Retrieval from EEG and MEG
RPA：用于从脑电图（EEG）和脑磁图（MEG）中提取图像的残差补丁-标记适配器
Abstract
Most existing MEG and EEG (M/EEG) visual decoding methods align brain signals with a single global embedding extracted from a pretrained visual encoder, leaving open whether intermediate patch representations, which preserve richer and more granular rich visual information, can improve representation learning. To address this question, we introduce the Residual Patch Adapter (RPA), a lightweight, modular adapter that leverages all patch tokens from an intermediate layer of a ViT visual encoder for alignment. Through extensive ablation analyses, we first show that pooling or masking patch tokens degrades the learned representation, demonstrating that retaining the full set of patch tokens is important for EEG alignment, while the CLS token provides little unique information. We then use a series of six quantitative feature analyses to show that both higher-level semantics and lower-level visual features, including color and texture, are essential for this EEG-to-image alignment. Under current protocols, our system achieves Top-1 accuracies of 95.4\% within-subject and 35.5\% cross-subject on THINGS-EEG2, and 65.2\% and 6.7\%, respectively, on THINGS-MEG, achieving state-of-the-art (SOTA) performance across both datasets. Evaluations with alternative brain encoders, including pretrained EEG foundation models, demonstrate that the approach extends beyond the projection-based EEG encoder. Furthermore, we provide a plug-and-play interface that allows RPA to be replaced by convolution, attention, or ConvNeXt alternatives. Together, these findings provide significant insight into M/EEG-to-image representation learning by establishing design principles for leveraging the latent space of visual encoders, and open new directions for brain--image alignment and non-invasive brain--computer interface (BCI).
Chinese Translation
大多数现有的脑磁图（MEG）和脑电图（EEG）视觉解码方法将脑信号与从预训练视觉编码器提取的单一全局嵌入对齐，尚未探讨保留更丰富和更细粒度视觉信息的中间补丁表示是否能改善表示学习。为了解决这个问题，我们引入了残差补丁适配器（Residual Patch Adapter, RPA），这是一种轻量级的模块化适配器，利用来自视觉编码器（ViT）中间层的所有补丁标记进行对齐。通过广泛的消融分析，我们首先表明，池化或掩蔽补丁标记会降低学习到的表示，证明保留完整的补丁标记集对于EEG对齐的重要性，而CLS标记提供的独特信息较少。然后，我们使用六项定量特征分析表明，高层语义和低层视觉特征（包括颜色和纹理）对于EEG到图像的对齐至关重要。在当前的协议下，我们的系统在THINGS-EEG2数据集上实现了95.4%的个体内Top-1准确率和35.5%的个体间准确率，在THINGS-MEG数据集上分别为65.2%和6.7%，在这两个数据集上均达到了最先进的（SOTA）性能。与其他脑编码器（包括预训练EEG基础模型）的评估表明，该方法超越了基于投影的EEG编码器。此外，我们提供了一个即插即用的接口，允许用卷积、注意力或ConvNeXt替代RPA。这些发现为M/EEG到图像的表示学习提供了重要的见解，建立了利用视觉编码器潜在空间的设计原则，并为脑-图像对齐和非侵入式脑-计算机接口（BCI）开辟了新的方向。
cs.CV / 22 / 2609.31700
Modernising the Compressed-Domain Video Captioner: A Controlled Study of SigLIP2 and GPT-2 Substitutions
现代化压缩域视频字幕生成器：SigLIP2与GPT-2替代的对照研究
Abstract
Compressed-domain video captioning avoids full video decoding by operating directly on I-frames, motion vectors and residuals, trading a small amount of accuracy for a large gain in inference speed. CoCap established this pipeline using a CLIP vision encoder and a shallow BERT-style multimodal decoder. Both components predate substantially stronger alternatives. We ask a narrow, controlled question: how much of CoCap's accuracy is limited by these two components, and which of the two is the binding constraint? We replace the CLIP I-frame encoder with SigLIP2 and the BERT-style decoder with GPT-2, and evaluate three configurations (the original pairing, the encoder substitution alone, and both substitutions together) under identical data, sampling budget and optimisation schedule. All comparisons are made against our own reproduction of CoCap rather than its published numbers, because we train on a 4,999-clip subset of VATEX at a reduced sampling budget; absolute values are therefore not comparable with the original work. Our reproduction tracks the published result closely: CIDEr and METEOR run slightly above it (54.9 against 52.7; 23.4 against 23.2), BLEU-4 and ROUGE-L slightly below (29.7 against 31.4; 48.9 against 49.4). We attribute the differences to our evaluation subset rather than to any improvement in either direction. We find the two substitutions pull in opposite directions: SigLIP2 alone improves every metric (+4.5 CIDEr), while adding GPT-2 on top erodes that gain, because a pretrained decoder overfits 4,999 clips within two epochs. We additionally report inference latency for each configuration, since speed is the property that motivates compressed-domain captioning in the first place, and an accuracy gain purchased at a latency cost should be reported as such.
Chinese Translation
压缩域视频字幕生成通过直接处理I帧、运动矢量和残差，避免了完整视频解码，从而在牺牲少量准确性的同时获得了显著的推理速度提升。CoCap使用CLIP视觉编码器和浅层BERT风格的多模态解码器建立了这一流程。这两个组件在很大程度上早于更强大的替代品。我们提出一个狭窄的、受控的问题：CoCap的准确性有多少是受到这两个组件的限制，以及这两者中哪个是主要约束？我们用SigLIP2替换CLIP I帧编码器，用GPT-2替换BERT风格解码器，并在相同的数据、采样预算和优化计划下评估三种配置（原始组合、仅编码器替换以及两者同时替换）。所有比较都是基于我们自己对CoCap的重现，而不是其发布的数字，因为我们在减少的采样预算下对VATEX的4,999个片段子集进行训练；因此，绝对值无法与原始工作进行比较。我们的重现结果与已发布结果非常接近：CIDEr和METEOR略高于其值（54.9对52.7；23.4对23.2），而BLEU-4和ROUGE-L略低（29.7对31.4；48.9对49.4）。我们将这些差异归因于我们的评估子集，而非任何方向上的改进。我们发现这两种替代品的效果相反：单独使用SigLIP2改善了每个指标（+4.5 CIDEr），而在此基础上添加GPT-2则削弱了这一增益，因为预训练的解码器在两个周期内过拟合了4,999个片段。我们还报告了每种配置的推理延迟，因为速度是推动压缩域字幕生成的根本属性，而以延迟成本换取的准确性提升应当如实报告。
cs.CV / 23 / 2609.31701
Integrated Deep Learning Framework Designed on Hybrid Optimization Strategies for Automated Health Detection and Analysis in Silkworms
基于混合优化策略的集成深度学习框架用于蚕的自动健康检测与分析
Abstract
A Hybrid Residual Attention Network is proposed for accurately classifying silkworm images into six different classes, including healthy and diseased states. It uses residual blocks for deep feature extraction and attention to focus on disease related features. A novel Integrated Adaptive Momentum Optimizer was introduced to enhance convergence and improve training efficiency. The dataset of silkworm images underwent preprocessing techniques such as normalization, resizing, and noise reduction, along with augmentation strategies to improve data quality and diversity. It is optimized using IAMO, achieved an accuracy of 98.67%.The integration of spatial and channel wise attention mechanisms, coupled with IAMO, significantly enhanced the model ability to recognize subtle differences between classes. Results indicate that HRAN can be used to detect disease at an early stage in sericulture, and future work will enhance scalability and efficiency in different environments.
Chinese Translation
提出了一种混合残差注意力网络（Hybrid Residual Attention Network），用于将蚕图像准确分类为六个不同类别，包括健康和病态。该网络利用残差块进行深度特征提取，并通过注意力机制聚焦于与疾病相关的特征。引入了一种新颖的集成自适应动量优化器（Integrated Adaptive Momentum Optimizer），以增强收敛性并提高训练效率。蚕图像数据集经过了归一化、调整大小和降噪等预处理技术，以及数据增强策略，以提高数据质量和多样性。使用IAMO进行优化，达到了98.67%的准确率。空间和通道注意力机制的结合，以及IAMO的应用，显著增强了模型识别类别间细微差异的能力。结果表明，HRAN可以用于蚕业早期阶段的疾病检测，未来的工作将提升其在不同环境中的可扩展性和效率。
cs.CV / 24 / 2609.31702
CLC-YOLO: A Compact Channel-Gated Prototype Network for Real-Time Leakage-Aware Breast Ultrasound Lesion Segmentation
CLC-YOLO：一种紧凑的通道门控原型网络用于实时漏泄感知乳腺超声病灶分割
Abstract
Reliable breast ultrasound lesion segmentation requires accurate boundaries and evaluation that prevents patients or duplicate images from crossing data splits. We propose Channel Local Contrast (CLC), a compact refinement of the YOLO26 segmentation prototype head. CLC adds a fixed local high-pass residual controlled by 64 zero-initialized, bounded channel gates. Baseline and CLC were compared in five matched folds on each of four breast ultrasound datasets. BUS-BRA used patient-disjoint outer tests with separate inner validation. BUS-UCLM and BrEaST used patient-grouped validation folds; BUSI used duplicate-component groups because patient identifiers are unavailable. Group-macro Dice increased by 1.68, 3.11, 1.12, and 2.45 percentage points on BUS-BRA, BUS-UCLM, BUSI, and BrEaST, respectively. Only the BUS-BRA paired 95% confidence interval excluded zero. CLC adds 64 parameters and 0.0049 giga floating-point operations (GFLOPs). At 640 pixels, single-T4, batch-one, 16-bit floating-point (FP16) TensorRT graph times were 2.1478 ms for CLC and 2.0512 ms for baseline, excluding preprocessing and postprocessing. CLC increased group-macro Dice across all four datasets with a measured T4 forward-pass overhead of 0.0966 ms. Code: https://github.com/mfazrinizar/CLC-YOLO
Chinese Translation
可靠的乳腺超声病灶分割需要准确的边界和评估，以防止患者或重复图像跨越数据划分。我们提出了通道局部对比（Channel Local Contrast, CLC），这是对YOLO26分割原型头的紧凑改进。CLC增加了一个由64个零初始化的、有界通道门控控制的固定局部高通残差。在四个乳腺超声数据集的每个五个匹配折叠中比较了基线和CLC。BUS-BRA使用了患者不重叠的外部测试，并进行了单独的内部验证。BUS-UCLM和BrEaST使用了患者分组的验证折叠；由于患者标识符不可用，BUSI使用了重复组件组。GROUP-MACRO Dice在BUS-BRA、BUS-UCLM、BUSI和BrEaST上分别增加了1.68、3.11、1.12和2.45个百分点。只有BUS-BRA的95%置信区间排除了零。CLC增加了64个参数和0.0049亿浮点运算（GFLOPs）。在640像素下，单个T4、批量为1、16位浮点（FP16）TensorRT图的时间为CLC 2.1478毫秒，基线为2.0512毫秒，不包括预处理和后处理。CLC在所有四个数据集上都提高了GROUP-MACRO Dice，测得的T4前向传递开销为0.0966毫秒。代码：https://github.com/mfazrinizar/CLC-YOLO
cs.CV / 25 / 2609.31705
MDL-Calibrated Significance-Gain Pair Encoding: Replication-Aware Automatic Stopping for Subword Tokenization
MDL校准的显著性增益对编码：面向复制的子词分词自动停止
Abstract
Byte-Pair Encoding (BPE) constructs subword vocabularies through greedy pair merging, but conventional BPE requires the number of merges or target vocabulary size to be specified externally. Significance-Gain Pair Encoding (SG-BPE) replaces frequency-only selection with a statistical criterion based on how strongly an observed pair exceeds its expected co-occurrence under an independence model. This paper introduces MDL-Calibrated Significance-Gain Pair Encoding (MDL-SG), a three-stage procedure separating discovery, replication, and utility. Candidate pairs are ranked by Significance-Gain on a discovery partition, tested for replication on a separate partition using an exact one-sided hypergeometric test with per-iteration Benjamini-Hochberg correction, and then evaluated on a utility partition using a Minimum Description Length (MDL) criterion. Merging stops automatically when no replicated candidate yields positive held-out MDL gain. On WikiText-103, MDL-SG stops at 209, 433, and 847 merges for 120K, 250K, and 500K-character tokenizer-training samples, respectively. At 500K characters, it selects a stored vocabulary of 1,017 tokens without prescribing the vocabulary size in advance. In a compute-matched TinyGPT experiment with identical 2,024,448-parameter models and 500 optimizer updates per language model, MDL-SG achieves validation/test BPC of 3.2612/3.2436, compared with test BPC of 3.2894 for SG-BPE and 3.3493 for frequency BPE. Frequency BPE achieves stronger raw compression, while MDL-SG achieves lower BPC, showing that compression-oriented merge selection and language-model utility need not coincide.
Chinese Translation
字节对编码（Byte-Pair Encoding, BPE）通过贪婪的对合并构建子词词汇，但传统的BPE需要外部指定合并次数或目标词汇大小。显著性增益对编码（Significance-Gain Pair Encoding, SG-BPE）用基于观察对在独立模型下超出预期共现程度的统计标准替代了仅基于频率的选择。本文介绍了MDL校准的显著性增益对编码（MDL-SG），这是一种将发现、复制和效用分开的三阶段程序。候选对在发现分区中根据显著性增益进行排名，在单独的分区中使用精确的单边超几何检验和每次迭代的Benjamini-Hochberg校正进行复制测试，最后在效用分区中使用最小描述长度（Minimum Description Length, MDL）标准进行评估。当没有复制候选对产生正的保留MDL增益时，合并自动停止。在WikiText-103上，MDL-SG在120K、250K和500K字符的分词训练样本中分别在209、433和847次合并时停止。在500K字符时，它选择了一个包含1,017个标记的存储词汇，而无需提前规定词汇大小。在一个计算匹配的TinyGPT实验中，使用相同的2,024,448参数模型和每个语言模型500次优化更新，MDL-SG实现了3.2612/3.2436的验证/测试BPC，而SG-BPE的测试BPC为3.2894，频率BPE为3.3493。频率BPE实现了更强的原始压缩，而MDL-SG实现了更低的BPC，表明以压缩为导向的合并选择和语言模型效用不必一致。
cs.CV / 26 / 2609.31706
Can't Find Waldo: Evaluating VLMs' Sensitivity to Image Resolution and Detail Level
找不到沃尔多：评估视觉语言模型对图像分辨率和细节水平的敏感性
Abstract
Visual Language Models (VLMs) have achieved remarkable success across diverse tasks, yet they struggle with high-resolution inputs where critical information resides in small regions or detailed, cluttered scenes. While several approaches address this limitation, a systematic understanding of why models fail at high resolutions is lacking. We introduce a controlled evaluation framework that disentangles resolution-related performance degradation from task difficulty through semantics-preserving transformations. We propose two simple metrics: Area Under the Scaling Curve (AUSC), which quantifies scaling robustness independent of baseline accuracy, and Prediction Variance Score (PVS), which measures resolution-induced prediction instability. Through comprehensive experiments across 5 model families and 5 benchmarks, we identify three primary failure modes: (1) information loss from downsampling at vision token limits, (2) tokenization artifacts from patch boundary shifts and positional encoding fragility under non-standard aspect ratios, and (3) attention dilution as token counts increase. Our analysis reveals that even state-of-the-art models suffer from performance drops when processing high-resolution images, with degradation patterns varying systematically by architectural family. We provide actionable insights for model architecture design and data augmentation strategies to mitigate these limitations.
Chinese Translation
视觉语言模型（VLMs）在多种任务中取得了显著成功，但在高分辨率输入下，它们在小区域或细致、杂乱场景中关键的信息处理上表现不佳。尽管已有多种方法试图解决这一局限性，但对模型在高分辨率下失败原因的系统理解仍然缺乏。我们引入了一种受控评估框架，通过保持语义不变的变换，将与分辨率相关的性能下降与任务难度区分开来。我们提出了两个简单的指标：缩放曲线下的面积（Area Under the Scaling Curve, AUSC），该指标量化了与基线准确性无关的缩放鲁棒性，以及预测方差得分（Prediction Variance Score, PVS），该指标测量了由分辨率引起的预测不稳定性。通过对5个模型系列和5个基准的全面实验，我们识别出三种主要的失败模式：（1）在视觉令牌限制下，由于下采样导致的信息丢失，（2）由于补丁边界移动和在非标准宽高比下位置编码脆弱性引起的令牌化伪影，以及（3）随着令牌数量增加而导致的注意力稀释。我们的分析表明，即使是最先进的模型在处理高分辨率图像时也会遭遇性能下降，且降级模式在不同的架构系列中系统性地变化。我们提供了可操作的见解，以指导模型架构设计和数据增强策略，从而减轻这些局限性。
cs.CV / 27 / 2609.31709
Cross-Dataset Generalization of Bangladeshi Rice Leaf Disease Classifiers: Benchmark, Diagnosis, and Mitigation
孟加拉国水稻叶病分类器的跨数据集泛化：基准、诊断与缓解
Abstract
Cross-dataset transfer in rice leaf disease classification remains a significant challenge, with models trained on one image collection performing substantially worse when deployed on another. We conduct a systematic benchmark across three Bangladeshi rice leaf disease datasets (5,419 images, 6 transfer pairs, 3 CNN backbones, 3 random seeds) to characterize and diagnose this failure. Strong augmentation recovers a mean cross-dataset macro-F1 improvement of +0.070 (Wilcoxon p < 0.001, 15 of 18 transfer pairs positive). Removing non-leaf image content via segmentation shows directional benefit (mean +0.066, p = 0.062, n = 36 paired observations) that is consistent across two independent segmentation methods but does not reach conventional significance. A self-supervised ViT control (DINOv2 linear probe) exhibits equivalent cross-dataset collapse to CNNs, ruling out architecture inductive bias as the primary driver and pointing to acquisition-condition shift. Adaptive batch normalization uniformly harms transfer performance, with harm magnitude correlating with source-target label-prior divergence and model depth (Spearman rho = 0.621, p = 0.009). Grad-CAM attribution analysis on 12 sampled predictions does not distinguish correct from incorrect cross-domain predictions (p = 0.462), indicating that common attribution proxies are insufficient for diagnosing shift at practical sample sizes. We document all frozen results, prespecified analysis criteria, and reproducibility artifacts in a public repository with SHA-256 integrity verification. This work establishes a rigorous empirical baseline for understanding cross-dataset generalization in agricultural computer vision and identifies both effective (augmentation) and ineffective (AdaBN) adaptation strategies.
Chinese Translation
水稻叶病分类中的跨数据集迁移仍然是一个重大挑战，模型在一个图像集合上训练后，在另一个数据集上的表现显著下降。我们在三个孟加拉国水稻叶病数据集（5,419张图像，6对迁移，3个CNN骨干网络，3个随机种子）上进行系统基准测试，以表征和诊断这一失败。强增强技术使得跨数据集的宏观F1均值提高了+0.070（Wilcoxon p < 0.001，18对迁移中有15对为正）。通过分割去除非叶片图像内容显示出方向性好处（均值+0.066，p = 0.062，n = 36对观察），这一结果在两种独立的分割方法中一致，但未达到传统显著性水平。自监督的ViT控制（DINOv2线性探针）表现出与CNN相同的跨数据集崩溃，排除了架构归纳偏差作为主要驱动因素，指出了获取条件的变化。自适应批量归一化普遍损害了迁移性能，损害程度与源目标标签先验的差异及模型深度相关（Spearman rho = 0.621，p = 0.009）。对12个采样预测的Grad-CAM归因分析未能区分正确与错误的跨领域预测（p = 0.462），表明常见的归因代理不足以在实际样本量下诊断变化。我们在一个公共库中记录了所有冻结结果、预先设定的分析标准和可重复性文献，并进行了SHA-256完整性验证。本研究为理解农业计算机视觉中的跨数据集泛化建立了严格的实证基准，并识别了有效（增强）和无效（AdaBN）适应策略。
cs.CV / 28 / 2609.31712
Statistical Testing for Multiple Instance Learning via Selective Inference with Applications to Computational Pathology
通过选择性推断进行多实例学习的统计检验及其在计算病理学中的应用
Abstract
Multiple instance learning (MIL) is widely used in computational pathology because it enables weakly supervised analysis of whole-slide images (WSIs) without requiring patch-level annotations. In attention-based MIL, instances with high attention scores are often interpreted as diagnostically important regions and used as visual explanations. However, attention scores alone cannot determine whether selected high-attention instances are significantly different from normal instances, limiting the reliability of attention-based explanations. In this paper, we formulate the evaluation of high-attention instances as a statistical hypothesis testing problem. Specifically, we assess whether a selected high-attention instance significantly deviates from a representative normal reference instance selected based on feature similarity. A major challenge is that both the target instance and the reference instance are selected through data-dependent procedures, rendering standard hypothesis testing invalid. To address this issue, we introduce a selective inference (SI) framework that explicitly accounts for the selection events induced by attention-based instance selection and adaptive reference selection, thereby enabling the computation of valid selective $p$-values conditional on these events. Experiments demonstrate Type-I error control on synthetic and MNIST-based data and practical applicability to pathological WSIs, with higher statistical power than the conventional over-conditioning approach.
Chinese Translation
多实例学习（MIL）在计算病理学中被广泛应用，因为它能够在不需要补丁级注释的情况下，对全切片图像（WSIs）进行弱监督分析。在基于注意力的多实例学习中，具有高注意力分数的实例通常被解释为具有诊断重要性的区域，并用作视觉解释。然而，仅凭注意力分数无法判断所选的高注意力实例是否与正常实例显著不同，这限制了基于注意力的解释的可靠性。本文将高注意力实例的评估形式化为统计假设检验问题。具体而言，我们评估所选的高注意力实例是否显著偏离基于特征相似性选择的代表性正常参考实例。一个主要挑战是，目标实例和参考实例都是通过依赖数据的程序选择的，这使得标准假设检验无效。为了解决这个问题，我们引入了一个选择性推断（SI）框架，该框架明确考虑了由基于注意力的实例选择和自适应参考选择引起的选择事件，从而使得能够在这些事件的条件下计算有效的选择性$p$-值。实验表明，在合成数据和基于MNIST的数据上控制了第一类错误，并在病理WSIs中具有实际应用性，其统计功效高于传统的过度条件化方法。
cs.CV / 29 / 2609.31713
Agentic Video Understanding: A Survey
主动视频理解：一项综述
Abstract
As large language models (LLMs) become capable of processing increasingly diverse modalities and longer temporal contexts, an emerging line of work is moving beyond fixed video-language inference toward agentic systems that actively decide what information to inspect, retain, verify, and act upon. This survey reviews video understanding agents: systems that use video as the primary information source and solve understanding tasks through adaptive state construction and action selection. We first formalize an agent loop for video understanding, then address a central question: why do agents matter for video understanding? To answer this, we organize the literature through a challenge-to-design taxonomy, linking context bottlenecks to hierarchical evidence memory, evidence sparsity to active evidence acquisition, temporal causality to state and process tracking, and multimodal ambiguity to role-specialized coordination. We further review state space paradigms, learning paradigms, supervision signals, benchmarks, and evaluation protocols. Finally, we identify open directions toward agentic-native temporal modeling and video-native agents. Project page: https://github.com/DXY0711/Awesome-Agentic-Video-Understanding
Chinese Translation
随着大型语言模型（LLMs）能够处理越来越多样化的模态和更长的时间上下文，一项新兴的研究方向正在超越固定的视频-语言推理，朝着主动系统发展，这些系统能够主动决定检查、保留、验证和采取行动的信息。本文综述了视频理解代理：这些系统以视频作为主要信息源，通过自适应状态构建和行动选择来解决理解任务。我们首先为视频理解形式化一个代理循环，然后探讨一个核心问题：代理在视频理解中为何重要？为了解答这一问题，我们通过挑战-设计分类法组织文献，将上下文瓶颈与层次证据记忆关联，将证据稀疏性与主动证据获取关联，将时间因果关系与状态和过程跟踪关联，将多模态歧义与角色专门协调关联。我们进一步回顾了状态空间范式、学习范式、监督信号、基准测试和评估协议。最后，我们确定了朝着主动原生时间建模和视频原生代理的开放方向。项目页面：https://github.com/DXY0711/Awesome-Agentic-Video-Understanding
cs.CV / 30 / 2609.31714
OmniFysics-Captioner Technical Report: Grounding Omni-Modal Understanding in the Physical World for Better Captioning
OmniFysics-Captioner 技术报告：在物理世界中扎根的全模态理解以实现更好的字幕生成
Abstract
Building omni-modal models with physical intelligence requires fine-grained supervision that captures physical evidence such as contact, support, deformation, and state transitions. However, existing omni-modal captioners primarily model general audiovisual semantics and often overlook transient or spatially localized physical evidence. We present a unified framework for physics-aware audiovisual captioning spanning data construction, training, and evaluation. Firstly, we build a data construction pipeline that identifies physics-rich clips and leverages OmniFysics-Agent to coordinate audio, visual, and physical-perception tools for collecting spatiotemporally aligned and traceable cross-modal evidence; within the Agent, a physical perception model (PPM) fine-tuned on approximately 2M image-level samples serves as a dedicated tool for extracting object-interaction and state-change cues. Secondly, we build the Daily-Physics 50K dataset and introduce the evidence-driven OmniPhysCap (OPC) benchmark to evaluate the recovery of physical and cross-modal evidence from generated captions. Finally, we train OmniFysics-Captioner from the resulting data. Our Captioner matches Gemini 3.1 Pro on audiovisual captioning, achieves state-of-the-art results on multiple video-captioning benchmarks, and substantially outperforms other open-source models. Ablations show that PPM evidence improves physical coverage and produces finer-grained, more reliable cross-modal descriptions.
Chinese Translation
构建具有物理智能的全模态模型需要细致的监督，以捕捉接触、支撑、变形和状态转变等物理证据。然而，现有的全模态字幕生成器主要建模一般的视听语义，往往忽视了瞬态或空间局部的物理证据。我们提出了一个统一的框架，用于物理感知的视听字幕生成，涵盖数据构建、训练和评估。首先，我们建立了一个数据构建管道，识别物理丰富的片段，并利用 OmniFysics-Agent 协调音频、视觉和物理感知工具，以收集时空对齐且可追踪的跨模态证据；在该代理中，经过大约 200 万图像级样本微调的物理感知模型（PPM）作为专用工具，用于提取物体交互和状态变化线索。其次，我们构建了 Daily-Physics 50K 数据集，并引入证据驱动的 OmniPhysCap (OPC) 基准，以评估从生成的字幕中恢复物理和跨模态证据的能力。最后，我们从生成的数据中训练 OmniFysics-Captioner。我们的字幕生成器在视听字幕生成上与 Gemini 3.1 Pro 相匹配，在多个视频字幕生成基准上取得了最先进的结果，并显著超越其他开源模型。消融实验表明，PPM 证据提高了物理覆盖率，并生成了更细致、更可靠的跨模态描述。
cs.CV / 31 / 2609.31715
Fysiverse-3D-SimReady Technical Report: Agentic Physical Simulation for Pragmatic 3D World Reconstruction
Fysiverse-3D-SimReady技术报告：用于务实3D世界重建的代理物理仿真
Abstract
Agentic recognition requires visual perception to move beyond static scene understanding and produce structured scene representations that support the perception--reasoning--action loop. Existing single-image 3D generation methods, however, mainly produce visually plausible object assets rather than simulation-ready scene states. When independently generated meshes are composed in a shared space, they may fail to align with the input camera, violate gravity, interpenetrate nearby objects, or become unstable under physics simulation. We present Fysiverse-3D-SimReady, a grounded refinement framework for reconstructing simulation-ready multi-object scenes from a single RGB image with instance and ground prompts. The method places generated object meshes into a shared gravity-aligned scene, refines their camera-space poses through differentiable rendering, and corrects scene-level supports and contacts for stable physical execution. The scene is then used by an agentic simulation workflow, which converts a scene-specific task goal into an executable physics rollout rendered from the original camera view. Experiments show that Fysiverse-3D-SimReady improves input-view alignment, contact plausibility, and physical stability over existing single-image reconstruction and scene generation baselines, while enabling goal-conditioned physical interactions from a single image.
Chinese Translation
代理识别需要视觉感知超越静态场景理解，并生成支持感知-推理-行动循环的结构化场景表示。然而，现有的单图像3D生成方法主要生成视觉上可信的物体资产，而非适合仿真的场景状态。当独立生成的网格在共享空间中组合时，它们可能无法与输入相机对齐，违反重力，穿透附近物体，或在物理仿真下变得不稳定。我们提出了Fysiverse-3D-SimReady，这是一个基于实例和地面提示的框架，用于从单个RGB图像重建适合仿真的多物体场景。该方法将生成的物体网格放置在一个共享的重力对齐场景中，通过可微渲染细化其相机空间姿态，并纠正场景级的支撑和接触，以实现稳定的物理执行。然后，该场景被用于代理仿真工作流程，该工作流程将场景特定的任务目标转换为从原始相机视角渲染的可执行物理展开。实验表明，Fysiverse-3D-SimReady在输入视图对齐、接触可信度和物理稳定性方面优于现有的单图像重建和场景生成基准，同时实现了从单个图像出发的目标条件物理交互。
cs.CV / 32 / 2609.31716
PanOVOcc: Panoramic Embodied Open-Vocabulary Occupancy Mapping with Long-term Spatial Voxel Memory
PanOVOcc：具有长期空间体素记忆的全景具身开放词汇占用映射
Abstract
Persistent semantic occupancy mapping is essential for embodied scene understanding. However, perspective-based systems provide limited spatial coverage, while existing panoramic methods primarily predict local volumes from single observations. We introduce PanOVOcc, a training-free framework for persistent open-vocabulary semantic occupancy mapping from panoramic sequences. PanOVOcc unifies panoramic SLAM, open-vocabulary perception, and long-term spatial voxel memory within an online architecture, continuously integrating geometric and semantic evidence into a global, language-queryable map. To facilitate systematic evaluation of this setting, we establish Pan-Replica and Pan-Holo360D, two benchmarks pairing continuous panoramic RGB-D sequences with scene-level semantic occupancy ground truth across synthetic and real-world scenes. Compared with the strongest evaluated baseline for each metric, PanOVOcc improves occupancy IoU and semantic mIoU by absolute +20.03 and +7.06 on Pan-Replica, and by +43.26 and +20.16 on Pan-Holo360D, respectively. The source code and the established benchmarks will be available at https://github.com/bakereet/PanOVOcc.
Chinese Translation
持久的语义占用映射对于具身场景理解至关重要。然而，基于视角的系统提供的空间覆盖有限，而现有的全景方法主要从单一观测中预测局部体积。我们提出了PanOVOcc，这是一个无需训练的框架，用于从全景序列中进行持久的开放词汇语义占用映射。PanOVOcc在在线架构中统一了全景SLAM、开放词汇感知和长期空间体素记忆，持续将几何和语义证据整合到一个全球可查询的语言地图中。为了便于对该设置进行系统评估，我们建立了Pan-Replica和Pan-Holo360D两个基准，将连续的全景RGB-D序列与合成和真实场景中的场景级语义占用真实值配对。与每个指标评估的最强基线相比，PanOVOcc在Pan-Replica上将占用IoU和语义mIoU分别提高了绝对+20.03和+7.06，在Pan-Holo360D上分别提高了+43.26和+20.16。源代码和建立的基准将可在https://github.com/bakereet/PanOVOcc获取。
cs.CV / 33 / 2609.31717
PanoFuse: Panorama-Enhanced Vision-Language-Action Learning with Decoupled Semantic-Geometric Routing
PanoFuse：基于全景增强的视觉-语言-动作学习与解耦语义-几何路由
Abstract
Vision-Language-Action (VLA) policies have shown promising performance in language-conditioned robotic manipulation. However, most existing VLA systems rely on conventional perspective cameras with limited fields of view, often missing global scene context and leading to unreliable manipulation under visual occlusions, distractors, and unseen environments. In this work, we propose PanoFuse, a panorama-enhanced VLA framework that complements local manipulation observations with global panoramic perception. PanoFuse introduces a dedicated panoramic branch that leverages a pretrained panoramic foundation model to extract complementary semantic and geometric representations from omnidirectional observations. Rather than directly mixing these heterogeneous features, we introduce Decoupled Semantic-Geometric Routing (DSGR), which maintains semantic and geometric representations as separate context streams and selectively routes both to downstream state and action representations through structured block-wise attention. This design provides the action expert with global spatial context while preserving task-relevant semantic information from the pretrained VLA backbone. We further develop a synchronized data collection pipeline and construct a new real-world manipulation dataset containing panoramic RGB observations, wrist-view images, language instructions, robot states, and actions. Across seven evaluation settings, PanoFuse achieves an average success rate of 52.9%, outperforming the evaluated baselines and achieving consistent gains under novel-object, unseen-background, and distractor-rich settings. Code and data will be released publicly at https://xux-hnu.github.io/PanoFuse.
Chinese Translation
视觉-语言-动作（VLA）策略在语言条件下的机器人操作中表现出良好的性能。然而，现有的大多数VLA系统依赖于传统的透视相机，其视野有限，常常缺乏全球场景上下文，导致在视觉遮挡、干扰物和未知环境下的操作不可靠。在本研究中，我们提出了PanoFuse，一个基于全景增强的VLA框架，它通过全球全景感知来补充局部操作观察。PanoFuse引入了一个专门的全景分支，利用预训练的全景基础模型从全向观察中提取互补的语义和几何表示。我们并不直接混合这些异构特征，而是引入了解耦语义-几何路由（Decoupled Semantic-Geometric Routing, DSGR），该方法将语义和几何表示作为独立的上下文流进行维护，并通过结构化的块级注意力选择性地将两者路由到下游状态和动作表示。这一设计为动作专家提供了全球空间上下文，同时保留了来自预训练VLA主干的任务相关语义信息。我们进一步开发了一个同步数据收集管道，并构建了一个新的真实世界操作数据集，其中包含全景RGB观察、手腕视图图像、语言指令、机器人状态和动作。在七个评估设置中，PanoFuse的平均成功率达到52.9%，超越了评估的基线，并在新物体、未知背景和干扰物丰富的设置下取得了一致的提升。代码和数据将公开发布于https://xux-hnu.github.io/PanoFuse。
cs.CV / 34 / 2609.31720
LatentReRig: An SDF-Based VAE with Dual Decoders for Latent-Space Deformation Conditioning
LatentReRig：一种基于SDF的变分自编码器，具有双解码器用于潜在空间变形条件
Abstract
Transferring deformation between characters with different geometry and topology is challenging because conventional rigs encode behaviour through character-specific structures and correspondences. We present LatentReRig, an experimental framework that investigates whether pose-associated changes can instead be represented as reusable directions in a learned geometric latent space. An SDF-based variational autoencoder is coupled with two decoders: one reconstructs the implicit field, while the other predicts target vertex positions from source geometry and latent deformation conditioning. The source geometry may be neutral or already deformed. Experiments on a controlled humanoid dataset show that several poses induce coherent latent directions across identities, particularly for broad articulated motions. These signals can guide deformation of unseen characters, but explicit predictions remain less accurate for localized changes and corrective contributions. Diagnostic comparisons with repeated SDF sampling show that inter-identity distances exceed same-geometry resampling variability on average, while pose signals exhibit different margins above this baseline. The results support the presence of reusable pose-related structure and identify stable local conditioning and accurate mesh decoding as complementary requirements for improving transfer.
Chinese Translation
在具有不同几何形状和拓扑结构的角色之间转移变形是具有挑战性的，因为传统的绑定通过角色特定的结构和对应关系来编码行为。我们提出了LatentReRig，这是一个实验框架，旨在研究姿态相关的变化是否可以表示为学习的几何潜在空间中的可重用方向。该框架结合了一个基于SDF（Signed Distance Function）的变分自编码器和两个解码器：一个重建隐式场，另一个根据源几何形状和潜在变形条件预测目标顶点位置。源几何形状可以是中性的或已经变形的。在一个受控的人形数据集上的实验表明，多个姿态在不同身份之间引发了一致的潜在方向，特别是在广泛的关节运动中。这些信号可以指导未见角色的变形，但对于局部变化和修正贡献的明确预测仍然不够准确。与重复的SDF采样进行的诊断比较显示，身份间的距离平均超过同几何重采样的变异性，而姿态信号在此基线之上表现出不同的边际。结果支持可重用的姿态相关结构的存在，并确定稳定的局部条件和准确的网格解码是改善转移的互补要求。
cs.CV / 35 / 2609.31722
Where Does the Watermark Hide? Push-Pull Disentanglement for Invisible Watermark Removal
水印隐藏在哪里？用于隐形水印去除的推拉解耦方法
Abstract
Fixed image distortions do not cover an attacker that learns from paired clean and watermarked images. We study this paired-training threat with single-image inference: deployment uses neither the clean reference nor the watermark key, payload, or decoder. An encoder maps each image to a structural latent $g$ and an auxiliary residual latent $u$. Push supervision reconstructs the watermarked image from $D(g_w,u_w)$. Pull supervision trains the zero-auxiliary output $D(A_g(g_w;k),0)$ toward the paired clean image. At $k=1.10,u=0$, the four-method sweep gives an average BER of $0.3958$, PSNR of $31.07$ dB, and SSIM of $0.9554$. Restoring $u$ from $0$ to $0.15$ moves average BER from $0.3893$ to $0.3357$, while PSNR falls from $30.99$ to $28.23$ dB. The intervention supports decoder dependence on the auxiliary input in the evaluated setting. The accompanying theory is a conditional, post-hoc account of this behavior rather than an experimentally verified information-relocation result.
Chinese Translation
固定的图像失真无法掩盖从配对的干净图像和水印图像中学习的攻击者。我们研究这种配对训练威胁下的单图像推断：部署过程中不使用干净参考图像、水印密钥、负载或解码器。编码器将每个图像映射到结构潜变量 $g$ 和辅助残差潜变量 $u$。推送监督从 $D(g_w,u_w)$ 重建水印图像。拉取监督则将零辅助输出 $D(A_g(g_w;k),0)$ 训练朝向配对的干净图像。在 $k=1.10,u=0$ 时，四种方法的平均比特错误率（BER）为 $0.3958$，峰值信噪比（PSNR）为 $31.07$ dB，结构相似性指数（SSIM）为 $0.9554$。将 $u$ 从 $0$ 恢复到 $0.15$ 时，平均 BER 从 $0.3893$ 降至 $0.3357$，而 PSNR 从 $30.99$ 降至 $28.23$ dB。该干预支持解码器在评估设置中对辅助输入的依赖。随附的理论是对这种行为的条件后验解释，而不是经过实验验证的信息重定位结果。
cs.CV / 36 / 2609.31723
Frequency-Domain AI-Generated Image Detection: Exploring Decoder and Channel Attention for Feature Refinement
频域AI生成图像检测：探索解码器和通道注意力进行特征细化
Abstract
With the rapid progress of AI, the number of AI-generated images has increased significantly in recent years. However, the increasing variety of image generation models makes detection more difficult. In this work, we use Fast Fourier Transform (FFT) representation with EfficientNet-B0 for AI-generated image detection. EfficientNet-B0 provides a lightweight architecture that can be useful for resource-limited applications. Most frequency-domain detectors use a standard encoder to extract features from the FFT spectrum and directly pass them to a classifier. We explored a different approach by investigating ECA, U-Net, and Attention U-Net as alternatives to this direct encoder-to-classifier approach. ECA applies channel attention, while U-Net and Attention U-Net use decoder-based architectures to recover and refine spatial information in the extracted frequency features. We used a balanced subset of the MS COCOAI dataset that includes AI-generated images from five different models. Three runs were carried out for each experiment, and the average values were recorded. Experimental results indicate that EfficientNet-B0 obtained an accuracy of 84.64%, which is 4.50 percentage points higher than the ResNet-50 baseline reported in the dataset paper. EfficientNet-B0 with U-Net provided a small improvement, achieving an accuracy of 84.85%, while ECA did not increase the overall performance. EfficientNet-B0 with Attention U-Net achieved the best overall performance, with an accuracy of 85.51% and an ROC-AUC of 92.99%. This represents an improvement of 0.87 percentage points in accuracy compared to the EfficientNet-B0 baseline and 5.37 percentage points over the ResNet-50 baseline reported in the dataset paper.
Chinese Translation
随着人工智能的快速发展，近年来AI生成图像的数量显著增加。然而，图像生成模型的多样性使得检测变得更加困难。在本研究中，我们使用快速傅里叶变换（Fast Fourier Transform, FFT）表示法结合EfficientNet-B0进行AI生成图像的检测。EfficientNet-B0提供了一种轻量级架构，适用于资源有限的应用。大多数频域检测器使用标准编码器从FFT频谱中提取特征，并直接将其传递给分类器。我们探索了一种不同的方法，研究了ECA、U-Net和Attention U-Net作为替代直接编码器到分类器的方法。ECA应用了通道注意力，而U-Net和Attention U-Net则使用基于解码器的架构来恢复和细化提取的频率特征中的空间信息。我们使用了MS COCOAI数据集中一个平衡的子集，其中包含来自五种不同模型的AI生成图像。每个实验进行了三次运行，并记录了平均值。实验结果表明，EfficientNet-B0的准确率为84.64%，比数据集论文中报告的ResNet-50基线高出4.50个百分点。EfficientNet-B0与U-Net的组合提供了小幅改进，达到了84.85%的准确率，而ECA未能提高整体性能。EfficientNet-B0与Attention U-Net的组合实现了最佳整体性能，准确率为85.51%，ROC-AUC为92.99%。与EfficientNet-B0基线相比，准确率提高了0.87个百分点，较数据集论文中报告的ResNet-50基线提高了5.37个百分点。
cs.CV / 37 / 2609.31725
SWT: Self-Supervised Video Object Segmentation via Sliding, Wavelet and Transportation
SWT：通过滑动窗口、小波变换和最优运输实现自监督视频目标分割
Abstract
Video Object Segmentation (VOS) aims to accurately segment target objects from consecutive video frames and track the changes of the objects in each frame of the video. Conventional VOS methods typically demand substantial quantities of pixel-level labeled video sequences for fully supervised learning, which limits the performance of the model in sparse video scenes, while existing VOS methods have limited adaptability to global changes in objects. Based on this observation, in this paper, we propose self-supervised VOS with Sliding window, Wavelet transform and optimal Transport (SWT), a self-supervised VOS framework entirely trained on static dataset using contrastive learning. Firstly, a rolling sample buffer reuses overlapping groups of independently sampled images across successive updates. Secondly, to address the long-distance modeling difficulty caused by simple convolutional structures, we introduce wavelet transform to expand the receptive field of convolutional kernels, thus improving the model's representational capability. Finally, we incorporate optimal transport to assist the model in finding the globally optimal match between the target across two frames, improving the model's ability to handle nonrigid deformations of objects. SWT only requires training on the COCO dataset once and achieves excellent results on five VOS datasets as well as an additional body part propagation dataset. The code will be released soon at [https://github.com/machine928/SWT.git](https://github.com/machine928/SWT.git).
Chinese Translation
视频目标分割（VOS）旨在从连续的视频帧中准确分割目标对象，并跟踪视频中每一帧对象的变化。传统的VOS方法通常需要大量的像素级标注视频序列进行完全监督学习，这限制了模型在稀疏视频场景中的性能，而现有的VOS方法对对象的全局变化适应性有限。基于这一观察，本文提出了一种基于滑动窗口、小波变换和最优运输（SWT）的自监督VOS框架，该框架完全在静态数据集上使用对比学习进行训练。首先，滚动样本缓冲区在连续更新中重用独立采样图像的重叠组。其次，为了解决简单卷积结构造成的长距离建模困难，我们引入小波变换以扩展卷积核的感受野，从而提高模型的表征能力。最后，我们结合最优运输，帮助模型在两帧之间找到目标的全局最优匹配，提高模型处理对象非刚性变形的能力。SWT仅需在COCO数据集上训练一次，即可在五个VOS数据集以及一个额外的身体部位传播数据集上取得优异的结果。代码将很快发布在 [https://github.com/machine928/SWT.git](https://github.com/machine928/SWT.git)。
cs.CV / 38 / 2609.31726
High-Capacity Robust Medical Image Exfiltration via Neural Network Weight Replacement
通过神经网络权重替换实现高容量稳健的医学图像外泄
Abstract
Collaborative medical AI platforms allow researchers to train models on sensitive imaging data while restricting data export. However, trained models can serve as covert carriers of patient information: medical images may be encoded within model parameters and reconstructed outside the secure environment. Existing defenses rely on lightweight sanitization (e.g., fine-tuning, pruning, quantization) and limited statistical auditing, creating a realistic insider exfiltration risk. We introduce a high-capacity neural steganography attack that encodes medical images as continuous latent representations embedded into model initialization. A StyleGAN2-based adversarial autoencoder learns compact latent codes regularized to match standard weight initialization statistics, keeping embedded parameters statistically consistent with clean models. Noise injection during training improves robustness to export-time mitigation. The carrier model remains functional on its intended task and hidden images can be reconstructed directly from its weights after export. This continuous encoding enables robust and scalable exfiltration, allowing up to 99 brain MRI volumes to be embedded within a 30MB model, and remains recoverable under mitigations that disrupt prior bit-level schemes. While reconstructions are approximate rather than pixel-exact, embedded content remains anatomically recognizable and recoverable at scale, exposing a privacy risk distinct from prior bit-level approaches. Experiments on MIMIC-CXR, BraTS, and LiTS demonstrate effectiveness across modalities, tasks, and architectures, highlighting the need for structural defenses beyond parameter-level sanitization. Code is available at https://github.com/ElieThellier/high-capacity-robust-medical-image-exfiltration.
Chinese Translation
协作医学人工智能平台允许研究人员在限制数据导出的情况下对敏感影像数据进行模型训练。然而，训练后的模型可能成为患者信息的隐秘载体：医学图像可能被编码在模型参数中，并在安全环境之外被重构。现有的防御措施依赖于轻量级的清洗（例如，微调、剪枝、量化）和有限的统计审计，造成了现实的内部外泄风险。我们提出了一种高容量的神经隐写攻击，将医学图像编码为嵌入模型初始化的连续潜在表示。基于StyleGAN2的对抗自编码器学习紧凑的潜在编码，并通过正则化使其符合标准权重初始化统计，保持嵌入参数在统计上与干净模型一致。训练期间的噪声注入提高了对导出时缓解措施的稳健性。载体模型在其预期任务上仍然功能正常，隐藏图像可以在导出后直接从其权重中重构。这种连续编码实现了稳健且可扩展的外泄，允许在一个30MB的模型中嵌入多达99个脑部MRI体积，并在破坏先前比特级方案的缓解措施下仍然可恢复。尽管重构是近似的而非逐像素精确的，但嵌入内容在解剖上仍然可识别，并且在规模上可恢复，暴露出与先前比特级方法不同的隐私风险。在MIMIC-CXR、BraTS和LiTS上的实验展示了在不同模态、任务和架构中的有效性，突显了超越参数级清洗的结构性防御的必要性。代码可在 https://github.com/ElieThellier/high-capacity-robust-medical-image-exfiltration 获取。
cs.CV / 39 / 2609.31731
GERIS: A Game-Theoretic Framework for Filtering Instance-Dependent Label Noise in License Plate Data Augmentation
GERIS：一种基于博弈论的框架，用于过滤车牌数据增强中的实例依赖标签噪声
Abstract
In this paper, we propose GERIS, a game-theoretic framework for instance selection in the data augmentation phase of license plate recognition systems. During augmentation, synthetic license plate images are generated and transformed using stochastic noise to simulate real-world conditions. However, certain noise configurations lead to highly distorted, unreadable images that degrade model performance by introducing instance-dependent label noise. GERIS formulates a non-cooperative game in which each noise vector competes for inclusion in the training set based on its similarity to labeled data and its contribution to model reliability. By identifying and pruning low-quality instances, GERIS improves the overall quality of the augmented dataset. Unlike traditional black-box learning methods, GERIS offers a transparent, theoretically grounded mechanism for data filtering. Experimental results demonstrate that GERIS outperforms existing instance selection methods in terms of classification accuracy and robustness.
Chinese Translation
本文提出了GERIS，一种用于车牌识别系统数据增强阶段的实例选择的博弈论框架。在增强过程中，生成合成车牌图像并使用随机噪声进行变换，以模拟现实世界条件。然而，某些噪声配置会导致高度扭曲、不可读的图像，从而通过引入实例依赖的标签噪声来降低模型性能。GERIS构建了一个非合作博弈，其中每个噪声向量根据其与标记数据的相似性及其对模型可靠性的贡献竞争进入训练集。通过识别和剔除低质量实例，GERIS提高了增强数据集的整体质量。与传统的黑箱学习方法不同，GERIS提供了一种透明的、理论基础扎实的数据过滤机制。实验结果表明，GERIS在分类准确性和鲁棒性方面优于现有的实例选择方法。
cs.CV / 40 / 2609.31733
When Retrieval Hurts: Measuring and Explaining Retrieval-Induced Hallucination in Chest X-ray Report Generation
当检索带来伤害：测量和解释胸部X光报告生成中的检索诱导幻觉
Abstract
Retrieval-augmented generation is an attractive way to improve chest X-ray reporting, because reports from similar prior studies supply clinical context a general vision-language model lacks. We show the same mechanism is a reliable source of clinical error. Over 100 MIMIC-CXR studies with a frozen LLaVA-1.5 generator and BioMedCLIP retrieval, CheXbert clinical F1 doubles under relevant retrieval (0.201 to 0.402) and collapses to 0.043 under clinically mismatched retrieval, a fifth of the image-only score; the retrieval-induced hallucination rate, counting only unsupported findings traceable to retrieved evidence, rises from 0.00 to 0.76 and 0.98. To show this is not an artefact of evidence-set coverage, we introduce a coincidental-overlap control that scores image-only generations against evidence they never saw, placing the chance base rate at 0.18, four to five times below the observed rates. The generator does not merely acquire findings, it transcribes text: 95% of reports produced under relevant retrieval contain an eight-word span occurring verbatim in the retrieved evidence but absent from the reference, against 0% without retrieval. We then explain the mechanism: a normal chest X-ray retrieves at least one abnormal precedent in 24 of 28 cases, because medical image-embedding similarity is dominated by anatomy and acquisition rather than by the presence of disease. This has a direct design consequence. Retrieval similarity does not predict harm (RIH rates 0.80/0.80/0.60/0.84 across similarity quartiles), so relevance gates conditioned on embedding similarity cannot work; gating on predicted pathology agreement removes 61% of unsupported evidence at no cost to useful coverage. We argue that retrieval-augmented clinical systems must be evaluated under retrieval failure, not only under retrieval success.
Chinese Translation
检索增强生成是一种改善胸部X光报告的有效方法，因为来自类似先前研究的报告提供了通用视觉-语言模型所缺乏的临床背景。我们表明，相同的机制也是临床错误的可靠来源。在超过100个MIMIC-CXR研究中，使用冻结的LLaVA-1.5生成器和BioMedCLIP检索，CheXbert临床F1在相关检索下翻倍（从0.201到0.402），而在临床不匹配的检索下崩溃至0.043，仅为图像评分的五分之一；检索诱导幻觉率（仅计算可追溯至检索证据的无支持发现）从0.00上升至0.76和0.98。为了证明这不是证据集覆盖的伪影，我们引入了一个偶然重叠控制，评估图像生成与它们从未见过的证据的匹配度，将机会基线率设定为0.18，远低于观察到的比率的四到五倍。生成器不仅仅是获取发现，它还转录文本：在相关检索下生成的报告中，95%的报告包含在检索证据中逐字出现的八个单词的片段，而在没有检索的情况下则为0%。我们接着解释了这一机制：正常的胸部X光在28个案例中的24个案例中检索到至少一个异常先例，因为医学图像嵌入的相似性主要受解剖结构和获取方式的主导，而不是疾病的存在。这具有直接的设计后果。检索相似性并不能预测伤害（RIH率在相似性四分位数中为0.80/0.80/0.60/0.84），因此基于嵌入相似性的相关性门控无法发挥作用；基于预测病理一致性的门控在没有损失有用覆盖的情况下去除了61%的无支持证据。我们认为，检索增强的临床系统必须在检索失败的情况下进行评估，而不仅仅是在检索成功的情况下。
cs.CV / 41 / 2609.31734
Measuring the evolution of camera distance across a century of film
跨越一个世纪的电影中相机距离演变的测量
Abstract
The rise of computer vision and artificial intelligence has made possible new forms of large-scale computational measurement. We apply these techniques to a deep collection of 5,205 digitized films viewed in theaters between 1922-2025 (covering popular, prestigious, and independent movies) to trace the development of one of the most fundamental ways through which film communicates: by manipulating the space between the camera and its subject. This work finds abrupt changes with the rise of new technologies in sound and television, and allows us to shed an empirical light on gender disparity (women, despite having substantially less screentime than men, are disproportionately the subject of closer shots), and illustrate how animated films both inherit and break free from the norms of live-action filmmaking.
Chinese Translation
计算机视觉和人工智能的兴起使得大规模计算测量的新形式成为可能。我们将这些技术应用于1922年至2025年间在影院观看的5205部数字化电影的深度收藏（涵盖流行、知名和独立电影），以追踪电影交流的最基本方式之一的发展：通过操控相机与其拍摄对象之间的空间。研究发现，随着新技术（如声音和电视）的出现，电影中相机距离的变化出现了突变，并使我们能够对性别差异进行实证分析（尽管女性的镜头时间明显少于男性，但她们却不成比例地成为特写镜头的对象），同时展示了动画电影如何继承并突破实景拍摄的规范。
cs.CV / 42 / 2609.31736
LukeNet: A lightweight CNN integrated with an XAI model for Smart acute lymphoblastic leukemia detection and management
LukeNet：一种与可解释人工智能模型集成的轻量级卷积神经网络，用于智能急性淋巴细胞白血病的检测与管理
Abstract
Acute Lymphoblastic Leukemia (ALL) patients require early, accurate detection to enable timely treatment and effective patient management. A Convolutional Neural Network (CNN) is well-suited for creating an end-to-end enabling environment for ALL detection and classification. However, most CNN-based ALL detection systems are theoretical and unsuitable for deployment on edge devices due to high computational demands. The Internet of Medical Things (IoMT)-enabled devices offer an opportunity to monitor ALL patients in real time. Wearables that track temperature, heart rate, oxygen saturation, and activity can deliver critical data to support timely clinical intervention and improve patient outcomes. In smart IoMT environments, a lightweight CNN is essential because connected devices often operate under limited computational power, memory, and latency constraints. To address this need, this study proposes LukeNet, a lightweight CNN integrated with explainable artificial intelligence (XAI) for an IoMT-based SMART Acute Lymphoblastic Leukemia Detection and Management System. Trained on three (3) ALL datasets and five-fold cross-validation, LukeNet achieved an impressive 99% model accuracy as well as 99% unseen test accuracy, which is higher than six state-of-the-art (SOTA) CNNs, such as DenseNet121, MobileNet, ResNet50, InceptionV3, Xception, and VGG16, as well as transfer learning models. Furthermore, LukeNet was compared with two ensemble models. In addition, explainable AI methods are integrated to highlight relevant regions in microscopic images. The novelty of this study lies in the architecture of LukeNet, which balances model depth and computational efficiency by using depthwise separable convolutions, mitigates the risk of gradient loss in deeper layers, and provides strong global and local feature extraction capabilities.
Chinese Translation
急性淋巴细胞白血病（ALL）患者需要尽早、准确的检测，以便及时治疗和有效管理患者。卷积神经网络（CNN）非常适合为ALL的检测和分类创建端到端的支持环境。然而，大多数基于CNN的ALL检测系统都是理论性的，因其高计算需求而不适合在边缘设备上部署。医疗物联网（IoMT）设备提供了实时监测ALL患者的机会。能够跟踪温度、心率、血氧饱和度和活动的可穿戴设备可以提供关键数据，以支持及时的临床干预并改善患者结果。在智能IoMT环境中，轻量级CNN是必不可少的，因为连接设备通常在有限的计算能力、内存和延迟约束下运行。为满足这一需求，本研究提出了LukeNet，一种与可解释人工智能（XAI）集成的轻量级CNN，用于基于IoMT的智能急性淋巴细胞白血病检测与管理系统。在三个（3）ALL数据集上进行训练并采用五折交叉验证，LukeNet达到了令人印象深刻的99%的模型准确率以及99%的未见测试准确率，超过了六种最先进（SOTA）的CNN，如DenseNet121、MobileNet、ResNet50、InceptionV3、Xception和VGG16，以及迁移学习模型。此外，LukeNet还与两个集成模型进行了比较。此外，集成了可解释AI方法，以突出显微图像中的相关区域。本研究的创新之处在于LukeNet的架构，通过使用深度可分离卷积平衡模型深度和计算效率，减轻了深层中的梯度消失风险，并提供了强大的全局和局部特征提取能力。
cs.CV / 43 / 2609.31737
Seeing the Heat: Synthesizing High-Resolution Wood Thermal Responses from Optical Imagery
观察热量：从光学图像合成高分辨率木材热响应
Abstract
The thermal behavior of wood is a critical factor in advanced material assembly. However, pixel-level thermal analysis remains fundamentally constrained by the low resolution and noise inherent to infrared thermography. To address this, we introduce an end-to-end computational framework that synthesizes high-resolution thermal responses directly from wood RGB images. We first establish a core physical linkage: because spatial color variation in natural wood is driven by cellular anatomy, optical intensity serves as a reliable geometric proxy for the localized solid volume fraction. By leveraging this theoretical insight, we develop an automated finite-element-method data engine that maps pixel-level optical intensity to a 3D thermodynamic voxel grid, generating high-fidelity synthetic thermal responses. We find that 1) when the thermal conductivity along the thickness direction is uniform or linear, wood RGB images and their corresponding thermal responses exhibit extreme morphological similarities, and the lateral thermal diffusion acts as a low-pass filter that smooths out high-frequency details; 2) when the thermal conductivity along the thickness direction is random, such morphological similarities are destroyed, and wood's 3D structure dominantly governs its thermal response. We further utilize these synthetic thermal responses to supervise a neural surrogate model built upon the DINOv3 foundation model. Our results demonstrate that the neural surrogate model successfully internalizes the governing thermodynamic laws, thereby bypassing computationally expensive simulations and enabling high-resolution thermal inference. This methodology effectively bridges the semantic and thermodynamic domains, unlocking systematic, pixel-level analysis of fine-grained wood thermal responses. Project: https://zekifayes.github.io/seeheat
Chinese Translation
木材的热行为是先进材料组装中的一个关键因素。然而，像素级的热分析仍然受到红外热成像固有的低分辨率和噪声的限制。为了解决这个问题，我们提出了一种端到端的计算框架，能够直接从木材的RGB图像合成高分辨率的热响应。我们首先建立了一个核心物理联系：由于自然木材中的空间颜色变化是由细胞结构驱动的，光学强度可以作为局部固体体积分数的可靠几何代理。通过利用这一理论见解，我们开发了一个自动化的有限元方法数据引擎，将像素级的光学强度映射到3D热力学体素网格，从而生成高保真的合成热响应。我们的研究发现：1）当厚度方向的热导率均匀或线性时，木材的RGB图像及其对应的热响应表现出极端的形态相似性，横向热扩散充当低通滤波器，平滑掉高频细节；2）当厚度方向的热导率随机时，这种形态相似性被破坏，木材的3D结构主导其热响应。我们进一步利用这些合成热响应来监督基于DINOv3基础模型构建的神经代理模型。我们的结果表明，神经代理模型成功地内化了主导的热力学规律，从而绕过了计算成本高昂的模拟，实现了高分辨率的热推断。这一方法有效地架起了语义与热力学领域之间的桥梁，解锁了对细粒度木材热响应的系统性像素级分析。项目链接：https://zekifayes.github.io/seeheat
cs.CV / 44 / 2609.31740
Beyond Volume Overlap: Surface Matching for Topology-Aware Coronary Artery Segmentation
超越体积重叠：面匹配用于拓扑感知的冠状动脉分割
Abstract
Accurate coronary artery segmentation on coronary computed tomography angiography (CCTA) is essential for diagnosing coronary artery disease. Deep networks are conventionally trained and evaluated with the Dice coefficient, but volume-overlap metrics are poorly suited to thin, tubular anatomy: since most voxels belong to a few thickproximal segments, a missing distal branch barely affects Dice despite severely disrupting the connectivity required for clinical use. We introduce a surface metric that matches predicted and reference surface points via bipartite assignment under a localized, vessel-radius tolerance, reporting precision, recall, and F1 with decoupled false positives (spurious branches) and false negatives (missed branches) a distinction the symmetric Dice cannot make. With it we show that a strong Dice-trained baseline omits far more vessel surface than it hallucinates, an asymmetry its high Dice hides. Building on this, we propose a differentiable surface loss that simultaneously suppresses spurious mass and recovers absent structure, validated by fine-tuning three backbones (nnU-Net, SwinUNETR, NexToU) on two public benchmarks (Image-CAS, ASOCA). Against a matched-epoch control, it significantly improves surface F1 by recovering missed distal vessels at comparable Dice. Our findings argue for measuring and optimizing the vessel surface, not the volume it overlaps. Code
Chinese Translation
在冠状计算机断层扫描血管造影（CCTA）中，准确的冠状动脉分割对于诊断冠状动脉疾病至关重要。深度网络通常使用Dice系数进行训练和评估，但体积重叠指标不适合细长的管状解剖结构：由于大多数体素属于少数几个较厚的近端段，缺失的远端分支几乎不会影响Dice，尽管这严重破坏了临床使用所需的连通性。我们引入了一种表面度量，通过在局部血管半径容差下的二分匹配来匹配预测和参考表面点，并报告精确度、召回率和F1分数，同时区分假阳性（虚假分支）和假阴性（漏检分支），这是对称Dice无法做到的。我们展示了一个强大的Dice训练基线遗漏了远比其虚构的更多的血管表面，这种不对称性被其高Dice所掩盖。在此基础上，我们提出了一种可微分的表面损失，能够同时抑制虚假质量并恢复缺失结构，通过在两个公共基准（Image-CAS，ASOCA）上微调三种主干网络（nnU-Net，SwinUNETR，NexToU）进行了验证。与匹配的训练周期对照组相比，它显著提高了表面F1分数，同时恢复了漏检的远端血管，且Dice相当。我们的研究结果主张测量和优化血管表面，而不是其重叠的体积。代码
cs.CV / 45 / 2609.31742
PEEL-DDPM: Physics-Enabled Evidential Learning for the Denoising Diffusion Probabilistic Model
PEEL-DDPM：基于物理的证据学习框架用于去噪扩散概率模型
Abstract
Normal-inverse-gamma (NIG) regression is not identifiable from its marginal Student-t likelihood: three combinations of four NIG parameters are determined, leaving one degree of freedom. We introduce PEEL-DDPM, a physics-enabled evidential learning framework for denoising diffusion probabilistic models. A measurement-conditioned DDPM is first trained with epsilon-MSE and then frozen. Its complete reverse trajectory produces a reconstruction, whose residual from the known training object is modeled by a final-image evidential network. The network learns only the identifiable Student-t coordinates, while repeated scanner-noise realizations and repeated diffusion trajectories provide a nested Monte Carlo estimate of final-image aleatoric variance, separated into scanner-induced and sampler-induced components. This measured variance resolves the remaining NIG ambiguity and yields a decomposition of predictive uncertainty into measurement, diffusion, and epistemic terms. The method uses sequential training without a cross-loss weighting coefficient. In a feasibility study on eight held-out objects, empirical central-interval coverages were 49.3%, 80.3%, 90.1%, and 95.6% for nominal 50%, 80%, 90%, and 95% intervals. The mean squared residual was 0.967 times the mean predicted variance, and a single-image aleatoric head achieved pooled Spearman correlation 0.785 against an independent nested reference. Across five dose levels, scanner-induced variance showed a log-log dose slope of -1.15, whereas sampler-induced variance remained nearly dose independent with slope -0.01. These results support PEEL-DDPM as a practical route to identifiable and physically interpretable uncertainty quantification in diffusion-based image reconstruction.
Chinese Translation
常逆伽马（NIG）回归无法从其边际学生-t 似然中识别：四个 NIG 参数的三种组合被确定，留下一个自由度。我们提出了 PEEL-DDPM，一种用于去噪扩散概率模型的基于物理的证据学习框架。首先，使用 epsilon-MSE 训练一个测量条件的 DDPM，然后将其冻结。其完整的反向轨迹生成一个重建，其与已知训练对象的残差由最终图像证据网络建模。该网络仅学习可识别的学生-t 坐标，而重复的扫描噪声实现和重复的扩散轨迹提供了最终图像随机方差的嵌套蒙特卡洛估计，分离为扫描仪引起的和采样器引起的成分。该测量方差解决了剩余的 NIG 模糊性，并将预测不确定性分解为测量、扩散和认知项。该方法使用顺序训练，无需交叉损失加权系数。在对八个保留对象的可行性研究中，经验中心区间覆盖率分别为 49.3%、80.3%、90.1% 和 95.6%，对应名义的 50%、80%、90% 和 95% 区间。均方残差为 0.967 倍于均值预测方差，单图像随机头在独立嵌套参考下达到了合并的斯皮尔曼相关系数 0.785。在五个剂量水平中，扫描仪引起的方差显示出 -1.15 的对数-对数剂量斜率，而采样器引起的方差几乎与剂量无关，斜率为 -0.01。这些结果支持 PEEL-DDPM 作为在基于扩散的图像重建中实现可识别和物理可解释的不确定性量化的实用途径。
cs.CV / 46 / 2609.31746
VisionPsy-Nano: Improving Accuracy, Efficiency, and Reliability in On-Device Vision-Language Models
VisionPsy-Nano：提升设备端视觉语言模型的准确性、效率和可靠性
Abstract
Sub-billion-parameter Vision-Language Models are increasingly viable for on-device deployment, yet compact model size alone does not guarantee usability. On a phone, such a model can still require more than two minutes to produce its first token. On-device usability depends on three axes: accuracy, efficiency, and behavioral reliability; standard benchmarks miss the third, with answers too short to expose doom loops and prompts too benign to probe adversarial safety. We introduce a diagnosis-driven post-training recipe in which a teacher VLM stress-tests the student, uncovers failure modes beyond human priors, and converts them into targeted supervision and preference alignment, supplementing generic data scaling with failure-driven optimization. Coupled with two visual-token policies, the recipe yields two accuracy-efficiency variants with improved behavioral reliability. \textbf{\NanoFull} attains a 62.3 normalized average over 17 benchmarks, the highest among openly released $\sim$0.5B models, +7.4 over its base at identical architecture and token budget, with doom-loop rates at or below the strongest baseline's. \textbf{\FlashFull} retains 61.4 while cutting warm time-to-first-token on a Pixel 9 from 138\,s to 6.1\,s (23$\times$). By jointly addressing all three axes, we move compact VLMs toward practical on-device usability.
Chinese Translation
子亿参数的视觉语言模型在设备端部署日益可行，但仅靠紧凑的模型大小并不能保证其可用性。在手机上，这样的模型仍然可能需要超过两分钟才能生成第一个标记。设备端的可用性依赖于三个方面：准确性、效率和行为可靠性；标准基准测试忽略了第三个方面，答案过于简短以至于无法暴露出灾难循环，而提示过于温和以至于无法探测对抗性安全性。我们提出了一种基于诊断驱动的后训练方法，其中教师视觉语言模型（VLM）对学生进行压力测试，揭示超出人类先验的失败模式，并将其转化为针对性的监督和偏好对齐，补充通用数据扩展与失败驱动的优化。结合两种视觉标记策略，该方法产生了两种具有改进行为可靠性的准确性-效率变体。 extbf{
anoFull} 在17个基准测试中获得62.3的标准化平均值，是公开发布的约0.5B模型中最高的，比其基础模型在相同架构和标记预算下提高了7.4，灾难循环率与最强基线相当或更低。 extbf{ lashFull} 在将Pixel 9的首次标记的预热时间从138秒缩短至6.1秒（23倍）的同时，保持了61.4的准确性。通过共同解决这三个方面，我们使紧凑的视觉语言模型朝着实际的设备端可用性迈进。
cs.CV / 47 / 2609.31747
The Earth in One Gaze: Training-Free Active Focus for UHR Remote Sensing Understanding
一瞥地球：无训练的主动聚焦用于超高分辨率遥感理解
Abstract
Multimodal large language models (MLLMs) must balance local detail against scene context when interpreting ultra-high-resolution (UHR) remote sensing (RS) imagery within a limited visual-input budget. Existing selection-based methods either prune tokens and select patches through relevance scoring, or crop actively through repeated inspection. Neither strategy directly redistributes pixels within a continuous full-scene view: the first retains selected tokens or patches, and the second re-encodes a crop detached from its surroundings. Our pilot study finds that a frozen MLLM already produces useful question-guided spatial requests, yet crop-based inspection of the selected regions does not consistently improve its answers. We therefore formulate UHR understanding as a question of where to spend a fixed pixel budget. Based on this, we introduce GazeEarth, a simple-yet-effective training-free framework that couples question-guided region selection with full-scene foveated observation. The MLLM selects evidence cells from an indexed overview; a deterministic, topology-preserving warp resamples the original image onto a fixed-size canvas, enlarging their shared neighborhood while compressing the periphery; the same frozen model answers from this focused view, using at most two MLLM calls and no external selector or iterative search. Across three UHR remote sensing benchmarks and four frozen backbones, GazeEarth improves benchmark-averaged accuracy by 4.6 to 9.4 percentage points over direct answering and 3.4 to 4.3 over overview answering, outperforming task-trained methods. Our analyses show that existing MLLMs can guide where to look in UHR images on their own, and that what they can infer from the selected evidence depends on how that evidence is presented.
Chinese Translation
多模态大型语言模型（MLLMs）在解读超高分辨率（UHR）遥感（RS）图像时，必须在有限的视觉输入预算内平衡局部细节与场景上下文。现有的基于选择的方法要么通过相关性评分修剪标记并选择补丁，要么通过反复检查主动裁剪。两种策略都未能直接在连续的全场景视图中重新分配像素：第一种保留所选的标记或补丁，第二种则重新编码与周围环境脱离的裁剪区域。我们的初步研究发现，冻结的MLLM已经能够生成有用的基于问题的空间请求，但对所选区域的裁剪检查并未始终改善其回答。因此，我们将UHR理解表述为在固定像素预算内的支出问题。在此基础上，我们引入了GazeEarth，这是一个简单而有效的无训练框架，将基于问题的区域选择与全场景的注视观察相结合。MLLM从索引概述中选择证据单元；一个确定性的、保持拓扑的变形将原始图像重新采样到固定大小的画布上，扩大其共享邻域，同时压缩周边区域；同一冻结模型从这个聚焦视图中回答，最多使用两个MLLM调用且不需要外部选择器或迭代搜索。在三个UHR遥感基准和四个冻结骨干网络上，GazeEarth在基准平均准确率上比直接回答提高了4.6到9.4个百分点，比概述回答提高了3.4到4.3个百分点，超越了任务训练的方法。我们的分析表明，现有的MLLM可以自主指导在UHR图像中查看的位置，而它们可以从所选证据中推断出的内容则依赖于该证据的呈现方式。
cs.CV / 48 / 2609.31749
Fusion Under Component Failure: Negative Results and Failure Modes in Ensemble AI-Generated Image Detection
组件故障下的融合：集成AI生成图像检测中的负面结果和故障模式
Abstract
We built an ordinary stacking ensemble for AI-generated image detection -- three open detectors producing five scores, fused by a gradient-boosted meta-learner that treats a detector's failure as missing data -- deployed it, and then evaluated it against three controls it should have faced first. This paper reports what the controls found, including where they overturned our own earlier conclusions. Fusion is worth its cost only when refitted on the target domain. The shipped meta-learner, fitted on a separate corpus, does not beat its best single member on 2000 StyleGAN faces (AUC 0.9896 vs 0.9961; McNemar p = 1.000). But a stacker refitted in-domain beats that member plus a post-hoc calibrator (Delta-AUC = +0.0025, [+0.0014, +0.0038]; p = 3.4e-4). An earlier draft claimed the calibrated single detector won outright; that comparison mixed regimes and we correct it here. One corpus is not an evaluation. On 80 screenshots every model's AUC interval contains 0.5. We can say nothing stronger: the difference between the ensemble's drop and its best member's is [-0.185, +0.115]. Abstention is real, correlated, and mishandled. With four of five detectors silent and the survivor reporting "real", the system returns P(AI) = 0.9985, because an all-NaN input scores 0.9995 in a learner never fitted with missingness. The three AIDE checkpoints fail together, sharing one preprocessing path. A quorum rule requiring two distinct architectures prevents both failures with no retraining. We also find Corpus A carries a class-conditional JPEG bias severe enough to separate the classes from the header alone, which limits every in-domain number we report. Code, harness, hash-identified artifacts and all corrections are released.
Chinese Translation
我们构建了一个普通的堆叠集成模型用于AI生成图像检测——三个开放检测器产生五个评分，通过一个将检测器故障视为缺失数据的梯度提升元学习器进行融合，随后进行了部署，并对其进行了评估，比较了它应首先面对的三个控制组。本文报告了控制组的发现，包括它们推翻我们早期结论的地方。只有在针对目标领域重新调整后，融合才值得其成本。已交付的元学习器在一个独立语料库上训练，并未在2000个StyleGAN人脸上超越其最佳单一成员（AUC 0.9896对0.9961；McNemar p = 1.000）。但是，在领域内重新调整的堆叠器超越了该成员及一个后验校准器（Delta-AUC = +0.0025，[+0.0014, +0.0038]；p = 3.4e-4）。早期草稿声称经过校准的单一检测器完全胜出；该比较混淆了不同的模式，我们在此进行了修正。一个语料库并不是评估。在80个截图中，每个模型的AUC区间都包含0.5。我们无法说出更强的结论：集成的下降与其最佳成员之间的差异为[-0.185, +0.115]。弃权是真实的，相关的，并且处理不当。当五个检测器中有四个静默，幸存者报告“真实”时，系统返回P(AI) = 0.9985，因为一个全为NaN的输入在从未适应缺失的学习器中得分为0.9995。三个AIDE检查点共同失败，共享一个预处理路径。要求两个不同架构的法定规则在不重新训练的情况下防止了这两种故障。我们还发现语料库A存在一种类别条件的JPEG偏差，严重到仅凭标题就能将类别分开，这限制了我们报告的每个领域内数字。代码、工具、哈希识别的工件及所有修正均已发布。
cs.CV / 49 / 2609.31750
Dimension-Specific Imbalance and an Adaptive Hybrid Label Strategy for Multi-Task Affective State Recognition in Classroom Video
针对课堂视频中多任务情感状态识别的维度特定不平衡与自适应混合标签策略
Abstract
Recognition of student affective states from classroom video is constrained by a class imbalance problem whose true nature is, we argue, under-analyzed. We show that on DAiSEE -- the de facto benchmark for this task -- class imbalance is fundamentally a dimension-specific phenomenon: raw imbalance ratios reach as high as approximately 79:1 (Frustration), and the structure of imbalance -- not only its magnitude -- varies across dimensions, so uniform algorithmic treatments (aggressive class reweighting, focal loss, uniform binary simplification) fail in dimension-dependent ways. We propose the Adaptive Hybrid Label Strategy (AHLS), which assigns four-level classification to dimensions with manageable imbalance and binary classification to severely long-tailed dimensions, coupled with a null-class placeholder mechanism that stabilizes multi-task optimization by compressing the maximum effective inverse-frequency weight ratio from as high as ~79$\times$ to at most ~8.5$\times$ across the four dimensions. Validated on a lightweight FERShuffleNetV2 + LTCN architecture (0.39 M parameters, 0.07 G FLOPs), the strategy attains the highest mean Macro-F1 of 47.70% across all four affective dimensions among 11 compared methods, including five state-of-the-art deep models (up to 167$\times$ larger) and five traditional machine-learning baselines. A cross-model behavioral analysis on DAiSEE indicates that, in this setting, classification strategy may influence minority-state recognition more strongly than further backbone scaling alone. We also report evidence of an annotation-density limitation of DAiSEE on minority affective states, which suggests that benchmark design may bound future progress as much as algorithmic refinement. Index Terms Affective computing, classroom video analysis, class imbalance, multi-task learning, lightweight deep learning, DAiSEE, engagement recognition.
Chinese Translation
从课堂视频中识别学生的情感状态受到类别不平衡问题的限制，而我们认为这一问题的真实性质尚未得到充分分析。我们展示了在 DAiSEE（该任务的事实标准基准）上，类别不平衡本质上是一种维度特定现象：原始不平衡比率高达约 79:1（挫败感），而不平衡的结构——不仅仅是其幅度——在不同维度之间存在差异，因此统一的算法处理（如激进的类别重加权、焦点损失、统一的二元简化）在维度依赖性方面表现不佳。我们提出了自适应混合标签策略（Adaptive Hybrid Label Strategy, AHLS），该策略对具有可管理不平衡的维度分配四级分类，对严重长尾的维度进行二元分类，并结合一个空类占位机制，通过将四个维度的最大有效逆频率权重比从高达 ~79× 压缩至最多 ~8.5×，从而稳定多任务优化。经过在轻量级 FERShuffleNetV2 + LTCN 架构（0.39 M 参数，0.07 G FLOPs）上的验证，该策略在 11 种比较方法中获得了所有四个情感维度的最高平均宏 F1 值 47.70%，包括五种最先进的深度模型（最大可达 167× 更大）和五种传统机器学习基线。对 DAiSEE 的跨模型行为分析表明，在这种情况下，分类策略可能比进一步的主干扩展更强烈地影响少数状态的识别。我们还报告了 DAiSEE 在少数情感状态上的标注密度限制的证据，这表明基准设计可能在未来进展中与算法优化同样重要。关键词：情感计算，课堂视频分析，类别不平衡，多任务学习，轻量级深度学习，DAiSEE，参与度识别。
cs.CV / 50 / 2609.31751
EgoTSR++: Egocentric Spatiotemporal Reasoning for Task Progress Understanding
EgoTSR++：以自我为中心的时空推理用于任务进展理解
Abstract
Vision-Language Models (VLMs) have advanced rapidly in static visual understanding, yet remain unreliable when judging how an egocentric task is progressing. Given a task instruction and two visual observations, a model should determine which state is closer to the goal by analyzing task-relevant object configurations and spatial relations, rather than relying on timestamps or presentation order. This distinction is critical in manipulation, where retries, corrective actions, and temporary regressions make progress inherently non-monotonic. We introduce EgoTSR, a unified framework for diagnosing and improving order-robust task-progress understanding. First, SpatialLogic-Bench evaluates each physical state pair in both original and order-swapped presentations across short- and long-horizon settings, exposing whether a model follows task-state evidence or chronological shortcuts. Second, our data construction pipeline converts successful, approximately monotonic manipulation and first-person trajectories into bidirectional supervision; LongTag further preserves intermediate subtask structure for long-horizon comparison, while failure-aware data extend learning to regressions and recoveries. Third, a progressive CoT-to-Tag curriculum first supervises evidence-grounded interpretation of task-relevant state changes and then consolidates the comparison rule through scalable label-only training. Experiments reveal substantial input-order bias in representative VLMs. EgoTSR achieves 92.4% long-horizon accuracy with a 0.1-point forward-inverse Gap. Failure-aware supervision further improves accuracy on non-monotonic trajectories by 11.8 points and Recovery Accuracy by 11.2 points, while maintaining broad visual and spatial capabilities. These results establish goal-conditioned state comparison as an explicit formulation of egocentric spatiotemporal reasoning for task-progress understanding.
Chinese Translation
视觉-语言模型（VLMs）在静态视觉理解方面迅速发展，但在判断以自我为中心的任务进展时仍然不可靠。给定任务指令和两个视觉观察，模型应通过分析与任务相关的物体配置和空间关系来确定哪个状态更接近目标，而不是依赖时间戳或呈现顺序。这一区别在操作中至关重要，因为重试、纠正措施和暂时性回退使得进展本质上是非单调的。我们提出了EgoTSR，这是一个统一框架，用于诊断和改善对任务进展理解的顺序鲁棒性。首先，SpatialLogic-Bench在短期和长期设置中评估每对物理状态在原始和顺序交换呈现下的表现，揭示模型是否遵循任务状态证据或时间顺序捷径。其次，我们的数据构建管道将成功的、近似单调的操作和第一人称轨迹转换为双向监督；LongTag进一步保留中间子任务结构以进行长期比较，而关注失败的数据则扩展学习到回退和恢复。第三，逐步的CoT-to-Tag课程首先监督基于证据的任务相关状态变化的解释，然后通过可扩展的仅标签训练巩固比较规则。实验表明，代表性VLMs存在显著的输入顺序偏差。EgoTSR在长期准确率上达到了92.4%，前向-逆向差距为0.1点。关注失败的监督进一步提高了非单调轨迹的准确率11.8点和恢复准确率11.2点，同时保持广泛的视觉和空间能力。这些结果确立了目标条件状态比较作为以自我为中心的时空推理在任务进展理解中的明确表述。
cs.CV / 51 / 2609.31754
Calibration-Free Surface Normals Estimation in Vision-Based Tactile Sensing using Universal Photometric Stereo
基于视觉的触觉传感中的无校准表面法线估计：使用通用光度立体法
Abstract
Vision-based tactile sensors are a popular solution for capturing rich contact surface geometry. However, to obtain high-detail contact surface normals and depth, it is necessary to calibrate the sensor by physically pressing a probe with known geometry against the sensor elastomer and mapping tactile images onto the ground truth probe's shape. This approach does not scale across different tactile sensors, and the calibration effort can be complex depending on the sensor shape and optical system. Instead, we propose a calibration-free procedure for the estimation of contact surface normals using Universal Photometric Stereo neural networks. In a series of real-world experiments, we evaluate our approach on 3 sensors with different optical systems, demonstrating that universal methods are a suitable approach for estimating surface normals at the contact patch from tactile images, thereby alleviating the need for tactile sensor calibration. Controlled experiments with a metal ball show that universal methods match the calibrated method, with a mean angular error of $6.56^{\Large\circ}$. We show that the proposed framework recovers high-frequency surface details of objects with natural textures, achieving an overall mean angular error of $10.66^{\Large\circ}$. Universal method robustly recovers the contact surface normals captured with dome-shaped Digit 360, achieving a low angular discrepancy of $10.18^{\Large\circ}$ relative to the calibrated baseline. This experiment demonstrates that with sufficient illumination settings surface normals could be estimated using a model trained solely on synthetic data. By providing a unified representation of contact surfaces across different vision-based tactile sensor designs, Universal Photometric Stereo neural networks lay the foundation for transferable tactile perception across sensors.
Chinese Translation
基于视觉的触觉传感器是捕捉丰富接触表面几何形状的热门解决方案。然而，为了获得高细节的接触表面法线和深度，必须通过将已知几何形状的探针物理压入传感器的弹性体并将触觉图像映射到真实探针的形状上来校准传感器。这种方法在不同的触觉传感器之间无法扩展，并且校准工作可能因传感器形状和光学系统的不同而变得复杂。相反，我们提出了一种无校准的程序，通过使用通用光度立体法神经网络来估计接触表面法线。在一系列实际实验中，我们在三种具有不同光学系统的传感器上评估了我们的方法，证明通用方法是从触觉图像中估计接触点表面法线的合适方法，从而减轻了对触觉传感器校准的需求。与金属球的控制实验表明，通用方法与校准方法匹配，平均角度误差为 $6.56^{ ext{°}}$。我们展示了所提出的框架能够恢复具有自然纹理的物体的高频表面细节，整体平均角度误差为 $10.66^{ ext{°}}$。通用方法稳健地恢复了用圆顶形状的 Digit 360 捕获的接触表面法线，相对于校准基线实现了低角度差异 $10.18^{ ext{°}}$。该实验表明，在足够的照明条件下，可以使用仅在合成数据上训练的模型来估计表面法线。通过提供不同基于视觉的触觉传感器设计的接触表面的统一表示，通用光度立体法神经网络为跨传感器的可转移触觉感知奠定了基础。
cs.CV / 52 / 2609.31755
Gauge-Equivariant Attention for Rotation-Stable $360^\circ$ Scene Understanding
旋转稳定的 $360^ ext{°}$ 场景理解中的规范等变注意力
Abstract
Panoramic $360^\circ$ scene understanding increasingly relies on icosphere transformers, but a state-of-the-art spherical model loses more than half of its segmentation accuracy when the camera rotates by $90^\circ$, and controlled ablations identify gauge dependence in its relative-position bias as a major contributor. We propose gauge-equivariant relative position encoding (GE-RPE): a parameter-free Reynolds average of the bias over a finite cyclic subgroup $C_n\!\subset\!\mathrm{SO}(2)$ of gauge rotations. Plugged into a SphereUFormer backbone the change is invisible at deployment---zero added parameters and $1.5$--$4.4\%$ forward latency---and the matched three-seed GE-RPE model records a $1.3\%$ drop; the published SphereUFormer checkpoint records $53\%$ under the same stress protocol but a different training recipe. Once the gauge defect is removed and a teacher-token permutation $\pi_R$ aligns the SSL views to the rotated student frame, iBOT$+$MAE pretraining stops being a liability and becomes a clean low-label lever: the full framework EquiSSL (GE-RPE $+$ $\pi_R$ $+$ iBOT$+$MAE) tightens the drop to $0.8\%$ at $68.30\%$ val mIoU and lifts $1\%$-label fine-tuning by $+2.39$ mIoU on the $N{=}373$ test split (and by $+4.10$ on the smaller $N{=}40$ val split); the same fix carries over to monocular depth and to zero-shot Structured3D segmentation. The construction is provably $C_n$-invariant and $\mathcal{O}(n^{-2})$-close to the continuous $\mathrm{SO}(2)$ average, making the resulting model a usable $360^\circ$ visual-computing primitive across panoramic relighting, immersive video, and cross-dataset transfer. Code is available at https://github.com/Jaywalk18/equissl-release.
Chinese Translation
全景 $360^ ext{°}$ 场景理解日益依赖于球面变换器，但一种最先进的球面模型在相机旋转 $90^ ext{°}$ 时，其分割精度损失超过一半，控制性消融实验表明其相对位置偏差中的规范依赖是主要原因。我们提出了规范等变相对位置编码 (GE-RPE)：这是对有限循环子群 $C_n ext{ } ext{⊆} ext{ } ext{SO}(2)$ 的偏差进行无参数的雷诺平均。将其嵌入到 SphereUFormer 主干网络中，部署时变化不可见——没有增加参数且前向延迟为 $1.5$--$4.4 ext{％}$——而匹配的三种种子 GE-RPE 模型记录了 $1.3 ext{％}$ 的下降；已发布的 SphereUFormer 检查点在相同的压力协议下记录为 $53 ext{％}$，但训练配方不同。一旦去除了规范缺陷，并且教师令牌置换 $ ext{π}_R$ 将 SSL 视图与旋转的学生框架对齐，iBOT$+$MAE 预训练不再是负担，而成为一个干净的低标签杠杆：完整框架 EquiSSL (GE-RPE $+$ $ ext{π}_R$ $+$ iBOT$+$MAE) 将下降幅度缩小至 $0.8 ext{％}$，在 $68.30 ext{％}$ 的验证 mIoU 上提升了 $1 ext{％}$ 标签微调的 $+2.39$ mIoU（在较小的 $N=40$ 验证集上提升 $+4.10$）；同样的修正也适用于单目深度和零样本 Structured3D 分割。该构造在理论上是 $C_n$-不变的，并且与连续的 $ ext{SO}(2)$ 平均值的距离为 $ ext{O}(n^{-2})$，使得所得到的模型在全景重光照、沉浸式视频和跨数据集迁移中成为可用的 $360^ ext{°}$ 视觉计算原语。代码可在 https://github.com/Jaywalk18/equissl-release 获取。
cs.CV / 53 / 2609.31756
3dgs-sc: a controlled static screen-content benchmark for 3d gaussian splatting
3DGS-SC：用于3D高斯溅射的受控静态屏幕内容基准
Abstract
Does high image fidelity imply readable screen content in 3D Gaussian Splatting (3DGS)? We introduce 3DGS-SC, a controlled static screen-content dataset and benchmark for examining this mismatch. Ten procedural scenes provide fixed multi-view splits, exact cameras, screen masks, text boxes, and transcripts. The protocol separates whole-image fidelity, screen-region fidelity, OCR readability, and edge preservation. In the reported five-method comparison, LightGaussian exceeds Mip-Splatting in screen PSNR by only 0.08 dB, yet trails it in OCR accuracy by 12.7 percentage points. Across all ten method pairs, screen-PSNR and OCR orderings disagree in seven cases. These aggregate results expose a method-selection failure of fidelity-only evaluation. Complementing prior text-aware 3DGS research, 3DGS-SC targets controlled monitor interfaces and exact annotations; scene-wise robustness and acquisition effects remain open validation questions.
Chinese Translation
高图像保真度是否意味着在3D高斯溅射（3DGS）中可读的屏幕内容？我们引入了3DGS-SC，这是一个受控的静态屏幕内容数据集和基准，用于检验这种不匹配。十个程序生成的场景提供了固定的多视图分割、精确的相机、屏幕掩码、文本框和转录文本。该协议将整个图像的保真度、屏幕区域的保真度、光学字符识别（OCR）可读性和边缘保留分开。在报告的五种方法比较中，LightGaussian在屏幕峰值信噪比（PSNR）上仅比Mip-Splatting高出0.08 dB，但在OCR准确性上落后12.7个百分点。在所有十对方法中，屏幕PSNR和OCR的排序在七个案例中存在不一致。这些综合结果揭示了仅依赖保真度评估的选择方法的失败。作为对先前文本感知3DGS研究的补充，3DGS-SC针对受控的监视器界面和精确的注释；场景级的鲁棒性和获取效果仍然是开放的验证问题。
cs.CV / 54 / 2609.31765
HGPTrans: Hierarchical Graph-Pooling Transolver for Automotive Aerodynamic Drag Coefficient Prediction
HGPTrans：用于汽车空气动力学阻力系数预测的层次图池化转解算器
Abstract
Accurate and rapid prediction of the aerodynamic drag coefficient ($C_D$) is essential for vehicle design, particularly during early-stage styling iterations where a large number of candidate geometries must be evaluated. Although computational fluid dynamics (CFD) provides reliable aerodynamic estimates, its high computational cost, typically requiring hours to days for a single configuration, limits its use in large-scale design exploration. This paper proposes HGPTrans, a hierarchical graph-pooling network with Transolver-based attention, to directly predict $C_D$ from vehicle surface meshes. Motivated by the fact that vehicle aerodynamics depends on both local geometric features and long-range interactions among spatially distant surface regions, HGPTrans integrates three complementary components. Graph isomorphism convolutions encode discriminative local geometry, physics-aware slice attention captures global interactions with linear computational complexity, and information-redundancy-aware hierarchical pooling progressively removes redundant nodes while preserving informative geometric structures. The model is trained and evaluated on the large-scale DrivAerNet and DrivAerNet++ datasets, where it achieves the lowest mean absolute error and mean squared error among the evaluated baselines. Its generalization capability is further assessed through transfer learning on a real-vehicle dataset containing both sedans and SUVs, achieving relative $L_1$ errors of 1.56% (sedans) and 2.12% (SUVs) with an inference time of approximately $0.293$ s per vehicle. This corresponds to an acceleration of several orders of magnitude relative to high-fidelity CFD while keeping the predicted drag coefficients within a few percent of the CFD reference. Ablation studies confirm each component's contribution and reveal the effects of depth and pooling ratio.
Chinese Translation
准确快速地预测空气动力学阻力系数 ($C_D$) 对于车辆设计至关重要，尤其是在早期样式迭代阶段，需要评估大量候选几何形状。尽管计算流体动力学 (CFD) 提供可靠的空气动力学估计，但其高计算成本，通常需要数小时到数天才能完成单一配置的计算，限制了其在大规模设计探索中的应用。本文提出了 HGPTrans，一种基于转解算器 (Transolver) 注意力机制的层次图池化网络，旨在直接从车辆表面网格预测 $C_D$。HGPTrans 的设计受到车辆空气动力学依赖于局部几何特征和空间上远离的表面区域之间的长距离交互这一事实的启发，集成了三个互补组件。图同构卷积编码了具有区分性的局部几何特征，物理感知切片注意力以线性计算复杂度捕获全局交互，而信息冗余感知的层次池化则逐步去除冗余节点，同时保留信息丰富的几何结构。该模型在大规模的 DrivAerNet 和 DrivAerNet++ 数据集上进行训练和评估，在所评估的基线中实现了最低的平均绝对误差和均方误差。通过在包含轿车和 SUV 的真实车辆数据集上进行迁移学习，进一步评估了其泛化能力，取得了相对 $L_1$ 误差为 1.56%（轿车）和 2.12%（SUV），每辆车的推理时间约为 $0.293$ 秒。这相对于高保真 CFD 加速了几个数量级，同时保持预测的阻力系数在 CFD 参考值的几个百分点内。消融研究确认了每个组件的贡献，并揭示了深度和池化比例的影响。
cs.CV / 55 / 2609.31766
UNMATCH: Selective Unbalanced Token-Patch Matching for Forensic Image-Claim Verification
UNMATCH：用于法医图像声明验证的选择性不平衡令牌-补丁匹配
Abstract
Contextual image misuse pairs an image with a misleading claim. We study image-claim correspondence in fact-checked pairs containing out-of-context reuse, visual manipulation, or both. Existing pair-based detectors often compress the two modalities into a global compatibility score or learn a highly flexible interaction module, which can obscure a decisive local mismatch. We introduce directional multiscale coverage, a compact representation that summarizes local image-claim affinity in both directions and at three spatial scales. At each scale, each direction is summarized by its mean, lower quartile, and two thresholded support ratios; the signed difference between directional means completes a nine-dimensional scale descriptor. Concatenating the three scales yields a compact local representation for a lightweight global-local classifier. Under leakage-aware three-fold, three-seed evaluation on the Snopes subset of the Fauxtography benchmark, UNMATCH achieves 69.82 Macro-F1 and 71.05 balanced accuracy, exceeding the MCOT adaptation by 2.60 and 2.16 points. A matched-reassigned intervention shows that breaking the observed pairing lowers coverage and increases both discrepancy and false-pair probability.
Chinese Translation
上下文图像误用将图像与误导性声明配对。我们研究包含上下文外重用、视觉操控或两者兼有的事实核查对中的图像-声明对应关系。现有的基于配对的检测器通常将这两种模态压缩为一个全局兼容性得分，或学习一个高度灵活的交互模块，这可能会掩盖决定性的局部不匹配。我们引入了方向性多尺度覆盖（directional multiscale coverage），这是一种紧凑表示，能够在两个方向和三个空间尺度上总结局部图像-声明亲和力。在每个尺度上，每个方向通过其均值、下四分位数和两个阈值支持比率进行总结；方向均值之间的符号差异完成了一个九维尺度描述符。将三个尺度连接起来，形成一个紧凑的局部表示，用于轻量级的全局-局部分类器。在对Fauxtography基准的Snopes子集进行泄漏感知的三折、三种种子评估下，UNMATCH达到了69.82的宏观F1和71.05的平衡准确率，分别比MCOT适应提高了2.60和2.16点。一个匹配重新分配的干预显示，打破观察到的配对会降低覆盖率，并增加差异性和错误配对概率。
cs.CV / 56 / 2609.31780
Panoptic Scene Program Diffusion Transformer
全景场景程序扩散变换器
Abstract
Modern text-to-image models produce high-fidelity images but still struggle with compositional prompts that require instance identity, attribute ownership, counting, spatial ordering, and role-sensitive relations. We introduce Panoptic Scene Program Diffusion Transformer (PSP-DiT), a diffusion-transformer architecture that treats a panoptic scene program as a first-class latent variable rather than an external control signal or post-hoc parse. PSP-DiT jointly denoises image latents and scene-program latents through coupled transformer streams, while panoptic grounding and cycle-consistency objectives tie object instances, attributes, relations, and counts to visual support in the generated image. Under matched training and inference settings, PSP-DiT improves over a strong flat-text baseline across GenEval 2, SANEval-Simple, PSG-Score, and DetailMaster, with the largest gains on counting, attribute binding, role-sensitive relations, and long structured prompts. The method preserves image quality, adds modest inference overhead, and remains robust to imperfect scene programs.
Chinese Translation
现代文本到图像模型能够生成高保真图像，但在处理需要实例身份、属性归属、计数、空间排序和角色敏感关系的组合提示时仍然存在困难。我们提出了全景场景程序扩散变换器（Panoptic Scene Program Diffusion Transformer，PSP-DiT），这是一种将全景场景程序视为一类重要的潜变量的扩散-变换器架构，而非外部控制信号或事后解析。PSP-DiT通过耦合的变换器流共同去噪图像潜变量和场景程序潜变量，同时全景基础和循环一致性目标将对象实例、属性、关系和计数与生成图像中的视觉支持联系起来。在匹配的训练和推理设置下，PSP-DiT在GenEval 2、SANEval-Simple、PSG-Score和DetailMaster等基准测试中，相较于强大的平面文本基线表现出显著提升，尤其在计数、属性绑定、角色敏感关系和长结构提示方面取得了最大的进展。该方法保持了图像质量，增加了适度的推理开销，并对不完美的场景程序保持鲁棒性。
cs.CV / 57 / 2609.31788
SelfCue: Making a 3D CT Report Generator Say What It Already Knows
SelfCue：让3D CT报告生成器说出它已经知道的内容
Abstract
Progress in 3D CT report generation is usually sought in increasingly sophisticated architectures and larger pools of training data. We find instead that a 3D CT report generator already holds what its report leaves out, and loses it when the hidden state becomes tokens. Over the 18 CT-RATE abnormalities, this hidden-to-report surfacing gap is reflected by a drop in macro AUROC from 0.848 in the hidden states to 0.739 in the generated report. We propose SelfCue based on contrastive decoding. It promotes what the hidden state already supports and suppresses what it does not. It raises clinical efficacy F1 to 0.481 and the LLM-judged GREEN score to 0.510. Distilling that behaviour into the weights gives SelfCue-KD, a student that keeps most of the gain, needs nothing extra at inference, and drops into any pipeline already serving the baseline. Code is available at https://github.com/renjie-liang/SelfCue-CT.
Chinese Translation
在3D CT报告生成的进展中，通常寻求更复杂的架构和更大规模的训练数据。然而，我们发现3D CT报告生成器实际上已经掌握了报告中遗漏的信息，而当隐藏状态转化为标记时，这些信息便会丢失。在18种CT-RATE异常中，这种隐藏状态与报告之间的差距体现在宏观AUROC的下降，从隐藏状态的0.848降至生成报告的0.739。我们提出了基于对比解码的SelfCue。它促进隐藏状态已经支持的内容，抑制不支持的内容。它将临床效能F1提高至0.481，并将LLM评估的GREEN评分提高至0.510。将这种行为提炼到权重中，形成了SelfCue-KD，一个在推理时保持大部分收益、无需额外内容并能融入任何已有基线的管道的学生模型。代码可在 https://github.com/renjie-liang/SelfCue-CT 获取。
cs.CV / 58 / 2609.31795
Rate-Adaptive One-Step Diffusion Compression for AIGC Images
适应速率的一步扩散压缩用于AIGC图像
Abstract
We describe our entry to the LoViF 2026 AIGC Image Compression Challenge, a benchmark for ultra-low-bitrate coding of AI-generated images under a strict global rate budget of 0.025 bits per pixel (BPP). Generated imagery poses a distinct challenge for compression: it frequently contains rendered typography, synthetic edges, repeated motifs, UI-like layout, and stylized micro-texture that conventional distortion-oriented codecs erase at this rate, while unconstrained generative decoders can restore plausible-looking detail that no longer matches the source geometry or symbols. We treat this as a rate-perception allocation problem. Our system fine-tunes four rate-specialized checkpoints of the AEIC one-step diffusion codec, generates per-image candidates from all four, including one latent refined through encoder-side test-time optimization (TTO) with periodic entropy-conditioning refresh, entropy-codes every candidate with practical rANS coding, and selects exactly one bitstream per image with an exact multiple-choice knapsack solved over true coded file sizes. A fixed, zero-additional-bit residual restoration network is applied at decode time. Every submitted bitstream is independently decodable by the shipped decoder, which uses no source image or external side information. We report the full pipeline, an ablation history spanning 77 logged experiments, and a set of negative results, including why PSNR could not be pushed to parity with rate-distortion-oriented competitors under this architecture, useful to future participants. This is a challenge report: our entry scored 31.527739 (PSNR 27.02 dB, MS-SSIM 0.9176, LPIPS 0.0778, DISTS 0.0390 at 0.02495 BPP), the second-best DISTS on the leaderboard, ranking 5th on the final test-phase leaderboard announced August 4, 2026.
Chinese Translation
我们描述了我们在LoViF 2026 AIGC图像压缩挑战赛中的参赛作品，这是一个针对在严格的全球速率预算0.025比特每像素（BPP）下对AI生成图像进行超低比特率编码的基准测试。生成的图像在压缩方面面临独特的挑战：它们通常包含渲染的排版、合成边缘、重复的图案、类似用户界面的布局以及风格化的微纹理，而传统的以失真为导向的编解码器在这个速率下会抹去这些细节，而不受限制的生成解码器则可以恢复看似合理的细节，但这些细节不再与源几何或符号匹配。我们将此视为一个速率-感知分配问题。我们的系统微调了AEIC一步扩散编解码器的四个速率专用检查点，从所有四个检查点生成每个图像的候选项，其中一个通过编码器端的测试时间优化（TTO）进行潜在的细化，并定期进行熵条件刷新，使用实用的rANS编码对每个候选项进行熵编码，并通过对真实编码文件大小进行精确的多选背包问题求解，选择每个图像的一个比特流。在解码时应用一个固定的、零额外比特的残差恢复网络。每个提交的比特流都可以由发货的解码器独立解码，该解码器不使用源图像或外部侧信息。我们报告了完整的流程、跨越77个记录实验的消融历史，以及一组负结果，包括为什么在这种架构下PSNR无法与以速率-失真为导向的竞争者持平，这对未来的参与者有用。这是一份挑战报告：我们的参赛作品得分为31.527739（PSNR 27.02 dB，MS-SSIM 0.9176，LPIPS 0.0778，DISTS 0.0390，速率为0.02495 BPP），在排行榜上DISTS排名第二，在2026年8月4日公布的最终测试阶段排行榜上排名第五。
cs.CV / 59 / 2609.31821
A Surgical Foundation Model Reveals Task-Dependent Label Efficiency
手术基础模型揭示任务依赖的标签效率
Abstract
Developing label-efficient models is a central challenge in surgical AI due to the high cost and scarcity of expert annotation. While self-supervised foundation models adapt well to new tasks with minimal data, how label efficiency varies across different surgical tasks remains largely unexplored. Here, we introduce SURGE, a surgical foundation model trained on SurgSpectrum-30M+, the largest pretraining dataset comprising over 30 million frames, with checkpoints released to enable further research. We systematically evaluate label efficiency across 5 task categories and 15 benchmarks. These range from temporal and spatial scene understanding to fine-grained reasoning tied to instrument-anatomy interactions and safety-critical maneuvers. SURGE outperforms prior state-of-the-art on all benchmarks, even surpassing task-specific models on complex reasoning tasks. Crucially, we reveal a task-dependent scaling behavior: while scene understanding tasks saturate with minimal supervision, fine-grained reasoning tasks continue improving with substantially larger annotation budgets, providing a blueprint for allocating expert effort in complex domains. Code: https://github.com/CAMMA-public/SURGE
Chinese Translation
开发标签高效的模型是手术人工智能领域的一个核心挑战，因为专家标注的成本高且稀缺。尽管自监督基础模型能够在最小数据下良好适应新任务，但不同手术任务之间标签效率的变化仍然很大程度上未被探索。在此，我们介绍了SURGE，一个在SurgSpectrum-30M+上训练的手术基础模型，该数据集是最大的预训练数据集，包含超过3000万帧，并发布了检查点以便于进一步研究。我们系统地评估了5个任务类别和15个基准上的标签效率。这些任务涵盖了时间和空间场景理解，以及与器械-解剖交互和安全关键操作相关的细粒度推理。SURGE在所有基准上均超越了之前的最先进技术，甚至在复杂推理任务上超越了特定任务模型。重要的是，我们揭示了任务依赖的扩展行为：尽管场景理解任务在最小监督下达到饱和，细粒度推理任务在显著更大的标注预算下仍持续改善，为在复杂领域分配专家努力提供了蓝图。代码：https://github.com/CAMMA-public/SURGE
cs.CV / 60 / 2609.31823
SynDORBench: Evaluating LVLM Perceptual Robustness Under Physically Constrained Visibility Conditions
SynDORBench：在物理约束可见性条件下评估大型视觉语言模型的感知鲁棒性
Abstract
Large vision-language models (LVLMs) have demonstrated remarkable performance on multimodal reasoning benchmarks, yet their perceptual reliability under physically constrained imaging conditions remains poorly understood. Existing evaluations predominantly assume ideal visual inputs and therefore fail to characterize how camera distance, illumination, viewpoint, and pixel density fundamentally affect semantic recoverability. We introduce SynDORBench, the first physically grounded benchmark for evaluating LVLM perceptual robustness under DORI-calibrated conditions aligned with human visual capability standards. SynDORBench comprises over 54k question--answer pairs generated through a controllable synthetic pipeline that systematically varies viewing distance, lighting, camera geometry, and action pose according to physically interpretable pixel-density regimes. To support scalable low-visibility supervision, we further propose a discernibility annotation framework that propagates human perceptual labels using mask-conditioned statistical features and ensemble learning. We evaluate 16 open-source LVLMs, a commercial LVLM baseline, and YOLO11x across human-presence classification and action recognition tasks under progressively degraded visibility conditions. Our results reveal that perceptual failure in LVLMs is strongly governed by pixel density and physical imaging constraints rather than model scale alone. Surprisingly, several compact open-source LVLMs outperform larger commercial baselines and substantially exceed YOLO11x robustness under long-range and low-light conditions. SynDORBench establishes a new benchmark paradigm for physically grounded multimodal evaluation, enabling systematic analysis of LVLM reliability under real-world perceptual constraints and direct comparison against human visibility thresholds.
Chinese Translation
大型视觉语言模型（LVLMs）在多模态推理基准测试中表现出色，但它们在物理约束成像条件下的感知可靠性仍然不够清楚。现有评估主要假设理想的视觉输入，因此未能表征相机距离、照明、视角和像素密度如何从根本上影响语义可恢复性。我们提出了SynDORBench，这是第一个基于物理的基准，用于在与人类视觉能力标准对齐的DORI校准条件下评估LVLM的感知鲁棒性。SynDORBench包含超过54,000对通过可控合成管道生成的问题-答案对，该管道系统地根据物理可解释的像素密度模式变化观看距离、照明、相机几何和动作姿态。为了支持可扩展的低可见性监督，我们进一步提出了一种可辨识性注释框架，该框架使用掩码条件的统计特征和集成学习传播人类感知标签。我们在逐步降低可见性条件下评估了16个开源LVLM、一个商业LVLM基线和YOLO11x在人体存在分类和动作识别任务中的表现。我们的结果揭示，LVLM的感知失败主要受像素密度和物理成像约束的强烈影响，而不仅仅是模型规模。令人惊讶的是，几个紧凑的开源LVLM在长距离和低光照条件下的表现超过了更大的商业基线，并显著超越了YOLO11x的鲁棒性。SynDORBench建立了一种新的基准范式，用于基于物理的多模态评估，使得在现实世界感知约束下系统分析LVLM的可靠性成为可能，并与人类可见性阈值进行直接比较。
cs.CV / 61 / 2609.31873
CueKFS: Agentic Cue-Driven Keyframe Selection for Long Video Understanding
CueKFS：基于代理线索驱动的长视频理解关键帧选择
Abstract
Keyframe selection (KFS) has long produced compact video summaries for browsing and retrieval, and representative frames for thumbnails. More recently, when conditioned on a question, KFS provides an alternative to uniform sampling for long-video question answering by selecting frames that are more relevant to the question. Most methods rank frames by similarity to the question. Yet a relevant frame may score poorly when the question combines subjects or moments that no single frame shows, or requires implicit information absent from its wording. Other methods try to break down the question into subqueries, but suffer from inaccurate decomposition due to their static initial context. Therefore, we propose CueKFS, a training-free method that reformulates question--frame matching as comparing frames against a set of dynamically generated visual cues. From an initial set of salient frames, we decompose the question into cues. Each cue concurrently probes the video to navigate to its own evidence. A reasoning VLM then agentically revises the cue set against its evidence to re-explore the video. CueKFS then allocates the budget across the surviving cues. Across three benchmarks, CueKFS establishes state-of-the-art results in all 27 evaluated settings with available prior results, achieving budget-averaged gains of up to +4.54% over the previous baseline and a median of only two VLM calls. We further provide a detailed behavioral analysis of CueKFS, showing that agentic cue refinement drives active re-exploration of the video, yielding relative similarity gains of up to 92% over the initial context.
Chinese Translation
关键帧选择（KFS）长期以来为浏览和检索生成紧凑的视频摘要，并为缩略图提供代表性帧。最近，当基于问题进行条件设置时，KFS通过选择与问题更相关的帧，为长视频问答提供了一种替代均匀采样的方法。大多数方法通过与问题的相似性对帧进行排名。然而，当问题结合了没有单一帧显示的主题或时刻，或者需要缺失于其措辞中的隐含信息时，相关帧可能得分较低。其他方法试图将问题分解为子查询，但由于其静态初始上下文，分解的准确性受到影响。因此，我们提出了CueKFS，这是一种无训练的方法，将问题-帧匹配重新表述为将帧与一组动态生成的视觉线索进行比较。从一组初始显著帧中，我们将问题分解为线索。每个线索同时探测视频以导航到其自身的证据。然后，一个推理的视觉语言模型（VLM）主动修正线索集以根据其证据重新探索视频。CueKFS随后在存活的线索之间分配预算。在三个基准测试中，CueKFS在所有27个评估设置中建立了最先进的结果，相较于之前的基线，预算平均增益高达+4.54%，且中位数仅为两个VLM调用。我们进一步提供了CueKFS的详细行为分析，显示代理线索的细化驱动了视频的主动重新探索，相较于初始上下文，产生了高达92%的相对相似性增益。
cs.CV / 62 / 2609.31896
Auditing Quality Filters for Long-Tail Human Data Curation
长尾人类数据策展的审计质量过滤器
Abstract
Robots on construction sites must detect workers who are kneeling or bending, which we call low poses. These workers can be lost from training datasets during automatic labeling. We study a pipeline that detects people, estimates their body joints using NLF, and groups similar poses. Low poses account for only about 2 percent of the retained examples. This low share may partly reflect the pipeline's quality filter, which rejects examples with low detection confidence or uncertain joint estimates. We examine this filtering using four alternative pose clues: bounding-box shape, vertical body span, pose grouping aligned to the scene's vertical direction, and image appearance. All four suggest that low poses are rejected by the filter more often. Separately, controlled simulated scenes show that a person detector fine-tuned on a public construction dataset misses more workers in these poses even when we correct their bounding box height is matched to that of standing workers. These findings suggest that low poses are scarce and hard to find, and we cannot rely on bounding boxes or poses for long-tail human data curation.
Chinese Translation
建筑工地上的机器人必须检测跪着或弯腰的工人，我们称之为低姿态。这些工人在自动标注过程中可能会从训练数据集中丢失。我们研究了一种管道，该管道检测人员，使用 NLF 估计他们的身体关节，并对相似姿态进行分组。低姿态仅占保留样本的约 2%。这一低比例可能部分反映了管道的质量过滤器，该过滤器会拒绝低检测置信度或不确定关节估计的样本。我们使用四种替代姿态线索来检查这种过滤：边界框形状、垂直身体跨度、与场景垂直方向对齐的姿态分组以及图像外观。所有四种线索都表明，低姿态更常被过滤器拒绝。另一方面，受控的模拟场景显示，经过公共建筑数据集微调的人体检测器在这些姿态下漏检更多工人，即使我们纠正了他们的边界框高度，使其与站立工人相匹配。这些发现表明，低姿态稀缺且难以找到，我们不能依赖边界框或姿态进行长尾人类数据策展。
cs.CV / 63 / 2609.31911
Enhancing Visual Reasoning in Chest X-Ray Report Generation Using Reinforcement Learning
利用强化学习增强胸部X光报告生成中的视觉推理
Abstract
Medical report generation has made significant progress with the rise of modern vision-language models and the growing availability of large-scale medical datasets. However, hallucinations remain a major challenge, largely due to the limitations of supervised fine-tuning (SFT), which prioritizes lexical similarity to reference reports rather than clinical correctness. While reinforcement learning has shown strong performance in domains with verifiable rewards such as mathematics and code generation, its application to open-ended medical tasks remains limited. Existing work focuses on evaluating final answers, overlooking the model's reasoning, despite evidence that flawed reasoning can degrade overall performance. In this study, we propose a framework that verifies the model's reasoning process by integrating anatomical regions, bounding boxes, and region-level textual descriptions. We design spatial and factual reward mechanisms to ensure that the model's reasoning is both visually grounded and factually accurate. Starting from Qwen3-VL-8B-Instruct as our base model, we adapt it to the medical domain using supervised fine-tuning, introduce reasoning capability through a cold-start SFT stage, and refine it with reinforcement learning. We find that RL provides performance gains beyond those achievable through SFT alone, and that jointly verifying both reasoning steps and final outputs yields larger improvements than verifying either in isolation. We further identify multiple modes of reward hacking in the RL stage. Finally, the model's structured think traces enhance interpretability, making its outputs easier to audit for clinical use.
Chinese Translation
随着现代视觉-语言模型的兴起和大规模医学数据集的日益可用，医学报告生成取得了显著进展。然而，幻觉现象仍然是一个主要挑战，这在很大程度上是由于监督微调（SFT）的局限性，后者优先考虑与参考报告的词汇相似性，而非临床正确性。尽管强化学习在数学和代码生成等可验证奖励的领域表现出色，但其在开放式医学任务中的应用仍然有限。现有研究主要集中在评估最终答案，忽视了模型的推理过程，尽管有证据表明，推理缺陷可能会降低整体性能。在本研究中，我们提出了一个框架，通过整合解剖区域、边界框和区域级文本描述来验证模型的推理过程。我们设计了空间和事实奖励机制，以确保模型的推理既有视觉基础又在事实准确性上得到保证。以Qwen3-VL-8B-Instruct作为基础模型，我们通过监督微调将其适应医学领域，通过冷启动SFT阶段引入推理能力，并通过强化学习进行优化。我们发现，强化学习提供的性能提升超出了仅通过监督微调所能实现的水平，并且联合验证推理步骤和最终输出所带来的改进大于单独验证任一方面的效果。此外，我们在强化学习阶段识别出多种奖励黑客模式。最后，模型的结构化思维轨迹增强了可解释性，使其输出更易于进行临床审计。
cs.CV / 64 / 2609.31912
PredRA: Fast Medical Image Translation by Deterministic Component Extraction and Controlled Stochastic Refinement
PredRA：通过确定性组件提取和受控随机细化实现快速医学图像翻译
Abstract
Strongly paired medical image translation contains a substantial component that can be predicted directly from the source. The current reality is that pure deterministic prediction can smooth away fine detail, while generative models can recover detail but may also introduce unnecessary or potentially harmful variation. We propose PredRA, a fast framework that uses the deterministic prediction as a stable reference and further extracts additional deterministic components from the residual during generative refinement, thereby improving fidelity while maintaining perceptual quality. The goal is simple: we view the entire residual-based generative process as an optimization problem and derive a practical solution that recovers useful residual detail through controlled refinement while keeping the prediction close to the paired target. Mechanism studies further show that useful residual information follows structured patterns, but its usefulness is difficult to estimate reliably at the voxel level. We therefore globally control how much of the residual refinement is added to the deterministic prediction, thereby reducing the accumulation of unnecessary uncertainty. PredRA therefore combines deterministic component extraction with controlled stochastic refinement for fast, fidelity-preserving, and perceptually strong medical image translation. We validate the approach across multiple real-world datasets, showing that controlled refinement improves fidelity to the paired target compared with full residual refinement while retaining much of the perceptual benefit of generative modeling. PredRA achieves competitive or superior performance to substantially larger state-of-the-art models with 1.4-11.9x fewer total parameters and 3.1-26.2x fewer trainable parameters, while its 32-step flow sampler requires 31.25x fewer sampling steps than the matched 1000-step DDPM.
Chinese Translation
强配对医学图像翻译包含一个可以直接从源图像预测的显著组件。目前的现实是，纯确定性预测可能会抹去细节，而生成模型能够恢复细节，但也可能引入不必要或潜在有害的变异。我们提出了PredRA，一个快速框架，利用确定性预测作为稳定参考，并在生成细化过程中进一步从残差中提取额外的确定性组件，从而在保持感知质量的同时提高保真度。目标很简单：我们将整个基于残差的生成过程视为一个优化问题，并推导出一个实用的解决方案，通过受控细化恢复有用的残差细节，同时保持预测接近配对目标。机制研究进一步表明，有用的残差信息遵循结构化模式，但在体素级别上可靠估计其有用性是困难的。因此，我们全局控制添加到确定性预测中的残差细化量，从而减少不必要的不确定性累积。因此，PredRA将确定性组件提取与受控随机细化相结合，实现快速、保真且感知强的医学图像翻译。我们在多个真实世界数据集上验证了该方法，结果表明，与完全残差细化相比，受控细化提高了与配对目标的保真度，同时保留了生成建模的大部分感知优势。PredRA在参数总数上比大得多的最先进模型减少了1.4-11.9倍的参数，且可训练参数减少了3.1-26.2倍，而其32步流采样器所需的采样步骤比匹配的1000步DDPM少31.25倍。
cs.CV / 65 / 2609.31915
Facial classification Using Hybrid Quantum Machine Learning
使用混合量子机器学习进行面部分类
Abstract
Hybrid quantum methods have received limited study for resource-constrained facial biometrics. We present a hybrid quantum-classical facial recognition pipeline designed to run on standard computing hardware. Images undergo gamma correction, contrast enhancement, and principal component analysis before their features are encoded into an eight-qubit variational quantum classifier. Classical image matching then performs recognition. In experiments with 50,000 images, comprising 25,000 faces from CelebA and 25,000 non-face images from CIFAR-10, the method outperformed the reported CPU-trained FaceNet baseline in accuracy and training efficiency. The pipeline was also evaluated on GPU and quantum hardware. In an attendance monitoring deployment at Mahindra University in collaboration with Lloyds Technology Centre, CPU inference took 0.2 to 0.5 seconds per person, and the system remained robust to the use of spectacles. These findings support the feasibility of deploying hybrid quantum methods for facial recognition on existing CPU hardware.
Chinese Translation
混合量子方法在资源受限的面部生物识别领域的研究相对有限。我们提出了一种混合量子-经典面部识别流程，旨在在标准计算硬件上运行。图像经过伽马校正、对比度增强和主成分分析后，其特征被编码到一个八量子比特变分量子分类器中。随后，经典图像匹配进行识别。在包含50,000张图像的实验中，其中包括来自CelebA的25,000张面部图像和来自CIFAR-10的25,000张非面部图像，该方法在准确性和训练效率上均优于报告的CPU训练的FaceNet基线。该流程还在GPU和量子硬件上进行了评估。在与Lloyds技术中心合作的Mahindra大学的出勤监测部署中，CPU推理每人耗时0.2到0.5秒，且系统在使用眼镜的情况下依然保持稳健。这些发现支持在现有CPU硬件上部署混合量子方法进行面部识别的可行性。
cs.CV / 66 / 2609.31957
CaptchaArena: A Large-Scale, Fine-Grained Dataset for Training Computer-Use Agents on Interactive CAPTCHAs
CaptchaArena：用于训练计算机使用代理的交互式 CAPTCHA 的大规模细粒度数据集
Abstract
Interactive CAPTCHAs remain challenging for computer-use agents, while existing datasets face trade-offs among type coverage, interaction fidelity, and trajectory supervision. To address these gaps, we present CaptchaArena, the first large-scale, fine-grained training dataset for interactive CAPTCHA solving. It contains 50K puzzles across 20 CAPTCHA types and 5 interaction modes, with every solution verified through execution. CaptchaArena provides 50K screenshot-action trajectories, including 46K with step-by-step reasoning annotations. It also includes fine-grained pixel-mask annotations for irregular targets. Using CaptchaArena, we train CaptchaAgent, a single 9B policy for all 20 CAPTCHA types, with supervised fine-tuning followed by reinforcement learning. The environment verifier directly provides the RL reward. Supervised fine-tuning reaches 70.5 Pass@1, and reinforcement learning further improves it to 71.7, while also improving performance on two external benchmarks. These results demonstrate the value of large-scale, fine-grained computer-use supervision for training interactive CAPTCHA agents. We release CaptchaArena and CaptchaAgent at https://github.com/X0X0X00/CaptchaArena.
Chinese Translation
交互式 CAPTCHA 对计算机使用代理仍然具有挑战性，而现有数据集在类型覆盖、交互真实性和轨迹监督之间存在权衡。为了解决这些问题，我们提出了 CaptchaArena，这是第一个大规模、细粒度的交互式 CAPTCHA 解题训练数据集。它包含 50,000 个谜题，涵盖 20 种 CAPTCHA 类型和 5 种交互模式，每个解答都经过执行验证。CaptchaArena 提供了 50,000 个屏幕截图-动作轨迹，其中 46,000 个具有逐步推理注释。它还包括不规则目标的细粒度像素掩码注释。利用 CaptchaArena，我们训练了 CaptchaAgent，这是一个针对所有 20 种 CAPTCHA 类型的单一 90 亿参数策略，经过监督微调后再进行强化学习。环境验证器直接提供 RL 奖励。监督微调达到 70.5 的 Pass@1，而强化学习进一步将其提高到 71.7，同时也提升了在两个外部基准上的表现。这些结果证明了大规模、细粒度的计算机使用监督在训练交互式 CAPTCHA 代理中的价值。我们在 https://github.com/X0X0X00/CaptchaArena 发布了 CaptchaArena 和 CaptchaAgent。
cs.CV / 67 / 2609.31970
Double-Edged Sword of Mediated Visibility: How Visual Framing Undermines Congresswomen's Perceived Competence
媒介可见性的双刃剑：视觉框架如何削弱女国会议员的能力感知
Abstract
Women's presence in Congress has stalled below 30%. That institutional underrepresentation is mirrored by their limited visibility on televised news, where visual presentation can shape perceived competence and authority. Yet research tells us more about whether congresswomen appear than how they are visually framed. Facial recognition analysis of 695,464 image-text segments from CNN and Fox News (2011-2021) reveals that congresswomen appear disproportionately in split-screen rather than solo shots. Two pre-registered experiments with 6,220 participants show that, in static frames, split-screen framing reduces congresswomen's perceived political competence, but not congressmen's. In dynamic videos, the competence penalty disappears; instead, outraged language reduces a congresswoman's perceived warmth roughly three times as much as a congressman's. Suggestive evidence (p = .057) indicates that women viewers report greater external political efficacy after watching a congresswoman appear alone. We conclude that visual framing shapes both congresswomen's mediated visibility and their perceived capabilities.
Chinese Translation
女性在国会中的比例停滞在30%以下。这种制度性代表性不足在电视新闻中的可见性有限，视觉呈现能够影响人们对能力和权威的感知。然而，研究更多关注女国会议员的出现频率，而非她们的视觉框架。对2011年至2021年间CNN和Fox News的695,464个图像-文本片段进行的面部识别分析显示，女国会议员在分屏展示中出现的比例明显高于单独镜头。两项预注册实验共涉及6,220名参与者，结果表明，在静态画面中，分屏框架降低了女国会议员的政治能力感知，而对男国会议员则没有影响。在动态视频中，能力惩罚消失；相反，愤怒的语言使女国会议员的温暖感知降低的程度大约是男国会议员的三倍。暗示性证据（p = .057）表明，女性观众在观看女国会议员单独出现后报告的外部政治效能感更强。我们得出结论，视觉框架不仅影响女国会议员的媒介可见性，还影响她们的能力感知。
cs.CV / 68 / 2609.31985
Does Vision-Language Pretraining Granularity Matter? A Controlled Evaluation of Vision-Language Objectives Across Chest X-Ray Interpretation Tasks
视觉-语言预训练粒度是否重要？对胸部X光解读任务中视觉-语言目标的控制评估
Abstract
Vision-language pretraining objectives differ in the spatial granularity of their supervision, yet the implications of this distribution for chest X-ray interpretation remain underexplored. We present a controlled study that isolates the pretraining objective: holding the encoder and pretraining data fixed, we train nine objectives spanning global and local contrastive learning, captioning, and their combinations, and evaluate across five chest X-ray tasks of increasing spatial granularity. We find that (i) pretraining granularity aligns with task granularity at the extremes, with local objectives leading on abnormality detection and global objectives on classification; (ii) local objectives are surprisingly competitive on global-level generation and question answering tasks; (iii) the merits of captioning and contrastive learning reverse across granularity levels; and (iv) among combinations, mixing captioning and contrastive supervision is strongest on classification and in distribution generation, while pairing two captioning objectives generalizes best on zero-shot report generation. These results show that no single objective is universally optimal, and that the interaction of objective type, granularity, and task governs downstream performance.
Chinese Translation
视觉-语言预训练目标在监督的空间粒度上存在差异，但这种分布对胸部X光解读的影响仍未得到充分探讨。我们提出了一项控制研究，隔离预训练目标：在固定编码器和预训练数据的情况下，我们训练了九个目标，涵盖全球和局部对比学习、描述生成及其组合，并在五个逐渐增加空间粒度的胸部X光任务中进行评估。我们发现：（i）预训练粒度在极端情况下与任务粒度一致，局部目标在异常检测上表现优越，而全球目标在分类上表现更佳；（ii）局部目标在全球级生成和问答任务中表现出意外的竞争力；（iii）描述生成和对比学习的优劣在不同粒度水平上发生逆转；（iv）在组合目标中，混合描述生成和对比监督在分类和分布生成上表现最强，而将两个描述生成目标配对在零样本报告生成上具有最佳泛化能力。这些结果表明，没有单一的目标是普遍最优的，目标类型、粒度和任务之间的相互作用决定了下游性能。
cs.CV / 69 / 2609.31998
Type-Balanced Federated Learning for Visual Analog Meter Reading
面向视觉模拟表计读数的类型平衡联邦学习
Abstract
Analog dial meters are widely deployed in industrial applications and utility sites, where environments and meter types vary and inspection data may be sensitive. Currently, automatic meter readers must be individually developed and deployed for each environment and meter type in practice. Deep learning could handle this variability but requires diverse labeled data that are costly to collect and update. In practice, meter images are distributed across independent sites, each with limited labels, while raw images often cannot be pooled because of ownership, governance, or privacy constraints. To address these challenges, we present a federated framework for visual analog meter reading that enables multiple sites to collaboratively train a reading model without sharing their raw images. Our framework consists of a four-stage pipeline: (1) dial localization, (2) thin-structure segmentation trained federatively across clients, (3) polar unwrapping, and (4) tick-counting decoding for final reading. To enable systematic evaluation of this setting, we release MeterFL, a 1,382-image mask-annotated dataset organized into deployment-motivated pseudo-clients derived from visual attributes via deterministic rules, with dHash near-duplicate control between the segmentation train and test splits. We evaluate both segmentation quality and end-to-end reading accuracy. MeterFL is publicly available at https://github.com/weidazhaoooo/Meter-FL.
Chinese Translation
模拟拨盘表计广泛应用于工业和公用事业现场，这些环境和表计类型各异，且检查数据可能敏感。目前，自动表计读取器在实践中必须针对每种环境和表计类型单独开发和部署。深度学习能够处理这种变异性，但需要多样化的标注数据，而这些数据的收集和更新成本高昂。在实践中，表计图像分布在独立的站点上，每个站点的标注数量有限，而由于所有权、治理或隐私限制，原始图像通常无法集中处理。为了解决这些挑战，我们提出了一种用于视觉模拟表计读数的联邦框架，使多个站点能够在不共享原始图像的情况下协同训练读取模型。我们的框架由四个阶段组成：(1) 拨盘定位，(2) 薄结构分割，跨客户端进行联邦训练，(3) 极坐标展开，和 (4) 刻度计数解码以获得最终读数。为了系统评估这一设置，我们发布了MeterFL，这是一个包含1,382张图像的掩码标注数据集，按照视觉属性通过确定性规则组织成以部署为动机的伪客户端，并在分割训练和测试拆分之间进行dHash近重复控制。我们评估了分割质量和端到端的读取准确性。MeterFL可在https://github.com/weidazhaoooo/Meter-FL公开获取。
cs.CV / 70 / 2609.32013
TriO: Tri-Modal Unsupervised Occupancy World Model for Anything Perception
TriO：用于任何感知的三模态无监督占用世界模型
Abstract
We present TriO, a multi-modal unsupervised world model that predicts 4D occupancy, obstacle segmentation, flow and LiDAR. In contrast to prior work, TriO utilizes three distinct sensor modalities (camera, LiDAR, and RADAR) as both inputs and sources of self-supervision, eliminating the need for additional human annotations. Thanks to its novel supervision, the model is able to segment any occupancy from the drivable surface, overcoming the limitations of existing open-set methods in handling long-tail objects. TriO achieves state-of-the-art results in multiple 3D and 4D tasks, including occupancy, flow, and LiDAR prediction, as well as zero-shot road obstacle segmentation across multiple datasets such as Argoverse 2, and Spotting the Unexpected.
Chinese Translation
我们提出了TriO，一种多模态无监督世界模型，能够预测4D占用、障碍物分割、流动和激光雷达（LiDAR）。与之前的工作相比，TriO利用三种不同的传感器模态（相机、激光雷达和雷达）作为输入和自我监督的来源，消除了对额外人工标注的需求。得益于其新颖的监督方式，该模型能够从可行驶表面中分割出任何占用，克服了现有开放集方法在处理长尾物体时的局限性。TriO在多个3D和4D任务中取得了最先进的结果，包括占用、流动和激光雷达预测，以及在多个数据集（如Argoverse 2和Spotting the Unexpected）上进行零样本道路障碍物分割。
cs.CV / 71 / 2609.32027
Depth Any Seen: Which Surfaces and How Far?
深度任意可见：哪些表面以及多远？
Abstract
When several surfaces are visible along a ray, recovering visible 3D structure from one image requires jointly estimating their presence and metric depth. Depth Any Seen represents these surfaces as image-conditioned multi-Bernoulli depth sets, whose components each contribute one depth or remain absent. Its auxiliary-free Exact Multi-Bernoulli objective (ExactMB) learns depth and presence by marginalizing one-to-one assignments to complete, distinct targets. Our analysis shows that matching expected count can leave component-surface assignment unresolved. We extend real and synthetic layered-depth benchmarks to evaluate depth accuracy, recovered support, and overprediction. Compared to depth stacking, ExactMB reduces overprediction by a relative 88.2% on LD-Real and 80.5% on MD-3K while retaining most ordinal accuracy, with comparable conditional metric-depth error on LD-Syn. Further ablation studies show that ordered assignment improves depth-accurate recall and precision over marginalization, whereas the count-regularized configuration achieves higher deeper-rank precision than ordered assignment at lower recall. Our code will be publicly released.
Chinese Translation
当沿着一条光线可见多个表面时，从一幅图像中恢复可见的三维结构需要共同估计它们的存在性和度量深度。深度任意可见将这些表面表示为图像条件的多伯努利深度集，其组成部分每个贡献一个深度或保持缺失。其无辅助的精确多伯努利目标（Exact Multi-Bernoulli，ExactMB）通过边际化一对一分配来学习深度和存在性，以完成不同的目标。我们的分析表明，匹配预期计数可能导致组件-表面分配未解决。我们扩展了真实和合成的分层深度基准，以评估深度准确性、恢复支持和过度预测。与深度堆叠相比，ExactMB在LD-Real上将过度预测减少了相对88.2%，在MD-3K上减少了80.5%，同时保持了大部分序数准确性，并在LD-Syn上具有可比的条件度量深度误差。进一步的消融研究表明，有序分配在边际化的基础上提高了深度准确的召回率和精确度，而计数正则化配置在较低召回率下实现了比有序分配更高的深层排名精确度。我们的代码将公开发布。
cs.CV / 72 / 2609.32036
ScreenHaystack: Finding Blind Zones in GUI Grounding
ScreenHaystack：在GUI定位中发现盲区
Abstract
We introduce ScreenHaystack, a dynamic needle-in-a-haystack benchmark for evaluating spatial reliability in GUI grounding. Instead of testing each target at a fixed position, ScreenHaystack systematically relocates controlled target icons across high-resolution GUI backgrounds and measures whether models can localize them consistently. Using this benchmark, we find that leading GUI grounding models, including Qwen3-VL, UI-TARS, GTA, and UI-Venus, exhibit blind zones: spatial regions where grounding accuracy drops sharply despite fixed target appearance and instruction. These blind zones transfer to unseen ScreenSpot-Pro examples: targets inside blind zones are consistently harder to ground, with Qwen3-VL-8B dropping by 16.1 percentage points, and controlled relocation shows that moving targets into blind zones decreases accuracy while moving them out improves accuracy. We further show through controlled synthetic experiments that uneven spatial coverage in training data can induce such blind zones. Therefore, we propose a simple strategy, blind-zone-oriented augmentation, which adds supervision in blind zones and improves ScreenSpot-Pro accuracy over both original and randomly augmented Click-100k fine-tuning.
Chinese Translation
我们介绍了ScreenHaystack，这是一个动态的“干草堆中的针”基准，用于评估GUI定位中的空间可靠性。与其在固定位置测试每个目标，ScreenHaystack系统地将受控目标图标重新定位于高分辨率的GUI背景中，并测量模型是否能够一致地定位它们。通过这个基准，我们发现领先的GUI定位模型，包括Qwen3-VL、UI-TARS、GTA和UI-Venus，表现出盲区：尽管目标外观和指令固定，定位准确性在这些空间区域内急剧下降。这些盲区在未见过的ScreenSpot-Pro示例中也会转移：盲区内的目标始终更难以定位，Qwen3-VL-8B的准确性下降了16.1个百分点，而受控重新定位显示，将目标移入盲区会降低准确性，而将其移出则提高准确性。我们进一步通过受控的合成实验表明，训练数据中的不均匀空间覆盖可以导致此类盲区。因此，我们提出了一种简单的策略，盲区导向增强，旨在在盲区内增加监督，从而提高ScreenSpot-Pro在原始和随机增强的Click-100k微调上的准确性。
cs.CV / 73 / 2609.32038
ControlGS: Conditioning Neural Gaussians for Downstream-Processing-Aware XR Rendering
ControlGS：为下游处理感知的神经高斯渲染控制
Abstract
Extended Reality (XR) users do not directly perceive the output of a rendering engine. Instead, rendered images pass through a post-processing pipeline and the physical display-optics path before reaching the eye. Critically, the exact downstream processing can vary significantly at run time, influenced by, for instance, camera pose and display power budget. Traditional 3DGS methods either implicitly assume that this downstream pipeline preserves image quality or cannot adapt to downstream processing changes. To bridge this gap, we present ControlGS, an XR Gaussian rendering pipeline that optimizes end-to-end visual quality. ControlGS models and integrates the entire downstream processing, between the rendering output and the human eye, into the optimization objective. To adapt to downstream processing at run time, ControlGS dynamically generates Gaussian primitives conditioned upon the downstream processing parameters. Experiments show that ControlGS consistently improves end-to-end post-optics XR quality across different neural Gaussian backbones and datasets, with minimal overhead. Code is available at https://horizon-lab.org/controlgs/.
Chinese Translation
扩展现实（XR）用户并不直接感知渲染引擎的输出。相反，渲染的图像在到达眼睛之前会经过后处理管道和物理显示光学路径。关键是，确切的下游处理在运行时可能会显著变化，例如受到相机姿态和显示功率预算的影响。传统的3DGS方法要么隐含假设该下游管道保持图像质量，要么无法适应下游处理的变化。为了解决这一问题，我们提出了ControlGS，这是一种优化端到端视觉质量的XR高斯渲染管道。ControlGS将渲染输出与人眼之间的整个下游处理建模并整合到优化目标中。为了在运行时适应下游处理，ControlGS根据下游处理参数动态生成条件高斯原语。实验表明，ControlGS在不同的神经高斯骨干网和数据集上始终提高端到端后光学XR质量，且开销最小。代码可在 https://horizon-lab.org/controlgs/ 获取。
cs.CV / 74 / 2609.32041
Amnesia by Design, Memory By Necessity: Persistent State for Document Intelligence
设计中的遗忘，必要的记忆：文档智能的持久状态
Abstract
Modern Document AI reads contracts, extracts fields, reasons over tables, and grounds answers to page regions, then forgets everything. Processing an amendment the next day begins from scratch: no schema retained, no contradiction detected, no experience carried forward. This is a structural choice, not a scale failure: current systems are stateless functions. We call this the statelessness bottleneck. This bottleneck lies beyond parameter scaling, context extension, and retrieval augmentation: storage provides persistence and retrieval provides access, but neither consolidates observations into knowledge that improves future processing. This survey formalizes persistent evidence-grounded document state as a unifying framework, specifying the operations and invariants required to convert multimodal evidence into durable, provenance-linked state. We introduce a statefulness audit showing that ten representative benchmarks, coded against eight statefulness criteria, leave cross-session state evolution untested, and derive a longitudinal benchmark harness with five counterfactual metrics: Experience Gain, Cost Efficiency, Memory Harm, Forgetting Fidelity, Coverage Retention, to characterize the benefit, cost, risk, and governability of persistent document state. Document AI lacks mechanisms coupling persistent state to document-native structure, provenance, and temporal validity. The next era of Document AI will be defined by what systems retain across documents, sessions, and time.
Chinese Translation
现代文档人工智能（Document AI）读取合同，提取字段，对表格进行推理，并将答案与页面区域关联，然后忘记一切。处理第二天的修正案时从头开始：没有保留模式，没有检测到矛盾，没有经验得以延续。这是一种结构性选择，而非规模失败：当前系统是无状态函数。我们称之为无状态瓶颈。这个瓶颈超越了参数扩展、上下文延伸和检索增强：存储提供持久性，检索提供访问，但两者都无法将观察整合为改善未来处理的知识。本调查将持久的证据基础文档状态形式化为一个统一框架，指定将多模态证据转换为耐用、来源关联状态所需的操作和不变性。我们引入了一种状态审计，显示十个代表性基准，依据八个状态性标准进行编码，未测试跨会话状态演变，并推导出一个具有五个反事实指标的纵向基准工具：经验增益、成本效率、记忆损害、遗忘忠诚度、覆盖保留，以表征持久文档状态的益处、成本、风险和可治理性。文档人工智能缺乏将持久状态与文档本土结构、来源和时间有效性相结合的机制。文档人工智能的下一个时代将由系统在文档、会话和时间中保留的内容所定义。
cs.CV / 75 / 2609.32068
ReFM: Semantic-Aware Refinement Flow Model for Motion Retargeting
ReFM：用于动作重定向的语义感知精炼流模型
Abstract
Motion retargeting transfers motion across characters with different skeletal structures while preserving semantic intent and physical plausibility. Despite recent progress, two fundamental questions remain: (i) how can reliable source-motion semantics be learned without high-quality paired retargeting data, and (ii) how should retargeting be formulated when no reliable paired motion can serve as a definitive regression objective? Existing methods commonly preserve semantics by constraining predictions toward copied motions. However, such initializations entangle useful articulation cues with artifacts caused by mismatched skeletal proportions and body geometry. Moreover, directly regressing a final motion in one forward pass is restrictive because retargeting is inherently underdetermined, and the desired solution must balance semantic fidelity with target-specific physical and temporal constraints rather than match a unique paired target. Motivated by these limitations, we propose ReFM, a source-mesh-agnostic, energy-guided model that reformulates motion retargeting as progressive refinement. First, an SO(3) canonicalizer removes redundant global-orientation variations. Second, a cross-character semantic encoder, pretrained through contrastive learning, provides a character-invariant representation for both optimization guidance and semantic evaluation. ReFM then progressively refines an initialized target motion through a learned flow guided by semantic consistency, physical plausibility, temporal coherence, and minimal motion modification. The framework is compatible with different initialization strategies, including both direct motion copying and Autodesk HumanIK, an industry-standard full-body inverse-kinematics retargeting system.
Chinese Translation
动作重定向在不同骨骼结构的角色之间转移动作，同时保持语义意图和物理合理性。尽管近期取得了一些进展，但仍然存在两个基本问题：（i）如何在没有高质量配对重定向数据的情况下学习可靠的源动作语义，以及（ii）当没有可靠的配对动作可以作为明确的回归目标时，重定向应如何被表述？现有方法通常通过将预测约束到复制的动作上来保持语义。然而，这种初始化将有用的关节提示与由于骨骼比例和身体几何不匹配而造成的伪影纠缠在一起。此外，在一次前向传递中直接回归最终动作是有限制的，因为重定向本质上是欠定的，所需的解决方案必须在语义保真度与特定目标的物理和时间约束之间取得平衡，而不是匹配一个独特的配对目标。基于这些局限性，我们提出了ReFM，这是一种源网格无关的、能量引导的模型，将动作重定向重新表述为渐进精炼。首先，SO(3)标准化器去除了冗余的全局方向变化。其次，通过对比学习预训练的跨角色语义编码器为优化指导和语义评估提供了角色不变的表示。然后，ReFM通过一个学习的流，逐步精炼初始化的目标动作，该流受到语义一致性、物理合理性、时间连贯性和最小动作修改的引导。该框架与不同的初始化策略兼容，包括直接动作复制和Autodesk HumanIK，这是一种行业标准的全身逆运动学重定向系统。
cs.CV / 76 / 2609.32131
Gaussian Image Steganography via Parameter-Domain Keyed Embeddings
基于高斯参数域的图像隐写术
Abstract
2D Gaussian-based image representation is becoming increasingly popular, and our work proposes a new approach to steganography by embedding information within Gaussian parameters rather than image pixels. We first fit the parameters of this representation to the target image and employ a secret key to select a subset of these parameters for fine-tuning, allowing us to embed an 8-bit message while maintaining high visual fidelity in the reconstructed image. Thirty fitting experiments on three synthetic images show that the correct key can recover the message without error, while decoding with incorrect keys yields a Bit Error Rate (BER) of $0.543$, close to random guessing. Compared with random selection using the key, selecting the least-disturbing edits recovers the message more reliably (one-sided $p=0.031$), and the average PSNR cost is only $0.091$ dB in visual quality. The embedding method transfers to 112 natural images at $256\times256$ using 4,096 Gaussians. The correct-key BER is $0.000$, and decoding under a wrong key stays close to random guessing at $0.520$. Our method embeds the payload through three Gaussian parameter types: log-anisotropy, opacity, and color luminance. In a separate nine-fit reduced setting, color luminance is removed, so the payload uses two instead of three parameter types, a $33.3\%$ reduction; all 256 Gaussians remain in the fitted representation. The correct key still recovers the message without error. However, wrong-key BER rises from $0.514$ to $0.571$, moving farther from random guessing ($0.5$).
Chinese Translation
基于二维高斯的图像表示正变得越来越流行，我们的工作提出了一种新的隐写术方法，通过在高斯参数中嵌入信息，而不是图像像素。我们首先将这种表示的参数拟合到目标图像，并使用一个秘密密钥选择这些参数的子集进行微调，从而在保持重建图像高视觉保真度的同时嵌入一个8位消息。在对三幅合成图像进行的三十次拟合实验中，正确的密钥能够无误地恢复消息，而使用错误密钥解码的比特错误率（BER）为$0.543$，接近随机猜测。与使用密钥进行随机选择相比，选择最小干扰的编辑能够更可靠地恢复消息（单侧$p=0.031$），且在视觉质量上平均PSNR成本仅为$0.091$ dB。该嵌入方法在$256 imes256$的112幅自然图像中转移，使用了4096个高斯。正确密钥的BER为$0.000$，而在错误密钥下的解码接近随机猜测，BER为$0.520$。我们的方法通过三种高斯参数类型嵌入负载：对数各向异性、透明度和颜色亮度。在一个单独的九次拟合减少设置中，去除了颜色亮度，因此负载使用两种而不是三种参数类型，减少了$33.3\%$；所有256个高斯仍保留在拟合表示中。正确的密钥仍然能够无误地恢复消息。然而，错误密钥的BER从$0.514$上升到$0.571$，与随机猜测的距离进一步加大（$0.5$）。
cs.CV / 77 / 2609.32157
CausalDriveBench: Evaluating Causal Reasoning in Vision-Language-Action Models for Autonomous Driving
CausalDriveBench：评估自主驾驶中视觉-语言-行动模型的因果推理
Abstract
Vision-Language-Action (VLA) models for autonomous driving produce natural-language reasoning alongside predicted trajectories, but whether this reasoning reflects the causal structure of the scene remains untested. We introduce CausalDriveBench, an evaluation framework grounded in Pearl's Causal Hierarchy (PCH) that tests causal reasoning in driving-specific VLAs through structured visual question answering (QA) and alternative-trajectory prediction. To this end, we construct causal scene graphs over nuScenes that distinguish causally active, dormant, and distractor entities, separating perceptual salience from causal relevance. The benchmark spans all four rungs of PCH (association, intervention, and counterfactual along with causal discovery) for QA generation. For the higher rungs, we additionally provide reference trajectories under specified scene modifications, enabling action-level verification that complements reasoning-level evaluation. In total, the benchmark contains 7,285 verified causal QA pairs and 1,000 counterfactual trajectories derived from nuScenes. We evaluate 10 driving-specific VLAs and 3 general-purpose VLMs, and report three findings. First, the best model reaches only 70.6% QA accuracy, and 4 of 13 models score below random chance. Second, comparing each driving VLA to the general-purpose VLM that shares its language backbone, the cost of driving fine-tuning ranges from 2 to 34 percentage points on causal QA, with post-training design explaining the spread. Third, causal QA and trajectory accuracy are statistically uncorrelated across models: under counterfactual prompts, predicted trajectories either over-react or collapse onto the observed-scene baseline. Taken together, these results show that neither fluent rationales nor accurate observed-scene trajectories constitute evidence of causal understanding.
Chinese Translation
自主驾驶的视觉-语言-行动（VLA）模型在预测轨迹的同时生成自然语言推理，但这种推理是否反映场景的因果结构尚未经过检验。我们引入了CausalDriveBench，一个基于Pearl的因果层级（PCH）的评估框架，通过结构化的视觉问答（QA）和替代轨迹预测来测试驾驶特定VLA的因果推理。为此，我们在nuScenes上构建了因果场景图，区分因果活跃、休眠和干扰实体，将感知显著性与因果相关性分开。该基准涵盖了PCH的四个层级（关联、干预、反事实及因果发现）用于QA生成。对于较高层级，我们还提供了在特定场景修改下的参考轨迹，使得行动级别的验证能够补充推理级别的评估。总的来说，该基准包含7,285对经过验证的因果QA对和1,000条来自nuScenes的反事实轨迹。我们评估了10个驾驶特定的VLA和3个通用VLM，并报告了三个发现。首先，最佳模型的QA准确率仅为70.6%，而13个模型中有4个得分低于随机机会。其次，将每个驾驶VLA与共享语言基础的通用VLM进行比较，驾驶微调的成本在因果QA上范围从2到34个百分点，后训练设计解释了这种差异。第三，因果QA和轨迹准确性在模型之间统计上不相关：在反事实提示下，预测轨迹要么过度反应，要么收敛到观察场景基线。综合来看，这些结果表明，流畅的推理或准确的观察场景轨迹都不能构成因果理解的证据。
cs.CV / 78 / 2609.32162
PruneForget: Joint Unlearning and Pruning of Vision Models
PruneForget：视觉模型的联合遗忘与剪枝
Abstract
Machine unlearning and model pruning are increasingly coupled in the real world. Models must support unlearning requests, e.g., for safety concerns, while also meeting requirements in latency and memory budget. Until recently, existing works have studied each aspect as an independent problem, e.g., running unlearning and pruning sequentially. In this work, we show that unlearning and pruning are naturally aligned and should be solved jointly to be made aware of each other. Intuitively, parameters that encode information of the unlearned samples are natural pruning targets, as unlearning and pruning both call for the "deletion" of such parameters. We propose PruneForget, a method that uses the unlearn set as a guide for pruning, so that unlearning and pruning mutually benefit each other. Extensive experiments on image classifiers and generative models show that PruneForget removes the influence of the unlearned samples while producing a more compact model with reduced inference cost. It achieves a negligible performance gap relative to an oracle that retrains from scratch for unlearning and then prunes.
Chinese Translation
机器遗忘与模型剪枝在现实世界中越来越紧密地结合在一起。模型必须支持遗忘请求，例如出于安全考虑，同时还需满足延迟和内存预算的要求。直到最近，现有研究将每个方面视为独立问题进行研究，例如顺序运行遗忘和剪枝。在本研究中，我们表明遗忘与剪枝自然是相互关联的，应当联合解决以相互感知。直观上，编码遗忘样本信息的参数是自然的剪枝目标，因为遗忘和剪枝都要求“删除”这些参数。我们提出了PruneForget，一种利用遗忘集作为剪枝指导的方法，使遗忘与剪枝相互促进。在图像分类器和生成模型上的大量实验表明，PruneForget在去除遗忘样本影响的同时，生成了一个更紧凑的模型，并降低了推理成本。相较于从头开始重训练以进行遗忘然后再进行剪枝的理想情况，它实现了微不足道的性能差距。
cs.CV / 79 / 2609.32163
PQR3D: Progressive Query Refinement over Reference-Conditioned Temporal Windows for Multi-View 3D Object Detection
PQR3D：基于参考条件时间窗口的多视角3D物体检测中的渐进查询细化
Abstract
Temporal context is essential for camera-only multi-view 3D object detection. Existing streaming detectors maintain and propagate query states from one frame to the next, requiring sequence-aware training and chronological inference. We propose PQR3D, which performs progressive query refinement within referenceconditioned temporal windows. This design enables random frame sampling and independent inference without persistent query memory. Within each window, PQR3D progressively transfers motion-aligned high-confidence queries from earlier timestamps toward the target frame. We further introduce masked selfattention to regulate interactions among regular, propagated, and denoising queries while keeping denoising supervision isolated from detection queries. In addition, a stage-decoupled anchor embedding injects position before self-attention and size, orientation, and velocity afterward, reducing interference from temporally inconsistent attributes. With a ViT-L backbone, PQR3D sets a new state of the art on the nuScenes test set, achieving 71.6 NDS and 64.9 mAP. Source code is available at https://github.com/huiyegit/PQR3D
Chinese Translation
时间上下文对于仅使用相机的多视角3D物体检测至关重要。现有的流式检测器在帧与帧之间维护并传播查询状态，这要求进行序列感知训练和时间顺序推理。我们提出了PQR3D，它在参考条件时间窗口内执行渐进查询细化。该设计支持随机帧采样和独立推理，而无需持久的查询记忆。在每个窗口内，PQR3D逐步将早期时间戳中的运动对齐的高置信度查询转移到目标帧。我们进一步引入了掩蔽自注意力机制，以调节常规查询、传播查询和去噪查询之间的交互，同时保持去噪监督与检测查询的隔离。此外，阶段解耦的锚嵌入在自注意力之前注入位置信息，并在之后注入大小、方向和速度，从而减少时间不一致属性的干扰。使用ViT-L主干，PQR3D在nuScenes测试集上设定了新的最先进水平，达到了71.6的NDS和64.9的mAP。源代码可在https://github.com/huiyegit/PQR3D获取。
cs.CV / 80 / 2609.32175
OneFixer: High-Quality and Consistent One-Step Autoregressive 3DGS Refinement for Driving Scenes
OneFixer：高质量且一致的一步自回归3D高斯点云（3DGS）驾驶场景精细化
Abstract
Autoregressive video diffusion is a promising render-time fixer for 3D Gaussian Splatting (3DGS) in autonomous-driving simulation, but deployment demands high visual quality and temporal consistency at low latency. This is especially hard for one-step causal generation, where each imperfect prediction immediately becomes context for subsequent frames. Existing approaches stabilize rollouts through staged training with multiple modules and rollout-aware regularization, yet one-step quality still falls short of what deployment requires. We introduce OneFixer, a one-step autoregressive video-diffusion fixer trained in a single task-specific adaptation stage. Our key idea is a deployment-matched shared rollout: the model's own one-step predictions serve as the causal context for flow matching, exposing training to deployment-time errors, while the same rollout receives direct pixel-space perceptual supervision to preserve fine detail. Because the predictions optimized for current-frame quality are exactly those reused as future context, fidelity and autoregressive robustness are learned jointly, without bidirectional-to-causal conversion or teacher-student distillation. OneFixer further exploits cues that driving simulation readily provides, lane geometry and dynamic-agent states, to improve geometric fidelity. On Waymo and proprietary driving scenes with 900-frame rollouts, OneFixer achieves the lowest FVD, LPIPS, and DISTS among all baselines at one step, with temporal consistency matching or exceeding multi-stage DMD pipelines. Under identical backbone and conditioning, it matches a multi-stage DMD-with-Self-Forcing pipeline in under half the GPU-hours and keeps improving beyond its plateau. In closed-loop simulation with a driving policy, OneFixer reduces the collision rate by a third relative to raw 3DGS rendering. Project page: https://onefixer-web.vercel.app/
Chinese Translation
自回归视频扩散是一种有前景的渲染时间修复器，用于自主驾驶模拟中的3D高斯点云（3DGS），但部署要求在低延迟下实现高视觉质量和时间一致性。这对于一步因果生成尤其困难，因为每个不完美的预测立即成为后续帧的上下文。现有方法通过多模块的分阶段训练和滚动预测感知正则化来稳定生成过程，但一步的质量仍然未能达到部署的要求。我们提出了OneFixer，一种在单一任务特定适应阶段训练的一步自回归视频扩散修复器。我们的关键思想是与部署相匹配的共享滚动：模型自身的一步预测作为流匹配的因果上下文，使训练暴露于部署时的错误，同时相同的滚动接收直接的像素空间感知监督，以保留细节。由于为当前帧质量优化的预测正是作为未来上下文重复使用的，因此保真度和自回归鲁棒性是共同学习的，无需双向到因果的转换或教师-学生蒸馏。OneFixer进一步利用驾驶模拟提供的线道几何和动态代理状态等线索，以提高几何保真度。在包含900帧滚动的Waymo和专有驾驶场景中，OneFixer在一步中实现了所有基线中最低的FVD、LPIPS和DISTS，时间一致性与多阶段DMD管道相匹配或超过。在相同的骨干网络和条件下，它在不到一半的GPU小时内匹配了多阶段DMD与自我强制管道，并在其平台上持续改进。在与驾驶策略的闭环模拟中，OneFixer将碰撞率相对于原始3DGS渲染降低了三分之一。项目页面：https://onefixer-web.vercel.app/
cs.CV / 81 / 2609.32177
Federated 3D Gaussian Splatting for Large-Scale Scene Reconstruction at Wireless Edge
无线边缘的大规模场景重建的联邦三维高斯点云技术
Abstract
Three-dimensional (3D) Gaussian splatting (3D-GS) has emerged as a promising technique for large-scale scene reconstruction due to its high rendering efficiency and fidelity. However, the training of large-scale 3D-GS models at wireless edge faces various technical challenges including the limited communication, computation, and graphics processing unit (GPU) memory resources at edge devices, the structural inconsistency issue across local models hindering their effective aggregation, as well as privacy leakage risks associated with raw visual content and camera parameters. To address these challenges, this paper proposes a novel resource-efficient federated learning framework for efficiently training 3D-GS models of large scenes under severe resource constraints. First, we propose an on-device model lightweighting mechanism that adaptively selects and prunes Gaussian points to balance the rendering quality and training efficiency. In this mechanism, we quantitatively evaluate the importance of different Gaussian points at each device to facilitate the pruning, and use a novel importance-to-latency ratio criterion to determine the number of pruned Gaussian points under GPU memory and computation/communication latency constraints. Furthermore, we develop a 3D-GS model recovery mechanism that restores structural consistency across local 3D-GS models without accessing private camera parameters, enabling their effective aggregation towards a global model. Finally, extensive experiments show that our approach significantly accelerates convergence, maintains high rendering quality, and reduces training latency compared to state-of-the-art federated 3D-GS baselines.
Chinese Translation
三维高斯点云技术（3D Gaussian Splatting, 3D-GS）因其高渲染效率和高保真度而成为大规模场景重建的有前景的技术。然而，在无线边缘训练大规模3D-GS模型面临多种技术挑战，包括边缘设备上有限的通信、计算和图形处理单元（GPU）内存资源、局部模型之间的结构不一致性问题妨碍其有效聚合，以及与原始视觉内容和相机参数相关的隐私泄露风险。为了解决这些挑战，本文提出了一种新颖的资源高效的联邦学习框架，以在严峻的资源限制下高效训练大场景的3D-GS模型。首先，我们提出了一种设备端模型轻量化机制，该机制自适应地选择和修剪高斯点，以平衡渲染质量和训练效率。在该机制中，我们定量评估每个设备上不同高斯点的重要性，以促进修剪，并使用一种新颖的重要性与延迟比标准来确定在GPU内存和计算/通信延迟限制下修剪的高斯点数量。此外，我们开发了一种3D-GS模型恢复机制，该机制在不访问私有相机参数的情况下恢复局部3D-GS模型之间的结构一致性，从而使其能够有效聚合为全局模型。最后，大量实验表明，与最先进的联邦3D-GS基线相比，我们的方法显著加速了收敛，保持了高渲染质量，并减少了训练延迟。
cs.CV / 82 / 2609.32180
Binaural Audio-Visual Instance Segmentation
双耳音频-视觉实例分割
Abstract
Audio-visual segmentation (AVS) aims to segment sounding objects at the pixel level by integrating auditory and visual cues. However, existing methods are predominantly developed under the monaural setting and primarily rely on cross-modal semantic correspondence, which limits their ability to distinguish visually similar instances of the same semantic class. In contrast, humans naturally exploit binaural hearing, where interaural differences and direction-dependent acoustic filtering introduced by the head and pinnae provide physically grounded spatial cues for accurate sound source localization. Motivated by this observation, we introduce binaural audio-visual instance segmentation (BiAVIS), a new task that leverages synchronized binaural audio and video frames to segment sounding instances. To advance research on this task, we establish two benchmarks by manually annotating an existing binaural audio-visual dataset and collecting a new real-world dataset, BiAVIS-Bench, in more challenging and diverse scenarios. We further propose a BiAVIS model, which leverages an audio-only sound source localization network to learn spatial and semantic priors for sounding instances from binaural audio. A query-level audio-visual fusion strategy is subsequently introduced to inject these informative priors into the instance segmentation decoder. Extensive experiments conducted on the two proposed benchmarks demonstrate the superior performance of the BiAVIS model over previous monaural AVS methods, especially in resolving instance-level intra-class ambiguity. On the more challenging BiAVIS-Bench, the proposed BiAVIS model outperforms the best-performing monaural baselines by 17.22\% in mAP and 7.71\% in FSLA, respectively.
Chinese Translation
音频-视觉分割（AVS）旨在通过整合听觉和视觉线索，在像素级别上对发声物体进行分割。然而，现有方法主要是在单耳设置下开发，主要依赖于跨模态语义对应，这限制了它们区分同一语义类别的视觉相似实例的能力。相比之下，人类自然利用双耳听觉，耳间差异和由头部及耳廓引入的方向依赖性声学过滤提供了物理基础的空间线索，以实现准确的声源定位。基于这一观察，我们引入了双耳音频-视觉实例分割（BiAVIS），这是一项利用同步的双耳音频和视频帧对发声实例进行分割的新任务。为了推动该任务的研究，我们通过手动标注现有的双耳音频-视觉数据集并在更具挑战性和多样化的场景中收集新的现实世界数据集BiAVIS-Bench，建立了两个基准。我们进一步提出了BiAVIS模型，该模型利用仅音频的声源定位网络，从双耳音频中学习发声实例的空间和语义先验。随后引入了一种查询级音频-视觉融合策略，将这些信息丰富的先验注入到实例分割解码器中。在两个提出的基准上进行的广泛实验表明，BiAVIS模型在解决实例级同类模糊性方面的表现优于之前的单耳AVS方法。在更具挑战性的BiAVIS-Bench上，所提出的BiAVIS模型在mAP上比表现最好的单耳基线高出17.22%，在FSLA上高出7.71%。
cs.CV / 83 / 2609.32182
KeyRec: Bounded Visual Memory for Streaming and Long-Video Understanding
KeyRec：用于流媒体和长视频理解的有限视觉记忆
Abstract
Vision-language models are increasingly used to understand long videos and continuous streams. However, dense visual tokens accumulate with video duration, making long-context inference prohibitively expensive. Existing training-free visual-token selection methods reduce this cost by retaining informative tokens, but may lose coherent event evidence and fail to distinguish detailed recent observations from long-range history. We propose \textbf{KeyRec}, a training-free framework for constructing bounded visual memory. During query-agnostic writing, KeyRec preserves fine-grained recent observations in a visual cache and organizes historical evidence into a structured event bank. Candidate events are proposed according to their novelty relative to previously stored events and maintained through an online add--merge--evict update. When a question arrives, a text-only router adaptively allocates a fixed readout budget between recent and event memory, without reprocessing historical frames. KeyRec operates on model-facing visual embeddings and supports both modular encoder--projector VLMs and the encoder- and projector-free NEO-ov architecture. Across four streaming and long-video benchmarks and three VLM backbones, KeyRec achieves the best compressed performance in 13 of 15 settings using only 10\% of the dense decoder-facing visual-token budget. It outperforms the strongest compressed baseline by 2.21--18.37 points on real-time questions, achieves the best compressed result in five of six long-video settings, and performs best in every NEO-ov 2B setting.
Chinese Translation
视觉-语言模型越来越多地被用于理解长视频和连续流。然而，随着视频时长的增加，密集的视觉标记会不断累积，使得长上下文推理的成本变得过于昂贵。现有的无训练视觉标记选择方法通过保留有信息的标记来降低这一成本，但可能会丧失连贯事件证据，并且无法区分详细的近期观察与长时间历史。我们提出了 extbf{KeyRec}，一个构建有限视觉记忆的无训练框架。在与查询无关的写入过程中，KeyRec 在视觉缓存中保留细粒度的近期观察，并将历史证据组织成结构化的事件库。候选事件根据其相对于先前存储事件的新颖性进行提议，并通过在线的添加-合并-驱逐更新进行维护。当问题到达时，文本路由器自适应地在近期记忆和事件记忆之间分配固定的读取预算，而无需重新处理历史帧。KeyRec 在面向模型的视觉嵌入上运行，并支持模块化编码器-投影器的视觉语言模型（VLMs）以及无编码器和投影器的NEO-ov架构。在四个流媒体和长视频基准测试以及三个VLM骨干网络中，KeyRec 在15个设置中有13个使用仅10%的密集解码器面向视觉标记预算实现了最佳压缩性能。它在实时问题上比最强的压缩基线提高了2.21至18.37分，在六个长视频设置中取得了五个最佳压缩结果，并在每个NEO-ov 2B设置中表现最佳。
cs.CV / 84 / 2609.32183
Scalable In-Domain Self-Supervised Foundation Model for Dense Representation Transfer in High-Resolution Plant Imaging
可扩展的领域内自监督基础模型用于高分辨率植物成像中的密集表示转移
Abstract
High-resolution plant imaging enables detailed characterization of plant morphology, but dense scientific analysis remains limited by costly pixel-level annotations, large image pixel dimensions, and substantial variation in imaging conditions. This work proposes a scalable in-domain self-supervised pretrained foundation model for high-resolution, high-pixel-dimension multi-species plant imagery. A masked autoencoder with a ViT backbone is pretrained on more than 10 million multi-view plant image tiles using distributed training. Following scalable pretraining, the learned foundation-model representations are comprehensively benchmarked across fine-grained dense prediction and coarse global feature recognition, with particular emphasis on limited supervision and realistic downstream imaging conditions. This work focuses on the domain gap of existing foundation models in dense feature representation and transfer. Through extensive experiments involving limited annotations, cross-view variation, and resolution degradation, the in-domain FM achieves a Mean Dice of 0.8686 and a Pooled Dice of 0.8959, outperforming an MAE counterpart pretrained on large-scale natural-image data by 0.0694 and 0.0613, respectively. The results further indicate that increasing pretraining scale produces consistent improvements in dense feature transfer. Overall, these findings suggest that scaling in-domain self-supervised pretraining can reduce the domain gap and improve transferable dense representations for high-pixel-dimension scientific imaging.
Chinese Translation
高分辨率植物成像能够详细描述植物形态，但由于昂贵的像素级注释、大量的图像像素维度以及成像条件的显著变化，密集科学分析仍然受到限制。本研究提出了一种可扩展的领域内自监督预训练基础模型，旨在处理高分辨率、高像素维度的多物种植物图像。该模型采用带有 ViT（Vision Transformer）主干的掩码自编码器，在超过 1000 万个多视角植物图像块上进行分布式训练的预训练。在可扩展的预训练之后，学习到的基础模型表示在细粒度密集预测和粗略全局特征识别方面进行了全面基准测试，特别强调了有限监督和现实下游成像条件下的表现。本研究关注现有基础模型在密集特征表示和转移中的领域差距。通过涉及有限注释、视角变化和分辨率降级的广泛实验，领域内的基础模型在平均 Dice 值上达到了 0.8686，汇总 Dice 值达到了 0.8959，分别超越了在大规模自然图像数据上预训练的 MAE（Masked Autoencoder）对比模型 0.0694 和 0.0613。这些结果进一步表明，增加预训练规模能在密集特征转移中产生一致的改善。总体而言，这些发现表明，扩展领域内自监督预训练可以减少领域差距，并改善高像素维度科学成像的可转移密集表示。
cs.CV / 85 / 2609.32185
Contamination, Prior, or Evidence? Decomposing and Training Evidence Use in Whole-Slide Vision-Language Models
污染、先验还是证据？分解与训练全幻灯片视觉语言模型中的证据使用
Abstract
Pathology vision-language models (VLMs) are conventionally evaluated by accuracy, but accuracy alone does not measure evidence use: it may conflate dataset contamination, prior knowledge, and image evidence. In a motivating study of lymph-node metastasis prediction, we found that most public pathology VLMs showed minimal differences when changing from feeding the models with whole-slide images, an annotated lesion, or no image at all. To better understand the specific features leveraged by these models, this paper presents two contributions aimed at disentangling these factors. First, we present CleanSlide, a TCGA-based VQA benchmark designed to eliminate image- and question-side contamination. It contains 149K audited multiple-choice questions over 9,985 slides, with patient- and tissue-source-disjoint splits. Every question is audited for option shortcuts, stem leakage, cross-split duplication, and blind solvability. Second, we propose Pair-DPO, a preference loss over counterfactual slide pairs from the same question and source. By controlling for shared confounding factors, Pair-DPO cancels out the question-attributable signal and leaves image evidence as the source of preference. Specifically, each pair consists of two real slides with opposite, verified findings, introducing neither editing artifacts nor unverified labels for diffuse or graded features such as invasion, necrosis, and tumor grade. Experiments show that our method gains 15.29% from image evidence on the CleanSlide, compared with 2.81% for the best published model. On the external CPTAC and BCNB cohorts, our method achieves accuracies of 57.6% and 59.0%, outperforming all other evaluated models by 9.7% and 3.4%, respectively. We will release the benchmark and code.
Chinese Translation
病理视觉语言模型（VLMs）通常通过准确性进行评估，但仅靠准确性无法衡量证据使用：它可能混淆数据集污染、先验知识和图像证据。在一项关于淋巴结转移预测的激励性研究中，我们发现大多数公共病理VLM在使用全幻灯片图像、标注病灶或根本不使用图像时，表现出最小的差异。为了更好地理解这些模型所利用的具体特征，本文提出了两个旨在解开这些因素的贡献。首先，我们提出了CleanSlide，这是一个基于TCGA的视觉问答（VQA）基准，旨在消除图像和问题侧的污染。它包含149K经过审核的多项选择题，覆盖9,985张幻灯片，并且患者和组织来源之间是互不重叠的分割。每个问题都经过审核，以检查选项捷径、题干泄漏、跨分割重复和盲解性。其次，我们提出了Pair-DPO，这是一种针对来自同一问题和来源的反事实幻灯片对的偏好损失。通过控制共享的混淆因素，Pair-DPO消除了可归因于问题的信号，并将图像证据作为偏好的来源。具体而言，每对包含两张具有相反、经过验证的发现的真实幻灯片，既不引入编辑伪影，也不为侵袭、坏死和肿瘤分级等扩散或分级特征引入未经验证的标签。实验表明，与最佳已发布模型的2.81%相比，我们的方法在CleanSlide上从图像证据中获得了15.29%的提升。在外部CPTAC和BCNB队列中，我们的方法分别达到了57.6%和59.0%的准确率，分别比所有其他评估模型高出9.7%和3.4%。我们将发布基准和代码。
cs.CV / 86 / 2609.32188
Presence Is Not Faithfulness: Figurative Vehicle Intrusion in Text-to-Image Generation
存在并非忠实：文本到图像生成中的比喻载体入侵
Abstract
Text-to-image (TTI) models increasingly generate high-quality images from natural-language prompts, yet figurative language exposes a failure: a vehicle that should guide the depiction of a tenor may instead be rendered as a visible object. We call this failure Figurative Vehicle Intrusion: the intruding content is textually licensed, but it is assigned the wrong visual role, showing that visual presence is not always faithfulness and that presence-oriented evaluation can miss such errors. To study it systematically, we introduce Vehicle Intrusion and Semantic Tenor Assessment (VISTA), a multilingual benchmark of figurative prompts organized by Figurative Form and Mapping Mechanism. We further propose V-Score, a diagnostic question-answering metric that evaluates role-aware figurative faithfulness in generated images. Evaluations on recent high-performing TTI models show that vehicle intrusion persists across languages and figurative categories. As a lightweight mitigation, we introduce VISTA-Guard, which partially reduces vehicle intrusion and suggests a practical path toward more figuratively faithful TTI generation. All resources will be released publicly.
Chinese Translation
文本到图像（TTI）模型越来越能够从自然语言提示中生成高质量图像，但比喻语言暴露出一个失败：一个应当引导比喻对象描绘的载体，可能反而被呈现为一个可见的物体。我们称这种失败为比喻载体入侵（Figurative Vehicle Intrusion）：入侵的内容在文本上是被许可的，但却被赋予了错误的视觉角色，表明视觉存在并不总是忠实的，而以存在为导向的评估可能会忽视这种错误。为了系统地研究这一现象，我们引入了载体入侵与语义比喻评估（Vehicle Intrusion and Semantic Tenor Assessment, VISTA），这是一个按比喻形式和映射机制组织的多语言比喻提示基准。此外，我们提出了V-Score，这是一种诊断性问答指标，用于评估生成图像中的角色意识比喻忠实度。对近期高性能TTI模型的评估显示，载体入侵在不同语言和比喻类别中普遍存在。作为一种轻量级的缓解措施，我们引入了VISTA-Guard，部分减少了载体入侵，并为更具比喻忠实度的TTI生成提供了实际路径。所有资源将公开发布。
cs.CV / 87 / 2609.32190
Evaluating Single and Multi-Omics Based Explainable Artificial Intelligence (MOXAI) for Molecular Subclass Classification of Adult-Type Diffuse Gliomas
评估基于单一和多组学的可解释人工智能（MOXAI）在成人型弥漫性胶质瘤分子亚类分类中的应用
Abstract
DNA methylation (DNAM) profiling has emerged as a powerful diagnostic tool for classifying brain and solid tumors. However, existing computational models typically analyze methylation and copy number variation (CNV) data separately, failing to capture the complementary information their integration could provide. Moreover, current classification models lack mechanisms for within-class risk assessment analogous to traditional tumor grading, and no established explainability method can attribute classification decisions to specific genomic loci. In this paper, we present MOXAI (Multi-Omics Based Explainable AI), a deep learning framework that integrates DNA methylation and copy number data from methylation arrays to classify molecular subtypes of adult-type diffuse gliomas, alongside single-modality variants for comparison. Using a cohort from The Cancer Genome Atlas (TCGA), we trained ResNet50, DINOv2, and Graph Attention Network (GAT) models on methylation data alone, copy number data alone, and combined multimodal data. We further developed explainable AI (XAI) methods based on class activation maps (CAMs) and gradient-weighted CAM (Grad-CAM) to identify the specific CpG sites, genes, and chromosomal regions most relevant to each classification decision. The multimodal model achieved up to 92.98% cross-validation accuracy, outperforming models trained on CNV data alone. DINOv2 showed the strongest generalization, reaching 94.25% accuracy (confidence >0.9) on independent validation sets. XAI results aligned with established molecular features of adult-type diffuse glioma subtypes, confirming the biological interpretability of the framework.
Chinese Translation
DNA甲基化（DNAM）分析已成为分类脑部和实体肿瘤的强大诊断工具。然而，现有的计算模型通常分别分析甲基化和拷贝数变异（CNV）数据，未能捕捉其整合所能提供的互补信息。此外，当前的分类模型缺乏类似于传统肿瘤分级的类内风险评估机制，且没有建立的可解释性方法能够将分类决策归因于特定的基因组位点。在本文中，我们提出了MOXAI（基于多组学的可解释人工智能），这是一个深度学习框架，整合来自甲基化阵列的DNA甲基化和拷贝数数据，以分类成人型弥漫性胶质瘤的分子亚型，并与单模态变体进行比较。我们使用癌症基因组图谱（TCGA）中的一个队列，分别在仅使用甲基化数据、仅使用拷贝数数据以及结合多模态数据上训练了ResNet50、DINOv2和图注意网络（GAT）模型。我们进一步开发了基于类激活图（CAMs）和梯度加权类激活图（Grad-CAM）的可解释人工智能（XAI）方法，以识别与每个分类决策最相关的特定CpG位点、基因和染色体区域。多模态模型的交叉验证准确率高达92.98%，优于仅在CNV数据上训练的模型。DINOv2显示出最强的泛化能力，在独立验证集上达到了94.25%的准确率（置信度>0.9）。XAI结果与已建立的成人型弥漫性胶质瘤亚型的分子特征一致，确认了该框架的生物学可解释性。
cs.CV / 88 / 2609.32193
Devol-ONE: One Autoregressive Mixture of Transformers to Unify Vision-Language-Action and Latent World Modeling
Devol-ONE：一种自回归混合变换器以统一视觉-语言-动作和潜在世界建模
Abstract
Vision Language Action (VLA) models condition actions directly on current visual and language context, without an explicit account of how the scene evolves under candidate actions. World Action Models (WAM) attempt to address this limitation by predicting future states, but existing designs keep prediction and policy learning architecturally separate, connecting them only through the predicted output, whether through pixel space video generation or a latent forecasting module trained independently of the policy. We present Devol-ONE, a Mixture of Transformers architecture that unifies vision language understanding, latent world dynamics prediction, and action generation within a single autoregressive framework. Instead of encoding vision language tokens once and feeding them to the action expert, Devol-ONE runs autoregressive prediction jointly across a vision language stream and a V-JEPA pretrained dynamics stream, attending to the vision language key-value cache at every layer to forecast future latent states under language guidance. The action expert is in turn shaped continuously by semantic reasoning and predicted physical dynamics rather than by a fixed representation computed in advance. Extensive experiments are conducted on LIBERO, LIBERO-PLUS, RoboTwin2.0 along with real-world evaluation on Flexiv single-arm and dual-arm setups. Ablation studies show the effectiveness of dynamic stream prediction and layer-wise unified attention to validate our model architectural coherency.
Chinese Translation
视觉语言动作（VLA）模型直接基于当前的视觉和语言上下文来条件化动作，而不明确考虑场景在候选动作下如何演变。世界动作模型（WAM）试图通过预测未来状态来解决这一限制，但现有设计在架构上将预测和策略学习分开，仅通过预测输出连接它们，无论是通过像素空间视频生成还是通过独立于策略训练的潜在预测模块。我们提出了Devol-ONE，一种混合变换器架构，它在一个自回归框架内统一了视觉语言理解、潜在世界动态预测和动作生成。Devol-ONE不是一次性编码视觉语言标记并将其输入到动作专家，而是通过视觉语言流和经过V-JEPA预训练的动态流共同进行自回归预测，在每一层关注视觉语言的键值缓存，以在语言指导下预测未来的潜在状态。动作专家则通过语义推理和预测的物理动态不断塑造，而不是通过预先计算的固定表示。我们在LIBERO、LIBERO-PLUS、RoboTwin2.0上进行了广泛的实验，并在Flexiv单臂和双臂设置上进行了现实世界评估。消融研究表明动态流预测和层级统一注意力的有效性，从而验证了我们模型的架构一致性。
cs.CV / 89 / 2609.32203
Kernel-Based Steering of CLIP with Vision-Language Model Preferences
基于核的CLIP引导与视觉-语言模型偏好
Abstract
Large vision-language models (VLMs) can judge visual similarity, but their judgments are not directly available as compact image embeddings for efficient comparison. We study how to transfer these preferences into CLIP while retaining its image--text capabilities. We introduce ASK, a kernel-based steering method that learns from elicited pairwise judgments without accessing teacher embeddings or collecting new human similarity annotations. ASK constructs positive semidefinite target kernels within small image groups and combines visual kernel matching with an image--text distributional anchor. Low-rank adapters jointly update the visual and text encoders while regularizing predictions toward frozen CLIP. After adaptation, retrieval uses CLIP image embeddings and cosine similarity, with no VLM calls. Experiments across five image domains, four CLIP backbones, and six judges evaluate teacher agreement, retrieval, and recognition retention. For ViT-B/16, mean retrieval mAP on classes excluded from adaptation increases from 53.8 to 75.0, compared with 71.7 for DINOv2 targets with KL anchoring. Mean zero-shot accuracy with jointly adapted encoders increases from 61.8\% to 62.4\%, averaged over 12 benchmarks and the five adaptation domains. Prompting provides an additional capability: selecting which visual distinctions the student learns. Human-annotated evaluations across four datasets support this criterion-specific control.
Chinese Translation
大型视觉-语言模型（VLMs）能够判断视觉相似性，但其判断结果并不能直接作为紧凑的图像嵌入用于高效比较。我们研究如何将这些偏好转移到CLIP中，同时保留其图像-文本能力。我们提出了ASK，一种基于核的引导方法，它通过引导的成对判断进行学习，而无需访问教师嵌入或收集新的人工相似性注释。ASK在小型图像组内构建正半定目标核，并结合视觉核匹配与图像-文本分布锚定。低秩适配器共同更新视觉和文本编码器，同时将预测正则化到冻结的CLIP上。适应后，检索使用CLIP图像嵌入和余弦相似性，无需调用VLM。在五个图像领域、四个CLIP主干和六个评审者的实验中评估了教师一致性、检索和识别保留。对于ViT-B/16，排除在适应之外的类别的平均检索mAP从53.8提高到75.0，而DINOv2目标与KL锚定的mAP为71.7。经过联合适配的编码器的平均零-shot准确率从61.8%提高到62.4%，在12个基准和五个适应领域中进行平均。提示提供了额外的能力：选择学生学习哪些视觉区分。跨四个数据集的人类注释评估支持这一特定标准的控制。
cs.CV / 90 / 2609.32222
Geometry-Preserving Blind Watermarking for Raw 3D Point Clouds
保持几何特性的盲水印技术用于原始三维点云
Abstract
Raw 3D point clouds are a core geometric representation. Establishing their ownership is challenging because point sets are irregular, unstructured, and frequently altered by resampling and geometric preprocessing. We present a blind watermarking framework that operates directly on xyz coordinates and supports both object-level shapes and scene-scale scans. At verification time, the embedded message is recovered from the observed point cloud alone, without access to the original point cloud, color, normals, or mesh connectivity. The method jointly learns watermark embedding and extraction through a feed-forward octree-based architecture, enabling efficient multi-scale geometric reasoning on large point sets. During training, a stochastic transformation layer exposes the decoder to common geometric perturbations, while progressive pose alignment improves robustness to pose changes. Experiments on object-level and scene-level benchmarks demonstrate reliable message recovery under common geometric processing while maintaining low geometric distortion. Qualitative comparisons further show that the learned perturbations are less visually conspicuous and less spatially structured than those of handcrafted alternatives.
Chinese Translation
原始三维点云是核心的几何表示。由于点集不规则、无结构，并且经常受到重采样和几何预处理的影响，确立其所有权具有挑战性。我们提出了一种盲水印框架，该框架直接在xyz坐标上操作，并支持对象级形状和场景级扫描。在验证时，嵌入的信息仅从观察到的点云中恢复，而无需访问原始点云、颜色、法线或网格连接性。该方法通过前馈八叉树架构共同学习水印嵌入和提取，从而实现对大规模点集的高效多尺度几何推理。在训练过程中，随机变换层使解码器暴露于常见的几何扰动，而渐进的姿态对齐提高了对姿态变化的鲁棒性。在对象级和场景级基准测试中的实验表明，在保持低几何失真下，能够在常见的几何处理下可靠地恢复信息。定性比较进一步表明，学习到的扰动在视觉上不那么显眼，且空间结构不如手工制作的替代方案明显。
cs.CV / 91 / 2609.32231
Skeletons in Flow: Graph Structured Flow Matching for Human Motion Prediction
流中的骨架：用于人类运动预测的图结构流匹配
Abstract
Human motion prediction requires diverse future trajectories that remain consistent with observed motion and the articulated physical structure of the body. Skeletal constraints restrict individual poses, while coordinated motion depends on spatial interactions (between connected joints) and temporal interactions (between time instants). To facilitate human motion prediction in light of these constraints and interactions, we introduce Graph Structured Flow Matching (GSFM), which transports the complete future skeletal trajectory through a single conditional velocity field. The trajectory produces a spatiotemporal skeleton graph, and spatial and temporal attention couple its evolution according to skeletal relations and physical time offsets. Bone directions lie on unit spheres relative to a root joint, and tangent evolution preserves input bone lengths throughout generation. We train a learned velocity field through conditional flow matching along geodesic paths connecting random trajectories centered on the last-observed pose to recorded future trajectories. Experiments on the Archive of Motion capture As Surface Shapes (AMASS) dataset evaluate prediction accuracy, diversity calibration, and motion statistics. We demonstrate the contributions of spatial and temporal message passing in the developed architecture through an ablation study. GSFM models trained on AMASS also perform competitively on the Human3.6M skeleton without parameter updates or retraining, demonstrating applicability to an unseen skeletal structure.
Chinese Translation
人类运动预测需要多样化的未来轨迹，这些轨迹必须与观察到的运动和身体的关节物理结构保持一致。骨骼约束限制了个体姿势，而协调运动则依赖于空间交互（连接关节之间）和时间交互（时间瞬间之间）。为了在这些约束和交互的背景下促进人类运动预测，我们提出了图结构流匹配（Graph Structured Flow Matching, GSFM），该方法通过单一条件速度场传输完整的未来骨骼轨迹。该轨迹生成一个时空骨架图，空间和时间注意力根据骨骼关系和物理时间偏移耦合其演变。骨骼方向相对于根关节位于单位球面上，切线演变在生成过程中保持输入骨骼长度。我们通过条件流匹配训练一个学习的速度场，沿着连接以最后观察到的姿势为中心的随机轨迹与记录的未来轨迹之间的测地线路径进行训练。对运动捕捉表面形状档案（Archive of Motion capture As Surface Shapes, AMASS）数据集的实验评估了预测准确性、多样性校准和运动统计。我们通过消融研究展示了在所开发架构中空间和时间信息传递的贡献。基于AMASS训练的GSFM模型在Human3.6M骨架上也表现出竞争力，无需参数更新或重新训练，展示了其对未见骨架结构的适用性。
cs.CV / 92 / 2609.32244
Two-Stage Multi-View Gait Recognition with a Re-Embedding Network
基于重嵌入网络的两阶段多视角步态识别
Abstract
Gait recognition always remains challenging due to severe overfitting and the rigid view constraints common in single-stage approaches. We propose a two-stage framework, termed Translate-First-Then-Reason (TFTR), to address these issues. In the first stage, a shallow Siamese convolutional network with triplet loss maps Gait Energy Images (GEIs) into a 128-dimensional view-specific embedding space. In the second stage, these per-view embeddings are treated as tokens and processed by a 12-layer Transformer encoder, which re-projects them into a new space with improved cosine separability. This design enables flexible fusion of an arbitrary number of views at inference, overcoming the fixed-input limitations of prior methods. Trained on the OU-MVLP dataset (6,000 subjects) and evaluated on unseen CASIA-B across normal, bag-carrying, and coat-wearing conditions, our pipeline achieves 96.91\% single-view and 99.49\% three-view accuracy on OU-MVLP, and attains 100\% accuracy on CASIA-B with three views.
Chinese Translation
步态识别一直面临着严重的过拟合和单阶段方法中常见的严格视角约束等挑战。为了解决这些问题，我们提出了一种名为“先转换再推理”（Translate-First-Then-Reason, TFTR）的两阶段框架。在第一阶段，采用浅层的Siamese卷积网络结合三元组损失，将步态能量图（Gait Energy Images, GEIs）映射到128维的视角特定嵌入空间。在第二阶段，这些每视角的嵌入被视为标记，并由一个12层的Transformer编码器处理，重新投影到一个具有更好余弦可分离性的新的空间。该设计使得在推理时能够灵活融合任意数量的视角，克服了先前方法的固定输入限制。在OU-MVLP数据集（6,000个受试者）上训练，并在未见过的CASIA-B数据集上进行评估，涵盖正常、携带包和穿外套的条件，我们的管道在OU-MVLP上达到了96.91%的单视角准确率和99.49%的三视角准确率，并在CASIA-B上以三视角达到了100%的准确率。
cs.CV / 93 / 2609.32250
RoboSTAR: Next-Scale Autoregressive Sign Language Translation for Humanoid Robots
RoboSTAR：下一代自回归手语翻译系统用于类人机器人
Abstract
Sign-language interpretation in public communication relies on qualified professional interpreters and can be difficult to scale, motivating robotic signing as a complementary accessibility interface. We present RoBoSTAR, a text-conditioned sign language production (SLP) framework for generating human-centric sign motion that can be retargeted for robotic execution, with speech supported optionally through an external ASR front end. Conventional autoregressive approaches flatten motion into a single full-resolution token sequence, forcing long-range and local dependencies to be modeled at a uniform temporal granularity. RoBoSTAR instead combines part-wise Finite Scalar Quantization with next-scale autoregression, generating motion over progressively finer temporal resolutions while predicting synchronized body and hand tokens in parallel within each step. This coarse-to-fine formulation provides compact long-range context before progressively refining motion details, while self-conditioning and context corruption improve robustness to cross-scale prediction errors. The generated motion is subsequently retargeted for physical humanoid execution. Extensive qualitative and quantitative evaluations are conducted to demonstrate the effectiveness of RoBoSTAR.
Chinese Translation
公共交流中的手语翻译依赖于合格的专业翻译员，且难以扩展，这促使了机器人手语作为一种补充的无障碍接口。我们提出了RoBoSTAR，一个基于文本条件的手语生成（SLP）框架，旨在生成以人为中心的手势动作，并可重新定向以供机器人执行，同时可通过外部自动语音识别（ASR）前端选择性地支持语音。传统的自回归方法将动作压缩为单一的全分辨率标记序列，迫使长距离和局部依赖在统一的时间粒度下建模。而RoBoSTAR则结合了部分有限标量量化与下一代自回归，在逐步细化的时间分辨率上生成动作，同时在每一步中并行预测同步的身体和手部标记。这种粗到细的表述在逐步细化动作细节之前提供了紧凑的长距离上下文，同时自我条件化和上下文干扰提高了对跨尺度预测错误的鲁棒性。生成的动作随后被重新定向以供物理类人执行。我们进行了广泛的定性和定量评估，以证明RoBoSTAR的有效性。
cs.CV / 94 / 2609.32273
FSS-UBrain: Multi-region Few-Shot Brain Tumor MRI Segmentation
FSS-UBrain：多区域少样本脑肿瘤MRI分割
Abstract
Accurate delineation of whole tumor (WT), tumor core (TC), and enhancing tumor (ET) from multimodal magnetic resonance imaging remains challenging under limited annotation, cross-cohort variation, and severe target sparsity. We propose FSS-UBrain, a region-wise one-shot segmentation framework that uses a labeled positive support slice to condition binary query segmentation separately for WT, TC, and ET. Support-derived foreground and background descriptors guide query-feature adaptation, bottleneck interaction, decoder-side reconstruction, and boundary refinement. Episodic training additionally incorporates hard-negative and fully negative queries with empty-query regularization to suppress spurious foreground activation when the selected region is absent. Although inference operates on two-dimensional support--query slice pairs, checkpoint selection, threshold calibration, and final evaluation are performed after volumetric reconstruction. FSS-UBrain is evaluated on a held-out BraTS 2020 split and under target-supported cross-cohort protocols on BraTS 2023 and BraTS-Africa. Cases used as target support are excluded from the query cohorts, and no target-domain fine-tuning or test-time parameter updates are performed. On BraTS 2020, FSS-UBrain achieves volumetric Dice scores of 89.82%, 82.14%, and 77.42% for WT, TC, and ET, respectively, with corresponding 95th-percentile Hausdorff distance (HD95) values of 11.12, 9.01, and 4.46 mm. It also achieves the highest mean Dice and lowest finite-pair mean HD95 point estimates on BraTS 2023 and BraTS-Africa among the compared few-shot methods. These findings support target-conditioned few-shot segmentation while highlighting sensitivity to support selection and cohort-specific variation.
Chinese Translation
在有限标注、跨队列变异和严重目标稀疏的情况下，从多模态磁共振成像中准确划分整个肿瘤（WT）、肿瘤核心（TC）和增强肿瘤（ET）仍然具有挑战性。我们提出了FSS-UBrain，这是一种区域级的一次性分割框架，利用标记的正支持切片分别对WT、TC和ET进行二元查询分割。支持导出的前景和背景描述符引导查询特征适应、瓶颈交互、解码器侧重建和边界细化。情节训练额外结合了困难负样本和完全负样本查询，并通过空查询正则化来抑制在所选区域缺失时虚假前景激活。尽管推理是在二维支持-查询切片对上进行的，但检查点选择、阈值校准和最终评估是在体积重建后进行的。FSS-UBrain在保留的BraTS 2020分割上以及在BraTS 2023和BraTS-Africa的目标支持跨队列协议下进行了评估。用作目标支持的案例被排除在查询队列之外，并且未进行目标领域的微调或测试时参数更新。在BraTS 2020上，FSS-UBrain分别实现了89.82%、82.14%和77.42%的体积Dice分数，及对应的95百分位Hausdorff距离（HD95）值为11.12、9.01和4.46毫米。在与比较的少样本方法中，它还在BraTS 2023和BraTS-Africa上获得了最高的平均Dice和最低的有限对平均HD95点估计。这些发现支持目标条件的少样本分割，同时突显了对支持选择和队列特定变异的敏感性。
cs.CV / 95 / 2609.32280
HeroFrame-Bench: Reference-Anchored Evaluation via Rubric--Ranking Co-Evolution for Movie Hero Frame Selection
HeroFrame-Bench：通过评分标准-排名共同演化的参考锚定评估用于电影英雄画面的选择
Abstract
Hero frames are in-film stills used as source imagery for theatrical posters, streaming cover art, film database listings, and other promotional placements. As the first visual entry point, they shape audiences' initial impressions of the movie and their subsequent willingness to watch it. Selecting these frames, a task we term hero frame selection, requires balancing content relevance with aesthetic appeal. A related task is keyframe selection, yet its benchmarks prioritize relevance over aesthetics, using either finite annotations that exclude valid alternatives or VideoQA that entangles selection quality with downstream model capability. We therefore introduce HeroFrame-Bench, built through a scalable VLM-as-a-Judge framework. We construct multimodal contexts from diverse metadata to ground a VLM judge that scores selected frames using our Reference-anchored Percentile. The percentile is obtained by inserting each frame into reusable, pre-ranked reference chains, enabling direct, extensible, and reliable evaluation. To reduce ambiguity and improve consistency in these subjective judgements, we further propose Rubric-Ranking Co-Evolution, which generates movie-specific rubrics to condition the VLM judge and refines rubrics jointly with the resulting rankings. Within this process, we introduce several verifiable signals, most notably the Inverted Rubric Attack, to select robust rubrics. Finally, HeroFrame-Bench is instantiated over 204 movies with 2,031 reference chains and 1,970 learned rubrics. We build an annotation interface for human-alignment studies which show that our construction design improves VLM agreement with human from 77.56% to 83.78%. Evaluation on multiple methods show that hero frame selection remains challenging.
Chinese Translation
英雄画面是用于电影海报、流媒体封面艺术、电影数据库列表及其他宣传位置的影片静帧。作为观众的首个视觉切入点，它们塑造了观众对电影的初步印象及其后续观看意愿。选择这些画面，我们称之为英雄画面选择，需在内容相关性与美学吸引力之间取得平衡。一个相关的任务是关键帧选择，但其基准更侧重于相关性而非美学，使用有限的注释排除了有效的替代选项，或使用视频问答（VideoQA）将选择质量与下游模型能力纠缠在一起。因此，我们引入了HeroFrame-Bench，基于可扩展的VLM-as-a-Judge框架构建。我们从多样化的元数据中构建多模态上下文，以支持VLM评审对所选画面的评分，使用我们的参考锚定百分位（Reference-anchored Percentile）。该百分位通过将每个画面插入可重用的、预先排名的参考链获得，从而实现直接、可扩展和可靠的评估。为了减少模糊性并提高这些主观判断的一致性，我们进一步提出了评分标准-排名共同演化（Rubric-Ranking Co-Evolution），该方法生成特定于电影的评分标准以调节VLM评审，并与结果排名共同优化评分标准。在这一过程中，我们引入了多个可验证的信号，最显著的是反向评分标准攻击（Inverted Rubric Attack），以选择稳健的评分标准。最后，HeroFrame-Bench在204部电影上实例化，包含2,031条参考链和1,970个学习到的评分标准。我们构建了一个注释接口用于人类对齐研究，结果表明我们的构建设计使VLM与人类的协议从77.56%提高到83.78%。对多种方法的评估显示，英雄画面选择仍然具有挑战性。
cs.CV / 96 / 2609.32316
One Perception, All Maneuvers: Directional Traffic Signal Understanding for Maneuver-Level Signal Intent Prediction
一种感知，所有操控：用于操控级信号意图预测的方向性交通信号理解
Abstract
Traffic lights are a key regulatory signal for autonomous driving at urban intersections, yet existing traffic signal perception is still predominantly formulated as instance-level detection or color recognition. Such formulations identify where traffic lights are and what colors they display, but leave a critical semantic gap before downstream planning: which ego maneuver is controlled by each visible signal and what dynamic permission the signal expresses for that maneuver. In this paper, we formulate Directional Traffic Signal Understanding, a decision-oriented task that predicts structured signal states for straight, left-turn, right-turn, and U-turn maneuvers from a front-view image. Each state contains the associated signal color and signal-implied passability. Based on OpenLane-V2, we provide a direction-level benchmark with maneuver-level supervision and metrics for color recognition, passability, full-frame consistency, and safety-critical errors. A direction-aware baseline combines global context, localized traffic-light evidence, and maneuver-specific representations. Experiments show that direction-level modeling improves passability prediction over image-level classifiers and detection-oriented pipelines, particularly at complex multi-signal intersections. The resulting representation provides a direct and interpretable traffic-signal interface for downstream planning together with topology, route, and surrounding-agent information.
Chinese Translation
交通信号灯是城市交叉口自动驾驶的重要调控信号，但现有的交通信号感知仍主要以实例级检测或颜色识别为主。这种表述方式识别了交通信号灯的位置及其显示的颜色，但在下游规划中留下了一个关键的语义空白：每个可见信号控制的自我操控是什么，以及该信号对该操控所表达的动态许可是什么。本文提出了方向性交通信号理解（Directional Traffic Signal Understanding），这是一项面向决策的任务，旨在从前视图图像中预测直行、左转、右转和掉头操控的结构化信号状态。每个状态包含相关的信号颜色和信号隐含的通行性。基于OpenLane-V2，我们提供了一个方向级基准，具有操控级监督和针对颜色识别、通行性、全帧一致性及安全关键错误的评估指标。方向感知基线结合了全局上下文、本地化的交通信号证据和特定操控的表示。实验表明，方向级建模在复杂多信号交叉口的通行性预测上优于图像级分类器和检测导向的管道。最终的表示为下游规划提供了一个直接且可解释的交通信号接口，并结合了拓扑、路线和周围代理的信息。
cs.CV / 97 / 2609.32323
FoundDSR: A Generalizable Foundation Model with Guided 2D Gaussian Splatting for Depth Super-Resolution
FoundDSR：一种具有引导性2D高斯溅射的可泛化基础模型用于深度超分辨率
Abstract
We introduce FoundDSR, a generalizable foundation model for robust depth reconstruction across unseen data distributions using RGB-D pairs. FoundDSR begins with a guided 2D Gaussian Splatting strategy to model depth representations with Gaussian primitives. This strategy employs high-resolution RGB as prompts to optimize the Gaussian parameters, thereby encouraging each Gaussian primitive to anisotropically deform along high-frequency structural directions. The resulting Gaussian-upsampled representations are then mapped to high-resolution depth through an effective depth reconstruction branch. Furthermore, to mitigate training instability and bias toward dominant sources caused by distribution gaps in large-scale heterogeneous data, we introduce heterogeneous federated learning that allocates each data source to an independent client for local optimization and global aggregation. This design effectively endows FoundDSR with stable scalability to diverse and large-scale training data. Extensive zero-shot evaluations on synthetic, real-world, arbitrary-scale, and noisy conditions demonstrate that FoundDSR consistently outperforms existing state-of-the-art approaches, confirming its strong robustness and generalization to unknown scenes.
Chinese Translation
我们介绍了FoundDSR，这是一种可泛化的基础模型，旨在利用RGB-D对在未见数据分布中进行稳健的深度重建。FoundDSR采用引导性2D高斯溅射策略，通过高斯原语建模深度表示。该策略利用高分辨率RGB作为提示来优化高斯参数，从而促使每个高斯原语沿高频结构方向各向异性变形。生成的高斯上采样表示随后通过有效的深度重建分支映射到高分辨率深度。此外，为了减轻由于大规模异构数据中的分布差距导致的训练不稳定性和对主导源的偏倚，我们引入了异构联邦学习，将每个数据源分配给独立客户端进行本地优化和全局聚合。这一设计有效地赋予了FoundDSR在多样化和大规模训练数据上的稳定可扩展性。在合成、真实世界、任意尺度和噪声条件下的广泛零-shot评估表明，FoundDSR始终优于现有的最先进方法，确认了其对未知场景的强鲁棒性和泛化能力。
cs.CV / 98 / 2609.32333
Progressive-View On-Policy Distillation for Regional-to-Global Transfer in Multimodal LLMs
用于多模态大语言模型的区域到全局迁移的渐进视图在线蒸馏
Abstract
Regional-to-global distillation uses crop-conditioned guidance to improve full-image understanding. The challenge is to effectively transfer the teacher's crop-based advantage to the student's full-image inference. We propose progressive-view on-policy distillation (PVD), which shifts the student's view distribution from the crop toward the full image through an intermediate aspect-preserving padded crop. The padded crop preserves regional content while matching the full image's visual-token grid. Across stages, the view mixture assigns increasing probability to the full image. A lightweight regional-advantage weighting reallocates token-level supervision using the crop-conditioned teacher-student log-probability gap. Evaluated under each sampled input, it applies mild reweighting when the gap is small and emphasizes higher-gap tokens when the gap widens. A Jensen-Shannon metric decomposition interprets this schedule as a shift from matched-input imitation toward the deployment objective. Across benchmarks spanning perception, visual mathematics and general multimodal question answering, PVD-full reaches an average accuracy of 77.51 over three seeds, improving on the reward-free distillation baseline by 2.01 points and on its reward-matched variant by 1.00 point. In the reward-free setting, PVD-distill still gains 1.16 points.
Chinese Translation
区域到全局的蒸馏利用基于裁剪的条件指导来改善全图理解。挑战在于如何有效地将教师在裁剪基础上的优势转移到学生的全图推理中。我们提出了渐进视图在线蒸馏（Progressive-View On-Policy Distillation, PVD），通过一个中间的保持纵横比的填充裁剪，将学生的视图分布从裁剪向全图转移。填充裁剪在匹配全图的视觉标记网格的同时，保留了区域内容。在各个阶段，视图混合为全图分配越来越高的概率。一个轻量级的区域优势加权利用基于裁剪的教师-学生对数概率差重新分配标记级监督。在每个采样输入下进行评估时，当差距较小时，它会进行轻微的重加权，而当差距扩大时，则强调高差距的标记。一个詹森-香农度量分解将这一调度解释为从匹配输入模仿向部署目标的转变。在涵盖感知、视觉数学和一般多模态问答的基准测试中，PVD-full在三个种子上达到了77.51的平均准确率，比无奖励蒸馏基线提高了2.01点，比其奖励匹配变体提高了1.00点。在无奖励设置下，PVD-distill仍然获得了1.16点的提升。
cs.CV / 99 / 2609.32340
SGA-Flow-GRPO: Spatial Gradient-Guided Credit Assignment for Flow-GRPO
SGA-Flow-GRPO：基于空间梯度引导的Flow-GRPO信用分配
Abstract
Reinforcement Learning (RL) has proven effective in aligning flow-based generative models with human preferences. Recently, Flow-GRPO has emerged as an efficient critic-free paradigm by calculating advantages over sampled candidate trajectories. However, standard Flow-GRPO applies a uniform scalar advantage across both temporal denoising steps and spatial latent dimensions, without explicitly accounting for the spatial structure of generated images, which may lead to sub-optimal policy updates. To address this, we propose a novel gradient-guided spatial credit assignment framework tailored for Diffusion Transformers (DiTs). We first reformulate the transition-level log-likelihood in Flow-GRPO into a token-wise representation natively aligned with DiT patch architectures, constructing spatially fine-grained importance sampling ratios. To allocate localized credit without rigid, boundary-sensitive segmentation heuristics, we introduce a continuous spatial credit map derived from reward gradients. Crucially, we employ an outlier-robust normalization scheme based on Median Absolute Deviation (MAD) coupled with temperature scaling, effectively eliminating gradient noise while highlighting functional prompt-aligned regions. Extensive evaluations on GenEval show that our approach delivers SOTA alignment quality, achieving a convergence rate comparable to top-tier methods like DiffusionNFT while substantially improving upon Flow-GRPO-based methods in alignment performance.
Chinese Translation
强化学习（RL）已被证明在将基于流的生成模型与人类偏好对齐方面有效。最近，Flow-GRPO作为一种高效的无评论者范式，通过计算采样候选轨迹的优势而崭露头角。然而，标准的Flow-GRPO在时间去噪步骤和空间潜在维度上应用统一的标量优势，而未明确考虑生成图像的空间结构，这可能导致次优的策略更新。为了解决这个问题，我们提出了一种新颖的基于梯度引导的空间信用分配框架，专门针对扩散变换器（Diffusion Transformers, DiTs）。我们首先将Flow-GRPO中的过渡级对数似然重新表述为与DiT补丁架构本质上对齐的逐标记表示，构建空间细粒度的重要性采样比率。为了在不依赖于严格的、对边界敏感的分割启发式方法的情况下分配局部信用，我们引入了一个基于奖励梯度推导的连续空间信用图。关键是，我们采用了一种基于中位数绝对偏差（Median Absolute Deviation, MAD）结合温度缩放的异常值鲁棒归一化方案，有效消除了梯度噪声，同时突出了功能提示对齐区域。在GenEval上的广泛评估表明，我们的方法提供了最先进的对齐质量，其收敛速度可与顶尖方法如DiffusionNFT相媲美，同时在对齐性能上显著优于基于Flow-GRPO的方法。
cs.CV / 100 / 2609.32343
OpenMASC: An Open-Source Pipeline for Cross-Trajectory Metal-Aware Sampling and Correction in Accelerated MRI
OpenMASC：一个用于加速MRI中跨轨迹金属感知采样与校正的开源管道
Abstract
Metal implants corrupt MRI measurements throughout $k$-space, yet existing accelerated MRI methods assume clean data and most metal artifact reduction approaches assume fully sampled acquisitions. No public dataset provides paired $k$-space and images with and without metal for the same anatomy, and no framework jointly addresses artifact-aware acquisition and reconstruction across sampling trajectories. We present OpenMASC, an open-source pipeline covering the full workflow from data generation to deployment. A physics-based data generation module converts public CT volumes into paired clean and metal-corrupted MRI data in both Cartesian and radial formats. MA-VarNet, an unrolled reconstruction network with a per-cascade DC Rectifier, corrects artifacts that data-consistency steps reintroduce from corrupted measurements. A reinforcement learning agent actively selects $k$-space readouts and co-trains with the reconstruction network through a decoupled three-stage procedure. The framework is trajectory-agnostic except for the data-consistency operator, supporting both Cartesian and radial acquisition without architectural changes. Experiments on two datasets at $4\times$ and $8\times$ acceleration demonstrate consistent improvements over conventional and learned baselines on both trajectories.
Chinese Translation
金属植入物在$k$-空间中破坏MRI测量，但现有的加速MRI方法假设数据是干净的，而大多数金属伪影减少方法则假设完全采样的获取。没有公开数据集提供相同解剖结构下带有和不带有金属的配对$k$-空间和图像，也没有框架能够共同解决跨采样轨迹的伪影感知获取和重建。我们提出了OpenMASC，这是一个覆盖从数据生成到部署的完整工作流的开源管道。基于物理的数据生成模块将公共CT体积转换为配对的干净和金属污染的MRI数据，支持笛卡尔和径向格式。MA-VarNet是一个展开的重建网络，配备每个级联的直流整流器，校正数据一致性步骤从污染测量中重新引入的伪影。一个强化学习代理主动选择$k$-空间读出，并通过解耦的三阶段程序与重建网络共同训练。该框架对轨迹是无关的，除了数据一致性算子，支持笛卡尔和径向获取而无需架构更改。在$4 imes$和$8 imes$加速的两个数据集上的实验表明，在这两种轨迹上均显著优于传统和学习基线。
cs.CV / 101 / 2609.32352
EyeVQA: Benchmarking Ophthalmic Vision-Language Models from Recognition to Spatial Grounding
EyeVQA：从识别到空间定位的眼科视觉语言模型基准测试
Abstract
Vision-language models (VLMs) have shown increasing potential for medical image understanding, yet their capabilities in ophthalmic imaging remain insufficiently characterized. Existing ophthalmic datasets are typically designed for individual diseases or specialized tasks, making it difficult to systematically evaluate whether VLMs can move beyond disease recognition toward comparative reasoning and fine-grained spatial grounding. We introduce EyeVQA, a unified visual question answering benchmark for comprehensive evaluation of ophthalmic VLMs. EyeVQA is constructed from 21 available ophthalmic datasets and contains 20,000 question-answer pairs spanning six disease groups and seven question types: Single-Choice, Multi-Select, Variable-Select, True-False, Ranking, Point Location, and Bounding Box. Gold answers are deterministically derived from source-provided diagnoses, severity grades, clinical findings, segmentation masks, bounding boxes, and anatomical landmarks, enabling reproducible evaluation without relying on model-generated annotations. Notably, 44.5% of the questions require reasoning across multiple images, extending evaluation beyond conventional single-image medical VQA. We benchmark fourteen representative general-purpose, scientific, and medically specialized VLMs under a unified zero-shot protocol. The best-performing model only achieves an overall score of 62.8, while substantial gaps remain in spatial grounding and cross-task generalization. These results highlight the limitations of current VLMs in comprehensive ophthalmic visual understanding and establish EyeVQA as a diagnostic benchmark for developing more reliable and spatially grounded ophthalmic multimodal models. The project page is available at https://github.com/PKUTHM/EyeVQA.
Chinese Translation
视觉语言模型（VLMs）在医学图像理解方面展现出越来越大的潜力，但其在眼科成像中的能力仍未得到充分表征。现有的眼科数据集通常针对单一疾病或特定任务设计，这使得系统性评估 VLMs 是否能够超越疾病识别，向比较推理和细粒度空间定位发展变得困难。我们提出了 EyeVQA，这是一个统一的视觉问答基准，用于全面评估眼科 VLMs。EyeVQA 由 21 个可用的眼科数据集构建而成，包含 20,000 对问题-答案，涵盖六个疾病组和七种问题类型：单选题、多选题、可变选择题、真假题、排名题、点位定位题和边界框题。金标准答案是从源提供的诊断、严重程度等级、临床发现、分割掩膜、边界框和解剖标志中确定的，从而实现可重复的评估，而无需依赖模型生成的注释。值得注意的是，44.5% 的问题需要跨多个图像进行推理，扩展了评估范围，超越了传统的单图像医学 VQA。我们在统一的零样本协议下对十四个具有代表性的通用、科学和医学专业 VLMs 进行了基准测试。表现最佳的模型仅获得了 62.8 的总体分数，而在空间定位和跨任务泛化方面仍存在显著差距。这些结果突显了当前 VLMs 在全面眼科视觉理解中的局限性，并确立了 EyeVQA 作为开发更可靠和空间定位的眼科多模态模型的诊断基准。项目页面可访问 https://github.com/PKUTHM/EyeVQA。
cs.CV / 102 / 2609.32353
Fewer Tokens, More Self-Teaching: On-Policy Self-Distillation for Extreme Visual Token Reduction
更少的标记，更多的自我教学：极端视觉标记减少的在线自蒸馏
Abstract
Visual token reduction is an effective way to accelerate multimodal large language models (MLLMs), but performance deteriorates rapidly under extremely low token budgets. Existing work has explored both visual-token selection and training-based adaptation to reduced visual inputs. We take a step further by asking how a heavily compressed MLLM should learn from the states induced by its own generations. This setting naturally calls for on-policy self-distillation: a heavily compressed model is supervised on the states induced by its own generations, while its full-token counterpart serves as an information-rich teacher. Based on this insight, we propose LT-OPD, a training framework for extreme visual-token reduction. The student rolls out responses with only a small fraction of visual tokens, and a frozen full-token copy of the same MLLM provides distributional supervision along these student-generated trajectories. To stabilize on-policy learning when visual evidence is severely limited, we further introduce a budget-level curriculum that progressively decreases the token budget during training. Across nine benchmarks on Qwen3.5-4B, LT-OPD raises average retained performance under 5% visual-token retention from 68.6% to 82.3%, outperforming training-free, training-based, and reinforcement-learning baselines at the same budget. The gains transfer consistently to Qwen3.5-9B, GLM-4.6V-9B, and LLaVA-OV-1.5-4B. LT-OPD also reduces KV-cache usage by 85.2% and prefill FLOPs by 85.4% without additional inference overhead, demonstrating that on-policy learning can substantially recover capabilities lost to extreme visual-token reduction.
Chinese Translation
视觉标记减少是一种加速多模态大语言模型（MLLMs）的有效方法，但在极低的标记预算下，性能迅速下降。现有研究探讨了视觉标记选择和基于训练的适应以减少视觉输入。我们进一步提出一个问题：一个高度压缩的MLLM应如何从其自身生成的状态中学习。这个设置自然要求在线自蒸馏：一个高度压缩的模型在其自身生成的状态上进行监督，而其完整标记的对应模型则作为信息丰富的教师。基于这一见解，我们提出了LT-OPD，一个用于极端视觉标记减少的训练框架。学生模型仅用少量视觉标记生成响应，而同一MLLM的冻结完整标记副本在这些学生生成的轨迹上提供分布式监督。为了在视觉证据严重有限的情况下稳定在线学习，我们进一步引入了一个预算级别的课程，该课程在训练过程中逐步减少标记预算。在Qwen3.5-4B的九个基准测试中，LT-OPD将5%视觉标记保留下的平均保留性能从68.6%提高到82.3%，在相同预算下超越了无训练、基于训练和强化学习的基线。这些提升在Qwen3.5-9B、GLM-4.6V-9B和LLaVA-OV-1.5-4B中也得到了持续转移。LT-OPD还将KV-cache的使用减少了85.2%，将预填充FLOPs减少了85.4%，且没有额外的推理开销，证明在线学习能够显著恢复因极端视觉标记减少而失去的能力。
cs.CV / 103 / 2609.32362
StegGNN: Learning Graphical Representation for Image Steganography
StegGNN：用于图像隐写的图形表示学习
Abstract
Image steganography refers to embedding secret messages within cover images while maintaining imperceptibility. Recent advances in deep learning - primarily driven by Convolutional Neural Networks (CNNs) and architectures such as inverse neural networks, autoencoders, and generative adversarial networks - have led to notable progress. However, these frameworks are primarily built on CNN architectures, which treat images as regular grids and are limited by their receptive field size and a bias toward spatial locality. In parallel, Graph Neural Networks (GNNs) have recently demonstrated strong adaptability in several computer vision tasks, achieving state-of-the-art performance with architectures such as Vision GNN (ViG). This work moves in that direction and introduces StegGNN - a novel autoencoder-based, cover-agnostic image steganography framework based on GNNs. By modeling images as graph structures, our approach leverages the representational flexibility of GNNs over the grid-based rigidity of conventional CNNs. We conduct extensive experiments on standard benchmark datasets to evaluate visual quality and imperceptibility. Our results show that our GNN-based method performs comparably to existing CNN benchmarks. These findings suggest that GNNs provide a promising alternative representation for steganographic embedding and open the field of deep learning-based steganography to further exploration of GNN-based architectures.
Chinese Translation
图像隐写是指在保持不可察觉性的情况下，将秘密信息嵌入到载体图像中。近年来，深度学习的进步——主要由卷积神经网络（CNNs）及逆神经网络、自编码器和生成对抗网络等架构推动——取得了显著进展。然而，这些框架主要基于CNN架构，将图像视为常规网格，受限于其感受野大小和对空间局部性的偏向。与此同时，图神经网络（GNNs）最近在多个计算机视觉任务中展现出了强大的适应性，使用如视觉GNN（ViG）等架构达到了最先进的性能。本研究朝着这个方向发展，提出了StegGNN——一种基于GNN的、基于自编码器的新型隐写框架，具有对载体无关的特性。通过将图像建模为图结构，我们的方法利用了GNN在表示灵活性方面的优势，克服了传统CNN在网格基础上的局限性。我们在标准基准数据集上进行了广泛的实验，以评估视觉质量和不可察觉性。我们的结果表明，我们的基于GNN的方法在性能上与现有的CNN基准相当。这些发现表明，GNN为隐写嵌入提供了一种有前景的替代表示，并为基于深度学习的隐写领域的进一步探索GNN架构打开了新的可能性。
cs.CV / 104 / 2609.32376
An End-to-End Latent-Rollout Approach for Pushing Few-Step ImageNet-$256$ Generation to FID $1.11$ without Fr\'echet Losses
一种端到端的潜在展开方法，将少步ImageNet-$256$生成推向FID $1.11$，无需Fréchet损失
Abstract
Iterative generation poses a joint optimization problem across steps, as intermediate predictions shape subsequent computations and ultimately determine the final output distribution. Few-step generators distilled from pretrained diffusion and flow-matching models make such optimization computationally practical end to end. We build on this opportunity with a distill-then-refine approach that uses teacher imitation to establish a strong initialization for a few-step rollout in latent space, then shifts to end-to-end refinement of the complete latent rollout against real data. We introduce FiST (Flow-in-Stage Transformer), an architecture that composes learned latent-state transitions in a few stages using a shared Transformer, with optional cross-stage hidden communication. Distillation applies teacher-forced regression to selected states along teacher trajectories; refinement replaces this supervision with adversarial and auxiliary classification objectives on the final latent output. A trainable discriminator module operates on semantically rich features extracted from clean real and generated latents by a frozen SiT backbone pretrained with REPA. All training takes place in latent space, without image decoding. During refinement, FiST consumes its own intermediate predictions, and endpoint gradients pass through every generation stage. For class-conditional generation on ImageNet at $256\times256$, our approach achieves FID 1.11 (IS 282) with three stages and FID 1.15 (IS 280) with two. These results demonstrate competitive few-step generation through learned distribution-level supervision, without explicit Fr\'echet-distance minimization. Ablations characterize how distillation, pretrained checkpoint choices, refinement supervision, and cross-stage hidden communication affect generation quality.
Chinese Translation
迭代生成在步骤之间提出了一个联合优化问题，因为中间预测会影响后续计算，并最终决定最终输出分布。从预训练的扩散和流匹配模型中提炼出的少步生成器使得这种优化在计算上变得可行。我们基于这一机会，采用了先提炼后精炼的方法，通过教师模仿为潜在空间中的少步展开建立强初始化，然后转向针对真实数据的完整潜在展开的端到端精炼。我们引入了FiST（Flow-in-Stage Transformer）架构，该架构使用共享的Transformer在几个阶段中组合学习到的潜在状态转换，并可选地进行跨阶段的隐藏通信。提炼通过在教师轨迹上对选定状态进行教师强制回归来应用；精炼则用对最终潜在输出的对抗性和辅助分类目标替代这种监督。一个可训练的判别器模块在从干净的真实和生成潜在中提取的语义丰富特征上操作，这些潜在是通过一个冻结的SiT主干网络预训练的，使用REPA进行训练。所有训练都在潜在空间中进行，无需图像解码。在精炼过程中，FiST消耗其自身的中间预测，端点梯度通过每个生成阶段传递。对于ImageNet的类条件生成，在$256 imes256$的情况下，我们的方法在三个阶段中实现了FID 1.11（IS 282），在两个阶段中实现了FID 1.15（IS 280）。这些结果展示了通过学习的分布级监督实现竞争性的少步生成，而无需显式的Fréchet距离最小化。消融实验表征了提炼、预训练检查点选择、精炼监督和跨阶段隐藏通信如何影响生成质量。
cs.CV / 105 / 2609.32389
RefCompose: Multi-Reference Image Generation via LoRA-Conditioned Diffusion
RefCompose：基于LoRA条件扩散的多参考图像生成
Abstract
Filmmakers and visual artists routinely need to compose multiple references, actors, locations, props, cultural elements, into a single coherent shot, but existing tools either fail to scale past a handful of references or destroy fine grained subject identity in the process, since per reference tokenization scales memory linearly with reference count $N$ and generated content often departs from the given references rather than reproducing them. We propose \textbf{RefCompose}, a pixel space compositional conditioning framework that decouples \emph{where} things go from \emph{what} they look like, via a single fixed resolution reference canvas that keeps conditioning size constant regardless of reference count. Spatial layout is induced at inference time from a frozen diffusion transformer and extracted via Grounding DINO, requiring no LLM or dedicated layout model, while dual stream LoRA adapters inject a layout derived depth map and the encoded canvas through separate low rank streams, disentangling geometric scaffolding from localized appearance. On the Dense Layout protocol, RefCompose consistently outperforms layout based and state of the art multi reference baselines on color, texture, shape, spatial accuracy, and identity/content preservation at higher reference counts, all with constant inference memory, making it a practical building block for multi subject cinematic composition at production scale.
Chinese Translation
电影制作人和视觉艺术家通常需要将多个参考、演员、地点、道具和文化元素组合成一个连贯的镜头，但现有工具要么无法扩展到超过少量参考，要么在此过程中破坏了细致的主题身份，因为每个参考的标记化在参考数量 $N$ 上线性扩展内存，而生成的内容往往偏离给定的参考，而不是重现它们。我们提出了 extbf{RefCompose}，一种像素空间组合条件框架，通过一个固定分辨率的参考画布将 extit{物体的位置}与 extit{物体的外观}解耦，保持条件大小在参考数量变化时恒定。空间布局在推理时由冻结的扩散变换器诱导，并通过Grounding DINO提取，无需大型语言模型或专用布局模型，而双流LoRA适配器通过独立的低秩流注入基于布局的深度图和编码画布，将几何支架与局部外观分离。在Dense Layout协议上，RefCompose在颜色、纹理、形状、空间准确性和身份/内容保留方面，在更高的参考数量下始终优于基于布局的和最先进的多参考基线，且在推理内存上保持恒定，使其成为在生产规模下进行多主体电影组合的实用构建模块。
cs.CV / 106 / 2609.32399
Endo-TSR: Temporal Spectral Modeling of Appearance and Motion for Endoscopic Reconstruction
Endo-TSR：内窥镜重建中外观与运动的时间谱建模
Abstract
Endoscopic scene reconstruction requires modeling tissue motion and temporal appearance while recovering fine surface detail. Deformable Gaussian models provide explicit trajectories, but their fixed colour coefficients lack a dedicated temporal representation for photometric changes. We propose Endo-TSR, which augments deformable Gaussian splatting with bounded Fourier colour residuals and independent translation residuals on shared temporal frequencies. The colour residuals capture local appearance changes, while a Mat\'ern spectral prior regularises motion corrections. Multi-scale Laplacian supervision guides tissue-detail recovery during joint image fitting. Extensive experiments on the EndoNeRF and StereoMIS datasets demonstrate state-of-the-art rendering quality, with the highest PSNR across all evaluated sequences. Ablation studies show that temporal appearance yields the largest PSNR gain among the tested component additions, while appearance and detail supervision jointly improve rendering with fixed Gaussian counts.
Chinese Translation
内窥镜场景重建需要在恢复细腻表面细节的同时对组织运动和时间外观进行建模。可变形高斯模型提供了明确的轨迹，但其固定的颜色系数缺乏对光度变化的专门时间表示。我们提出了Endo-TSR，它通过有界傅里叶颜色残差和共享时间频率上的独立平移残差增强了可变形高斯点云。颜色残差捕捉局部外观变化，而Matérn谱先验则对运动校正进行正则化。多尺度拉普拉斯监督在联合图像拟合过程中指导组织细节的恢复。在EndoNeRF和StereoMIS数据集上的大量实验表明，该方法实现了最先进的渲染质量，在所有评估序列中获得了最高的峰值信噪比（PSNR）。消融研究表明，时间外观在测试的组件添加中带来了最大的PSNR增益，而外观和细节监督共同改善了固定高斯数量的渲染效果。
cs.CV / 107 / 2609.32405
Toward On-Chip Training of Spiking Neural Networks for Dense Event-Based Vision
面向密集事件驱动视觉的片上脉冲神经网络训练
Abstract
Event cameras provide low-latency, asynchronous visual sensing for resource-constrained robotics. Spiking neural networks (SNNs) process event streams naturally, but training deep SNNs with backpropagation through time (BPTT) requires substantial memory and remains difficult on neuromorphic hardware. Local learning avoids this by restricting error propagation to local blocks, but existing methods mainly target classification rather than dense prediction. We introduce DELL (Dense Event-driven Local Learning), a block wise scheme for dense event-based vision that replaces global gradient propagation with local dense supervision. Learnable, spatially structured local heads supervise each block at its appropriate resolution while preserving temporal dynamics within blocks. We evaluate DELL on optical-flow regression and semantic segmentation with a fully spiking U-shaped architecture. On DSEC optical flow, DELL reduces peak training memory by 39.6% relative to end-to-end BPTT while improving accuracy, reaching 1.670 px endpoint error on the official test benchmark versus 1.941 px for the same backbone trained end-to-end. Block detachment behaves more like a regularizer than a constraint. DECOLLE, the existing local-learning baseline, relies on fixed random local read-outs poorly suited to dense regression, resulting in a 3.9x higher endpoint error; learnable local heads recover this loss and outperform end-to-end training across all optical-flow metrics. On segmentation, they recover most of the performance gap, although DELL remains a few mIoU points behind end-to-end training. With 2.3M parameters, 24x fewer than the strongest SNN baseline, the backbone remains competitive with the SNN state of the art on DSEC. These results extend local learning to dense event-based prediction while substantially reducing training memory.
Chinese Translation
事件相机为资源受限的机器人提供低延迟、异步的视觉感知。脉冲神经网络（SNNs）自然处理事件流，但通过时间反向传播（BPTT）训练深层SNNs需要大量内存，并且在神经形态硬件上仍然困难。局部学习通过将误差传播限制在局部块中来避免这一问题，但现有方法主要针对分类而非密集预测。我们提出了DELL（密集事件驱动局部学习），这是一种针对密集事件驱动视觉的块级方案，它用局部密集监督替代了全局梯度传播。可学习的、空间结构化的局部头在适当的分辨率下监督每个块，同时保持块内的时间动态。我们在光流回归和语义分割上评估了DELL，采用完全脉冲的U型架构。在DSEC光流数据集上，DELL相对于端到端的BPTT将峰值训练内存减少了39.6%，同时提高了准确性，在官方测试基准上达到了1.670 px的端点误差，而同一骨干网络的端到端训练则为1.941 px。块分离的行为更像是一种正则化器而非约束。现有的局部学习基线DECOLLE依赖于固定随机的局部读取，这对于密集回归不太适合，导致端点误差高出3.9倍；可学习的局部头弥补了这一损失，并在所有光流指标上超越了端到端训练。在分割任务中，它们恢复了大部分性能差距，尽管DELL仍比端到端训练低几mIoU点。具有230万个参数，较最强SNN基线少24倍，该骨干网络在DSEC上仍与SNN领域的最先进技术具有竞争力。这些结果将局部学习扩展到密集事件驱动预测，同时显著减少了训练内存。
cs.CV / 108 / 2609.32415
RefAdapt-DiT: Adaptive Joint Attention for Reference-Conditioned Diffusion Transformers
RefAdapt-DiT：参考条件扩散变换器的自适应联合注意力
Abstract
Diffusion Transformers (DiTs) have become the standard backbone for high-quality generative modeling, yet deploying them in conditional generation tasks remains computationally prohibitive because bidirectional joint attention repeatedly processes large reference streams. While existing optimization schemes mitigate generic temporal redundancy, they typically rely on coarse-grained static reuse and overlook the distinct dynamics of references and targets. Specifically, we observe that reference representations often evolve slowly along the generation trajectory, while the target often assigns little attention mass to them; reference drift and this target-to-reference exposure jointly shape how strongly stale reference states affect the target. To exploit these patterns, we introduce \RefAdapt, a training-free framework for adaptive control of joint attention between references and targets. Instead of rigid static strategies, \RefAdapt combines consecutive target-Q change with previously observed target-to-reference attention mass to control reference computation adaptively at block granularity. Under ultra-few-step settings, \RefAdapt enables speedups of up to $2.097\times$ on 4-step MiniMax H3 and $3.54\times$ on 8-step Qwen Image Edit, while maintaining comparable visual quality.
Chinese Translation
扩散变换器（DiTs）已成为高质量生成建模的标准骨干，但在条件生成任务中部署它们仍然在计算上具有挑战性，因为双向联合注意力会反复处理大量参考流。虽然现有的优化方案减轻了通用的时间冗余，但它们通常依赖于粗粒度的静态重用，并忽视了参考和目标的独特动态。具体而言，我们观察到参考表示通常在生成轨迹上缓慢演变，而目标往往对其分配较少的注意力；参考漂移和这种目标对参考的暴露共同影响着过时的参考状态对目标的影响强度。为了利用这些模式，我们引入了 extit{RefAdapt}，一个用于自适应控制参考与目标之间联合注意力的无训练框架。与僵化的静态策略不同， extit{RefAdapt}结合了连续的目标-Q变化和先前观察到的目标对参考的注意力质量，以在块粒度上自适应地控制参考计算。在超少步设置下， extit{RefAdapt}在4步MiniMax H3上实现了高达$2.097 imes$的加速，在8步Qwen图像编辑上实现了$3.54 imes$的加速，同时保持了可比的视觉质量。
cs.CV / 109 / 2609.32427
CityToolVQA: Tool-Augmented Visual Question Answering for 3D Spatial Cognition in Urban Low-Altitude Environments
CityToolVQA：用于城市低空环境中3D空间认知的工具增强视觉问答
Abstract
CityToolVQA addresses the weak performance of Vision-Language Models (VLMs) on quantitative tasks in urban low-altitude visual question answering. We divide the seven tasks into qualitative and quantitative groups: qualitative questions are answered directly by the VLM, whereas quantitative questions are handled by an external visual-geometric toolchain that performs object grounding, segmentation, depth back-projection, and spatial computation. The toolchain can be attached to different VLMs in a zero-shot manner; a Depth-Assisted Prompt Inference (DAPI) fallback is triggered when the main-chain detection is invalid or unreliable, and CityToolVQA-SFT adapts the 8B backbone to tool-conditioned inputs. On the 73,324-question Open3D-VQA-v2 test set, CityToolVQA-SFT (Qwen3-VL-8B) reaches 67.6% overall accuracy, and attaching the toolchain to ten open-source VLMs improves quantitative-task accuracy by 12.7-36.9 percentage points. These results indicate that externalizing explicit 3D geometric computation effectively complements the limited ability of RGB-only VLMs to estimate metric distances and object sizes.
Chinese Translation
CityToolVQA 解决了视觉-语言模型（VLM）在城市低空视觉问答中定量任务表现不佳的问题。我们将七个任务分为定性和定量两组：定性问题由 VLM 直接回答，而定量问题则通过一个外部视觉几何工具链处理，该工具链执行对象定位、分割、深度反投影和空间计算。该工具链可以以零样本的方式附加到不同的 VLM 上；当主链检测无效或不可靠时，会触发深度辅助提示推理（Depth-Assisted Prompt Inference, DAPI）回退，并且 CityToolVQA-SFT 将 8B 主干适配为工具条件输入。在 73,324 个问题的 Open3D-VQA-v2 测试集中，CityToolVQA-SFT（Qwen3-VL-8B）达到了 67.6% 的总体准确率，并且将工具链附加到十个开源 VLM 上使定量任务的准确率提高了 12.7-36.9 个百分点。这些结果表明，外部化显式的 3D 几何计算有效地补充了仅依赖 RGB 的 VLM 在估计度量距离和物体大小方面的有限能力。
cs.CV / 110 / 2609.32454
De-biasing Skeleton-based Action Recognition with Convex Hull Adaptive Shift
基于骨架的动作识别去偏差方法：凸包自适应偏移
Abstract
Skeleton sequences can represent both individual actions and multi-entity interactions, encompassing human bodies, hands, objects, and robots. Existing approaches to recognize skeleton-based actions and interactions usually adopt a late fusion strategy, which expects individuals are independent and identically distributed to train a robust weight-shared entity encoder. However, observed entity bias in various skeletal data violates this assumption, leading to suboptimal optimization of backbone models that might produce wrong recognition results. This bias arises from the world coordinate system's initial configuration, where the choice of origin often creates bias in representation. To this end, we propose a Convex Hull Adaptive Shift based normalization method to reduce Entity bias (CHASE), improving performance across a variety of skeleton-based action and interaction recognition tasks. To adaptively apply plausible shifts to the input skeletons, we formulate a plug-and-play parameterized network that ensures the relocated world origin lies within the skeleton convex hull, which avoids non-convergence by limiting the search space. To further minimize entity bias, we incorporate an auxiliary objective that leverages pair-wise distribution distances to guide network optimization. To support both single- and multi-entity actions, we propose a sub-entity strategy that offers a consistent formulation for both scenarios. Moreover, CHASE demonstrates compatibility with various intra-skeleton modalities, such as bones and velocities, highlighting its adaptability. Essentially, our method works as a normalization approach to reduce entity bias, enabling subsequent classifiers to achieve improved recognition performance across diverse settings. Extensive experiments on 7 datasets verify our approach by seamlessly integrating with various backbones and significantly boosting their performance.
Chinese Translation
骨架序列可以表示个体动作和多实体交互，包括人体、手、物体和机器人。现有的基于骨架的动作和交互识别方法通常采用后期融合策略，假设个体是独立同分布的，以训练一个稳健的权重共享实体编码器。然而，在各种骨架数据中观察到的实体偏差违反了这一假设，导致主干模型的优化效果不佳，可能产生错误的识别结果。这种偏差源于世界坐标系的初始配置，原点的选择往往会在表示上造成偏差。为此，我们提出了一种基于凸包自适应偏移的归一化方法（CHASE），以减少实体偏差，提高各种基于骨架的动作和交互识别任务的性能。为了自适应地对输入骨架施加合理的偏移，我们构建了一个即插即用的参数化网络，确保重新定位的世界原点位于骨架的凸包内，从而通过限制搜索空间避免非收敛。为了进一步最小化实体偏差，我们引入了一个辅助目标，利用成对分布距离来指导网络优化。为了支持单实体和多实体动作，我们提出了一种子实体策略，为这两种场景提供一致的公式。此外，CHASE展示了与各种骨架内部模态（如骨骼和速度）的兼容性，突显了其适应性。本质上，我们的方法作为一种归一化方法，减少了实体偏差，使后续分类器在多种环境下实现更好的识别性能。在7个数据集上的大量实验验证了我们的方法，通过与各种主干模型的无缝集成，显著提升了它们的性能。
cs.CV / 111 / 2609.32455
QuacamFM: Quaternion-Constrained Flow Matching for Camera Pose Estimation
QuacamFM：用于相机姿态估计的四元数约束流匹配
Abstract
Camera pose estimation from multi-view images remains a challenge in computer vision. Traditional methods often address this problem using Structure-from-Motion (SfM) with bundle adjustment. However, camera poses estimated from sparse views are inherently ambiguous due to insufficient geometric constraints. Recent work leverages probabilistic models, such as diffusion models, to generate multiple camera pose hypotheses and therefore capture this uncertainty better. Most of these methods represent camera rotations using unit quaternions, but treat them as unconstrained 4D vectors during the generative processes, thereby ignoring the unit-norm constraint of quaternions. Unconstrained quaternions create non-smooth and suboptimal generation trajectories. To this end, we propose *QuacamFM*, a quaternion-constrained flow matching framework for camera pose estimation that preserves unit quaternion representations throughout the entire flow trajectory. We design the optimal transport of the quaternion flows using smooth spherical linear interpolation. Experiments on CO3Dv2 demonstrate our method's advantage in camera pose accuracy over diffusion-based methods and classical SfM approaches. We further show that our quaternion-constrained formulation outperforms the naive application of standard flow matching to 4D quaternion vectors on sparse-view camera pose estimation. Finally, it is observed that QuacamFM generalizes well across datasets and in-the-wild examples.
Chinese Translation
从多视角图像中进行相机姿态估计仍然是计算机视觉中的一个挑战。传统方法通常通过运动重建（Structure-from-Motion, SfM）结合束调整来解决这个问题。然而，由于几何约束不足，从稀疏视图估计的相机姿态本质上是模糊的。最近的研究利用概率模型，如扩散模型，生成多个相机姿态假设，从而更好地捕捉这种不确定性。这些方法大多数使用单位四元数来表示相机旋转，但在生成过程中将其视为无约束的4D向量，从而忽略了四元数的单位范数约束。无约束的四元数会导致生成轨迹不平滑且次优。为此，我们提出了*QuacamFM*，一个用于相机姿态估计的四元数约束流匹配框架，能够在整个流动轨迹中保持单位四元数表示。我们使用平滑的球面线性插值设计四元数流的最优传输。在CO3Dv2数据集上的实验表明，我们的方法在相机姿态精度上优于基于扩散的方法和经典的SfM方法。我们进一步展示了我们的四元数约束公式在稀疏视图相机姿态估计中优于将标准流匹配简单应用于4D四元数向量的效果。最后，我们观察到QuacamFM在不同数据集和真实场景示例中具有良好的泛化能力。
cs.CV / 112 / 2609.32456
Seeing Parts, Reasoning about Worlds: Visual Inference under Partial Observation
观察部分，推理世界：在部分观察下的视觉推理
Abstract
World modeling under partial observation requires reasoning about the complete worlds that remain compatible with limited visual evidence. Occluded objects and unseen regions can leave several world states possible; additional views can exclude alternatives and strengthen the conclusions supported by the observations. We introduce WorldScope to study this process through possible-world semantics, evidence-grounded data, and learned visual representations. WorldScope-1.2M provides 1.2 million English question-answer pairs spanning eight world properties, ten task interfaces, and three observation protocols. Its answers encode confirmed facts, supported bounds, and unresolved possibilities. Complementary supervision comprises 4,800 certified counterworld groups with equivalent base observations and different hidden object configurations and query answers. These groups provide physical witnesses of ambiguity and training-only labels for world compatibility and view-induced exclusions. We propose WorldFlow, which composes cross-view entity evidence and support-surface coverage into an image-subset evidence lattice. Counterworld compatibility and transition objectives train subset representations to reflect how new observations constrain possible worlds. A shared answer generator uses these representations to predict the strongest supported conclusion. WorldScope-Bench evaluates claim judgments and evidence-dependent conclusions as views are selected, combined, removed, or ordered. On its 5,000-question test set, WorldFlow reaches 64.34% exact accuracy, improving over the same backbone trained on QA alone by 24.88 percentage points. It retains 50.43% accuracy on the 3,000 questions from structure-disjoint scenes.
Chinese Translation
在部分观察下的世界建模需要推理与有限视觉证据相兼容的完整世界。被遮挡的物体和未见区域可能导致多个世界状态；额外的视角可以排除替代方案，并加强由观察支持的结论。我们引入了WorldScope来通过可能世界语义、基于证据的数据和学习的视觉表示来研究这一过程。WorldScope-1.2M提供了120万个涵盖八个世界属性、十个任务接口和三个观察协议的英语问答对。其答案编码了确认的事实、支持的界限和未解决的可能性。补充监督包括4800个经过认证的反世界组，这些组具有等效的基础观察和不同的隐藏物体配置及查询答案。这些组提供了模糊性的物理见证和仅用于训练的世界兼容性及视图引起的排除标签。我们提出了WorldFlow，它将跨视图实体证据和支持表面覆盖组合成一个图像子集证据格。反世界兼容性和过渡目标训练子集表示，以反映新观察如何限制可能的世界。一个共享答案生成器利用这些表示来预测最强支持的结论。WorldScope-Bench评估声明判断和依赖证据的结论，因为视图被选择、组合、移除或排序。在其5000个问题的测试集中，WorldFlow达到了64.34%的准确率，比仅在问答上训练的相同骨干提高了24.88个百分点。在来自结构不相交场景的3000个问题中，它保持了50.43%的准确率。
cs.CV / 113 / 2609.32460
REMEDY: How Far Is Video Generation from Medical Education World Models?
REMEDY：视频生成距离医学教育世界模型还有多远？
Abstract
Recent video generation models produce realistic videos and show potential as a foundation for world models. These advances create opportunities for generating medical teaching demonstrations, which requires both convincing visual quality and precise procedural actions. However, whether current generators can meet these requirements has not been measured. To address this problem, we introduce Readiness Evaluation of Medical Education Demonstration sYnthesis (REMEDY), to our knowledge, the first benchmark for AI-generated medical teaching demonstrations. REMEDY provides 900 first frames from real demonstration videos, covering 12 tasks across four scenarios: operating room, imaging, clinic and bedside, and resuscitation. Five contemporary open-source video generation models produce 4,500 videos from these frames. We combine task-specific clinical checklists with video and motion quality metrics. Evaluation covers four dimensions: clinical action following, clinical profiles, video quality, and motion quality. Our results show that realistic appearance and temporal consistency do not ensure correct clinical actions. Even the most advanced MiniMax-H3 achieves only 28.25% on strict clinical success rate, and fine-grained clinical actions remain challenging. These findings establish a foundation and roadmap for developing future medical education world models.
Chinese Translation
近期的视频生成模型能够生成逼真的视频，并显示出作为世界模型基础的潜力。这些进展为生成医学教学演示提供了机会，但这需要既具说服力的视觉质量，又要精准的操作动作。然而，目前的生成器是否能够满足这些要求尚未被评估。为了解决这一问题，我们引入了医学教育演示合成的准备评估（Readiness Evaluation of Medical Education Demonstration sYnthesis，REMEDY），据我们所知，这是首个针对AI生成医学教学演示的基准。REMEDY提供了900个来自真实演示视频的首帧，涵盖了四种场景下的12个任务：手术室、影像学、诊所和床边急救。五个当代开源视频生成模型从这些帧中生成了4500个视频。我们将特定任务的临床检查表与视频和运动质量指标相结合。评估涵盖四个维度：临床动作遵循、临床特征、视频质量和运动质量。我们的结果表明，逼真的外观和时间一致性并不能确保正确的临床动作。即使是最先进的MiniMax-H3在严格的临床成功率上也仅达到28.25%，而细粒度的临床动作仍然具有挑战性。这些发现为未来医学教育世界模型的开发奠定了基础和路线图。
cs.CV / 114 / 2609.32462
Can Motion-Language Models Ground Structure? STRIDE for Evaluating the Evaluators
运动语言模型能否扎根于结构？用于评估评估者的 STRIDE
Abstract
Motion-language models are typically scored by motion-language evaluators, but how well these evaluators ground language structure remains unclear. Here, we introduce the Structure grounding via Temporal-order, Reflection, and Identity Diagnostic Evaluation (STRIDE) benchmark to systematically evaluate the ability of evaluators to track temporal order, mirror reflection, and action identity. STRIDE comprises $5{,}869$ triples, each consisting of a motion, its original caption, and a perturbed caption, spanning both short and long descriptions. We likelihood-balance caption pairs to reduce text-only bias and estimate each evaluator's caption preference under unrelated motions to measure the discrimination gain from matched motions relative to this baseline. Our experiments reveal weak structural grounding and severe deficits in mirror sensitivity among the audited evaluators, which commonly used evaluation protocols fail to expose. To understand why these limitations go undetected in standard tests, we examine the evaluators more closely. We find that text-only priors alone can solve naive perturbation tests on existing datasets, while common retrieval and distributional metrics barely respond to structural corruption introduced by mirroring ground-truth motions. These findings suggest a natural intervention: structural hard negatives. Our experiments show that a simple modification to contrastive learning substantially improves performance on temporal order and mirror reflection. The benchmark and code will be released.
Chinese Translation
运动语言模型通常由运动语言评估者进行评分，但这些评估者在多大程度上能够扎根于语言结构仍然不清楚。在此，我们引入了通过时间顺序、镜像反射和动作身份诊断评估（Structure grounding via Temporal-order, Reflection, and Identity Diagnostic Evaluation，STRIDE）基准，以系统地评估评估者跟踪时间顺序、镜像反射和动作身份的能力。STRIDE 包含 $5{,}869$ 个三元组，每个三元组由一个动作、其原始标题和一个扰动标题组成，涵盖了短描述和长描述。我们对标题对进行可能性平衡，以减少仅文本的偏差，并在无关动作下估计每个评估者的标题偏好，以测量相对于这一基线的匹配动作所带来的区分增益。我们的实验揭示了被审计评估者在结构扎根方面的薄弱表现和镜像敏感性方面的严重缺陷，而常用的评估协议未能揭示这些问题。为了理解为何这些局限性在标准测试中未被发现，我们对评估者进行了更深入的检查。我们发现，仅依靠文本先验就能解决现有数据集上的简单扰动测试，而常见的检索和分布度量几乎对镜像真实动作引入的结构损坏没有反应。这些发现提示了一种自然的干预：结构硬负样本。我们的实验表明，对对比学习的简单修改显著提高了在时间顺序和镜像反射方面的表现。该基准和代码将被发布。
cs.CV / 115 / 2609.32477
Back-Tracking from Clarity: Self-Learning to See Text from Afar
从清晰回溯：自学习远距离识别文本
Abstract
We propose a self-supervised framework designed to enhance the capability of scene text detectors in identifying and recognizing text in scenarios where instances are shown at significant distances, typically small, blurred, and frequently missed by conventional models. Our approach leverages the high-fidelity performance of existing text spotting models on large, clear text as a foundational supervisor. By temporally back-tracking these high-confidence detections through video sequences, we automatically synthesize pseudo-labels for preceding frames where the distant text is still visually degraded or undersized. These pseudo-labels enable training a student model specialized for early text detection, without requiring any manual annotation. The success of this approach depends on accurate pseudo-label generation, for which we develop a dedicated scene text tracker capable of maintaining consistent text identities across challenging video sequences. In addition, we propose SceneText50, a diverse multilingual outdoor dataset to facilitate training and evaluation. Experiments show that our framework significantly improves early detection accuracy and robustness across varied scenes and languages. Code and data are at \href{https://github.com/trid2912/BackTrackingText}{https://github.com/trid2912/BackTrackingText}.
Chinese Translation
我们提出了一种自监督框架，旨在增强场景文本检测器在识别和识别远距离文本时的能力，这些文本通常较小、模糊，并且常常被传统模型遗漏。我们的方法利用现有文本检测模型在大而清晰文本上的高保真性能作为基础监督。通过在视频序列中对这些高置信度检测结果进行时间回溯，我们自动合成前帧的伪标签，在这些帧中，远处的文本仍然在视觉上退化或尺寸较小。这些伪标签使得我们能够训练一个专注于早期文本检测的学生模型，而无需任何人工标注。该方法的成功依赖于准确的伪标签生成，为此我们开发了一个专用的场景文本跟踪器，能够在具有挑战性的视频序列中保持一致的文本身份。此外，我们提出了SceneText50，一个多样化的多语言户外数据集，以促进训练和评估。实验表明，我们的框架显著提高了在不同场景和语言中的早期检测准确性和鲁棒性。代码和数据可在 [https://github.com/trid2912/BackTrackingText](https://github.com/trid2912/BackTrackingText) 获取。
cs.CV / 116 / 2609.32481
JEPA Learns What the Mask Leaves Unrecoverable
JEPA学习掩码留下的不可恢复内容
Abstract
Joint-embedding predictive architectures are unusually sensitive to how the input is masked: block masks work, scattered masks do not, and the explanations are empirical. We give a measurement account. A mask is a linear measurement, and in a compactly supported wavelet basis every atom whose support lies inside the hidden region falls in the measurement's null space and leaves no trace in the data. The JEPA loss asks only that the encoded context suffice for the target, so a target that a low-level prior can recover admits a shortcut, one the moving-average target encoder can make self-consistent. What removes the shortcut is the coarse-scale content the mask leaves unrecoverable, provided enough context stays within reach of each target. We score that content before training and test the account's distinctive predictions in 151 pre-training runs. On ImageNet-100, strip masks match blocks in area and contiguity yet are recoverable, and they land at 40.3% linear top-1, beside random masks at 40.8%, against 64.3% for blocks; within one geometry family, the placements that leave the least unrecoverable content lose 6.5 points to those that leave the most, over five seed pairs; pixel targets span 7 points where latent targets span 25; and against a frozen target the gap between random and block masks, 19 points on the same kind of GPU, closes to 1.5, so the geometry acts through the target the encoder produces for itself. On UCF101 the masking ratio decides which condition, content or reach, binds; removing whole frames, unrecoverable in space but recoverable from neighbouring frames, is worst at both ratios; and on V-JEPA's own masks, batching them intact instead of truncated changes little (36.0% against 35.1%), whereas making 100 target tokens inside the blocks visible lifts them to 48.7% and hiding 100 context tokens outside the blocks does not (33.7%).
Chinese Translation
联合嵌入预测架构对输入的掩码方式异常敏感：块状掩码有效，而分散掩码无效，且其解释是经验性的。我们提供了一种测量解释。掩码是一种线性测量，在紧支撑小波基中，所有支持位于隐藏区域内的原子都落入测量的零空间，并在数据中不留痕迹。JEPA损失只要求编码的上下文足以满足目标，因此一个低级先验能够恢复的目标允许存在捷径，而移动平均目标编码器可以使其自洽。移除捷径的是掩码留下的粗尺度内容，只要足够的上下文保持在每个目标的可及范围内。我们在训练前对该内容进行评分，并在151次预训练中测试该解释的独特预测。在ImageNet-100上，条形掩码在面积和连通性上与块状掩码匹配，但是可恢复的，其线性top-1准确率为40.3%，而随机掩码为40.8%，块状掩码为64.3%；在同一几何族中，留下最少不可恢复内容的放置比留下最多的低6.5分，跨越五对种子；像素目标跨越7分，而潜在目标跨越25分；在固定目标下，随机掩码与块状掩码之间的差距（在同一类型的GPU上为19分）缩小至1.5，因此几何通过编码器为自身生成的目标发挥作用。在UCF101上，掩码比例决定了内容或可及性哪个条件约束；移除整个帧，在空间上不可恢复但可从相邻帧恢复，在两个比例下都是最糟的；而在V-JEPA自己的掩码上，完整批处理而不是截断几乎没有变化（36.0%对35.1%），而在块内使100个目标标记可见将其提升至48.7%，而将100个上下文标记隐藏在块外则没有提升（33.7%）。
cs.CV / 117 / 2609.32510
GeoCR: Learning a Generalist Cloud Removal Prior from Heterogeneous Observations
GeoCR：从异构观测中学习通用云去除先验
Abstract
Cloud removal methods are typically specialized to individual datasets and input configurations, limiting reuse across sensors, spectral bands, and observation settings. We introduce GeoCR, a generalist model that unifies RGB-only-based CR and multispectral-based CR from single- or multi-temporal cloudy observations, with optional SAR guidance, within a single network. To accommodate different spectral and sensing domains, compact input and output stems extend a pretrained RGB autoencoder while keeping its encoder and decoder trunks frozen. This shared latent interface enables a single flow transformer to jointly model clean RGB and non-RGB latents, conditioned on separate cloudy-observation streams and optional SAR tokens. Through joint pretraining on the training splits of ten datasets comprising 883,331 cloud-free target images, GeoCR learns a shared cloud removal prior across these heterogeneous configurations. The same pretrained checkpoint supports direct inference without dataset-specific fine-tuning and efficient adaptation through low-rank adaptation (LoRA). We evaluate GeoCR against general image restoration and cloud removal methods on test splits of the contributing datasets under full-band and RGB-only settings. GeoCR achieves the best FID and DISTS on full-band SEN12MS-CR and Sen2_MTC_New and RGB-only CUHK-CR2, outperforming existing models and demonstrating the effectiveness of a reusable generative model across diverse settings.
Chinese Translation
云去除方法通常针对单个数据集和输入配置进行专业化，限制了在传感器、光谱波段和观测设置之间的重用。我们提出了GeoCR，这是一种通用模型，统一了基于仅RGB的云去除（CR）和基于多光谱的CR，适用于单时相或多时相的多云观测，并可选择性地结合SAR指导，均在一个网络中实现。为了适应不同的光谱和传感领域，紧凑的输入和输出结构扩展了一个预训练的RGB自编码器，同时保持其编码器和解码器的主干不变。这个共享的潜在接口使得单一流变换器能够联合建模干净的RGB和非RGB潜在变量，条件是基于不同的多云观测流和可选的SAR标记。通过在十个数据集的训练分割上进行联合预训练，这些数据集包含883,331张无云目标图像，GeoCR学习了在这些异构配置下的共享云去除先验。相同的预训练检查点支持直接推断，而无需特定于数据集的微调，并通过低秩适应（LoRA）实现高效适应。我们在贡献数据集的测试分割上评估GeoCR，相较于一般图像恢复和云去除方法，在全波段和仅RGB设置下表现出色。GeoCR在全波段SEN12MS-CR和Sen2_MTC_New以及仅RGB的CUHK-CR2上取得了最佳的FID和DISTS，超越了现有模型，展示了可重用生成模型在多样化设置中的有效性。
cs.CV / 118 / 2609.32513
Cross-Domain Few-Shot Writer Adaptation for Real-World Handwritten Mathematical Expression Recognition
跨域少样本写作者适应用于现实世界手写数学表达识别
Abstract
Handwritten mathematical expression recognition (HMER) refers to the task of recognizing and converting handwritten mathematics into a parsable markup language, usually LaTeX. No current state-of-the-art-competitive system adjusts to the way a specific person writes, and the domain gap between training images (usually digital or perfectly binarized) and images physically taken with a camera used in inference has received fairly little attention for this specific problem. We characterize this domain gap through fragmentation and stroke-width analyses of the images as well as introduce a writer-adaptive fine-tuning pipeline to MFH-CoMER in an attempt to address it. We further introduce sample author-specific datasets, consisting of five handwriting category subsets from two authors, and evaluate using a McNemar's test and permutation tests adapted to limited data. Results suggest an increase in expression recognition rate and a decrease in CER for digital handwriting but more varied results for physical handwriting, with the model struggling for handwriting articles that the base model can already evaluate well. Nonetheless, adapted models were found to have improved results for four out of five subsets. Statistical testing results suggest a consistency in improvement for two of the tested author-specific subsets. Our results point towards the potential feasibility of writer adaptation for the HMER task.
Chinese Translation
手写数学表达识别（HMER）是指将手写数学识别并转换为可解析的标记语言，通常是 LaTeX 的任务。目前没有任何具有竞争力的系统能够适应特定个体的书写方式，并且训练图像（通常是数字化或完美二值化的）与用于推理的通过相机拍摄的图像之间的领域差距在这个特定问题上受到的关注相对较少。我们通过对图像的碎片化和笔画宽度分析来表征这一领域差距，并引入了一种针对 MFH-CoMER 的写作者自适应微调管道以尝试解决该问题。我们进一步引入了样本作者特定数据集，由两位作者的五个书写类别子集组成，并使用 McNemar 检验和适应于有限数据的置换检验进行评估。结果表明，数字手写的表达识别率有所提高，字符错误率（CER）有所降低，但对于物理手写的结果则更为多样，模型在评估基础模型已经能够良好评估的手写文章时遇到困难。尽管如此，经过适应的模型在五个子集中的四个子集上取得了更好的结果。统计检验结果表明，在两个经过测试的作者特定子集中改进的一致性。我们的结果指向了手写数学表达识别任务中写作者适应的潜在可行性。
cs.CV / 119 / 2609.32518
UnStep: Training-Free Acceleration of Causal Video Diffusion with Fewer Steps Than Distillation
UnStep：无训练加速因果视频扩散，步骤少于蒸馏
Abstract
Distilling bidirectional multi-step video diffusion transformers into few-step causal models has become a common approach for streaming video generation. While these few-step students are significantly faster than the teachers they are distilled from, they remain slow for real-time generation. In this work we present UnStep, a training-free wrapper that accelerates few-step causal video models at inference by running them with fewer diffusion transformer (DiT) steps than during distillation and limiting the temporal window retained in the attention KV cache. We propose two inference-only mechanisms to recover quality lost by step reduction and attention windowing: renoising the generated latent frames to a near-clean level and reusing the existing clean-cache pass to refine them, and applying truncated SVD to the DiT attention value and output projections. We also accelerate inference with a quality-preserving runtime stack for the DiT and VAE decoder, including more efficient attention calls and KV indexing, fused Triton RoPE with cached coefficients, and VAE decoding with optimized memory layout, precision, and convolution kernels. By reducing computation and optimizing the runtime stack, UnStep sets a new throughput regime for causal video diffusion, by running substantially faster than current methods, reaching 50 FPS on a single H100 without quality loss, and 77 FPS on GB200, all without retraining.
Chinese Translation
将双向多步视频扩散变换器蒸馏为少步因果模型已成为流媒体视频生成的常见方法。尽管这些少步学生模型的生成速度显著快于其蒸馏来源的教师模型，但在实时生成方面仍然较慢。在本研究中，我们提出了UnStep，这是一种无训练的封装器，通过以比蒸馏过程中更少的扩散变换器（DiT）步骤运行少步因果视频模型，并限制在注意力KV缓存中保留的时间窗口，从而加速推理过程。我们提出了两种仅用于推理的机制，以恢复因步骤减少和注意力窗口限制而损失的质量：将生成的潜在帧重新去噪至接近干净的水平，并重用现有的干净缓存传递进行精细化处理，以及对DiT注意力值和输出投影应用截断奇异值分解（SVD）。我们还通过为DiT和变分自编码器（VAE）解码器加速推理，采用更高效的注意力调用和KV索引、融合的Triton RoPE与缓存系数，以及优化的内存布局、精度和卷积核进行VAE解码，来保持质量的运行时堆栈。通过减少计算和优化运行时堆栈，UnStep为因果视频扩散设定了新的吞吐量标准，其运行速度显著快于当前方法，在单个H100上达到50 FPS且没有质量损失，在GB200上达到77 FPS，且无需重新训练。
cs.CV / 120 / 2609.32529
SetOPD: From Few Visual Exemplars to Multimodal Candidate Sets for Remote-Sensing Open-Prompt Detection
SetOPD：从少量视觉示例到多模态候选集的遥感开放提示检测
Abstract
Open-prompt detectors allow users to specify targets with text, visual exemplars, or both. We argue that existing designs underuse the visual modality in two ways. First, multiple exemplars are commonly compressed into a single class-level embedding. This textualizes visual prompting: the resulting vector plays the role of another class name, is often aligned to or injected into the text pathway, and may be suboptimal when only a few heterogeneous exemplars are available. Second, existing methods interact primarily in prompt or representation space, before modality-specific detection states are formed. We address both issues from a set perspective: \setopd preserves modality-specific decoding states from a shared prompt-conditioned initialization and performs explicit multimodal collaboration at the candidate-state level. For the first issue, we introduce \br prompting, which reads every boxed exemplar in its full scene context and decomposes the pooled support evidence into a Base anchor and a learned Residual correction; the resulting prompt has fixed capacity regardless of the number of examples and drives its own visual detection pathway. For the second, we recast text--visual collaboration from representation-level fusion into a candidate-set modeling problem. Paired-Query Arbitration (\pqa) then performs explicit cross-modal state arbitration only after modality-specific candidate states have been formed. The two readers share query initialization so their candidates are paired by index; a learned gate arbitrates within each pair, followed by a permutation-equivariant module that reasons over the fused set.
Chinese Translation
开放提示检测器允许用户通过文本、视觉示例或两者指定目标。我们认为现有设计在两个方面未充分利用视觉模态。首先，多个示例通常被压缩为单一的类别级嵌入。这使得视觉提示文本化：生成的向量充当另一个类别名称，通常与文本路径对齐或注入，并且在仅有少量异质示例可用时可能表现不佳。其次，现有方法主要在提示或表示空间中进行交互，而在形成特定模态的检测状态之前。我们从集合的角度解决这两个问题： extit{SetOPD} 保留来自共享提示条件初始化的模态特定解码状态，并在候选状态层面上执行显式的多模态协作。针对第一个问题，我们引入了 extit{br prompting}，它在完整场景上下文中读取每个框选示例，并将汇总的支持证据分解为基础锚点和学习的残差修正；生成的提示具有固定的容量，无论示例数量如何，并驱动其自身的视觉检测路径。对于第二个问题，我们将文本-视觉协作从表示级融合重新构建为候选集建模问题。配对查询仲裁（ extit{PQA}）则在形成特定模态的候选状态后，执行显式的跨模态状态仲裁。两个读取器共享查询初始化，因此它们的候选通过索引配对；一个学习的门控在每对内进行仲裁，随后是一个置换等变模块，对融合集进行推理。
cs.CV / 121 / 2609.32534
DepthBench: Measuring How Residual Connections Enable More Computational Depth
DepthBench：测量残差连接如何使计算深度更大
Abstract
Depth is a natural way to increase the computational capacity in Transformers, yet the contribution of deeper layers can diminish as depth grows larger. Recent approaches enhance normalization (\text{e.g.}, LayerNorm Scaling) or residual connections (\text{e.g.}, mHC, AttnRes) to enable better information flow and depth utilization. However, it remains unclear whether they truly translate increased architectural depth into effective computational depth, and whether their reported gains stem from better access to information across depth, or unaccounted-for confounding factors. In this paper, we introduce \textbf{DepthBench}, a controlled benchmark for studying computational depth across various architectures. We systematically vary the width--depth aspect ratio ($d_{\text{model}}/n_{\text{layer}}$) from shallow--wide to deep--narrow shapes, while keeping the model size and pre-training recipe fixed. Across 10 representative architectures, we find that the benefit of allocating more capacity to depth is strongly architecture-dependent. Standard Pre-LN and most of its norm- and scaling-based variants provide little benefit and can even degrade performance as models become deeper and narrower, whereas HC and Full AttnRes improve consistently even at extreme deep shapes. These gains extend beyond pre-training loss and consistently translate into improved domain-specific performance and effective computation. Controlled layer-level analyses further show that the gains of HC and Full AttnRes are associated with more effective utilization of additional layers, revealing distinct mechanisms of computational depth across architectures. Overall, our results identify residual connection design as a key determinant of whether depth can serve as a meaningful scaling axis by enabling additional architectural depth to translate into effective computation.
Chinese Translation
深度是增加变换器（Transformers）计算能力的自然方式，但随着深度的增加，深层的贡献可能会减小。最近的方法通过增强归一化（例如，LayerNorm Scaling）或残差连接（例如，mHC，AttnRes）来实现更好的信息流动和深度利用。然而，目前尚不清楚这些方法是否真正将增加的架构深度转化为有效的计算深度，以及它们报告的收益是否源于更好的信息访问，还是由于未考虑的混杂因素。在本文中，我们介绍了 extbf{DepthBench}，一个用于研究各种架构中计算深度的受控基准。我们系统地改变宽度-深度比（$d_{ ext{model}}/n_{ ext{layer}}$），从浅-宽到深-窄的形状，同时保持模型大小和预训练方案不变。在10种代表性架构中，我们发现将更多容量分配给深度的好处在很大程度上依赖于架构。标准的Pre-LN及其大多数基于归一化和缩放的变体提供的好处很小，甚至在模型变得更深更窄时会降低性能，而HC和Full AttnRes在极端深度形状下也能持续改善。这些收益超越了预训练损失，并始终转化为改进的领域特定性能和有效计算。受控的层级分析进一步表明，HC和Full AttnRes的收益与额外层的更有效利用相关，揭示了不同架构中计算深度的独特机制。总体而言，我们的结果确定了残差连接设计是深度能否作为一个有意义的扩展轴的关键因素，从而使额外的架构深度转化为有效计算。
cs.CV / 122 / 2609.32537
RIPE-MambaSpike: Resolution-Independent Spiking-State-Space Interfaces for Parameter-Efficient Event-Based Vision
RIPE-MambaSpike：用于参数高效事件驱动视觉的分辨率无关脉冲状态空间接口
Abstract
Spiking-Mamba hybrids reach strong accuracy on event-based vision, but existing designs often require tens of millions of parameters. Much of that cost comes from how the spiking front-end is connected to the state-space backbone rather than from the hybrid architecture itself. In a representative model, a single resolution-dependent projection accounts for 33.55M of 36.25M parameters. To that end, we introduce RIPE-MambaSpike (Resolution-Independent, Parameter-Efficient), which replaces that projection with a hierarchical multi-resolution bridge of fixed channel width. Its deployed footprint is 0.870M parameters, constant at fixed time steps and widths across a 43x range of input areas. Reparameterized spiking stages, temporal decoupled modulation, and a dynamic convex-hull-bounded dual-stream membrane-potential attention preserve accuracy under this compact design. Result-wise, RIPE-MambaSpike is pareto-optimal on CIFAR10-DVS, N-Caltech101, and DailyDVS-200. Notably, on the 200-class DailyDVS-200, a scaled 8.04M configuration achieves 45.7% top-1 accuracy, the best reported spiking result on that benchmark, and outperforms prior spiking methods with 3.0-15.1x fewer parameters than dense ANNs. Overall, our findings demonstrate that competitive event-based recognition does not require resolution-dependent parameter growth. Code is available at https://github.com/MuhiminOsim/RIPE-MambaSpike.
Chinese Translation
脉冲-Mamba混合模型在事件驱动视觉上达到了较高的准确性，但现有设计通常需要数千万个参数。大部分成本来自脉冲前端与状态空间主干的连接方式，而非混合架构本身。在一个代表性模型中，单个依赖分辨率的投影占据了36.25M参数中的33.55M。为此，我们提出了RIPE-MambaSpike（分辨率无关，参数高效），它用固定通道宽度的层次多分辨率桥替代了该投影。其部署的参数占用为0.870M，在固定时间步和宽度下，适用于43倍范围的输入区域。重新参数化的脉冲阶段、时间解耦调制以及动态凸包约束的双流膜电位注意机制在这一紧凑设计下保持了准确性。结果显示，RIPE-MambaSpike在CIFAR10-DVS、N-Caltech101和DailyDVS-200上均为帕累托最优。值得注意的是，在200类的DailyDVS-200上，缩放后的8.04M配置达到了45.7%的顶级准确率，这是该基准上报告的最佳脉冲结果，并且在参数数量上比稠密人工神经网络少3.0-15.1倍。总体而言，我们的研究结果表明，竞争性的事件驱动识别并不需要依赖分辨率的参数增长。代码可在https://github.com/MuhiminOsim/RIPE-MambaSpike获取。
cs.CV / 123 / 2609.32540
In-Flight KV Cache with Clean Anchors for Faster Autoregressive Video Diffusion
具有清晰锚点的飞行KV缓存以加快自回归视频扩散
Abstract
Few-step autoregressive video diffusion generates a long video by splitting the video into temporal chunks and generating chunk-by-chunk, each through a short sequence of denoising stages. To memorize chunks that are already generated, previous methods reconstruct a clean or less-noisy key--value (KV) cache by additional forwards to build the cache without advancing an output latent. However, every denoising forward itself already computes the in-flight KV of the current chunk. We introduce FlashForward, which directly reuses this cache to avoid the heavy cache-update-only model forwards. After the current chunk completes one denoising stage, its stage-specific cache is already available for the next chunk. Assigning one GPU to each stage therefore lets different chunks occupy different stages concurrently. This early availability has a quality cost: the resulting stage-matched history is noisy, causing appearance and motion drift among chunks. To complement it, FlashForward produces sparse auxiliary clean anchor latents before the corresponding region is generated so the generation trajectories can be stabilized by this two-sided conditioning. The two memories operate at different temporal scales: sparse clean anchor KV supplies coarse, long-range two-sided structural guidance, while dense stage-matched history preserves fine, recent evolution. With up to four GPUs, FlashForward runs $1.16$--$1.69\times$ faster than HiAR and $1.42$--$2.92\times$ faster than Self-Forcing for 16 FPS videos of 20 seconds or longer across 1.3B and 14B backbone scales at 480p and 720p. On VBench, for the 1.3B model at 480p, it achieves higher scores and remains stable at longer durations, demonstrating that FlashForward generates high-quality and temporally consistent videos across durations of 20s, 35s and 65s at a much faster generation speed.
Chinese Translation
少步自回归视频扩散通过将视频分割为时间块并逐块生成来生成长视频，每个块通过一系列短的去噪阶段进行生成。为了记忆已经生成的块，之前的方法通过额外的前向计算重建一个干净或低噪声的键值（KV）缓存，以在不推进输出潜变量的情况下构建缓存。然而，每个去噪前向计算本身已经计算了当前块的飞行KV。我们提出了FlashForward，它直接重用这个缓存，以避免繁重的仅缓存更新模型的前向计算。在当前块完成一个去噪阶段后，其阶段特定的缓存已经可用于下一个块。因此，将一个GPU分配给每个阶段可以让不同的块同时占用不同的阶段。这种提前可用性有一个质量成本：结果的阶段匹配历史是噪声的，导致块之间的外观和运动漂移。为了补充这一点，FlashForward在生成相应区域之前产生稀疏的辅助清晰锚点潜变量，以便通过这种双向条件稳定生成轨迹。这两种记忆在不同的时间尺度上操作：稀疏的清晰锚点KV提供粗略的、长距离的双向结构指导，而密集的阶段匹配历史则保留细致的、近期的演变。在使用多达四个GPU的情况下，FlashForward的运行速度比HiAR快$1.16$--$1.69 imes$，比Self-Forcing快$1.42$--$2.92 imes$，适用于480p和720p的16 FPS视频，时长为20秒或更长，涵盖了1.3B和14B的主干规模。在VBench上，对于480p的1.3B模型，它获得了更高的分数，并在更长的持续时间内保持稳定，证明FlashForward在20秒、35秒和65秒的持续时间内生成高质量且时间一致的视频，且生成速度更快。
cs.CV / 124 / 2609.32551
Harnessing Coupled Stream Completion For Human-Object Ineraction Modeling
利用耦合流完成技术进行人机交互建模
Abstract
Text-conditioned human-object interaction (HOI) generation requires body motion, object trajectories & rotations, and hand articulation to remain coordinated. These components differ in scale and dynamics, but must agree on contact, relative pose, and timing. A shared representation may limit the distinct structure of each stream, while independent generation prevents each stream from responding to changes in the others. Latent supervision alone also does not directly constrain contact after decoding. We propose TRACE, a continuous latent framework that keeps stream states separate and couples their updates. TRACE encodes body, object, and hand motion into separate latents and predicts each stream velocity from the complete current interaction state. Geometric losses on decoded motion further constrain contact and object-relative motion over time. The same model supports completion of any single absent stream from the other two. Frozen flow features also serve as input to a language model for HOI understanding. Experiments on InterAct, OMOMO, and BEHAVE show that joint completion training improves generation and that frozen flow features improve understanding over raw-motion encoding. On InterAct, TRACE achieves the highest contact precision, recall, and F1 among the compared methods.
Chinese Translation
基于文本的人机交互（HOI）生成需要身体运动、物体轨迹与旋转以及手部动作保持协调。这些组件在尺度和动态上有所不同，但必须在接触、相对姿态和时序上达成一致。共享表示可能限制每个流的独特结构，而独立生成则使每个流无法响应其他流的变化。仅依靠潜在监督也无法在解码后直接约束接触。我们提出了TRACE，一个连续的潜在框架，保持流状态的独立性并耦合它们的更新。TRACE将身体、物体和手部运动编码为独立的潜在变量，并从完整的当前交互状态预测每个流的速度。对解码运动的几何损失进一步约束了接触和物体相对运动随时间的变化。同一模型支持从其他两个流完成任何单一缺失流。冻结流特征也作为输入用于HOI理解的语言模型。在InterAct、OMOMO和BEHAVE上的实验表明，联合完成训练改善了生成效果，而冻结流特征在理解上优于原始运动编码。在InterAct上，TRACE在比较方法中实现了最高的接触精度、召回率和F1值。
cs.CV / 125 / 2609.32555
Feature Space Guidance for Breast Cancer Classification in DCE-MRI
基于特征空间的动态对比增强乳腺MRI乳腺癌分类指导
Abstract
Dynamic contrast enhanced breast MRI (DCE-MRI) is a powerful clinical tool for breast cancer detection, providing high resolution anatomical detail together with rich temporal contrast information. However, high dimensional 4D inputs, small lesions, and heterogeneous acquisition protocols across clinical sites hinder robust automated classification of healthy, benign, and malignant cases. To address these challenges, we propose a framework that dynamically analyzes latent representations to adapt to protocol-specific characteristics. Spatial variability is mitigated by reducing confounding background uptake and compensating for misalignment caused by deformable soft tissue. Additionally, relationships in the latent space across phases are leveraged to select the most informative temporal features, improving robustness to protocol-specific temporal variability. Finally, task specific discriminative features are promoted through large scale supervised lesion segmentation pretraining, which substantially enhances downstream finetuning. Evaluated under leave-one-center-out validation on the ODELIA dataset and the held-out AMBL cohort, the proposed framework substantially outperforms finetuned radiology foundation models and prior methods, improving mean AUROC by nearly 8 points and balanced accuracy by 4 points over the strongest baseline. Additionally, our method achieved first place in the MICCAI ODELIA Breast MRI Challenge 2025, further demonstrating its effectiveness for robust breast cancer classification. We publicly release our codebase under https://github.com/MIC-DKFZ/CURIAtor.
Chinese Translation
动态对比增强乳腺MRI（DCE-MRI）是一种强大的临床工具，用于乳腺癌检测，能够提供高分辨率的解剖细节以及丰富的时间对比信息。然而，高维的4D输入、小病灶以及临床场所之间异质的采集协议，阻碍了健康、良性和恶性病例的稳健自动分类。为了解决这些挑战，我们提出了一种框架，动态分析潜在表示以适应特定协议的特征。通过减少混淆的背景摄取并补偿由可变形软组织引起的错位，降低了空间变异性。此外，利用各阶段潜在空间中的关系选择最具信息量的时间特征，从而提高了对特定协议时间变异性的稳健性。最后，通过大规模监督病灶分割预训练，促进了任务特定的判别特征，这显著增强了下游微调的效果。在ODELIA数据集和保留的AMBL队列上进行的留一中心验证中，所提框架显著优于微调的放射学基础模型和先前的方法，平均AUROC提高了近8个点，平衡准确率提高了4个点。此外，我们的方法在2025年MICCAI ODELIA乳腺MRI挑战赛中获得第一名，进一步证明了其在稳健乳腺癌分类中的有效性。我们公开发布了我们的代码库，网址为https://github.com/MIC-DKFZ/CURIAtor。
cs.CV / 126 / 2609.32559
CFCH: Coarse-Fine Collaborative Hierarchical Learning for Anterior Segment Disease Analysis
CFCH：用于前节疾病分析的粗细协作层次学习
Abstract
Accurate classification of anterior segment diseases is crucial for ophthalmic screening and diagnosis. However, slit-lamp image analysis remains challenging due to substantial variability in imaging conditions and the intrinsic anatomical-disease hierarchy of ocular pathologies. Existing methods typically formulate this task as a flat multi-class classification problem, ignoring the structured dependency between anatomical regions (e.g., cornea, conjunctiva, and lens) and disease manifestations.To address these limitations, we propose CFCH, a Coarse-Fine Collaborative Hierarchical learning framework that explicitly models anatomical context and disease semantics through a dual-branch architecture. To enable effective cross-granularity collaboration, CFCH introduces semantic and cross-granularity attention consistency constraints, encouraging aligned yet complementary feature learning across branches. In addition, we construct AS-9K, a large-scale anterior segment dataset with 8975 images covering 12 common disease categories. To the best of our knowledge, AS-9K is the largest publicly available dataset for anterior segment image classification. Extensive experiments on two anterior segment datasets demonstrate that CFCH outperforms state-of-the-art methods. Qualitative visualizations further show more focused and lesion-relevant activation responses, validating the effectiveness of the proposed framework. Code will be available at https://github.com/ybupengwang/CFCH.
Chinese Translation
准确分类前节疾病对眼科筛查和诊断至关重要。然而，由于成像条件的显著变化以及眼部病理的内在解剖-疾病层次结构，裂隙灯图像分析仍然面临挑战。现有方法通常将此任务表述为一个平面的多类分类问题，忽视了解剖区域（如角膜、结膜和晶状体）与疾病表现之间的结构依赖关系。为了解决这些局限性，我们提出了CFCH，一种粗细协作层次学习框架，通过双分支架构明确建模解剖上下文和疾病语义。为了实现有效的跨粒度协作，CFCH引入了语义和跨粒度注意力一致性约束，鼓励跨分支对齐但互补的特征学习。此外，我们构建了AS-9K，一个大型前节数据集，包含8975幅图像，覆盖12个常见疾病类别。根据我们所知，AS-9K是目前最大的公开可用的前节图像分类数据集。在两个前节数据集上的广泛实验表明，CFCH优于最先进的方法。定性可视化进一步显示出更集中和与病灶相关的激活响应，验证了所提框架的有效性。代码将可在 https://github.com/ybupengwang/CFCH 获取。
cs.CV / 127 / 2609.32567
Attribution Gaps in Zero-Training LLM+OVOD Pipelines: A Fine-Grained Analysis of the CAAP--SNAP Discrepancy
零训练LLM+OVOD流水线中的归因差异：CAAP–SNAP差异的细粒度分析
Abstract
LAOD and similar zero-training LLM+open-vocabulary-detector (OVOD) pipelines score two things separately: class-agnostic localization accuracy (CAAP) and semantic naming accuracy (SNAP). The two consistently diverge, and nobody has asked why. This paper asks why, on the full 5,000-image COCO-Val split (27,273 detections) rather than the small subset the original work evaluated on. Object visual complexity turns out not to be the driver -- small and occluded objects are, if anything, localized better than large ones. Vocabulary novelty is: once the LLM's wording falls outside the detector's native category set, localization accuracy falls from 80.9% to 31.6%. That drop is not spread evenly across unfamiliar phrasing, though. Almost all of it comes from cases where the novel wording actually names a different object than the one COCO annotated (true synonyms still score 89.3%; semantically unrelated "noise" labels score 12.0%). A closer look at a further failure subset tells a similar story: 78-88% of what looks like complete localization failure is really the model correctly finding a real object that COCO's non-exhaustive 80-category scheme simply never labeled, not hallucination. Swap the detector backbone (YOLO-World for Grounding DINO) or the LLM (Gemma-3 for Qwen2.5-VL) and both the effect and its rough size hold up, so this looks like a general property of the pipeline family rather than a quirk of one model pairing. The upshot is that a large share of the apparent CAAP--SNAP gap traces back to closed-category annotation limits rather than a real grounding failure, which matters for how we detect hallucination, analyze failure modes, and design evaluation for grounded multimodal systems meant to work in the open world.
Chinese Translation
LAOD及类似的零训练LLM+开放词汇检测器（OVOD）流水线分别评估两项指标：类别无关定位准确率（CAAP）和语义命名准确率（SNAP）。这两者持续存在差异，但此前无人探究其原因。本文基于完整的5,000张图像COCO-Val分割集（27,273个检测结果），而非原始工作评估的小子集，探讨了这一差异的原因。研究发现，物体视觉复杂度并非主因——小型和遮挡物体的定位准确率反而优于大型物体。词汇新颖性才是关键：一旦LLM的表述超出检测器的原生类别集合，定位准确率便从80.9%骤降至31.6%。然而，这种下降并非均匀分布于所有不熟悉的表述中。几乎所有下降均来自于新颖表述实际上指代了与COCO标注不同的物体（真正的同义词仍能获得89.3%的准确率；语义无关的“噪声”标签准确率仅为12.0%）。对进一步失败子集的深入分析揭示类似情况：78%-88%的看似完全定位失败，实际上是模型正确识别了COCO非穷尽的80类别方案中未标注的真实物体，而非幻觉。更换检测器骨干（从YOLO-World换成Grounding DINO）或LLM（从Gemma-3换成Qwen2.5-VL）后，该效应及其大致规模依然存在，表明这是一类流水线的普遍特性，而非单一模型组合的偶发现象。结论是，显著的CAAP–SNAP差异很大程度上源于封闭类别标注的限制，而非真实的定位失败，这对于我们如何检测幻觉、分析失败模式以及设计面向开放世界的多模态系统的评估方法具有重要意义。
cs.CV / 128 / 2609.32569
OmniSmartHome: A Multimodal Reasoning Benchmark for Smart-Home Agents
OmniSmartHome：一个用于智能家居代理的多模态推理基准
Abstract
Smart-home assistants are expected to handle diverse, realistic requests that arise in daily life. In such interactions, users often rely on the surrounding multimodal context-pointing at objects or referring to what they see or hear, leaving their requests underspecified in language alone. Existing smart-home benchmarks, however, express user requests solely through language, leaving context-dependent real-world requests underexplored. To bridge this gap, we introduce OmniSmartHome, a multimodal smart-home benchmark where each spoken request is paired with the surrounding visual and spatial-audio context, providing complementary cues to disambiguate underspecified requests. OmniSmartHome comprises 1,360 synthetic and 272 real-world episodes. We evaluate 16 omnimodal large language models (Omni-LLMs) and reveal that, while they perform strongly when speech alone sufficiently conveys the user's intent, performance drops substantially when resolving it requires reasoning over multimodal contextual cues. As a simple agent baseline, we provide PROME (PROcedural Memory for multimodal Evidence gathering), which equips agents with specialized audio-visual perception tools and procedural memory for orchestrating their use. PROME generally improves performance across six Omni-LLMs. Demos and examples are available at https://omni-smart-home.github.io
Chinese Translation
智能家居助手预计能够处理日常生活中出现的多样化、现实的请求。在这样的互动中，用户常常依赖周围的多模态上下文——指向物体或提及他们所见或所听的内容，从而使他们的请求在语言上显得不够明确。然而，现有的智能家居基准仅通过语言表达用户请求，导致依赖上下文的现实请求未得到充分探索。为了解决这一问题，我们推出了OmniSmartHome，一个多模态智能家居基准，其中每个口头请求都与周围的视觉和空间音频上下文配对，提供补充线索以消除不明确请求的歧义。OmniSmartHome包含1360个合成和272个真实世界的情节。我们评估了16个全模态大型语言模型（Omni-LLMs），发现当仅靠语言足以传达用户意图时，它们的表现强劲，但当需要对多模态上下文线索进行推理时，性能显著下降。作为一个简单的代理基线，我们提供了PROME（用于多模态证据收集的程序记忆），该工具为代理配备了专门的视听感知工具和程序记忆，以协调其使用。PROME通常在六个Omni-LLMs中提高了性能。演示和示例可在https://omni-smart-home.github.io获取。
cs.CV / 129 / 2609.32590
Retrieved but Not Delivered: Multimodal Memory Delivery for Long-Term Agents
已检索但未交付：长期智能体的多模态记忆交付
Abstract
Work on memory for multimodal agents optimizes what is written, updated and retrieved. Between retrieval and the answer, however, is a stage that multimodal memory evaluations do not isolate: what of the retrieved memory reaches the model, and in what form. We call it delivery, and a controlled decomposition on MemLens locates the remaining room there. With the retrieved evidence set exactly fixed, delivering the original pixels instead of withholding them raises accuracy by 13.87 points on an 8B backbone, whereas making retrieval perfect on those same messages improves it by 2.31. Delivery is the larger term on all three MemLens backbones and grows with backbone strength; retrieval grows too, without closing the gap. We propose DeliverMem, an instantiation of delivery as three decisions: keep the original modality, give each item a readable identity, and state when it was seen, with a retrieval-side adapter for the one property delivery cannot supply. Each is measured against a delivery-matched control that alters only its own variable. DeliverMem leads the strongest published memory agent on MemLens at all four context lengths, and beats DMV-Bench's own strongest method at every setting on both backbones. On MemLens it does this on a tenth to a seventieth of the input. Each decision helps only where the question lacks what it supplies, and is null elsewhere. A single fixed configuration nonetheless leads both benchmarks, without training any component or modifying the stored records. Project page: https://avalon-s.github.io/DeliverMem/
Chinese Translation
关于多模态智能体的记忆研究优化了写入、更新和检索的内容。然而，在检索与答案之间存在一个阶段，而多模态记忆评估并未单独考虑：检索到的记忆中哪些内容到达了模型，以及以何种形式到达。我们称之为交付，通过对 MemLens 的受控分解定位了其中的剩余空间。在检索证据集完全固定的情况下，交付原始像素而不是保留它们使得在 8B 主干上准确率提高了 13.87 分，而在同样消息上使检索完美则提高了 2.31 分。交付在所有三个 MemLens 主干上都是更大的术语，并随着主干强度的增强而增长；检索也在增长，但未能缩小差距。我们提出了 DeliverMem，将交付实例化为三个决策：保持原始模态、为每个项目赋予可读身份，以及说明何时被看到，并为交付无法提供的唯一属性设计了一个检索侧适配器。每个决策都与一个仅改变其自身变量的交付匹配控制进行比较。DeliverMem 在所有四个上下文长度上领先于 MemLens 上发布的最强记忆智能体，并在两个主干的每个设置中击败了 DMV-Bench 自身的最强方法。在 MemLens 上，它在输入量的十分之一到七十分之一的情况下实现了这一点。每个决策仅在问题缺乏其提供的内容时有帮助，而在其他情况下则无效。尽管如此，一个固定的配置仍然在两个基准测试中领先，而无需训练任何组件或修改存储记录。项目页面：https://avalon-s.github.io/DeliverMem/
cs.CV / 130 / 2609.32592
SPACE: Sparse Predictive Attractor via Counterfactual Eviction for Streaming Video Memory
SPACE：通过反事实驱逐实现稀疏预测吸引子的流媒体视频记忆
Abstract
Fixed-capacity streaming video memory requires repeated eviction decisions whose effects accumulate over time. Yet existing policies are evaluated primarily in terms of retained information or downstream accuracy, leaving how repeated updates alter the futures supported by memory largely unexamined. We define a memory's predictive state as the future representations supported by its retained history and formulate eviction as counterfactual control over transitions in this space. We introduce SPACE (Sparse Predictive Attractor via Counterfactual Eviction), which uses a frozen multi-horizon JEPA to predict the future representations induced by alternative eviction actions. Counterfactual utility identifies future-useful alternatives, while slow predictive-basin geometry determines when to correct avoidable drift and when to adapt to sustained predictive change, without online parameter updates. We further introduce MABS-Bench, which evaluates future-task sufficiency, within-regime predictive stability, transition responsiveness, and perturbation recovery under matched causal streams and memory budgets. Across multiple video datasets, SPACE yields consistent improvements in dataset-native task performance while reducing predictive-state drift.
Chinese Translation
固定容量的流媒体视频记忆需要反复做出驱逐决策，其影响会随着时间的推移而累积。然而，现有政策主要通过保留的信息或下游准确性进行评估，导致反复更新如何改变记忆支持的未来未得到充分研究。我们将记忆的预测状态定义为由其保留历史支持的未来表征，并将驱逐形式化为对这一空间中转变的反事实控制。我们引入了SPACE（通过反事实驱逐实现稀疏预测吸引子），该方法使用冻结的多视角JEPA来预测由替代驱逐动作引发的未来表征。反事实效用识别未来有用的替代方案，而缓慢的预测盆地几何则决定何时纠正可避免的漂移以及何时适应持续的预测变化，而无需在线参数更新。我们进一步引入了MABS-Bench，用于评估未来任务的充分性、制度内预测稳定性、转变响应性和在匹配因果流和记忆预算下的扰动恢复。在多个视频数据集上，SPACE在数据集本地任务性能上持续改善，同时减少了预测状态的漂移。
cs.CV / 131 / 2609.32596
GAUGE: Group-Wise View-Inconsistency Rectification for Feed-Forward 4D Tracking
GAUGE：基于组的视图不一致性校正用于前馈4D跟踪
Abstract
Feed-forward models regress dense 3D point trajectories directly from monocular video, yet the residual after global alignment is substantial and lacks a structural explanation. Measured on dynamic query points across models and datasets, the error concentrates along the view direction, while the scale correction each motion group requires differs. The predicted displacement direction nevertheless supports reliable grouping, with a median angle far below the 90{\deg} random baseline. The systematic part of the residual is therefore a family of radial degrees of freedom per motion group, along directions 2D observations cannot constrain. We call it group-wise view inconsistency. We present GAUGE (Group-wise Adaptive Unsupervised Gauge Estimation), a training-free and model-agnostic post-hoc module. It recovers motion groups from direction consistency and spatial connectivity, then estimates a per-frame radial scale and group-level translation from 1% to 5% metric anchors, four degrees of freedom per group and frame. On dynamic query points of eight trackers, including D4RT, 4RC and SM4RT, our correction lowers endpoint error by 15.1% to 62.6% over the uncorrected predictions, while spending the same anchors on gradient fine-tuning improves the same models by only -1.1% to 15.2%. Code is publicly available at https://github.com/HCPLab-SYSU/GAUGE.
Chinese Translation
前馈模型直接从单目视频回归稠密的3D点轨迹，但全局对齐后的残差相当显著，且缺乏结构性解释。在不同模型和数据集上测量的动态查询点显示，误差主要集中在视图方向上，而每个运动组所需的尺度校正则各不相同。然而，预测的位移方向支持可靠的分组，其中位角度远低于90°的随机基线。因此，残差的系统部分是每个运动组的一系列径向自由度，沿着二维观测无法约束的方向。我们称之为基于组的视图不一致性。我们提出了GAUGE（基于组的自适应无监督量测估计），这是一个无训练且与模型无关的后处理模块。它通过方向一致性和空间连通性恢复运动组，然后从1%到5%的度量锚点中估计每帧的径向尺度和组级平移，每组和每帧有四个自由度。在包括D4RT、4RC和SM4RT在内的八个跟踪器的动态查询点上，我们的校正将端点误差降低了15.1%至62.6%，而在相同锚点上进行梯度微调仅使相同模型改善了-1.1%至15.2%。代码已公开发布在https://github.com/HCPLab-SYSU/GAUGE。
cs.CV / 132 / 2609.32612
Levy-Driven Correspondence Estimation for Registration
基于Lévy驱动的配准对应关系估计
Abstract
Finding reliable point correspondences is difficult when point clouds have low overlap or undergo non-rigid deformation. Iterative refinement can correct uncertain matches, but costly network evaluations limit the number of updates. We present LevyMatch, a L\'evy-driven method that uses random jumps to refine a soft matching matrix. At each step, a network uses the current matching state and geometric information to predict a target matching matrix. A Brownian reference bridge gives an explicit formula for the update toward this target. A Gamma random clock sets the time step for each update. The updated matches provide new geometric feedback for the next target prediction. We further propose a fixed front-loaded Gamma policy that assigns more expected clock time to early updates and less to later ones, without retraining or extra network evaluations. Reordering the same sampled Gamma increments shows that placing larger increments early gives higher accuracy than placing them late. On 4DMatch and 4DLoMatch, our method improves both non-rigid feature matching recall (NFMR) and inlier ratio (IR) over the compared methods. The front-loaded policy achieves 93.09% NFMR and 92.11% IR on 4DMatch, and 82.79% NFMR and 79.07% IR on 4DLoMatch.
Chinese Translation
当点云重叠较少或经历非刚性变形时，寻找可靠的点对应关系是困难的。迭代精炼可以纠正不确定的匹配，但昂贵的网络评估限制了更新的次数。我们提出了LevyMatch，这是一种基于Lévy过程的方法，通过随机跳跃来精炼软匹配矩阵。在每一步中，网络利用当前的匹配状态和几何信息来预测目标匹配矩阵。布朗桥提供了一个明确的公式，用于向该目标的更新。伽马随机时钟为每次更新设置时间步长。更新后的匹配为下一个目标预测提供新的几何反馈。我们进一步提出了一种固定的前加载伽马策略，该策略为早期更新分配更多的期望时钟时间，而为后期更新分配更少的时间，无需重新训练或额外的网络评估。重新排序相同的采样伽马增量表明，早期放置较大的增量比后期放置能获得更高的准确性。在4DMatch和4DLoMatch上，我们的方法在非刚性特征匹配召回率（NFMR）和内点比例（IR）方面均优于比较方法。前加载策略在4DMatch上达到了93.09%的NFMR和92.11%的IR，在4DLoMatch上达到了82.79%的NFMR和79.07%的IR。
cs.CV / 133 / 2609.32628
DraftAttention2: Fast Video Diffusion with Low-Resolution-Guided Mixed-Precision Attention
DraftAttention2：基于低分辨率引导的混合精度注意力的快速视频扩散
Abstract
Video generation has broad applications in content creation and entertainment. Diffusion transformers have advanced the quality of generated videos, but attention over spatiotemporal tokens becomes increasingly expensive as video resolution and duration increase. We present DraftAttention2, a training-free framework that uses the low-resolution draft attention map to jointly select attention blocks and assign their numerical precision. Specifically, spatial 2D average- and max-pooled queries and keys capture complementary regional statistics to estimate block importance, and a shared ranking assigns higher precision to important blocks, lower precision to less important retained blocks, and skips the rest under configurable budgets. Our analysis separates sparsification error from attention-weighted quantization error, establishing when recovering skipped interactions with low-bit computation tightens the output-error bound. This analysis motivates retaining more interactions at low precision while reserving higher precision for blocks with larger attention mass. To translate these fine-grained assignments into practical speedups, we further develop fused operand preparation and a single attention kernel with precision-specific phases, sharing data movement, softmax statistics, and output accumulation across precisions. Experiments demonstrate that our method achieves a superior quality-efficiency trade-off over existing efficient video generation methods. Notably, its advantage is particularly pronounced for few-step video diffusion, where jointly combining sparsity with 4- and 8-bit mixed-precision computation substantially improves generation quality while retaining significant acceleration. Code is available at https://github.com/anemoi-project/anemoi
Chinese Translation
视频生成在内容创作和娱乐领域具有广泛的应用。扩散变换器提高了生成视频的质量，但随着视频分辨率和时长的增加，时空标记的注意力计算变得越来越昂贵。我们提出了DraftAttention2，这是一个无训练的框架，利用低分辨率草图注意力图共同选择注意力块并分配其数值精度。具体而言，空间二维平均池化和最大池化的查询和键捕捉互补的区域统计，以估计块的重要性，共享排名将更高的精度分配给重要块，将较低的精度分配给不太重要的保留块，并在可配置的预算下跳过其余块。我们的分析将稀疏化误差与注意力加权量化误差分开，建立了在低位计算恢复跳过的交互时如何收紧输出误差界限。这一分析激励我们在低精度下保留更多交互，同时为具有更大注意力质量的块保留更高的精度。为了将这些细粒度的分配转化为实际的加速效果，我们进一步开发了融合操作数准备和具有精度特定阶段的单一注意力内核，共享数据移动、softmax统计和跨精度的输出累积。实验表明，我们的方法在现有高效视频生成方法中实现了优越的质量效率权衡。值得注意的是，其优势在少步视频扩散中尤为明显，在这种情况下，将稀疏性与4位和8位混合精度计算结合在一起，显著提高了生成质量，同时保持了显著的加速效果。代码可在 https://github.com/anemoi-project/anemoi 获取。
cs.CV / 134 / 2609.32660
InterTab: Interleaved Visual-Structure Alignment for Multi-Modal Table Reasoning
InterTab：用于多模态表格推理的交错视觉-结构对齐
Abstract
Table images preserve structural information that are often lost in text serialization, and reasoning over them requires locating relevant rows, columns, and cells step by step. Current multimodal large language models (MLLMs) encode the whole image once before reasoning, so they cannot pick up row-, column-, and cell-level evidence as the question unfolds. Encoder-side table structure and generic interleaved visual chain-of-thought still do not bind each reasoning step to that structure. We propose \textbf{InterTab}, an \textbf{Inter}leaved structure-aware framework for CoT reasoning over \textbf{Tab}le images, interleaves chain-of-thought with tool calls that crop structure-aligned table regions. First, we build InterTab-22K, includes reasoning trajectories in which each step is tied to both a structural location and a bounding box. InterTab is trained in two stages: supervised structure-aware alignment (SSA) on InterTab-22K teaches the model to interleave reasoning with structure-aligned crops, and active localization optimization (ALO) further optimizes answer correctness, localization IoU, and output format, while penalizing missing or excessive tool calls. Experiments on nine table benchmarks show that InterTab improves the average accuracy of its backbone from 68.28% to 73.17% and achieves the best average performance among all compared methods. Code and data will be released soon.
Chinese Translation
表格图像保留了在文本序列化中常常丢失的结构信息，而对其进行推理需要逐步定位相关的行、列和单元格。目前的多模态大型语言模型（MLLMs）在推理之前对整个图像进行一次编码，因此在问题展开时无法捕捉到行、列和单元格级别的证据。编码器侧的表格结构和通用的交错视觉思维链仍然无法将每个推理步骤与该结构绑定。我们提出了 extbf{InterTab}，一个用于对 extbf{表格}图像进行思维链推理的 extbf{交错}结构感知框架，将思维链与裁剪结构对齐的表格区域的工具调用交错在一起。首先，我们构建了InterTab-22K，其中包含每个步骤与结构位置和边界框相关联的推理轨迹。InterTab的训练分为两个阶段：在InterTab-22K上进行的监督结构感知对齐（SSA）教会模型将推理与结构对齐的裁剪交错在一起，而主动定位优化（ALO）进一步优化答案的正确性、定位的IoU和输出格式，同时惩罚缺失或过多的工具调用。在九个表格基准测试上的实验表明，InterTab将其基础模型的平均准确率从68.28%提高到73.17%，并在所有比较方法中实现了最佳的平均性能。代码和数据将很快发布。
cs.CV / 135 / 2609.32681
RCVLA: 4D Radar-Grounded Semantic Reasoning and Trajectory Arbitration for Autonomous Driving
RCVLA：基于4D雷达的语义推理与自主驾驶轨迹仲裁
Abstract
4D radar provides geometric and motion cues that complement visual semantics, but integrating it into vision-language-action (VLA) models requires both radar--language alignment for semantic reasoning and explicit use of radar measurements for trajectory refinement and selection. To support these capabilities, we construct Cap4DR with 86,016 radar-image-text samples for alignment pretraining and OmniHD-QA with 520,161 question-answer pairs for instruction tuning across scene description, key-object reasoning, occupancy understanding, and trajectory planning. Building on these datasets, we propose RCVLA, a radar-camera VLA framework consisting of a radar-grounded semantic reasoning stage (RCVLA-Sem) and a trajectory arbitration stage (RCVLA-Phys). RCVLA-Sem performs gated bidirectional interaction between camera and radar tokens for driving question answering and reference trajectory generation, while auxiliary heads provide object and occupancy queries. RCVLA-Phys refines reference-guided trajectory candidates through truncated diffusion conditioned on these queries and cluster-level radar measurements, then calibrates candidate scores using radar-derived time-to-collision risk. On OmniHD-QA, RCVLA-Sem improves CIDEr by 9.92 points and reduces key-object velocity error by $21.9\%$ relative to OmniDrive. RCVLA-Phys further reduces average L2 error from $0.348$ to $0.259\,\mathrm{m}$ and average open-loop collision rate from $0.576\%$ to $0.175\%$ relative to RCVLA-Sem. Ablation studies further show that language-aligned radar tokens improve semantic reasoning, while cluster-level radar measurements and risk calibration improve trajectory arbitration. Code will be released.
Chinese Translation
4D雷达提供几何和运动线索，补充视觉语义，但将其整合到视觉-语言-动作（VLA）模型中需要雷达与语言的对齐以进行语义推理，以及明确使用雷达测量进行轨迹细化和选择。为了支持这些功能，我们构建了Cap4DR，包含86,016个雷达-图像-文本样本用于对齐预训练，以及OmniHD-QA，包含520,161个问答对用于场景描述、关键对象推理、占用理解和轨迹规划的指令调优。在这些数据集的基础上，我们提出了RCVLA，一个雷达-摄像头VLA框架，包括一个基于雷达的语义推理阶段（RCVLA-Sem）和一个轨迹仲裁阶段（RCVLA-Phys）。RCVLA-Sem在摄像头和雷达标记之间执行门控双向交互，以进行驾驶问答和参考轨迹生成，同时辅助头提供对象和占用查询。RCVLA-Phys通过基于这些查询和集群级雷达测量的截断扩散细化参考引导的轨迹候选，然后使用雷达推导的碰撞风险时间对候选分数进行校准。在OmniHD-QA上，RCVLA-Sem将CIDEr提高了9.92分，并相对于OmniDrive减少了关键对象速度误差$21.9\%$。RCVLA-Phys进一步将平均L2误差从$0.348$减少到$0.259 \, ext{m}$，将平均开环碰撞率从$0.576\\%$减少到$0.175\\%$，相对于RCVLA-Sem。消融研究进一步表明，语言对齐的雷达标记改善了语义推理，而集群级雷达测量和风险校准改善了轨迹仲裁。代码将会发布。
cs.CV / 136 / 2609.32690
MM-OPD: Towards One More Bottleneck Between Perception and Reasoning
MM-OPD：在感知与推理之间再添一个瓶颈
Abstract
Recent multimodal large language models (MLLMs) advance visual reasoning by strengthening both perception and reasoning, implicitly assuming a process that transitions seamlessly from perception to reasoning. However, we observe a counterintuitive phenomenon that challenges this assumption: holding the model, question, and decoding fixed, we replace images with their caption or code representations (symbolic views), which seems to be redundant given the clear image structures, but the performance surprisingly improves by 10.2% to 23.6% across model scales and datasets. We term this performance gap as the Symbolic Visual Gap and then take a closer look at it. Through experiments, we find that although the visual evidence can already appear in the reasoning trace for the image-input model, the symbolic-view-input model shows much higher attention to the correct evidence than the image-input model. This suggests that despite good capabilities from current works in perception and reasoning themselves, another bottleneck exists between perception and reasoning in selecting perceived visual information as appropriate evidence for subsequent reasoning. To handle this bottleneck, since the symbolic view steers attention toward correct evidence and is readily obtained at scale, it provides supervision for evidence selection without manually labeled evidence. Building on this, we introduce MM-OPD, a multimodal on-policy self-distillation framework for symbolic-to-visual correction that transfers guidance from symbolic-conditioned behavior to the image-conditioned policy through residual token-level targets, steering the model toward correct visual evidence. Experiments across benchmarks and model scales show that MM-OPD improves a broad range of multimodal abilities, with gains in visual perception, chart and document understanding, mathematical reasoning, and general VQA.
Chinese Translation
最近的多模态大语言模型（MLLMs）通过增强感知和推理来推动视觉推理的发展，隐含地假设了一个从感知到推理无缝过渡的过程。然而，我们观察到一个与直觉相悖的现象，这一现象挑战了这一假设：在固定模型、问题和解码的情况下，我们用图像的标题或代码表示（符号视图）替换图像，尽管考虑到清晰的图像结构，这似乎是多余的，但在不同模型规模和数据集上，性能意外地提高了10.2%到23.6%。我们将这一性能差距称为符号视觉差距，并对此进行了更深入的研究。通过实验，我们发现，尽管视觉证据已经可以在图像输入模型的推理轨迹中出现，但符号视图输入模型对正确证据的关注程度远高于图像输入模型。这表明，尽管当前在感知和推理方面的能力较强，但在选择感知的视觉信息作为后续推理的适当证据时，感知与推理之间仍然存在另一个瓶颈。为了解决这一瓶颈，由于符号视图能够引导注意力朝向正确证据，并且可以大规模获取，它为证据选择提供了监督，而无需手动标记证据。在此基础上，我们引入了MM-OPD，一个用于符号到视觉校正的多模态在线自蒸馏框架，通过残差令牌级目标将符号条件行为的指导转移到图像条件策略，引导模型朝向正确的视觉证据。跨基准和模型规模的实验表明，MM-OPD改善了广泛的多模态能力，在视觉感知、图表和文档理解、数学推理以及一般视觉问答（VQA）方面均取得了显著提升。
cs.CV / 137 / 2609.32705
DPAMixerSR: An Efficient Degradation-Pattern-Aware Model for Image Super-Resolution
DPAMixerSR：一种高效的降解模式感知图像超分辨率模型
Abstract
While content-adaptive schemes have delivered notable advances in image super-resolution (SR), existing approaches typically focus on texture complexity and ignore intrinsic degradation factors (e.g., blur kernels or noise patterns), leading to suboptimal computation allocation and reconstruction performance. To remedy this, we propose DPAMixerSR, a degradation-pattern-aware framework that enables efficient SR through adaptive sparse computation. We design a lightweight Perceptual Degradation Ranking (PDR) module partitions the image into severely and mildly degraded patches, which are routed to the Adaptive Sparse Processing (ASP) and a lightweight convolutional branch, respectively. ASP performs structure-aligned, multi-scale sparse propagation and bidirectional refinement, while the convolutional branch enhances efficiency in mildly degraded regions. By coupling degradation-driven routing with structure-aligned sparse processing, DPAMixerSR establishes a self-regulating framework that dynamically balances computational efficiency and reconstruction fidelity. Extensive experiments on various SR tasks demonstrate that our DPAMixerSR achieves superior structural restoration and perceptual fidelity with markedly reduced computational overhead, providing a novel and scalable framework for degradation-aware, resource-efficient SR.
Chinese Translation
尽管内容自适应方案在图像超分辨率（SR）方面取得了显著进展，但现有方法通常侧重于纹理复杂性，而忽视了内在的降解因素（例如，模糊核或噪声模式），导致计算分配和重建性能不佳。为了解决这一问题，我们提出了DPAMixerSR，一种降解模式感知框架，通过自适应稀疏计算实现高效的超分辨率。我们设计了一个轻量级的感知降解排序（PDR）模块，将图像划分为严重降解和轻度降解的补丁，分别路由到自适应稀疏处理（ASP）和轻量级卷积分支。ASP执行结构对齐的多尺度稀疏传播和双向精炼，而卷积分支则在轻度降解区域提高效率。通过将降解驱动的路由与结构对齐的稀疏处理相结合，DPAMixerSR建立了一个自我调节的框架，动态平衡计算效率和重建保真度。在各种超分辨率任务上的广泛实验表明，我们的DPAMixerSR在结构恢复和感知保真度方面表现优越，同时显著降低了计算开销，为降解感知、资源高效的超分辨率提供了一个新颖且可扩展的框架。
cs.CV / 138 / 2609.32711
ProDyGS: Dynamic Gaussian Splatting from a Single Static Monocular Camera
ProDyGS：基于单一静态单目相机的动态高斯点云生成
Abstract
We present ProDyGS, a novel dynamic 3D Gaussian Splatting framework for high-quality novel view synthesis from videos captured by a single static camera. While existing methods rely on multi-view setups or significant camera motion for geometric constraints, our approach addresses the challenging scenario where multi-view supervision is completely absent. We overcome this limitation by generating synthetic multi-view supervision through depth-guided proxy image synthesis. Specifically, we estimate temporally consistent depth maps using foundational monocular depth networks, then construct 3D Gaussian representations that generate proxy images from arbitrary viewpoints. A deformation network learns temporal dynamics by warping canonical Gaussians using this augmented supervision. Experiments on the DyNeRF dataset demonstrate that our method achieves state-of-the-art performance while requiring only monocular depth estimation as external supervision, outperforming approaches that rely on stronger priors such as scene flow.
Chinese Translation
我们提出了ProDyGS，一种新颖的动态3D高斯点云生成框架，用于从单一静态相机捕获的视频中合成高质量的新视角图像。现有方法依赖于多视角设置或显著的相机运动来获取几何约束，而我们的方法则解决了完全缺乏多视角监督的挑战场景。我们通过深度引导的代理图像合成来克服这一限制。具体而言，我们使用基础的单目深度网络估计时间一致的深度图，然后构建3D高斯表示，从任意视点生成代理图像。变形网络通过使用这种增强的监督信息对标准高斯进行变形，学习时间动态。在DyNeRF数据集上的实验表明，我们的方法在仅需单目深度估计作为外部监督的情况下，达到了最先进的性能，超越了依赖于更强先验（如场景流）的方法。
cs.CV / 139 / 2609.32716
Region-Local Copula Evidence Fusion for Heterogeneous Remote Sensing Change Detection
异构遥感变化检测的区域局部Copula证据融合
Abstract
Superpixel copula models provide stable regional evidence for heterogeneous remote sensing change detection, but a single label per region limits localization within mixed superpixels. This letter develops a region-local copula evidence fusion method that retains the regional decision structure while introducing spatially varying local dependence anomalies. Independently fitted local models characterize departures from unchanged cross-image relationships. Reference ranking and an upper-tail gate transform these anomalies for fusion with continuous regional confidence. We derive the resulting regiondependent local decision threshold and identify a condition under which gating is equivalent to reparameterizing ungated fusion. On Lake and UK, whole-image optimized configurations achieve kappa coefficients of 0.78136 and 0.90817 and improve mixedregion and boundary decisions. Four-fold retrospective spatial validation over ten training subsets confirms complementary local information, with ungated reference fusion increasing mean kappa by 0.00693 and 0.01793. Fixed gating yields a larger UK gain of 0.03353 but only 0.00041 on Lake. These results support regional-local dependence interaction, while showing that calibration and gating have scene-dependent benefits.
Chinese Translation
超像素Copula模型为异构遥感变化检测提供了稳定的区域证据，但每个区域仅有一个标签限制了混合超像素内的定位。本文提出了一种区域局部Copula证据融合方法，该方法保留了区域决策结构，同时引入了空间变化的局部依赖异常。独立拟合的局部模型表征了未变化的跨图像关系的偏离。参考排名和上尾门控将这些异常转化为与连续区域置信度融合。我们推导出结果的区域依赖局部决策阈值，并确定了一个条件，在该条件下，门控等同于重新参数化未门控融合。在湖泊和英国的全图优化配置中，kappa系数分别达到了0.78136和0.90817，并改善了混合区域和边界决策。对十个训练子集进行的四折回顾性空间验证证实了互补的局部信息，未门控参考融合使平均kappa提高了0.00693和0.01793。固定门控在英国获得了更大的增益0.03353，但在湖泊上仅为0.00041。这些结果支持区域局部依赖交互，同时表明标定和门控具有场景依赖的好处。
cs.CV / 140 / 2609.32734
REALIS: A Curated Dataset for Studying the Challenges of AI Image Detection
REALIS：用于研究人工智能图像检测挑战的精选数据集
Abstract
AI-generated image detectors are often evaluated on benchmarks where real and synthetic images differ in content, quality, or generation artifacts, allowing models to rely on dataset-specific cues and fail on unfamiliar generators or processed images. Existing datasets provide limited support for evaluating these challenges jointly across diverse visual content. We introduce REALIS, a dataset of 1.43 million real and synthetic images generated by 42 modern text-to-image models, including the latest proprietary systems such as Nano Banana 2. REALIS combines prompts derived from real images, quality filtering, and stratified sampling to reduce class-specific shortcuts while preserving content diversity. We further introduce REALIS-Expert, a stress-test subset for high-quality synthetic images, where real and generated samples are selected with closely matched semantic and visual characteristics. We also propose a robustness protocol covering 35 transformations at five severity levels to analyze detector behavior under image processing. Based on REALIS, our benchmark evaluates pretrained detectors, fine-tuned models, and zero-shot vision-language models under generator and post-processing shifts. On the hardest processed split, the best pretrained conventional detector achieves 0.550 ROC-AUC, compared with 0.752 for the best REALIS-trained detector. REALIS provides a unified framework for measuring and improving the reliability of AI-image detectors under conditions that better reflect real-world use.
Chinese Translation
人工智能生成的图像检测器通常在真实图像与合成图像在内容、质量或生成伪影上存在差异的基准上进行评估，这使得模型能够依赖于特定数据集的线索，而在面对不熟悉的生成器或处理过的图像时失败。现有数据集在跨多样视觉内容共同评估这些挑战方面支持有限。我们介绍了REALIS，一个包含143万张由42种现代文本到图像模型生成的真实和合成图像的数据集，包括最新的专有系统如Nano Banana 2。REALIS结合了源自真实图像的提示、质量过滤和分层抽样，以减少特定类别的捷径，同时保持内容的多样性。我们进一步推出了REALIS-Expert，一个针对高质量合成图像的压力测试子集，其中真实样本和生成样本在语义和视觉特征上紧密匹配。我们还提出了一种鲁棒性协议，涵盖35种变换和五个严重程度级别，以分析检测器在图像处理下的表现。基于REALIS，我们的基准评估了预训练检测器、微调模型和零样本视觉-语言模型在生成器和后处理变化下的表现。在最难的处理分割上，最佳的预训练传统检测器达到了0.550的ROC-AUC，而最佳的REALIS训练检测器则达到了0.752。REALIS提供了一个统一的框架，用于测量和提高人工智能图像检测器在更好反映现实世界使用条件下的可靠性。
cs.CV / 141 / 2609.32737
Gradient-Guided Decoupled Adaptation for Geospatial Vision-Language Models
基于梯度引导的解耦适应用于地理空间视觉-语言模型
Abstract
Existing geospatial vision-language models (Geo-VLMs) typically optimize diverse geospatial tasks through a unified multi-task adaptation paradigm without explicitly accounting for the heterogeneous optimization characteristics. Our empirical observations reveal heterogeneous gradient characteristics across tasks, including vision-language differences, intra-branch gradient relationships, and task interference, which hinder effective multi-task optimization. Motivated by these observations, we propose Gradient-Guided Decoupled Adaptation (G2DA), a gradient-aware optimization framework for multi-task Geo-VLM learning. G2DA first partitions tasks into vision- and language-centric groups through gradient-guided cross-modal decoupling. It then constructs modality-specific curricula based on task gradient similarity and employs bidirectional rehearsal to mitigate the recency effects introduced by sequential optimization. We evaluate G2DA on three Geo-VLM benchmarks using six InternVL3 and Qwen3.5-VL variants, along with GeoChat and GeoLLaVA. Across all 24 benchmark-model combinations, G2DA consistently outperforms representative baselines, improving over the strongest competitor by 3.08, 4.30, and 2.81 percentage points on UrBench-MCQ, XLRS-Bench-Lite, and VRS-Bench-VQA, respectively. These results demonstrate the effectiveness of gradient-guided task organization for Geo-VLM adaptation.
Chinese Translation
现有的地理空间视觉-语言模型（Geo-VLMs）通常通过统一的多任务适应范式来优化多样的地理空间任务，而未明确考虑异构优化特性。我们的实证观察揭示了任务之间的异构梯度特性，包括视觉-语言差异、分支内部梯度关系和任务干扰，这些因素阻碍了有效的多任务优化。基于这些观察，我们提出了基于梯度引导的解耦适应（G2DA），这是一种针对多任务Geo-VLM学习的梯度感知优化框架。G2DA首先通过梯度引导的跨模态解耦将任务划分为以视觉和语言为中心的组。然后，它基于任务梯度相似性构建特定模态的课程，并采用双向复习来减轻顺序优化引入的近期效应。我们在三个Geo-VLM基准上评估了G2DA，使用了六个InternVL3和Qwen3.5-VL变体，以及GeoChat和GeoLLaVA。在所有24个基准-模型组合中，G2DA始终优于代表性基线，在UrBench-MCQ、XLRS-Bench-Lite和VRS-Bench-VQA上分别提高了3.08、4.30和2.81个百分点。这些结果证明了基于梯度引导的任务组织在Geo-VLM适应中的有效性。
cs.CV / 142 / 2609.32740
AnesTRACE: Benchmarking Intraoperative Anesthesia from Multimodal Perception to Multi-step Decision-Making
AnesTRACE：从多模态感知到多步骤决策的术中麻醉基准测试
Abstract
Intraoperative anesthesia requires systems to interpret evolving multimodal evidence, recommend timely management, and revise decisions as patient states change, yet existing benchmarks usually isolate perception or single-point reasoning. We introduce AnesTRACE, an evaluation suite comprising AnesTRACE-Bench and AnesTRACE-Eval. Built from public perioperative datasets with anesthesiologist annotation, AnesTRACE-Bench evaluates Intraoperative Perception, Single-point Anesthesia Decision-Making, and Multi-step Anesthesia Decision-Making. AnesTRACE-Eval assesses open-ended responses through anesthesiologist-defined criteria for Clinical Correctness, Evidence Grounding, Task Completeness, and Safety, with Temporal Consistency for multi-step decisions; its domain-specific evaluator is trained by supervised fine-tuning and preference alignment on expert-reviewed judgments. Across more than 30 models, fine-grained visual grounding and intervention selection remain difficult: the leading model reaches only 32.2 mIoU for TEE visual grounding and retains a 17.5\% Major/Critical Safety Error Rate in multi-step management. Evaluator alignment with anesthesiologists improves across both training stages, while the best decision quality is accompanied by a 74.3-second P95 Latency. These results show that aggregate performance alone does not establish safe, timely longitudinal decision-making. We release our code at https://github.com/zjuDBxAI/AnesTRACE.
Chinese Translation
术中麻醉要求系统能够解读不断变化的多模态证据，及时推荐管理方案，并在患者状态变化时修正决策，而现有的基准测试通常孤立感知或单点推理。我们引入了AnesTRACE，一个评估套件，包括AnesTRACE-Bench和AnesTRACE-Eval。AnesTRACE-Bench基于公共围手术期数据集和麻醉师注释，评估术中感知、单点麻醉决策和多步骤麻醉决策。AnesTRACE-Eval通过麻醉师定义的临床正确性、证据基础、任务完整性和安全性标准评估开放式响应，并针对多步骤决策的时间一致性进行评估；其领域特定评估器通过监督微调和与专家审查判断的偏好对齐进行训练。在超过30个模型中，细粒度的视觉基础和干预选择仍然困难：领先模型在TEE视觉基础上仅达到32.2 mIoU，并在多步骤管理中保持17.5%的重大/关键安全错误率。评估器与麻醉师的对齐在两个训练阶段均有所改善，而最佳决策质量伴随着74.3秒的P95延迟。这些结果表明，单靠聚合性能并不能建立安全、及时的纵向决策。我们的代码已发布在https://github.com/zjuDBxAI/AnesTRACE。
cs.CV / 143 / 2609.32742
LoCoVSR: Local Context Diffusion Posterior Sampling for Video Super-Resolution
LoCoVSR：用于视频超分辨率的局部上下文扩散后验采样
Abstract
Video super-resolution (VSR) is an ill-posed inverse problem that aims to reconstruct a high-resolution (HR) video from a noisy, low-resolution (LR) version of it. We present LoCoVSR, a diffusion-based VSR framework that leverages pixel-space denoising diffusion probabilistic models. LoCoVSR integrates the Diffusion Posterior Sampling technique with spatio-temporal context learning, operating in a moving-average form. A localized window of adjacent LR frames is used for recovering each center frame, while applying a shared noise trajectory across all frames. The localized windowing enables processing of long videos without length limitations, supports parallel inference, and prevents error accumulation that may occur in recursive processing. Unlike prior methods, LoCoVSR offers a simple yet very effective VSR solution, avoiding explicit optical flow estimation, or information loss caused by latent space processing. Trained on the VFHQ face dataset, LoCoVSR achieves accurate, temporally consistent and high-quality upscaling with competitive results against recent diffusion-based VSR approaches.
Chinese Translation
视频超分辨率（VSR）是一个不适定的逆问题，旨在从其噪声较大、分辨率较低（LR）的版本中重建高分辨率（HR）视频。我们提出了LoCoVSR，这是一种基于扩散的VSR框架，利用像素空间去噪扩散概率模型。LoCoVSR将扩散后验采样技术与时空上下文学习相结合，以移动平均的形式运行。每个中心帧的恢复使用相邻LR帧的局部窗口，同时在所有帧上应用共享的噪声轨迹。局部窗口处理使得可以在没有长度限制的情况下处理长视频，支持并行推理，并防止在递归处理过程中可能出现的误差累积。与之前的方法不同，LoCoVSR提供了一种简单而有效的VSR解决方案，避免了显式光流估计或潜在空间处理造成的信息损失。在VFHQ人脸数据集上训练后，LoCoVSR实现了准确、时间一致且高质量的放大，并在与近期基于扩散的VSR方法的比较中表现出竞争力的结果。
cs.CV / 144 / 2609.32761
From Feed-Forward to Flow: Unifying Reconstruction and Generation Is Easier Than You Think
从前馈到流动：统一重建与生成比你想象的更简单
Abstract
Reconstruct where the images provide evidence, and generate where they do not: recent success of spatial world models such as Atlas (World Labs Team, 2026) highlights the value of unifying reconstruction and generation in one model. Yet the two have long lived in separate paradigms with distinctive failure modes: feed-forward reconstruction averages ambiguity into blur, while conditional generation invents plausible but scene-inconsistent detail. In this work, we present a unified flow-based formulation for reconstruction and generation, where a shared clean-target predictor performs direct reconstruction at its single-step endpoint and unfolds conditional generation through multi-step flow. A controlled toy study reveals the mechanism: with a single step, the predictor collapses to the conditional mean just like feed-forward methods, favoring consistency over diversity. With multi-step inference, the fidelity of generated details grows with context richness: closer observations reduce ambiguity and yield better-matched details. We further instantiate the formulation in appearance and geometry 3D tasks. JiT-LVSM improves perceptual and distributional quality in novel view synthesis, while JUSt3R retains competitive single-step geometry prediction with additional multi-step inference capabilities that reduces veil and flying-pixel artifacts, producing cleaner surface structure with greater test-time compute. Together, they show that reconstruction and generation can share both a formulation and a backbone, with their behavior governed by denoising configuration---making unification surprisingly simple.
Chinese Translation
在图像提供证据的地方进行重建，在图像没有证据的地方进行生成：最近空间世界模型如 Atlas (World Labs Team, 2026) 的成功突显了在一个模型中统一重建与生成的价值。然而，这两者长期以来一直生活在不同的范式中，具有独特的失败模式：前馈重建将模糊的歧义平均化，而条件生成则创造出合理但与场景不一致的细节。在本研究中，我们提出了一种统一的基于流的重建与生成的公式，其中共享的干净目标预测器在其单步端点执行直接重建，并通过多步流展开条件生成。一项受控的玩具研究揭示了这一机制：在单步中，预测器像前馈方法一样崩溃到条件均值，偏向于一致性而非多样性。通过多步推理，生成细节的保真度随着上下文丰富度的增加而提高：更近的观察减少了歧义，产生更匹配的细节。我们进一步在外观和几何3D任务中实例化该公式。JiT-LVSM 在新视图合成中提高了感知和分布质量，而 JUSt3R 在保持竞争性的单步几何预测的同时，增加了多步推理能力，减少了面纱和飞像素伪影，产生了更干净的表面结构，且在测试时计算量更大。总的来说，它们表明重建与生成可以共享同一公式和主干，其行为由去噪配置所主导——使得统一变得出乎意料的简单。
cs.CV / 145 / 2609.32780
OmniMoE-VL: A Sparse Vision-Language Model with Coupled Visual-Depth Routing
OmniMoE-VL：一种具有耦合视觉-深度路由的稀疏视觉-语言模型
Abstract
Vision-language models (VLMs) increasingly use sparse mixture-of-experts (MoE) to scale language-side computation, yet visual information is typically routed only after passing through a fixed cross-modal interface. This leaves an important decision unresolved: which intermediate visual representations should be exposed to language computation for a given question? We introduce OmniMoE-VL, a sparse VLM with a coupled visual-depth routed projector. For each image-prompt pair, the projector selects a sparse set of intermediate visual depths and reuses the resulting global preference to guide both local patch fusion and dynamic visual injection into the language model. This design enables question-dependent visual access while preserving the native visual-token sequence, and complements token-level expert routing in the vision and language stacks. Across eight image-based benchmarks, OmniMoE-VL achieves an average score of 85.9 with 28B total and 9B activated parameters. Controlled comparisons show that the routed visual interface provides the dominant architectural gain, while matched route and component controls, same-image route analysis, and route interventions further support the value of coupling and question-conditioned visual access.
Chinese Translation
视觉-语言模型（VLMs）越来越多地使用稀疏专家混合（MoE）来扩展语言侧的计算，然而视觉信息通常仅在通过固定的跨模态接口后进行路由。这留下了一个重要的决策未得到解决：对于给定的问题，应该暴露哪些中间视觉表示给语言计算？我们提出了OmniMoE-VL，一种具有耦合视觉-深度路由投影器的稀疏VLM。对于每对图像-提示，投影器选择一组稀疏的中间视觉深度，并重用由此产生的全局偏好来指导局部补丁融合和动态视觉注入到语言模型中。这种设计使得依赖问题的视觉访问成为可能，同时保留了原生的视觉标记序列，并补充了视觉和语言堆栈中的标记级专家路由。在八个基于图像的基准测试中，OmniMoE-VL的平均得分为85.9，使用了28B的总参数和9B的激活参数。受控比较显示，路由的视觉接口提供了主要的架构增益，而匹配的路由和组件控制、同图像路由分析以及路由干预进一步支持了耦合和问题条件视觉访问的价值。
cs.CV / 146 / 2609.32781
CT-OPD: Counterfactual Trace On-Policy Distillation for Diffusion Vision-Language Models
CT-OPD：用于扩散视觉语言模型的反事实轨迹在线蒸馏
Abstract
Diffusion vision-language models generate answers by gradually resolving masked tokens, making accurate conditional prediction in partially resolved states central to post-training. Masking completed answers yields coherent contexts and targets, but prescribed masks do not reflect the model's reveal decisions. Its trajectories capture these decisions, yet their provisional visible tokens can conflict with the target response. Outcome-based reinforcement learning follows these trajectories but provides only response-level feedback, which loses contrast when sampled rewards tie. To align coherent token-level supervision with the model's reveal decisions, we introduce Counterfactual Trace On-Policy Distillation (CT-OPD), which combines completed teacher responses with trajectory masks from the current student. CT-OPD retokenizes each teacher response in the student's vocabulary and extracts unresolved-position masks at successive stages of the student's reverse process. For each mask, it discards provisional rollout values and reconstructs the partial state from the teacher endpoint, so the supervised positions follow the current trajectory while the visible context and targets remain consistent with the same response. The student is trained on these reconstructed states with its native categorical loss, and trajectories are refreshed as the model evolves. Across dense and sparse diffusion architectures, CT-OPD consistently enhances multimodal understanding and reasoning capabilities, with gains of up to 9.80 points on the nine-benchmark average. On the unified understanding-and-generation architecture, it also improves both visual understanding and image generation, showing that the same principle transfers across architectures and modalities. Ablations further attribute these gains to coherent reconstruction and current-model trajectory masks.
Chinese Translation
扩散视觉语言模型通过逐步解析被遮蔽的标记生成答案，使得在部分解析状态下进行准确的条件预测成为后训练的核心。遮蔽完整的答案产生连贯的上下文和目标，但规定的遮蔽并不反映模型的揭示决策。其轨迹捕捉了这些决策，但其临时可见标记可能与目标响应相冲突。基于结果的强化学习遵循这些轨迹，但仅提供响应级别的反馈，当采样奖励相互关联时，反馈的对比性会丧失。为了将连贯的标记级监督与模型的揭示决策对齐，我们提出了反事实轨迹在线蒸馏（Counterfactual Trace On-Policy Distillation, CT-OPD），该方法将完成的教师响应与当前学生的轨迹遮蔽相结合。CT-OPD在学生的词汇中重新标记每个教师响应，并在学生反向过程的连续阶段提取未解析位置的遮蔽。对于每个遮蔽，它丢弃临时的展开值，并从教师端点重建部分状态，使得监督位置遵循当前轨迹，同时可见的上下文和目标与相同的响应保持一致。学生在这些重建状态上使用其本地的分类损失进行训练，随着模型的演变，轨迹也会被刷新。在密集和稀疏的扩散架构中，CT-OPD始终增强多模态理解和推理能力，在九个基准平均上提高了多达9.80分。在统一的理解与生成架构上，它也改善了视觉理解和图像生成，表明相同的原则可以在不同架构和模态之间迁移。消融实验进一步将这些增益归因于连贯的重建和当前模型的轨迹遮蔽。
cs.CV / 147 / 2609.32789
Hierarchical Frequency-Domain Compression of Implicit Geometric Representations for Large-Scale Point Clouds
大规模点云隐式几何表示的层次频域压缩
Abstract
Large-scale point cloud representations of complex geome tries incur prohibitive computational and memory costs, necessitating compressed implicit representations. To ad dress this, we propose a unified framework comprising im plicit geometric field representation, hierarchical frequency domain compression, and conditional high-frequency predic tion. Specifically, an unordered point cloud is mapped to an implicit field defined within its physical bounding box. A smooth Fourier pyramid is then constructed, where com pact low-frequency components capture the global geometry. Inter-scale high-frequency residuals are encoded to preserve the spatial information required for reconstructing fine geo metric details. To restore the high-frequency information lost during compression, we develop a hierarchical 3D neural net work. The reconstructed implicit field is converted back into a point cloud through isosurface extraction. Experiments on a complex-boundary point cloud with more than eight mil lion points demonstrate that the proposed method achieves a higher compression ratio than existing point cloud compres sion methods while maintaining comparable reconstruction quality.
Chinese Translation
复杂几何体的大规模点云表示会产生高昂的计算和内存成本，因此需要压缩的隐式表示。为此，我们提出了一个统一框架，包括隐式几何场表示、层次频域压缩和条件高频预测。具体而言，无序点云被映射到其物理边界框内定义的隐式场中。然后构建一个平滑的傅里叶金字塔，其中紧凑的低频成分捕捉全局几何信息。跨尺度的高频残差被编码，以保留重建细致几何细节所需的空间信息。为了恢复在压缩过程中丢失的高频信息，我们开发了一个层次化的3D神经网络。重建的隐式场通过等值面提取转换回点云。在一个包含超过八百万个点的复杂边界点云上的实验表明，所提出的方法在保持可比重建质量的同时，达到了比现有点云压缩方法更高的压缩比。
cs.CV / 148 / 2609.32794
Latent Space Is Not Flat: Rethinking Latent Structure for 3D Medical Image Synthesis
潜在空间并非平坦：重新思考3D医学图像合成的潜在结构
Abstract
Latent generative models make 3D medical image synthesis computationally practical by generating in a compressed space. However, we show that the common flat Euclidean assumption induced by $\ell_2$ objectives is imprecise: latent-space geometry is so strongly anisotropic that equal-magnitude errors can produce drastically different decoded distortions. We further find that this anisotropy has a clear feature: sensitive variation concentrates in a low-rank subspace. The dominant low-rank components capture the overall structure, encoding long-range, spatially coordinated variation while remaining resistant to local noise. Its orthogonal residual, in contrast, mainly captures local and image-specific variation. Motivated by this asymmetry, we introduce Latent Structure Flow (LSF). At each block, LSF decomposes the latent state into structure and residual, models structural changes with global context, and predicts residual variation locally while preserving a direct path for the input structure. LSF changes only the generator, leaving the frozen codec and pointwise training objective unchanged. Across cross-modality synthesis and tumor inpainting tasks, LSF outperforms all compared baselines on both global and tumor-specific metrics, demonstrating the benefit of explicitly modeling latent-space structure for 3D medical image synthesis.
Chinese Translation
潜在生成模型通过在压缩空间中生成，使得3D医学图像合成在计算上变得可行。然而，我们表明，由$ ext{l}_2$目标引起的常见平坦欧几里得假设是不准确的：潜在空间的几何形状是如此强烈的各向异性，以至于相同幅度的误差可以产生截然不同的解码失真。我们进一步发现，这种各向异性具有一个明显的特征：敏感的变化集中在一个低秩子空间中。主导的低秩成分捕捉整体结构，编码长距离、空间协调的变化，同时对局部噪声保持抵抗。相反，其正交残差主要捕捉局部和图像特定的变化。基于这种不对称性，我们引入了潜在结构流（Latent Structure Flow, LSF）。在每个模块中，LSF将潜在状态分解为结构和残差，利用全局上下文建模结构变化，并在保留输入结构的直接路径的同时局部预测残差变化。LSF仅改变生成器，保持冻结的编解码器和逐点训练目标不变。在跨模态合成和肿瘤修复任务中，LSF在全球和肿瘤特定指标上均优于所有比较基线，证明了显式建模潜在空间结构对3D医学图像合成的益处。
cs.CV / 149 / 2609.32811
Progressive Risk Estimation for Accident Anticipation
渐进式事故预警风险评估
Abstract
Accident anticipation aims to recognize anomalous driving cues before a crash while avoiding false alarms during normal driving. Existing approaches typically formulate this task as binary classification, focusing on whether an accident will occur rather than when it will occur. We propose PRE-ACT, a framework that models accident risk as a continuously evolving signal that increases as the crash approaches. By explicitly enforcing temporal ordering and distance-to-accident awareness, our method progressively raises risk while suppressing premature alarms, leading to significant improvements on MM-AU subsets and Nexar. We further introduce a Separation Score to evaluate the global behavior of predicted risk curves beyond local temporal windows. Code and visualizations are available at https://github.com/giddyyupp/PRE-ACT.
Chinese Translation
事故预警旨在在碰撞发生之前识别异常驾驶信号，同时在正常驾驶过程中避免误报。现有方法通常将此任务表述为二分类问题，关注的是事故是否会发生，而非何时发生。我们提出了PRE-ACT框架，该框架将事故风险建模为一个持续演变的信号，随着碰撞的临近而增加。通过明确强制时间顺序和对事故距离的意识，我们的方法逐步提高风险，同时抑制过早的警报，从而在MM-AU子集和Nexar上实现显著改进。我们进一步引入了分离评分（Separation Score），以评估预测风险曲线在局部时间窗口之外的全局行为。代码和可视化结果可在 https://github.com/giddyyupp/PRE-ACT 获取。
cs.CV / 150 / 2609.32813
USAI-Quant: A Quantitative Reasoning Benchmark for Vision-Language Models in Built Environments
USAI-Quant：用于建筑环境中视觉-语言模型的定量推理基准
Abstract
Large vision-language models (VLMs) have emerged as a powerful paradigm for urban and spatial AI. However, current state-of-the-art large VLMs still struggle with quantitative reasoning on remote sensing imagery. Existing benchmarks and algorithms are predominantly based on qualitative Visual Question Answering (VQA), providing limited insights into the quantitative reasoning capabilities of VLMs for built environment metrics. To address this gap, we develop Quantitative Urban and Spatial AI benchmark (USAI-Quant), the first benchmark designed to quantitatively evaluate VLM's reasoning capabilities on built environment metrics via remote sensing imagery. USAI-Quant is curated from the 335 largest U.S. cities, aligning high-resolution remote sensing images with quantitative built environment metrics. We then evaluate both general-purpose and remote sensing VLMs (RS-VLMs) by applying VQAs to tens of built environment metrics across three complexity levels. Our results reveal that current state-of-the-art models consistently fall short on numeric reasoning tasks. We further conduct in-depth analyses across models, question types, and geographic locations, uncovering insights into performance variability and task-specific challenges.
Chinese Translation
大型视觉-语言模型（VLMs）已成为城市和空间人工智能的强大范式。然而，目前最先进的大型VLM在遥感图像上的定量推理仍然面临挑战。现有的基准和算法主要基于定性的视觉问答（VQA），对VLM在建筑环境指标上的定量推理能力提供的见解有限。为了解决这一问题，我们开发了定量城市和空间人工智能基准（USAI-Quant），这是第一个旨在通过遥感图像定量评估VLM在建筑环境指标上推理能力的基准。USAI-Quant是从美国335个最大城市中整理而成，将高分辨率遥感图像与定量建筑环境指标对齐。随后，我们通过对三个复杂性水平的数十个建筑环境指标应用VQA，评估了通用和遥感VLM（RS-VLMs）。我们的结果显示，当前最先进的模型在数字推理任务上始终表现不佳。我们进一步对模型、问题类型和地理位置进行了深入分析，揭示了性能变异性和任务特定挑战的见解。
cs.CV / 151 / 2609.32824
Unlocking Geodesic Gromov-Wasserstein Distances for 3D Modeling
解锁三维建模中的测地线 Gromov-Wasserstein 距离
Abstract
\textit{Gromov-Wasserstein Distances} (GWDs) provide quantitative ways of comparing probabilistic distributions defined on different metric spaces by applying techniques from the optimal transport theory. As such, GWD can be potentially useful in a large variety of applications ranging from graph matching problems to 3D object detection. However its practical use at scale is significantly limited by cubic time complexity computations involving dense intra-space distance matrices. Even though in the Euclidean metric spaces several techniques (e.g. involving scalable kernel methods) were proposed to address it, to the best of our knowledge, analogous techniques for general geodesic distances on manifolds, or shortest-path distance on graphs in their discretized variants, were not developed. In this paper, we present \textbf{E}fficient \textbf{G}eodesic \textbf{Gro}mov-\textbf{W}asserstein methods (EGGroW), a new class of efficient algorithms designed to calculate geodesic Gromov-Wasserstein distances with entropic Sinkhorn-like approaches, leveraging recently introduced \textit{GenusSink} methods \citep{genussink} and the theory of random features. We provide important downstream applications, namely: 3D pose estimation and 3D template detection. In the latter setting, we formulate a partial 3D template recovery as a staged problem: capacity-constrained scene selection is followed by semi-relaxed recovery of template visibility and correspondence. Our empirical findings show that EGGroW provides accurate solutions when standard Euclidean-based techniques fail and is characterized by light computational footprint, as our theoretical analysis predicts.
Chinese Translation
测地线 Gromov-Wasserstein 距离（GWDs）提供了一种定量比较定义在不同度量空间上的概率分布的方法，采用了最优传输理论中的技术。因此，GWD 在从图匹配问题到三维物体检测等各种应用中具有潜在的实用价值。然而，其在大规模实际应用中的使用受到涉及稠密内部空间距离矩阵的立方时间复杂度计算的显著限制。尽管在欧几里得度量空间中提出了几种技术（例如涉及可扩展核方法）来解决这一问题，但据我们所知，尚未开发出适用于流形上的一般测地线距离或其离散变体中图的最短路径距离的类似技术。在本文中，我们提出了一种新的高效算法类——高效测地线 Gromov-Wasserstein 方法（EGGroW），旨在通过采用类熵的 Sinkhorn 方法计算测地线 Gromov-Wasserstein 距离，利用最近引入的 GenusSink 方法和随机特征理论。我们提供了重要的下游应用，即三维姿态估计和三维模板检测。在后者的设置中，我们将部分三维模板恢复形式化为一个分阶段问题：容量受限的场景选择后是半放松的模板可见性和对应关系恢复。我们的实证结果表明，当标准的基于欧几里得的方法失败时，EGGroW 提供了准确的解决方案，并且其计算开销较小，正如我们的理论分析所预测的那样。
cs.CV / 152 / 2609.32840
VCRE-Fib: View-Conditioned Regional Evidence for Fine-Grained Ultrasound Grading of Schistosoma japonicum-Associated Liver Fibrosis
VCRE-Fib：基于视图条件的区域证据用于细粒度超声分级的日本血吸虫相关肝纤维化
Abstract
Accurate assessment of Schistosoma japonicum-associated liver fibrosis is essential for disease management and long-term follow-up in endemic regions. Ultrasound provides non-invasive imaging, but complex local echogenic patterns and anatomical structures make fine-grained grading challenging. Existing deep learning methods can predict fibrosis scores, yet directly incorporating acquisition views and regional cues into grading while retaining spatial information for inspection remains an open problem. Here we present VCRE-Fib, a view-conditioned regional evidence framework that integrates anatomical context, local information, and global image assessment for fine-grained ultrasound grading. The framework forms view-conditioned local grading evidence before spatial pooling, uses weak localization to guide its aggregation, and combines it with global predictions. Image-only inference jointly returns a fibrosis score, acquisition view, and candidate abnormal-region map. We developed and evaluated the method on a re-curated cohort of 108,709 ultrasound images from 6,373 patients across 35 centers. On a patient-disjoint test set of 4,107 images from 240 patients across four centers, VCRE-Fib reduced the prespecified composite grading risk by 7.115% relative to SFibAI trained and evaluated on the same data split. Image-level mean absolute error decreased from 0.391 to 0.378, alongside lower patient-max, patient-median, and center-balanced risks. The full model also achieved lower composite grading risk than variants that separately removed view conditioning or weak localization. These results support incorporating anatomical context and regional evidence into ultrasound grading while exposing spatial predictions for inspection alongside severity estimates.
Chinese Translation
准确评估日本血吸虫相关肝纤维化对于流行地区的疾病管理和长期随访至关重要。超声提供了非侵入性成像，但复杂的局部回声模式和解剖结构使得细粒度分级变得具有挑战性。现有的深度学习方法可以预测纤维化评分，但在保留空间信息的同时，直接将采集视图和区域线索纳入分级仍然是一个未解决的问题。在此，我们提出了VCRE-Fib，一个基于视图条件的区域证据框架，整合了解剖背景、局部信息和全局图像评估以实现细粒度超声分级。该框架在空间池化之前形成视图条件的局部分级证据，利用弱定位指导其聚合，并将其与全局预测相结合。仅基于图像的推断共同返回纤维化评分、采集视图和候选异常区域图。我们在来自35个中心的6,373名患者的108,709张超声图像的重新整理队列上开发并评估了该方法。在来自四个中心的240名患者的4,107张图像的患者不重叠测试集上，VCRE-Fib将预设的综合分级风险降低了7.115%，相对于在相同数据划分上训练和评估的SFibAI。图像级的平均绝对误差从0.391降至0.378，同时患者最大、患者中位数和中心平衡风险也降低。完整模型的综合分级风险也低于分别去除视图条件或弱定位的变体。这些结果支持在超声分级中纳入解剖背景和区域证据，同时暴露空间预测以供检查以及严重程度评估。
cs.CV / 153 / 2609.32841
Beyond Temporal Smoothing: Spatial Energy Budgets Stabilize One-Step Diffusion Editing
超越时间平滑：空间能量预算稳定一步扩散编辑
Abstract
One-step text-guided diffusion editing is efficient but prone to spatially misallocated updates that distort the edited object and alter the background. Existing methods often improve stability by averaging the editing field across timesteps. We instead identify spatial energy misallocation as a distinct and measurable failure mode: across two independent noise draws, the residual field is essentially unrepeatable, making the background field unreliable for direct transport, while its total energy still sets a usable magnitude for the draw at hand. BudEdit turns that magnitude into an explicit budget and reallocates it to edit-relevant regions selected jointly by residual energy and cross-attention, controlling where editing energy is spent rather than averaging over timesteps. The resulting training-free, inversion-free editor spends the budget on transport and reuses it to scale a correction in a lower-noise gated refinement. The budgeted injection field matches its prescribed budget exactly and vanishes on the identified background support, by construction. On PIE-Bench with SD-Turbo, BudEdit outperforms ChordEdit under each method's reported default settings on all 11 evaluated metrics, including a $2.1$\,dB gain in background PSNR, $31$\% lower DINO, and $36$\% lower LPIPS, while improving all five editing-quality metrics and reporting the lowest runtime in the comparison.
Chinese Translation
一步文本引导的扩散编辑效率高，但容易出现空间上错误分配的更新，从而扭曲编辑对象并改变背景。现有方法通常通过在时间步之间平均编辑场来提高稳定性。我们则将空间能量错误分配识别为一种独特且可测量的失败模式：在两个独立的噪声抽样中，残差场基本上是不可重复的，这使得背景场在直接传输时不可靠，而其总能量仍然为当前抽样设定了可用的幅度。BudEdit将该幅度转化为一个明确的预算，并将其重新分配给通过残差能量和交叉注意力共同选择的与编辑相关的区域，控制编辑能量的使用位置，而不是在时间步之间进行平均。最终生成的无训练、无反演的编辑器将预算用于传输，并在较低噪声的门控细化中重用该预算以缩放修正。预算注入场精确匹配其规定的预算，并在识别的背景支撑上消失，这是构造的结果。在使用SD-Turbo的PIE-Bench上，BudEdit在所有11个评估指标上超越了ChordEdit，且在每种方法报告的默认设置下，包括背景PSNR提高$2.1$ dB、DINO降低$31\%$和LPIPS降低$36\\%$，同时改善了所有五个编辑质量指标，并在比较中报告了最低的运行时间。
cs.CV / 154 / 2609.32846
SynCo: Learning Cross-Modal Synergy by Contrasting Interaction Residuals
SynCo：通过对比交互残差学习跨模态协同
Abstract
Multimodal contrastive learning is a dominant paradigm for learning transferable representations from unlabeled data, but standard objectives primarily capture information that is redundant between modalities. Partial Information Decomposition (PID) shows that task-relevant information in multimodal data decomposes into three components: redundancy shared between modalities, uniqueness specific to each modality, and synergy available only from their joint observation. Recent frameworks extend contrastive learning to capture all three components, yet synergy remains undertrained in practice. We propose SynCo (Synergy Contrastive Learning), a method that directly addresses synergy undertraining through dedicated supervision on an interaction residual. SynCo fits a linear projector to predict the fused representation from independently computed unimodal features, and the resulting interaction residual, which removes the linearly unimodal-predictable component, receives dedicated contrastive supervision at negligible computational cost. On the controlled Trifeature benchmark, SynCo achieves state-of-the-art synergy capture with a $+5.98\%$ gain over the baseline, and on real-world benchmarks from MultiBench, DARai, and MM-IMDb, SynCo consistently outperforms or matches prior methods across diverse modality combinations and task types. The method operates as a plug-in to existing contrastive multimodal frameworks without modifying the underlying fusion architecture and can further improve synergy capture when combined with other methods.
Chinese Translation
多模态对比学习是从未标记数据中学习可转移表示的主导范式，但标准目标主要捕捉模态之间冗余的信息。部分信息分解（Partial Information Decomposition, PID）表明，多模态数据中的任务相关信息可以分解为三个组成部分：模态之间共享的冗余、特定于每个模态的独特性，以及仅通过联合观察获得的协同。最近的框架扩展了对比学习，以捕捉这三种组成部分，但在实践中，协同仍然未得到充分训练。我们提出了SynCo（协同对比学习），一种通过对交互残差的专门监督直接解决协同未训练的问题的方法。SynCo拟合一个线性投影器，以预测从独立计算的单模态特征融合而来的表示，得到的交互残差去除了线性可预测的单模态成分，并以微不足道的计算成本接受专门的对比监督。在受控的Trifeature基准测试中，SynCo实现了最先进的协同捕捉，相较于基线提升了$+5.98\%$，在来自MultiBench、DARai和MM-IMDb的真实世界基准测试中，SynCo在各种模态组合和任务类型上始终优于或与先前的方法持平。该方法作为插件操作于现有的对比多模态框架中，无需修改底层融合架构，并且在与其他方法结合时可以进一步改善协同捕捉。
cs.CV / 155 / 2609.32856
PolyTopoBench: A Benchmark for Complex Vector Polygon Generation from Remote Sensing Imagery
PolyTopoBench：一种基于遥感影像的复杂矢量多边形生成基准测试
Abstract
Vector polygon generation converts visual inputs, e.g., remote sensing (RS) images, into vectorized polygonal geometries, supporting applications such as autonomous driving, vector map construction, and remote sensing. Early pipelines predict raster masks and post-process them into polygons, which prevents end-to-end optimization and may miss small objects or introduce inaccurate vertices. Recent methods directly generate vector polygons, but most focus on simple exterior contours, while they either cannot represent complex polygons with holes or fail to preserve their topology. In this paper, we propose PolyTopoBench, a unified evaluation framework for vector polygon generation from RS images with explicit emphasis on complex polygons. PolyTopoBench evaluates both exterior and interior rings, and benchmarks 11 representative methods, including segmentation-based polygonization pipelines, vision foundation model baselines, and specialized vector polygon generators, on two RS-image datasets covering buildings, roads, vegetation, and unvegetated regions. Experiments show that existing methods often recover simple exterior boundaries but degrade substantially on polygons with holes or multiple rings. These results reveal complex polygon generation as an unresolved challenge and motivate topology-aware benchmarks and model designs. Code and data are available at https://github.com/seai-lab/PolyTopoBench.
Chinese Translation
矢量多边形生成将视觉输入（例如遥感（RS）图像）转换为矢量化的多边形几何体，支持自主驾驶、矢量地图构建和遥感等应用。早期的处理流程预测栅格掩膜并将其后处理为多边形，这阻碍了端到端的优化，并可能错过小物体或引入不准确的顶点。最近的方法直接生成矢量多边形，但大多数集中于简单的外部轮廓，而无法表示带孔的复杂多边形或未能保持其拓扑结构。本文提出了PolyTopoBench，一个统一的评估框架，用于从RS图像生成矢量多边形，明确强调复杂多边形。PolyTopoBench评估外部和内部环，并在涵盖建筑、道路、植被和非植被区域的两个RS图像数据集上对11种代表性方法进行基准测试，包括基于分割的多边形化流程、视觉基础模型基线和专门的矢量多边形生成器。实验表明，现有方法通常能够恢复简单的外部边界，但在带孔或多个环的多边形上表现显著下降。这些结果揭示了复杂多边形生成作为一个未解决的挑战，并激励了对拓扑感知基准和模型设计的需求。代码和数据可在 https://github.com/seai-lab/PolyTopoBench 获取。
cs.CV / 156 / 2609.32857
Is H&E Image-to-Spatial Transcriptomics Simpler Than It Looks?
H&E 图像到空间转录组学是否比看起来更简单？
Abstract
Predicting spatial gene expression from routine H&E histology offers a scalable route toward spatial molecular profiling. Recent work has pursued increasingly sophisticated architectures to capture spatial context and richer expression structure. At the same time, simple estimators have shown strong performance in several studies, but what they already solve and where additional complexity is needed remain unclear. We study this behavior through the structure of prediction error under the mean-squared error (MSE) objective. Differences in average expression across genes can account for a substantial part of aggregate prediction performance, while a key unresolved error lies in recovering variation within each slide. Decomposing MSE into slide-level and within-slide components, we find that the within-slide component has lower residual-normalized parameter sensitivity in controlled neural experiments. This motivates Component-Guided Loss (CGL), which increases supervision of the within-slide component. CGL-Linear is a closed-form affine instantiation that achieves overall state-of-the-art performance across HEST-1k cohorts and gene-panel sizes. The same within-slide supervision improves existing neural models. These results suggest that substantial gains can come from aligning the training objective with prediction-error structure rather than increasing model complexity.
Chinese Translation
从常规的 H&E 组织学预测空间基因表达为空间分子分析提供了一条可扩展的路径。近期的研究追求越来越复杂的架构以捕捉空间上下文和更丰富的表达结构。同时，简单的估计器在多个研究中表现出强劲的性能，但它们已经解决了什么问题以及何处需要额外的复杂性仍然不清楚。我们通过均方误差（MSE）目标下的预测误差结构研究这种行为。基因之间的平均表达差异可以占据整体预测性能的相当大一部分，而一个关键的未解决误差在于恢复每个切片内的变异。将 MSE 分解为切片级和切片内成分，我们发现切片内成分在受控神经实验中具有较低的残差归一化参数敏感性。这激励了成分引导损失（Component-Guided Loss, CGL），该方法增加了对切片内成分的监督。CGL-Linear 是一种封闭形式的仿射实例，在 HEST-1k 队列和基因面板规模上实现了整体的最先进性能。相同的切片内监督改善了现有的神经模型。这些结果表明，通过将训练目标与预测误差结构对齐，而不是增加模型复杂性，可以获得显著的提升。
cs.CV / 157 / 2609.32863
SV2V-RSim: A Comprehensive Benchmark for Self-Selective V2V Cooperative Perception with Near-Realistic Data
SV2V-RSim：自选择V2V协作感知的综合基准，基于近真实数据
Abstract
Vehicle-to-Vehicle (V2V) cooperative perception enhances autonomous driving by enabling vehicles to share information beyond their direct line of sight. However, existing V2V datasets are limited by a small number of participating agents, static collaborator selection strategies, and a significant domain gap between simulated and real-world environments. To overcome these challenges, we introduce SV2V-RSim, a large-scale, multi-modal, near-realistic simulation dataset engineered to elevate agent diversity and realism. Additionally, we present the Select Vehicles Adaptively (SVA) module, which optimizes collaborator selection to balance perception performance against communication bandwidth constraints. Our dataset is generated using the Unreal Engine 5-based simulator that integrates high-fidelity 3D assets, diverse environments, and intricate traffic scenarios. All vehicles within a specified range of the ego vehicle are equipped with sensor suites, enabling dynamic and adaptive collaborator selection. SV2V-RSim encompasses four maps, four weather conditions, six time periods from sunrise to night, 203K LiDAR frames, 402K RGB frames, and 788K annotated 3D bounding boxes across 17 object classes, supporting a range of cooperative perception tasks such as 3D object detection, segmentation, and depth estimation. Benchmarking on recent cooperative perception algorithms demonstrates that SVA achieves a superior performance-bandwidth trade-off, while sim-to-real experiments and No-Reference Image Quality Assessment validate the dataset's high realism and practical effectiveness. Our dataset and code will be publicly available.
Chinese Translation
车对车（V2V）协作感知通过使车辆能够共享超出其直接视线的信息，从而增强了自主驾驶。然而，现有的V2V数据集受限于参与代理数量少、静态协作选择策略以及模拟与现实环境之间显著的领域差距。为了解决这些挑战，我们引入了SV2V-RSim，这是一个大规模、多模态、近真实的模拟数据集，旨在提升代理的多样性和真实感。此外，我们还提出了自适应选择车辆（SVA）模块，该模块优化了协作伙伴选择，以在感知性能与通信带宽限制之间取得平衡。我们的数据集是使用基于虚幻引擎5的模拟器生成的，该模拟器集成了高保真3D资产、多样化环境和复杂的交通场景。所有在自我车辆指定范围内的车辆都配备了传感器套件，使得动态和自适应的协作伙伴选择成为可能。SV2V-RSim涵盖了四个地图、四种天气条件、从日出到夜晚的六个时间段、203K LiDAR帧、402K RGB帧以及跨17个物体类别的788K标注3D边界框，支持一系列协作感知任务，如3D物体检测、分割和深度估计。对近期协作感知算法的基准测试表明，SVA在性能与带宽的权衡上表现优越，而模拟到现实的实验和无参考图像质量评估验证了数据集的高真实感和实际有效性。我们的数据集和代码将公开发布。
cs.CV / 158 / 2609.32876
Multimodal LLMs Outperform Pathology Foundation Models in Cross-Domain Histological Similarity
多模态大语言模型在跨领域组织学相似性方面优于病理基础模型
Abstract
State-of-the-art pathology foundation models, trained on millions of histology tiles, can fail to preserve tissue similarity when comparisons cross slide or institution boundaries. We show that general-purpose multimodal LLMs, without being trained as pathology foundation models, consistently outperform these specialized models in cross-domain histological similarity judgments. Using a relative similarity framework that we release as the MOSAIC (Model Similarity Assessment across Institutions and Cohorts) benchmark, we evaluate 17 models across 6 datasets and find that pathology encoders often rank same-institution, different-disease tiles as more similar than same-disease, different-institution tiles, a clinically dangerous failure mode invisible to standard within-domain evaluations. LLMs appear less susceptible to this failure, likely because they perform semantic visual comparison of morphology and tissue architecture rather than relying on shortcut features tied to acquisition context. Scaling training data does not resolve the problem for pathology encoders, implicating the learning objective rather than data coverage. Our results expose a fundamental robustness gap in current pathology foundation models and establish multimodal LLMs as a viable alternative for cross-institutional retrieval, dataset harmonization, and multi-site quality control. Code and data will be released upon acceptance.
Chinese Translation
最先进的病理基础模型在数百万个组织学切片上进行训练，但在跨切片或机构边界的比较中可能无法保持组织相似性。我们展示了通用的多模态大语言模型（LLMs），即使没有作为病理基础模型进行训练，在跨领域组织学相似性判断中始终优于这些专业模型。我们发布了一个相对相似性框架，命名为MOSAIC（跨机构和队列的模型相似性评估）基准，评估了17个模型在6个数据集上的表现，并发现病理编码器通常将同一机构、不同疾病的切片排名为比同一疾病、不同机构的切片更相似，这是一种在标准领域内评估中无法察觉的临床危险失效模式。大语言模型似乎对这种失效的敏感性较低，这可能是因为它们执行形态和组织结构的语义视觉比较，而不是依赖于与获取背景相关的捷径特征。扩大训练数据并未解决病理编码器的问题，暗示学习目标而非数据覆盖是根本原因。我们的结果揭示了当前病理基础模型的基本鲁棒性差距，并确立了多模态大语言模型作为跨机构检索、数据集协调和多地点质量控制的可行替代方案。代码和数据将在接受后发布。
cs.CV / 159 / 2609.32882
Improving Video Sparse Attention with Fine-grained Router and Sparse Rebasing
通过细粒度路由器和稀疏重基准提高视频稀疏注意力
Abstract
We present VSA2, a frontier trainable sparse attention for video DiTs. VSA2 includes a variety of new architectural features and training procedures that we apply across all stages of the DiT development cycle, including pretraining, RL, and inference, to produce a DiT with comparable or better quality than a full attention counterpart. Architecturally, VSA2 introduces a fine-grained router that improves the precision of identifying critical tokens and supports dynamic computation by allowing each query to attend to a variable number of key-value pairs. In training, we identify a Hard-to-Easy Curriculum, where models trained under high sparsity and later evaluated with lower sparsity during inference not only generalize effectively, but also outperform models trained with full attention in motion quality. VSA2 is also flexible: it can replace full attention during the middle of progressive low-to-high resolution pretraining, rebasing early-stage full-attention checkpoints. Experiments show that VSA2 reduces attention computation by half over VSA with lower loss. On 720p videos, it accelerates attention by 8.9x and end-to-end generation by 4.62x compared to the FlashAttention-3 baseline, while achieving comparable or better video quality.
Chinese Translation
我们提出了 VSA2，一种可训练的前沿视频稀疏注意力机制，适用于视频 DiTs。VSA2 包含多种新的架构特征和训练程序，我们在 DiT 开发周期的各个阶段应用这些特征和程序，包括预训练、强化学习（RL）和推理，以生成具有与全注意力模型相当或更好质量的 DiT。在架构上，VSA2 引入了细粒度路由器，改善了识别关键标记的精度，并通过允许每个查询关注可变数量的键值对来支持动态计算。在训练中，我们识别出一种“难到易”的课程，其中在高稀疏度下训练的模型在推理时以较低稀疏度进行评估，不仅能够有效泛化，还在运动质量上超越了使用全注意力训练的模型。VSA2 也具有灵活性：它可以在渐进式低到高分辨率预训练的中间阶段替代全注意力，重基准早期全注意力检查点。实验表明，VSA2 在降低损失的同时，将注意力计算减少了一半，相较于 VSA。在 720p 视频上，它比 FlashAttention-3 基线加速了 8.9 倍的注意力计算和 4.62 倍的端到端生成，同时实现了相当或更好的视频质量。
cs.CV / 160 / 2609.32899
Rethinking the Fully Hyperbolic Vision Transformer in Polar Coordinates
重新思考极坐标下的完全双曲视觉变换器
Abstract
Hyperbolic space can embed intrinsic hierarchies in data with low distortion due to the exponential growth of volume with distance from the origin. However, current Lorentz transformer blocks are formulated in ambient coordinates, where numerical errors increase at large radii due to instability in the Lorentzian inner product, leading many models to limit the radius to avoid this issue. This prevents us from utilizing the regions of hyperbolic space that motivate the geometry. To address this, we revisit the components of the transformer block in polar coordinates, where hyperbolic operations such as distance calculation and attention centroids can be computed without the numerical cancellation in their ambient-coordinate formulations. Specifically, we propose a polar fully connected layer that separately maps an embedding's direction and radius, allowing the radius of a feature to be learned rather than determined by the norm of a linear map. We additionally introduce horospherical shifts as relative positional encodings whose query-key distances grow logarithmically with the token gap. Finally, we reformulate the residual connection as average radius Lorentz boosts. Combining these components, we develop a fully hyperbolic transformer that substantially improves performance over Euclidean and hyperbolic baselines on standard vision tasks. We further evaluate our model on ImageNet and demonstrate its ability to generalize to other datasets using pre-trained weights, similar to Euclidean counterparts.
Chinese Translation
双曲空间能够以低失真的方式嵌入数据中的内在层次结构，这是由于距离原点越远，体积呈指数增长。然而，当前的洛伦兹变换器块是在环境坐标中构建的，在大半径下，由于洛伦兹内积的不稳定性，数值误差会增加，导致许多模型限制半径以避免此问题。这阻碍了我们利用激励几何的双曲空间区域。为了解决这个问题，我们重新审视了极坐标下变换器块的组成部分，其中双曲操作，如距离计算和注意力中心，可以在其环境坐标公式中无数值抵消地计算。具体而言，我们提出了一种极坐标全连接层，分别映射嵌入的方向和半径，使特征的半径可以被学习，而不是由线性映射的范数决定。我们还引入了作为相对位置编码的水平球面位移，其查询-键距离随着令牌间隔以对数方式增长。最后，我们将残差连接重新表述为平均半径洛伦兹提升。结合这些组件，我们开发了一种完全双曲变换器，在标准视觉任务上显著提高了相较于欧几里得和双曲基线的性能。我们进一步在ImageNet上评估了我们的模型，并展示了其在使用预训练权重时能够推广到其他数据集的能力，类似于欧几里得对应物。
cs.CV / 161 / 2609.32944
Synthetic Thermal Image Generation for Real-Time Animal Detection Under Low-Visibility Conditions
低能见度条件下实时动物检测的合成热图像生成
Abstract
Wildlife-vehicle collisions remain a significant road safety concern, particularly during nighttime and low-visibility conditions when RGB-based perception systems are often unreliable. Thermal imaging offers a promising alternative for detecting animals under poor illumination. However, the limited availability of annotated infrared animal datasets restricts the development of robust deep learning-based detection models. This paper investigates synthetic thermal image generation as a scalable approach for real-time animal detection under low-visibility conditions. A subset of 514 annotated visible-spectrum animal images from the NTLNP dataset is translated into synthetic thermal representations using CycleGAN-Turbo, while a limited real thermal dataset of 60 images is expanded through thermal-focused augmentation. Multiple object detection architectures, including YOLOv8, YOLOv9, YOLOv10, and RT-DETR, are trained independently on synthetic and real thermal datasets and evaluated using precision, recall,
[email protected],
[email protected]:0.95, model size, and inference latency. Experimental results show that synthetic thermal images provide competitive detection performance, with RT-DETR achieving the highest synthetic-data
[email protected] of 0.9613. Models trained on augmented real thermal data achieve the strongest overall performance, with YOLOv10s obtaining 0.9879
[email protected] and 0.9571
[email protected]:0.95. Computational analysis further indicates that lightweight YOLO variants provide favorable inference latency, supporting their potential for real-time deployment. These findings demonstrate that synthetic thermal imagery can reduce dependence on scarce infrared datasets and support the development of efficient animal detection systems for future vehicle-mounted wildlife collision mitigation applications.
Chinese Translation
野生动物与车辆的碰撞仍然是一个重要的道路安全问题，尤其是在夜间和低能见度条件下，基于RGB的感知系统往往不可靠。热成像为在光照不足的情况下检测动物提供了一种有前景的替代方案。然而，标注的红外动物数据集的有限可用性限制了基于深度学习的检测模型的开发。本文研究了合成热图像生成作为一种可扩展的方法，用于在低能见度条件下进行实时动物检测。我们从NTLNP数据集中提取了514幅标注的可见光谱动物图像，并使用CycleGAN-Turbo将其转换为合成热表示，同时通过热成像专注的增强扩展了一个有限的真实热数据集，包含60幅图像。包括YOLOv8、YOLOv9、YOLOv10和RT-DETR在内的多种目标检测架构分别在合成和真实热数据集上进行训练，并使用精确度、召回率、
[email protected]、
[email protected]:0.95、模型大小和推理延迟进行评估。实验结果表明，合成热图像提供了具有竞争力的检测性能，其中RT-DETR在合成数据上达到最高的
[email protected]为0.9613。基于增强真实热数据训练的模型表现出最强的整体性能，YOLOv10s获得了0.9879的
[email protected]和0.9571的
[email protected]:0.95。计算分析进一步表明，轻量级的YOLO变体提供了良好的推理延迟，支持其在实时部署中的潜力。这些发现表明，合成热成像可以减少对稀缺红外数据集的依赖，并支持未来车辆安装的野生动物碰撞缓解应用中高效动物检测系统的发展。
cs.CV / 162 / 2609.32957
DynamicDx: Evaluating Evidence Acquisition in Video-Based Diagnosis
DynamicDx：评估基于视频的诊断中的证据获取
Abstract
Diagnosing a patient from video requires more than recognizing the sign: a vision-language model must turn what it sees into hypotheses, questions and tests. DynamicDx evaluates each step in 71 neurological consultations across 11 sign categories, linking authentic patient videos to confirmed diagnoses and fixed charts built from the same case reports, so that every model queries the same evidence. Across five such models, video improves accuracy by 9.9-22.5 percentage points over blind input, but neither recognition alone nor temporal order explains the gain: the cause is usually missing from the model's video-only differential diagnosis even when the sign is recognized, and shuffling the frames produces no reliable accuracy loss. Instead, a trajectory replay traces most of the gain to the investigation results the video prompts. Evidence acquisition is the bottleneck: supplying the decisive investigations raises accuracy to 73.2-93.0%. Two interventions act on it. A post-trained 4B video describer improves sign descriptions, especially from a short, densely sampled segment, and source-clean literature retrieval expands initial hypotheses; both bring the tests a model orders closer to those the treating clinicians documented and, through them, raise accuracy. For video-based diagnosis, seeing better helps when it leads to asking better.
Chinese Translation
从视频中诊断患者不仅仅需要识别症状：视觉-语言模型必须将其所见转化为假设、问题和测试。DynamicDx评估了71个神经科咨询中的每一个步骤，涵盖11个症状类别，将真实患者视频与确认的诊断和基于相同病例报告构建的固定图表关联起来，以便每个模型查询相同的证据。在五个这样的模型中，视频的使用使准确率比盲输入提高了9.9-22.5个百分点，但仅凭识别或时间顺序并不能解释这一提升：即使症状被识别，模型的仅视频差异诊断中通常缺少原因，而打乱帧序列也不会导致可靠的准确率损失。相反，轨迹回放将大部分提升归因于视频所促发的调查结果。证据获取是瓶颈：提供决定性的调查将准确率提高到73.2-93.0%。有两个干预措施可以改善这一点。经过后训练的4B视频描述器改善了症状描述，特别是来自短而密集采样段的描述，而源清晰的文献检索则扩展了初始假设；这两者都使模型所排序的测试更接近治疗临床医生记录的测试，并通过这些测试提高了准确率。对于基于视频的诊断，改善视觉效果有助于提出更好的问题。
cs.CV / 163 / 2609.32971
Distributed Hydrological Modeling in the Feature Space
特征空间中的分布式水文建模
Abstract
Accurate forecasting of river discharge and floods is very challenging. River dynamics are affected by storage, meteorological forcing, and flow propagation at different spatial and temporal scales. Forecasting thus requires a framework that considers the upstream-to-downstream flow through river networks across grid cells and catchments. This modeling is known in hydrology as distributed modeling and routing. Existing deep learning approaches either ignore this topology, operate on lumped catchments, or route predicted physical quantities through a separate graph or physical routing model. We instead introduce feature-space routing: a topology-aware state-space operator embedded directly in the forecasting dynamics. At every forecast step, the operator gathers latent states from upstream grid cells and causally updates the downstream state according to the known river network. This preserves the physical connectivity of the river system while allowing the propagated state itself to be learned end-to-end and allows the model to predict river discharge considering both local dynamics and neighboring upstream contributions. To address uncertainty and provide probabilistic forecasts, we minimize the fair continuous ranked probability score (fCRPS) as a training objective. Our experiments on the European Flood Awareness System (EFAS) and observational data for river discharge forecasting demonstrate that encoding the physical structure of river networks explicitly in the feature space substantially improves the forecasting skill, particularly in an ungauged setting. Our approach achieves state-of-the-art results on both reanalysis and observational data and is able to forecast maps of river discharge at 1 arcminute and 6-hourly resolution up to 10 days lead time.
Chinese Translation
准确预测河流流量和洪水是非常具有挑战性的。河流动态受存储、气象强迫和在不同空间和时间尺度上的流动传播影响。因此，预测需要一个框架，考虑通过网格单元和流域的河流网络进行的上游到下游的流动。这种建模在水文学中被称为分布式建模和路由。现有的深度学习方法要么忽略这种拓扑结构，要么在汇流区上进行操作，或者通过单独的图形或物理路由模型来路由预测的物理量。相反，我们提出了特征空间路由：一种嵌入在预测动态中的拓扑感知状态空间算子。在每个预测步骤中，该算子从上游网格单元收集潜在状态，并根据已知的河流网络因果更新下游状态。这保留了河流系统的物理连通性，同时允许传播的状态本身以端到端的方式进行学习，并使模型能够在考虑局部动态和邻近上游贡献的情况下预测河流流量。为了应对不确定性并提供概率预测，我们将公平连续排名概率评分（fCRPS）最小化作为训练目标。我们在欧洲洪水意识系统（EFAS）和河流流量预测的观测数据上的实验表明，在特征空间中明确编码河流网络的物理结构显著提高了预测技能，特别是在无测站的情况下。我们的方法在再分析和观测数据上均达到了最先进的结果，并能够在10天的预报期内以1弧分和6小时的分辨率预测河流流量图。
cs.CV / 164 / 2609.32984
ReVision3D: Attribution-Guided Recursive Self-Improvement for 3D Medical Perception
ReVision3D：基于归因的递归自我改进用于三维医学感知
Abstract
Recursive self-improvement (RSI) offers a promising path for overcoming the limited visual capability of current medical imaging agents. Yet applying RSI to volumetric imaging remains difficult: failures can arise from acquisition, perception, training recipe, or downstream inference, while self-generated feedback and logged trajectories provide little guidance on which component should change. We introduce ReVision3D, an RSI system that leverages 3D volumes with spatially grounded annotations to determine where visual evidence is lost and recursively improve the corresponding visual capability. A frozen language-model designer proposes revisions to acquisition, perception, training, or inference, while the verifier and system-level objective remain fixed. Our key insight is that an annotated volume forms an exact replay world for view rendering and spatial verification: unvisited views can be rendered on demand, and localized predictions can be checked directly against reference masks. This grounded feedback directs targeted revision, while only changes that improve beyond measured seed noise are retained. Each accepted change triggers renewed attribution, allowing the dominant bottleneck to shift across rounds. On abdominal CT, attribution identifies perception as the dominant remaining limitation. Revising that level enables ReVision3D to achieve 79% liver recall and 83% kidney recall at under 0.4 false positives per patient, outperforming the evaluated frozen multimodal foundation models, with the largest gains on small lesions.
Chinese Translation
递归自我改进（RSI）为克服当前医学成像代理的有限视觉能力提供了一条有前景的路径。然而，将RSI应用于体积成像仍然困难：失败可能源于采集、感知、训练方案或下游推理，而自生成的反馈和记录的轨迹对哪些组件应进行更改提供的指导有限。我们提出了ReVision3D，一个利用具有空间基础注释的三维体积来确定视觉证据丢失位置并递归改善相应视觉能力的RSI系统。一个冻结的语言模型设计师提出对采集、感知、训练或推理的修订，而验证者和系统级目标保持不变。我们的关键见解是，带注释的体积形成了一个精确的重放世界，用于视图渲染和空间验证：未访问的视图可以按需渲染，局部预测可以直接与参考掩膜进行检查。这种有依据的反馈指导了有针对性的修订，同时仅保留那些超过测量种子噪声的改进变化。每个被接受的变化都会触发新的归因，使得主瓶颈在多个回合中发生转移。在腹部CT上，归因识别出感知是主要的剩余限制。修订该层级使ReVision3D能够在每位患者低于0.4个假阳性的情况下实现79%的肝脏召回率和83%的肾脏召回率，超越了评估的冻结多模态基础模型，并在小病变上获得了最大的提升。
cs.CV / 165 / 2609.32996
Oracle Gaps in Reliability Coverage: Sampling Noise or Policy Specialization?
可靠性覆盖中的Oracle差距：采样噪声还是策略专业化？
Abstract
Policies trained from the same base model can appear to solve different problems. An oracle that chooses the best policy for each problem may therefore appear much stronger than any single policy. Selecting the largest estimated success rate also selects favorable sampling errors. We study this effect through reliability coverage, the fraction of problems whose success probability reaches a chosen threshold. Our first test redistributes stored correctness outcomes across policies within each problem. A second also preserves each policy's total successes, accounting for overall quality differences under a specified statistical model. For five training seeds of a seven-billion-parameter vision-language model, redistribution reproduces 0.096 of an estimated 0.113 oracle gap at threshold 0.10. Neither test finds significant evidence at this threshold. Small advantages remain unresolved. Mixtures, routers, voting, and weight averaging show no detectable improvement over their corresponding single-policy baselines. Training policies on different datasets shows little detectable specialization under light post-training, and no router gain. A stronger recipe does create it, both tests detect it, and a router gain appears only at the high thresholds where the specialists separate. In a control with predictable specialization, a router recovers about half the oracle gap. Coverage bounds explain why even a genuine oracle advantage need not yield a deployment gain. The tests assess apparent specialization from stored responses before investment in routing. Code: https://github.com/KurbanIntelligenceLab/oracle-gaps
Chinese Translation
从同一基础模型训练出的策略可能看似解决不同的问题。因此，选择每个问题最佳策略的oracle可能显得比任何单一策略强得多。选择最大估计成功率也会选择有利的采样误差。我们通过可靠性覆盖来研究这一效应，即成功概率达到所选阈值的问题比例。我们的第一次测试在每个问题内重新分配存储的正确性结果。第二次测试则保留每个策略的总成功次数，同时考虑在指定统计模型下的整体质量差异。对于一个拥有七十亿参数的视觉-语言模型的五个训练种子，重新分配在阈值0.10下再现了0.096的估计0.113的oracle差距。两个测试在该阈值下均未发现显著证据。小的优势仍未得到解决。混合、路由、投票和权重平均在其对应的单一策略基线下未显示出可检测的改善。在不同数据集上训练策略在轻微的后训练下显示出很少的可检测专业化，且没有路由增益。一个更强的方案确实创造了它，两个测试均检测到了它，且路由增益仅在专家分离的高阈值下出现。在一个具有可预测专业化的控制中，路由器大约恢复了oracle差距的一半。覆盖界限解释了即使是真正的oracle优势也不一定带来部署增益的原因。这些测试评估了在投资路由之前存储响应的明显专业化。代码链接：https://github.com/KurbanIntelligenceLab/oracle-gaps
cs.CV / 166 / 2609.33003
Certified Interface Aliases: Exact Collisions in Vision-Language Preprocessing, and When They Exist
认证接口别名：视觉-语言预处理中的精确冲突及其存在条件
Abstract
Vision-language verifiers and routers must distinguish errors repairable by more reasoning from those caused by visual evidence never reaching the language model. This distinction lacks ground truth because annotators see full-resolution images while models receive preprocessed tensors. We introduce AliasForge to create cases where the relevant fact is provably absent from the interface. Fixed-point resampling makes the pre-rounding resize an exact integer linear map that can send nonzero integer perturbations to zero. Hiding a label-flipping perturbation there produces images with opposite step-correctness labels but bit-identical interface states. Every verifier therefore has the same output law on both members, giving pair-balanced accuracy exactly one half and zero gain from language-side repair. We prove that every fixed-point downscaler has such null vectors and bound their smallest size at the most common ratios, which rules out an 8-bit fit whenever the bound exceeds 255. From the resize configuration alone, a lattice criterion supplies realizable collisions and certifies their absence within the specified construction family. It resolves all 20 screened configurations, 17 as constructible and 3 as non-constructible. We construct certified pairs across three architectures and certify four additional processors, with zero decision-logit gap on all 18 scored pairs and none of the 18 controls. The pairs also screen routers that waste computation on re-attention or further reasoning. On natural items, per-item routing headroom exists, but no tested interface-only router improves over stopping. Our fiber ceiling bounds the headroom recoverable from the interface. Code: https://github.com/KurbanIntelligenceLab/aliasforge.
Chinese Translation
视觉-语言验证器和路由器必须区分可通过更多推理修复的错误与因视觉证据未能到达语言模型而导致的错误。这一区分缺乏真实依据，因为标注者看到的是全分辨率图像，而模型接收到的是预处理的张量。我们引入AliasForge来创建相关事实在接口中可证明缺失的情况。固定点重采样使得预缩放调整成为一个精确的整数线性映射，可以将非零整数扰动发送到零。在此处隐藏标签翻转扰动会产生具有相反步骤正确性标签但位相同的接口状态的图像。因此，每个验证器在两个成员上都有相同的输出法则，使得成对平衡准确率恰好为一半，并且从语言侧修复中没有增益。我们证明每个固定点降采样器都有这样的零向量，并在最常见的比率下限制其最小大小，这排除了在界限超过255时的8位适配。从缩放配置本身，晶格标准提供可实现的冲突并证明它们在指定构造家族内的缺失。它解决了所有20个筛选配置，其中17个为可构造，3个为不可构造。我们在三种架构中构建认证对，并认证四个额外的处理器，所有18对得分对的决策逻辑差距为零，18个对照组均无此差距。这些对还筛选出在重新关注或进一步推理上浪费计算的路由器。在自然项目上，每个项目的路由余量存在，但没有测试的仅接口路由器能改善停止策略。我们的纤维上限限制了从接口可恢复的余量。代码链接： https://github.com/KurbanIntelligenceLab/aliasforge.
cs.CV / 167 / 2609.33005
Safety-Constrained Cascade Inference for Robust Malaria Cell Classification Under Field Corruptions
安全约束的级联推断用于在现场干扰下的稳健疟疾细胞分类
Abstract
Automated malaria diagnosis from thin blood-smear microscopy could meaningfully reduce the burden on under-resourced laboratories, but a model that maximises accuracy on clean laboratory images fails badly the moment an inexpensive smartphone camera introduces sensor noise. This paper introduces MalariaCascade, a two-stage inference system in which a lightweight MobileNetV2 sentinel (2,225,153 parameters) makes confident classifications at low compute cost and escalates uncertain cases to an EfficientNet-B3 expert (10,697,769 parameters) that sees only clean, standardised images regardless of how corrupted the incoming frame is. Structural isolation of the expert stage, not learned robustness, is the mechanism. The sentinel is trained under a safety-score objective that places an explicit floor on Recall(Parasitised) (>=0.95) and Recall(Uninfected) (>=0.40) before precision is optimised, ensuring the checkpoint satisfies clinical safety constraints by construction. On the NIH Malaria Cell Images Dataset (27,558 cells), the cascade reaches Accuracy=0.9736, Recall(Parasitised)=0.9570, Precision(Parasitised)=0.9912, F1=0.9738, and AUROC=0.9955 on the clean test set. Under Gaussian sensor noise at full severity, cascade Recall(Parasitised) degrades by only 2.4 pp (0.9570 to 0.9329), while the flat single-model baseline collapses by 63.0 pp (0.9584 to 0.3281). McNemar's test confirms the cascade improvement is statistically significant (chi^2=11.14, p=0.00085). An ablation isolating the structural property shows that routing clean images to the expert is responsible for a 16.8 pp Recall(Parasitised) advantage under sensor noise relative to a cascade where the expert also sees corrupted inputs.
Chinese Translation
通过薄血涂片显微镜进行自动化疟疾诊断可以显著减轻资源不足实验室的负担，但在干净的实验室图像上最大化准确性的模型在廉价智能手机相机引入传感器噪声时表现不佳。本文介绍了MalariaCascade，一个两阶段的推断系统，其中轻量级的MobileNetV2哨兵（2,225,153个参数）在低计算成本下做出自信的分类，并将不确定的案例升级到仅查看干净、标准化图像的EfficientNet-B3专家（10,697,769个参数），无论输入帧受到多大程度的干扰。专家阶段的结构隔离，而非学习的鲁棒性，是其机制。哨兵在一个安全评分目标下进行训练，该目标对Recall(Parasitised)（>=0.95）和Recall(Uninfected)（>=0.40）设定了明确的下限，然后优化精确度，确保检查点在构建时满足临床安全约束。在NIH疟疾细胞图像数据集（27,558个细胞）上，级联达到了Accuracy=0.9736，Recall(Parasitised)=0.9570，Precision(Parasitised)=0.9912，F1=0.9738，以及AUROC=0.9955的干净测试集结果。在全强度的高斯传感器噪声下，级联的Recall(Parasitised)仅下降了2.4个百分点（从0.9570降至0.9329），而单模型基线则崩溃了63.0个百分点（从0.9584降至0.3281）。McNemar检验确认级联的改进在统计上是显著的（chi^2=11.14，p=0.00085）。一个孤立结构特性的消融实验表明，将干净图像路由到专家负责在传感器噪声下相对于专家也看到干扰输入的级联，带来了16.8个百分点的Recall(Parasitised)优势。
cs.CV / 168 / 2609.33020
Residual Diffusion Implicit Models
残差扩散隐式模型
Abstract
Diffusion models achieve state-of-the-art results across multiple tasks. However, in inverse problems, standard initialization from pure Gaussian noise misaligns the generative process with real-world degradations. More recent methods such as diffusion bridges impose strict endpoint constraints and often require long reverse processes that are prone to hallucinations. Alternative consistency models provide noise-invariant, one-step mappings but lack inherent variance modeling and can degrade under severe corruption. Hence, residual diffusion implicit models (RDIMs) are proposed, constituting a generalized framework that explicitly models the residuals between high-quality (HQ) and low-quality (LQ) images, aligning the forward process with the actual degradation. A non-Markovian implicit reverse sampler is derived, which can skip intermediate timesteps, enabling accurate few-step or even single-step reconstruction, while mitigating the hallucinations inherent to long diffusion chains. RDIM also introduces a controllable variance mechanism that interpolates between deterministic and stochastic sampling, balancing fidelity and diversity. Furthermore, it enables the straightforward use of perceptual losses, when needed. Experiments on denoising and super-resolution benchmarks demonstrate that RDIMs consistently outperforms the state of the art, including bridge and consistency models, in terms of PSNR, SSIM, and LPIPS, reducing hallucinations while requiring only a few sampling steps (often just one). The results position RDIMs as an efficient solution for a broad range of image restoration tasks.
Chinese Translation
扩散模型在多个任务中取得了最先进的结果。然而，在逆问题中，来自纯高斯噪声的标准初始化使生成过程与现实世界的降质不一致。最近的方法如扩散桥施加了严格的端点约束，通常需要较长的反向过程，这容易导致幻觉。替代的一致性模型提供了噪声不变的一步映射，但缺乏固有的方差建模，并且在严重损坏的情况下可能会退化。因此，提出了残差扩散隐式模型（RDIMs），构成了一个广义框架，明确建模高质量（HQ）和低质量（LQ）图像之间的残差，使前向过程与实际降质对齐。推导出一种非马尔可夫隐式反向采样器，可以跳过中间时间步，从而实现准确的少步甚至单步重建，同时减轻长扩散链固有的幻觉。RDIM还引入了一种可控的方差机制，在确定性和随机采样之间进行插值，平衡保真度和多样性。此外，当需要时，它还能够简单地使用感知损失。在去噪和超分辨率基准测试中的实验表明，RDIM在PSNR、SSIM和LPIPS方面始终优于最先进的方法，包括桥和一致性模型，减少了幻觉，同时仅需少量采样步骤（通常仅一步）。这些结果将RDIM定位为广泛图像恢复任务的高效解决方案。
cs.CV / 169 / 2609.33076
NutriVision: Ingredient-Conditioned Fusion and Prediction for Single-Image Food Nutrition Estimation
NutriVision：基于成分条件的融合与单图像食品营养估计的预测
Abstract
Nutrition estimation is a fundamental task in consumer diet tracking, clinical dietetics, chronic disease management, sports and hospital nutrition, and broader food computing systems. The existing approaches have progressed along two largely separate axes, vision models that rely on calibrated RGB-depth captures and ingredient-aware methods that use textual cues but use limited multimodal fusion. We introduce NutriVision, an end-to-end framework that leverages visual geometry and ingredient semantics to estimate calories, mass, fat content, carbohydrates, and protein from a single RGB image and an optional ingredient list. It obtains the unavailable depth modality using DepthAnything-V3 and encodes ingredient descriptions using CLIP. It integrates three complementary mechanisms: (1) an \emph{Ingredient-Conditioned Frequency-Aligned Fusion Module (IC-FAFM)}, which uses textual guidance to reweight and align RGB-depth frequency components; (2) an \emph{Ingredient-Aware Mask-based Prediction Head (IA-MPH)}, whose gating and channel masks are conditioned on food identity; and (3) modality-specific \emph{Internal Semantic Modeling (ISM)} blocks. On the Nutrition5k dataset, NutriVision achieves a mean PMAE of $\mathbf{13.60\pm0.10\%}$, outperforming our IGSMNet implementation by $0.89$ percentage points and OmniFood8k by $2.90$ percentage points (both $p<0.001$). The module-level ablations identify the ingredient-aware prediction head as the primary architectural contributor, improving mean PMAE by $1.50\pm0.17$ percentage points ($p<0.001$). These results demonstrate that ingredient-conditioned prediction and frequency-aware RGB-depth fusion provide measurable gains for single-image nutrient estimation. More broadly, NutriVision offers a practical route toward nutrition-assessment systems that exploit geometric and semantic cues without requiring specialized depth-sensing hardware
Chinese Translation
营养估计是消费者饮食追踪、临床营养学、慢性疾病管理、体育和医院营养以及更广泛的食品计算系统中的一项基础任务。现有的方法主要沿着两个相对独立的方向发展：依赖于校准的RGB-深度捕捉的视觉模型和利用文本线索但多模态融合有限的成分感知方法。我们提出了NutriVision，一个端到端框架，利用视觉几何和成分语义从单个RGB图像及可选的成分列表中估计卡路里、质量、脂肪含量、碳水化合物和蛋白质。它使用DepthAnything-V3获取缺失的深度模态，并使用CLIP编码成分描述。它集成了三种互补机制：（1） extit{成分条件频率对齐融合模块（IC-FAFM）}，利用文本指导重新加权和对齐RGB-深度频率分量；（2） extit{成分感知基于掩码的预测头（IA-MPH）}，其门控和通道掩码基于食品身份条件；（3）模态特定的 extit{内部语义建模（ISM）}模块。在Nutrition5k数据集上，NutriVision实现了平均PMAE为$ extbf{13.60 extpm0.10 ext{ extperthousand}}$，比我们的IGSMNet实现提高了$0.89$个百分点，比OmniFood8k提高了$2.90$个百分点（均为$p<0.001$）。模块级消融实验确定成分感知预测头是主要的架构贡献者，使平均PMAE提高了$1.50 extpm0.17$个百分点（$p<0.001$）。这些结果表明，基于成分条件的预测和频率感知的RGB-深度融合为单图像营养估计提供了可测量的提升。更广泛地说，NutriVision为利用几何和语义线索的营养评估系统提供了一条实用的途径，而无需专门的深度传感硬件。
cs.CV / 170 / 2609.33077
Parameter-Efficient 3D Segmentation of Liver and Liver tumors: Depthwise factorization Scales Better Than Dense Convolution with Spatial Dimensionality
参数高效的肝脏及肝肿瘤三维分割：深度可分离因式分解优于具有空间维度的稠密卷积
Abstract
Three-dimensional dense convolutional networks are the strongest performers on volumetric medical image segmentation, but their parameter counts scale poorly: moving a dense k x k convolution to k x k x k multiplies its weights by k. We observe that depthwise separable factorization does not share this penalty. Because the cubic kernel term applies only to the depthwise stage while the pointwise projection, which dominates the parameter count, is unchanged, the same architecture grows by 5 % from 2D to 3D where a dense convolutional U-Net grows by 200 %. We exploit this asymmetry to build a 3D U-Net with 536,990 parameters, 24x fewer than an identical dense 3D U-Net. On MSD Task03 Liver (the Medical Segmentation Decathlon liver task, derived from LiTS), evaluated per case under five-fold cross-validation over all 131 public volumes, the model reaches a tumor Dice of 0.577 (95% CI [0.518, 0.633]) and a liver Dice of 0.947 (95% CI [0.941, 0.952]). Its liver Dice exceeds previously reported performance. Its tumor Dice exceeds their low-resolution configuration (0.4701) by 0.107 and their 2D configuration (0.5394), at approximately one twenty-fourth of the parameters and roughly half the in-plane resolution. Trained under identical conditions on a common held-out split, it exceeds a dense 3D U-Net on liver by +0.031 Dice (paired p = 0.006) and on tumor by +0.041 (95 % CI [+0.005, +0.081], paired p = 0.056), suggesting the factorization also acts as a regularizer in the small-data regime characteristic of medical imaging. We further show, on both LiTS and a 2D endoscopy benchmark, that a large fraction of the network's learnable spatial filters can be replaced by fixed shifts at no cost in accuracy, but that replacing all of them is measurably worse, the placement of spatial capacity matters more than its total amount.
Chinese Translation
三维稠密卷积网络在体积医学图像分割中表现最为出色，但其参数数量的扩展效果较差：将稠密的 k x k 卷积扩展到 k x k x k 时，其权重会增加 k 倍。我们观察到，深度可分离因式分解并不受此限制。由于立方核项仅适用于深度阶段，而主导参数数量的逐点投影保持不变，因此相同架构在从二维到三维时仅增长 5%，而稠密卷积 U-Net 则增长 200%。我们利用这种不对称性构建了一个具有 536,990 个参数的三维 U-Net，比相同的稠密三维 U-Net 少 24 倍。在 MSD Task03 Liver（医学分割十项全能肝脏任务，源自 LiTS）上，在对所有 131 个公共体积进行五折交叉验证的情况下，该模型在肿瘤 Dice 指标上达到了 0.577（95% CI [0.518, 0.633]），在肝脏 Dice 指标上达到了 0.947（95% CI [0.941, 0.952]）。其肝脏 Dice 超过了之前报告的性能。其肿瘤 Dice 超过了低分辨率配置（0.4701）0.107，且超出了其二维配置（0.5394），参数数量约为其 1/24，平面分辨率大约为其一半。在相同条件下对一个共同的保留分割进行训练时，其肝脏 Dice 超过了稠密三维 U-Net +0.031（配对 p = 0.006），肿瘤 Dice 超过了 +0.041（95% CI [+0.005, +0.081]，配对 p = 0.056），这表明因式分解在医学成像特有的小数据环境中也起到了正则化的作用。我们进一步展示，在 LiTS 和一个二维内窥镜基准上，网络的大部分可学习空间滤波器可以被固定的位移替代，而不影响准确性，但替换所有滤波器的效果明显更差，空间容量的布置比其总量更为重要。
cs.CV / 171 / 2609.33082
SemReward-VL: Semantic Reward-Guided Video-Language Adaptation for Developmental Behavior Assessment
SemReward-VL：基于语义奖励的视频-语言适应用于发展行为评估
Abstract
Developmental screening videos show how children perform specific behaviors, but clinical records usually contain outcomes rather than descriptions of what happened. We present SemReward-VL, which learns to describe item-specific behavior from these outcomes. A vision-language model generates a description, and a frozen language model scores its agreement with the clinical outcome, relevance to the item, abstention on unrelated video-item pairs, and clarity. Group relative policy optimization (GRPO) updates LoRA adapters using this semantic reward. On 13,379 videos covering 41 items, the method improves aggregate accuracy and the number of items with recall above 0.5. Errors remain in temporal direction, duration, and age-specific interpretations of behavior.
Chinese Translation
发展筛查视频展示了儿童如何表现特定行为，但临床记录通常包含结果而非发生的描述。我们提出了SemReward-VL，它从这些结果中学习描述特定项目的行为。一个视觉-语言模型生成描述，而一个冻结的语言模型则对其与临床结果的一致性、与项目的相关性、对无关视频-项目对的回避以及清晰度进行评分。组相对策略优化（GRPO）使用这种语义奖励更新LoRA适配器。在覆盖41个项目的13,379个视频中，该方法提高了整体准确性和召回率超过0.5的项目数量。错误仍然存在于时间方向、持续时间和年龄特定的行为解释中。
cs.CV / 172 / 2609.33083
dKFD: Phase-Structured Evidence Allocation for Fixed-Budget Localized Event Understanding
dKFD：用于固定预算局部事件理解的相位结构证据分配
Abstract
Sparse video understanding often requires selecting a small set of visual evidence under a fixed frame budget. Most sparse selectors allocate this budget globally, allowing all frames to compete with one another. For temporally localized events, this can be a poor inductive bias: useful evidence is often distributed across pre-event context, the event itself, and post-event consequences. We study fixed-budget evidence allocation for localized event videos and show that globally competitive Top-$K$ selectors can preserve recognition and grounding while producing unstable event evidence. On DoTA Video Anomaly Recognition, Global Top-$K$ obtains competitive recognition and temporal grounding, but low selector-event alignment at $K=12$ (Frame AUC $51.5 \pm 10.8$). We propose dKFD, a phase-structured differentiable selector that reserves evidence capacity across pre-event, event, and post-event phases after full-sequence temporal encoding. Under matched-budget multi-seed evaluation, dKFD improves Frame AUC by $+30.97$ over a matched Global Top-$K$ selector at $K=12$ ($p<0.01$), while yielding modest but statistically significant recognition gains and comparable temporal grounding. Mechanism ablations show that phase supervision is load-bearing: removing it reduces Frame AUC to $41.1 \pm 12.0$ even when phase-partitioned budgets are retained. Downstream diagnostics on VRU-Accident show consistent gains over learned Global Top-$K$ across VLM families, while dense captioning reveals a boundary condition where uniform sampling remains competitive. These results support phase-structured allocation as a controlled fixed-budget approach for event-centric sparse evidence selection, not as a universal video summarization strategy.
Chinese Translation
稀疏视频理解通常需要在固定帧预算下选择一小组视觉证据。大多数稀疏选择器在全局范围内分配这一预算，使得所有帧相互竞争。对于时间局部化事件，这可能是一种较差的归纳偏差：有用的证据通常分布在事件前的背景、事件本身以及事件后的后果中。我们研究了局部事件视频的固定预算证据分配，并表明全局竞争的 Top-$K$ 选择器可以在产生不稳定事件证据的同时保持识别和定位。在 DoTA 视频异常识别中，全球 Top-$K$ 获得了具有竞争力的识别和时间定位，但在 $K=12$ 时选择器与事件的对齐度较低（帧 AUC 为 $51.5 imes 10.8$）。我们提出了 dKFD，一种相位结构的可微选择器，在完整序列时间编码后在事件前、事件和事件后阶段保留证据容量。在匹配预算的多种子评估下，dKFD 在 $K=12$ 时相比于匹配的全球 Top-$K$ 选择器提高了 $+30.97$ 的帧 AUC（$p<0.01$），同时带来了适度但统计显著的识别增益和可比的时间定位。机制消融实验表明相位监督是承重的：即使保留相位划分预算，去除相位监督会将帧 AUC 降低到 $41.1 imes 12.0$。在 VRU-Accident 的下游诊断中，dKFD 在不同 VLM 家族上相较于学习的全球 Top-$K$ 显示出一致的增益，而密集字幕则揭示了一个边界条件，在该条件下均匀采样仍然具有竞争力。这些结果支持相位结构分配作为一种受控的固定预算方法，用于事件中心的稀疏证据选择，而不是作为一种通用的视频摘要策略。
cs.CV / 173 / 2609.33088
QSCP: Beyond Class-Name Prompts for Query-Guided Semantic Change Parsing
QSCP：超越类别名称提示的查询引导语义变化解析
Abstract
Traditional change detection (CD) identifies changes between bi-temporal remote sensing images, while semantic change detection (SCD) assigns predefined land-cover classes. However, mapping all changes may not meet a user's specific needs. Referring change detection (RCD) enables selective retrieval through category prompts. However, existing category-prompted RCD uses the queried category to specify the destination of a change and returns only a binary mask of the corresponding regions. Users may instead request a particular transition and paired semantic maps to understand what changed into what. Such requests require explicit source and target reasoning beyond target-class localization. To address these needs, we propose query-guided semantic change parsing (QSCP), which supports category names, synonyms, and intent-bearing sentences and returns a query-specific mask with paired temporal semantic maps. QSCP parses requests into intents and semantic slots, composes bidirectional visual evidence, and predicts both temporal states with a query-conditioned decoder. On SECOND, QSCP outperforms RCDNet on synonym, sentence, and transition queries and improves end-to-end semantic prediction over evaluated semantic baselines. WHU-CDC experiments further assess cross-dataset transfer and consistency across equivalent expressions without target-domain training. Code is available at https://github.com/qianyuancs/QSCP
Chinese Translation
传统的变化检测（CD）识别双时相遥感图像之间的变化，而语义变化检测（SCD）则为预定义的土地覆盖类别分配标签。然而，映射所有变化可能无法满足用户的特定需求。参考变化检测（RCD）通过类别提示实现选择性检索。然而，现有的类别提示RCD使用查询类别来指定变化的目标，并仅返回相应区域的二进制掩模。用户可能会请求特定的转变和配对的语义图，以理解变化的内容。这类请求需要超越目标类别定位的明确源和目标推理。为满足这些需求，我们提出了查询引导语义变化解析（QSCP），该方法支持类别名称、同义词和具有意图的句子，并返回查询特定的掩模和配对的时态语义图。QSCP将请求解析为意图和语义槽，组合双向视觉证据，并使用查询条件解码器预测两个时态。在SECOND数据集上，QSCP在同义词、句子和转变查询中优于RCDNet，并在评估的语义基准上提高了端到端的语义预测。WHU-CDC实验进一步评估了跨数据集的迁移和在没有目标领域训练的情况下对等表达的一致性。代码可在 https://github.com/qianyuancs/QSCP 获取。
cs.CV / 174 / 2609.33097
Query, Align, and Distill: Navigation-Aware Cross-Modal Interaction for Efficient Vision-and-Language Navigation
查询、对齐与提炼：基于导航的跨模态交互以实现高效的视觉-语言导航
Abstract
Recent large-scale Vision-and-Language Navigation (VLN) models deliver strong accuracy but remain costly to deploy due to their large parameter counts and computational requirements. We tackle efficient VLN in two steps. First, we build a high-performing teacher that makes navigation evidence selection explicit and compressible. The teacher introduces a small set of learnable query slots to extract global and local action-sufficient navigable evidence from panoramic observations via a Navigable Query Generator, then progressively grounds these evidence tokens to the instruction with an Instruction-Query Aligner for policy prediction. Second, using this explicit query bottleneck as a distillation interface, we train a compact student by transferring both where to attend and what to do. We distill the teacher's global and local navigable queries with a navigation-aware token-adaptive objective, then further match action distributions during fine-tuning. Experiments on standard VLN benchmarks demonstrate that our student nearly matches the teacher's navigation performance while reducing the number of parameters by 93.65% compared to the teacher.
Chinese Translation
最近的大规模视觉-语言导航（VLN）模型在准确性上表现出色，但由于其庞大的参数量和计算需求，部署成本依然较高。我们通过两个步骤来解决高效的VLN问题。首先，我们构建一个高性能的教师模型，使导航证据选择变得明确且可压缩。该教师模型引入了一小组可学习的查询槽，通过可导航查询生成器（Navigable Query Generator）从全景观察中提取全局和局部的足够导航证据，然后通过指令-查询对齐器（Instruction-Query Aligner）逐步将这些证据标记与指令进行对接，以进行策略预测。其次，利用这一明确的查询瓶颈作为提炼接口，我们通过转移注意力的位置和行动内容来训练一个紧凑的学生模型。我们使用基于导航的令牌自适应目标对教师的全局和局部可导航查询进行提炼，然后在微调过程中进一步匹配行动分布。在标准VLN基准上的实验表明，我们的学生模型在导航性能上几乎与教师模型相当，同时与教师模型相比，参数数量减少了93.65%。
cs.CV / 175 / 2609.33100
Octree-based Video Representation
基于八叉树的视频表示
Abstract
Video models commonly use uniform grids even though visual complexity varies substantially across space and time. We introduce OctVideo, which approximates a video clip with an octree. This hierarchy recursively partitions a spatio-temporal volume into eight subvolumes, so that smooth regions remain coarse while detailed regions receive finer cells. Each leaf stores local RGB values and spatio-temporal gradients, supplemented by a lightweight learned residual. For reconstruction, a Conv1D VAE maps the serialized cells to a regular latent grid and selectively refines details during decoding. Our VAE achieves 36.12 dB PSNR with 38.2M parameters and 189.4 GFLOPs per clip on Kinetics-400 (K400). It also generalizes zero-shot to the high-resolution Densely Annotated VIdeo Segmentation (DAVIS) 2016 dataset with reconstruction quality comparable to the best evaluated models. On both datasets, it requires the fewest model FLOPs and achieves the fastest encoding and decoding among the evaluated models. OctVideo also supports video understanding, achieving competitive recognition performance with few input tokens when trained from scratch. By exploiting the redundancy already present in video signals and efficiently processing sparse structures, OctVideo provides an efficient representation for video.
Chinese Translation
视频模型通常使用均匀网格，尽管视觉复杂性在空间和时间上有显著变化。我们提出了OctVideo，它通过八叉树来近似视频片段。该层次结构递归地将时空体积划分为八个子体积，使得平滑区域保持粗糙，而细节区域则获得更细的单元。每个叶节点存储局部RGB值和时空梯度，并辅以轻量级的学习残差。为了重建，Conv1D变分自编码器（VAE）将序列化的单元映射到一个规则的潜在网格，并在解码过程中选择性地细化细节。我们的VAE在Kinetics-400（K400）数据集上以38.2M参数和189.4 GFLOPs每个片段实现了36.12 dB的峰值信噪比（PSNR）。它还能够零-shot泛化到高分辨率的密集标注视频分割（DAVIS）2016数据集，重建质量可与最佳评估模型相媲美。在这两个数据集上，它所需的模型FLOPs最少，并且在评估的模型中实现了最快的编码和解码。OctVideo还支持视频理解，在从零开始训练时，使用较少的输入标记实现了具有竞争力的识别性能。通过利用视频信号中已经存在的冗余并高效处理稀疏结构，OctVideo为视频提供了一种高效的表示。
cs.CV / 176 / 2609.33109
Toward Comprehensive 3D Grounding: Orientation Grounding through Vision-Language Models
迈向全面的三维定位：通过视觉-语言模型进行方向定位
Abstract
Grounding is a core capability of spatial vision-language models, yet most existing work focuses only on where a referred object is. Many 3D tasks also require knowing how it is oriented. Although existing 3D VLMs may predict oriented boxes, box pose does not explicitly capture object-centric orientation or symmetry-induced ambiguities. We introduce orientation grounding, a referring grounding task that predicts an object's 6D orientation and axial symmetry from a language or box query in single-view or multi-view scenes. To support this task, we construct ReferOri, with 331K multi-view and 387K single-view orientation-grounding queries obtained through scalable reconstruction, consistency checking, and human verification. We further present OG-VLM, which adapts a 3D VLM with structured box/orientation outputs, sign and symmetry tokens, and geometry-aware auxiliary losses. Across single-view and multi-view benchmarks, OG-VLM substantially outperforms orientation-aware VLM baselines and surpasses object-level orientation foundation models on scene-level referring benchmarks, showing that explicit orientation grounding is a distinct and learnable capability beyond localization. Downstream results validate its benefit for orientation-related spatial reasoning.
Chinese Translation
定位是空间视觉-语言模型的核心能力，但现有的大多数研究仅关注所提及物体的位置。许多三维任务还需要了解物体的方向。尽管现有的三维视觉-语言模型（VLM）可能预测定向框，但框的姿态并未明确捕捉物体中心的方向或对称性引起的模糊性。我们引入了方向定位，这是一种通过语言或框查询在单视图或多视图场景中预测物体的6D方向和轴对称性的定位任务。为了支持这一任务，我们构建了ReferOri，包含331K多视图和387K单视图的方向定位查询，这些查询通过可扩展重建、一致性检查和人工验证获得。我们进一步提出了OG-VLM，该模型适配了具有结构化框/方向输出、符号和对称性标记以及几何感知辅助损失的三维视觉-语言模型。在单视图和多视图基准测试中，OG-VLM显著超越了方向感知的视觉-语言模型基线，并在场景级的引用基准上超越了物体级方向基础模型，显示出显式方向定位是一种超越定位的独特且可学习的能力。下游结果验证了其在与方向相关的空间推理中的益处。
cs.CV / 177 / 2609.33125
Train Together or Merge Later? Unifying VLA Experts via a Shared Action Interface
一起训练还是后期合并？通过共享动作接口统一 VLA 专家
Abstract
Co-training offers a straightforward way to build a multi-task vision-language-action (VLA) policy, but can fall short of the performance achieved by training each task independently. The challenge is to retain these task-specific gains in a multi-task policy without joint post-training. Combining independently trained experts through model merging is a natural approach, yet strong individual experts do not necessarily yield a strong merged policy. We identify one source of this incompatibility: task-specific changes to the action interface, comprising action normalization and the action encoder and decoder. We propose PolicyWeave, combining merge-compatible post-training with context-guided sparse merging. During post-training, all experts retain the common base policy's action interface, while task adaptation is restricted to LoRA updates in the hidden layers of the action model. This makes the experts more compatible with existing model merging methods. However, merging all experts can still introduce interference from unrelated tasks at deployment. PolicyWeave scores each expert's LoRA updates using the initial visual-language context, determines the expert set through leave-one-layer-out ranking stability, and forms a sparse weighted merge of the selected updates that remains fixed for current task. We evaluate PolicyWeave with GR00T N1.5 on 18 RoboCasa365 tasks, using only 10% of the target-task demonstrations for supervised fine-tuning (SFT). Preserving the shared action interface raises the average success rate across four static merging methods from 17.0% to 52.8%. PolicyWeave achieves 64.7% success with these SFT experts and 74.1% after task-specific reinforcement learning (RL), compared with 60.7% for joint RL. Further evaluations on LIBERO-10 and an AgileX Piper arm support the deployment of independently learned skills in long-horizon and real-world manipulation.
Chinese Translation
共同训练提供了一种构建多任务视觉-语言-动作（VLA）策略的简单方法，但其性能可能不及独立训练每个任务所达到的效果。挑战在于如何在多任务策略中保留这些任务特定的优势，而无需联合后期训练。通过模型合并将独立训练的专家结合起来是一种自然的方法，但强大的个体专家不一定能产生强大的合并策略。我们确定了这种不兼容性的一个来源：动作接口的任务特定变化，包括动作归一化以及动作编码器和解码器。我们提出了 PolicyWeave，结合了与合并兼容的后期训练和基于上下文的稀疏合并。在后期训练过程中，所有专家保留公共基础策略的动作接口，而任务适应仅限于动作模型隐藏层中的 LoRA 更新。这使得专家与现有模型合并方法更兼容。然而，合并所有专家在部署时仍可能引入来自无关任务的干扰。PolicyWeave 使用初始视觉-语言上下文对每个专家的 LoRA 更新进行评分，通过留一层排名稳定性确定专家集，并形成一个针对当前任务固定的稀疏加权合并的选定更新。我们在 18 个 RoboCasa365 任务上使用 GR00T N1.5 评估了 PolicyWeave，仅使用 10% 的目标任务演示进行监督微调（SFT）。保留共享动作接口使得四种静态合并方法的平均成功率从 17.0% 提升至 52.8%。使用这些 SFT 专家的 PolicyWeave 成功率达到 64.7%，经过任务特定的强化学习（RL）后达到 74.1%，而联合 RL 的成功率为 60.7%。在 LIBERO-10 和 AgileX Piper 臂上的进一步评估支持独立学习技能在长时间和现实世界操作中的部署。
cs.CV / 178 / 2609.33148
DroneWAM: Efficient World Action Model for Drone Visual Navigation
DroneWAM：高效的无人机视觉导航世界动作模型
Abstract
World-action models give visual navigation agents a way to anticipate how candidate actions will change future observations and to act from the predicted consequences. For drones, this capability must operate under tight accuracy and efficiency constraints. We present DroneWAM, an efficient world-action model for drone visual navigation. DroneWAM adopts a JEPA-based architecture to model future states directly in representation space, avoiding the cost of explicit future image generation. A pretrained Resampler further compresses dense encoder features into fewer latent tokens, reducing the computation repeated at each imagined step. We also introduce adaptive rollout, where a preference-trained Gate adaptively allocates prediction depth according to the current scene. To support learning under richer aerial motion, we construct DroneNav-6D, a simulated visual navigation dataset with synchronized RGB observations, 6-DoF flight trajectories, control commands, and randomized wind disturbances. On DroneNav-6D, DroneWAM achieves the best trajectory accuracy among the compared methods. Adaptive rollout further reduces the average prediction depth from 8 to 4.58 while improving trajectory accuracy, demonstrating that predictive computation can be allocated more effectively across scenes. \href{https://github.com/1e12Leon/DroneWAM}{Codes and data} will be released.
Chinese Translation
世界动作模型为视觉导航代理提供了一种预测候选动作如何改变未来观测的方法，从而根据预测的结果进行行动。对于无人机而言，这种能力必须在严格的准确性和效率限制下运行。我们提出了DroneWAM，一种高效的无人机视觉导航世界动作模型。DroneWAM采用基于JEPA的架构，直接在表征空间中建模未来状态，避免了显式生成未来图像的成本。一个预训练的重采样器进一步将密集编码器特征压缩为更少的潜在标记，减少了在每个想象步骤中重复的计算。我们还引入了自适应展开，其中一个经过偏好训练的门控（Gate）根据当前场景自适应地分配预测深度。为了支持在更丰富的空中运动下进行学习，我们构建了DroneNav-6D，这是一个模拟的视觉导航数据集，包含同步的RGB观测、6自由度飞行轨迹、控制命令和随机风扰动。在DroneNav-6D上，DroneWAM在比较方法中实现了最佳的轨迹准确性。自适应展开进一步将平均预测深度从8减少到4.58，同时提高了轨迹准确性，证明了预测计算可以在场景之间更有效地分配。代码和数据将被发布。
cs.CV / 179 / 2609.33158
FOCUS: Benchmarking Retinal Model Generalization from Foundation Vision Encoders to Multimodal LLMs
FOCUS：从基础视觉编码器到多模态大语言模型的视网膜模型泛化基准测试
Abstract
Progress in AI-based retinal image analysis has advanced with foundation models, yet evaluating their reliability remains challenging. Performance reported on a single dataset does not capture how models behave under dataset shift, across clinical definitions, or for different patient subgroups. This limitation is particularly critical in medical imaging analysis, where robustness, calibration, and fairness are essential for safe deployment. We introduce FOCUS (Foundation Ophthalmic Cross-Dataset Understanding under Shift), a cross-dataset benchmark for evaluating retinal fundus models that considers vision-only encoder models (VM), vision-language dual-encoder models (VLM), and multimodal large language models (MLLM). FOCUS harmonizes binary diabetic retinopathy, referable diabetic retinopathy, and glaucomatous optic neuropathy tasks across ten public datasets spanning diverse geographies, acquisition conditions, and label protocols. The benchmark evaluates models through a unified analysis layer that measures ranking performance, calibration, subgroup disparities, and image-quality robustness. We present a large-scale evaluation covering 532 base configurations and 228 MLLM configurations adapted through supervised fine-tuning with low-rank adaptation (LoRA). Results show that no model family consistently dominates across tasks and datasets: general VM encoders achieve the strongest average ranking performance, medical MLLMs are competitive but variable, and dual encoder VLMs benefit substantially from lightweight adaptation. Fine-tuning improves in-domain performance but exhibits heterogeneous transfer to external datasets, particularly in calibration. These findings demonstrate that retinal model evaluation is inherently multidimensional. FOCUS provides a practical framework and public benchmark to assess generalization, reliability, and robustness beyond single-dataset leaderboards
Chinese Translation
基于人工智能的视网膜图像分析随着基础模型的发展而取得了进展，但评估其可靠性仍然具有挑战性。仅在单一数据集上报告的性能无法反映模型在数据集转移、临床定义差异或不同患者子群体中的表现。这一局限性在医学影像分析中尤为重要，因为在安全部署中，模型的鲁棒性、校准性和公平性至关重要。我们引入了FOCUS（基础眼科跨数据集理解），这是一个用于评估视网膜眼底模型的跨数据集基准，考虑了仅视觉编码器模型（VM）、视觉-语言双编码器模型（VLM）和多模态大语言模型（MLLM）。FOCUS在十个公共数据集中协调了二元糖尿病视网膜病变、可转介糖尿病视网膜病变和青光眼性视神经病变任务，这些数据集涵盖了不同的地理区域、采集条件和标签协议。该基准通过统一的分析层评估模型，测量排名性能、校准性、子群体差异和图像质量鲁棒性。我们展示了一个大规模评估，涵盖532个基础配置和228个通过低秩适应（LoRA）进行监督微调的MLLM配置。结果表明，没有任何模型家族在任务和数据集上始终占据主导地位：一般的VM编码器在平均排名性能上表现最强，医学MLLM具有竞争力但表现不稳定，而双编码器VLM在轻量适应中受益显著。微调提高了领域内性能，但在外部数据集的转移表现不均，特别是在校准方面。这些发现表明，视网膜模型评估本质上是多维的。FOCUS提供了一个实用框架和公共基准，以评估超越单一数据集排行榜的泛化、可靠性和鲁棒性。
cs.CV / 180 / 2609.33167
FloodDiffusion 2: Efficient and Path Controllable Streaming Motion Generation
FloodDiffusion 2：高效且可路径控制的流式运动生成
Abstract
We present FloodDiffusion 2 (FD2), an efficient and controllable framework that builds upon FloodDiffusion (FD1), a state-of-the-art streaming motion generation model. While FD1 produces plausible motion, it suffers from low efficiency and limited controllability, as its attention design requires repeated computation over the entire history, and it lacks precise trajectory control for real-world applications. To address these limitations and improve generation quality, FD2 introduces three advances. First, Partial Attention makes finalized history representations independent of the active window, enabling KV-cached inference and shared-history packing for efficient training. Second, we establish a necessary-and-sufficient Bregman criterion for regression losses to preserve diffusion's conditional-mean velocity field. This criterion guides an FK-induced quadratic loss that incorporates motion geometry without online FK evaluation. Third, FD2 introduces precise path conditioning to control the character's root trajectory while preserving natural body motion. Experiments show that FD2 reduces training computation by 4.6$\times$ and accelerates denoising by 11.29$\times$, reaching 2.303 ms per update on long sequences. Alongside these efficiency gains, FD2 improves motion quality over FD1 and achieves state-of-the-art FID scores among streaming methods, with 0.048 on SEED and 0.053 on HumanML3D.
Chinese Translation
我们提出了FloodDiffusion 2 (FD2)，这是一个高效且可控的框架，基于FloodDiffusion (FD1)这一最先进的流式运动生成模型。虽然FD1能够生成合理的运动，但其效率低下且可控性有限，因为其注意力设计需要对整个历史进行重复计算，并且缺乏对现实应用中精确轨迹的控制。为了解决这些局限性并提高生成质量，FD2引入了三项进展。首先，部分注意力（Partial Attention）使得最终的历史表示独立于活动窗口，从而实现KV缓存推理和共享历史打包以提高训练效率。其次，我们建立了一个必要且充分的Bregman标准，用于回归损失，以保持扩散的条件均值速度场。该标准指导了一种FK诱导的二次损失，结合了运动几何而无需在线FK评估。第三，FD2引入了精确的路径条件，以控制角色的根轨迹，同时保持自然的身体运动。实验表明，FD2将训练计算减少了4.6倍，加速去噪过程达11.29倍，在长序列上每次更新达到2.303毫秒。在这些效率提升的同时，FD2在运动质量上超过了FD1，并在流式方法中实现了最先进的FID分数，在SEED上为0.048，在HumanML3D上为0.053。
cs.CV / 181 / 2609.33176
ABO-Med: Accelerated Bilevel Optimization for Few-Shot Medical Image Classification
ABO-Med：用于少样本医学图像分类的加速双层优化
Abstract
In recent years, bilevel optimization has been widely used in a variety of machine learning tasks. However, prior bilevel optimization algorithms generally require the computation of second-order information, which limits their practical scalability. Only recently has a first-order paradigm for bilevel optimization been established, attaining near-optimal theoretical guarantees for solving bilevel optimization problems. In this paper, we propose ABO-Med, a scalable instantiation of this paradigm for few-shot learning, by incorporating it into the model-agnostic meta-learning (MAML) framework and tailoring it to medical image classification. We also introduce Medical Adaptive RandomAugment (MedRAug), a modality-aware augmentation strategy designed for medical images. Theoretically, ABO-Med establishes the optimality of MAML-type meta-learning approaches. Empirically, ABO-Med outperforms prior baselines on several public medical datasets, with gains of 1.99% to 18.76%, while MedRAug further improves the average accuracy by 2.20% to 6.34%. Additional cross-domain experiments, augmentation ablation studies, backbone ablation studies, and training efficiency analysis further validate the effectiveness and efficiency of the proposed method.
Chinese Translation
近年来，双层优化在各种机器学习任务中得到了广泛应用。然而，先前的双层优化算法通常需要计算二阶信息，这限制了它们的实际可扩展性。直到最近，才建立了一种一阶范式用于双层优化，达到了近似最优的理论保证，以解决双层优化问题。在本文中，我们提出了ABO-Med，这是这一范式在少样本学习中的可扩展实例，通过将其纳入模型无关的元学习（MAML）框架，并针对医学图像分类进行调整。我们还引入了医学自适应随机增强（MedRAug），这是一种针对医学图像设计的模态感知增强策略。从理论上讲，ABO-Med确立了MAML类型元学习方法的最优性。从实证上看，ABO-Med在多个公共医学数据集上超越了先前的基线，提升幅度为1.99%至18.76%，而MedRAug进一步将平均准确率提高了2.20%至6.34%。额外的跨领域实验、增强消融研究、骨干网络消融研究和训练效率分析进一步验证了所提方法的有效性和效率。
cs.CV / 182 / 2609.33178
Can Protein-Derived Knowledge Improve Pathology Foundation Models?
蛋白质衍生知识能否改善病理基础模型？
Abstract
Molecularly guided pathology foundation models (PFMs) exploit transcriptomic or proteomic information to enrich whole-slide image (WSI) representations, yet effectively leveraging large standalone molecular corpora remains challenging. First, existing molecular foundation models encode protein sequences or single-cell states, not the patient-level bulk expression profiles paired with WSIs. Second, because cross-modal supervision is restricted to paired WSI-omics samples, knowledge from standalone molecular corpora reaches the pathology encoder only indirectly, creating a paired-support bottleneck. To address these challenges, we propose a three-stage framework that decouples proteomic knowledge acquisition from cross-modal transfer, yielding ProSlide, a slide-level hierarchical pathology foundation model. First, to close the modality gap, we pretrain a Proteomic Foundation Encoder (PFE) on 12,695 sample-level bulk protein profiles using virtual profile generation and expression-space multi-view pretraining. Second, we pretrain ProSlide, a patch-region-slide encoder, to predict protein expression from paired WSI-protein samples. Third, to relax the paired-support bottleneck, we introduce Prot2Path, a cross-modal relational distillation objective. For each paired sample, it aligns the similarity distributions of the WSI and its protein profile over a shared, frozen bank of PFE-encoded paired and standalone profiles. We evaluate ProSlide on 12 downstream tasks across breast, lung, and renal cancers. Despite being pretrained with only 2,229 WSIs and 12,695 sample-level protein profiles, ProSlide achieves the highest mean accuracy and AUC within each cancer group.
Chinese Translation
分子引导的病理基础模型（PFMs）利用转录组或蛋白组信息来丰富全切片图像（WSI）表示，但有效利用大型独立分子语料库仍然具有挑战性。首先，现有的分子基础模型编码蛋白质序列或单细胞状态，而不是与WSI配对的患者级别整体表达谱。其次，由于跨模态监督仅限于配对的WSI-组学样本，来自独立分子语料库的知识只能间接传递给病理编码器，从而形成配对支持瓶颈。为了解决这些挑战，我们提出了一个三阶段框架，将蛋白质组知识获取与跨模态转移解耦，产生了ProSlide，一个切片级层次病理基础模型。首先，为了缩小模态差距，我们在12,695个样本级整体蛋白质谱上预训练了一个蛋白质基础编码器（PFE），使用虚拟谱生成和表达空间多视图预训练。其次，我们预训练了ProSlide，一个补丁-区域-切片编码器，以从配对的WSI-蛋白样本中预测蛋白质表达。第三，为了放宽配对支持瓶颈，我们引入了Prot2Path，一个跨模态关系蒸馏目标。对于每个配对样本，它对WSI及其蛋白质谱的相似性分布进行对齐，基于一个共享的、冻结的PFE编码的配对和独立谱库。我们在乳腺癌、肺癌和肾癌的12个下游任务上评估了ProSlide。尽管仅使用2,229个WSI和12,695个样本级蛋白质谱进行预训练，ProSlide在每个癌症组中都达到了最高的平均准确率和AUC。
cs.CV / 183 / 2609.33190
FocusDrive: Reasoning with Visual Focus for Autonomous Driving
FocusDrive：基于视觉焦点的自主驾驶推理
Abstract
Driving decisions depend on both where to focus and how to act on what is seen. Effective driving reasoning must establish which objects matter, where they are, and how they inform the intended action. Text-based rationales can describe a driving response while leaving its correspondence to specific visual evidence implicit. Visual focus provides a concrete starting point for this connection by identifying what matters in the scene and where it is. We propose FocusDrive, a structured multimodal reasoning framework that organizes end-to-end planning around explicit visual focus. It pairs descriptions of decision-relevant objects with image-patch references, bringing explicit visual focus into the reasoning that generates driving plans and trajectories. We first assess this focus representation through driver gaze prediction, then investigate its role in planning reasoning using driving-focus annotations within existing NAVSIM training scenes. Experiments on W3DA and NAVSIM demonstrate competitive gaze prediction and end-to-end planning performance, with FocusDrive improving over text-based chain-of-thought. These results support visual focus as an effective link between scene understanding and driving action.
Chinese Translation
驾驶决策依赖于关注的对象以及如何对所见进行反应。有效的驾驶推理必须确定哪些对象重要、它们的位置以及它们如何影响预期的行动。基于文本的推理可以描述驾驶反应，但其与特定视觉证据的对应关系往往是隐含的。视觉焦点为这一联系提供了一个具体的起点，通过识别场景中重要的内容及其位置。我们提出了FocusDrive，一个结构化的多模态推理框架，围绕显式视觉焦点组织端到端的规划。它将与决策相关的对象描述与图像块引用相结合，将显式视觉焦点引入生成驾驶计划和轨迹的推理中。我们首先通过驾驶员注视预测评估这一焦点表示，然后利用现有NAVSIM训练场景中的驾驶焦点注释研究其在规划推理中的作用。在W3DA和NAVSIM上的实验表明，FocusDrive在注视预测和端到端规划性能上表现出竞争力，且优于基于文本的思维链。这些结果支持视觉焦点作为场景理解与驾驶行动之间有效的联系。
cs.CV / 184 / 2609.33202
ReAL: Accelerating Flow Matching through Segment Advancement with Shared Lookahead
ReAL：通过共享前瞻加速流匹配的段推进
Abstract
Flow-matching models generate high-quality images and videos, but repeated neural network evaluations make sampling expensive. Skipping evaluations reduces this cost by extending an available velocity estimate over a longer span. However, local velocity agreement alone does not determine a suitable span, and checking each candidate endpoint adds costly model calls. We introduce ReAL, a training-free sampler that selects how far to advance using one shared lookahead. Our key insight is that the discrepancy between uncorrected and lookahead-corrected endpoint proposals can be computed directly from the observed velocity mismatch and the candidate span beyond the lookahead. This relation provides a span-dependent selection criterion without additional endpoint evaluations. The same lookahead selects the longest passing candidate span, corrects the accepted update, and supplies its velocity as the next starting estimate. After initialization, each regular iteration requires only one fresh evaluation. ReAL uses the pretrained velocity output and original noise schedule, with no additional training or access to internal features. Experiments cover four image-generation backbones, video generation, and image editing. ReAL achieves 4.91x measured speedup on FLUX.1-dev while retaining 97.0% of dense mean ImageReward. On HunyuanVideo, it achieves a 5.49x speedup while maintaining a VBench score close to that of dense sampling.
Chinese Translation
流匹配模型生成高质量的图像和视频，但重复的神经网络评估使得采样成本高昂。跳过评估通过在更长的范围内扩展可用的速度估计来降低这一成本。然而，仅依靠局部速度一致性并不能确定合适的范围，检查每个候选端点会增加昂贵的模型调用。我们提出了ReAL，这是一种无训练的采样器，利用一个共享的前瞻来选择推进的距离。我们的关键见解是，未校正的端点提议与前瞻校正的端点提议之间的差异可以直接从观察到的速度不匹配和超出前瞻的候选范围中计算得出。这一关系提供了一种基于范围的选择标准，而无需额外的端点评估。同一前瞻选择最长的通过候选范围，校正接受的更新，并将其速度作为下一个起始估计。在初始化后，每次常规迭代只需一次新的评估。ReAL使用预训练的速度输出和原始噪声调度，无需额外的训练或访问内部特征。实验涵盖了四个图像生成基础模型、视频生成和图像编辑。在FLUX.1-dev上，ReAL实现了4.91倍的测量加速，同时保留了97.0%的密集均值图像奖励。在HunyuanVideo上，它实现了5.49倍的加速，同时保持了接近于密集采样的VBench得分。
cs.CV / 185 / 2609.33203
Structured Residual Connectivity Matters for Diffusion Transformers
结构化残差连接对扩散变换器的重要性
Abstract
Diffusion Transformers (DiTs) have established themselves as a scalable backbone for high-fidelity image synthesis. However, unlike U-Net based diffusion models that rely on rigid, hand-crafted skip connections, DiTs predominantly use a uniform residual stream that integrates all preceding layers as a monolithic state. In this work, we rethink residual connections in diffusion transformers and propose to transform them from passive summation into an active retrieval mechanism optimized for image denoising. First, we conduct a systematic analysis of DiT's internal representation, revealing a latent preference for early-layer feature reuse and symmetric layer guidance. Motivated by this, we introduce a structured connectivity design that explicitly integrates local residual connections with long-range pathways. Instead of static skip connections or dense all-layer routing, our method enables each transformer block to selectively ``attend'' to critical earlier representations, dynamically retrieving spatial and semantic cues through direct, differentiable cross-depth paths. Experiments show that our adaptive connectivity leads to faster convergence, with up to $1.73\times$ fewer training iterations, and significant gains in FID and visual quality with less than $0.1\%$ additional parameters, further improving a strong REPA-XL/2 model from $5.9$ to $4.34$ FID without guidance and reaching $1.39$ FID with classifier-free guidance. Our findings suggest that adaptive cross-layer connectivity is a critical yet underexplored factor in diffusion transformers, and that incorporating structured information pathways provides a simple and effective direction for improving scalable generative models.
Chinese Translation
扩散变换器（Diffusion Transformers, DiTs）已被确立为高保真图像合成的可扩展基础。然而，与依赖于刚性、手工制作的跳跃连接的基于 U-Net 的扩散模型不同，DiTs 主要使用统一的残差流，将所有前置层作为一个整体状态进行整合。在本研究中，我们重新思考了扩散变换器中的残差连接，并提出将其从被动求和转变为一种针对图像去噪优化的主动检索机制。首先，我们对 DiT 的内部表示进行了系统分析，揭示了对早期层特征重用和对称层引导的潜在偏好。基于此，我们引入了一种结构化连接设计，明确地将局部残差连接与长程路径集成在一起。我们的方法使每个变换器块能够选择性地“关注”关键的早期表示，通过直接、可微分的跨深度路径动态检索空间和语义线索，而不是静态的跳跃连接或密集的全层路由。实验表明，我们的自适应连接导致了更快的收敛，训练迭代次数减少了多达 $1.73 imes$，并在 FID 和视觉质量上显著提升，额外参数少于 $0.1 ext{ extperthousand}$，进一步将强大的 REPA-XL/2 模型的 FID 从 $5.9$ 提升至 $4.34$（无引导），并在无分类器引导下达到 $1.39$ FID。我们的研究结果表明，自适应跨层连接是扩散变换器中一个关键但未被充分探索的因素，而整合结构化信息路径为改善可扩展生成模型提供了一个简单而有效的方向。
cs.CV / 186 / 2609.33210
Background Gradients Shape Memorization in Flow Matching
背景梯度塑造流匹配中的记忆
Abstract
Repetition is closely associated with memorization in generative models, but how other training images affect the retention and copying of targets remains unclear. We study this question in class-conditioned flow matching, where images outside the target set form the background. At fixed target repetition and same-class background row count, replacing repeated same-class images with distinct images reduces the target extraction rate from 80.7% to 18.0%. To explain this effect, we develop a paired-trajectory framework that isolates target-induced parameter displacement and the background gradient response to it. This response has an exact path-integrated curvature representation, connecting background loss geometry to target learning. Reciprocal response transfer between repeated and distinct backgrounds changes target retention and copying in both directions, establishing the response's causal role. After target removal, the response correction parallel to the target-induced displacement preserves approximately 90% of the copying effects of full response transfer. Directly scaling the displacement also changes copying without further training. The post-removal copying effects of reciprocal transfer are reproduced across datasets and architectures. Together, these results identify the background gradient response as a mechanism through which same-class training data shape the retention of target learning and the reproduction of target images.
Chinese Translation
重复与生成模型中的记忆密切相关，但其他训练图像如何影响目标的保留和复制仍不清楚。我们在类别条件的流匹配中研究这个问题，其中目标集外的图像形成背景。在固定的目标重复和同类背景行数下，用不同的图像替换重复的同类图像会将目标提取率从80.7%降低到18.0%。为了解释这一效应，我们开发了一个配对轨迹框架，隔离目标引起的参数位移和背景梯度对此的响应。该响应具有精确的路径积分曲率表示，连接背景损失几何与目标学习。重复背景与不同背景之间的相互响应转移改变了目标的保留和复制，确立了响应的因果作用。在目标移除后，平行于目标引起的位移的响应修正保留了大约90%的完整响应转移的复制效应。直接缩放位移也会在没有进一步训练的情况下改变复制。相互转移的移除后复制效应在不同数据集和架构中得到了重现。综合来看，这些结果将背景梯度响应确定为同类训练数据塑造目标学习保留和目标图像再现的机制。
cs.CV / 187 / 2609.33217
RepFlow: Reciprocal Supervision Improves Generation and Representation in Flow Models
RepFlow：互惠监督提升流模型中的生成与表征
Abstract
Generative models learn visual structure through denoising, yet their internal states are entangled with both noise level and network depth, making it difficult to obtain a stable visual representation from the generator itself. We introduce RepFlow, which learns such a representation from the generator's evolving computation and uses it to guide generation. Specifically, a separate timestep-free encoder is trained, through a timestep-conditioned predictor, to recover generator states across depths and noise levels from masked clean images. By excluding the reference noise and masked-out content from the encoder's input, we encourage the encoder to distill visual information that is recoverable from the visible image context and predictive of generator states. The representation learned from the generator's evolving states is then fed back to guide and improve the generator, whose updated states provide supervision for further representation learning, forming a reciprocal learning process. This reciprocal process improves multi-step generation and representation quality, as measured by frozen linear probing on ImageNet with latent-space SiT and pixel-space JiT, without an externally pretrained representation teacher. Across the two unconditional settings, FID decreases by $19.2$--$40.8\%$ relative to native training, while linear-probe accuracy improves by 6.80--10.04 percentage points over the best searched raw generator features. The learned representation also serves as a distributional metric for one-step JiT post-training, extending its role from instance-level alignment to distribution-level supervision.
Chinese Translation
生成模型通过去噪学习视觉结构，但其内部状态与噪声水平和网络深度交织在一起，使得从生成器本身获得稳定的视觉表征变得困难。我们提出了RepFlow，它从生成器不断演变的计算中学习这种表征，并利用它来指导生成。具体而言，通过一个时间步条件预测器，训练一个独立的无时间步编码器，以从被遮蔽的干净图像中恢复生成器在不同深度和噪声水平下的状态。通过排除参考噪声和被遮蔽内容的输入，我们鼓励编码器提取可从可见图像上下文中恢复的视觉信息，并预测生成器状态。然后，从生成器不断演变的状态中学习到的表征被反馈用于指导和改善生成器，其更新后的状态为进一步的表征学习提供监督，形成一个互惠学习过程。该互惠过程改善了多步生成和表征质量，通过在ImageNet上使用潜在空间的SiT和像素空间的JiT进行冻结线性探测进行测量，而无需外部预训练的表征教师。在这两种无条件设置中，FID相较于原生训练降低了19.2%至40.8%，而线性探测准确率在最佳搜索的原始生成器特征上提高了6.80至10.04个百分点。所学习的表征还作为一步JiT后训练的分布度量，扩展了其从实例级对齐到分布级监督的角色。
cs.CV / 188 / 2609.33218
Scope-WM: Scoped Computation for Efficient Visual World Models
Scope-WM：高效视觉世界模型的范围计算
Abstract
Visual world models enable robotic planning by predicting future observations, but dense latent-state propagation and sample-intensive trajectory optimization incur high inference latency and peak memory usage, limiting real-time deployment on resource-constrained platforms. Existing sparse world-model acceleration methods either rely on unguided token sparsification, which may discard planning-relevant information and restrict achievable sparsity, or introduce heavy auxiliary modules and cumbersome multi-stage training pipelines. In this work, we present Scope-WM, an efficient visual world model that scopes computation to prediction-relevant latent regions and promising action sequences. Scope-WM distills prediction relevance into a lightweight action-conditioned selector and applies full dynamics prediction only to a compact subset of selected tokens. It updates the remaining tokens using a compact summary of foreground states and their changes, allowing the background to perceive foreground dynamics without costly token-to-token interactions. During planning, Scope-WM preserves and reuses high-quality action sequences discovered during the initial MPC search, focusing subsequent search under reduced rollout budgets. The resulting pipeline requires only a one-off selector distillation followed by a single joint training stage for the sparse world model. On the challenging Push-T task, Scope-WM reduces peak GPU memory usage and planning time to $18.1\%$ and $14.3\%$ of those of dense DINO-WM, respectively, corresponding to a $6.97\times$ planning speedup, while maintaining competitive task performance. Further evaluations across five diverse visual planning tasks demonstrate the general applicability of Scope-WM. Code is available at https://github.com/ChunZheng2022/Scope-WM.
Chinese Translation
视觉世界模型通过预测未来观察来支持机器人规划，但密集的潜在状态传播和样本密集的轨迹优化导致高推理延迟和峰值内存使用，限制了在资源受限平台上的实时部署。现有的稀疏世界模型加速方法要么依赖于无指导的标记稀疏化，这可能会丢弃与规划相关的信息并限制可实现的稀疏性，要么引入重型辅助模块和繁琐的多阶段训练流程。在本研究中，我们提出了Scope-WM，一种高效的视觉世界模型，它将计算范围限定在与预测相关的潜在区域和有前景的动作序列上。Scope-WM将预测相关性提炼为轻量级的动作条件选择器，仅对选定标记的紧凑子集应用完整的动态预测。它使用前景状态及其变化的紧凑摘要来更新其余标记，使背景能够感知前景动态，而无需昂贵的标记间交互。在规划过程中，Scope-WM保留并重用在初始模型预测控制（MPC）搜索中发现的高质量动作序列，集中后续搜索在减少的展开预算下。最终的流程仅需一次选择器提炼，随后进行一次联合训练阶段以训练稀疏世界模型。在具有挑战性的Push-T任务中，Scope-WM将峰值GPU内存使用和规划时间分别减少到密集DINO-WM的$18.1\%$和$14.3\\%$，对应于$6.97 imes$的规划加速，同时保持竞争性的任务性能。对五个不同视觉规划任务的进一步评估展示了Scope-WM的广泛适用性。代码可在 https://github.com/ChunZheng2022/Scope-WM 获取。
cs.CV / 189 / 2609.33230
AevaScenes: An FMCW LiDAR Dataset and Benchmark for Long-Range Perception
AevaScenes：一种用于长距离感知的FMCW LiDAR数据集和基准测试
Abstract
FMCW LiDAR measures per-point radial Doppler velocity alongside range, providing a motion cue unavailable in conventional time-of-flight sensors. Exploiting this signal at long range remains understudied. We present an FMCW LiDAR dataset of 575 sequences (57.5K frames) with over 8 million annotated 3D boxes across 16 detection classes and per-point labels across 24 semantic classes, captured by six commercial FMCW LiDAR sensors and six paired 4K cameras across eight Bay Area cities, including 237 nighttime sequences, with annotations extending to 400m. We define a benchmark with three tasks: 3D object detection, scene flow estimation, and semantic segmentation. Detection and scene flow are evaluated across three range bins to 400m, with a public evaluation server. We explore the impact of Doppler measurements on flagship recognition tasks, and find significant improvements up to 2X in detection AP of far-away vehicles and pedestrians, particularly in low-latency single-frame settings. We similarly find scene flow accuracy is significantly improved with Doppler measurements across all ranges. Our dataset and benchmark have been publicly released at https://scenes.aeva.com.
Chinese Translation
FMCW LiDAR在测量每个点的径向多普勒速度的同时提供了距离信息，这为传统的飞行时间传感器所无法提供的运动线索。利用这一信号进行长距离感知的研究仍然较少。我们呈现了一个FMCW LiDAR数据集，包含575个序列（57.5K帧），在16个检测类别中标注了超过800万个3D框，并在24个语义类别中提供了每点标签。该数据集由六个商业FMCW LiDAR传感器和六个配对的4K摄像头在八个旧金山湾区城市捕获，包括237个夜间序列，标注范围延伸至400米。我们定义了一个基准测试，包含三个任务：3D物体检测、场景流估计和语义分割。在三个范围区间内（最高至400米）评估检测和场景流，并设立了公共评估服务器。我们探讨了多普勒测量对主要识别任务的影响，发现远处车辆和行人的检测平均精度（AP）显著提高，最高可达2倍，尤其是在低延迟的单帧设置中。我们同样发现，所有范围内的场景流准确性在有多普勒测量的情况下显著提高。我们的数据集和基准测试已在 https://scenes.aeva.com 上公开发布。
cs.CV / 190 / 2609.33236
PARSEE-VAD: Efficient Training-Free Online Video Anomaly Detection via Proposition-Aware Reasoning and Streaming Evidence Escalation
PARSEE-VAD：通过命题感知推理和流式证据升级实现高效的无训练在线视频异常检测
Abstract
Training-free online video anomaly detection (VAD) with frozen multimodal language models faces two coupled challenges: extracting reliable current-window semantics under causal and computational constraints, and maintaining temporal continuity without repeatedly transmitting high-dimensional history. Encoding history through text can compress visual evidence and introduce semantic bias, whereas retaining visual history expands multimodal context. We introduce PARSEE-VAD, a two-module framework that separates semantic evidence acquisition from score-state evolution. Proposition-Aware Reasoning (PAR) extracts structured propositional evidence from the current causal window and conditionally activates more specific queries when coarse evidence warrants further refinement. By sharing a reusable causal visual prefix across queries, PAR reduces redundant computation through selective execution. Streaming Evidence Escalation (SEE) maps the acquired proposition evidence into a compact score-domain event state through current evidence escalation, then propagates only the resulting bounded state across decisions to support temporal continuity. Experiments on four benchmarks demonstrate strong training-free online performance while selective routing reduces specialist computation and score-state propagation remains sparse. These results support a current-first principle for streaming multimodal inference: resolve present semantics first, then use compact historical state only to repair residual continuity gaps.
Chinese Translation
无训练的在线视频异常检测（VAD）在冻结的多模态语言模型下面临两个相互关联的挑战：在因果和计算约束下提取可靠的当前窗口语义，以及在不重复传输高维历史的情况下保持时间连续性。通过文本编码历史可以压缩视觉证据并引入语义偏差，而保留视觉历史则扩展了多模态上下文。我们提出了PARSEE-VAD，一个将语义证据获取与评分状态演变分开的双模块框架。命题感知推理（PAR）从当前因果窗口中提取结构化的命题证据，并在粗略证据需要进一步细化时有条件地激活更具体的查询。通过在查询之间共享可重用的因果视觉前缀，PAR通过选择性执行减少冗余计算。流式证据升级（SEE）通过当前证据升级将获取的命题证据映射到紧凑的评分域事件状态，然后仅在决策之间传播结果的有界状态，以支持时间连续性。在四个基准上的实验表明，强大的无训练在线性能，同时选择性路由减少了专业计算，评分状态传播保持稀疏。这些结果支持流式多模态推理的当前优先原则：首先解决当前语义，然后仅使用紧凑的历史状态来修复剩余的连续性缺口。
cs.CV / 191 / 2609.33253
VGGT-Diff: Visual Geometry Meets Diffusion for Sparse-View Novel View Synthesis
VGGT-Diff：视觉几何与扩散相结合的稀疏视图新视图合成
Abstract
We present VGGT-Diff, a geometry-routed multi-view diffusion model for sparse-view novel view synthesis. Existing novel view synthesis (NVS) methods face a fundamental trade-off: reconstruction-based approaches preserve observed geometry but struggle to synthesize unseen regions, while diffusion-based methods provide strong generative priors yet rely on implicit source-to-query correspondence. VGGT-Diff bridges these regimes by routing visual geometry latents from VGGT-{\Omega} into a pretrained video diffusion model. Each visual token is associated with a 3D point and confidence, then transformed into query-aligned latent conditions through a confidence-aware Visual Geometry Router (VGR) that preserves front and back surface evidence. These conditions guide joint target-view denoising, while Point-Track Residual Consistency (PTRC) regularizes predicted-clean residuals along reliable 3D tracks, improving multi-view stability. We further introduce robust geometry conditioning, combining training-time regularization with inference-time guidance for improved robustness. Experiments show competitive or state-of-the-art performance across interpolation and extrapolation under different viewpoint difficulties. Our code is available at https://github.com/chenkangjie1123/VGGT-Diff.
Chinese Translation
我们提出了VGGT-Diff，一种基于几何引导的多视图扩散模型，用于稀疏视图的新视图合成。现有的新视图合成（NVS）方法面临一个基本的权衡：基于重建的方法能够保留观察到的几何信息，但在合成未见区域时表现不佳，而基于扩散的方法提供了强大的生成先验，但依赖于隐式的源到查询对应关系。VGGT-Diff通过将来自VGGT-{ ext{Ω}}的视觉几何潜变量引导到预训练的视频扩散模型中，弥合了这两种方法。每个视觉标记都与一个三维点和置信度相关联，然后通过一个关注置信度的视觉几何路由器（VGR）转换为与查询对齐的潜在条件，该路由器保留了前后表面的证据。这些条件指导联合目标视图去噪，而点轨迹残差一致性（PTRC）则沿着可靠的三维轨迹对预测的干净残差进行正则化，从而提高多视图的稳定性。我们进一步引入了稳健的几何条件，结合训练时的正则化与推理时的指导，以提高鲁棒性。实验结果表明，在不同视角难度下，我们的方法在插值和外推方面表现出竞争力或达到最先进的性能。我们的代码可在 https://github.com/chenkangjie1123/VGGT-Diff 获取。
cs.CV / 192 / 2609.33261
EngIntervene: Benchmarking Multimodal Engineering State Understanding and Design Intervention Reasoning
EngIntervene：多模态工程状态理解与设计干预推理的基准评估
Abstract
Multimodal engineering benchmarks largely evaluate static understanding, such as recognizing components, interpreting diagrams, or answering technical questions. This leaves a missing middle between engineering perception and full design generation: whether a model can use an understood system state to reason about relations, constraints, and the consequences of design changes. We introduce \textsc{EngIntervene}, a benchmark for this capability. It contains 3,229 questions across seven engineering domains and organizes evaluation into four levels: state grounding (T1), relational and mechanistic reasoning (T2), constraint-aware diagnosis (T3), and intervention reasoning (T4), which asks whether a proposed modification achieves its target while preserving required constraints. The tasks instantiate a unified engineering state representation spanning objects, relations, constraints, and design objectives, and T2--T4 are scored against structured reference answers with atomic criteria. Across open- and closed-weight multimodal models, stronger grounding does not reliably translate into better diagnosis or intervention, and the best open-weight model trails the best closed model by 14.7 percentage points on the T2--T4 average. Removing or shuffling visual evidence consistently degrades performance, while benchmark-specific supervised fine-tuning improves T1 but not T2--T4. T4 further exposes a large gap between satisfying individual revision criteria and producing a fully valid intervention. Engineering reasoning thus requires not only recovering the current state, but also reliably using it to reason about constraints and post-intervention consequences. Code and benchmark artifacts are available at https://github.com/changcv2021/EngIntervene
Chinese Translation
多模态工程基准主要评估静态理解，例如识别组件、解释图表或回答技术问题。这在工程感知与完整设计生成之间留下了一个缺失的环节：模型是否能够利用理解的系统状态推理关系、约束和设计变更的后果。我们引入了 extsc{EngIntervene}，这是一个针对这一能力的基准。它涵盖了七个工程领域的3,229个问题，并将评估组织为四个层次：状态基础（T1）、关系和机械推理（T2）、约束感知诊断（T3）和干预推理（T4），后者询问所提议的修改是否在保留所需约束的同时达成其目标。这些任务实例化了一个统一的工程状态表示，涵盖对象、关系、约束和设计目标，而T2至T4的评分则基于结构化参考答案和原子标准。在开放权重和闭合权重的多模态模型中，较强的基础并不可靠地转化为更好的诊断或干预，且最佳开放权重模型在T2至T4平均分上落后于最佳闭合模型14.7个百分点。移除或打乱视觉证据会持续降低性能，而基准特定的监督微调提高了T1的表现，但对T2至T4没有改善。T4进一步揭示了满足单个修订标准与产生完全有效干预之间的巨大差距。因此，工程推理不仅需要恢复当前状态，还需要可靠地利用该状态推理约束和干预后的后果。代码和基准文档可在 https://github.com/changcv2021/EngIntervene 获取。
cs.CV / 193 / 2609.33264
VehDyn: A Driving World Model Benchmark for Vehicle Dynamics
VehDyn：一种用于车辆动力学的驾驶世界模型基准
Abstract
Video world models are emerging as data engines, action planners, and generative simulators for autonomous driving, but existing benchmarks primarily assess visual fidelity and coarse physical plausibility, providing limited evidence on whether generated driving futures obey realistic vehicle kinematics and dynamics. This limitation is further compounded by the lack of datasets in which vehicle, road, maneuver, and speed conditions are independently controlled, and ground-truth vehicle states are recorded in synchrony with videos. We introduce VehDyn, a driving world model benchmark for vehicle dynamics. VehDyn is built on a CARLA-CarSim co-simulation platform where photorealistic rendering is coupled with a validated multi-body dynamics model, and it contains 10,080 configurations from a full factorial design over five vehicle types, four tire-road friction coefficients, three maneuvers, four target speeds, 14 scenes, and three illuminations, each paired with synchronized position, velocity, and attitude sequences. Built on this dataset, VehDyn introduces a hierarchical evaluation framework that measures trajectory alignment, kinematic consistency, and dynamic consistency, and benchmarks 12 state-of-the-art video world models. We further assess the video quality using two established protocols and correlate it with the VehDyn score. Trajectory-level metrics are nearly saturated, with ten of twelve models within 20\% of ground truth, while no model reaches 92\% of ground truth on dynamic consistency, and visual-quality metrics are only weakly correlated with vehicle-dynamics fidelity. DrivingWorld achieves the highest VehDyn score, followed by Cosmos 3 Nano and LTX-Video 2.5, and the VehDyn score agrees closely with human judgment. VehDyn provides a systematic foundation for developing driving world models that are physically consistent and visually realistic.
Chinese Translation
视频世界模型正在成为自动驾驶的数据引擎、行动规划者和生成模拟器，但现有基准主要评估视觉逼真度和粗略的物理合理性，提供的证据有限，无法判断生成的驾驶未来是否遵循现实的车辆运动学和动力学。这一局限性进一步加剧，原因在于缺乏能够独立控制车辆、道路、机动和速度条件的数据集，并且地面真实车辆状态与视频同步记录。我们介绍了VehDyn，一种用于车辆动力学的驾驶世界模型基准。VehDyn建立在CARLA-CarSim协同仿真平台上，其中光线逼真的渲染与经过验证的多体动力学模型相结合，包含来自五种车辆类型、四种轮胎-道路摩擦系数、三种机动、四种目标速度、14个场景和三种光照条件的10,080种配置，每种配置都配有同步的位置、速度和姿态序列。在此数据集的基础上，VehDyn引入了一个分层评估框架，用于测量轨迹对齐、运动学一致性和动力学一致性，并对12种最先进的视频世界模型进行基准测试。我们进一步使用两种已建立的协议评估视频质量，并将其与VehDyn评分相关联。轨迹级别的指标几乎达到饱和，12种模型中有十种在动态一致性上与地面真实值相差20 ext{%}以内，但没有模型在动态一致性上达到92 ext{%}的地面真实值，而视觉质量指标与车辆动力学的逼真度仅呈弱相关。DrivingWorld获得了最高的VehDyn评分，其次是Cosmos 3 Nano和LTX-Video 2.5，VehDyn评分与人类判断高度一致。VehDyn为开发在物理上一致且视觉上真实的驾驶世界模型提供了系统性的基础。
cs.CV / 194 / 2609.33286
InfoEdit: Probing Global Layout Reasoning in Infographic Editing
InfoEdit：探讨信息图编辑中的全球布局推理
Abstract
Multimodal foundation models edit natural photographs at production quality, yet the same models struggle with structured visual content such as infographics. Unlike photographs, infographics encode information through logical relations; editing one element often requires surrounding elements to be adapted. We refer to this global layout reasoning capability as reflow. Existing image-editing benchmarks neither provide a dedicated setting for structured visual content nor evaluate the reflow capability. We introduce InfoEdit, a novel benchmark of 1,000 infographics across eight logical-relation families, paired with 4,000 editing instructions across four editing tasks, and a reflow-aware evaluation protocol. Across eight frontier editors, only GPT-Image-2 clears 60% average success rate; most models fall below 7%, and no editor exceeds 36% on the Swap-Block task even with perfect target localization. We further show that code-level editing can match the strongest pixel-level editor, revealing complementary strengths across tasks. InfoEdit identifies reflow as a central challenge in structured visual content editing and provides a diagnostic benchmark to facilitate future progress.
Chinese Translation
多模态基础模型能够以生产质量编辑自然照片，但同样的模型在处理结构化视觉内容（如信息图）时却表现不佳。与照片不同，信息图通过逻辑关系编码信息；编辑一个元素通常需要调整周围的元素。我们将这种全球布局推理能力称为重排（reflow）。现有的图像编辑基准既没有为结构化视觉内容提供专门的设置，也没有评估重排能力。我们引入了InfoEdit，这是一个包含1,000个信息图的全新基准，涵盖八种逻辑关系类别，并配有4,000条编辑指令，涉及四个编辑任务，以及一个关注重排的评估协议。在八个前沿编辑器中，只有GPT-Image-2达到了60%的平均成功率；大多数模型的成功率低于7%，即使在Swap-Block任务中，且目标定位完美，也没有编辑器超过36%。我们进一步展示了代码级编辑可以与最强的像素级编辑器相匹配，揭示了不同任务间的互补优势。InfoEdit将重排识别为结构化视觉内容编辑中的核心挑战，并提供了一个诊断基准，以促进未来的进展。
cs.CV / 195 / 2609.33288
Informative Viewpoint Selection for Episodic-Memory Embodied Question Answering using Omnidirectional Images
基于全景图像的情节记忆具身问答的信息视角选择
Abstract
Embodied Question Answering (EQA) requires agents to answer natural language questions about surrounding environments from visual observations. In this work, we focus on open-vocabulary episodic-memory EQA (EM-EQA), where an agent answers free-form questions using recorded observation histories. Omnidirectional images are promising for this task, as they provide wide field-of-view observations that can capture surrounding context without requiring explicit camera rotations. However, omnidirectional images introduce two challenges for EQA: (i) equirectangular projection causes severe geometric distortion that degrades vision-language model (VLM) recognition accuracy, and (ii) feeding equirectangular images directly into VLMs introduces excessive irrelevant background information, reducing answer accuracy and increasing the visual-token burden. To address these challenges, we propose a viewpoint selection method for EM-EQA using omnidirectional images. Our method converts equirectangular observations into perspective views via cubemap projection, estimates question-conditioned relevance with fine-tuned BLIP-2, and selects informative and diverse viewpoints through diversity-aware greedy selection. Experiments on the Habitat-Matterport 3D (HM3D) subset of OpenEQA show that our method achieves state-of-the-art model performance among the reported model results with equirectangular observations. Moreover, after removing rotation views, which reduces observation frames by 65.5%, our method largely maintains its answer accuracy.
Chinese Translation
具身问答（EQA）要求智能体根据视觉观察回答关于周围环境的自然语言问题。在本研究中，我们专注于开放词汇的情节记忆具身问答（EM-EQA），其中智能体利用记录的观察历史回答自由形式的问题。全景图像在这一任务中具有很大潜力，因为它们提供了广阔的视野观察，可以捕捉周围的上下文，而无需显式的相机旋转。然而，全景图像给EQA带来了两个挑战：（i）等矩形投影导致严重的几何失真，从而降低了视觉-语言模型（VLM）的识别准确性；（ii）将等矩形图像直接输入VLM会引入过多无关的背景信息，降低回答的准确性并增加视觉标记的负担。为了解决这些挑战，我们提出了一种基于全景图像的EM-EQA视角选择方法。我们的方法通过立方体映射投影将等矩形观察转换为透视视图，利用微调后的BLIP-2估计与问题相关的相关性，并通过关注多样性的贪婪选择方法选择信息丰富且多样的视角。在OpenEQA的Habitat-Matterport 3D（HM3D）子集上的实验表明，我们的方法在报告的等矩形观察模型结果中实现了最先进的模型性能。此外，在去除旋转视图后，观察帧减少了65.5%，我们的方法在很大程度上保持了其回答的准确性。
cs.CV / 196 / 2609.33304
Relevance Does Not Imply Applicability: Experience Activation for Personal GUI Agents
相关性并不意味着适用性：个人图形用户界面的经验激活
Abstract
Personal Graphical User Interface (GUI) agents rely on interaction history to infer what a user wants from ambiguous instructions and to anticipate recurring routines. Existing approaches retrieve task-relevant history and append it to the policy's context, implicitly assuming that experience relevant to a task remains useful for each decision within it. We find that this help is largely spent at the first decision: retrieved history strongly improves the opening step of an episode, yet provides little sustained benefit over the remaining 90\% of steps, and offers weak guidance on whether a proactive suggestion is warranted. A relevant record may tell the agent where to begin, but not which past action applies to the current screen or whether a routine is due now. The underlying issue is that relevance does not imply applicability}: relevance is determined at the task level, whereas applicability depends on the situation at decision time. We therefore recast personalization as experience activation and introduce ExpActivator, a training-free framework that activates only the experience applicable to the current situation. During execution, ExpActivator matches each new screen to historical states in the frozen GUI backbone's latent space and supplies the corresponding action as a reference. Before execution, it activates a recurring intent only when the current time and scenario provide sufficient support, and otherwise abstains. Across four GUI backbones, ExpActivator improves within-trajectory step success by 28\% on average, achieves the best personalized execution on every backbone while using about one-fifth as many history tokens, and reaches approximately 2.3$\times$ the Matthews correlation coefficient of the strongest proactive baseline. Experience pays where it is activated, not where it is appended.
Chinese Translation
个人图形用户界面（GUI）代理依赖于交互历史来推断用户从模糊指令中想要什么，并预测重复的例行程序。现有的方法检索与任务相关的历史记录并将其附加到策略的上下文中，隐含地假设与任务相关的经验在每个决策中仍然是有用的。我们发现这种帮助在第一次决策时大多被消耗：检索的历史记录显著改善了一个情节的开头步骤，但在剩余的90%步骤中提供的持续收益很少，并且在是否需要主动建议方面提供的指导也很弱。相关记录可能会告诉代理从哪里开始，但并不能说明哪个过去的行动适用于当前屏幕或是否现在该进行例行操作。根本问题在于，相关性并不意味着适用性：相关性是在任务层面上确定的，而适用性则取决于决策时的情况。因此，我们将个性化重新构建为经验激活，并引入了ExpActivator，这是一个无训练的框架，仅激活适用于当前情况的经验。在执行过程中，ExpActivator将每个新屏幕与冻结的GUI主干的潜在空间中的历史状态进行匹配，并提供相应的动作作为参考。在执行之前，仅在当前时间和场景提供足够支持时激活重复意图，否则则不激活。在四个GUI主干上，ExpActivator平均提高了轨迹内步骤成功率28%，在每个主干上实现了最佳个性化执行，同时使用的历史令牌约为五分之一，并达到了最强主动基线的马修斯相关系数约为2.3倍。经验在被激活的地方发挥作用，而不是在被附加的地方。
cs.CV / 197 / 2609.33306
LoopTrack: A Simple Baseline for Parameter-Efficient Transformer Tracking
LoopTrack：一种简单的参数高效Transformer跟踪基线
Abstract
Current Transformer-based tracking methods typically stack multiple Transformer blocks with separate parameters to model interactions between the target template and the search region for target localization. These trackers often incur substantial parameter overhead from stacked blocks, making their deployment on resource-limited devices difficult. To address this, we propose a parameter-efficient Transformer tracking framework, dubbed LoopTrack, which repeatedly applies a set of Transformer blocks with shared parameters to interact features in a looped architecture for tracking, significantly reducing the number of parameters. To further exploit target cues, we present two lightweight designs, including target-aware looping (TAL) and gated target memory (GTM). The former applies intermediate target information generated by one loop to guide feature interaction in the subsequent loop, enabling progressive feature refinement, while the latter maintains a compact memory across frames, which is incorporated into the loop process to provide long-term information to the tracker, mitigating temporal drift in tracking. Compared to existing Transformer trackers, LoopTrack enables multiple rounds of feature interaction with fewer model parameters, making it resource-friendly for deployment. In extensive experiments on multiple datasets, LoopTrack shows a favorable accuracy-parameter trade-off. In particular, our LoopTrack$_{\rm One}$, with a single shared Transformer block, achieves 66.2\% SUC score on LaSOT with only 3.4M parameters, while LoopTrack$_{\rm Three}$, using three shared blocks, achieves 69.3\% SUC score with 6.4M parameters, surpassing existing parameter-efficient tracking methods with comparable or larger model size. With LoopTrack, we aim to establish a simple yet strong baseline for parameter-efficient Transformer tracking. Our code and models will be released.
Chinese Translation
当前基于Transformer的跟踪方法通常堆叠多个具有独立参数的Transformer模块，以建模目标模板与搜索区域之间的交互，从而实现目标定位。这些跟踪器往往因堆叠模块而产生大量参数开销，使其在资源有限的设备上部署变得困难。为了解决这个问题，我们提出了一种参数高效的Transformer跟踪框架，称为LoopTrack，该框架在循环架构中重复应用一组共享参数的Transformer模块，以交互特征，从而显著减少参数数量。为了进一步利用目标线索，我们提出了两种轻量级设计，包括目标感知循环（Target-Aware Looping, TAL）和门控目标记忆（Gated Target Memory, GTM）。前者利用一个循环生成的中间目标信息来指导后续循环中的特征交互，实现渐进式特征精炼，而后者在帧间保持紧凑的记忆，并将其纳入循环过程中，为跟踪器提供长期信息，从而减轻跟踪中的时间漂移。与现有的Transformer跟踪器相比，LoopTrack以更少的模型参数实现多轮特征交互，使其在部署时更加资源友好。在多个数据集上的广泛实验表明，LoopTrack在准确性与参数之间达成了良好的平衡。特别地，我们的LoopTrack$_{
m One}$使用单个共享Transformer模块，在LaSOT上获得了66.2\%的SUC分数，仅需3.4M参数，而LoopTrack$_{
m Three}$使用三个共享模块，在6.4M参数下达到了69.3\\%的SUC分数，超越了现有的参数高效跟踪方法，且模型规模相当或更大。通过LoopTrack，我们旨在建立一种简单而强大的参数高效Transformer跟踪基线。我们的代码和模型将会发布。
cs.CV / 198 / 2609.33318
PIC-UIE: Predicting Image-Adaptive Corrections for Lightweight Underwater Image Enhancement
PIC-UIE：预测轻量级水下图像增强的图像自适应校正
Abstract
Underwater image enhancement (UIE) aims to restore visibility, color fidelity, and structural detail from images degraded by wavelength-dependent attenuation and backscatter. State-of-the-art UIE methods often rely on large backbones and dense image-to-image prediction, limiting their practicality for edge deployment. Moreover, operating entirely in a single color space couples degradation estimation with luminance and chroma correction. To address these challenges, we propose PIC-UIE, a lightweight predictor--executor framework that predicts image-adaptive corrections from a fixed $256\times256$ RGB thumbnail and applies them to the native-resolution input in the YCbCr color space. The predictor produces seven outputs, organized into spatial correction, nonlinear luminance and coupled chroma mapping, and image-level color calibration. A depth map regularizes the transmission proxy during training, whereas inference uses only the RGB input. With 9,486 parameters and 0.094 GFLOPs at $256\times256$, PIC-UIE achieves 24.137 dB PSNR and 0.9216 SSIM on UIEB-90 and 21.320 dB PSNR on zero-shot LSUI. It further processes native 4K images at 55.0 FPS under the comparison protocol. These results show that structured correction prediction provides an effective and practical alternative to dense RGB reconstruction for underwater image enhancement.
Chinese Translation
水下图像增强（UIE）旨在恢复因波长依赖的衰减和反向散射而退化的图像的可见性、色彩保真度和结构细节。最先进的UIE方法通常依赖于大型骨干网络和密集的图像到图像预测，这限制了它们在边缘部署中的实用性。此外，完全在单一颜色空间中操作将退化估计与亮度和色度校正耦合在一起。为了解决这些挑战，我们提出了PIC-UIE，这是一种轻量级的预测-执行框架，它从固定的$256 imes256$ RGB缩略图中预测图像自适应校正，并将其应用于YCbCr颜色空间中的原始分辨率输入。预测器生成七个输出，组织为空间校正、非线性亮度和耦合色度映射，以及图像级颜色校正。深度图在训练过程中对传输代理进行正则化，而推理仅使用RGB输入。在$256 imes256$的情况下，PIC-UIE具有9486个参数和0.094 GFLOPs，在UIEB-90上达到24.137 dB PSNR和0.9216 SSIM，在零样本LSUI上达到21.320 dB PSNR。它进一步以55.0 FPS的速度处理原生4K图像，符合比较协议。这些结果表明，结构化的校正预测为水下图像增强提供了一种有效且实用的替代方案，优于密集的RGB重建。
cs.CV / 199 / 2609.33325
VisionHOPE: Visual Backbones as Self-Modifying Learning Systems
VisionHOPE：视觉主干作为自我修改学习系统
Abstract
Visual backbones have evolved from Convolutional Neural Networks (CNNs) with local aggregation to Vision Transformers (ViTs) with global interactions, State-Space Models (SSMs) with input-dependent state transitions, and Test-Time Training (TTT) layers that adapt an inner learner while processing an image. Across this progression, visual computation has become increasingly adaptive to each input, yet the rules governing that adaptation remain largely prescribed by the trained backbone. We introduce VisionHOPE, the first generic visual backbone formulated as a self-modifying learning system, in which what the model remembers and how it learns co-evolve within an image. Building on the self-referential construction of Nested Learning (NL), VisionHOPE realizes this co-evolution through five coupled memories that store content, generate key and value representations, and govern learning rate and retention. These memories evolve jointly as visual context accumulates along each scan. However, directly applying the unconstrained self-referential update to a visual backbone leads to instability. We therefore derive a stability-matched step-size control scheme that combines a soft cap on self-referential injection with a spectral clamp on the retained memory transition, and prove that the resulting memory dynamics are non-expansive along each scan. For two-dimensional feature maps, we adapt NL's chunk formulation by aligning chunks with image rows and columns across four directional scans. The proposed VisionHOPE achieves competitive results on ImageNet-1K, COCO, and ADE20K, establishing self-modifying learning systems as a practical foundation for general-purpose visual backbones. The code is available at https://github.com/PSRben/VisionHOPE.
Chinese Translation
视觉主干已经从具有局部聚合的卷积神经网络（CNN）演变为具有全局交互的视觉变换器（ViT）、具有输入依赖状态转移的状态空间模型（SSM）以及在处理图像时自适应内部学习者的测试时训练（TTT）层。在这一演变过程中，视觉计算变得越来越适应每个输入，但支配这种适应的规则在很大程度上仍由训练好的主干所规定。我们提出了VisionHOPE，这是第一个被构建为自我修改学习系统的通用视觉主干，其中模型记忆的内容和学习方式在图像中共同演化。基于嵌套学习（Nested Learning, NL）的自我参照构造，VisionHOPE通过五个耦合记忆实现这种共同演化，这些记忆存储内容，生成键和值表示，并控制学习率和保留。随着每次扫描的视觉上下文的累积，这些记忆共同演化。然而，直接将不受约束的自我参照更新应用于视觉主干会导致不稳定。因此，我们推导出一种稳定匹配的步长控制方案，该方案结合了对自我参照注入的软上限和对保留记忆转移的谱限制，并证明所得到的记忆动态在每次扫描中都是非扩展的。对于二维特征图，我们通过将块与图像的行和列对齐，适应了NL的块构造，涵盖了四个方向的扫描。所提出的VisionHOPE在ImageNet-1K、COCO和ADE20K上取得了具有竞争力的结果，确立了自我修改学习系统作为通用视觉主干的实用基础。代码可在 https://github.com/PSRben/VisionHOPE 获取。
cs.CV / 200 / 2609.33330
FeCoSplat: Feedback-Guided Compression for Feed-Forward 3D Gaussian Splatting
FeCoSplat：用于前馈3D高斯点云的反馈引导压缩
Abstract
Feed-forward 3D Gaussian Splatting (3DGS) enables efficient novel-view synthesis from sparse multi-view images, yet its representations remain costly to store and transmit. Existing approaches compress either the input images, incurring heavy receiver-side reconstruction, or the reconstructed Gaussian primitives, which are difficult to compress due to their heterogeneous and irregular attributes. We instead compress compact intermediate features, providing a better balance between compression efficiency and receiver-side complexity. Based on this paradigm, we propose FeCoSplat, a feedback-guided compression framework for feed-forward 3DGS. FeCoSplat first compresses multi-view features to obtain an intermediate 3DGS, whose rendered views are used as feedback to guide a second-stage compression for further refinement. The resulting bitstreams are decoded into a compact implicit state, from which the final Gaussian primitives are reconstructed with a lightweight predictor. Experiments demonstrate that FeCoSplat achieves favorable rate--distortion performance, particularly at low bitrates, while requiring only 3.45M parameters for receiver-side Gaussian reconstruction. Code will be released soon.
Chinese Translation
前馈3D高斯点云（3DGS）能够从稀疏的多视图图像中高效合成新视角，但其表示仍然在存储和传输上成本高昂。现有方法要么压缩输入图像，导致接收端重建负担沉重，要么压缩重建的高斯原语，由于其异质和不规则特性，难以压缩。我们则选择压缩紧凑的中间特征，从而在压缩效率和接收端复杂性之间提供更好的平衡。基于这一范式，我们提出了FeCoSplat，一种用于前馈3DGS的反馈引导压缩框架。FeCoSplat首先压缩多视图特征以获得中间3DGS，其渲染视图被用作反馈，以指导第二阶段的压缩以进一步优化。最终生成的比特流被解码为紧凑的隐式状态，从中使用轻量级预测器重建最终的高斯原语。实验表明，FeCoSplat在低比特率下实现了良好的率-失真性能，同时仅需3.45M参数用于接收端的高斯重建。代码将很快发布。
cs.CV / 201 / 2609.33338
OPERA: A Unified Omnimodal Progressive Spatio-Temporal Reasoning Agent for Referring Video Segmentation
OPERA：一种统一的全模态渐进时空推理代理用于指称视频分割
Abstract
Referring video segmentation with heterogeneous multimodal queries---spanning text, audio, and reference images---demands both robust cross-modal understanding and precise spatio-temporal reasoning. We propose OPERA (Omnimodal Progressive spatio-tEmporal Reasoning Agent), a unified reasoning agent built on a single MLLM that performs dual-axis progressive reasoning via three specialized stages. Along the temporal axis, a Temporal Reasoning Agent narrows the frame search space through coarse-to-fine filtering to identify the most informative key frame. Along the spatial axis, a Distillation Agent establishes what to locate via cross-modal semantic distillation, and a Grounding Agent enhanced with GRPO determines where the target appears, with dense mask propagation completing the pixel-level output. OPERA sets a new state of the art on OmniAVS and Ref-AVS and transfers zero-shot to standard referring video segmentation benchmarks.
Chinese Translation
指称视频分割涉及异构多模态查询——包括文本、音频和参考图像——需要强大的跨模态理解能力和精确的时空推理能力。我们提出了OPERA（全模态渐进时空推理代理），这是一个基于单一多语言大模型（MLLM）构建的统一推理代理，通过三个专业阶段进行双轴渐进推理。在时间轴上，时间推理代理通过粗到细的过滤缩小帧搜索空间，以识别最具信息量的关键帧。在空间轴上，蒸馏代理通过跨模态语义蒸馏确定需要定位的内容，而增强了GRPO的定位代理则确定目标出现的位置，密集的掩膜传播完成了像素级输出。OPERA在OmniAVS和Ref-AVS上设定了新的最先进水平，并在标准指称视频分割基准上实现零样本迁移。
cs.CV / 202 / 2609.33344
ReLoc: Rethinking Scene Coordinate Regression Architecture for Robust Outdoor LiDAR-based Localization
ReLoc：重新思考场景坐标回归架构以实现稳健的户外基于激光雷达的定位
Abstract
Scene Coordinate Regression (SCR) has recently emerged as a promising approach for LiDAR-based localization, achieving accurate localization without requiring an explicit 3D map. Despite their effectiveness, existing SCR methods rely on scene classification-based global embedding that struggles to provide fine-grained discrimination among nearby locations. Moreover, their reliance on uniform sampling of local features during training assigns equal importance to all points, thereby inadvertently propagating features from dynamic objects or unstable regions and potentially degrading training stability. In this paper, we present ReLoc, a revamped SCR architecture that can effectively address these limitations. First, we redesign the global embedding module by combining learnable context tokens with a feature aggregator to capture richer and more discriminative scene context. Second, we introduce an attention-based local feature enhancement module to mitigate the impact of noisy local features while encouraging context-consistent structures, yielding more robust local feature representations. Experimental results on two large-scale outdoor datasets demonstrate that our approach achieves state-of-the-art accuracy over previous SCR-based methods while maintaining real-time inference performance.
Chinese Translation
场景坐标回归（SCR）最近作为一种有前景的激光雷达（LiDAR）定位方法出现，能够在不需要显式三维地图的情况下实现准确定位。尽管现有的SCR方法有效，但它们依赖于基于场景分类的全局嵌入，难以在邻近位置之间提供细粒度的区分。此外，它们在训练过程中对局部特征的均匀采样使所有点的重要性相等，从而无意中传播动态物体或不稳定区域的特征，可能导致训练稳定性下降。在本文中，我们提出了ReLoc，一种改进的SCR架构，能够有效解决这些局限性。首先，我们通过将可学习的上下文标记与特征聚合器相结合，重新设计了全局嵌入模块，以捕捉更丰富和更具区分性的场景上下文。其次，我们引入了一种基于注意力的局部特征增强模块，以减轻噪声局部特征的影响，同时鼓励上下文一致的结构，从而产生更稳健的局部特征表示。在两个大规模户外数据集上的实验结果表明，我们的方法在保持实时推理性能的同时，达到了比以往SCR方法更先进的准确性。
cs.CV / 203 / 2609.33353
Focus and Supplement: Dual-Enhanced Vision Transformer for Multi-Class Anomaly Classification
聚焦与补充：用于多类别异常分类的双重增强视觉变换器
Abstract
Multi-class anomaly classification in industrial vision remains challenging due to noisy/incomplete anomaly representations and the unknown number of anomaly classes. To overcome this, we propose MACO, a novel multi-class anomaly classification framework that learns comprehensive representations and dynamically estimates class number without prior knowledge. First, a soft-focus attention uses anomaly maps to concentrate on relevant abnormal regions, while suppressing background noise. Second, auxiliary classification ([A-CLS]) tokens complement the [CLS] token. They collectively attend to diverse anomaly sub-regions, yielding more holistic and discriminative features. These [A-CLS] tokens are also effective across more tasks and domains. To infer the class number, we propose Correlation-based Number Estimation strategy. It computes the average correlation among labeled classes and transfers its separability cue to the unlabeled set. Experiments on MVTec AD and MTD datasets demonstrate our superiority. Under known class number, MACO improves ARI by 6.5% and $\textbf{16.3%}$ on both datasets, respectively. In the more challenging unknown number scenario, it achieves an $\textbf{11.2%}$ NMI gain on MTD and outperforms existing number estimation strategies by $\textbf{24.1%}$ UPS on MVTec AD. Code will be released at https://github.com/HUST-SLOW/MACO.
Chinese Translation
工业视觉中的多类别异常分类仍然面临挑战，主要由于异常表示的噪声/不完整性以及异常类别数量的不确定性。为了解决这一问题，我们提出了MACO，一个新颖的多类别异常分类框架，它能够学习全面的表示并在没有先验知识的情况下动态估计类别数量。首先，软聚焦注意力机制利用异常图集中关注相关的异常区域，同时抑制背景噪声。其次，辅助分类（[A-CLS]）标记补充了[CLS]标记。它们共同关注多样的异常子区域，从而产生更全面和具有区分性的特征。这些[A-CLS]标记在更多任务和领域中也表现出色。为了推断类别数量，我们提出了基于相关性的数量估计策略。该策略计算标记类别之间的平均相关性，并将其可分离性线索转移到未标记集上。在MVTec AD和MTD数据集上的实验结果证明了我们的优越性。在已知类别数量的情况下，MACO在这两个数据集上分别提高了ARI 6.5%和16.3%。在更具挑战性的未知数量场景中，它在MTD上实现了11.2%的NMI增益，并在MVTec AD上超越现有的数量估计策略，提升了24.1%的UPS。代码将发布在https://github.com/HUST-SLOW/MACO。
cs.CV / 204 / 2609.33359
When Does Geometric View Synthesis Help Wine Label Retrieval? A Public One-Shot Benchmark Across Self-Supervised and Vision-Language Backbones
几何视图合成何时有助于葡萄酒标签检索？基于自监督和视觉-语言骨干网的公共单次基准测试
Abstract
Geometric view synthesis can expand a single wine-label photograph into a training set, but its value with pretrained image encoders is unclear. We study this on a public WineSensed-derived benchmark of 1,000 classes, one enrollment photograph per class, and 4,295 real queries. With the earlier DINO vision transformer (ViT-S/16) recipe, geometric views raise top-1 accuracy from 34.1% to 62.6-63.7%, about three times the gain from two-dimensional (2D) augmentation. Frozen SigLIP 2-B already reaches 94.7%. A linear head over its frozen features gains 1.2-1.3 percentage points with the two geometric pipelines localized by the Segment Anything Model (SAM), while the other pipelines gain an inconclusive 0.3-0.6 points. Low-rank adaptation (LoRA) and validation-selected full fine-tuning show no clear gain within the reported confidence intervals; fixed-budget full fine-tuning loses 9-24 points. SAM localization supplies all six views for 99% of sources, compared with 43% for the edge-based front end. Recognition differences between the two cylinder constructions depend on the training recipe and are confounded by their crop and canvas conventions. Rendered-cylinder tests show different responses to source tilt, but an uncalibrated rim-ratio proxy establishes no corresponding trend in recognition on real photographs. An author-confirmed audit of 50 residual errors identifies 21 query-enrollment appearance mismatches, without establishing an irreducible error rate. These results support geometric synthesis for the tested self-supervised recipe and a smaller benefit through frozen-feature adaptation of the text-supervised encoder.
Chinese Translation
几何视图合成可以将单张葡萄酒标签照片扩展为训练集，但其在预训练图像编码器中的价值尚不明确。我们在一个公共的基于WineSensed的基准测试上进行研究，该基准测试包含1,000个类别，每个类别一张入库照片，以及4,295个真实查询。在早期的DINO视觉变换器（ViT-S/16）方案中，几何视图将top-1准确率从34.1%提升至62.6-63.7%，约为二维（2D）增强的三倍增益。冻结的SigLIP 2-B已经达到了94.7%。在其冻结特征上使用线性头与通过Segment Anything Model（SAM）本地化的两个几何管道相比，获得了1.2-1.3个百分点的提升，而其他管道的增益则不确定，仅为0.3-0.6点。低秩适应（LoRA）和验证选择的完全微调在报告的置信区间内没有明显增益；固定预算的完全微调损失了9-24点。与基于边缘的前端相比，SAM本地化为99%的源提供了六个视图，而后者仅为43%。两种圆柱构造之间的识别差异依赖于训练方案，并受到其裁剪和画布规范的混淆。渲染圆柱测试对源倾斜的响应不同，但未校准的边缘比率代理未能在真实照片的识别中建立相应的趋势。对50个残余错误的作者确认审计识别出21个查询-入库外观不匹配，但未能确定不可减少的错误率。这些结果支持在测试的自监督方案中使用几何合成，并通过冻结特征适应文本监督编码器获得较小的益处。
cs.CV / 205 / 2609.33384
PulseQuant: Propagation-Guided Subspace Correction for 4-Bit Video Diffusion Transformers
PulseQuant：基于传播引导的4位视频扩散变换器子空间修正
Abstract
Quantization errors in video diffusion transformers can be amplified or attenuated by subsequent denoising updates, making local reconstruction error an incomplete predictor of final impact. We introduce PulseQuant, a 4-bit post-training quantization method that combines trajectory sensitivity with activation geometry to guide offline calibration. Isolated block--step interventions estimate propagation risk, which prioritizes sensitive trajectory states during row-radius selection. With these radii fixed, response-subspace correction uses neighboring-code edits to reduce residual components along dominant activation directions. Both stages preserve the original 4-bit weight representation. Controlled interventions show that short-horizon propagated error predicts final latent error more reliably than immediate block-output error, supporting calibration beyond local reconstruction objectives. Evaluations on Wan models, Self Forcing, and MiniMax-H3 demonstrate improvements in key consistency and dense-reference metrics while remaining competitive on other attributes across model scales and generation paradigms.
Chinese Translation
视频扩散变换器中的量化误差可以通过后续的去噪更新被放大或减弱，使得局部重建误差无法完全预测最终影响。我们提出了PulseQuant，这是一种结合轨迹敏感性与激活几何的4位后训练量化方法，以指导离线校准。孤立的块步干预估计传播风险，在行半径选择过程中优先考虑敏感的轨迹状态。在这些半径固定的情况下，响应子空间修正利用邻近代码编辑来减少沿主导激活方向的残差分量。这两个阶段均保持原始的4位权重表示。受控干预显示，短期传播误差比即时块输出误差更可靠地预测最终潜在误差，支持超越局部重建目标的校准。在Wan模型、自我强迫和MiniMax-H3上的评估显示，在关键一致性和密集参考指标上有所改善，同时在模型规模和生成范式的其他属性上保持竞争力。
cs.CV / 206 / 2609.33399
SciGen-Verifier: A Multimodal Reasoner for Explainable Verification in Scientific Image Generation
SciGen-Verifier：一种用于科学图像生成可解释验证的多模态推理器
Abstract
In realistic education, a solution is often expressed not only in words but in a drawing--a circuit, a geometric construction, a function plot--and a teacher must grade the drawing as carefully as the text. Recent advances in unified multimodal models have enabled scientific image generation, yet verifying the correctness of these specialized visual outputs remains a critical bottleneck: errors often arise from intricate domain knowledge, structural reasoning, and multi-step instruction rather than surface-level artifacts. Existing verifiers mainly target natural images and compress judgement into scalar scores, leaving scientific coverage and explainable feedback for error correction underexplored. To bridge this gap, we make three main contributions. (1) We construct SciGen-Verify, a benchmark dedicated to explainable verification of scientific image generation, spanning instruction following, multidisciplinary reasoning, and world knowledge domains. It contains a three-tier hierarchical protocol over the binary judgement, supporting explanation, and corrective editing instruction. (2) We develop SciGen-Verifier, a reasoning-driven multimodal verifier trained via cold-start supervised fine-tuning followed by a curriculum-based two-stage reinforcement learning pipeline. The rubric-guided process rewards first strengthen scientific reasoning exploration and outcome rewards subsequently align output with ground-truth annotation. (3) On SciGen-Verify, SciGen-Verifier achieves competitive performance against much larger proprietary models. It further serves as a practical online critic for iterative image rectification.
Chinese Translation
在现实教育中，解决方案往往不仅以文字表达，还以图形呈现——电路图、几何构造、函数图——教师必须像评估文本一样仔细评估这些图形。近期统一多模态模型的进展使得科学图像生成成为可能，但验证这些专业视觉输出的正确性仍然是一个关键瓶颈：错误通常源于复杂的领域知识、结构性推理和多步骤指令，而非表面级别的伪影。现有的验证器主要针对自然图像，并将判断压缩为标量分数，导致科学覆盖和可解释反馈在错误修正方面尚未得到充分探索。为了解决这一问题，我们做出了三项主要贡献。(1) 我们构建了SciGen-Verify，这是一个专门用于科学图像生成可解释验证的基准，涵盖了指令遵循、多学科推理和世界知识领域。它包含一个关于二元判断的三层层次协议，支持解释和纠正编辑指令。(2) 我们开发了SciGen-Verifier，这是一种基于推理的多模态验证器，通过冷启动监督微调训练，随后采用基于课程的两阶段强化学习流程。该评分指导过程首先奖励科学推理探索，随后将结果奖励与真实标注对齐。(3) 在SciGen-Verify上，SciGen-Verifier在性能上与更大规模的专有模型相媲美，并进一步作为迭代图像修正的实用在线评论者。
cs.CV / 207 / 2609.33400
Groupwise Selective State-Space Filtering for Accurate and Streaming Action Boundary Detection
基于组选择的状态空间滤波的准确流式动作边界检测
Abstract
Action boundary detection partitions untrimmed video into intervals without assigning action classes. We present a boundary-detection adapter operating on pre-extracted video features, learning temporal representations via groupwise selective scans. Learned group fusion and temporal modeling convert these into transition scores, which are decoded into boundary timestamps. Trained with boundary-time supervision, the class-agnostic model is evaluated on Breakfast, GTEA, and 50Salads using temporal tolerances and bipartite matching, achieving boundary $F_1$ scores of 0.457, 0.622, and 0.611. A stateful variant enables feature-streaming inference with zero neural look-ahead, one-sample peak confirmation, and bounded memory. Downstream systems can subsequently assign s
Chinese Translation
动作边界检测将未修剪的视频划分为不分配动作类别的时间间隔。我们提出了一种在预提取视频特征上操作的边界检测适配器，通过组选择扫描学习时间表示。学习到的组融合和时间建模将这些表示转换为过渡得分，并解码为边界时间戳。在边界时间监督下训练的无类别模型在Breakfast、GTEA和50Salads数据集上进行评估，使用时间容忍度和二分匹配，获得了边界 $F_1$ 分数分别为0.457、0.622和0.611。状态保持变体实现了零神经前瞻、单样本峰值确认和有限内存的特征流推理。下游系统随后可以分配相应的动作类别。
cs.CV / 208 / 2609.33402
VaME: Exploring Variational Latent Reasoning for Multimodal Embeddings
VaME：探索变分潜在推理用于多模态嵌入
Abstract
Universal multimodal retrieval requires compact embeddings that preserve task-relevant semantic information across diverse modalities. Prior works have incorporated latent reasoning into multimodal embedding learning to refine this information before embedding extraction. However, most existing approaches remain confined to deterministic latent paths, without exploring alternative trajectories to discover better embeddings. Thus, we propose VaME (Variational Multimodal Embeddings), a framework that models latent reasoning as a learnable distribution over trajectories. Specifically, we first introduce Variational Latent Reasoning (VLR) to enable autoregressive exploration in latent space, guided by answer reconstruction through a lightweight decoder. Meanwhile, we augment the original embedding-token readout with a latent-fused embedding to facilitate exploration during subsequent reinforcement learning. Finally, we optimize latent reasoning over stochastic variational trajectories through reinforcement learning, using Semantic Decoding Reward (SDR) to favor semantically meaningful trajectories with interpretable decoded outcomes. On the 78-task MMEB-V2 benchmark, spanning image, video, and visual-document retrieval, VaME outperforms most explicit CoT-based models and all latent-reasoning baselines. VaME also demonstrates robust performance on reasoning-intensive benchmarks such as MRMR, with substantial gains after reinforcement learning. Importantly, VaME achieves these gains with at least a 4.25x inference speedup over the deterministic latent autoregressive baselines. The code will be made publicly available.
Chinese Translation
通用多模态检索需要紧凑的嵌入，以在不同模态中保留与任务相关的语义信息。先前的研究将潜在推理纳入多模态嵌入学习，以在嵌入提取之前精炼这些信息。然而，大多数现有方法仍然局限于确定性的潜在路径，没有探索替代轨迹以发现更好的嵌入。因此，我们提出了VaME（变分多模态嵌入），这是一个将潜在推理建模为可学习的轨迹分布的框架。具体而言，我们首先引入变分潜在推理（VLR），以实现潜在空间中的自回归探索，通过轻量级解码器引导答案重构。同时，我们通过潜在融合嵌入增强原始嵌入标记的读取，以促进后续强化学习中的探索。最后，我们通过强化学习在随机变分轨迹上优化潜在推理，使用语义解码奖励（SDR）来偏向具有可解释解码结果的语义上有意义的轨迹。在涵盖图像、视频和视觉文档检索的78个任务MMEB-V2基准上，VaME的表现超过了大多数显式基于CoT的模型和所有潜在推理基线。VaME在推理密集型基准MRMR上也表现出强大的性能，在强化学习后取得了显著的提升。重要的是，VaME在推理速度上比确定性潜在自回归基线快至少4.25倍。代码将公开发布。
cs.CV / 209 / 2609.33412
Resolving State-Representation Mismatch: State-Space Visual Reasoning for Open-Loop VLA Planning
解决状态表示不匹配问题：用于开环VLA规划的状态空间视觉推理
Abstract
Despite rapid progress in vision-language-action (VLA) models, existing reasoning paradigms still face a fundamental \emph{state-representation mismatch} in open-loop planning. Given only an initial observation, models must internally simulate action-conditioned state transitions, whereas text-, pixel-, and latent-space reasoning can suffer from lossy spatial compression, error-accumulating visual generation, and bypass of intermediate latent tokens, respectively, undermining reliable long-horizon planning. We propose \textbf{State-Space Visual Reasoning} (SSVR), which decouples static visual context, language constraints, and a recurrent latent state. SSVR encodes the initial image and instruction once, then conditions each action prediction on the latent state and updates it with an action-conditioned GRU. Using Qwen2.5-VL as the backbone, SSVR achieves 99.5/99.6, 96.3/98.0, and 83.9/90.6 EM/PR on FrozenLake, Maze, and MiniBehavior, substantially outperforming prior methods. Extensive experiments support the effectiveness of recurrent state modeling for VLA open-loop planning across input transformations and transfer settings. By reusing static visual-textual context and updating a compact recurrent state, SSVR supports efficient multi-step inference, achieving up to $98.58\times$ faster Maze decoding rollouts than the evaluated baselines with the prefix cache prebuilt.
Chinese Translation
尽管视觉-语言-动作（VLA）模型取得了快速进展，现有的推理范式在开环规划中仍面临根本性的“状态表示不匹配”问题。仅凭初始观测，模型必须在内部模拟基于动作的状态转移，而基于文本、像素和潜在空间的推理分别可能遭受空间压缩损失、误差累积的视觉生成以及跳过中间潜在标记的问题，从而削弱了可靠的长时序规划能力。我们提出了“状态空间视觉推理”（State-Space Visual Reasoning，SSVR），该方法将静态视觉上下文、语言约束和递归潜在状态解耦。SSVR对初始图像和指令进行一次编码，然后基于潜在状态条件进行每一步动作预测，并通过基于动作的门控循环单元（GRU）更新该状态。以Qwen2.5-VL为骨干，SSVR在FrozenLake、Maze和MiniBehavior数据集上分别实现了99.5/99.6、96.3/98.0和83.9/90.6的EM/PR指标，显著优于先前方法。大量实验验证了递归状态建模在不同输入变换和迁移设置下对VLA开环规划的有效性。通过重用静态视觉-文本上下文并更新紧凑的递归状态，SSVR支持高效的多步推理，在预构建前缀缓存的条件下，Maze解码推理速度最高提升至基线方法的98.58倍。
cs.CV / 210 / 2609.33414
TTRSD: Test-Time Reinforcement Learning with Self-Distillation for Vision-Language Models
TTRSD：用于视觉语言模型的自蒸馏测试时强化学习
Abstract
Test-time reinforcement learning enables vision-language models (VLMs) to adapt using unlabeled inputs. However, repeated sampling under fixed visual conditions can reinforce shared perceptual errors, while sequence-level rewards fail to isolate visual perception the foundational bottleneck that anchors multimodal reasoning risking the degradation of pre-trained reasoning capabilities. We propose TTRSD, a test-time reinforcement learning framework combining multi-view answer-level self-distillation with visual contrastive token selection. A shared policy aggregates teacher predictions across original, cropped, and downsampled views into an answer distribution. Student trajectories generated from the original image receive rewards based on the support for their final answers in this distribution. To allocate this feedback precisely toward perceptual bottlenecks, we compare the log-probabilities of the same sampled tokens under original and visually ablated inputs while holding their textual prefixes fixed, selecting visually sensitive positions for policy-gradient updates. TTRSD separates update direction, determined by group-relative advantages, from update position, determined by visual sensitivity, without requiring ground-truth labels, external verifiers, or a separate teacher. With only 20 unlabeled adaptation samples, TTRSD improves performance across seven benchmarks and three VLMs, raising InternVL3-2B's MMMU accuracy from 35.79% to 49.32%(+13.53%), demonstrating cross-dataset generalization while preserving inherent reasoning integrity.
Chinese Translation
测试时强化学习使视觉语言模型（VLMs）能够利用未标记输入进行适应。然而，在固定视觉条件下的重复采样可能会强化共享的感知错误，而序列级奖励未能隔离视觉感知这一基础瓶颈，从而影响多模态推理，风险是降低预训练的推理能力。我们提出了TTRSD，一个结合多视角答案级自蒸馏与视觉对比令牌选择的测试时强化学习框架。共享策略将来自原始、裁剪和下采样视图的教师预测聚合成答案分布。来自原始图像生成的学生轨迹根据其在该分布中对最终答案的支持程度获得奖励。为了将这种反馈精确分配到感知瓶颈，我们比较在固定文本前缀的情况下，原始输入和视觉消融输入下相同采样令牌的对数概率，选择视觉敏感的位置进行策略梯度更新。TTRSD将由组相对优势决定的更新方向与由视觉敏感性决定的更新位置分开，而无需真实标签、外部验证者或单独的教师。仅使用20个未标记的适应样本，TTRSD在七个基准和三个VLM上提高了性能，将InternVL3-2B的MMM准确率从35.79%提升至49.32%(+13.53%)，展示了跨数据集的泛化能力，同时保持了固有的推理完整性。
cs.CV / 211 / 2609.33419
TT-VidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining
TT-VidT：解耦时间轴以实现高效的运动中心视频预训练
Abstract
Comparisons in video self-supervised learning often evaluate complete training recipes rather than isolating the method itself: architecture, objective, data exposure, schedule, scale, and decoder capacity can all vary at once. This makes it hard to identify which choices yield motion-prioritized representations, whose gains concentrate on frame-to-frame change while retaining useful appearance. We address this with a matched $4 \times 6 = 24$ architecture-objective study at roughly 170M ~ 190M encoder scale on $\sim$1.7M OpenVid and Moments-in-Time v2 clips for 8 epochs, and propose TT-VidT. TT-VidT combines a DINOv3-initialized ViT-B/16 per-frame spatial path with a compact Temporal Transfer Layer, trained by Diff Compression to reconstruct target frames from a first-frame appearance anchor and frame-specific motion tokens. The sweep shows that TT3D with Diff Compression, not either component alone, enters the strongest motion-sensitive regime, and decoder ablations favor a compact video-pretrained decoder. In final comparison, TT-VidT leads Jester, Something-Something V2, ARID, and Diving48 fine-tuning simultaneously, improving over the strongest non-TT row by 54% ~ 121%, while using 48% fewer encoder FLOPs than DisMo and 55% fewer than VideoMAE or V-JEPA2. HMDB51, IARD, and EPIC-Kitchens bound the claim.
Chinese Translation
视频自监督学习中的比较通常评估完整的训练方案，而不是单独隔离方法本身：架构、目标、数据暴露、调度、规模和解码器容量都可能同时变化。这使得很难识别出哪些选择能够产生以运动为优先的表示，这些表示的增益集中在帧与帧之间的变化，同时保留有用的外观。我们通过在大约170M ~ 190M编码器规模上对约1.7M OpenVid和Moments-in-Time v2片段进行的匹配$4 imes 6 = 24$架构-目标研究，进行为期8个周期的训练，来解决这个问题，并提出了TT-VidT。TT-VidT结合了一个DINOv3初始化的ViT-B/16每帧空间路径和一个紧凑的时间传输层，通过差异压缩训练，以从第一帧外观锚点和帧特定运动标记重建目标帧。实验结果表明，结合差异压缩的TT3D，而不是单独的任一组件，进入了最强的运动敏感区域，而解码器消融实验则支持紧凑的视频预训练解码器。在最终比较中，TT-VidT在Jester、Something-Something V2、ARID和Diving48的微调中同时领先，较最强的非TT行提高了54% ~ 121%，同时使用的编码器FLOPs比DisMo少48%，比VideoMAE或V-JEPA2少55%。HMDB51、IARD和EPIC-Kitchens为该声明提供了界限。
cs.CV / 212 / 2609.33432
TC-ADA: One-Shot Active Domain Adaptation for Semantic Segmentation
TC-ADA：用于语义分割的一次性主动领域适应
Abstract
Manual dense annotation remains a major obstacle to deploying semantic segmentation models in new driving environments. Active domain adaptation (ADA) seeks label-efficient transfer by annotating only a selected portion of the target domain. Existing ADA methods commonly implement this process through multiple rounds of acquisition, annotation, and retraining. We study a practical one-shot image-level setting that selects and densely annotates a fixed target subset in a single round, followed by uninterrupted adaptation. Within this setting, we develop Target-Calibrated Active Domain Adaptation (TC-ADA) as a joint design of complete-image acquisition and target-calibrated adaptation. Stage~1 uses visual representations from a vision foundation model (VFM) together with semantic predictions from a fixed unsupervised domain adaptation model to select representative and informative target images without target annotations. Stage~2 jointly uses labeled source data, labeled target data, and the remaining unlabeled target data, while calibrating source and target supervision under limited target labels. Extensive experiments across five synthetic-to-real and real-to-real driving transfers show consistent improvements over representative ADA baselines. With only 23 to 46 labeled target images on four transfers and 140 on Mapillary, TC-ADA stays within 1.9 mean intersection over union (mIoU) points of target-only full supervision. Code will be available at https://github.com/ywher/TC-ADA.
Chinese Translation
手动密集标注仍然是将语义分割模型部署到新驾驶环境中的主要障碍。主动领域适应（ADA）通过仅对目标领域的选定部分进行标注，寻求标签高效的迁移。现有的ADA方法通常通过多轮获取、标注和再训练来实现这一过程。我们研究了一种实用的一次性图像级设置，该设置在单轮中选择并密集标注固定的目标子集，随后进行不间断的适应。在这一设置中，我们开发了目标校准主动领域适应（TC-ADA），作为完整图像获取和目标校准适应的联合设计。第一阶段使用来自视觉基础模型（VFM）的视觉表示以及来自固定无监督领域适应模型的语义预测，以在没有目标标注的情况下选择具有代表性和信息量的目标图像。第二阶段共同使用标注的源数据、标注的目标数据和剩余的未标注目标数据，同时在有限的目标标签下校准源和目标监督。在五个合成到真实和真实到真实的驾驶迁移中的广泛实验表明，相较于代表性的ADA基线，TC-ADA consistently shows consistent improvements。仅在四个迁移中使用23到46个标注的目标图像以及在Mapillary上使用140个，TC-ADA的平均交并比（mIoU）仅比目标仅全监督低1.9个点。代码将可在https://github.com/ywher/TC-ADA获取。
cs.CV / 213 / 2609.33445
Concept Score Relearning: A Unified Cross-Architecture Attack on Concept Erasure
概念分数重学习：对概念抹除的统一跨架构攻击
Abstract
Concept erasure aims to suppress undesirable knowledge in text-to-image generative models. However, existing robustness evaluations typically rely on relearning attacks tailored to specific model architectures. We study concept reactivation across two substantially different generative paradigms: noise-prediction U-Nets and flow-matching Transformers. We introduce \textbf{Concept Score Relearning (CSR)}, a unified parameter-level framework that reactivates erased concepts by optimizing each model within its native prediction space. CSR requires no external target-concept image dataset and applies the same concept-directed objective to both U-Net-based Stable Diffusion and Transformer-based FLUX. Experiments across diverse concepts and multiple erasure methods demonstrate consistent concept reactivation across both architectures, highlighting the cross-architecture applicability of CSR and the persistent recoverability of apparently erased concepts. For strict nudity, CSR reaches average ASRs of 50.47\% on FLUX and 40.29\% on Stable Diffusion, consistently ranking first across all evaluated safety settings.
Chinese Translation
概念抹除旨在抑制文本到图像生成模型中不良知识的存在。然而，现有的鲁棒性评估通常依赖于针对特定模型架构量身定制的重学习攻击。我们研究了在两种显著不同的生成范式中的概念再激活：噪声预测的 U-Net 和流匹配的 Transformer。我们引入了 extbf{概念分数重学习（Concept Score Relearning, CSR）}，这是一个统一的参数级框架，通过在每个模型的本地预测空间中优化来再激活被抹除的概念。CSR 不需要外部目标概念图像数据集，并对基于 U-Net 的稳定扩散（Stable Diffusion）和基于 Transformer 的 FLUX 应用相同的概念导向目标。针对多种概念和多种抹除方法的实验表明，两种架构之间的一致概念再激活，突显了 CSR 的跨架构适用性以及明显被抹除概念的持久可恢复性。在严格的裸体内容方面，CSR 在 FLUX 上达到了 50.47\% 的平均攻击成功率（ASR），在稳定扩散上达到了 40.29\\%，在所有评估的安全设置中始终排名第一。
cs.CV / 214 / 2609.33449
A Visual Classification Dataset and Model Evaluation for Historical Manuscript Illustrations
历史手稿插图的视觉分类数据集与模型评估
Abstract
Historical manuscript illustrations preserve rich visual evidence of past cultures. They depict people, animals, plants, diagrams, music notations, and decorative forms. Although large digitization projects have made many manuscripts available online, the material itself remains difficult to explore at scale. Extraction systems can find illustrations on manuscript pages, but without meaningful categories, large collections remain hard to search and explore. We address this gap by introducing a manually labeled dataset of 15,000 illustrations from manuscripts dating back hundreds of years across 22 categories, and evaluating modern vision models for image classification on this task. The problem is challenging due to stylistic diversity, degradation, and semantic ambiguity, with many images that fit more than one category. We compare fine-tuned CNN and Transformer-based classifiers, zero-shot CLIP, embedding-based classifiers, and direct vision-language models. Results show that fine-tuned image classifiers perform best overall, with ConvNeXt reaching 88.9% accuracy and 81.3% macro-F1. Using CLIP embeddings with XGBoost provides a strong alternative. In contrast, zero-shot CLIP and direct vision-language classification perform substantially worse, highlighting the limits of general-purpose models in this domain. Beyond overall performance, the analysis reveals which categories are visually separable and where errors reflect genuine semantic overlap, suggesting that some limitations arise from the taxonomy itself.
Chinese Translation
历史手稿插图保存了丰富的过去文化的视觉证据。它们描绘了人物、动物、植物、图表、音乐符号和装饰形式。尽管大型数字化项目使许多手稿可以在线获取，但材料本身在大规模探索时仍然困难。提取系统可以在手稿页面上找到插图，但缺乏有意义的分类，使得大型收藏难以搜索和探索。我们通过引入一个手动标记的数据集，包含来自数百年前的手稿的15,000幅插图，涵盖22个类别，来填补这一空白，并评估现代视觉模型在此任务上的图像分类能力。由于风格多样性、退化和语义模糊，问题具有挑战性，许多图像适合多个类别。我们比较了微调的卷积神经网络（CNN）和基于变换器（Transformer）的分类器、零样本CLIP、基于嵌入的分类器以及直接的视觉-语言模型。结果表明，微调的图像分类器整体表现最佳，其中ConvNeXt达到了88.9%的准确率和81.3%的宏观F1值。使用CLIP嵌入与XGBoost结合提供了一个强有力的替代方案。相比之下，零样本CLIP和直接的视觉-语言分类表现明显较差，突显了通用模型在这一领域的局限性。除了整体性能，分析还揭示了哪些类别在视觉上是可分的，以及错误反映了真正的语义重叠，表明某些限制源于分类法本身。
cs.CV / 215 / 2609.33450
Native Association: Confidence-Aware Human Perception in the Wild with a Foundation VLM
原生关联：基于基础视觉语言模型的野外信心感知人类识别
Abstract
Extracting who is where, on which team, wearing which number from a broadcast frame is typically done by stitching a detector, an OCR engine, and classifiers together -- and the stitching step swaps identities under occlusion. We make association native instead: a 0.77B vision-language model (Florence-2) is fine-tuned to emit all per-person attributes as one grammar-constrained sequence, with each attribute generated inside its owner's block. Output is therefore schema-valid on every frame by construction, and no post-hoc binding step exists to attach a correctly read number to the wrong player: residual misassociation is pure perception error, $\approx4\times$ rarer than zero-shot-prompted frontier APIs' (0.057 vs. 0.21-0.24). On a frozen multi-sport test set, this single pass reaches 0.95 detection F1 (APIs: 0.65-0.75). A single extra forward pass yields a per-field confidence that supports a reject option (jersey precision $0.71\rightarrow0.96$ at half coverage) and routes a training-free zoom-and-re-read for small players. Surprisingly, once the grammar is learned, further parameter-efficient tuning yields no measurable gain under the adaptation configurations we test; the identical recipe on WIDER-Attribute reaches 93.1 mAP given-box, yields the first detection-coupled end-to-end results under its standard test protocol (84.5 mAP), and reproduces the same tuning result. In this regime, the gains live in the structure, not in added weights.
Chinese Translation
从广播画面中提取谁在何处、属于哪个队伍、穿着哪个号码通常是通过将检测器、光学字符识别（OCR）引擎和分类器拼接在一起完成的——而拼接步骤在遮挡情况下会交换身份。我们使关联变得原生：一个7.7亿参数的视觉语言模型（Florence-2）经过微调，以发出所有每个人的属性作为一个语法约束序列，每个属性在其所有者的区块内生成。因此，输出在每一帧上都是结构有效的，并且不存在后期绑定步骤来将正确读取的号码附加到错误的球员上：残余的错误关联纯粹是感知错误，约为零-shot 提示前沿API的四分之一（0.057 对比 0.21-0.24）。在一个冻结的多运动项目测试集上，这个单次传递达到了0.95的检测F1（API：0.65-0.75）。一次额外的前向传递产生了每个字段的置信度，支持拒绝选项（球衣精度从0.71提升至0.96，覆盖率为一半），并为小球员提供了一种无训练的缩放和重读方式。令人惊讶的是，一旦语法被学习，进一步的参数高效调优在我们测试的适应配置下没有产生可测量的增益；在WIDER-Attribute上使用相同的配方达到了93.1的mAP，在给定框的情况下，产生了在其标准测试协议下的首个检测耦合端到端结果（84.5 mAP），并重现了相同的调优结果。在这种情况下，增益存在于结构中，而不是在增加的权重中。
cs.CV / 216 / 2609.33462
SphMind: Towards Robust, Training-Free VLM-based Spatial Reasoning with a 360 Camera
SphMind：朝着基于VLM的稳健、无训练空间推理的360度摄像头
Abstract
Omnidirectional or 360 cameras provide embodied AI agents with a holistic, wide field-of-view (FoV) view of their surroundings, motivating the use of Multi-modal Large Language Models (MLLMs) for omnidirectional spatial reasoning. However, most MLLMs are trained on conventional 2D perspective images and struggle with the severe distortions and wrap-around discontinuities induced by spherical geometry. Enabling them to generalize to non-Euclidean 3D spaces without retraining therefore remains challenging. We propose SphMind, a training-free, plug-and-play framework that decouples semantic perception from geometric reasoning. Rather than requiring MLLMs to learn spherical geometry internally, SphMind preserves their semantic capabilities while handling geometry externally. We introduce a Spherical Harmonics-based Spatial Graph (SHSG) that models spatial relationships through equivariant transformations on the sphere, together with Inference-Time Geometric Grounding (IGG), a model-agnostic closed-loop optimization process that aligns MLLM representations with spherical geometric constraints during inference. Experiments on three benchmarks show that SphMind achieves over 21.4% average improvement in directional reasoning on MP3D and Stanford2D-3D, outperforms prompt-engineering baselines by 8.7% on the real-world ODI-Bench, and improves rotational invariance by 5.9% under panorama rotations, without additional training or dataset-specific tuning. In-the-wild evaluations further show that SphMind resolves directional reasoning queries that baseline vision-language models fail to answer correctly.
Chinese Translation
全向或360度摄像头为具身人工智能代理提供了对其周围环境的整体宽视场（FoV）视图，这激励了多模态大型语言模型（MLLMs）在全向空间推理中的应用。然而，大多数MLLMs是在传统的二维透视图像上训练的，难以应对由球面几何引起的严重失真和环绕不连续性。因此，使它们能够在不重新训练的情况下推广到非欧几里得三维空间仍然具有挑战性。我们提出了SphMind，这是一种无训练、即插即用的框架，解耦了语义感知与几何推理。SphMind在处理几何时保持MLLM的语义能力，而不是要求MLLM内部学习球面几何。我们引入了一种基于球谐函数的空间图（SHSG），通过在球面上的等变变换建模空间关系，并结合推理时几何基础（IGG），这是一种模型无关的闭环优化过程，在推理过程中将MLLM表示与球面几何约束对齐。在三个基准测试上的实验表明，SphMind在MP3D和Stanford2D-3D上的方向推理平均提升超过21.4%，在真实世界的ODI-Bench上比提示工程基线提高8.7%，在全景旋转下提高了5.9%的旋转不变性，且无需额外训练或数据集特定调优。野外评估进一步表明，SphMind能够解决基线视觉-语言模型无法正确回答的方向推理查询。
cs.CV / 217 / 2609.33513
Printability-Constrained Adversarial Decals for Near-Nadir Aerial Perception: Measured Ink Gamuts, Nested Realism Constraints, and a Physical-World Bound
受打印性限制的对抗性贴花用于近天顶空中感知：测量的墨水色域、嵌套现实约束和物理世界界限
Abstract
Adversarial patches for aerial perception are typically evaluated as digital composites, with printing left as an implementation detail. This study imposes three physical constraints during optimization rather than after it: the color range a particular printer can reproduce, the size of the flat panel a vehicle offers, and the loss of fine detail incurred when the patch is imaged from altitude. The principal comparison isolates the ink set. Two patches share all seventeen recorded optimization settings and differ only in the colors available to them. One is constrained to a uniform color cube; the other to a gamut measured by printing and scanning a 216-patch chart. Each was optimized at three seeds and evaluated against thirteen victim conditions, with every rate reported against a size-matched optimized control. The effect of the measured gamut is victim-dependent rather than uniform. Net attack success rises on three of six closed-set segmentation victims, and for these the seed ranges of the two ink sets are disjoint: $+0.120$ on DeepLabv3-R101 and $+0.041$ on SegFormer-B0. The color-cube patch is consistently stronger on the open-vocabulary segmenter and on two of four detectors, though no detector exceeds a net of $+0.026$ under either ink set. The natural explanation is that a printable palette is simply less chromatic and lower in frequency than a digital one. Eleven further patches test this account and it does not hold. Once cardinality is matched, a palette as chromatic as the cube attacks equally well. Cardinality itself shows no trend from three inks to thirty-two. Palettes matched on cardinality, lightness and chroma, and differing only in hue placement, span $0.035$ to $0.136$. A physical evaluation with printed decals did not detect transfer; it bounds the transferred rate at $0.133$, which does not exclude the simulated value of $0.121$.
Chinese Translation
空中感知的对抗性贴花通常作为数字合成物进行评估，打印被视为实施细节。本研究在优化过程中施加三种物理约束，而不是在优化之后：特定打印机可再现的颜色范围、车辆提供的平面尺寸，以及从高空拍摄时贴花所损失的细节。主要比较集中在墨水集上。两个贴花共享所有十七个记录的优化设置，仅在可用颜色上有所不同。一个受到均匀颜色立方体的限制；另一个则受到通过打印和扫描216个贴花图表测量的色域的限制。每个贴花在三个种子下进行了优化，并针对十三种受害者条件进行了评估，所有比率均与大小匹配的优化控制进行比较。测量色域的影响因受害者而异，而非统一。在六个封闭集分割受害者中，有三个的净攻击成功率上升，对于这些受害者，两个墨水集的种子范围是互不重叠的：在DeepLabv3-R101上增加了$+0.120$，在SegFormer-B0上增加了$+0.041$。在开放词汇分割器和四个检测器中的两个上，颜色立方体贴花的表现始终更强，尽管在任何墨水集下，没有一个检测器的净值超过$+0.026$。自然的解释是，可打印的调色板在色彩饱和度上低于数字调色板，且频率较低。进一步的十一种贴花测试了这一解释，但结果并不成立。一旦基数匹配，色彩饱和度与立方体一样的调色板攻击效果相当。基数本身从三种墨水到三十二种没有显示出趋势。基于基数、亮度和色度匹配的调色板，仅在色相位置上有所不同，范围在$0.035$到$0.136$之间。通过打印贴花进行的物理评估未检测到转移；它将转移率限制在$0.133$，这并不排除模拟值$0.121$。
cs.CV / 218 / 2609.33518
SceneScaffold: Active Scene-State Construction for Unified 3D Scene Understanding
SceneScaffold：统一3D场景理解的主动场景状态构建
Abstract
Recent 3D large multimodal models (3D-LMMs) rely on a visual bottleneck to compress complex 3D scene evidence into a limited number of visual tokens compatible with large language models (LLMs). Current visual bottlenecks, however, often passively compress heterogeneous 3D evidence into a homogeneous object-centric token sequence, leaving the spatial organization of the scene under-represented. This under-representation forces the LLM to recover spatial relations from a flattened token sequence, leading to unstable reasoning in relation-intensive and spatially ambiguous scenes. To address this issue, we propose SceneScaffold, an active scene-state construction framework for unified 3D scene understanding. SceneScaffold reformulates the visual bottleneck from a passive feature compressor into an active scene organizer, constructing a role-aware spatial scaffold before language reasoning. Specifically, SceneScaffold organizes superpoint-level visual evidence into scene-state components with distinct structural roles: entity states preserve core object semantics, scene-frame states maintain spatial references via boundary and region anchors, relation states encode object-environment interaction cues, and a global summary provides compact context. Through this role-aware construction, SceneScaffold provides the LLM with a spatially organized scene representation before language reasoning. Experiments on unified 3D scene understanding tasks, including 3D visual grounding, question answering, and dense captioning, demonstrate the effectiveness of SceneScaffold, while diagnostic results further show its applicability to relation-intensive and spatially ambiguous cases. Code is available at https://github.com/lixiangqi707/SceneScaffold.
Chinese Translation
近期的3D大型多模态模型（3D-LMMs）依赖于视觉瓶颈，将复杂的3D场景证据压缩为与大型语言模型（LLMs）兼容的有限数量的视觉标记。然而，目前的视觉瓶颈通常被动地将异构的3D证据压缩为同质的以对象为中心的标记序列，导致场景的空间组织未得到充分表示。这种不足的表示迫使LLM从扁平化的标记序列中恢复空间关系，从而在关系密集和空间模糊的场景中导致推理不稳定。为了解决这一问题，我们提出了SceneScaffold，一个用于统一3D场景理解的主动场景状态构建框架。SceneScaffold将视觉瓶颈从被动特征压缩器重新构造为主动场景组织者，在语言推理之前构建一个角色感知的空间支架。具体而言，SceneScaffold将超点级视觉证据组织为具有不同结构角色的场景状态组件：实体状态保留核心对象语义，场景框架状态通过边界和区域锚点维护空间参考，关系状态编码对象与环境的交互线索，而全局摘要提供紧凑的上下文。通过这种角色感知的构建，SceneScaffold在语言推理之前为LLM提供了一个空间组织的场景表示。在统一3D场景理解任务上的实验，包括3D视觉定位、问答和密集描述，展示了SceneScaffold的有效性，而诊断结果进一步表明其在关系密集和空间模糊情况下的适用性。代码可在 https://github.com/lixiangqi707/SceneScaffold 获取。
cs.CV / 219 / 2609.33520
Anatomy-Structured Hierarchical MIL for Weakly-Supervised Thoracic Disease Detection in Chest X-rays
基于解剖结构的层次化多实例学习用于胸部X光片中的弱监督胸部疾病检测
Abstract
Weakly-supervised thoracic disease detection in chest X-rays (CXR) is challenging due to subtle appearances and complex anatomical overlap, motivating anatomy-aware modeling for improved localization. However, prior anatomy-aware methods typically rely on coarse region proxies or static spatial priors, which may restrict dynamic instance discovery and limit precise localization of small abnormalities. We propose Anatomy-Structured Hierarchical Multiple Instance Learning (ASH-MIL), a framework that introduces parallel anatomy-structured observation branches (cardiac, pulmonary, and agnostic) combined with hierarchical MIL aggregation. Anatomical priors are injected as soft spatial biases into decoder cross-attention, enabling anatomically grounded evidence maps without disease bounding-box supervision. Instance localization is derived directly from MIL-weighted cross-attention maps without bounding box supervision. Experiments on CXR8 and cross-domain MIMIC-CXR held-out sets demonstrate consistent improvements over prior weakly-supervised and anatomy-aware approaches, particularly under stricter localization criteria. Our code is available at https://github.com/jn-kim/ash-mil.
Chinese Translation
在胸部X光片（CXR）中进行弱监督胸部疾病检测具有挑战性，因为其表现微妙且解剖结构复杂重叠，这促使我们进行解剖知识驱动的建模以改善定位。然而，先前的解剖知识驱动方法通常依赖于粗略的区域代理或静态空间先验，这可能限制动态实例发现并限制小异常的精确定位。我们提出了基于解剖结构的层次化多实例学习（ASH-MIL）框架，该框架引入了并行的解剖结构观察分支（心脏、肺部和无特定器官）并结合层次化的多实例学习聚合。解剖先验作为软空间偏置注入到解码器的交叉注意力中，使得能够生成基于解剖的证据图，而无需疾病边界框的监督。实例定位直接来源于加权的交叉注意力图，而无需边界框监督。在CXR8和跨域MIMIC-CXR的保留数据集上的实验表明，在更严格的定位标准下，相较于先前的弱监督和解剖知识驱动方法，ASH-MIL展现出了一致的改进。我们的代码可在https://github.com/jn-kim/ash-mil获取。
cs.CV / 220 / 2609.33523
In-Token Learning for High-Fidelity Image Restoration via Diffusion Transformers
基于扩散变换器的高保真图像恢复中的内标记学习
Abstract
We present In-Token Learning, an image restoration framework that adapts a pretrained diffusion transformer using conditional rectified flow matching. Clean targets paired with degraded inputs supervise transport from Gaussian noise to restored images. Spatially aligned degraded-image tokens are fused with evolving latent tokens along the channel dimension, preserving the image-token count at a given resolution. Direct Low-Quality Guidance (DLG) combines frozen degraded-image embeddings with a fixed task prompt through the native conditioning pathway, without a trainable ControlNet-style branch or image captioning. We evaluate super-resolution and denoising on DIV2K, LSDIR, FFHQ, RealLQ250, and RealPhoto60, and automatic colorization on DIV2K and LSDIR. The tasks use separately trained checkpoints under the same framework. Results show competitive fidelity and perceptual quality under the evaluated protocols, with weaker generalization on RealLQ250. We report full-image QHD ($2560{\times}1440$) inference and a tiled $12$K restoration demonstration of Along the River During the Qingming Festival. Attention cost still increases with resolution. This technical report preserves the early broader study underlying Fill2SR, which subsequently developed the real-world super-resolution direction.
Chinese Translation
我们提出了内标记学习（In-Token Learning），这是一种图像恢复框架，通过条件修正流匹配（conditional rectified flow matching）来适应预训练的扩散变换器。干净的目标图像与退化输入配对，监督从高斯噪声到恢复图像的传输。空间对齐的退化图像标记与沿通道维度演变的潜在标记融合，保持在给定分辨率下的图像标记数量。直接低质量引导（Direct Low-Quality Guidance, DLG）通过本地条件路径将冻结的退化图像嵌入与固定任务提示结合，而无需可训练的ControlNet风格分支或图像字幕。我们在DIV2K、LSDIR、FFHQ、RealLQ250和RealPhoto60上评估超分辨率和去噪，并在DIV2K和LSDIR上进行自动上色。这些任务在相同框架下使用单独训练的检查点。结果显示在评估协议下具有竞争力的保真度和感知质量，但在RealLQ250上的泛化能力较弱。我们报告了全图QHD（$2560{ imes}1440$）推理和《清明上河图》的$12$K瓷砖恢复演示。注意力成本仍随分辨率增加。本技术报告保留了Fill2SR背后的早期更广泛研究，该研究随后发展了现实世界超分辨率的方向。
cs.CV / 221 / 2609.33558
PGL-3D: Towards Progressive Geometric Learning for 3D Visual Query Localization
PGL-3D：面向3D视觉查询定位的渐进几何学习
Abstract
3D Visual Query Localization (3DVQL) retrieves the latest contiguous occurrence of a queried object in an RGB--point-cloud sequence and predicts a 9-DoF cuboid for every response frame. The query is captured independently of the search sequence, so its annotated pose may differ from how the object appears in the search frames. The benchmark baseline predicts cuboids after feature modeling, leaving their geometry unused for subsequent feature refinement. We investigate whether complete intermediate cuboids can improve query and proposal representations before final decoding. We introduce Progressive Geometric Learning for 3DVQL (PGL-3D), a predict--select--refine--re-predict framework that uses intermediate cuboids to guide the aggregation of search evidence and update query and proposal representations. A shared head first predicts a complete cuboid for every proposal. Query--Tube--Memory (QTM) then selects reference observations by combining proposal association, cuboid quality, frame response, and target absence, since association confidence alone establishes neither target presence nor geometric accuracy. The center, size, and orientation of each selected cuboid define soft pooling weights over query-conditioned proposal features. The pooled memory updates the query and proposal representations, and the head re-predicts from the updated features. A training-only objective, ST-D9O, supervises cuboid geometry at every stage by adding boundary, signed-distance, and soft-overlap terms to parameter regression. PGL-3D achieves a mean stAP of $0.270 \pm 0.004$ on 3DVQL, compared with $0.044$ reported for LaF. Ablations support the benefits of geometry-guided feature updates, while stage-wise analyses show improved cuboid accuracy. Replacing the geometry objective in our PROT3D reproduction with ST-D9O improves mAO on GSOT3D from $21.63\%$ to $25.78\%$. Our code and models will be released.
Chinese Translation
3D视觉查询定位（3DVQL）在RGB-点云序列中检索查询对象的最新连续出现，并为每个响应帧预测一个9自由度的立方体。查询独立于搜索序列进行捕捉，因此其标注姿态可能与对象在搜索帧中的外观不同。基准基线在特征建模后预测立方体，未能利用其几何信息进行后续特征细化。我们研究完整的中间立方体是否能够在最终解码之前改善查询和提议表示。我们提出了用于3DVQL的渐进几何学习（PGL-3D），这是一个预测-选择-细化-重新预测的框架，利用中间立方体指导搜索证据的聚合，并更新查询和提议表示。共享头部首先为每个提议预测一个完整的立方体。查询-管道-记忆（QTM）通过结合提议关联、立方体质量、帧响应和目标缺失来选择参考观察，因为仅凭关联置信度无法确定目标的存在或几何精度。每个选定立方体的中心、大小和方向定义了基于查询条件的提议特征的软池化权重。池化记忆更新查询和提议表示，头部从更新的特征中重新预测。仅用于训练的目标ST-D9O通过将边界、符号距离和软重叠项添加到参数回归中，监督每个阶段的立方体几何。PGL-3D在3DVQL上实现了平均stAP为$0.270 imes 0.004$，而LaF报告的为$0.044$。消融实验支持几何引导特征更新的好处，而阶段分析显示立方体精度的改善。在我们的PROT3D重现中用ST-D9O替换几何目标，使GSOT3D上的mAO从$21.63\%$提高到$25.78\\%$。我们的代码和模型将被发布。
cs.CV / 222 / 2609.33581
ForeFly: A Dual-Horizon World Action Model for Aerial Vision-Language Navigation
ForeFly：一种用于空中视觉-语言导航的双视界世界动作模型
Abstract
Aerial Vision-Language Navigation (AVLN) requires UAVs to maintain reliable instruction following over long trajectories in complex 3D environments. However, existing AVLN approaches are predominantly reactive or limited to single-horizon prediction, overlooking complementary future cues across different temporal horizons. To address this limitation, we propose ForeFly, a dual-horizon latent world action model that predicts both a proximal future for local continuity and an adaptive route-critical future for long-range guidance. Horizon-specific foresight queries are primed with recent and route-critical visual memories, providing history-aware context for future prediction. To exploit their distinct roles in action generation, we introduce Foresight-Guided Action Refinement (FGAR), which asymmetrically exploits proximal foresight for local action enhancement and route-critical foresight for feature-wise correction and route-level guidance. Experiments on the TravelUAV and UAV-ON benchmarks show that ForeFly consistently outperforms strong baselines across seen and unseen settings, validating the effectiveness of dual-horizon foresight and FGAR learning. The code is available at: https://github.com/kunhuiW/ForeFly
Chinese Translation
空中视觉-语言导航（AVLN）要求无人机在复杂的三维环境中沿着长轨迹可靠地执行指令。然而，现有的AVLN方法主要是反应式的或限于单一视界预测，忽视了不同时间视界之间的互补未来线索。为了解决这一局限，我们提出了ForeFly，一种双视界潜在世界动作模型，能够预测局部连续性的近端未来和用于长途指导的适应性路线关键未来。视界特定的前瞻查询与近期和路线关键的视觉记忆相结合，为未来预测提供了历史感知的背景。为了充分利用它们在动作生成中的不同角色，我们引入了前瞻引导的动作精炼（FGAR），该方法不对称地利用近端前瞻进行局部动作增强，并利用路线关键前瞻进行特征级修正和路线级指导。在TravelUAV和UAV-ON基准测试中的实验表明，ForeFly在已见和未见的设置中始终优于强基线，验证了双视界前瞻和FGAR学习的有效性。代码可在以下链接获取：https://github.com/kunhuiW/ForeFly
cs.CV / 223 / 2609.33582
Fill2SR: Repurposing Inpainting Diffusion Transformers for Real-World Super-Resolution
Fill2SR：将图像修复扩散变换器重新用于真实世界超分辨率
Abstract
Recent real-world image super-resolution (SR) methods often adapt text-to-image (T2I) backbones with ControlNet-style branches or spatial conditioning tokens, which increases memory and computes with resolution and often constrains training to a fixed scale. We propose Fill2SR, which repurposes a masked-inpainting Diffusion Transformer for SR without extra spatial branches. Our Inpainting-Interface Evidence Adapter (IIEA) writes the low-quality (LQ) observation into the native masked-image slot under a full-image mask, turning inpainting into a reverse-degradation conditional rectified flow trained with LoRA-only tuning. We further introduce RCDT, an offline pipeline that distills degradation descriptors from unpaired real images and transfers them onto clean targets using frozen open-source models. Fill2SR supports mixed-resolution training up to QHD and yields stable performance across $512/1024/2048$ outputs. On synthetic benchmarks, our base model with IIEA achieves the best LPIPS on DIV2K and LSDIR; adding RCDT trades a small LPIPS drop for consistently stronger no-reference quality on RealLQ250 and RealPhoto60. Fill2SR remains memory-predictable, running $1536^2$ inference on a single 32GB GPU and extending to multi-megapixel outputs via tiled restoration.
Chinese Translation
最近的真实世界图像超分辨率（SR）方法通常采用文本到图像（T2I）主干网络，并结合ControlNet风格的分支或空间条件标记，这在分辨率提高时增加了内存和计算需求，并且通常将训练限制在固定的尺度。我们提出了Fill2SR，该方法重新利用了掩码图像修复扩散变换器进行超分辨率，而无需额外的空间分支。我们的修复接口证据适配器（IIEA）将低质量（LQ）观测值写入全图掩码下的原生掩码图像槽，将图像修复转变为一种逆降解条件修正流，并通过仅使用LoRA调优进行训练。我们进一步引入了RCDT，一个离线管道，从未配对的真实图像中提取降解描述符，并利用冻结的开源模型将其转移到干净的目标上。Fill2SR支持高达QHD的混合分辨率训练，并在$512/1024/2048$输出上产生稳定的性能。在合成基准测试中，我们的基础模型与IIEA在DIV2K和LSDIR上达到了最佳的LPIPS；添加RCDT则在RealLQ250和RealPhoto60上以小幅LPIPS下降换取了更强的无参考质量。Fill2SR保持内存可预测性，在单个32GB GPU上运行$1536^2$推理，并通过平铺恢复扩展到多百万像素输出。
cs.CV / 224 / 2609.33585
IVT-Guard: All-in-One Reasoning Model for AI-Generated Content Detection
IVT-Guard：用于AI生成内容检测的全能推理模型
Abstract
The rapid proliferation of highly realistic AI-Generated Content (AIGC) necessitates robust and interpretable detection mechanisms. However, existing detectors are predominantly confined to single modalities and provide binary outputs without reasoning. While Multimodal Large Language Models (MLLMs) present a promising solution, their development is constrained by the scarcity of multimodal reasoning data and the reasoning-detection optimization dilemma, where explicit reasoning supervision can compromise detection accuracy. To this end, we introduce IVT-Set, a comprehensive dataset comprising over 152K diverse image, video, and text samples equipped with multi-granularity Chain-of-Thought (CoT) reasoning trajectories. Based on it, we propose IVT-Guard, a pioneering framework for unified and interpretable AIGC detection across image, video, and text modalities. Furthermore, to overcome the aforementioned optimization dilemma, we design a novel three-stage training paradigm: Artifact-Aware Pre-training, Artifact-to-Evidence Supervised Fine-Tuning via artifact-aware injection, and Evidence-Verdict Consistency Group Relative Policy Optimization. Extensive experiments demonstrate that IVT-Guard achieves state-of-the-art detection performance across in-domain, out-of-domain, and cross-dataset settings while delivering faithful reasoning. Code and data will be released.
Chinese Translation
高度逼真的AI生成内容（AIGC）的快速传播需要强大且可解释的检测机制。然而，现有的检测器主要局限于单一模态，并且提供没有推理的二元输出。尽管多模态大型语言模型（MLLMs）展现了有希望的解决方案，但其发展受到多模态推理数据稀缺和推理-检测优化困境的限制，其中显式推理监督可能会影响检测准确性。为此，我们引入了IVT-Set，这是一个综合数据集，包含超过152K个多样化的图像、视频和文本样本，并配备多粒度的思维链（CoT）推理轨迹。在此基础上，我们提出了IVT-Guard，这是一个开创性的框架，用于在图像、视频和文本模态中统一且可解释的AIGC检测。此外，为了克服上述优化困境，我们设计了一种新颖的三阶段训练范式：工件感知预训练、通过工件感知注入进行的工件到证据的监督微调，以及证据-裁决一致性组相对策略优化。大量实验表明，IVT-Guard在领域内、领域外和跨数据集设置中实现了最先进的检测性能，同时提供了可靠的推理。代码和数据将会发布。
cs.CV / 225 / 2609.33593
LoopLUT: 3D Lookup Tables with Progressive Region Refinement for Real-Time 4K Image Enhancement
LoopLUT：具有渐进区域细化的三维查找表用于实时4K图像增强
Abstract
Color enhancement of 4K images must meet a quality target under a tight compute budget. Three-dimensional lookup tables (3D LUTs) dominate real-time enhancement because they decide at low resolution and apply a per-pixel lookup at full resolution. A single global LUT, however, is spatially invariant, so an underexposed shadow and a well-exposed region that share a pixel value receive identical corrections. Spatially heterogeneous demands cannot be expressed by such a mapping. We propose LoopLUT, a region-cascaded 3D LUT with progressive refinement. A global LUT performs the overall correction, followed by K-1 loop iterations. In each iteration a gating head predicts at low resolution the region that still needs correction, then builds a residual LUT from the color statistics of that region alone. The cascaded gates form a partition of unity, so the output is a per-pixel convex combination of the K lookup results. Fusion is therefore performed by the gates themselves, with no separate fusion module and no interpolation error accumulating across rounds. The decision stage runs at a fixed 256x256 resolution, independent of output resolution, so a 4K image costs only K pure lookups. Extensive experiments across four benchmarks show that LoopLUT improves PSNR by up to 2.81 dB over the strongest prior method, while keeping real-time throughput at 4K. The same decomposition also generalizes well to underwater enhancement datasets.
Chinese Translation
4K图像的色彩增强必须在紧凑的计算预算下达到质量目标。三维查找表（3D LUT）在实时增强中占据主导地位，因为它们在低分辨率下决定，并在全分辨率下应用逐像素查找。然而，单一的全局LUT在空间上是不变的，因此一个曝光不足的阴影区域和一个曝光良好的区域如果共享相同的像素值，则会收到相同的修正。这种映射无法表达空间异质的需求。我们提出了LoopLUT，一种具有渐进细化的区域级联3D LUT。全局LUT执行整体修正，随后进行K-1次循环迭代。在每次迭代中，门控头在低分辨率下预测仍需修正的区域，然后仅根据该区域的颜色统计构建残差LUT。级联门控形成了一个统一的划分，因此输出是K个查找结果的逐像素凸组合。因此，融合由门控本身执行，没有单独的融合模块，也没有在轮次间累积的插值误差。决策阶段以固定的256x256分辨率运行，与输出分辨率无关，因此处理一幅4K图像仅需K次纯查找。针对四个基准的广泛实验表明，LoopLUT在保持4K实时吞吐量的同时，相较于最强的先前方法提高了高峰信噪比（PSNR）最多2.81 dB。相同的分解方法在水下增强数据集上也表现良好。
cs.CV / 226 / 2609.33603
ViCoR: Reliable Molecular Structure Extraction via Spatially Aligned Verification and Executable Revision
ViCoR：通过空间对齐验证和可执行修订实现可靠的分子结构提取
Abstract
Reliable optical chemical structure recognition (OCSR) is essential for building high-quality chemical data from scientific literature, yet even small recognition errors can propagate into chemical databases and downstream models. In practice, recognized structures often require manual inspection and correction before use, making large-scale data curation costly and difficult to scale. We therefore study Selective Structure Recognition (SSR), a post-recognition setting that automatically produces reliable structured outputs while rejecting unresolved cases. Selection-only approaches can improve reliability by rejection, but cannot create additional correct outputs beyond those produced by the base recognizer. We propose ViCoR, a repair-before-rejection framework for iterative VerIfiCatiOn and Revision. Its key idea is to make observation-prediction correspondence explicit: coordinate-preserving rendering establishes spatial correspondence between the source image and predicted structure, while index anchoring maps localized visual discrepancies to executable graph edits without full-structure regeneration. A shared VLM is progressively trained from verification to revision. On two real-world OCSR benchmarks, ViCoR improves overall accuracy from 73.53\% to 88.26\% and from 61.83\% to 84.32\%, while achieving over 97\% accepted accuracy at 85--89\% coverage. The resulting molecular data further improve reaction-extraction F1 by 15.5 points and literature-sourced reaction prediction accuracy by 7.7 and 5.8 points, demonstrating the value of automated reliability control for scientific data curation and downstream chemical learning.
Chinese Translation
可靠的光学化学结构识别（OCSR）对于从科学文献中构建高质量的化学数据至关重要，然而即使是小的识别错误也可能传播到化学数据库和下游模型中。在实践中，识别出的结构通常需要手动检查和修正后才能使用，这使得大规模数据整理变得昂贵且难以扩展。因此，我们研究了选择性结构识别（SSR），这是一种后识别设置，能够自动生成可靠的结构化输出，同时拒绝未解决的案例。仅选择的方法可以通过拒绝提高可靠性，但无法生成超出基础识别器所产生的额外正确输出。我们提出了ViCoR，一种在拒绝之前进行修复的框架，用于迭代的验证和修订。其关键思想是明确观察-预测对应关系：保持坐标的渲染建立了源图像与预测结构之间的空间对应关系，而索引锚定将局部视觉差异映射到可执行的图形编辑，而无需完全重建结构。一个共享的视觉语言模型（VLM）在验证到修订的过程中逐步训练。在两个真实世界的OCSR基准测试中，ViCoR将整体准确率从73.53%提高到88.26%，从61.83%提高到84.32%，同时在85%到89%的覆盖率下实现了超过97%的接受准确率。由此产生的分子数据进一步将反应提取的F1值提高了15.5个百分点，文献来源的反应预测准确率提高了7.7和5.8个百分点，证明了自动化可靠性控制在科学数据整理和下游化学学习中的价值。
cs.CV / 227 / 2609.33611
Anguinus Sculpturae: Compositional Synthesis of Peak-Enhancement Breast DCE-MRI Scans
Anguinus Sculpturae：峰值增强乳腺DCE-MRI扫描的组合合成
Abstract
Dynamic contrast-enhanced breast MRI (DCE-MRI) is rich in anatomical and perfusion information, but its reliance on gadolinium-based contrast agents raises safety concerns and adds cost. Virtual contrast enhancement, synthesizing post-contrast from pre-contrast images, is a promising alternative. We address the MAMA-SYNTH challenge task of predicting peak-enhancement breast MRI. Rather than adopting the full machinery of diffusion or flow matching, we observe that under a rectified, straight-line path the generative process collapses to a single difference prediction: the synthetic peak image is the pre-contrast image plus a predicted enhancement map, recovered in one forward pass. Around this we build Anguinus Sculpturae, a compositional pipeline in which nnU-Net segmentations of lesion, foreground and breast region guide two generators - one optimized for global fidelity, one for lesion structure through an asymmetric Tversky term routed via a frozen segmenter - composited region-wise with Gaussian-weighted blending. On the held-out Duke subset of MAMA-MIA our model achieves the best FRD and Dice among all evaluated variants, showing that single-step difference prediction with segmentation guidance suffices to recover both global fidelity and lesion structure. Code is available at https://github.com/MIC-DKFZ/AnguinusSculpturae.
Chinese Translation
动态对比增强乳腺MRI（DCE-MRI）富含解剖和灌注信息，但其对基于钆的对比剂的依赖引发了安全性担忧并增加了成本。虚拟对比增强，即从对比前图像合成对比后图像，是一种有前景的替代方案。我们针对MAMA-SYNTH挑战任务，预测峰值增强乳腺MRI。我们观察到，在经过整流的直线路径下，生成过程简化为单一差异预测：合成的峰值图像是对比前图像加上一个预测的增强图，且在一次前向传递中恢复。基于此，我们构建了Anguinus Sculpturae，这是一个组合管道，其中nnU-Net对病变、前景和乳腺区域的分割引导两个生成器——一个优化全局保真度，另一个通过一个不对称的Tversky项优化病变结构，该项通过一个冻结的分割器进行路由——并通过高斯加权混合进行区域合成。在MAMA-MIA的Duke子集上，我们的模型在所有评估变体中实现了最佳的FRD和Dice，显示出带有分割指导的单步差异预测足以恢复全局保真度和病变结构。代码可在https://github.com/MIC-DKFZ/AnguinusSculpturae获取。
cs.CV / 228 / 2609.33616
SpatialSpeak: QA-Native Reconstruction with Local and Global Context for Spatial Chain-of-Thought Reasoning
SpatialSpeak：结合局部和全局上下文的空间链式思维推理的QA原生重建
Abstract
Vision-language models (VLMs) can benefit from geometric priors for multi-view spatial reasoning, yet answer-only training does not directly supervise the intermediate geometric estimates and their use in deriving quantitative spatial answers. We hypothesize that spatial chain-of-thought (CoT) supervision becomes more effective when the VLM first jointly learns complementary local geometry and global scene context through multi-view reconstruction. We introduce SpatialSpeak, a two-stage framework that connects QA-native reconstruction pretraining with spatial CoT learning. In Stage I, QA-Native Reconstruction Pretraining (QA-RP) combines marked-point 3D queries for fine-grained local geometry with object-center queries for global scene context across views. Both tasks are formulated as text-based question answering, allowing geometric estimation and subsequent reasoning to share the same autoregressive output interface. In Stage II, spatial CoT with Visual Compensation (CoT-VC) trains the model to express question-relevant geometric estimates and use them to derive answers, with reliability assessment and visual compensation supporting answer refinement when needed. On ReVSI, QA-RP increases the gain from CoT-VC from 2.6 to 6.9 points, and ablations show that both local and global reconstruction supervision are beneficial. SpatialSpeak achieves state-of-the-art results on ReVSI, VSI-Bench, and SPAR-Bench, with a ReVSI score of 62.8 that exceeds the strongest compared baseline by 8.7 points.
Chinese Translation
视觉语言模型（VLMs）可以通过几何先验来提升多视角空间推理的能力，但仅基于答案的训练并未直接监督中间几何估计及其在推导定量空间答案中的应用。我们假设，当VLM首先通过多视角重建共同学习互补的局部几何和全局场景上下文时，空间链式思维（CoT）监督会变得更加有效。我们提出了SpatialSpeak，一个将QA原生重建预训练与空间CoT学习相结合的两阶段框架。在第一阶段，QA原生重建预训练（QA-RP）结合了用于细粒度局部几何的标记点3D查询和用于全局场景上下文的物体中心查询。两个任务都被表述为基于文本的问题回答，使得几何估计和后续推理可以共享相同的自回归输出接口。在第二阶段，带有视觉补偿的空间CoT（CoT-VC）训练模型表达与问题相关的几何估计，并利用这些估计推导答案，同时在需要时通过可靠性评估和视觉补偿支持答案的细化。在ReVSI上，QA-RP使得CoT-VC的增益从2.6提高到6.9分，消融实验表明，局部和全局重建监督都是有益的。SpatialSpeak在ReVSI、VSI-Bench和SPAR-Bench上取得了最先进的结果，ReVSI得分为62.8，超过了最强对比基线8.7分。
cs.CV / 229 / 2609.33627
StoryEngine: A State-Grounded Agentic Framework for Video Storytelling
StoryEngine：一个基于状态的代理框架用于视频叙事
Abstract
Despite recent progress in agentic multi-shot video generation, producing coherent and consistent long-form stories remains challenging. Existing agentic pipelines typically rely on textual shot plans or previously generated pixels, yet lack an explicit mechanism for propagating the consequences of story events and maintaining the video world state across shots. As a result, missing visual details may be reconstructed inaccurately, while visual drift may propagate across subsequent shots, undermining both narrative coherence and visual consistency. To address these challenges, we propose StoryEngine, a state-grounded agentic framework for video storytelling. StoryEngine establishes a separation between authoritative semantic plans and unreliable visual observations. Specifically, StoryEngine maintains a structured representation of entity placement and story-relevant states, and propagates event-induced changes to define the intended start and end states of each shot. To visually realize these states, StoryEngine constructs canonical references for recurring entities and environments, and compiles state and visual constraints into executable render plans. Meanwhile, to realize these states correctly, a bounded evaluation-guided repair loop further corrects local state inconsistencies. Together, these mechanisms preserve causal story progression and prevent local visual errors from propagating across shots. To comprehensively evaluate long-form storytelling, we construct a benchmark across diverse scenarios and visual styles, with metrics assessing storytelling quality, narrative coherence, and visual consistency. Experimental results demonstrate that StoryEngine consistently outperforms state-of-the-art methods across all evaluation dimensions, validating its effectiveness for coherent and consistent video storytelling.
Chinese Translation
尽管在代理多镜头视频生成方面取得了近期进展，但生成连贯且一致的长篇故事仍然具有挑战性。现有的代理流程通常依赖于文本镜头计划或先前生成的像素，但缺乏明确的机制来传播故事事件的后果，并在镜头之间维持视频世界状态。因此，缺失的视觉细节可能被不准确地重建，而视觉漂移可能在后续镜头中传播，从而破坏叙事的连贯性和视觉的一致性。为了解决这些挑战，我们提出了StoryEngine，一个基于状态的代理框架用于视频叙事。StoryEngine在权威语义计划和不可靠视觉观察之间建立了分离。具体而言，StoryEngine维护了一个结构化的实体位置和与故事相关状态的表示，并传播事件引起的变化以定义每个镜头的预期起始和结束状态。为了在视觉上实现这些状态，StoryEngine为重复出现的实体和环境构建了规范参考，并将状态和视觉约束编译成可执行的渲染计划。同时，为了正确实现这些状态，一个有界的评估引导修复循环进一步纠正局部状态不一致性。这些机制共同保持因果故事进展，并防止局部视觉错误在镜头之间传播。为了全面评估长篇叙事，我们在多种场景和视觉风格中构建了一个基准，使用指标评估叙事质量、叙事连贯性和视觉一致性。实验结果表明，StoryEngine在所有评估维度上始终优于最先进的方法，验证了其在连贯且一致的视频叙事中的有效性。
cs.CV / 230 / 2609.33659
Learning Multimodal Embeddings with Evidence-Aligned Readout
学习与证据对齐的多模态嵌入
Abstract
Multimodal large language models can expose task-relevant evidence through generation, but producing useful evidence does not by itself determine how it enters a retrieval embedding. We study whether the semantic organization of that evidence can also specify where representations are read. To address this question, we introduce EviAlign, which couples Semantic Evidence Generation with Boundary Readout in a shared multimodal large language model. It organizes evidence into five semantic units, reads the contextualized state at each unit boundary, and aggregates these states into a single normalized embedding. Generation and contrastive retrieval objectives jointly train this shared structure. With the same trailing readout, semantic evidence and free-form CoT yield nearly identical retrieval performance, suggesting that evidence organization alone does not explain the full gain. A controlled $2\times3$ study compares consistent and permuted evidence organization across three readout strategies, using training targets with matched evidence spans. With five readout states and the same mean pooling, the advantage of consistent semantic organization grows from 0.65 points at length-based training positions to 2.39 at evidence boundaries, yielding a 1.74-point co-design interaction. Across 12 MMEB retrieval tasks, EviAlign achieves 76.9 average Recall@1 with 500K training pairs while retaining single-vector indexing and scoring.
Chinese Translation
多模态大型语言模型可以通过生成暴露与任务相关的证据，但产生有用的证据本身并不能决定其如何进入检索嵌入。我们研究了该证据的语义组织是否也能指定表示的读取位置。为了解决这个问题，我们引入了EviAlign，它将语义证据生成与边界读取结合在一个共享的多模态大型语言模型中。它将证据组织成五个语义单元，在每个单元边界读取上下文化状态，并将这些状态聚合成一个单一的归一化嵌入。生成和对比检索目标共同训练这个共享结构。在相同的尾部读取下，语义证据和自由形式的链式推理（CoT）几乎产生相同的检索性能，表明证据组织本身并不能解释全部增益。一个控制的$2 imes3$研究比较了三种读取策略下的一致和置换证据组织，使用具有匹配证据跨度的训练目标。在五个读取状态和相同的均值池化下，一致的语义组织的优势从基于长度的训练位置的0.65点增长到在证据边界的2.39点，产生了1.74点的共同设计交互。在12个MMEB检索任务中，EviAlign在500K训练对的情况下实现了76.9的平均Recall@1，同时保持单向量索引和评分。
cs.CV / 231 / 2609.33668
When Noise Meets Long-Tail: Feature-Threshold Dual Calibration for Robust Pseudo-Labeling
当噪声遇到长尾：用于鲁棒伪标签的特征-阈值双重校准
Abstract
Pseudo-labeling has become a cornerstone of learning from unlabeled data in semantic segmentation. Yet its effectiveness drops sharply in real-world scenarios where strong imaging noise and long-tailed class distributions occur together. We trace this failure to a vicious cycle of pseudo-label degradation. Imaging noise entangles foreground and background features, lowering prediction confidence across all classes, while long-tailed distributions leave tail classes with far fewer training samples and inherently lower confidence. Under fixed high-threshold filtering, these tail-class predictions are systematically filtered out, so they receive no supervision from unlabeled data and thus features keep degrading in subsequent iterations. Critically, noise and long-tail are not independent obstacles but mutually amplifying ones, and addressing either alone is insufficient. To break this cycle, we propose FTC-Seg, a Feature-Threshold dual-Calibration framework built on a standard teacher-student framework. At the feature level, Orthogonal Prototype Reconstruction (OPR) uses a set of learnable orthogonal prototypes to residually purify pixel-wise features, widening the margin between weak foreground targets and noisy backgrounds. At the threshold level, Adaptive Threshold Calibration (ATC) dynamically adjusts class-specific thresholds based on learning difficulty and prediction-distribution bias, rescuing low-confidence pseudo-labels of tail classes from systematic exclusion. Extensive experiments on four public benchmarks spanning three distinct noise modalities show that FTC-Seg achieves strong performance against state-of-the-art methods, with particularly substantial gains on tail classes. Our results establish that jointly calibrating features and thresholds is essential for robust pseudo-labeling under compounded noise and class imbalance.
Chinese Translation
伪标签已经成为语义分割中从未标记数据学习的基石。然而，在强成像噪声和长尾类别分布同时出现的现实场景中，其有效性急剧下降。我们将这种失败追溯到伪标签退化的恶性循环。成像噪声将前景和背景特征纠缠在一起，降低了所有类别的预测置信度，而长尾分布使得尾部类别的训练样本数量远远不足，固有的置信度也较低。在固定的高阈值过滤下，这些尾部类别的预测系统性地被过滤掉，因此它们没有来自未标记数据的监督，从而在后续迭代中特征持续退化。关键是，噪声和长尾并不是独立的障碍，而是相互增强的障碍，单独解决其中一个是不够的。为了打破这一循环，我们提出了FTC-Seg，一个基于标准教师-学生框架构建的特征-阈值双重校准框架。在特征层面，正交原型重建（Orthogonal Prototype Reconstruction, OPR）使用一组可学习的正交原型来残余净化逐像素特征，扩大弱前景目标与噪声背景之间的间隔。在阈值层面，自适应阈值校准（Adaptive Threshold Calibration, ATC）根据学习难度和预测分布偏差动态调整类别特定的阈值，拯救尾部类别的低置信度伪标签，避免系统性排除。在四个涵盖三种不同噪声模式的公共基准上的大量实验表明，FTC-Seg在与最先进方法的比较中表现出色，尤其在尾部类别上取得了显著的提升。我们的结果表明，在复合噪声和类别不平衡的情况下，联合校准特征和阈值对于鲁棒伪标签至关重要。
cs.CV / 232 / 2609.33683
MAD-Guard: Controlled Study of Autoregressive Generation versus Direct Decision Interfaces for Closed Multimodal Forensic Tasks
MAD-Guard：自回归生成与直接决策接口在封闭多模态法医学任务中的对照研究
Abstract
When should multimodal foundation models generate tokens, and when should they directly output a decision? We present MAD-Guard, a controlled study of output-decision interfaces for closed multimodal forensic tasks. Once a multimodal representation is computed, is autoregressive generation necessary for closed forensic decisions with high input complexity but low output entropy? Under a matched Qwen3-VL-8B backbone, 2,400 FakeClue training samples, and LoRA budget ($r=16, \alpha=32$) on Huawei Ascend 910C NPUs, we evaluate a progression of decision interfaces (AR-SFT [generate] $\to$ Logit Slice $\to$ Binary Direct Head $\to$ +choice $\to$ +act $\to$ CLM-Head) and decompose latency into backbone representation (53.12 ms), 151,643-way vocabulary projection (+85.04 ms $\to$ 138.16 ms), and decoding (+248.26 ms $\to$ 386.42 ms). Under 1-to-1 binary supervision ($\mathcal{L}_{\mathrm{BCE}}$), a Binary Direct Head cuts latency by $2.60\times$-$7.27\times$ (53.12 ms) and lowers calibration error by $1.88\times$ (ECE = 0.0450 vs. 0.0845), with a -1.80% accuracy trade-off (93.10% vs. 94.90%; 0.9795 vs. 0.9871 ROC-AUC) from forfeiting token priors. Gains above AR-SFT arise either from multi-task attribution and uncertainty gating (+choice+act: 96.44% accuracy, 0.9940 ROC-AUC, 0.0187 ECE at 53.71 ms) or from a disaggregated contrastive head (CLM-Head: 96.55% binary and 96.44% multi-task accuracy, 0.0166 ECE, 98.79% 7-class attribution at 54.42 ms) retaining semantic priors without token decoding. Across 5,000 out-of-sample images from five benchmarks, our framework excels on synthetic, camouflage, and document forgeries (96.44% GenImage, 97.73% Chameleon, 91.84% Doc) while showing a clear boundary on compressed face manipulation (FF++ ROC-AUC = 0.5913).
Chinese Translation
多模态基础模型何时应生成标记，何时应直接输出决策？我们提出了MAD-Guard，这是一个针对封闭多模态法医学任务的输出决策接口的对照研究。一旦计算出多模态表示，自回归生成在输入复杂性高但输出熵低的封闭法医学决策中是否必要？在匹配的Qwen3-VL-8B骨干网络下，使用2400个FakeClue训练样本，以及在华为Ascend 910C NPU上进行的LoRA预算（$r=16, eta=32$），我们评估了一系列决策接口（AR-SFT [生成] $ o$ Logit Slice $ o$ Binary Direct Head $ o$ +choice $ o$ +act $ o$ CLM-Head），并将延迟分解为骨干表示（53.12毫秒）、151,643维词汇投影（+85.04毫秒 $ o$ 138.16毫秒）和解码（+248.26毫秒 $ o$ 386.42毫秒）。在1对1的二元监督下（$ ext{L}_{ ext{BCE}}$），Binary Direct Head将延迟减少了$2.60 imes$-$7.27 imes$（53.12毫秒），并将校准误差降低了$1.88 imes$（ECE = 0.0450对比0.0845），以牺牲-1.80%的准确率为代价（93.10%对比94.90%；0.9795对比0.9871 ROC-AUC）。相较于AR-SFT的增益来自于多任务归因和不确定性门控（+choice+act：96.44%准确率，0.9940 ROC-AUC，53.71毫秒时的0.0187 ECE）或来自于一个解构的对比头（CLM-Head：96.55%二元和96.44%多任务准确率，0.0166 ECE，54.42毫秒时的98.79% 7类归因），在不进行标记解码的情况下保留语义先验。在来自五个基准的5000个样本外图像中，我们的框架在合成、伪装和文档伪造任务中表现优异（96.44% GenImage，97.73% Chameleon，91.84% Doc），而在压缩人脸操控（FF++ ROC-AUC = 0.5913）上则显示出明显的界限。
cs.CV / 233 / 2609.33687
Resource-Aware Parameter-Efficient Model Adaptation for Onboard High-Dimensional Data
资源感知的参数高效模型适应用于在轨高维数据
Abstract
Onboard satellite models often require frequent updates, but the weights adapted to earlier data distributions can quickly become outdated. However, updating large-scale model parameters in orbit presents significant challenges due to the limited uplink bandwidth of Low Earth Orbit (LEO) satellite systems, particularly for hyperspectral satellite imagery, where high-dimensional spectral-spatial inputs lead to increased model size and update costs. Existing full fine-tuning methods are thus expensive to retrain and difficult to deploy under strict communication constraints. To address this challenge, we propose NE-LoRA, a parameter-efficient adaptation framework for bandwidth-constrained onboard hyperspectral model updates. NE-LoRA combines a primary low-rank branch with a nonlinear auxiliary branch to capture both global update trends and complex spectral-spatial variations. Additionally, we introduce a differentiated training strategy for multi-matrix adapters, motivated by the asymmetric initialization and gradient dynamics of different adapter matrices. Experiments on four hyperspectral datasets and three representative backbone models demonstrate that NE-LoRA consistently outperforms LoRA-based baselines and remains competitive with, and in several cases superior to, full fine-tuning. Across the evaluated settings, NE-LoRA updates only a small fraction of the total parameters on average while preserving low deployment overhead, offering a favorable accuracy-communication trade-off for onboard hyperspectral adaptation.
Chinese Translation
在轨卫星模型通常需要频繁更新，但适应于早期数据分布的权重很快会变得过时。然而，在轨更新大规模模型参数面临重大挑战，因为低地球轨道（LEO）卫星系统的上行带宽有限，尤其是对于高光谱卫星图像，高维光谱-空间输入导致模型规模和更新成本增加。因此，现有的完全微调方法在重新训练时成本高昂，并且在严格的通信限制下难以部署。为了解决这一挑战，我们提出了NE-LoRA，一种用于带宽受限的在轨高光谱模型更新的参数高效适应框架。NE-LoRA结合了一个主要的低秩分支和一个非线性辅助分支，以捕捉全球更新趋势和复杂的光谱-空间变化。此外，我们引入了一种针对多矩阵适配器的差异化训练策略，灵感来自于不同适配器矩阵的不对称初始化和梯度动态。在四个高光谱数据集和三个代表性主干模型上的实验表明，NE-LoRA始终优于基于LoRA的基线，并在多个情况下与完全微调相比保持竞争力，甚至表现更佳。在评估的设置中，NE-LoRA平均仅更新总参数的一小部分，同时保持低部署开销，为在轨高光谱适应提供了有利的准确性与通信的权衡。
cs.CV / 234 / 2609.33694
Seeing and Solving Are Not Enough for Vision-Language Models
视觉-语言模型仅有观察和解决问题是不够的
Abstract
Vision-language models (VLMs) answer visual questions by combining visual information extraction with downstream problem solving. We investigate a fundamental question: Does an incorrect answer necessarily reflect a failure in visual extraction or problem solving? A model may succeed at both abilities when tested separately yet still fail on the original multimodal question, a distinction that overall answer accuracy cannot reveal. To study this, we perform a question-level empirical analysis across multiple VLMs and visual domains. We define an exactly scorable task state (i.e., the visual information sufficient to solve a question) and use it to test whether the same model can extract the required state, solve the question from the ground-truth state, and answer the original multimodal question. We find that composition failures, where extraction and solving both succeed but direct answering fails, account for 17.7% to 75.6% of direct-answering errors across multiple VLMs and datasets. To address this failure mode, we introduce a simple yet effective method, termed State Realization Tuning (SRT). SRT fine-tunes LoRA adapters attached to the language-model layers while keeping the pretrained VLM weights frozen. It trains the model to output the ground-truth task state before the final answer in a single autoregressive response. SRT improves over standard supervised fine-tuning by 1.7 to 14.1 percentage points and repairs 92.5% to 98.1% of diagnosed composition failures. A single LoRA adapter trained with SRT also improves performance across substantially different task-state structures. Our work shows that having both visual extraction and problem-solving capabilities does not guarantee correct multimodal answering. Requiring the model to first output the visual information needed to solve the question can help bridge this gap.
Chinese Translation
视觉-语言模型（VLMs）通过结合视觉信息提取与下游问题解决来回答视觉问题。我们探讨一个基本问题：错误答案是否必然反映出视觉提取或问题解决的失败？一个模型在单独测试时可能在这两种能力上都表现良好，但在原始的多模态问题上仍可能失败，这一点是整体答案准确性无法揭示的。为此，我们在多个 VLMs 和视觉领域进行了一项问题级的实证分析。我们定义了一个可精确评分的任务状态（即解决问题所需的视觉信息），并利用它来测试同一模型是否能够提取所需状态，从真实状态中解决问题，并回答原始的多模态问题。我们发现，组合失败，即提取和解决均成功但直接回答失败，导致了多个 VLMs 和数据集中17.7%到75.6%的直接回答错误。为了解决这种失败模式，我们提出了一种简单而有效的方法，称为状态实现调优（State Realization Tuning, SRT）。SRT 在保持预训练 VLM 权重不变的情况下，对附加在语言模型层上的 LoRA 适配器进行微调。它训练模型在最终答案之前以单一自回归响应输出真实的任务状态。SRT 在标准监督微调上提高了1.7到14.1个百分点，并修复了92.5%到98.1%的诊断组合失败。使用 SRT 训练的单个 LoRA 适配器在结构差异显著的任务状态上也提高了性能。我们的研究表明，仅具备视觉提取和问题解决能力并不能保证正确的多模态回答。要求模型首先输出解决问题所需的视觉信息可以帮助弥补这一差距。
cs.CV / 235 / 2609.33701
Prompt-Anchored Residual Adaptation for Biomedical Vision-Language Models
基于提示锚的残差适应用于生物医学视觉-语言模型
Abstract
Pretrained biomedical vision-language models achieve strong zero-shot performance in biomedical image classification. However, downstream biomedical classification often depends on subtle visual differences between classes that may not be fully captured by pretrained representations. Few-shot adaptation addresses this mismatch by optimizing a task-specific predictor on a small labeled support set. Because the selected examples capture only part of the visual variation within the target classes, the adapted predictions can depend strongly on their composition. We propose Prompt-Anchored Residual Adaptation (PARA), which retains the frozen prompt prediction as a support-invariant semantic anchor and incorporates a visual prediction learned from the support set through an anchor-relative residual. The residual step is computed in a closed form from frozen support embeddings using anchor discrepancy and support agreement. Support-set dependence also limits evaluation: comparisons are fair within a shared draw but remain conditional on its composition. To obtain more reliable comparisons, we introduce a repeated-support protocol that separates support-selection variation from optimization randomness and reports both average and worst-20% performance. PARA achieves state-of-the-art performance in both few-shot classification and base-to-novel generalization.
Chinese Translation
预训练的生物医学视觉-语言模型在生物医学图像分类中实现了强大的零-shot性能。然而，下游生物医学分类通常依赖于类别之间微妙的视觉差异，而这些差异可能未被预训练表示充分捕捉。少样本适应通过在小型标记支持集上优化特定任务的预测器来解决这种不匹配。由于所选示例仅捕捉目标类别内视觉变化的一部分，适应后的预测可能会强烈依赖于其组成。我们提出了基于提示锚的残差适应（Prompt-Anchored Residual Adaptation, PARA），该方法保留冻结的提示预测作为支持不变的语义锚，并通过锚相关残差结合从支持集学习的视觉预测。残差步骤是通过使用锚差异和支持一致性从冻结的支持嵌入中以封闭形式计算的。支持集的依赖性也限制了评估：在共享样本内的比较是公平的，但仍然依赖于其组成。为了获得更可靠的比较，我们引入了一种重复支持协议，将支持选择的变化与优化随机性分开，并报告平均性能和最差的20%性能。PARA在少样本分类和基础到新颖泛化方面实现了最先进的性能。
cs.CV / 236 / 2609.33716
Revisiting Diffusion Fine-Tuning for Unsupervised Domain Adaptation
重新审视基于扩散的无监督领域适应的微调
Abstract
Diffusion-based unsupervised domain adaptation (UDA) improves cross-domain transfer by generating target-specific synthetic data for downstream adaptation. Existing methods are largely designed for single-target adaptation: when a model trained on one labeled source domain must be adapted to multiple unlabeled target domains, they typically require separate diffusion fine-tuning for each source--target pair, causing training, storage, and deployment costs to grow with the number of targets. In this paper, we study multi-target data generation for diffusion-based UDA, where a single source-guided diffusion fine-tuning process is reused to generate target-specific synthetic data for multiple target domains. We propose MUSE (Multi-target UDA-oriented Synthesis with Efficient diffusion fine-tuning), a decoupled adaptation framework that separates source-supervised semantic adaptation from target-specific style adaptation. MUSE uses a shared semantic branch updated by labeled source data and target-private style branches specialized to individual target domains, enabling target-specific generation while avoiding repeated source-guided fine-tuning for each target. Experiments on standard UDA benchmarks show that MUSE achieves a stronger accuracy--efficiency trade-off than repeated per-target diffusion adaptation, reducing diffusion fine-tuning cost while improving average target-domain accuracy. The project page is available at https://xuanqi99.github.io/MUSE/.
Chinese Translation
基于扩散的无监督领域适应（UDA）通过生成针对目标的合成数据来改善跨领域迁移，以便于下游适应。现有方法主要针对单一目标适应设计：当一个在某个标记源领域上训练的模型需要适应多个未标记的目标领域时，通常需要对每个源-目标对进行单独的扩散微调，这导致随着目标数量的增加，训练、存储和部署成本上升。本文研究了基于扩散的UDA的多目标数据生成，其中一个单一的源引导扩散微调过程被重用，以为多个目标领域生成特定于目标的合成数据。我们提出了MUSE（多目标UDA导向的高效扩散微调合成），这是一个解耦的适应框架，将源监督的语义适应与特定于目标的风格适应分开。MUSE使用由标记源数据更新的共享语义分支和专门针对各个目标领域的目标私有风格分支，从而实现特定于目标的生成，同时避免对每个目标进行重复的源引导微调。在标准UDA基准上的实验表明，MUSE在准确性与效率的权衡上优于重复的每目标扩散适应，降低了扩散微调成本，同时提高了平均目标领域的准确性。项目页面可访问：https://xuanqi99.github.io/MUSE/
cs.CV / 237 / 2609.33723
GeoShrink: Accelerating Diffusion Transformers with Two Lines of Code
GeoShrink：用两行代码加速扩散变换器
Abstract
Diffusion transformers incur substantial inference cost through repeated model evaluations along a sampling trajectory. We introduce GeoShrink, a training-free acceleration method that retains the original solver grid while evaluating the model only at a prescribed set of anchors. At skipped stages, GeoShrink predicts the solver-facing output by adding a geometrically retained fraction of the latest observed innovation to the most recent exact output. We derive this rule from chordal tangent transport and round-trip line projection, and establish a geometric anchor-spacing principle that minimizes the largest adjacent gap expansion under fixed coverage and first span. The analysis characterizes the geometric closure and propagation of prediction errors without assuming access to future model outputs. Experiments cover image, video, motion, and audio generation, together with adapted 3D backends. At approximately $5\times$ acceleration, GeoShrink improves FLUX PSNR by 3.10 dB over the strongest listed baseline. On HunyuanVideo, it achieves a reported $4.99\times$ speedup and improves ChronoMagic-Bench-150 PSNR by 5.44 dB over the strongest listed fidelity baseline. Comparisons at fixed evaluation budgets further show substantial gains on motion, audio, music, and 3D generation.
Chinese Translation
扩散变换器在采样轨迹上通过重复模型评估产生了可观的推理成本。我们提出了GeoShrink，这是一种无训练加速方法，它在仅在规定的锚点集上评估模型的同时保留原始求解器网格。在跳过的阶段，GeoShrink通过将最新观察到的创新的几何保留部分添加到最近的精确输出中，来预测面向求解器的输出。我们从弦切线传输和往返线投影中推导出这一规则，并建立了一个几何锚点间距原则，以在固定覆盖和首次跨度下最小化最大的相邻间隙扩展。该分析表征了预测误差的几何闭合和传播，而不假设访问未来模型输出。实验涵盖了图像、视频、运动和音频生成，以及适配的3D后端。在约$5 imes$的加速下，GeoShrink使FLUX PSNR比最强基线提高了3.10 dB。在HunyuanVideo上，它实现了报告的$4.99 imes$加速，并使ChronoMagic-Bench-150 PSNR比最强保真基线提高了5.44 dB。在固定评估预算下的比较进一步显示了在运动、音频、音乐和3D生成方面的显著提升。
cs.CV / 238 / 2609.33735
Constrained Edit Fields for Training-Free Flow Editing
无训练流编辑的约束编辑场
Abstract
Text-guided image editing aims to perform a desired edit while preserving source content unrelated to it. Pretrained rectified-flow models enable training-free editing of real images through modifications to their sampling trajectories. However, responses at locations unrelated to the desired edit can still accumulate along the editing trajectory and become visible in the final result. To overcome this, we propose Constrained Edit Fields (CEF), which assigns each spatial location a continuous edit responsibility that quantifies its relevance to the desired edit. CEF estimates edit responsibility directly from the source image when the relevant content is present. For edits whose target content is absent from the source, CEF first generates an unconstrained proposal to reveal its realized spatial support and then estimates responsibility from that proposal. At each editing step, CEF decomposes the base edit field into prompt-induced and trajectory-induced components, enabling edit responsibility to preserve instruction-relevant updates while suppressing unintended trajectory-induced changes. Evaluated on all 700 PIE-Bench examples, CEF achieves state-of-the-art Structure Distance, background LPIPS, and background MSE with both Stable Diffusion 3.5 Medium and FLUX, while retaining competitive instruction alignment. On Stable Diffusion 3.5 Medium, it reduces these metrics over the previous best results by 10.2%, 21.2%, and 48.0%, respectively.
Chinese Translation
文本引导的图像编辑旨在在保留与所需编辑无关的源内容的同时，执行所需的编辑。预训练的校正流模型通过对其采样轨迹的修改，使得对真实图像的无训练编辑成为可能。然而，与所需编辑无关的位置的响应仍然可能在编辑轨迹上累积，并在最终结果中变得可见。为了解决这个问题，我们提出了约束编辑场（Constrained Edit Fields, CEF），它为每个空间位置分配一个连续的编辑责任，以量化其与所需编辑的相关性。当相关内容存在于源图像中时，CEF直接从源图像中估计编辑责任。对于目标内容在源图像中缺失的编辑，CEF首先生成一个无约束的提案，以揭示其实现的空间支持，然后从该提案中估计责任。在每个编辑步骤中，CEF将基础编辑场分解为由提示引起的和由轨迹引起的组件，使得编辑责任能够保留与指令相关的更新，同时抑制意外的轨迹引起的变化。在对700个PIE-Bench示例的评估中，CEF在使用Stable Diffusion 3.5 Medium和FLUX时，达到了最先进的结构距离、背景LPIPS和背景均方误差，同时保持了竞争性的指令对齐。在Stable Diffusion 3.5 Medium上，它分别将这些指标较之前的最佳结果降低了10.2%、21.2%和48.0%。
cs.CV / 239 / 2609.33758
ENet-GP: Unified Document Image Restoration
ENet-GP: 统一文档图像修复
Abstract
Reliable document digitization in uncontrolled capture settings is challenging because real images exhibit multiple interacting degradations rather than a single isolated distortion. Documents thus captured are affected simultaneously by geometric distortions, like page warping, as well as photometric degradations such as non-uniform illumination, and blurring. However, most existing approaches address these factors independently and are evaluated on benchmarks containing only one distortion type, limiting their real-world applicability. We introduce GutenDoc, a large-scale dataset of high-resolution dense-text documents with physically grounded compound degradations. Using physics-based rendering, our dataset jointly models geometric warping and diverse photometric effects, enabling systematic evaluation under realistic capture conditions. We further propose a unified restoration framework that jointly corrects geometric and photometric distortions within a single-network and single-training setup, without the need for degradation-specific retraining or sequential inference passes. Extensive experiments show that our method remains competitive on established single-distortion benchmarks while substantially improving robustness under compound degradations, providing a practical solution for real-world document digitization.
Chinese Translation
在不受控的捕捉环境中，可靠的文档数字化面临挑战，因为真实图像表现出多种相互作用的退化，而不是单一孤立的失真。因此，捕获的文档同时受到几何失真（如页面变形）以及光度退化（如不均匀照明和模糊）的影响。然而，大多数现有方法独立处理这些因素，并在仅包含一种失真类型的基准上进行评估，这限制了它们在现实世界中的适用性。我们引入了GutenDoc，这是一个大规模的高分辨率密集文本文档数据集，具有物理基础的复合退化。通过基于物理的渲染，我们的数据集联合建模了几何变形和多样的光度效应，使得在现实捕捉条件下进行系统评估成为可能。我们进一步提出了一种统一的修复框架，该框架在单一网络和单一训练设置中联合校正几何和光度失真，无需针对特定退化的重新训练或顺序推理过程。大量实验表明，我们的方法在已建立的单一失真基准上保持竞争力，同时在复合退化下显著提高了鲁棒性，为现实世界的文档数字化提供了实用的解决方案。
cs.CV / 240 / 2609.33769
M3-Score: Fidelity, Memorization and Coverage as Separate Axes for Evaluating Generative Radiology Image Models
M3-Score：将保真度、记忆化和覆盖率作为评估生成放射学图像模型的独立维度
Abstract
Quantitative evaluation of generative models for radiology remains challenging. Clinically relevant structures are often small and infrequent, feature spaces learned from natural images may represent them poorly, and a single summary score cannot distinguish limited fidelity from limited diversity. This study proposes the Medical Multi-axis Maximum Mean Discrepancy score (M3-Score), an evaluation framework based on RadioDINO-s16, a frozen vision transformer pretrained on radiology images. M3-Score reports three complementary axes computed at pre-specified encoder depths: \emph{fidelity}, measured by an unbiased multi-bandwidth radial basis function (RBF) MMD$^2$ at the final block; \emph{memorization}, measured by nearest-neighbor distances at 75\% depth; and \emph{coverage}, defined as the fraction of real images with a generated neighbor within their $k$-nearest-neighbor radius at 33\% depth. Reference sets are sampled across subjects to limit the influence of correlated slices. On BraTS brain MRI, the fidelity axis ordered five comparison sets of increasing severity (Spearman $\rho = 1.00$), and a subject-disjoint real set yielded $\mathrm{MMD}^2 = 0$ (permutation $p = 1$). An unconditional denoising diffusion probabilistic model achieved $\mathrm{MMD}^2 = 0.073$ (95\% confidence interval $[0.071, 0.080]$) but covered only 38\% of the real distribution. Under progressive mode dropping, $k$-NN manifold recall increased at all twelve encoder blocks, whereas the proposed coverage estimator decreased monotonically ($\rho = -1.00$). RadioDINO-s16 features separated real brain MRI from generated samples with a ROC-AUC of 0.819, compared with 0.555 for InceptionV3 and 0.582 for CLIP. Across a twentyfold range of sample sizes, the mean M3 value varied by a factor of 1.05, compared with 2.52 for the Fr\'{e}chet Inception Distance.
Chinese Translation
对放射学生成模型的定量评估仍然具有挑战性。临床相关结构通常较小且不常见，从自然图像中学习的特征空间可能对其表示不佳，单一的总结评分无法区分有限的保真度与有限的多样性。本研究提出了医学多轴最大均值差异评分（M3-Score），这是一个基于RadioDINO-s16的评估框架，后者是一个在放射学图像上预训练的冻结视觉变换器。M3-Score报告在预先指定的编码器深度下计算的三个互补维度： extit{保真度}，通过最终块的无偏多带宽径向基函数（RBF）MMD$^2$进行测量； extit{记忆化}，通过75 ext{%}深度的最近邻距离进行测量；以及 extit{覆盖率}，定义为在33 ext{%}深度下，具有生成邻居的真实图像占其$k$-最近邻半径的比例。参考集在不同受试者之间抽样，以限制相关切片的影响。在BraTS脑MRI上，保真度维度将五个比较集按严重程度排序（Spearman $
ho = 1.00$），而一个受试者不重叠的真实集得到了$ ext{MMD}^2 = 0$（置换$p = 1$）。一个无条件去噪扩散概率模型达到了$ ext{MMD}^2 = 0.073$（95 ext{%}置信区间$[0.071, 0.080]$），但仅覆盖了真实分布的38 ext{%}。在逐步模式丢弃下，$k$-NN流形召回在所有十二个编码器块中均有所增加，而所提议的覆盖率估计器则单调下降（$
ho = -1.00$）。与InceptionV3的0.555和CLIP的0.582相比，RadioDINO-s16特征将真实脑MRI与生成样本分开，ROC-AUC为0.819。在样本大小的二十倍范围内，平均M3值变化因子为1.05，而Fréchet Inception Distance为2.52。
cs.CV / 241 / 2609.33811
Eyes on the Road: A Naturalistic Comparison of MTW Rider Gaze in Urban Indian Traffic
关注道路：城市印度交通中摩托车骑行者注视行为的自然比较
Abstract
Motorized two-wheelers (MTW) dominate Indian roads but remain underrepresented in driver behavior research. This study presents the first large-scale analysis of MTW driver gaze behavior in naturalistic, heterogeneous urban traffic, using the \textit{myEye2Wheeler} dataset. A semantic segmentation pipeline (YOLOv11 + SAM2) was used to extract object-level gaze metrics under two attention modes: direct gaze (foveal overlap) and central vision (parafoveal monitoring). Results reveal a functional division: central vision supports broad monitoring, while direct gaze enables brief, selective sampling. Novice riders exhibit road-anchored scanning, returning to the road between object fixations, while experienced riders form longer chains of attention across multiple objects. The findings suggest that experience primarily refines temporal rhythm rather than altering allocation strategy and reduces object-class effects in gaze patterns. These findings offer new insight into MTW attention structures and inform future work on behavior modeling and safety systems.
Chinese Translation
机动车两轮车（MTW）在印度道路上占据主导地位，但在驾驶行为研究中仍然被低估。本研究首次对自然环境中异质城市交通中的MTW驾驶者注视行为进行了大规模分析，使用了 extit{myEye2Wheeler}数据集。采用语义分割管道（YOLOv11 + SAM2）提取了在两种注意模式下的对象级注视指标：直接注视（中央重叠）和中央视觉（旁中央监测）。结果揭示了功能上的划分：中央视觉支持广泛监测，而直接注视则使得短暂的选择性采样成为可能。新手骑行者表现出以道路为锚点的扫描行为，在对象注视之间回归道路，而经验丰富的骑行者则在多个对象之间形成更长的注意链。这些发现表明，经验主要是精炼时间节奏，而不是改变分配策略，并减少了注视模式中对象类别的影响。这些发现为MTW注意结构提供了新的见解，并为未来的行为建模和安全系统研究提供了参考。
cs.CV / 242 / 2609.33818
Augmenting Visual Anomaly Detection with Automated Interpretability
通过自动可解释性增强视觉异常检测
Abstract
Visual anomaly detectors identify deviations from known-normal data, but their anomaly signals may mix evidence of actual anomalies with benign visual variation. We investigate whether automated interpretability can augment visual anomaly detectors by identifying and intervening on different components of this signal. We decompose PatchCore nearest-normal residuals into sparse features using Sparse Autoencoders (SAEs), and provide high-activation and contrastive non-active examples to a Multimodal LLM, which describes each feature and labels it as anomaly, distractor, or uncertain. These labels guide interventions in the SAE hidden representation, where distractor features are suppressed and anomaly features amplified. The edited representation is then used to reconstruct patch embeddings, which are rescored with PatchCore. Across 40 categories from four benchmarks, applying both interventions jointly improves macro-average image-level AUROC from 0.8724 to 0.8857 on source data and from 0.8066 to 0.8210 under synthetic corruptions. On three additional RobustAD categories with real acquisition shifts, the same interventions improve AUROC from 0.8745 to 0.9056 on source data and from 0.6069 to 0.6599 under real acquisition shifts. Finally, individual feature interventions across all 43 categories show that the MLLM labels are aligned in aggregate with how features differently affect normal and anomalous images.
Chinese Translation
视觉异常检测器识别与已知正常数据的偏差，但它们的异常信号可能将实际异常的证据与良性视觉变化混合在一起。我们研究了自动可解释性是否能够通过识别和干预这一信号的不同组成部分来增强视觉异常检测器。我们使用稀疏自编码器（Sparse Autoencoders, SAEs）将 PatchCore 最近邻残差分解为稀疏特征，并向多模态大语言模型（Multimodal LLM）提供高激活和对比的非活动示例，该模型描述每个特征并将其标记为异常、干扰或不确定。这些标签指导对 SAE 隐藏表示的干预，其中干扰特征被抑制，异常特征被放大。编辑后的表示随后用于重建补丁嵌入，并使用 PatchCore 重新评分。在四个基准的 40 个类别中，同时应用这两种干预将源数据的宏平均图像级 AUROC 从 0.8724 提高到 0.8857，并在合成损坏下从 0.8066 提高到 0.8210。在三个额外的 RobustAD 类别中，针对真实获取偏移的相同干预将源数据的 AUROC 从 0.8745 提高到 0.9056，并在真实获取偏移下从 0.6069 提高到 0.6599。最后，针对所有 43 个类别的个体特征干预显示，MLLM 标签在整体上与特征对正常和异常图像的不同影响一致。
cs.CV / 243 / 2609.33833
One Attack to Fool Them All: Highly Transferable Black-Box Adversarial Attacks on Frontier MLLMs
一招制胜：对前沿多模态大型语言模型的高度可转移黑箱对抗攻击
Abstract
Adversarial attacks have long posed a fundamental threat to machine learning systems. As multimodal large language models (MLLMs) rapidly evolve and become widely deployed, assessing their vulnerability to such attacks is essential for their safe use. In this work, we investigate whether a single adversarial image can consistently mislead diverse frontier MLLMs in black-box settings. We propose O-Attack, a highly transferable black-box attack framework. This framework builds on our insight that surrogate models contain a broad, high-level, cross-modally aligned semantic space. This space extends beyond final-layer outputs and provides multiple semantically consistent representations that remain underexploited by existing attacks. Within this space, O-Attack anchors aligned representations, progressively broadens semantic conditions, and optimizes perturbations through semantic consensus to promote consistent target alignment. By fully exploiting this space with the same surrogate models as M-Attack, O-Attack raises attack success rates on GPT-5.4 (29.1% to 77.2%), Claude-4.6 (42.8% to 81.6%), and Gemini-3.1 (38.2% to 80.9%). Extensive experiments across 24 MLLMs show that O-Attack outperforms six state-of-the-art methods in black-box transferability, with consistent effectiveness across prompts and improved efficiency and imperceptibility. This work exposes the practical safety risks posed by black-box adversarial attacks against frontier MLLMs, underscoring the need for more rigorous robustness evaluation and more effective defenses.
Chinese Translation
对抗攻击长期以来对机器学习系统构成了根本威胁。随着多模态大型语言模型（MLLMs）的快速发展和广泛部署，评估它们对这些攻击的脆弱性对于安全使用至关重要。在本研究中，我们探讨了单一对抗图像是否能够在黑箱环境中持续误导多样化的前沿 MLLMs。我们提出了 O-Attack，一个高度可转移的黑箱攻击框架。该框架基于我们的洞察，即替代模型包含一个广泛的、高层次的、跨模态对齐的语义空间。这个空间超越了最终层输出，提供了多个语义一致的表示，这些表示在现有攻击中尚未得到充分利用。在这个空间内，O-Attack 固定对齐的表示，逐步扩展语义条件，并通过语义共识优化扰动，以促进一致的目标对齐。通过充分利用与 M-Attack 相同的替代模型，O-Attack 提高了对 GPT-5.4（从 29.1% 提升至 77.2%）、Claude-4.6（从 42.8% 提升至 81.6%）和 Gemini-3.1（从 38.2% 提升至 80.9%）的攻击成功率。对 24 个 MLLMs 的广泛实验表明，O-Attack 在黑箱可转移性方面优于六种最先进的方法，在提示上表现出一致的有效性，并且提高了效率和不可察觉性。本研究揭示了黑箱对抗攻击对前沿 MLLMs 造成的实际安全风险，强调了对更严格的鲁棒性评估和更有效防御的需求。
cs.CV / 244 / 2609.33834
CLIMB-flow: Coupled Linear Inverse posterior sampling via Multiscale-Based flow
CLIMB-flow：基于多尺度流的耦合线性逆后验采样
Abstract
Diffusion models are now widely used in Bayesian inverse problems in imaging as priors, where latent diffusion models are often used for larger scale problems to keep the computational complexity and model-size manageable. Unfortunately, the auto-encoder based compression results in loss of spatial detail. In addition, the optimization is converted to a non-linear problem. In this paper, we introduce a posterior sampling algorithm customized for the pyramidal/cascaded architecture, which relies on a coarse to fine hierarchical strategy to generate images in the pixel domain. We present CLIMB-Flow which alternates between three steps: an end-point estimation from the current coarse and noisy image, data-consistent update of the clean image, and re-noising it back to the level the network expects. Together these steps sample the posterior at that scale using an approximate Gibbs sampling from two conditional distributions. Experiments on ImageNet, CelebA, AFHQ and fastMRI span inpainting, deblurring, super-resolution and accelerated MRI, with PSNR gains of 1.37-7.66 dB over the strongest competing method on CelebA and pixel-domain reconstruction up to 512x512.
Chinese Translation
扩散模型现已广泛应用于成像中的贝叶斯逆问题作为先验，其中潜在的扩散模型通常用于较大规模的问题，以保持计算复杂性和模型规模的可管理性。不幸的是，基于自编码器的压缩导致空间细节的丧失。此外，优化被转换为非线性问题。在本文中，我们介绍了一种针对金字塔/级联架构定制的后验采样算法，该算法依赖于从粗到细的层次策略在像素域生成图像。我们提出了CLIMB-Flow，它在三个步骤之间交替进行：从当前粗糙且嘈杂的图像进行终点估计、对干净图像进行数据一致性更新，以及将其重新加噪至网络期望的水平。这些步骤共同在该尺度上使用来自两个条件分布的近似吉布斯采样来采样后验。在ImageNet、CelebA、AFHQ和fastMRI上的实验涵盖了图像修复、去模糊、超分辨率和加速MRI，在CelebA上相较于最强竞争方法获得了1.37-7.66 dB的PSNR增益，并且在像素域重建达到了512x512的分辨率。
cs.CV / 245 / 2609.33854
ReDrive: Shaping Representations with World Modeling for End-to-End Driving
ReDrive：通过世界建模塑造表征以实现端到端驾驶
Abstract
Driving policies require capabilities of scene understanding and future evolution prediction. To achieve this goal, current end-to-end models typically construct complex perception-planning pipelines or introduce world models that explicitly predict future states, resulting in a complex system architecture. Inspired by the transferability of general-purpose visual representations, we argue that combining sufficiently strong visual representations with representation world modeling can support effective planning without relying on complex inference-time auxiliary modules. Based on this insight, we present ReDrive, an end-to-end driving framework that strengthens planning-oriented visual features via future representation prediction. To achieve this, ReDrive adopts a three-stage training pipeline consisting of driving video pretraining, joint world-modeling and planning training, and planner adaptation. This yields a strong planning-oriented representation and a high-performance planner, while requiring neither auxiliary perception modules nor future prediction at inference time. Experiments on NAVSIM demonstrate strong performance, achieving 91.0 PDMS on NAVSIM v1 and 90.8 EPDMS on NAVSIM v2. These results show that shaping representations with world modeling is sufficient to enable high-performance end-to-end planning while retaining a simple encoder-planner inference pipeline.
Chinese Translation
驾驶策略需要具备场景理解和未来演变预测的能力。为实现这一目标，目前的端到端模型通常构建复杂的感知-规划管道或引入明确预测未来状态的世界模型，从而导致复杂的系统架构。受到通用视觉表征可迁移性的启发，我们认为将足够强大的视觉表征与表征世界建模相结合，可以支持有效的规划，而无需依赖复杂的推理时辅助模块。基于这一见解，我们提出了ReDrive，一个端到端驾驶框架，通过未来表征预测增强规划导向的视觉特征。为此，ReDrive采用了一个三阶段的训练管道，包括驾驶视频预训练、联合世界建模与规划训练，以及规划器适应。这产生了一个强大的规划导向表征和高性能的规划器，同时在推理时既不需要辅助感知模块，也不需要未来预测。在NAVSIM上的实验表明了强大的性能，在NAVSIM v1上达到了91.0 PDMS，在NAVSIM v2上达到了90.8 EPDMS。这些结果表明，通过世界建模塑造表征足以实现高性能的端到端规划，同时保持简单的编码器-规划器推理管道。
cs.CV / 246 / 2609.33855
Program-Verified Self-Evolution for Vision-Language Models
程序验证的自我进化视觉-语言模型
Abstract
Self-evolving vision-language models train on questions they generate from unlabeled images. Since these questions have no gold answers, prior methods label them by majority vote over sampled answers or by a model judge. In a human evaluation, we find that 24\% of majority-vote labels and 18\% of model-judge labels produced during self-evolution are wrong. To address this problem, we present Verifiable QA Generation for Self-Evolving Models (VQS), which changes how the model judges answers. Instead of voting on an answer, the model parses each image into a structured record, such as a scene graph, a chart table, or a diagram graph. Fixed programs then write a question from the record and compute its answer. The model still acts as a visual checker, but it only confirms the individual facts the program reads, one short claim at a time. These claim-level checks select the parser's training targets, so the parser also improves without labels. Human raters find 94\% of VQS answers correct, against 76\% for majority voting. Across ten benchmarks, VQS improves Qwen3-VL by up to 3.18 points at the 2B, 4B, and 8B scales and outperforms the strongest self-evolving baseline at each. Gains keep growing over three training rounds, reaching 3.84 points at 2B. Code is released at https://github.com/ahmedheakl/VQS
Chinese Translation
自我进化的视觉-语言模型通过从未标记图像中生成的问题进行训练。由于这些问题没有标准答案，之前的方法通过对采样答案进行多数投票或由模型评判来标记它们。在人类评估中，我们发现自我进化过程中产生的24\%的多数投票标签和18\%的模型评判标签是错误的。为了解决这个问题，我们提出了可验证的自我进化模型问答生成（Verifiable QA Generation for Self-Evolving Models, VQS），该方法改变了模型评判答案的方式。模型不再对答案进行投票，而是将每个图像解析为结构化记录，例如场景图、表格或图示。固定程序随后从记录中生成问题并计算其答案。模型仍然充当视觉检查器，但它仅确认程序读取的每个事实，每次一个简短的声明。这些声明级检查选择了解析器的训练目标，因此解析器也在没有标签的情况下得到改进。人类评审员发现94\%的VQS答案是正确的，而多数投票的正确率为76\%。在十个基准测试中，VQS在2B、4B和8B规模上将Qwen3-VL的性能提高了最多3.18分，并在每个基准上超越了最强的自我进化基线。经过三轮训练，性能提升持续增长，在2B规模上达到3.84分。代码已发布在 https://github.com/ahmedheakl/VQS
cs.CV / 247 / 2609.33895
Residual-Stream Burden Shapes Representation Learning in Diffusion Transformers
残差流负担塑造扩散变换器中的表征学习
Abstract
In diffusion-based generation, a neural network can be trained to predict the clean data, the noise, or the velocity from a noisy input. These prediction targets are interconvertible and describe the same generative process, yet plain Diffusion Transformers operating on large pixel patches succeed with clean prediction and fail with noise or velocity prediction. We argue that this asymmetry arises because noisy targets require the residual stream to preserve noise-dependent input variation through depth for the final readout, forcing subsequent layers to compute on noisy representations. A spectrally concentrated clean target imposes a lighter demand, leaving greater freedom to organize hidden representations for subsequent computation. We call this preservation requirement *residual-stream burden* and show how it shapes representation learning in Diffusion Transformers. Controlled experiments indicate that the exploitable structure is spectral concentration in patch space and that the bandwidth of the persistent residual state is a key resource for noisy prediction. We further show that this account is consistent with recent decoupled pixel-space architectures, whose diverse designs all reduce the residual-stream burden on the main pathway. To examine this understanding from a complementary direction, we expand and reorganize the residual-stream bandwidth directly, introducing Spatially Indexed Hyper-Connections (SiHC) that reach FID 1.71 on ImageNet $256^2$. Together, these results identify residual-stream burden as a mechanism through which prediction targets and architecture jointly shape representation learning in Diffusion Transformers.
Chinese Translation
在基于扩散的生成中，神经网络可以被训练来预测干净数据、噪声或从嘈杂输入中提取的速度。这些预测目标是可互换的，描述了相同的生成过程，然而，普通的扩散变换器在处理大像素块时在干净预测上表现良好，而在噪声或速度预测上则表现不佳。我们认为，这种不对称性产生的原因在于，嘈杂目标需要残差流在深度上保留依赖于噪声的输入变化，以便进行最终的读取，从而迫使后续层在嘈杂的表征上进行计算。频谱集中型的干净目标对模型的要求较轻，给予了更大的自由度来组织隐藏表征以进行后续计算。我们将这种保留要求称为*残差流负担*，并展示它如何塑造扩散变换器中的表征学习。控制实验表明，可利用的结构是在块空间中的频谱集中性，而持久残差状态的带宽是进行噪声预测的关键资源。我们进一步表明，这一解释与最近的解耦像素空间架构是一致的，这些多样化的设计都减少了主路径上的残差流负担。为了从互补的角度检验这一理解，我们直接扩展和重组了残差流带宽，引入了空间索引超连接（Spatially Indexed Hyper-Connections, SiHC），在 ImageNet $256^2$ 上达到了 FID 1.71。总之，这些结果将残差流负担识别为一个机制，通过它，预测目标和架构共同塑造了扩散变换器中的表征学习。
cs.CV / 248 / 2609.33928
Preserving DEG Rankings for Gene Discovery in Histology-Based Spatial Gene Expression Prediction
在基于组织学的空间基因表达预测中保留差异表达基因排名以促进基因发现
Abstract
Predicting spatial gene expression from histology images could scale spatial transcriptomics (ST) to image-only cohorts, but conventional histology-based ST prediction is trained and evaluated mainly by per-gene spatial-profile reconstruction. This objective is misaligned with a key downstream use of ST: differentially expressed gene (DEG) discovery, where genes are ranked for a biological or morphology-defined contrast by evidence of between-group expression differences. We formulate image-based differential expression ranking (IDER), which asks whether predicted expression profiles preserve the contrast-specific ranked gene list obtained from measured profiles. IDER compares gene rankings induced by differential-expression statistics, rather than raw expression magnitudes or per-gene spatial correlations. We further introduce a differentiable IDER objective that aligns these statistics across genes and can be trained with morphology-derived proxy contrasts without predefined biological group labels. Experiments on public ST datasets show improved DEG-ranking agreement and pathway-enrichment overlap over conventional reconstruction objectives, including morphology-derived and pathologist-annotated tissue-region evaluations.
Chinese Translation
从组织学图像中预测空间基因表达可以将空间转录组学（ST）扩展到仅基于图像的队列，但传统的基于组织学的ST预测主要通过每个基因的空间特征重建进行训练和评估。这一目标与ST的一个关键下游用途不一致：差异表达基因（DEG）发现，其中基因根据组间表达差异的证据被排名，以进行生物学或形态学定义的对比。我们提出了基于图像的差异表达排名（IDER），该方法询问预测的表达特征是否保留了从测量特征获得的对比特定排名基因列表。IDER比较由差异表达统计引起的基因排名，而不是原始表达量或每个基因的空间相关性。我们进一步引入了一种可微分的IDER目标，该目标在基因之间对齐这些统计数据，并且可以在没有预定义生物组标签的情况下使用形态学派生的代理对比进行训练。对公共ST数据集的实验表明，与传统的重建目标相比，DEG排名一致性和通路富集重叠得到了改善，包括形态学派生和病理学家注释的组织区域评估。
cs.CV / 249 / 2609.33935
Video, Ergo Genero: Unifying Video Tasks via Spatiotemporal Analogy
视频，因此生成：通过时空类比统一视频任务
Abstract
Adapting video models to new tasks typically requires dedicated data curation and fine-tuning. While visual analogy provides a training-free alternative by specifying tasks in-context, it remains restricted to the image domain. To explore whether analogy-based methods can unify diverse video tasks and generalize to out-of-distribution scenarios, we introduce ViGeo, a framework that extends visual in-context learning to the video domain via spatiotemporal canvas completion. Evaluated on a diverse task taxonomy with a strict train-test split, ViGeo generalizes to unseen video manipulations and zero-shot modalities (e.g., event cameras). Finally, we identify task internalization, where a query format associated with a pretrained task overrides the demonstration, and show that this shortcut can be removed with a small amount of task-unrelated data, highlighting the need to decorrelate prompt format from task identity.
Chinese Translation
将视频模型适应于新任务通常需要专门的数据整理和微调。虽然视觉类比通过在上下文中指定任务提供了一种无训练的替代方案，但它仍然局限于图像领域。为了探索基于类比的方法是否能够统一多样的视频任务并推广到分布外场景，我们提出了ViGeo，一个通过时空画布补全将视觉上下文学习扩展到视频领域的框架。在严格的训练-测试分割下对多样化任务分类进行评估，ViGeo能够推广到未见过的视频操作和零-shot模态（例如，事件摄像头）。最后，我们识别了任务内化现象，即与预训练任务相关的查询格式覆盖了演示，并展示了通过少量与任务无关的数据可以消除这一捷径，强调了将提示格式与任务身份去相关化的必要性。
cs.CV / 250 / 2609.33937
Test-Time Generalized Category Discovery
测试时的广义类别发现
Abstract
Test-Time Adaptation (TTA) and Generalized Category Discovery (GCD) are traditionally treated as disjoint problems: the former adapts models to domain shift assuming all test classes are known, while the latter discovers novel categories assuming labeled training data for known classes. However, real-world deployment rarely fits either setting. Motivated by this gap, we introduce Test-Time Generalized Category Discovery (TT-GCD), a unified and more realistic scenario where a vision-language model must adapt to distribution shifts, classify known categories using only textual supervision, and discover novel categories, all during test time and without access to labeled data. To address this challenging scenario, we propose PACT (Prototype Assignment for Category discovery at Test time), a fully unsupervised framework that casts known-class recognition and novel-class discovery via prototype assignment. PACT first re-aligns shifted visual features with the text-derived class representations of the VLM using confident zero-shot predictions. Known and novel categories are then both represented by prototypes in the visual embedding space, estimated from the unlabeled test stream, and each test image is assigned to the category whose prototype is most similar to its visual feature. Extensive experiments across corruption and domain-shift benchmarks demonstrate that PACT outperforms adapted state-of-the-art TTA and GCD methods, effectively bridging the gap between adaptation and discovery.
Chinese Translation
测试时适应（Test-Time Adaptation, TTA）和广义类别发现（Generalized Category Discovery, GCD）传统上被视为两个不相交的问题：前者在假设所有测试类别已知的情况下对模型进行领域适应，而后者在假设已知类别的标记训练数据的情况下发现新类别。然而，现实世界的部署很少符合这两种设定。基于这一差距，我们引入了测试时广义类别发现（Test-Time Generalized Category Discovery, TT-GCD），这是一个统一且更为现实的场景，其中视觉-语言模型必须适应分布变化，仅通过文本监督对已知类别进行分类，并在测试时发现新类别，而无需访问标记数据。为了解决这一挑战性场景，我们提出了PACT（测试时类别发现的原型分配，Prototype Assignment for Category discovery at Test time），这是一个完全无监督的框架，通过原型分配将已知类别识别与新类别发现结合起来。PACT首先使用自信的零样本预测重新对齐偏移的视觉特征与来自视觉-语言模型的文本派生类别表示。已知类别和新类别在视觉嵌入空间中均由原型表示，这些原型是从未标记的测试流中估计得出的，每个测试图像被分配给与其视觉特征最相似的原型所代表的类别。广泛的实验结果显示，PACT在各种损坏和领域转移基准测试中表现优于适应后的最先进TTA和GCD方法，有效地弥合了适应与发现之间的差距。
cs.CV / 251 / 2609.33969
Gaussian Splatting-based Volumetric Video Compression with Sparse 4D Anchors
基于高斯点云的体积视频压缩与稀疏4D锚点
Abstract
Immersive video communication requires photorealistic, render-efficient, and compact dynamic scene representations. 3D Gaussian Splatting (3DGS) offers a promising representation, but dynamic 3DGS remains difficult to compress due to dense primitives and spatiotemporal redundancy. Anchor-based formulations improve compactness with sparse scaffolds that share geometry and appearance across primitives. However, existing designs often rely on deforming a single canonical scaffold and condition each primitive on its associated anchor in isolation, limiting their ability to handle non-local dynamics and disocclusion while under-exploiting inter-anchor correlations, particularly in motion- or texture-dense regions. To address these limitations, we propose SAGA, a volumetric video codec built upon Sparse Anchor-assisted GAussian splatting representations. SAGA represents dynamic 3D scenes using hierarchically organized sparse 4D anchors, where coordinate-based INR decoders generate fine anchors and Gaussian primitives from inter-anchor interpolations, enabling compact parameter sharing across spatiotemporal structures. For long-range dependencies among unstructured anchors, we further introduce fixed-size memory slots with orthogonality-informed updates for accurate entropy-context modeling. Experiments show that SAGA achieves strong rate-distortion performance against GIFStream, with PSNR BD-rate reductions of 80.39% and 83.94% on Neu3D and MPEG MIV, respectively.
Chinese Translation
沉浸式视频通信需要具有照片真实感、渲染高效且紧凑的动态场景表示。3D高斯点云（3DGS）提供了一种有前景的表示方式，但由于密集的原语和时空冗余，动态3DGS的压缩仍然困难。基于锚点的公式通过稀疏支架提高了紧凑性，这些支架在原语之间共享几何和外观。然而，现有设计通常依赖于变形单一的规范支架，并孤立地将每个原语与其关联的锚点进行条件化，这限制了它们处理非局部动态和去遮挡的能力，同时未充分利用锚点之间的相关性，尤其是在运动或纹理密集的区域。为了解决这些局限性，我们提出了SAGA，一种基于稀疏锚点辅助高斯点云表示的体积视频编解码器。SAGA使用分层组织的稀疏4D锚点表示动态3D场景，其中基于坐标的INR解码器通过锚点之间的插值生成精细的锚点和高斯原语，从而实现跨时空结构的紧凑参数共享。针对非结构化锚点之间的长程依赖，我们进一步引入固定大小的内存槽，并通过正交性信息更新进行准确的熵上下文建模。实验表明，SAGA在与GIFStream的比较中实现了强大的速率失真性能，在Neu3D和MPEG MIV上分别实现了80.39%和83.94%的PSNR BD-rate降低。
cs.CV / 252 / 2609.33991
A Multi-Dataset Benchmark of YOLO-Based Weed Detection in Precision Agriculture
基于YOLO的精准农业杂草检测多数据集基准测试
Abstract
Weed detection is an important component of precision agriculture, enabling site-specific weed management and reducing unnecessary herbicide use. Although deep learning methods have achieved strong results for crop and weed detection, many studies rely on single-dataset evaluation, making it difficult to assess robustness across different agricultural domains. This paper presents a multi-dataset benchmark of deep object detectors for weed detection in precision agriculture, with a focused evaluation of YOLO26 models. We evaluate nano, small, and medium variants on seven public weed-detection datasets covering different crops, weed species, field conditions, acquisition setups, and annotation protocols. The models are compared in terms of detection accuracy, model complexity, inference latency, FPS, and model size. In addition to in-dataset evaluation, we investigate cross-domain generalization using a unified one-class weed setup and evaluate multi-source training using the combined training subsets from all datasets. The results show that YOLO26 achieves strong in-dataset performance, with YOLO26m obtaining the highest average accuracy and YOLO26s providing the best practical accuracy-efficiency trade-off. However, cross-domain performance decreases substantially, with YOLO26s dropping from an average in-domain mAP$_{50:95}$ of 0.603 to 0.148 in the off-domain setting. Multi-source training improves performance on several datasets, but does not fully eliminate domain shift. Overall, the benchmark highlights the importance of dataset diversity, domain similarity, and target-domain adaptation for robust weed detection in real-world precision agriculture applications.
Chinese Translation
杂草检测是精准农业的重要组成部分，能够实现针对特定地点的杂草管理并减少不必要的除草剂使用。尽管深度学习方法在作物和杂草检测方面取得了良好的效果，但许多研究依赖于单一数据集的评估，这使得在不同农业领域中评估其稳健性变得困难。本文提出了一项针对精准农业中杂草检测的深度目标检测器的多数据集基准测试，重点评估YOLO26模型。我们在七个公共杂草检测数据集上评估了纳米、小型和中型变体，这些数据集涵盖了不同的作物、杂草种类、田间条件、采集设置和标注协议。模型在检测准确性、模型复杂性、推理延迟、每秒帧数（FPS）和模型大小等方面进行了比较。除了数据集内评估外，我们还使用统一的一类杂草设置研究跨领域泛化，并评估使用所有数据集的组合训练子集进行多源训练。结果表明，YOLO26在数据集内表现良好，其中YOLO26m获得了最高的平均准确率，而YOLO26s提供了最佳的实际准确性与效率的权衡。然而，跨领域性能显著下降，YOLO26s在域外设置中的平均mAP$_{50:95}$从0.603降至0.148。多源训练在多个数据集上提高了性能，但并未完全消除领域转移。总体而言，该基准测试突显了数据集多样性、领域相似性和目标领域适应性在现实世界精准农业应用中实现稳健杂草检测的重要性。
cs.CV / 253 / 2609.33996
UnfoldCRF: Structured Mask Refinement with Image-Conditioned Latent Regions
UnfoldCRF：基于图像条件的潜在区域的结构化掩膜精炼
Abstract
Learned mask refiners improve segmentation accuracy, but it is hard to tell how much of the improvement comes from explicit structure rather than from extra capacity, and whether it holds up when the mask generator or its error distribution changes. UnfoldCRF treats refinement as inference in a conditional random field over pixel labels and latent region variables. Its energy has a corrected unary term, learned local pairwise interactions, and image-conditioned latent-region consistency, with a null state that lets a region with weak label agreement withdraw from the consistency term; inference unrolls damped mean-field updates on this one energy. To isolate the effect of structure, we compare against recurrent black-box refiners that read the same inputs and receive the same parameter budget, stage count, and supervision. On COD10K, UnfoldCRF beats the strongest matched control by 1.0 $F^\omega_\beta$ point, improves all four COD metrics, and lowers the fraction of images made worse from 11.7\% to 8.5\%. Under a train-once protocol over five datasets and several mask sources, the 2.6M-parameter variant gains 4.2 mean $\Delta$IoU against 2.0 for its control, and a variant built on frozen DINOv2 features matches the strongest foundation-model refiner with about a seventh of its resident parameters while staying ahead of its own control. On mask generators never seen in training, the gain is 2.0 $F^\omega_\beta$ points against 0.6 for the control. Zeroing individual messages shows where the corrections come from: the pairwise messages mostly fix boundaries, the region messages mostly fix non-boundary errors. Code and supporting materials will be publicly released.
Chinese Translation
学习的掩膜精炼器提高了分割精度，但很难判断这种改善有多少来自于显式结构而非额外的容量，以及当掩膜生成器或其错误分布发生变化时，这种改善是否依然有效。UnfoldCRF将精炼视为在像素标签和潜在区域变量上的条件随机场中的推断。其能量包含一个修正的单项式项、学习的局部成对交互以及基于图像的潜在区域一致性，其中的空状态允许标签一致性较弱的区域从一致性项中撤回；推断在这一能量上展开阻尼均场更新。为了隔离结构的影响，我们与读取相同输入并获得相同参数预算、阶段计数和监督的递归黑箱精炼器进行比较。在COD10K数据集上，UnfoldCRF比最强匹配控制提高了1.0 $F^eta_eta$点，改善了所有四个COD指标，并将变差图像的比例从11.7\%降低到8.5\%。在五个数据集和多个掩膜源的单次训练协议下，2.6M参数的变体相比其控制获得了4.2的平均$ riangle$IoU，而控制为2.0；基于冻结的DINOv2特征构建的变体在参数数量约为其基础模型的七分之一的情况下，与最强的基础模型精炼器相匹配，同时仍然优于其自身的控制。在训练中从未见过的掩膜生成器上，增益为2.0 $F^eta_eta$点，而控制为0.6。将个别消息归零显示了修正的来源：成对消息主要修复边界，区域消息主要修复非边界错误。代码和支持材料将公开发布。
cs.CV / 254 / 2609.33998
MetaSampling: Making Frame Samplers Efficient for Long-Video Question Answering
元采样：提升长视频问答中的帧采样效率
Abstract
Frame selection is an important component of long-video question answering (VQA) with Multimodal Large Language Models (MLLMs). Existing frame-selection methods improve over simple top-$k$ embedding retrieval and uniform sampling, but are typically applied under a fixed global selection budget. We introduce \textbf{MetaSampling}, a training-free, plug-and-play sampling strategy that can be applied on top of existing frame selectors. MetaSampling improves downstream VQA efficiency by dynamically reducing the number of frames passed to the MLLM while preserving, and in some cases improving, answer accuracy. We evaluate MetaSampling across 36 paired frame-selector--MLLM-backbone--VQA-benchmark configurations. MetaSampling reduces the number of selected frames in all 36 configurations and improves accuracy in 25 of them, yielding an average frame reduction of $8.9\%$ while slightly improving accuracy overall.
Chinese Translation
帧选择是基于多模态大语言模型（MLLMs）的长视频问答（VQA）中的一个重要组成部分。现有的帧选择方法在简单的前$k$嵌入检索和均匀采样的基础上有所改进，但通常是在固定的全局选择预算下应用。我们提出了 extbf{MetaSampling}，一种无训练、即插即用的采样策略，可以在现有帧选择器之上应用。MetaSampling通过动态减少传递给MLLM的帧数来提高下游VQA的效率，同时保持并在某些情况下提高答案的准确性。我们在36个配对的帧选择器-MLLM骨干- VQA基准配置上评估了MetaSampling。MetaSampling在所有36个配置中减少了所选帧的数量，并在其中25个配置中提高了准确性，平均帧减少率为$8.9\%$，同时整体上略微提高了准确性。
cs.CV / 255 / 2609.34021
Position Aware Layer Queries for Test Time Training in Vision Language Models
面向位置的层查询在视觉语言模型中的测试时间训练
Abstract
Test-Time Training (TTT) adapts models to incoming test samples (e.g. out-of-distribution, (OOD)) when conventional fine-tuning is infeasible. Existing TTT methods for Vision-Language Models (VLMs) create supervision from several augmented views, each requiring forward (and often backward) passes through the entire VLM, incurring substantial computational cost. We observe that one forward pass with all the intermediate layer outputs already yields far more signal than the final embedding from all augmentations. We introduce Layer Query Network (LQN), a lightweight approach that can adapt a frozen VLM (teacher) in a single forward pass of the VLM via a small model (student). LQN uses Position-Aware Distillation (PAD) to mimic the teacher VLM's intermediate-layer spatial tokens by querying spatial coordinates of intermediate tokens. LQN additionally relies on Location Consistency Regularization (LCR), a self-supervision technique, replacing expensive O(H x W) image augmentation with O(1) coordinate sampling. Integrating these, LQN i) adapts and improves zero-shot CLIP ViT-B/16 by 9.8% Top-1 on OOD ImageNet, ii) outperforms the previous best GS-Bias on fine-grained classification by 3.9% Top-1, iii) achieves faster convergence than TPS for CLIP ResNet-50 (47 mins vs 55 mins), iv) generalizes adaptation to VLMs like SigLIP, EVA-CLIP, and CoCa, and lightweight students like MLP, ResNet, VGG, and v) extends to panoptic, instance, and semantic segmentation.
Chinese Translation
测试时间训练（TTT）在传统微调不可行时，能够使模型适应进入的测试样本（例如，分布外样本（OOD））。现有的视觉语言模型（VLMs）TTT方法通过多个增强视图生成监督，每个视图都需要对整个VLM进行前向（通常还包括反向）传播，导致巨大的计算成本。我们观察到，使用所有中间层输出进行一次前向传播已经能够提供比所有增强的最终嵌入更多的信号。我们提出了层查询网络（Layer Query Network, LQN），这是一种轻量级的方法，可以通过一个小模型（学生）在VLM的单次前向传播中适应一个冻结的VLM（教师）。LQN使用面向位置的蒸馏（Position-Aware Distillation, PAD）通过查询中间标记的空间坐标来模拟教师VLM的中间层空间标记。LQN还依赖于位置一致性正则化（Location Consistency Regularization, LCR），这是一种自监督技术，用O(1)的坐标采样替代了昂贵的O(H x W)图像增强。通过整合这些方法，LQN i) 在OOD ImageNet上使零-shot CLIP ViT-B/16的Top-1性能提高了9.8%；ii) 在细粒度分类上超越了之前最佳的GS-Bias，提升了3.9%的Top-1；iii) 相比于TPS，CLIP ResNet-50的收敛速度更快（47分钟对比55分钟）；iv) 将适应性推广到SigLIP、EVA-CLIP和CoCa等VLM，以及MLP、ResNet、VGG等轻量级学生模型；v) 扩展到全景、实例和语义分割。
cs.CV / 256 / 2609.34032
Re:Cognize -- Open-Set Comic Character Re-Identification
Re:Cognize -- 开放集漫画角色重识别
Abstract
A manga reader meets a character on one page and knows them on sight a hundred pages later, without ever being handed a cast list. Re-identifying comic characters demands the same, open-set and sequential: pages arrive as a stream in reading order, new faces appear before anyone names them, and the cast is assembled as the story is read. $\textbf{Re:Cognize}$ evaluates recognition as the story is read, not against a cast handed over in advance: four protocols on one query stream, from closed-set retrieval to a cast the model must build and grow itself. The surprise is where models fail. Recognising is close to solved: one reference image per character already ranks as well as a gallery built in advance. Knowing what to believe is not: a model that adds its own matches makes its cast worse, while the same growth with correct labels would gain over twenty points of top-1 accuracy. The bottleneck is acceptance, not vision, and one comparison decides it: an addition pays exactly when it is right more often than the cast already was on the queries it takes over. The comparison has nothing to fit, and measured on half of a new corpus it calls the other half correctly. $\textbf{ReCast}$ puts it to work with nothing fitted on data: a cast sheet of one running average per character, grown only where the page itself vouches for a crop. It recovers a third to two thirds of what perfect labels would, depending on whether the cast starts from random examples or from first appearances. Re:Cognize measures whether a model can read along; ReCast is a cast that does. Our claims are on identity maintenance, recognising characters already met; the emergence of new ones is measured as a diagnostic under a fixed reference rule, and we propose no method for it.
Chinese Translation
漫画读者在某一页遇到一个角色，并在一百页后凭借一瞥认出他们，而无需查看角色名单。重识别漫画角色要求具备相同的开放集和顺序性：页面以阅读顺序流式到达，新面孔在任何人命名之前就会出现，角色在故事阅读过程中逐渐形成。$ extbf{Re:Cognize}$ 评估在故事阅读时的识别，而不是基于提前提供的角色名单：在一个查询流上进行四种协议，从封闭集检索到模型必须自行构建和扩展的角色名单。令人惊讶的是模型失败的地方。识别几乎已解决：每个角色一个参考图像的表现已与提前构建的图库相当。知道该相信什么却不是：一个添加自己匹配的模型使其角色名单变得更糟，而如果使用正确标签进行相同的扩展，则可以提高超过二十个点的 top-1 准确率。瓶颈在于接受，而不是视觉，且一个比较决定了这一点：当添加的内容在其接管的查询中比已有角色更常正确时，添加才是有价值的。这个比较没有任何适配，在新语料库的一半上测量时，它正确地调用了另一半。$ extbf{ReCast}$ 在没有任何适配的数据上投入使用：每个角色一个运行平均的角色表，仅在页面本身为裁剪提供担保的地方进行扩展。它恢复了完美标签的三分之一到三分之二，具体取决于角色是从随机示例开始还是从首次出现开始。Re:Cognize 测量模型是否能够随故事阅读；ReCast 是一个能够做到的角色名单。我们的主张集中在身份维护上，识别已经遇到的角色；新角色的出现则作为在固定参考规则下的诊断进行测量，我们并未提出任何方法。
cs.CV / 257 / 2609.34035
3D Point Tracking with State Space Models
基于状态空间模型的三维点跟踪
Abstract
Tracking any point of a dynamic scene in metric 3D - in absolute meters, not up to an unknown scale - underpins 3D and 4D reconstruction, robot navigation, and autonomous driving, where decisions are made in meters, not pixels. Our objective is a 3D point tracker accurate in those absolute terms and operating within a single commodity GPU, pose-free, monocular budget. Our method rests on one observation: once a point's 2D image trajectory is fixed, the quantity that governs its metric accuracy is the depth along its pixel ray. Rather than learning tracking end-to-end, we therefore compose two frozen front-ends - dense optical flow for 2D correspondence and a monocular metric-depth network for the third dimension - and learn only the residual they cannot supply: that depth, refined by a compact state space model (Mamba-3) conditioned on appearance features (DINOv3). A state space model rather than the transformers the strongest 3D trackers adopt is what makes a single-GPU budget attainable: it summarises a track in a fixed-size recurrent state whose memory cost is constant in the number of frames, whereas attention requires a key-value cache that grows linearly with them. On the TAPVid-3D minival benchmark our best configuration attains the highest absolute metric accuracy among methods evaluated under identical conditions (mean metric Average Jaccard, 0.256), exceeding strong feed-forward trackers, while a companion analysis, reproduced with each competitor's own evaluator, explains why several published trackers lose most of their accuracy under this budget.
Chinese Translation
在绝对米制下（而非未知尺度下）跟踪动态场景中的任意点是三维和四维重建、机器人导航和自主驾驶的基础，其中决策是以米为单位，而非像素。我们的目标是开发一个在这些绝对条件下准确的三维点跟踪器，并且能够在单个普通GPU的、无需姿态估计的单目预算下运行。我们的方法基于一个观察：一旦一个点的二维图像轨迹被固定，决定其度量准确性的量是沿着其像素光线的深度。因此，我们并不是端到端地学习跟踪，而是组合两个固定的前端——用于二维对应的稠密光流和用于第三维的单目度量深度网络——并仅学习它们无法提供的残差：即由基于外观特征（DINOv3）条件化的紧凑状态空间模型（Mamba-3）细化的深度。选择状态空间模型而非当前最强三维跟踪器所采用的变换器，使得单GPU预算成为可能：它以固定大小的递归状态总结一个轨迹，其内存成本在帧数上是恒定的，而注意力机制则需要一个随着帧数线性增长的键值缓存。在TAPVid-3D minival基准测试中，我们的最佳配置在相同条件下评估的方法中达到了最高的绝对度量准确性（平均度量Jaccard为0.256），超越了强大的前馈跟踪器。同时，伴随的分析通过每个竞争者自己的评估器重现，解释了为什么一些已发布的跟踪器在该预算下失去了大部分准确性。
cs.CV / 258 / 2609.34044
SCOPD: Sparse-Context On-Policy Self-Distillation for Efficient Vision-Language Models
SCOPD：稀疏上下文在线自蒸馏用于高效的视觉-语言模型
Abstract
Reasoning vision-language models (VLMs) process images and videos as long sequences of visual tokens, making inference expensive. Training-free token pruning reduces this cost, but aggressive compression can sharply degrade performance, often attributed to irreversible loss of task-relevant visual information. We show that this explanation is incomplete. In a fixed-context Pass@K analysis, repeated sampling from the same pruned visual representation recovers many examples missed by greedy decoding, indicating that useful visual evidence can remain accessible but be used unreliably. We call this the representation-utilization gap. Motivated by this observation, we introduce SCOPD, a sparse-context on-policy self-distillation framework in which a student generates reasoning trajectories from pruned visual tokens while a privileged full-context teacher supervises the same on-policy prefixes. SCOPD requires no ground-truth responses, architectural changes, or additional inference-time computation. We further introduce SCOPD+, which uses a small visual-budget intervention to identify visually sensitive response positions and selectively distill them. At 10% visual-token retention, the Vanilla model retains 86.37% of its unpruned performance across 13 benchmarks. SCOPD raises this to 90.49%, while SCOPD+ further improves it to 92.43%. Across token budgets, benchmarks, and pruning operators, our results show that efficient reasoning depends not only on which visual information survives pruning, but also on how reliably the model learns to use it.
Chinese Translation
推理视觉-语言模型（VLMs）将图像和视频处理为长序列的视觉标记，这使得推理成本高昂。无训练的标记剪枝可以降低这一成本，但过于激进的压缩可能会显著降低性能，这通常归因于任务相关视觉信息的不可逆损失。我们表明，这一解释并不完整。在固定上下文的 Pass@K 分析中，从相同剪枝视觉表示中重复采样能够恢复许多被贪婪解码遗漏的例子，这表明有用的视觉证据可能仍然可用，但使用不可靠。我们称之为表示利用差距。基于这一观察，我们引入了 SCOPD，一种稀疏上下文在线自蒸馏框架，其中学生从剪枝的视觉标记生成推理轨迹，而特权的全上下文教师监督相同的在线前缀。SCOPD 不需要真实响应、架构变化或额外的推理时间计算。我们进一步引入 SCOPD+，它使用小规模视觉预算干预来识别视觉敏感的响应位置并选择性地进行蒸馏。在 10% 的视觉标记保留下，Vanilla 模型在 13 个基准测试中保留了 86.37% 的未剪枝性能。SCOPD 将这一比例提高到 90.49%，而 SCOPD+ 进一步改善至 92.43%。在不同的标记预算、基准测试和剪枝操作中，我们的结果表明，高效推理不仅依赖于哪些视觉信息在剪枝中幸存，还依赖于模型学习使用这些信息的可靠性。
cs.CV / 259 / 2609.34047
ARCH-B: Architectural Representation, Comprehension and Hierarchy Benchmark
ARCH-B：建筑表现、理解与层次基准
Abstract
Multimodal models increasingly interpret visual environments, but their ability to recognize the same building across photographs, floor plans, elevations, sections, and renderings remains poorly characterized. We introduce ARCH-B, a benchmark of 354 four-choice questions across 11 cross-representational archetypes, constructed from a building-linked corpus of 3.9 million architectural images using visually similar distractors, model-guided difficulty screening, and manual validation. We evaluate 25 multimodal models and collect 5,830 responses from non-expert human participants. Model accuracy ranges from 10.45% to 83.90%, compared with a human baseline of 35.35%. Models perform comparatively well on mixed-representation outlier detection and photograph matching, but remain weaker on floorplan-to-photograph correspondence. Human and model difficulty across archetypes is only weakly correlated (Spearman's (\rho=0.33)). Held-out evaluation confirms that the difficulty identified during screening generalizes beyond the curation models. ARCH-B provides a diagnostic evaluation of visual correspondence and representation transfer across architectural media.
Chinese Translation
多模态模型越来越能够解读视觉环境，但它们在不同照片、平面图、立面图、剖面图和渲染图中识别同一建筑的能力仍然缺乏明确的表征。我们提出了ARCH-B，这是一个包含354个四选一问题的基准，涵盖11种跨表现原型，基于一个包含390万建筑图像的建筑关联语料库构建，使用视觉相似的干扰项、模型引导的难度筛选和人工验证。我们评估了25个多模态模型，并收集了5830个来自非专家人类参与者的反馈。模型的准确率范围从10.45%到83.90%，而人类基线为35.35%。模型在混合表现的异常检测和照片匹配方面表现相对较好，但在平面图与照片的对应关系上仍较弱。不同原型之间人类与模型的难度仅呈弱相关（Spearman's (
ho=0.33)）。保留评估确认了在筛选过程中识别的难度超出了策展模型的范围。ARCH-B提供了对建筑媒体中视觉对应关系和表现转移的诊断评估。
cs.CV / 260 / 2609.34071
SNaP: One-Step Posterior Sampling for Noisy Inverse Problems
SNaP：用于噪声逆问题的一步后验采样
Abstract
Diffusion and flow-matching models can produce high-quality posterior samples for inverse problems, but typically require tens to thousands of network evaluations per draw. MeanFlow enables one-step generation, yet applying it to inverse problems leaves no intermediate steps at which to enforce measurement consistency. We introduce SNaP, a one-step MeanFlow posterior sampler for linear inverse problems with Gaussian noise. Its central innovation is a measurement-adapted source: a Gaussian distribution whose mean and anisotropic covariance are determined by the measurement operator, observation, and noise level. The source anchors well-measured directions while preserving variation where the measurements are weak or uninformative. We show that the exact conditional flow transports this source to the true posterior. Across natural-image restoration and multi-coil MRI, SNaP produces diverse, high-quality samples with one network evaluation per draw, 30 to 2250 $\times$ faster than iterative samplers.
Chinese Translation
扩散和流匹配模型能够为逆问题生成高质量的后验样本，但通常每次抽样需要进行数十到数千次网络评估。MeanFlow 实现了一步生成，然而将其应用于逆问题时，缺乏可以强制测量一致性的中间步骤。我们提出了 SNaP，这是一种针对具有高斯噪声的线性逆问题的一步 MeanFlow 后验采样器。其核心创新在于测量适应源：一个高斯分布，其均值和各向异性协方差由测量算子、观测值和噪声水平决定。该源在测量良好的方向上锚定，同时在测量较弱或信息不足的地方保留变异性。我们展示了精确的条件流将该源传输到真实后验。在自然图像恢复和多线圈 MRI 中，SNaP 以每次抽样一次网络评估的速度生成多样化的高质量样本，比迭代采样器快 30 到 2250 倍。
cs.CV / 261 / 2609.34078
WhiteCon: Semi-Supervised Domain Adaptation Regression Through Whitening Transform and Dual Consistency
WhiteCon：通过白化变换和双重一致性进行半监督领域适应回归
Abstract
Domain adaptation is crucial for addressing distributional shifts that degrade model performance across domains. While most existing research has centered on classification, semi-supervised domain adaptation regression (SSDAR) for continuous-output tasks remains largely unexplored, particularly in practical scenarios with limited labeled target data. To address this gap, we propose semi-supervised domain adaptation regression through whitening transform and dual consistency (WhiteCon), which combines domain-specific whitening transform (DWT) and dual consistency regularization to enhance training stability and domain adaptation. DWT reduces the variance of the model parameters by transforming the feature covariance matrix into an identity matrix, thus stabilizing training under ordinary least squares assumptions. In addition, variance consistency regularization, as part of dual consistency regularization, aligns the variances of weak, strong, and mixup-augmented features to improve resilience against augmentation-induced perturbations. Empirical evaluations on various benchmark datasets under SSDAR settings demonstrate that the proposed WhiteCon achieves state-of-the-art performance compared to existing methods, effectively addressing domain shifts in regression tasks. The code for WhiteCon is available at https://github.com/sejin-sim/WhiteCon.
Chinese Translation
领域适应对于解决跨领域模型性能下降的分布变化至关重要。尽管现有研究大多集中在分类任务上，但对于连续输出任务的半监督领域适应回归（SSDAR）仍然大部分未被探索，尤其是在标记目标数据有限的实际场景中。为了解决这一问题，我们提出了通过白化变换和双重一致性进行半监督领域适应回归（WhiteCon），该方法结合了领域特定的白化变换（DWT）和双重一致性正则化，以增强训练稳定性和领域适应性。DWT通过将特征协方差矩阵转化为单位矩阵来减少模型参数的方差，从而在普通最小二乘假设下稳定训练。此外，作为双重一致性正则化的一部分，方差一致性正则化使得弱、强和混合增强特征的方差一致，从而提高对增强引起的扰动的抵抗力。在SSDAR设置下对各种基准数据集的实证评估表明，所提出的WhiteCon相较于现有方法实现了最先进的性能，有效应对回归任务中的领域转变。WhiteCon的代码可在 https://github.com/sejin-sim/WhiteCon 获取。
cs.CV / 262 / 2609.34094
Advancing Wildlife Conservation through Multimodal Animal Re-Identification with Environmental Metadata
通过环境元数据的多模态动物重识别推动野生动物保护
Abstract
Identifying individual animals is crucial for effective wildlife monitoring and conservation efforts. Recent advancements in computer vision have shown promise in animal re-identification (Animal ReID) by leveraging data from camera traps. However, existing Animal ReID datasets rely exclusively on visual data, overlooking environmental metadata that ecologists have identified as highly correlated with animal behavior and identity, such as temperature and circadian rhythms. Meanwhile, modern vision-language models (VLMs) offer rich multimodal reasoning capabilities, but existing resources underutilize their text-processing potential. To address these limitations, we propose MetaWild, a multimodal Animal ReID dataset comprising 20,890 images across six species, paired with environmental metadata extracted from embedded camera trap overlays and scene contexts. Additionally, to facilitate the use of metadata in existing ReID methods, we propose the Meta-Feature Adapter (MFA), a lightweight module that can be incorporated into existing VLM-based ReID methods, allowing ReID models to leverage both environmental metadata and visual information to improve ReID performance. Experiments on MetaWild show that combining baseline ReID models with MFA to incorporate metadata consistently improves performance compared to using visual information alone, validating the effectiveness of incorporating metadata in re-identification.
Chinese Translation
识别个体动物对于有效的野生动物监测和保护工作至关重要。最近计算机视觉的进展在动物重识别（Animal ReID）方面展现了潜力，利用来自相机陷阱的数据。然而，现有的Animal ReID数据集完全依赖于视觉数据，忽视了生态学家已识别的与动物行为和身份高度相关的环境元数据，例如温度和昼夜节律。同时，现代视觉-语言模型（VLMs）提供了丰富的多模态推理能力，但现有资源未充分利用其文本处理潜力。为了解决这些局限性，我们提出了MetaWild，一个包含20,890张图像的多模态Animal ReID数据集，涵盖六个物种，并配有从嵌入式相机陷阱叠加层和场景上下文中提取的环境元数据。此外，为了促进在现有ReID方法中使用元数据，我们提出了Meta-Feature Adapter (MFA)，一个轻量级模块，可以集成到现有的基于VLM的ReID方法中，使ReID模型能够利用环境元数据和视觉信息来提高ReID性能。在MetaWild上的实验表明，将基线ReID模型与MFA结合以纳入元数据，性能始终优于仅使用视觉信息，验证了在重识别中纳入元数据的有效性。
cs.CV / 263 / 2609.34106
The Devil is in the Spectrum Bias: Spectrum-Balanced Feature Matching for Robust Representation Distillation
魔鬼藏在光谱偏差中：用于鲁棒表示蒸馏的光谱平衡特征匹配
Abstract
Large visual foundation models have demonstrated remarkable transferability across a wide range of downstream tasks. To deploy such models efficiently, feature matching has become a popular knowledge distillation approach that transfers teacher representations to smaller student models without requiring labeled data. However, we show that the conventional feature matching objective with L2-distance is inherently biased toward reconstructing dominant spectral directions of the teacher representation, while under-optimizing low-variance directions that often contain task-relevant information. To address this, we propose Spectrum-Balanced Feature Matching, SpecMatch, a simple objective that adaptively emphasizes under-optimized spectral directions while preserving the relative importance of dominant directions. SpecMatch is easy to implement and introduces negligible computational overhead. Extensive experiments on image recognition demonstrate that SpecMatch consistently improves downstream adaptation across diverse tasks, including image classification, anomaly detection, medical image analysis, and domain generalization. In particular, SpecMatch outperforms conventional feature matching in 40 of 42 teacher--student and training-setting combinations, while consistently improving over the original student model in all settings. We further demonstrate that the proposed objective generalizes beyond vision, improving downstream performance across six protein understanding tasks.
Chinese Translation
大型视觉基础模型在广泛的下游任务中展现了显著的迁移能力。为了高效部署这些模型，特征匹配已成为一种流行的知识蒸馏方法，它将教师表示转移到较小的学生模型中，而无需标记数据。然而，我们发现传统的L2距离特征匹配目标本质上偏向于重建教师表示的主导光谱方向，而对通常包含任务相关信息的低方差方向优化不足。为了解决这个问题，我们提出了光谱平衡特征匹配（Spectrum-Balanced Feature Matching，SpecMatch），这是一种简单的目标，能够自适应地强调未优化的光谱方向，同时保留主导方向的相对重要性。SpecMatch易于实现，并且引入的计算开销微乎其微。在图像识别方面的广泛实验表明，SpecMatch在包括图像分类、异常检测、医学图像分析和领域泛化等多种任务中，始终改善下游适应性。特别是，SpecMatch在42个教师-学生和训练设置组合中的40个中超越了传统特征匹配，同时在所有设置中持续改善了原始学生模型。我们进一步证明，该目标超越了视觉领域，在六个蛋白质理解任务中改善了下游性能。
cs.CV / 264 / 2609.34124
SpatialSkill: Self-Evolving Skills for Cross-View Spatial Reasoning
SpatialSkill：用于跨视角空间推理的自我演化技能
Abstract
Cross-view spatial reasoning requires a model to align different viewpoints into a coherent spatial representation, yet this ability remains challenging for vision-language models despite being natural to humans. Existing methods typically improve spatial reasoning by updating model weights, which keeps the acquired knowledge implicit and tied to a specific backbone. We propose \textit{SpatialSkill}, a weight-update-free framework that enables a frozen vision-language model to accumulate explicit natural-language reasoning skills from offline trajectories. Unlike symbolic tasks, perceptual skills cannot be reliably verified simply by executing them: a plausible spatial rule may lack visual support or require transformations that the frozen model cannot perform. SpatialSkill therefore admits candidate skills only after visual-grounding and executability checks, constrains manual evolution to prevent harmful regressions, and routes skills by spatial-reasoning category to reduce negative transfer. On CityCube, across four frozen executors, SpatialSkill yields consistent gains, and a 9B executor equipped with SpatialSkill surpasses the strongest closed-source reference in our evaluation. The skills are stored in a versioned natural-language manual, making the reasoning strategies explicit and auditable without modifying model parameters. Code at https://github.com/vindahi/SpatialSkill.
Chinese Translation
跨视角空间推理要求模型将不同的视角对齐为一致的空间表示，尽管这对人类而言是自然的，但对于视觉-语言模型而言，这一能力仍然具有挑战性。现有方法通常通过更新模型权重来提高空间推理能力，这使得获得的知识保持隐式，并与特定的基础模型紧密相关。我们提出了 extit{SpatialSkill}，一个无需更新权重的框架，使得一个冻结的视觉-语言模型能够从离线轨迹中积累显式的自然语言推理技能。与符号任务不同，感知技能不能仅通过执行来可靠验证：一个合理的空间规则可能缺乏视觉支持或需要冻结模型无法执行的变换。因此，SpatialSkill 仅在经过视觉基础和可执行性检查后接受候选技能，限制手动演化以防止有害的回归，并通过空间推理类别引导技能，以减少负迁移。在 CityCube 上，四个冻结的执行器中，SpatialSkill 一致地带来了增益，并且一个配备 SpatialSkill 的 9B 执行器在我们的评估中超越了最强的闭源参考。这些技能存储在一个版本化的自然语言手册中，使推理策略显式且可审计，而无需修改模型参数。代码可在 https://github.com/vindahi/SpatialSkill 获取。
cs.CV / 265 / 2609.34133
PrefLUT: Reusable and Refinable Personalized Color Editing from Pairwise Preferences
PrefLUT：可重用和可细化的基于成对偏好的个性化颜色编辑
Abstract
Photographic color editing is inherently personal: the same image can appear too warm, too muted, or already satisfactory to different users. Most lookup table (LUT) and reference-guided methods target a specified appearance rather than model persistent preferences from repeated user choices. To address this gap, we introduce PrefLUT, a reusable and refinable user-preference modeling framework for deployable 3D LUTs, encoding ordered preferred/non-preferred image pairs into a lightweight Reusable User Profile that is reused across queries and refined using additional user preference pairs, without per-user optimization. A Query-Conditioned LUT Predictor combines this profile with each image to predict a LUT latent vector and edit strength. An Identity-Residual LUT Decoder and Edit-Strength Controller then produce an exportable 3D LUT. Experiments on three datasets demonstrate effective personalized editing and general-purpose enhancement. Each quantized profile requires only 260 bytes, and editing takes 1.365 ms/image on an RTX 5090 GPU. We also introduce the Preference-Conditioning Verification Protocol (PCVP), an evaluation protocol to verify whether personalized image edits depend on user preferences and the query image through controlled changes to user profiles, preference orders, pair correspondences, and query images.
Chinese Translation
摄影颜色编辑本质上是个人化的：同一图像对于不同用户可能显得过于温暖、过于柔和或已经令人满意。大多数查找表（LUT）和参考引导方法针对特定外观，而不是从重复的用户选择中建模持久的偏好。为了解决这一问题，我们提出了PrefLUT，一种可重用和可细化的用户偏好建模框架，用于可部署的3D LUT，将有序的偏好/非偏好图像对编码为轻量级的可重用用户档案，该档案在查询中重复使用，并通过额外的用户偏好对进行细化，而无需针对每个用户进行优化。查询条件LUT预测器将此档案与每个图像结合，以预测LUT潜在向量和编辑强度。身份残差LUT解码器和编辑强度控制器随后生成可导出的3D LUT。在三个数据集上的实验表明了有效的个性化编辑和通用增强。每个量化档案仅需260字节，并且在RTX 5090 GPU上编辑每幅图像仅需1.365毫秒。我们还引入了偏好条件验证协议（PCVP），这是一种评估协议，用于验证个性化图像编辑是否依赖于用户偏好和查询图像，通过对用户档案、偏好顺序、对对应关系和查询图像的控制变化进行验证。
cs.CV / 266 / 2609.34142
Analytical and Convolutional Neural Network-Based Motion-Vector Propagation for Efficient Video Object Detection
基于分析和卷积神经网络的运动矢量传播以实现高效的视频目标检测
Abstract
Continuous video analytics requires accurate localization at low latency within embedded power budgets. This paper presents a hardware-software design methodology that reuses codec motion vectors (MVs) between detector invocations. Two alternative models support translation and scale changes: analytical motion-vector propagation (Analytical-MV) and learned propagation using a convolutional neural network (CNN) (CNN-MV). The learned model uses convolutional operations and independent object updates suited to parallel execution on an edge graphics processing unit (GPU). Analytical-MV combines a harmonic-mean precision-recall score (F1) of 0.909 with a mean end-to-end latency of 9.03 ms and an energy consumption of 0.177 J per frame, yielding the lowest latency and energy among the evaluated configurations. Relative to detection on every frame, it reduces mean latency by 25.9% and energy per frame by 36.4%. CNN-MV offers a different trade-off: its fastest configuration raises recall from 0.871 for Analytical-MV to 0.890 and lowers mean power from 19.64 to 17.32 W, while achieving a latency of 18.42 ms and an energy consumption of 0.319 J per frame. It is therefore useful when recall or operating power is more important than minimum latency and energy. Execution on a deep learning accelerator (DLA) further reduces time-averaged GPU utilization relative to GPU execution. Host-processing optimization substantially improves both latency and energy, demonstrating the value of jointly designing temporal models and their execution pipelines.
Chinese Translation
连续视频分析需要在嵌入式功耗预算内实现低延迟的准确定位。本文提出了一种硬件-软件设计方法，通过在检测器调用之间重用编解码器运动矢量（MVs）。两种替代模型支持平移和缩放变化：分析运动矢量传播（Analytical-MV）和使用卷积神经网络（CNN）进行的学习传播（CNN-MV）。学习模型利用卷积操作和适合于在边缘图形处理单元（GPU）上并行执行的独立对象更新。分析运动矢量传播结合了0.909的调和平均精确度-召回率得分（F1），平均端到端延迟为9.03毫秒，每帧能耗为0.177焦耳，在评估的配置中实现了最低的延迟和能耗。相较于对每帧进行检测，它将平均延迟减少了25.9%，每帧能耗降低了36.4%。CNN-MV提供了不同的权衡：其最快的配置将召回率从Analytical-MV的0.871提高到0.890，并将平均功率从19.64瓦降低到17.32瓦，同时实现了18.42毫秒的延迟和每帧0.319焦耳的能耗。因此，当召回率或操作功率比最低延迟和能耗更重要时，它是有用的。在深度学习加速器（DLA）上执行进一步减少了相对于GPU执行的时间平均GPU利用率。主机处理优化显著改善了延迟和能耗，展示了联合设计时间模型及其执行管道的价值。
cs.CV / 267 / 2609.34143
Beyond Geometry: Benchmarking and Consistency Reasoning for 3D Logical Anomaly Detection
超越几何：3D逻辑异常检测的基准测试与一致性推理
Abstract
Existing 3D industrial anomaly detection mainly targets local geometric deviations. In contrast, many industrial anomalies violate object-level design or assembly rules, which we define as 3D logical anomalies. To address these challenges, we introduce the Industrial Logical Anomaly Detection Dataset (ILGAD), the first scalable benchmark dedicated to logical anomalies in industrial point clouds. ILGAD contains 2,774 samples from 15 categories with point-level annotations and covers existence, specification, pose, and assembly-state errors. To detect such 3D logical anomalies, we propose a consistency reasoning framework that assesses whether local geometry, structure coverage, and spatial relations conform to the normal design. The framework detects geometric changes, unsupported expected structures, and abnormal local arrangements. Experiments on ILGAD, Anomaly-ShapeNet, and IEC3D demonstrate superior object-level detection and point-level localization, showing that the framework effectively detects logical anomalies and generalizes to conventional geometric defects.
Chinese Translation
现有的3D工业异常检测主要针对局部几何偏差。相比之下，许多工业异常违反了对象级的设计或装配规则，我们将其定义为3D逻辑异常。为了解决这些挑战，我们引入了工业逻辑异常检测数据集（Industrial Logical Anomaly Detection Dataset，ILGAD），这是首个专注于工业点云中逻辑异常的可扩展基准。ILGAD包含来自15个类别的2774个样本，具备点级注释，涵盖存在性、规格、姿态及装配状态错误。为了检测此类3D逻辑异常，我们提出了一种一致性推理框架，用于评估局部几何、结构覆盖及空间关系是否符合正常设计。该框架能够检测几何变化、不支持的预期结构及异常的局部排列。在ILGAD、Anomaly-ShapeNet和IEC3D上的实验表明，该框架在对象级检测和点级定位方面表现优越，证明其有效检测逻辑异常并能推广至传统几何缺陷。
cs.CV / 268 / 2609.34144
CAST: Reconstruction-Coupled Acceleration of Interactive World Models
CAST：重建耦合的互动世界模型加速
Abstract
Interactive world models must respond quickly to controls while preserving scene consistency. Existing acceleration methods can miss heterogeneous control responses and spatial transport when recovering skipped features. We observe that interaction-induced feature changes correlate with approximation error, while low-frequency interpolation errors are phase-sensitive and show more predictable phase progression. These findings motivate CAST, a reconstruction-coupled inference framework. CAST selects anchors by interaction sensitivity and cross-layer coverage, reconstructs skipped residuals with frequency- and confidence-aware Phase-Aware Reconstruction (PAR), and coordinates historical KV routing according to downstream reconstruction responsibility. On Matrix-Game 3.0 and HY-World 1.5, CAST achieves 2.15x and 3.48x speedups, respectively, while maintaining visual quality close to Native (Figure 1). It also attains the highest VBench scores among compared methods and leads non-native baselines on seven and six of thirteen WorldMark dimensions, demonstrating a balance of generation speed, visual quality, and interactive responsiveness under real-time control. Code is available at https://github.com/lokiniuniu/CAST.
Chinese Translation
互动世界模型必须快速响应控制，同时保持场景一致性。现有的加速方法在恢复跳过的特征时可能会错过异质控制响应和空间传输。我们观察到，交互引起的特征变化与近似误差相关，而低频插值误差则对相位敏感，并显示出更可预测的相位进展。这些发现促使我们提出CAST，一个重建耦合的推理框架。CAST通过交互敏感性和跨层覆盖选择锚点，利用频率和置信度感知的相位感知重建（Phase-Aware Reconstruction, PAR）重建跳过的残差，并根据下游重建责任协调历史KV路由。在Matrix-Game 3.0和HY-World 1.5上，CAST分别实现了2.15倍和3.48倍的加速，同时保持接近原生的视觉质量（见图1）。它还在比较方法中获得了最高的VBench得分，并在十三个WorldMark维度中领先于七个和六个非原生基线，展示了在实时控制下生成速度、视觉质量和互动响应之间的平衡。代码可在https://github.com/lokiniuniu/CAST获取。
cs.CV / 269 / 2609.34148
Geometric Encoding for Spatial Reasoning in Vision-Language Models
视觉语言模型中的空间推理几何编码
Abstract
Vision-Language Models (VLMs) are far more reliable at recognizing what appears in a video than at reasoning about its spatial and temporal properties, such as metric distances, object dimensions, and consistent object identities across frames. We present Geometric Code, a perception-to-geometry pipeline that computes explicit spatial structure from video and supplies it to VLMs as context to augment reasoning. A perception layer segments and classifies objects and recovers depth, camera pose, and intrinsics from monocular RGB video. A deterministic geometric engine then back-projects, merges, and cleans these outputs into a spatial code, including per-object positions, dimensions, counts, inter-object distances, appearance order, and room geometry. The code is serialized into VLMs' prompts, either alongside the video or replacing it entirely. Specifically, there is no component trained or fine-tuned in our approach. On VSI-Bench, augmenting 2B and 4B open models with the spatial code improves average accuracy by +4.1 points over the frames-only baseline, with the largest gains on numeric estimation tasks such as absolute distance (+24.1 points). The results suggest that explicitly computed geometry, delivered through the language channel, recovers spatial competence that small VLMs cannot extract from pixels alone.
Chinese Translation
视觉语言模型（VLMs）在识别视频中出现的内容方面远比在推理其空间和时间属性（如度量距离、物体尺寸以及跨帧的一致物体身份）方面更为可靠。我们提出了几何编码（Geometric Code），这是一种从视频计算显式空间结构并将其作为上下文提供给VLMs以增强推理的感知到几何的管道。感知层对物体进行分割和分类，并从单目RGB视频中恢复深度、相机姿态和内参。然后，一个确定性的几何引擎将这些输出进行反投影、合并和清理，生成空间编码，包括每个物体的位置、尺寸、数量、物体间距离、外观顺序和房间几何。该编码被序列化为VLMs的提示，既可以与视频一起使用，也可以完全替代视频。具体而言，我们的方法中没有任何组件经过训练或微调。在VSI-Bench上，将空间编码与2B和4B开放模型结合使用，使得平均准确率比仅使用帧的基线提高了+4.1个百分点，尤其在数值估计任务（如绝对距离）上获得了最大的提升（+24.1个百分点）。结果表明，通过语言通道传递的显式计算几何恢复了小型VLMs无法仅从像素中提取的空间能力。
cs.CV / 270 / 2609.34149
Functional Hand Type Prior for 3D Hand Pose Estimation and Action Recognition from Egocentric View Monocular Videos
基于功能手型先验的3D手势估计与自我中心视角单目视频中的动作识别
Abstract
Current methods for egocentric view action recognition often face challenges in perceiving dynamic hand movements relying solely on geometrical or physical information. In this work, we effectively address this problem by gaining insights into the correlation between functional hand configurations and objects, which improves the detailed interpretation of real-world scenarios. To this end, we introduce a practical taxonomy of hand types based on the functioning perspective and utilize it for per-frame hand type labeling on existing datasets. We also propose a novel hand action recognition framework considering semantic details of the hand type as prior. This approach boosts the network's understanding of the continuous hand interaction throughout the action sequence. Our whole pipeline consists of three main modules: (1) Feature Extraction, (2) Egocentric Knowledge Module, which estimates 3D hand pose, object category, and hand type leveraging short-term cues, and (2) Egocentric Action Module, which aggregates per-frame knowledge, including text embeddings of hand type, over a longer time. In our extensive experiments with large-scale benchmarks, FPHA and H2O, our model outperforms current state-of-the-art methods, demonstrating its superior performance.
Chinese Translation
当前自我中心视角的动作识别方法在仅依赖几何或物理信息来感知动态手部动作时，常常面临挑战。在本研究中，我们通过深入探讨功能手部配置与物体之间的关联，有效解决了这一问题，从而改善了对现实场景的详细解释。为此，我们基于功能视角引入了一种实用的手型分类法，并利用该分类法对现有数据集进行逐帧手型标注。我们还提出了一种新颖的手部动作识别框架，将手型的语义细节作为先验信息进行考虑。这种方法增强了网络对动作序列中连续手部交互的理解。我们的整个流程由三个主要模块组成：(1) 特征提取，(2) 自我中心知识模块，该模块利用短期线索估计3D手势、物体类别和手型，以及(3) 自我中心动作模块，该模块在更长时间内聚合逐帧知识，包括手型的文本嵌入。在我们与大规模基准数据集FPHA和H2O进行的广泛实验中，我们的模型超越了当前的最先进方法，展示了其卓越的性能。
cs.CV / 271 / 2609.34167
Natural Image Autoencoder-Based fMRI Representations for Trait and State Prediction
基于自然图像自编码器的功能性磁共振成像表示用于特质和状态预测
Abstract
Foundation models pre-trained on large-scale fMRI datasets have shown strong downstream performance, but at substantial data and computation cost. To investigate how much fMRI-specific pre-training is actually needed for such performance, we introduce FReD, which derives fMRI representations from a frozen Deep Compression AutoEncoder (DCAE) pre-trained exclusively on natural images and pairs them with a task specific readout. For trait prediction, FReD summarizes frame-wise representations by their temporal mean and log-standard deviation and applies linear probing, with late fusion across two normalization schemes. For state prediction, it represents each frame as a single token and models temporal dependencies with a shallow Transformer. Across four resting-state datasets spanning six trait-prediction targets, linear probes on frozen DCAE features generally outperform those on fMRI foundation model representations and remain competitive with fully fine-tuned fMRI foundation models. On three task-fMRI state-prediction tasks, a temporal readout on DCAE features performs comparably to the strongest foundation models evaluated. A Gaussian injection analysis further shows that localized signal changes are recovered more accurately from the frozen DCAE features than from the evaluated foundation-model representations. Together, these results show that strong performance on current fMRI benchmarks is possible without fMRI-specific representation pre-training, making frozen natural-image features as a useful baseline for assessing its added value.
Chinese Translation
在大规模功能性磁共振成像（fMRI）数据集上预训练的基础模型表现出强大的下游性能，但代价高昂，涉及大量数据和计算资源。为了探讨实现这种性能所需的fMRI特定预训练的实际需求，我们引入了FReD，该模型从一个仅在自然图像上预训练的冻结深度压缩自编码器（DCAE）中提取fMRI表示，并将其与特定任务的读取方式配对。对于特质预测，FReD通过时间均值和对数标准差总结逐帧表示，并应用线性探测，同时在两种归一化方案之间进行后期融合。对于状态预测，它将每帧表示为一个单一的标记，并使用浅层变换器建模时间依赖性。在涵盖六个特质预测目标的四个静息状态数据集中，冻结DCAE特征上的线性探测通常优于fMRI基础模型表示，并与完全微调的fMRI基础模型保持竞争力。在三个任务fMRI状态预测任务中，DCAE特征上的时间读取与评估的最强基础模型表现相当。高斯注入分析进一步表明，从冻结的DCAE特征中恢复局部信号变化的准确性优于评估的基础模型表示。综合来看，这些结果表明，在当前fMRI基准测试中实现强大性能是可能的，而无需fMRI特定的表示预训练，这使得冻结的自然图像特征成为评估其附加价值的有用基线。
cs.CV / 272 / 2609.34176
AGILE-GS: Anchor-Guided Fast Next-Best-View Selection for Active 3D Gaussian Splatting
AGILE-GS：基于锚点引导的快速下一最佳视图选择用于主动3D高斯点云渲染
Abstract
Radiance fields need hundreds of views, and their placement matters as much as their number. Next-best-view (NBV) selection for 3D Gaussian Splatting (3DGS) usually scores every candidate in the pool and keeps one. Searching for information and choosing a camera, however, are separable problems. We present AGILE-GS, an anchor-guided NBV method that separates the two. A virtual anchor pose is optimized on SE(3) by Riemannian gradient ascent on expected information gain. It need not be reachable or in the pool; it marks where the model is most uncertain. Candidates are scored against the anchor's viewing geometry, and a greedy ridge-leverage step distills the pool into a small, non-redundant shortlist without rendering any candidate. The shortlist can be used in two ways. AGILE-GS takes the first view on it as the next view, so no Fisher information is computed for any candidate. AGILE-GS+ computes the Fisher information gain of each shortlisted view and picks the best, so the expensive evaluation runs on a handful of views rather than the whole pool. On standard benchmarks and in closed-loop embodied acquisition, both match or exceed existing baselines while cutting selection latency by one to two orders of magnitude.
Chinese Translation
辐射场需要数百个视图，其位置与数量同样重要。3D高斯点云渲染（3DGS）的下一最佳视图（NBV）选择通常对池中的每个候选视图进行评分并保留一个。然而，信息搜索和相机选择是可分离的问题。我们提出了AGILE-GS，一种基于锚点引导的NBV方法，将这两者分开。通过在期望信息增益上进行黎曼梯度上升，优化一个虚拟锚点姿态于SE(3)空间。该锚点不必可达或在候选池中；它标记了模型最不确定的地方。候选视图根据锚点的视角几何进行评分，并通过贪婪的脊杠杆步骤将池中的候选视图提炼为一个小的、非冗余的候选列表，而无需渲染任何候选视图。该候选列表可以有两种使用方式。AGILE-GS将其上的第一个视图作为下一视图，因此不会为任何候选视图计算Fisher信息。AGILE-GS+计算每个入选视图的Fisher信息增益并选择最佳视图，因此昂贵的评估只在少数视图上进行，而不是在整个候选池中进行。在标准基准测试和闭环具身获取中，两者的表现均与现有基线相匹配或超越，同时将选择延迟缩短了一个到两个数量级。
cs.CV / 273 / 2609.34178
Enhanced Video Text Editing with Trajectory-Aligned Glyph Rendering
基于轨迹对齐的字形渲染增强视频文本编辑
Abstract
Video text editing aims to replace or add text in a video while keeping the rest of the video unchanged, which requires the edited text to be correct in every frame and to move coherently with the scene. Despite the remarkable progress of video diffusion models, they struggle to reproduce exact stroke structures and often produce garbled or wrong characters, especially for characters with complex strokes. To address this, we propose a trajectory-aligned glyph rendering reference that provides explicit per-frame glyph guidance following the position and perspective of the text, and a depth-normalized recognizer feature supervision that supervises the generated text on multi-depth features of a frozen text recognizer with per-depth normalized errors, targeting stroke errors overlooked by the diffusion loss. We further build VTEdit, a benchmark of 288 real-scene clips with 440 annotated text trajectories covering text replacement and text addition, which will be publicly released to facilitate future research. Experiments on VTEdit show that our method outperforms image text editing methods, video editing methods, and commercial models in text accuracy and background preservation, achieving a sentence accuracy of 0.9408, and receives the highest preference in a user study.
Chinese Translation
视频文本编辑旨在在视频中替换或添加文本，同时保持视频的其他部分不变，这要求编辑后的文本在每一帧中都要准确，并与场景协调移动。尽管视频扩散模型取得了显著进展，但它们在重现精确的笔画结构方面仍然存在困难，常常产生混乱或错误的字符，尤其是对于笔画复杂的字符。为了解决这一问题，我们提出了一种轨迹对齐的字形渲染参考，提供明确的逐帧字形指导，遵循文本的位置和视角，并引入了一种深度归一化的识别特征监督，监督生成文本在冻结文本识别器的多深度特征上的表现，针对扩散损失忽视的笔画错误。我们进一步构建了VTEdit，这是一个包含288个真实场景片段和440个标注文本轨迹的基准数据集，涵盖文本替换和文本添加，将公开发布以促进未来的研究。在VTEdit上的实验表明，我们的方法在文本准确性和背景保留方面优于图像文本编辑方法、视频编辑方法和商业模型，达到了0.9408的句子准确率，并在用户研究中获得了最高偏好。
cs.CV / 274 / 2609.34183
CRF Loss is How Networks Should Learn Boundaries in Weakly Supervised Segmentation
CRF损失是网络在弱监督分割中学习边界的方式
Abstract
Weakly Supervised Semantic Segmentation (WSSS) learns pixel-level predictions from image-level tags. Recent work focuses on improving coarse CAMs extracted from large vision-language models (commonly CLIP), but does little to improve their accuracy along segment boundaries. That job is instead delegated to a post-processing method like DenseCRF. However, because DenseCRF relies on low-level colour cues, it can flip correct labels to incorrect ones when neighbouring pixels share similar colours. SAM has recently been adopted as a natural alternative, yet it simply takes on DenseCRF's role as an intermediate "refinement" step that outputs one-hot pseudo-labels in prior work. By discarding the valuable uncertainty in CAMs, these one-hot pseudo-labels turn borderline errors into confidently wrong targets. Our key insight is that CAMs should supervise training alongside SAM boundaries, each through its own loss, rather than being fused together into a single hard target. Inspired by CRF potentials, we propose a framework that disentangles soft pseudo-labels as unary supervision and binary edge maps as pairwise supervision. We realize our framework in a single-stage model, DS-CRF, using CAMs from dino.txt and boundaries from SAM. DS-CRF sets a new state-of-the-art of 56.5% mIoU on MS COCO.
Chinese Translation
弱监督语义分割（WSSS）从图像级标签中学习像素级预测。近期的研究集中在改善从大型视觉-语言模型（通常是CLIP）提取的粗略CAM（类激活映射），但对改善分割边界的准确性关注较少。这个任务通常被委派给像DenseCRF这样的后处理方法。然而，由于DenseCRF依赖于低级颜色线索，当相邻像素共享相似颜色时，它可能会将正确标签错误地翻转。最近，SAM被作为一种自然的替代方案采用，但它在先前的工作中仅仅充当DenseCRF的中间“细化”步骤，输出一热伪标签。通过丢弃CAM中有价值的不确定性，这些一热伪标签将边界错误转化为自信的错误目标。我们的关键见解是，CAM应该与SAM边界一起监督训练，各自通过自己的损失，而不是融合成一个单一的硬目标。受到CRF潜力的启发，我们提出了一个框架，将软伪标签解耦为单一监督，将二元边缘图作为成对监督。我们在一个单阶段模型DS-CRF中实现了我们的框架，使用来自dino.txt的CAM和来自SAM的边界。DS-CRF在MS COCO上设定了56.5%的新状态-of-the-art mIoU。
cs.CV / 275 / 2609.34190
MotionSpaceFlow: Representation-Aware Flow Matching in Direct Motion Space
运动空间流：在直接运动空间中的表征感知流匹配
Abstract
Recent advances in diffusion and flow models have substantially improved text-driven human motion generation. Yet most methods generate in low-dimensional, temporally downsampled latent spaces learned primarily for reconstruction, a bottleneck that can limit generation quality and preclude direct manipulation of individual frames and joints. We introduce MotionSpaceFlow (MSFlow), a representation-aware flow-matching framework that predicts clean motion directly in continuous motion space without a learned encoder or decoder. To account for the anisotropic structure of direct motion representations, we propose representation-aware noise scaling and show how the initial Gaussian source scale governs the covariance of intermediate probability-path marginals. We further introduce a Representation-Aware Multimodal Diffusion Transformer (RA-MMDiT), which jointly updates token-level language and full-resolution motion features through joint attention while adapting temporal information flow to the motion representation: causal attention for incremental features defined by frame-to-frame changes, and bidirectional attention for global features such as absolute joint coordinates. Across different datasets and motion representations, MSFlow achieves state-of-the-art text-to-motion performance. Its global representation variant additionally enables zero-shot, inference-time control over any joint or frame through projection sampling without control-conditioned training, delivering leading motion quality with exact constraint satisfaction.
Chinese Translation
近期在扩散和流模型方面的进展显著提升了基于文本的人体运动生成。然而，大多数方法主要在为重建学习的低维、时间下采样的潜在空间中生成，这一瓶颈限制了生成质量，并阻碍了对单帧和关节的直接操作。我们提出了运动空间流（MotionSpaceFlow，MSFlow），这是一种表征感知的流匹配框架，能够直接在连续运动空间中预测干净的运动，而无需学习的编码器或解码器。为了考虑直接运动表征的各向异性结构，我们提出了表征感知噪声缩放，并展示了初始高斯源尺度如何影响中间概率路径边际的协方差。我们进一步引入了一种表征感知的多模态扩散变换器（Representation-Aware Multimodal Diffusion Transformer，RA-MMDiT），该变换器通过联合注意力同时更新标记级语言和全分辨率运动特征，同时调整时间信息流以适应运动表征：因果注意力用于由帧间变化定义的增量特征，而双向注意力则用于绝对关节坐标等全局特征。在不同的数据集和运动表征中，MSFlow实现了最先进的文本到运动的性能。其全局表征变体还能够通过投影采样在推理时实现对任何关节或帧的零-shot控制，无需控制条件训练，提供领先的运动质量并精确满足约束。
cs.CV / 276 / 2609.34196
ConvCue: Complementary Visual Inductive Biases for Vision-Language Models
ConvCue：视觉语言模型的互补视觉归纳偏置
Abstract
Modern vision-language models (VLMs) achieve strong performance across a broad range of multimodal tasks, yet still struggle with visual questions that require fine-grained discrimination and spatial understanding. These limitations motivate investigating whether supplementary visual representations can improve existing VLMs without replacing their native visual encoders. Pretrained convolutional networks offer a candidate feature source, motivated by their local connectivity and spatial weight sharing. We introduce CONVCUE, which augments the native visual representations of a pretrained VLM with final-stage features from a parallel, frozen pretrained CNN. A learnable adapter maps convolutional features to the native visual feature dimension, while gated cross-attention allows the original visual tokens to retrieve information from the CNN features. The enhanced tokens are passed through the original visual-to-language projector, and the model is adapted through a two-stage training procedure. We evaluate CONVCUE on Qwen3-VL-2B, Qwen3-VL-4B, and LLaVA-OneVision-7B across 13 multimodal benchmarks covering visual question answering, document and chart understanding, and multimodal reasoning. CONVCUE improves average benchmark performance over both the original models and matched two-stage fine-tuning controls on all three backbones. On Qwen3-VL-4B, it improves over the original model on all 13 benchmarks and raises the average score from 75.00 to 78.82 relative to the matched fine-tuning control. These results show that pretrained convolutional representations, when integrated through learned adaptation and fusion, can improve the visual understanding of existing VLMs without replacing their original visual encoders.
Chinese Translation
现代视觉语言模型（VLMs）在广泛的多模态任务中表现出色，但在需要细粒度区分和空间理解的视觉问题上仍然存在困难。这些局限性促使我们探讨是否可以通过补充视觉表示来改善现有的VLM，而不替换其原有的视觉编码器。预训练的卷积网络提供了一种候选特征源，基于其局部连接性和空间权重共享。我们引入了CONVCUE，它通过来自并行冻结的预训练CNN的最终阶段特征，增强了预训练VLM的原生视觉表示。一个可学习的适配器将卷积特征映射到原生视觉特征维度，而门控交叉注意力允许原始视觉标记从CNN特征中检索信息。增强后的标记通过原始的视觉到语言投影器传递，并通过两阶段训练程序对模型进行适应。我们在Qwen3-VL-2B、Qwen3-VL-4B和LLaVA-OneVision-7B上评估了CONVCUE，涵盖了视觉问答、文档和图表理解以及多模态推理的13个多模态基准。CONVCUE在所有三种骨干网络上，相较于原始模型和匹配的两阶段微调对照组，提升了平均基准性能。在Qwen3-VL-4B上，它在所有13个基准上均优于原始模型，并将平均得分从75.00提高到78.82，相较于匹配的微调对照。这些结果表明，通过学习适配和融合整合的预训练卷积表示，可以在不替换原始视觉编码器的情况下，提高现有VLM的视觉理解能力。
cs.CV / 277 / 2609.34206
WorldGuide: Learning Success-Failure Boundaries in Latent World Models for Vision-Language-Action Policies
WorldGuide：在潜在世界模型中学习视觉-语言-行动策略的成功-失败边界
Abstract
Latent world models offer a promising way to improve Vision-Language-Action policies by capturing the consequences of actions. However, models trained primarily on expert demonstrations have limited exposure to failure outcomes and may struggle to distinguish visually similar successful and failed interactions. We propose \textbf{WorldGuide}, a framework that learns these distinctions in latent space and uses them to guide policy training. WorldGuide combines predictive pretraining on successful and failed trajectories with contrastive learning on matched success--failure pairs. The learned predictor then provides a differentiable reward to guide joint optimization of the policy and visual encoder. The predictor is discarded after training, so deployment requires no additional world-model inference. Extensive experiments show that WorldGuide substantially improves VLA reliability and achieves state of the art performance on LIBERO 100 and SimplerEnv, reaching \textbf{96.8\%} and \textbf{72.0\%}, respectively. Code will be publicly available.
Chinese Translation
潜在世界模型为通过捕捉行动后果来改善视觉-语言-行动（VLA）策略提供了一种有前景的方法。然而，主要基于专家演示训练的模型对失败结果的接触有限，可能难以区分视觉上相似的成功和失败交互。我们提出了 extbf{WorldGuide}，一个在潜在空间中学习这些区分并利用它们指导策略训练的框架。WorldGuide结合了对成功和失败轨迹的预测预训练以及对匹配的成功-失败对的对比学习。学习到的预测器随后提供可微分的奖励，以指导策略和视觉编码器的联合优化。训练后，预测器被丢弃，因此部署不需要额外的世界模型推理。大量实验表明，WorldGuide显著提高了VLA的可靠性，并在LIBERO 100和SimplerEnv上达到了最先进的性能，分别达到 extbf{96.8\%}和 extbf{72.0\\%}。代码将公开发布。
cs.CV / 278 / 2609.34221
WorldWeave: Growing Persistent Geometric Worlds for Video Generation
WorldWeave：用于视频生成的持久几何世界构建
Abstract
Despite rapid progress, world models still lack explicit, persistent structural memory, making it difficult to preserve consistent world structure during continual scene expansion and cross-view revisits. To address this limitation, we present WorldWeave, a world generation framework that decouples world-state maintenance from visual rendering. Specifically, WorldWeave combines continual elevation-map generation with agent-guided scene organization and stitching to build an expandable explicit 3D world that incrementally extends structural memory while preserving existing structure. First, its terrain module uses diffusion-based image outpainting to generate continuous metric elevation maps under neighborhood conditioning and boundary constraints. Next, an agent integrates user intent, terrain evidence, and cross-region connectivity constraints to construct scenes through hierarchical semantic planning, deterministic geometry compilation, and local revision. Finally, during visual generation, planned camera trajectories query world geometry through a read-only interface, producing depth sequences that guide video synthesis without writing the generated results back into the world state. As a result, structural memory remains independent of short-window video generation, enabling continual expansion without predefined map boundaries and providing a consistent geometric basis for observations across trajectories and repeated visits.
Chinese Translation
尽管取得了快速进展，世界模型仍然缺乏明确的、持久的结构记忆，这使得在持续场景扩展和跨视角重访过程中保持一致的世界结构变得困难。为了解决这一限制，我们提出了WorldWeave，一个将世界状态维护与视觉渲染解耦的世界生成框架。具体而言，WorldWeave结合了持续的高程图生成与代理引导的场景组织和拼接，以构建一个可扩展的显式3D世界，该世界在保留现有结构的同时逐步扩展结构记忆。首先，其地形模块利用基于扩散的图像外推生成连续的度量高程图，遵循邻域条件和边界约束。接下来，代理整合用户意图、地形证据和跨区域连通性约束，通过分层语义规划、确定性几何编译和局部修正构建场景。最后，在视觉生成过程中，规划的相机轨迹通过只读接口查询世界几何，生成深度序列以指导视频合成，而不将生成的结果写回世界状态。因此，结构记忆独立于短时间窗口的视频生成，使得在没有预定义地图边界的情况下实现持续扩展，并为跨轨迹和重复访问的观察提供一致的几何基础。
cs.CV / 279 / 2609.34223
Uncovering Ordinal-Matching Bias in Audio-Visual LLMs
揭示音视频大型语言模型中的序数匹配偏差
Abstract
This work aims to improve how audio-visual large language models (AVLLMs) associate speech with the correct visible speaker in multi-speaker scenes. We find that current AVLLMs frequently fail at this task, and analyze the nature of these failures. To this end, we construct a synthetic diagnostic dataset in which multiple visible speakers each utter a single word. Analysis on this corpus reveals a consistent error pattern across three recent open-source AVLLMs: models attribute utterances by simply matching the order of spoken sentences with the left-to-right, top-to-bottom arrangement of visible faces, rather than relying on audio-visual cues such as lip synchronization. We term this behavior \emph{ordinal-matching bias}. We further show that this bias can be substantially mitigated through a simple remedy, Ordinal-Decoupled Fine-Tuning (OD-FT), in which models are fine-tuned on synthetic videos where spatial positions of speakers and speaking order are independently randomized. Despite using only 400 synthetic training videos, OD-FT not only suppresses ordinal-matching bias but also improves audio-visual understanding on real-world videos, yielding average gains of 8.27\% for Qwen2.5-Omni and 2.57\% for video-SALMONN2+ across three audio-visual benchmarks.
Chinese Translation
本研究旨在改善音视频大型语言模型（AVLLMs）在多说话场景中将语音与正确可见说话者关联的能力。我们发现当前的AVLLMs在这一任务上经常失败，并分析了这些失败的性质。为此，我们构建了一个合成诊断数据集，其中多个可见说话者各自说出一个单词。对该语料库的分析揭示了三个近期开源AVLLMs中一致的错误模式：模型通过简单地将所说句子的顺序与可见面孔的从左到右、从上到下的排列进行匹配，而不是依赖于诸如唇部同步等音视频线索。我们将这种行为称为“序数匹配偏差”（ordinal-matching bias）。我们进一步表明，通过一种简单的补救措施——序数解耦微调（Ordinal-Decoupled Fine-Tuning, OD-FT），这种偏差可以显著减轻。在这种方法中，模型在合成视频上进行微调，其中说话者的空间位置和说话顺序是独立随机化的。尽管仅使用了400个合成训练视频，OD-FT不仅抑制了序数匹配偏差，还改善了对真实视频的音视频理解，在三个音视频基准测试中，Qwen2.5-Omni的平均提升为8.27%，video-SALMONN2+的平均提升为2.57%。
cs.CV / 280 / 2609.34231
ReGDiff: Guided Diffusion in Regulated Latent Space for Exploring Metamaterial Voxel Geometry
ReGDiff：在受限潜在空间中引导扩散以探索超材料体素几何
Abstract
Metamaterials are artificially engineered structures whose mechanical and physical behaviors are strongly shaped by geometry rather than composition. Voxel representation provides a unified format for metamaterial geometry generation, as it can express diverse classes such as truss, shell, and porous structures within a single cubic discretization. However, voxel-based generation faces a plausibility-novelty trade-off: staying close to known geometries helps preserve geometric regularities, while moving away from them is necessary for novelty but may produce degenerate geometries. To address this challenge, we propose REGDIFF, a generative framework that couples voxel representation with latent space regulation and guided diffusion. REGDIFF introduces a repel-and-sink (RAS) mechanism to smooth the latent distribution of plausible geometries, and short-range repulsion (SRR) guidance to discourage generation overly close to known samples while maintaining geometric plausibility. We further contribute a voxel-based benchmark covering truss- and shell-type metamaterial geometries, together with an evaluation module for geometric plausibility, novelty, and diversity. Experiments show that REGDIFF outperforms voxel-based generative baselines, achieving +8.9% in geometric plausibility, +46.4% in novelty, and +128.6% in diversity on average across two datasets. These results suggest that REGDIFF is a strong geometry candidate generator for downstream evaluation. Our code is provided at https://github.com/wzhan24/ReGDiff.
Chinese Translation
超材料是人工设计的结构，其机械和物理行为主要受几何形状而非成分的影响。体素表示为超材料几何生成提供了一种统一的格式，因为它可以在单一立方体离散化中表达多种类别，如桁架、壳体和多孔结构。然而，基于体素的生成面临着合理性与新颖性的权衡：保持接近已知几何形状有助于保持几何规律性，而远离它们则是实现新颖性所必需的，但可能会产生退化的几何形状。为了解决这一挑战，我们提出了REGDIFF，一个将体素表示与潜在空间调节和引导扩散相结合的生成框架。REGDIFF引入了一种排斥与沉降（RAS）机制，以平滑合理几何形状的潜在分布，并采用短程排斥（SRR）引导，以避免生成过于接近已知样本的几何形状，同时保持几何合理性。我们还贡献了一个覆盖桁架和壳体型超材料几何的基于体素的基准，并提供了一个用于几何合理性、新颖性和多样性的评估模块。实验表明，REGDIFF在体素生成基线模型上表现优异，在两个数据集上平均实现了+8.9%的几何合理性、+46.4%的新颖性和+128.6%的多样性。这些结果表明REGDIFF是下游评估的强几何候选生成器。我们的代码可在https://github.com/wzhan24/ReGDiff获取。
cs.CV / 281 / 2609.34232
Trustworthy synthetic visual media: Evidence across the media lifecycle
可信的合成视觉媒体：跨媒体生命周期的证据
Abstract
Images and videos have long helped people understand what happened and how a work came into being. Generative systems complicate that role. Realistic media can now be produced and revised without leaving a stable history, so appearance no longer reveals whether a scene was captured, synthesized, or altered along the way. Trust must instead come from evidence that explains the path an asset has taken and the circumstances in which it was used. Some of this evidence can be recovered from the media, while some must be recorded during production and preserved as the asset circulates. This review brings those approaches together and asks when their claims remain meaningful after ordinary processing or deliberate manipulation. We argue that trustworthy media do not depend on one universal marker of authenticity. The evidence must suit the question at hand, reach the person making the judgment, and remain open to correction when better information emerges. The larger goal is to keep the history of media intelligible even as the media itself continues to change.
Chinese Translation
图像和视频长期以来帮助人们理解发生了什么以及一项作品是如何产生的。生成系统使这一角色变得复杂。现实主义媒体现在可以在没有稳定历史记录的情况下被生产和修订，因此外观不再能揭示一个场景是被捕捉、合成还是在过程中被修改。信任必须来自于能够解释资产所经历路径及其使用环境的证据。这些证据中的一部分可以从媒体中恢复，而另一部分则必须在生产过程中记录并在资产流通时保存。本文综述了这些方法，并探讨在普通处理或故意操控之后，它们的主张何时仍然具有意义。我们认为，可信的媒体并不依赖于一个普遍的真实性标记。证据必须适合当前的问题，能够到达做出判断的人，并在更好的信息出现时保持开放以进行修正。更大的目标是使媒体的历史在媒体本身持续变化的同时仍然能够被理解。
cs.CV / 282 / 2609.34235
SegBanana: Steering Unified Multimodal Models into Medical Segmenters
SegBanana：引导统一多模态模型进入医学分割领域
Abstract
Medical image segmentation remains challenging in practical deployment, as models often struggle to generalize beyond the distributions covered by their training data and high-quality pixel-level annotations are typically unavailable for adaptation. Inspired by the cross-task transferability of large language models, we investigate whether unified multimodal models (UMMs) can transfer their pretrained visual understanding, reasoning, and generation capabilities to medical image segmentation without task-specific post-training. By recasting segmentation as structured visual generation, we find that frontier UMMs (e.g., Nano Banana) already exhibit basic segmentation capabilities across diverse clinical scenarios, but still struggle with challenging tasks requiring specialized anatomical or domain-specific knowledge. We further show that these limitations can be effectively mitigated by incorporating visual anatomical knowledge from in-context exemplars, expanding candidate solutions through repeated sampling, and refining suboptimal predictions via targeted editing.Motivated by these observations, we propose SegBanana, to our knowledge, the first agentic visual generation framework for training-free medical image segmentation. SegBanana builds on a frozen UMM as the core generative model, augmented with Anatomy-Aware Knowledge Retrieval and Comparative Quality Critique to unlock its potential segmentation capability. A State-Aware Multimodal Controller maintains structured state and iteratively orchestrates these tools, repeatedly refining intermediate predictions toward higher-quality masks. Across eight medical segmentation datasets, SegBanana achieves an average mDice of 77.45%, outperforming representative generalist (SAM3 and SegGPT) and medical-specific (BiomedParse and MedSAM3) baselines by at least 14.93 points, while remaining robust to out-of-domain visual supports.
Chinese Translation
医学图像分割在实际应用中仍然面临挑战，因为模型往往难以超越其训练数据所覆盖的分布，并且通常缺乏高质量的像素级注释以进行适应。受到大型语言模型跨任务可迁移性的启发，我们研究了统一多模态模型（UMMs）是否能够在没有特定任务后训练的情况下，将其预训练的视觉理解、推理和生成能力转移到医学图像分割中。通过将分割重新表述为结构化视觉生成，我们发现前沿的UMMs（例如，Nano Banana）在多种临床场景中已经展现出基本的分割能力，但在需要专业解剖或领域特定知识的挑战性任务中仍然存在困难。我们进一步表明，通过结合来自上下文示例的视觉解剖知识、通过重复采样扩展候选解决方案，以及通过针对性编辑来优化次优预测，这些局限性可以得到有效缓解。基于这些观察，我们提出了SegBanana，作为我们所知的首个无训练医学图像分割的自主视觉生成框架。SegBanana以一个冻结的UMM作为核心生成模型，并结合解剖知识检索和比较质量评估，以释放其潜在的分割能力。状态感知多模态控制器维护结构化状态，并迭代协调这些工具，反复优化中间预测，以获得更高质量的掩膜。在八个医学分割数据集上，SegBanana的平均mDice达到77.45%，至少比代表性的通用基线（SAM3和SegGPT）和医学特定基线（BiomedParse和MedSAM3）高出14.93个百分点，同时对域外视觉支持保持鲁棒性。
cs.CV / 283 / 2609.34237
DecFlowEdit: Self-Localized Flow-based Image Editing via Guidance Decoupling
DecFlowEdit：通过引导解耦实现自我定位的基于流的图像编辑
Abstract
Flow-based image editing (FlowEdit) enables inversion-free semantic changes through the difference between source and target velocities. In this paper, we observe that FlowEdit's default classifier-free guidance (CFG) configuration, with asymmetric source and target scales, causes substantial background leakage. Matching these guidance scales, for example by removing CFG, improves edit-relevant localization but severely degrades editability. To get the best of both worlds, we propose DecFlowEdit, which decouples the optimal guidance scales for localization and for editing in flow-based generative models. In particular, DecFlowEdit first extracts an edit-relevant prior by temporally aggregating velocity differences evaluated without CFG, and then uses this prior to reweight the original updates under default CFG. Our method remains training-free and inversion-free, requiring neither external spatial masks nor attention manipulation. Experiments on PIE-Bench across FLUX, SD3, and SD3.5 show that DecFlowEdit improves background preservation, reducing structure distance by approximately 61 to 73 percent and background LPIPS by 68 to 80 percent relative to FlowEdit at comparable editing fidelity.
Chinese Translation
基于流的图像编辑（FlowEdit）通过源速度和目标速度之间的差异实现无反演的语义变化。本文观察到，FlowEdit的默认无分类器引导（CFG）配置由于源和目标尺度的不对称性，导致了显著的背景泄漏。通过匹配这些引导尺度，例如通过去除CFG，可以改善与编辑相关的定位，但会严重降低可编辑性。为了兼顾两者的优点，我们提出了DecFlowEdit，它在基于流的生成模型中解耦了定位和编辑的最佳引导尺度。具体而言，DecFlowEdit首先通过对未使用CFG评估的速度差异进行时间聚合，提取与编辑相关的先验信息，然后利用该先验信息对默认CFG下的原始更新进行重新加权。我们的方法保持无训练和无反演，不需要外部空间掩模或注意力操作。在FLUX、SD3和SD3.5的PIE-Bench实验中，DecFlowEdit在背景保留方面表现出色，相较于FlowEdit在可比编辑保真度下，结构距离减少约61%至73%，背景LPIPS减少68%至80%。
cs.CV / 284 / 2609.34271
Scaling Versatile 3D Assets Editing with a Million-Scale Dataset
基于百万规模数据集的多功能3D资产编辑的扩展
Abstract
Although recent 3D generative models produce increasingly realistic assets, controllable 3D asset editing remains challenging. Existing methods are limited by scarce training data, insufficient source-aware modeling, and a lack of practical evaluation protocols. To address these limitations, we present Alchemy3D, a unified framework for training and evaluating versatile 3D asset editors that covers data construction, model architecture, and benchmark evaluation. Specifically, we curate Alchemy3D-1M, a large-scale 3D editing dataset containing 1.25M assets and 1.38M editing pairs across seven editing types. On this data, we train a family of generative flow models for general-purpose 3D asset editing. The model family supports image- and text-conditioned editing, few-step inference, and transfer to multi-view 3D part segmentation. We further introduce GEdit3D-Bench, a large-scale, open-world benchmark with a multi-dimensional evaluation protocol. Across existing and newly introduced benchmarks, our method outperforms prior methods on most metrics of editing fidelity, source preservation, and visual quality.
Chinese Translation
尽管近期的3D生成模型能够生成越来越逼真的资产，但可控的3D资产编辑仍然面临挑战。现有方法受限于稀缺的训练数据、缺乏源感知建模以及缺乏实用的评估协议。为了解决这些限制，我们提出了Alchemy3D，这是一个统一的框架，用于训练和评估多功能3D资产编辑器，涵盖数据构建、模型架构和基准评估。具体而言，我们策划了Alchemy3D-1M，这是一个大规模的3D编辑数据集，包含125万个资产和138万个编辑对，涵盖七种编辑类型。在此数据上，我们训练了一系列用于通用3D资产编辑的生成流模型。该模型系列支持图像和文本条件的编辑、少步推理，并可转移到多视角3D部件分割。我们进一步推出了GEdit3D-Bench，这是一个大规模的开放世界基准，具有多维评估协议。在现有和新引入的基准测试中，我们的方法在大多数编辑保真度、源保留和视觉质量的指标上优于先前的方法。
cs.CV / 285 / 2609.34277
See, Measure, and Reason: Learning Visually Grounded Reasoning in Pathology
观察、测量与推理：在病理学中学习视觉基础的推理
Abstract
Pathological assessment relies on recognizing fine-grained visual details in histological images. Vision-language models (VLMs) increasingly support pathology interpretation, yet their ability to perceive these details remains inadequate. This weakness leads to inaccurate cellular observations that can persist even when final answers are correct. In this paper, we propose ASPECT to improve visually grounded reasoning through explicit supervision of cellular appearance and abundance. ASPECT trains intermediate visual tokens through pathology feature reconstruction, cell feature alignment, and count supervision. Three-stage supervised fine-tuning teaches the model to perceive, generate visual tokens, and reason, followed by reinforcement learning that rewards answer correctness and consistency with reported measurements. We also introduce PathoVernier, a benchmark of 759 expert-reviewed questions from five pathology datasets covering four cellular composition tasks. It evaluates both final answers and intermediate measurements to expose errors hidden by answer accuracy. On PathoVernier, ASPECT achieves relative accuracy gains of approximately 19.2% over the strongest baseline, Gemini-3.1-Pro, and 99.3% over its Qwen3-VL-8B backbone, while reducing RAWR, which measures counting errors within correct responses, by 28.1% and 42.7%, respectively. ASPECT also improves over its backbone on three external pathology benchmarks covering classification and question answering beyond cellular composition tasks.
Chinese Translation
病理评估依赖于识别组织学图像中的细微视觉细节。视觉-语言模型（VLMs）越来越多地支持病理解释，但它们感知这些细节的能力仍然不足。这一弱点导致细胞观察不准确，即使最终答案正确也可能持续存在。在本文中，我们提出了ASPECT，以通过对细胞外观和丰度的明确监督来改善视觉基础的推理。ASPECT通过病理特征重建、细胞特征对齐和计数监督来训练中间视觉标记。三阶段的监督微调教会模型感知、生成视觉标记和推理，随后进行强化学习，以奖励答案的正确性和与报告测量的一致性。我们还引入了PathoVernier，这是一个包含来自五个病理数据集的759个专家审查问题的基准，涵盖四个细胞组成任务。它评估最终答案和中间测量，以揭示被答案准确性掩盖的错误。在PathoVernier上，ASPECT相较于最强基线Gemini-3.1-Pro实现了约19.2%的相对准确性提升，相较于其Qwen3-VL-8B主干网络则提升了99.3%，同时将RAWR（测量正确响应中的计数错误）分别降低了28.1%和42.7%。ASPECT在三个外部病理基准上也优于其主干，涵盖了超出细胞组成任务的分类和问答。
cs.CV / 286 / 2609.34286
Dexterous Tactile World Model
灵巧触觉世界模型
Abstract
World models for manipulation are typically trained from video, yet the events that determine how manipulation unfolds, such as making and releasing contact, are difficult to observe visually and are often easier to sense through touch. We present the Dexterous Tactile World Model (DTWM), a video world model for future-frame prediction of egocentric manipulation from both observed video and tactile signals from a glove worn on each hand. We condition a pretrained video diffusion transformer on each hand's tactile signal through a zero-initialized residual at the corresponding hand location in the video tokens, while a causal mask prevents predicted frames from accessing future information. Compared with a vision-only model matched in architecture, parameters, and training, DTWM reduces the underestimation of hand motion from 23% to 9%, while reducing the perceptual error in the hand region by 7.4% across three training runs per model. The benefit also increases over the prediction horizon, with the improvement in the later predicted chunks being about 4.1x larger than in the first. DTWM also outperforms other visual-tactile world models under the same setting, and training with touch improves future-frame prediction even when no touch is available at inference. Ablations show that the model benefits from both the magnitude and spatial location of force: replacing the tactile signal with binary contact states, either per hand or per location, increases prediction error. The observed course of the force indicates whether the interaction will persist or change.
Chinese Translation
操控的世界模型通常是从视频中训练而来，然而决定操控展开的事件，例如接触和释放接触，往往难以通过视觉观察，且通过触觉感知更为容易。我们提出了灵巧触觉世界模型（Dexterous Tactile World Model, DTWM），这是一个用于从观察到的视频和佩戴在每只手上的手套的触觉信号中进行自我中心操控未来帧预测的视频世界模型。我们通过在视频标记中相应手的位置使用零初始化的残差，将预训练的视频扩散变换器与每只手的触觉信号相结合，同时因果掩码防止预测帧访问未来信息。与架构、参数和训练相匹配的仅视觉模型相比，DTWM将手部运动的低估从23%降低到9%，同时在每个模型的三次训练运行中，手部区域的感知误差减少了7.4%。这种益处在预测范围内也随之增加，后期预测块的改进约为前期的4.1倍。DTWM在相同设置下也优于其他视觉-触觉世界模型，并且即使在推理时没有触觉可用，触觉训练仍能改善未来帧预测。消融实验表明，该模型受益于力的大小和空间位置：用每只手或每个位置的二元接触状态替代触觉信号，会增加预测误差。观察到的力的变化表明交互将持续或改变。
cs.CV / 287 / 2609.34294
Semantic Modality Compensation for Unsupervised Visible-Infrared Person Re-identification under Unpaired Settings
无监督可见-红外行人重识别中的语义模态补偿
Abstract
Unsupervised visible-infrared person re-identification (USL-VI-ReID) learns person representations that can be compared across modalities without identity annotations. In the unpaired setting, however, identity correspondences between modalities are often incomplete, leaving many identities without an observed counterpart in the other modality. Existing unpaired methods bridge this gap by generating or mapping features for the other modality, mainly by exploiting the statistics of visual features without explicitly separating content that is discriminative for identity from style that is specific to modality. Consequently, the generated features may distort identity cues or inherit bias from the source modality, undermining the reliability of supervision across modalities. We formulate unpaired learning across modalities as a semantic compensation problem and propose Semantic Modality Compensation (SMC), a framework based on prompt composition that decouples identity semantics from modality style within a shared visual semantic space. SMC first constructs a discriminative ReID space through augmented dual contrastive learning, yielding pseudo labels, cluster prototypes, and memory banks for each modality. It then learns visible and infrared modality prompts in the CLIP semantic space and maps clusters obtained from pseudo labels to identity semantic tokens. For each cluster lacking a reliable match in the other modality, SMC combines its identity token with the prompt for the target modality to synthesize a semantic counterpart in the missing modality. The synthesized counterpart is then projected back into the ReID space and injected into a compensation memory through confidence gating. Extensive experiments under both paired and unpaired settings demonstrate that SMC consistently outperforms state-of-the-art methods, with particularly large gains when identity mismatch is severe.
Chinese Translation
无监督可见-红外行人重识别（USL-VI-ReID）学习可以跨模态比较的行人表征，而无需身份注释。然而，在无配对设置中，模态之间的身份对应关系往往不完整，导致许多身份在另一模态中没有观察到的对应物。现有的无配对方法通过生成或映射另一模态的特征来弥补这一差距，主要是通过利用视觉特征的统计信息，而未明确区分对身份具有判别性的内容与特定于模态的风格。因此，生成的特征可能会扭曲身份线索或继承源模态的偏差，从而削弱跨模态监督的可靠性。我们将跨模态的无配对学习形式化为一个语义补偿问题，并提出了语义模态补偿（Semantic Modality Compensation, SMC）框架，该框架基于提示组合，将身份语义与模态风格解耦，置于共享的视觉语义空间中。SMC首先通过增强的双重对比学习构建一个判别性的重识别空间，生成伪标签、聚类原型和每个模态的记忆库。然后，它在CLIP语义空间中学习可见和红外模态提示，并将从伪标签获得的聚类映射到身份语义标记。对于每个在另一模态中缺乏可靠匹配的聚类，SMC将其身份标记与目标模态的提示结合，以合成缺失模态中的语义对应物。合成的对应物随后被投影回重识别空间，并通过置信度门控注入到补偿记忆中。在配对和无配对设置下的广泛实验表明，SMC始终优于最先进的方法，尤其是在身份不匹配严重时，表现出显著的提升。
cs.CV / 288 / 2609.34299
PSM: Dataset Distillation Based on Precise Statistical Matching by Difficulty
PSM：基于难度的精确统计匹配的数据集蒸馏
Abstract
Dataset distillation (DD) condenses a large original dataset into a small distilled dataset with high training utility. Decoupled statistical matching methods substantially reduce distillation time and memory overhead while achieving strong performance. However, they typically supervise all distilled samples using running statistics estimated from the entire original dataset. These statistics mainly capture the average feature distribution while overlooking differences in sample difficulty, limiting their ability to characterize the difficulty structure of the original data. To address this issue, we propose Precise Statistical Matching (PSM) by difficulty. After pretraining, PSM uses the Global Precision Score (GPS) to estimate image difficulty, ranks the samples within each class, and partitions each class into IPC (images per class) difficulty groups. During distillation, Statistics Updated Again (SUA) updates the teacher's batch normalization (BN) running statistics through forward passes on original samples from each group, providing difficulty-specific supervision for the corresponding distilled batch. Meanwhile, Initial Sample Screening (ISS) initializes distilled samples using original images from the corresponding difficulty group, providing an effective starting point for precise matching. Experiments across multiple datasets and model architectures demonstrate that PSM broadens the difficulty range of distilled samples and improves downstream performance in most evaluated settings. Code will be released.
Chinese Translation
数据集蒸馏（Dataset Distillation, DD）将一个大型原始数据集浓缩为一个具有高训练效用的小型蒸馏数据集。解耦的统计匹配方法在实现强大性能的同时，显著减少了蒸馏时间和内存开销。然而，它们通常使用从整个原始数据集中估计的运行统计来监督所有蒸馏样本。这些统计主要捕捉平均特征分布，而忽视了样本难度的差异，限制了它们表征原始数据难度结构的能力。为了解决这个问题，我们提出了基于难度的精确统计匹配（Precise Statistical Matching, PSM）。在预训练后，PSM使用全局精度评分（Global Precision Score, GPS）来估计图像难度，对每个类别内的样本进行排名，并将每个类别划分为IPC（每类图像数量）难度组。在蒸馏过程中，统计更新（Statistics Updated Again, SUA）通过对每组的原始样本进行前向传播，更新教师的批量归一化（Batch Normalization, BN）运行统计，为相应的蒸馏批次提供特定于难度的监督。同时，初始样本筛选（Initial Sample Screening, ISS）使用来自相应难度组的原始图像初始化蒸馏样本，为精确匹配提供有效的起始点。跨多个数据集和模型架构的实验表明，PSM拓宽了蒸馏样本的难度范围，并在大多数评估设置中提高了下游性能。代码将会发布。
cs.CV / 289 / 2609.34302
CAT-Free: Multi-View Pedestrian Localization without Calibration, Annotations, or Target-Scene Training via Adaptive Geometric Filtering
CAT-Free：通过自适应几何滤波实现无校准、无标注或目标场景训练的多视角行人定位
Abstract
Multi-camera pedestrian localization is useful for wide-area monitoring in public and commercial spaces. However, deploying these systems often requires considerable setup for each new environment. Existing methods typically require camera calibration, position annotations, or target-scene training. CAT-Free removes all three requirements. It uses synchronized RGB video as its only scene-specific input. Camera configuration is estimated directly from the video. Pedestrian locations are then estimated by combining observations from multiple cameras. Automatic camera estimation is not always accurate. This can produce unreliable pedestrian locations. CAT-Free therefore introduces two adaptive geometric filters. They remove unreliable position estimates. Their thresholds are estimated from each input sequence. CAT-Free achieves 82.5, 84.5, and 65.7 MODA on WildTrack, MultiviewX, and GMVD. It uses no supplied calibration, position annotations, or target-scene training. Published methods using such scene-specific information report 88.2--95.0 MODA on WildTrack and 83.9--96.5 on MultiviewX under their respective protocols. CAT-Free also transfers without retuning. It reaches 74.9 MODA on four additional sequences and 78.6 on an unseen 8-camera installation. Finally, localization uncertainty predicts MODA with $r=-0.98$. This provides a label-free estimate of localization reliability.
Chinese Translation
多摄像头行人定位在公共和商业空间的广域监控中非常有用。然而，部署这些系统通常需要针对每个新环境进行大量的设置。现有方法通常需要摄像头校准、位置标注或目标场景训练。CAT-Free 消除了这三项要求。它仅使用同步的 RGB 视频作为唯一的场景特定输入。摄像头配置直接从视频中估计。然后，通过结合来自多个摄像头的观察来估计行人位置。自动摄像头估计并不总是准确，这可能导致不可靠的行人位置。因此，CAT-Free 引入了两个自适应几何滤波器，以去除不可靠的位置估计。它们的阈值是从每个输入序列中估计的。CAT-Free 在 WildTrack、MultiviewX 和 GMVD 上分别达到了 82.5、84.5 和 65.7 的 MODA。它不使用任何提供的校准、位置标注或目标场景训练。使用此类场景特定信息的已发布方法在 WildTrack 上报告了 88.2--95.0 的 MODA，在 MultiviewX 上报告了 83.9--96.5 的 MODA，均符合其各自的协议。CAT-Free 还可以在不重新调优的情况下进行迁移。在四个额外序列上达到了 74.9 的 MODA，并在一个未见的 8 摄像头安装上达到了 78.6 的 MODA。最后，定位不确定性以 $r=-0.98$ 预测 MODA。这提供了一个无标签的定位可靠性估计。
cs.CV / 290 / 2609.34309
MaLiang-Harness: A Programmable Path to Image and Video Generation
MaLiang-Harness：图像和视频生成的可编程路径
Abstract
Executable programs offer explicit control over how images and videos are constructed, but generating runnable code is only the beginning of visual creation. A program can execute correctly while violating the requested composition, appearance, or motion. We define this discrepancy as the Program-to-Visual (P2V) gap and introduce MaLiang-Harness, a unified framework for organizing MLLM-driven visual generation into a persistent process of construction, inspection, and revision. Its central design is to make the evolving visual program, its construction history, and its verification share a common revision reference. We define the Persistent Executable Generation (PEG) state as preserving programs and task context. Traceable Generation Process (TGP) connects edits to rendered evidence, and Revision-aware Editing and Verification (REV) supports restoration and checks the current revision before completion. Together, these mechanisms coordinate planning, execution, and visual feedback across rendering backends. We evaluate 11 powerful closed-source MLLMs on MaLiang-IBench and four on MaLiang-VBench, measuring generation success, visual quality, and computational cost. GPT-6-Astra achieves 100% generation success on both benchmarks, with 96.0% of image tasks and 76.9% of video tasks meeting all quality thresholds. The comparison also reveals a mismatch between general capability scores and visual generation performance, with similarly scored models differing substantially in their ability to satisfy visual requirements. MaLiang-Harness provides a systematic basis for studying how MLLMs translate executable code into visual outcomes, exposing both the potential of programmable generation and the limitations of general benchmarks as predictors of this ability. The project is available at https://github.com/gulucaptain/MaLiang-Harness.
Chinese Translation
可执行程序提供了对图像和视频构建方式的明确控制，但生成可运行代码仅仅是视觉创作的开始。一个程序可以在违反请求的构图、外观或运动的情况下正确执行。我们将这种差异定义为程序到视觉（Program-to-Visual, P2V）差距，并介绍了MaLiang-Harness，这是一个统一框架，用于将基于MLLM（多模态大语言模型）驱动的视觉生成组织成一个持续的构建、检查和修订过程。其核心设计是使不断演变的视觉程序、其构建历史和验证共享一个共同的修订参考。我们将持久可执行生成（Persistent Executable Generation, PEG）状态定义为保留程序和任务上下文。可追踪生成过程（Traceable Generation Process, TGP）将编辑与渲染证据连接起来，而关注修订的编辑和验证（Revision-aware Editing and Verification, REV）支持恢复并在完成之前检查当前修订。这些机制共同协调了渲染后端的规划、执行和视觉反馈。我们在MaLiang-IBench上评估了11个强大的闭源MLLM，在MaLiang-VBench上评估了4个，测量生成成功率、视觉质量和计算成本。GPT-6-Astra在两个基准上均实现了100%的生成成功率，其中96.0%的图像任务和76.9%的视频任务满足所有质量标准。比较还揭示了通用能力评分与视觉生成性能之间的不匹配，得分相似的模型在满足视觉要求的能力上存在显著差异。MaLiang-Harness为研究MLLM如何将可执行代码转化为视觉结果提供了系统基础，揭示了可编程生成的潜力以及通用基准作为这一能力预测者的局限性。该项目可在https://github.com/gulucaptain/MaLiang-Harness获取。
cs.CV / 291 / 2609.34314
PlaylistEval: Can Video-Language Judges Be Trusted at Day Scale and Beyond?
PlaylistEval：视频语言评判者在超长视频中的可信度如何？
Abstract
Video-language models are increasingly used as judges of video understanding, both for evaluating model outputs and for training reward models. Whether their judgments remain reliable when the evidence is buried in day-long videos has yet to be established. Existing benchmarks cannot answer this. Their videos are typically only a few minutes long, many answer pairs can be separated from the transcript alone, and collecting human judgments does not scale to ultra-long videos. We introduce PlaylistEval, an agentic framework that builds video-language judge benchmarks over 100-hour playlist collection without human annotation. It automatically generates questions with paired answers whose differences are controlled by causal degradation, so that every pair demands retrieval across the collection. The resulting benchmark contains 630 pairs across seven domains spanning both static and dynamic knowledge, and on a stratified subset of 152 pairs it agrees with human judgments 93.0% of the time (IAA 0.781). Evaluating 17 omnimodal and multimodal models from eight families reveals that frontier judges reach only 75.4% pairwise accuracy, while open-source judge models perform far behind. We further show that both retrieval and final judgment depend on using multiple modalities, and that judge accuracy degrades as the playlist set grows. We release our pipeline, benchmark, and evaluation code at https://playlisteval.github.io.
Chinese Translation
视频语言模型越来越多地被用作视频理解的评判者，既用于评估模型输出，也用于训练奖励模型。然而，当证据埋藏在长达一天的视频中时，它们的判断是否仍然可靠尚未得到证实。现有基准无法回答这个问题。它们的视频通常只有几分钟长，许多问答对可以仅通过转录文本分离，而收集人类判断在超长视频中并不具备可扩展性。我们引入了PlaylistEval，一个自主框架，构建了一个超过100小时播放列表集合的视频语言评判基准，无需人类注释。它自动生成问题及配对答案，其差异通过因果降解进行控制，以便每对都需要在集合中进行检索。最终的基准包含630对跨越静态和动态知识的七个领域，在一个分层子集的152对中，它与人类判断的符合率为93.0%（IAA 0.781）。对来自八个家族的17个全模态和多模态模型的评估显示，前沿评判者仅达到75.4%的成对准确率，而开源评判模型的表现远远落后。我们进一步表明，检索和最终判断都依赖于使用多种模态，并且随着播放列表集的增长，评判准确性会下降。我们在 https://playlisteval.github.io 发布了我们的管道、基准和评估代码。
cs.CV / 292 / 2609.34319
Text-Vision Synergistic Token Caching: A Training-Free Framework for Efficient Vision-Language-Action Inference
文本-视觉协同令牌缓存：一种无训练的高效视觉-语言-动作推理框架
Abstract
Vision-Language-Action (VLA) models enable generalizable robotic control but remain computationally expensive. Token caching provides a training-free, plug-and-play acceleration alternative. However, existing VLA caching does not fully exploit a key inductive bias of VLA models: text-vision synergy, wherein textual semantics guide the precise visual grounding of task-relevant regions. In particular, existing designs insufficiently account for head-wise reliability in attention aggregation and layer-wise stability in cache reuse. To address this, we propose Text-Vision Synergistic Token Caching (TVCache), a training-free framework for efficient VLA inference. TVCache filters attention heads based on text-vision information focus to improve task-relevant and physically consistent visual grounding. Concurrently, we introduce a reuse-layer selection mechanism guided by text-vision entropy differences to avoid caching unstable representations and improve cache resource allocation. Extensive experiments across four representative VLA models, two simulation benchmarks, and real-world robotic tasks demonstrate the effectiveness and generality of TVCache. At matched token-retention ratios, TVCache consistently improves task success over existing VLA caching with comparable computational cost. On OpenVLA-OFT, it improves average success by up to 14.5 percentage points over VLA-Cache at 12.5% retention while reducing FLOPs by 2.45x relative to full-token inference.
Chinese Translation
视觉-语言-动作（VLA）模型使得通用机器人控制成为可能，但计算成本仍然较高。令牌缓存提供了一种无训练的即插即用加速替代方案。然而，现有的VLA缓存并未充分利用VLA模型的一个关键归纳偏差：文本-视觉协同，其中文本语义指导任务相关区域的精确视觉定位。特别是，现有设计在注意力聚合中的头部可靠性和缓存重用中的层级稳定性方面考虑不足。为了解决这个问题，我们提出了文本-视觉协同令牌缓存（TVCache），一种无训练的高效VLA推理框架。TVCache根据文本-视觉信息的关注度过滤注意力头，以改善任务相关和物理一致的视觉定位。同时，我们引入了一种基于文本-视觉熵差异的重用层选择机制，以避免缓存不稳定的表示并改善缓存资源分配。在四个代表性的VLA模型、两个模拟基准和现实世界的机器人任务上进行的广泛实验表明，TVCache的有效性和通用性。在匹配的令牌保留比率下，TVCache在可比的计算成本下，始终提高了任务成功率。在OpenVLA-OFT上，它在12.5%的保留率下，相比于VLA-Cache提高了平均成功率高达14.5个百分点，同时相对于完整令牌推理减少了2.45倍的FLOPs。
cs.CV / 293 / 2609.34325
DORA: Dynamic Online Reinforcement Agent for Token Pruning in Vision Transformers
DORA：用于视觉变换器中令牌修剪的动态在线强化代理
Abstract
Vision Transformers (ViTs) incur quadratic self-attention cost in the number of tokens. Most token-reduction methods adapt token identities within a prescribed layer-wise compression schedule, or search a static mask offline, and thus limit online adaptation of when and how much to prune. We propose DORA (Dynamic Online Reinforcement Agent), which learns an input-adaptive pruning policy itself for frozen ViTs. At each eligible block, a hierarchical actor decides whether to prune, how many tokens to remove, and which tokens to remove from each image's evolving representation. Because early deletions change the states observed by later decisions, DORA formulates pruning as a finite-horizon Markov decision process. Complete-prefix shadow evaluations convert final-prediction fidelity into localized per-step credit, while closed-loop accuracy feedback adjusts the fidelity penalty toward a shared accuracy-drop target. A privileged critic and all shadow computations are training-only. Deployment retains the frozen backbone and a lightweight actor that applies hard deletion and packed variable-length FlashAttention, converting token reduction into measured speedups. On ImageNet-1K with DeiT-Base, DORA reduces FLOPs by 38.4% relative to the uncompressed backbone within one percentage point of accuracy loss. Averaged across four ViT-type backbones at matched accuracy, DORA uses 13.2% fewer FLOPs and achieves 32.4% higher throughput than the corresponding per-backbone baseline means. Under zero-shot transfer to ImageNet-A, these gains widen to 20.3% and 45.6%, respectively.
Chinese Translation
视觉变换器（ViTs）在令牌数量上产生二次自注意力成本。大多数令牌减少方法在规定的层级压缩计划内调整令牌身份，或离线搜索静态掩码，从而限制了在线适应何时以及修剪多少的能力。我们提出了DORA（动态在线强化代理），它为冻结的ViTs学习自适应的修剪策略。在每个合适的块中，分层的行为者决定是否修剪、移除多少令牌以及从每个图像的演变表示中移除哪些令牌。由于早期删除会改变后续决策观察到的状态，DORA将修剪公式化为有限时域的马尔可夫决策过程。完整前缀阴影评估将最终预测的保真度转换为局部每步的信用，而闭环准确性反馈则调整保真度惩罚以达到共享的准确性下降目标。特权评论员和所有阴影计算仅用于训练。部署时保留冻结的主干和一个轻量级行为者，该行为者应用硬删除和打包可变长度的FlashAttention，将令牌减少转化为可测量的加速。在ImageNet-1K上使用DeiT-Base，DORA相对于未压缩的主干减少了38.4%的FLOPs，同时准确性损失在一个百分点以内。在四个ViT类型主干的匹配准确性下，DORA使用了13.2%更少的FLOPs，并实现了32.4%更高的吞吐量，相较于相应的每个主干基准均值。在零样本迁移到ImageNet-A时，这些增益分别扩大到20.3%和45.6%。
cs.CV / 294 / 2609.34330
MiCo: Mutual Information Coverage Optimization through Semantic Erasure Modeling for Efficient MLLM Inference
MiCo：通过语义擦除建模实现的互信息覆盖优化，以提高多模态大语言模型推理的效率
Abstract
Multimodal large language models (MLLMs) have demonstrated impressive performance in multimodal understanding, but processing large numbers of visual tokens results in high computational costs. While many methods have been proposed to reduce the number of visual tokens, most of them rely on heuristics and are prone to discarding substantial visual information during pruning, leading to degradation in model performance. In this work, by using a semantic erasure model, we derive a general mutual information coverage objective from task log-loss and propose MiCo, a training-free two-stage pruning method. MiCo first uses visual signals to select a representative candidate pool before visual tokens enter the language model, then performs task-aware subset selection within it. At each stage, suitable observable proxies instantiate the derived objective as a monotone submodular coverage function, which MiCo greedily optimizes under the token budget. MiCo is evaluated on diverse MLLMs ranging from 7B to 13B parameters across a broad range of image and video benchmarks spanning general visual reasoning, fine-grained OCR and grounding, hallucination detection, and long-video understanding. MiCo consistently achieves the best performance across nearly all evaluated models under all pruning ratios. On LLaVA-NEXT-13B, MiCo uses only 5.6% visual tokens, retains 97.5% of baseline performance, and achieves a 3.8-fold inference speedup. Our experiments demonstrate the effectiveness of MiCo and our mutual information coverage objective for visual token pruning.
Chinese Translation
多模态大语言模型（MLLMs）在多模态理解方面表现出色，但处理大量视觉标记会导致高计算成本。虽然已经提出了许多方法来减少视觉标记的数量，但大多数方法依赖于启发式，容易在剪枝过程中丢弃大量视觉信息，从而导致模型性能下降。在本研究中，我们通过使用语义擦除模型，从任务日志损失中推导出一般的互信息覆盖目标，并提出了MiCo，一种无训练的两阶段剪枝方法。MiCo首先利用视觉信号在视觉标记进入语言模型之前选择一个代表性的候选池，然后在其中执行任务感知的子集选择。在每个阶段，适当的可观察代理将推导出的目标实例化为单调子模覆盖函数，MiCo在标记预算下贪婪地优化该函数。MiCo在参数范围从7B到13B的多种MLLM上进行了评估，涵盖了一系列广泛的图像和视频基准，包括一般视觉推理、细粒度OCR和定位、幻觉检测以及长视频理解。MiCo在几乎所有评估模型和所有剪枝比例下始终实现最佳性能。在LLaVA-NEXT-13B上，MiCo仅使用5.6%的视觉标记，保留了97.5%的基线性能，并实现了3.8倍的推理加速。我们的实验展示了MiCo及其互信息覆盖目标在视觉标记剪枝中的有效性。
cs.CV / 295 / 2609.34335
SkillPE: Creativity-Oriented Cinematic Skill Evolution for Text-to-Video Prompt Engineering
SkillPE：面向创造力的电影技能演变用于文本到视频提示工程
Abstract
Achieving high-quality, cinematic results in text-to-video generation remains challenging for non-experts, whose prompts often lack professional narrative and creative design. We propose SkillPE, a prompt engineering (PE) framework that evolves reusable cinematic skills from expert-authored seeds. SkillPE represents shot logic, composition, lighting, sound design, and other filmmaking cues in a fine-grained format, and retrieves movie references categorized as resonators (good matches), dissonants (weak matches), and divergents (creatively useful near-misses). The first two refine when and how a skill should be applied, while divergents inspire alternative cinematic realizations at different degrees of modification while preserving the user intent. Candidate skills are assessed through generated videos along prompt fidelity, cinematic quality, narrative appeal, and creativity to construct the final skill libraries. Experiments on StoryEval and VBench show improvements of up to 1.40 points over the strongest external baseline and 0.51 points over seed skills on 7-point four-dimensional evaluation, while remaining competitive on benchmark-native metrics. Overall, SkillPE offers a practical approach to balancing fidelity and creativity in cinematic text-to-video generation. Code is available at https://github.com/Ais0n/SkillPE .
Chinese Translation
在文本到视频生成中实现高质量的电影效果对非专业人士仍然具有挑战性，因为他们的提示往往缺乏专业的叙事和创造性设计。我们提出了SkillPE，一种从专家撰写的种子中演变可重用电影技能的提示工程（PE）框架。SkillPE以细粒度格式表示镜头逻辑、构图、照明、声音设计和其他电影制作线索，并检索被分类为共鸣者（良好匹配）、不和谐者（弱匹配）和发散者（创造性有用的近似值）的电影参考。前两者用于细化技能应用的时机和方式，而发散者则在不同程度的修改下激发替代的电影实现，同时保留用户意图。候选技能通过生成的视频在提示保真度、电影质量、叙事吸引力和创造力等方面进行评估，以构建最终的技能库。在StoryEval和VBench上的实验显示，在7分制四维评估中，相较于最强的外部基线提高了最多1.40分，相较于种子技能提高了0.51分，同时在基准原生指标上保持竞争力。总体而言，SkillPE为在电影文本到视频生成中平衡保真度和创造力提供了一种实用的方法。代码可在 https://github.com/Ais0n/SkillPE 获取。
cs.CV / 296 / 2609.34346
E-WAVE: Event-based Continuous Optical Flow via Warping-Aligned Visual Encoding
E-WAVE：基于事件的连续光流估计通过变形对齐的视觉编码
Abstract
Temporally dense optical flow is essential for dynamic perception in immersive VR/AR systems, where rapid head, hand, and object motion must be continuously captured and tracked. Existing frame-based optical flow estimation methods are constrained by the tradeoff between temporal resolution and computational cost; while event cameras, with their high temporal resolution and energy efficiency, serve as a natural solution to the dilemma. However, event-based approaches commonly rely on correlation volumes to capture pairwise voxel correspondences, which incur substantial memory and computation overhead. We present E-WAVE, a correlation-free framework for high-temporal-resolution (HTR) optical flow estimation from event streams. Instead of constructing all-pairs correlation volumes, E-WAVE employs global attention mechanism to model long-range feature dependencies and performs trajectory guided feature warping using B\'ezier curve. Through iterative updates, it predicts trajectories that allow for querying at arbitrary timestamps without repeated inference. Experiments on MultiFlow and DSEC-Flow demonstrate a 25% lower trajectory error and comparable endpoint flow estimation accuracy relative to state-of-the art baselines. Additional evaluations on self-captured data using a head-mounted prototype validate that E-WAVE remains robust under challenging real-world conditions.
Chinese Translation
时间密集型光流对于沉浸式虚拟现实/增强现实系统中的动态感知至关重要，在这些系统中，快速的头部、手部和物体运动必须被持续捕捉和跟踪。现有的基于帧的光流估计方法受到时间分辨率与计算成本之间权衡的限制；而事件相机凭借其高时间分辨率和能效，成为解决这一困境的自然选择。然而，基于事件的方法通常依赖于相关体积来捕捉成对的体素对应关系，这会带来大量的内存和计算开销。我们提出了E-WAVE，一个无相关性的框架，用于从事件流中进行高时间分辨率（HTR）光流估计。E-WAVE不构建所有对的相关体积，而是采用全局注意力机制来建模长距离特征依赖关系，并使用Bézier曲线执行轨迹引导的特征变形。通过迭代更新，它预测轨迹，使得可以在任意时间戳进行查询而无需重复推理。在MultiFlow和DSEC-Flow上的实验表明，相较于最先进的基线，轨迹误差降低了25%，并且端点光流估计精度相当。对使用头戴式原型自捕获数据的额外评估验证了E-WAVE在具有挑战性的真实世界条件下的鲁棒性。
cs.CV / 297 / 2609.34363
SyncRA: Learning Temporal Correspondence in Omni-Modal Models
SyncRA：在全模态模型中学习时间对应关系
Abstract
Recent omni-modal models demonstrate strong perception of audio and visual inputs, yet often struggle to connect what they hear with what they see at the same moment. This weakness in temporal correspondence can cause models to associate spoken cues with the wrong visual scenes, producing plausible answers grounded in incorrect audio-visual pairings. We diagnose this problem through controlled temporal swaps, revealing that model answers do not reliably follow changes in these pairings. To address it, we propose Synchrony-Guided Representation Alignment (SyncRA), a lightweight method for strengthening temporal correspondence between audio and vision. Specifically, SyncRA contrasts intermediate audio-visual representations within each video, aligning matching moments while separating mismatched ones to capture local temporal correspondence within a shared global context. The objective derives supervision directly from existing input timing, requiring no additional annotations and leaving inference unchanged. We evaluate SyncRA across four open omni-modal models spanning different sizes and architectures on five public video benchmarks. SyncRA consistently outperforms answer-only fine-tuning across all model-benchmark combinations, while substantially improving the ability to track changing audio-visual pairings in controlled evaluations. These results demonstrate that lightweight, targeted supervision can effectively strengthen temporal correspondence and translate into broad improvements in audio-visual question answering.
Chinese Translation
最近的全模态模型在音频和视觉输入的感知上表现出色，但往往难以将他们所听到的与他们在同一时刻所看到的联系起来。这种时间对应关系的弱点可能导致模型将口语提示与错误的视觉场景关联，从而产生基于不正确音频-视觉配对的合理答案。我们通过控制时间交换来诊断这个问题，揭示模型的答案并不可靠地跟随这些配对的变化。为了解决这个问题，我们提出了同步引导表示对齐（Synchrony-Guided Representation Alignment，SyncRA），这是一种轻量级的方法，用于增强音频与视觉之间的时间对应关系。具体而言，SyncRA 在每个视频中对比中间的音频-视觉表示，对齐匹配的时刻，同时分离不匹配的时刻，以捕捉共享全局上下文中的局部时间对应关系。该目标直接从现有输入时序中获取监督，不需要额外的注释，并且不改变推理过程。我们在五个公共视频基准上评估了四种不同大小和架构的开放全模态模型的 SyncRA。SyncRA 在所有模型-基准组合中始终优于仅基于答案的微调，同时在控制评估中显著提高了跟踪变化的音频-视觉配对的能力。这些结果表明，轻量级的、有针对性的监督可以有效增强时间对应关系，并转化为音频-视觉问答的广泛改进。
cs.CV / 298 / 2609.34367
Rate-Distortion Adaptive Primitive Selection for Omnidirectional Gaussian Splatting
面向全向高斯溅射的率失真自适应原语选择
Abstract
Learned image codecs (LICs) achieve high reconstruction quality, but their decoding speed is often insufficient for immersive virtual reality (VR). Gaussian splatting (GS) codecs render much faster, yet still lag in reconstruction quality and typically decide primitive allocation without considering the coding cost of each primitive. We introduce OIC-GS, an omnidirectional GS codec with a new hierarchical HEALPix primitive grid representation. Gaussian primitives are anchored at predefined spherical locations, eliminating explicit coordinate coding. Finer levels refine their coarser ancestors, naturally supporting coarse-to-fine reconstruction and layered transmission. The predefined grid also enables efficient viewport decoding by selecting only view-relevant primitives. We further introduce a lightweight entropy model for quantized primitives and optimize the codec under a spherical rate-distortion objective. Primitives with insufficient rate-distortion benefit are automatically removed when their quantized opacity becomes zero, allowing OIC-GS to adapt both primitive density and level of detail without a fixed primitive budget. A single bitstream supports full-sphere, viewport-dependent, and progressive decoding. The first viewport reaches final quality after decoding only 52% of the bitstream, and is then rendered at 1,270 FPS. On a 100-image omnidirectional benchmark, OIC-GS outperforms all evaluated GS codecs, reducing WS-PSNR BD-rate by 49.6% over GaussianImage++ and 68.6% over SGI, which uses a learned entropy model.
Chinese Translation
学习型图像编解码器（LICs）在重建质量上表现优异，但其解码速度往往不足以满足沉浸式虚拟现实（VR）的需求。高斯溅射（GS）编解码器的渲染速度更快，但在重建质量上仍显不足，并且通常在决定原语分配时未考虑每个原语的编码成本。我们提出了OIC-GS，一种具有新型分层HEALPix原语网格表示的全向GS编解码器。高斯原语锚定在预定义的球面位置，消除了显式坐标编码。更细的层级对其粗糙的祖先进行细化，自然支持粗到细的重建和分层传输。预定义的网格还通过仅选择与视图相关的原语来实现高效的视口解码。我们进一步引入了一种轻量级的熵模型用于量化原语，并在球面率失真目标下优化编解码器。当量化的不透明度变为零时，具有不足率失真收益的原语会被自动移除，从而使OIC-GS能够在没有固定原语预算的情况下自适应调整原语密度和细节层次。一个比特流支持全球面、视口依赖和渐进解码。第一个视口在解码仅52%的比特流后达到最终质量，并以1,270 FPS进行渲染。在一个包含100幅图像的全向基准测试中，OIC-GS超越了所有评估的GS编解码器，相对于GaussianImage++减少了49.6%的WS-PSNR BD率，相对于使用学习熵模型的SGI减少了68.6%。
cs.CV / 299 / 2609.34371
From Static to Dynamic: On-Policy Distillation from Image to Video Diffusion Models
从静态到动态：基于策略的图像到视频扩散模型的蒸馏
Abstract
On-policy distillation (OPD) specializes pretrained video diffusion models through teacher supervision along the student's own generation trajectory. Although large video models are natural teachers, developing specialized video experts can require costly video data and training, while querying them incurs substantially higher latency than querying image experts. More readily available and cheaper to query, image experts offer a cost-effective alternative, particularly for largely temporal-agnostic capabilities such as aesthetics and OCR that admit frame-level supervision. However, heterogeneous image and video latent spaces prevent direct supervision of intermediate student states, while image experts lack cross-frame motion supervision, making temporal consistency vulnerable to frame-level improvements. In this paper, we propose MILD, a Motion-Preserving Image-to-Video Latent Distillation framework that transfers specialized image expertise while preserving pretrained video dynamics. MILD uses a learnable linear connector that aligns student latent states and predicted updates with those of image experts, enabling supervision transfer across heterogeneous latent spaces. We further constrain image-guided corrections around the pretrained student's predictions to preserve video dynamics and incorporate an optical-flow-based motion reward to improve motion quality and temporal consistency. Across specialized image experts and multiple video-student backbones, our method consistently outperforms video-teacher OPD baselines, with further studies demonstrating effective transfer across connector designs and heterogeneous architectures. These results establish image-to-video distillation as an effective route to improving video generation by drawing on the diverse and evolving capabilities of the image-generation ecosystem.
Chinese Translation
基于策略的蒸馏（On-policy distillation, OPD）通过教师监督沿着学生自身的生成轨迹来专门化预训练的视频扩散模型。尽管大型视频模型是自然的教师，但开发专门的视频专家可能需要昂贵的视频数据和训练，而查询它们的延迟显著高于查询图像专家。图像专家更容易获取且查询成本低，为大多数时间无关的能力（如美学和光学字符识别（OCR））提供了一种具有成本效益的替代方案，这些能力允许逐帧监督。然而，异构的图像和视频潜在空间阻止了对中间学生状态的直接监督，而图像专家缺乏跨帧运动监督，使得时间一致性容易受到逐帧改进的影响。在本文中，我们提出了MILD，一个运动保持的图像到视频潜在蒸馏框架，该框架在保持预训练视频动态的同时转移专门的图像专业知识。MILD使用一个可学习的线性连接器，将学生潜在状态和预测更新与图像专家的状态对齐，从而实现跨异构潜在空间的监督转移。我们进一步约束图像引导的修正围绕预训练学生的预测，以保持视频动态，并结合基于光流的运动奖励来提高运动质量和时间一致性。在多个专门的图像专家和视频学生骨干网络上，我们的方法始终优于视频教师OPD基线，进一步研究表明在连接器设计和异构架构之间有效转移。这些结果确立了图像到视频的蒸馏作为通过借鉴图像生成生态系统的多样化和不断发展的能力来改善视频生成的有效途径。
cs.CV / 300 / 2609.34378
Marathoner: Ultra-Long-Horizon Autonomous Intelligence
马拉松者：超长视野自主智能
Abstract
Humans naturally possess the ability to work persistently toward long-term goals. Given a challenging task, humans can continuously work for months or even years to accomplish a specific objective. In this paper, we propose Marathoner, an autonomous agentic model possessing the ability of ultra-long-horizon execution. Specifically, we propose a comprehensive post-training pipeline to instill this critical capability into base model. For Ultra-Long-Horizon Task Synthesis, we leverage major release PRs containing 1000+ lines of new code from diverse GitHub repositories as the primary source for synthesizing challenging task-level data. Additionally, we introduce Multi-Task Chaining, which chains multiple generated tasks into a single more challenging task, enabling the synthesis of tasks with frontier-level difficulty. For rejection sampling finetuning, we combine strong teacher model with diverse harnesses to generate trajectories on our synthesized tasks and conduct supervised finetuning on base model with rejection sampled trajectories. For reinforcement learning, cold-started model performs real-world execution through harnesses in independent sandboxes during rollout process, effectively facilitating the acquisition of genuine ultra-long-horizon execution capability. We further propose a novel reward strategy, Later Stage Bonus Reward, which explicitly encourages model to perform meaningful maneuvers during later stages of execution. Through extensive evaluation on 5 benchmarks containing ultra-long-horizon tasks, Marathoner achieves consistent and substantial performance improvements over base model and even surpasses performance of strong proprietary model. Further analysis shows that Marathoner can consistently work for 10+ hours and conduct 1000+ tool calls on highly challenging tasks.
Chinese Translation
人类自然具备朝着长期目标持续努力的能力。在面对具有挑战性的任务时，人类可以连续工作数月甚至数年以实现特定目标。本文提出了马拉松者（Marathoner），一种具备超长视野执行能力的自主智能模型。具体而言，我们提出了一种全面的后训练流程，以将这一关键能力注入基础模型。对于超长视野任务合成，我们利用来自多样化GitHub代码库的包含1000多行新代码的主要发布PR作为合成具有挑战性的任务级数据的主要来源。此外，我们引入了多任务链（Multi-Task Chaining），将多个生成的任务链接成一个更具挑战性的单一任务，从而实现具有前沿难度的任务合成。在拒绝采样微调方面，我们结合强大的教师模型与多样化的工具生成轨迹，并在基础模型上进行监督微调，使用拒绝采样的轨迹。对于强化学习，冷启动模型在独立沙箱中通过工具进行真实世界执行，在推广过程中有效促进了真实超长视野执行能力的获得。我们进一步提出了一种新颖的奖励策略，即后期奖金奖励（Later Stage Bonus Reward），明确鼓励模型在执行的后期阶段进行有意义的操作。通过对包含超长视野任务的5个基准进行广泛评估，马拉松者在基础模型上实现了一致且显著的性能提升，甚至超越了强大的专有模型的性能。进一步分析表明，马拉松者能够持续工作超过10小时，并在高度挑战性的任务上进行1000多次工具调用。
cs.AI / 1 / 2609.31763
SMARtCARE: Privacy-Preserving Agentic AI Systems for Bounded-Autonomy Clinical Decision Support
SMARtCARE：用于有限自主临床决策支持的隐私保护代理人工智能系统
Abstract
Long-context clinical AI systems can miss relevant patient history when prior admissions fall outside the active reasoning context. In ICU monitoring, this can cause early vital-sign drift to appear nonspecific even when it resembles a prior deterioration pattern. SMARtCARE addresses this gap through a four-state clinical decision-support architecture: Stable, Meta-cognitive, Assisted, and Regulated (Revoked). Rather than automatically retrieving prior records, SMARtCARE uses a lossy six-channel fingerprint of the patient's prior trajectory. When current drift matches that fingerprint and the prior record is absent from context, the system raises a Meta-cognitive escalation for clinician review; full retrieval occurs only through clinician action in the Assisted state. A patient-identity guard is designed to enforce correct attribution across data loading, logging, and audit layers. Evaluation combines a synthetic Monte Carlo study that validates the state-transition logic and estimator stability, not clinical performance, with real-data runs on both the MIMIC-III and MIMIC-IV Clinical Database Demos. On MIMIC-III, one prior-pattern recurrence was identified among 14 two-admission patients; on MIMIC-IV, the same pipeline produced no fingerprint matches among 9 two-admission patients, which illustrates a key limitation of a fixed canonical pattern library. Across both runs all logged decisions were fully traceable and correctly attributed. The results support SMARtCARE as a traceable, privacy-aware mechanism for surfacing middle-context risk; they are not a clinical efficacy claim.
Chinese Translation
长上下文的临床人工智能系统在先前入院记录超出活跃推理上下文时可能会遗漏相关的患者历史。在重症监护室监测中，这可能导致早期生命体征漂移看起来不具特异性，即使它类似于先前的恶化模式。SMARtCARE通过一种四状态的临床决策支持架构来解决这一问题：稳定（Stable）、元认知（Meta-cognitive）、辅助（Assisted）和受限（Regulated，撤销）。SMARtCARE并不是自动检索先前记录，而是使用患者先前轨迹的有损六通道指纹。当当前漂移与该指纹匹配且上下文中缺少先前记录时，系统会提出元认知升级以供临床医生审查；完整检索仅在辅助状态下通过临床医生的操作进行。患者身份保护机制旨在确保数据加载、记录和审计层的正确归属。评估结合了一个合成的蒙特卡洛研究，验证状态转换逻辑和估计器稳定性，而不是临床表现，以及在MIMIC-III和MIMIC-IV临床数据库演示上的真实数据运行。在MIMIC-III中，在14名两次入院的患者中识别出一个先前模式的复发；在MIMIC-IV中，同一管道在9名两次入院的患者中未产生指纹匹配，这突显了固定典范模式库的一个关键限制。在这两次运行中，所有记录的决策均可完全追溯并正确归属。结果支持SMARtCARE作为一种可追溯的、关注隐私的机制，用于揭示中等上下文风险；这些结果并不构成临床有效性的声明。
cs.AI / 2 / 2609.31784
Witeness Overlap: Directional Provenance Inside Open-Weight Model Families
见证重叠：开放权重模型家族中的方向性来源
Abstract
Open-weight models are often released, fine-tuned, aligned, merged, and re-released, making provenance audits ask not only whether checkpoints are related, but also which checkpoint came first. Many existing model-provenance methods are designed for a base-known audit setting: given a victim or source model, they test whether a suspect model is related to it. Although these audits are framed as source-to-suspect tests, their underlying evidence is often symmetric, relying on representation similarity, weight similarity, behavioral fingerprints, or correlation statistics. Symmetric pairwise comparisons can detect relatedness, but they cannot by themselves orient relationship between checkpoints A and B. We therefore introduce a local geometric comparison: instead of comparing two checkpoints directly, we add a third same-family checkpoint as a witness and compare the geometry around each candidate endpoint. Direction is inferred by asking which candidate behaves more like a branching parent. Motivated by this idea, and by the empirically observed asymmetry between parent-anchored and child-anchored witness-overlap distributions, we propose Witness Overlap, a prompt-free, training-free white-box test for directional provenance. On 176 LLM checkpoints from 16 families, our one-witness test orients 95.3\% of parent-child decisions using Frobenius cosine. We further evaluate root identification, sibling discrimination, generalizations to VLM and diffusion families, and chain-structured ordering. The signal is robust to weight noise and sparse pruning, with a proposed SVD weight reduction variant showing greater robustness than Frobenius cosine.
Chinese Translation
开放权重模型经常被发布、微调、对齐、合并和重新发布，这使得来源审计不仅要询问检查点之间是否相关，还要确定哪个检查点先出现。许多现有的模型来源方法是为已知基础审计设置设计的：给定一个受害者或源模型，它们测试一个可疑模型是否与之相关。尽管这些审计被框架化为源到可疑模型的测试，但其基础证据往往是对称的，依赖于表示相似性、权重相似性、行为指纹或相关统计。对称的成对比较可以检测相关性，但它们无法单独确定检查点A和B之间的关系方向。因此，我们引入了一种局部几何比较：我们不直接比较两个检查点，而是添加一个来自同一家族的第三个检查点作为见证，并比较每个候选端点周围的几何形状。通过询问哪个候选者的行为更像一个分支父模型来推断方向。受到这一思路的启发，以及在经验上观察到的父锚和子锚见证重叠分布之间的非对称性，我们提出了见证重叠（Witness Overlap），这是一种无提示、无训练的白盒方向性来源测试。在来自16个家族的176个大型语言模型（LLM）检查点上，我们的单见证测试使用Frobenius余弦对95.3%的父子决策进行了方向性定位。我们进一步评估了根识别、兄弟区分、对视觉语言模型（VLM）和扩散家族的推广，以及链结构排序。该信号对权重噪声和稀疏修剪具有鲁棒性，提出的奇异值分解（SVD）权重减少变体显示出比Frobenius余弦更大的鲁棒性。
cs.AI / 3 / 2609.31790
CP-Agent: A Harness-Engineered Agent for Crystal Plasticity Simulation Workflows
CP-Agent：一种用于晶体塑性模拟工作流的工程化代理
Abstract
Crystal plasticity (CP) simulations predict the mechanical behavior of polycrystalline metals, yet their routine use is hindered by the manual effort of configuring heterogeneous tools, orchestrating multi-step data pipelines, and calibrating constitutive parameters against experiments. These bottlenecks impede productivity in systematic parameter studies, motivating interest in automated workflows. This study presents CP-Agent, a harness-engineered LLM-based agent that autonomously executes complete CP modeling workflows from natural-language tasks. Operating under the ReAct paradigm, the agent reasons about tool selection and sequencing while delegating numerical search to established optimizers. The harness comprises a minimal system prompt, typed tool definitions, a dispatcher, and a safety-bounded iteration loop, encoding domain knowledge through tool schemas rather than hard-coded logic. CP-Agent is demonstrated on four case studies: calibrating four slip parameters of additively manufactured stainless steel 316L against tensile data; validating the workflow against published copper benchmarks, reproducing stress-strain and texture evolution; recovering the initial crystallographic texture of copper, where the agent correctly identifies a diffuse initial texture; and reproducing the multi-pass rolling texture evolution of a Mg-Zn-Ca alloy, where the agent chains five deformation passes and recovers the experimentally observed weakened, split basal texture. In all cases, the agent inferred the correct execution sequence from the task statement, robustly across repeated runs, and delivered physically interpretable results. This work establishes harness engineering as a systematic approach to automating CP modeling workflows while maintaining physical interpretability and auditability through visible reasoning traces.
Chinese Translation
晶体塑性（CP）模拟预测多晶金属的机械行为，但其常规使用受到配置异构工具、协调多步骤数据管道以及根据实验校准本构参数的手动努力的阻碍。这些瓶颈妨碍了系统参数研究的生产力，激发了对自动化工作流的兴趣。本研究提出了CP-Agent，这是一种基于大语言模型（LLM）的工程化代理，能够自主执行完整的CP建模工作流，从自然语言任务出发。该代理在ReAct范式下运作，能够对工具选择和顺序进行推理，同时将数值搜索委托给已建立的优化器。该工程系统包括一个最小系统提示、类型化工具定义、调度器和安全边界迭代循环，通过工具模式编码领域知识，而不是硬编码逻辑。CP-Agent在四个案例研究中得到了验证：校准增材制造的不锈钢316L的四个滑移参数与拉伸数据；验证工作流与已发布的铜基准，重现应力-应变和织构演变；恢复铜的初始晶体织构，其中代理正确识别出弥散的初始织构；以及重现镁-锌-钙合金的多道次轧制织构演变，其中代理链式执行五个变形道次，并恢复实验观察到的减弱的、分裂的基面织构。在所有案例中，代理从任务声明中推断出正确的执行顺序，并在重复运行中表现出稳健性，提供了物理可解释的结果。这项工作确立了工程化作为自动化CP建模工作流的系统性方法，同时通过可见的推理痕迹保持物理可解释性和可审计性。
cs.AI / 4 / 2609.31792
ConflictVLA-Bench: Benchmarking Behavioral Responses of Vision-Language-Action Models to Premise Conflicts
ConflictVLA-Bench：视觉-语言-动作模型对前提冲突的行为响应基准测试
Abstract
While Vision-Language-Action (VLA) models perform strongly on manipulation tasks, their responses to invalid task premises remain underexplored. Existing evaluations of premise conflicts often focus on terminal task outcomes, yet task failure alone cannot distinguish behavioral disengagement from continued pursuit followed by an execution error. We call the latter pattern Failed Persistence. To study this phenomenon, we introduce ConflictVLA-Bench, which pairs conflict rollouts with premise-consistent reference rollouts and evaluates both outcomes and execution processes. Built on LIBERO, the benchmark contains 2,826 prompt-conditioned conflict tasks spanning four conflict families, four structural configurations, and two prompt conditions. Across all eight VLAs, invalid premises reduce original goal completion by at least 17.3 percentage points, with the reduction reaching 56.2 percentage points for OpenVLA. Crucially, even when models succeed on premise-consistent tasks and fail on their matched conflict tasks, they often continue to approach the original targets, retain early trajectory structure, and show limited action magnitude suppression. Failed Persistence therefore recurs across the evaluated models. Explicit premise checking does not consistently produce selective and coordinated behavioral changes. These findings show that terminal failure alone establishes neither behavioral disengagement nor refusal and that outcomes alone are insufficient for VLA evaluation. Experimental data and additional details are available on the project page: https://github.com/EmbodiedAISurvey/ConflictVLA-Bench
Chinese Translation
尽管视觉-语言-动作（VLA）模型在操作任务中表现出色，但它们对无效任务前提的响应仍然未得到充分探索。现有的前提冲突评估通常侧重于最终任务结果，而仅仅依赖任务失败无法区分行为脱离与在执行错误后继续追求的情况。我们称这种后者模式为失败的持续性（Failed Persistence）。为了研究这一现象，我们引入了ConflictVLA-Bench，该基准将冲突回放与前提一致的参考回放配对，并评估两者的结果和执行过程。基于LIBERO，该基准包含2,826个以提示为条件的冲突任务，涵盖四个冲突类别、四种结构配置和两种提示条件。在所有八个VLA中，无效前提使得原始目标完成率至少降低了17.3个百分点，其中OpenVLA的降幅达到56.2个百分点。关键是，即使模型在前提一致的任务上成功，而在其匹配的冲突任务上失败，它们仍然常常继续接近原始目标，保持早期轨迹结构，并表现出有限的动作幅度抑制。因此，失败的持续性在评估的模型中反复出现。明确的前提检查并未始终产生选择性和协调性的行为变化。这些发现表明，仅靠最终失败无法确立行为脱离或拒绝，且仅凭结果不足以评估VLA。实验数据和更多细节可在项目页面获取：https://github.com/EmbodiedAISurvey/ConflictVLA-Bench
cs.AI / 5 / 2609.31793
Working with AI: A Design Framework for Human-AI Collaboration
与人工智能合作：人机协作设计框架
Abstract
Artificial Intelligence (AI), particularly GenAI, is becoming an increasingly important part of modern work. In industrial settings, AI can support decision-making, automate routine activities, assist humans, and improve productivity. However, successful AI adoption depends on more than what the technology can do. It also depends on how people experience and work with it. This raises an important question: how should human-AI collaboration be designed so that it works well for both people and organisations? This white paper addresses that question by presenting a practical framework for designing human-AI collaboration. The framework considers the human, the AI system, the task, the organisation, and the wider societal environment. It explains what effective collaboration looks like, what conditions influence it, what requirements should be met, and what design decisions organisations should consider. The report also includes a human-AI collaborative assembly system with cobot use case to demonstrate how the framework can be applied in practice. The use case shows how design requirements can be translated into specific collaboration features and evaluated through a case study. The aim of this white paper is to provide a clear and practical guide for designing human-AI collaboration that is effective, human-centred, and responsible.
Chinese Translation
人工智能（AI），特别是生成性人工智能（GenAI），正日益成为现代工作的重要组成部分。在工业环境中，人工智能可以支持决策、自动化常规活动、协助人类并提高生产力。然而，成功的人工智能采纳不仅仅依赖于技术的能力，还取决于人们如何体验和使用它。这引发了一个重要问题：人机协作应该如何设计，以便对人和组织都能有效运作？本白皮书通过提出一个实用的人机协作设计框架来解决这个问题。该框架考虑了人、人工智能系统、任务、组织以及更广泛的社会环境。它解释了有效协作的表现形式、影响协作的条件、应满足的要求以及组织应考虑的设计决策。报告还包括一个人机协作装配系统的协作机器人（cobot）应用案例，以展示该框架如何在实践中应用。该案例展示了如何将设计要求转化为具体的协作特征，并通过案例研究进行评估。本白皮书的目的是提供一个清晰且实用的指南，以设计有效、人本和负责任的人机协作。
cs.AI / 6 / 2609.31814
DriveHierarchy: A Benchmark for Diagnosing VLM Driving Capabilities from Open-Loop Understanding to Closed-Loop Execution
DriveHierarchy：从开放环路理解到闭环执行的VLM驾驶能力诊断基准
Abstract
Evaluating VLM-based autonomous driving remains difficult because driving competence is composite, where a capable system must ground traffic participants and hazards, integrate context across views and time, reason about future evolution, and act appropriately under closed-loop interaction. Existing benchmarks usually assess either open-loop understanding or closed-loop driving but provide limited structure for explaining how these abilities are organized, how they relate, and how they may inform model diagnosis and improvement. We present \textsc{DriveHierarchy}, a hierarchical benchmark that organizes VLM-based autonomous driving into four ranks, spanning perceptual grounding, contextual memory, mental reasoning, and closed-loop execution. To instantiate this hierarchy, we integrate multiple open-source autonomous-driving datasets into a unified open-loop benchmark with 76,798 question-answer pairs over 84,279 frames and develop a closed-loop simulation platform with interactive scenario construction on a real-world road network, from which 100 driving scenarios are curated for embodied evaluation. Experiments on 15 VLMs show that \textsc{DriveHierarchy} captures structured but non-redundant capability variation, relates open-loop understanding to closed-loop driving, and provides a practical basis for diagnosis and benchmark-guided optimization. \textsc{DriveHierarchy} therefore serves as a unified framework for evaluating and improving VLM-based autonomous driving systems. An anonymized project has been released on https://github.com/PerfectXu88/DriveHierarchy
Chinese Translation
评估基于VLM的自主驾驶仍然困难，因为驾驶能力是复合的，能够胜任的系统必须能够识别交通参与者和危险，整合跨视角和时间的上下文，推理未来的发展，并在闭环交互中适当地行动。现有基准通常评估开放环路理解或闭环驾驶，但对这些能力如何组织、如何关联以及如何为模型诊断和改进提供信息的结构性支持有限。我们提出了 extsc{DriveHierarchy}，一个将基于VLM的自主驾驶组织为四个等级的分层基准，涵盖感知基础、上下文记忆、心理推理和闭环执行。为了实现这一层次结构，我们将多个开源自主驾驶数据集整合为一个统一的开放环路基准，其中包含76,798个问答对，覆盖84,279帧，并开发了一个闭环仿真平台，能够在真实世界道路网络上进行互动场景构建，从中策划出100个驾驶场景以进行具体评估。对15个VLM的实验表明， extsc{DriveHierarchy}捕捉到了结构化但不冗余的能力变化，将开放环路理解与闭环驾驶关联起来，并为诊断和基准指导的优化提供了实用基础。因此， extsc{DriveHierarchy}作为一个统一框架，服务于评估和改进基于VLM的自主驾驶系统。一个匿名项目已在https://github.com/PerfectXu88/DriveHierarchy上发布。
cs.AI / 7 / 2609.31857
LLM Judge Validation Under Sparse Overlap: From Inference to Design
稀疏重叠下的LLM评估验证：从推理到设计
Abstract
Validating an LLM-as-a-judge requires estimating its agreement with humans, yet annotation budgets rarely allow every item to be multiply labeled. We prove that this \emph{overlap sparsity} is the first-order determinant of wrong deployment decisions: at 5\% pairwise overlap, wrong-decision rates reach 25\% and the probability of selecting the wrong best judge among ten candidates is 65\%. The two actionable levers are overlap \emph{quantity} and \emph{allocation}. For quantity, we derive a minimum-overlap formula showing $\rho \geq 0.25$ suffices for non-borderline judges while borderline cases remain fundamentally hard. For allocation, a zero-cost stratified scheme halves false-rejection rates relative to random sampling when strata are informative. We validate on 10 LLM judges across four evaluation matrices spanning visual assessment, causal reasoning, and summarization.
Chinese Translation
验证作为评判者的LLM需要估计其与人类的协议，然而注释预算通常无法允许每个项目被多次标注。我们证明了这种 extit{重叠稀疏性}是错误部署决策的首要决定因素：在5\%的成对重叠下，错误决策率达到25\%，在十个候选者中选择错误最佳评判者的概率为65\\%。两个可操作的杠杆是重叠的 extit{数量}和 extit{分配}。对于数量，我们推导出一个最小重叠公式，表明$
ho \\geq 0.25$足以用于非边界评判者，而边界情况仍然在根本上是困难的。对于分配，当层次信息丰富时，一个零成本的分层方案相对于随机抽样将错误拒绝率减半。我们在四个评估矩阵上验证了10个LLM评判者，涵盖视觉评估、因果推理和摘要生成。
cs.AI / 8 / 2609.31868
Metro-WM: Long-Horizon Latent Planning with Realisable Sub-Goals
Metro-WM：具有可实现子目标的长时域潜在规划
Abstract
Model-predictive control with Joint-Embedding Predictive Architectures (JEPAs) provides a strong zero-shot goal-reaching planner, but it is only effective over short planning horizons. Hierarchical extensions attempt to bridge this gap by learning a macro planner to predict intermediate latent sub-goals to guide the micro planner. In this work, we demonstrate that unconstrained latent sub-goal prediction is fundamentally flawed. A rigorous evaluation reveals that a leading state-of-the-art macro planner routinely emits physically unrealisable sub-goals. To resolve this, we introduce Metro-WM, a hierarchical framework that issues sub-goals by retrieving genuine states from prior experience rather than generating ungrounded latent vectors. Specifically, Metro-WM constructs a graph whose vertices are observed frames from offline expert demonstrations or random-action trajectories, allowing frames from different episodes to be connected and stitched into routes to the goal. Planning over the full graph also makes the system highly robust to execution errors: if the micro planner drifts off course, Metro-WM instantly finds a new optimal path from the current state. Our experiments show that Metro-WM achieves superior long-horizon success rates of up to 37.33 percentage points over the next best hierarchical approach while being up to 10.9 times faster, requiring both 13-56 times less offline compute and fewer tuned hyperparameters. Additional analysis reveals that Metro-WM finds shorter paths than the offline demonstrations, outperforms an oracle relying on the query's own demonstration, and maintains robust performance under extremely sparse dataset conditions.
Chinese Translation
基于联合嵌入预测架构（Joint-Embedding Predictive Architectures, JEPAs）的模型预测控制提供了一种强大的零样本目标达成规划器，但其仅在短期规划范围内有效。层级扩展试图通过学习宏观规划器来预测中间潜在子目标，以指导微观规划器，从而弥合这一差距。在本工作中，我们证明了无约束的潜在子目标预测存在根本缺陷。严格评估表明，当前领先的最先进宏观规划器经常输出物理上不可实现的子目标。为了解决这一问题，我们提出了Metro-WM，一种层级框架，通过从先前经验中检索真实状态而非生成无根潜在向量来发布子目标。具体而言，Metro-WM 构建了一个图，其顶点为离线专家示范或随机动作轨迹中的观测帧，允许来自不同回合的帧相互连接并拼接成通往目标的路径。在整个图上进行规划还使系统对执行误差具有高度鲁棒性：如果微观规划器偏离轨道，Metro-WM 能立即从当前状态找到新的最优路径。我们的实验表明，Metro-WM 在长时域成功率上比下一佳层级方法高出最多37.33个百分点，同时速度提升最多达10.9倍，且离线计算需求减少13至56倍，调参数量也更少。额外分析显示，Metro-WM 找到的路径比离线示范更短，优于依赖查询自身示范的oracle，并在极度稀疏的数据集条件下仍保持稳健性能。
cs.AI / 9 / 2609.31871
IndustryLLM: Failure-Driven LLM Training for Industrial Procurement
IndustryLLM：面向工业采购的基于失败驱动的语言模型训练
Abstract
Industrial procurement requires language models to bridge informal buyer jargon, sparse marketplace attributes, and authoritative engineering standards under strict safety tolerances. We present IndustryLLM, an open-weight industrial language model trained from Qwen3.5-35B-A3B-Base (35B total parameters with ~3B activated per token, with the vision encoder frozen). Rather than relying on generic text scaling, we introduce a failure-driven adaptation recipe spanning continued pre-training (CPT) and supervised fine-tuning (SFT). CPT leverages a curated ~100B-token corpus integrating 5B tokens of national standards (e.g., GB/T) and technical archives, 10B tokens of de-identified real-world industrial transaction and inquiry records, and 60B tokens of general replay. To overcome register mismatch and factual brittleness, we systematically reconstruct an estimated 20B-token domain subset via multi-register rewriting across 10 genres and 8 writing styles, confidence-routed minimal factual editing, and error-targeted QA synthesis (resolving colloquial typos like '42-luo-mu' -> 42CrMo, expanding ambiguous codes like '16674' -> GB/T 16674, and clarifying conflicting dimensional specs). For downstream deployment, we formalize an evidence-gated constraint-evaluation interface enforcing three-valued logic where unverified product evidence remains unknown rather than satisfied. Offline evaluations demonstrate consistent gains on procurement-query structuring (+2.97 percentage points in exact match, 95% CI [2.11, 3.86] in No-Think mode), while randomized online A/B experiments in production yield substantial improvements (+4.25% GMV, +8.3% satisfied inquiries) alongside a latency reduction from 6-7 s to 1.5 s. Model weights and configs are released at https://huggingface.co/alibaba-multimodal-industrial-ai/IndustryLLM.
Chinese Translation
工业采购需要语言模型在严格的安全容忍度下，弥合非正式买方术语、稀疏的市场属性和权威的工程标准。我们提出了IndustryLLM，这是一个基于Qwen3.5-35B-A3B-Base（总参数为35B，每个token激活约3B的开放权重工业语言模型，视觉编码器被冻结）训练的模型。我们引入了一种基于失败驱动的适应性训练方案，涵盖了持续预训练（CPT）和监督微调（SFT），而不是依赖于通用文本扩展。CPT利用了一个经过精心策划的约100B-token语料库，其中包含5B个国家标准（如GB/T）和技术档案，10B个去标识化的真实工业交易和查询记录，以及60B个一般重播数据。为了克服注册不匹配和事实脆弱性，我们通过跨10个体裁和8种写作风格的多注册重写、基于信心的最小事实编辑和针对错误的QA合成系统地重建了一个估计为20B-token的领域子集（解决口语拼写错误，如'42-luo-mu' -> 42CrMo，扩展模糊代码，如'16674' -> GB/T 16674，并澄清冲突的维度规格）。为了下游部署，我们形式化了一个证据门控约束评估接口，强制执行三值逻辑，其中未经验证的产品证据保持未知而非满足。离线评估显示在采购查询结构化方面的一致提升（精确匹配提高2.97个百分点，95%置信区间[2.11, 3.86]，在无思考模式下），而在生产中的随机在线A/B实验则带来了显著改善（GMV提高4.25%，满意查询提高8.3%），同时延迟从6-7秒减少到1.5秒。模型权重和配置已发布在https://huggingface.co/alibaba-multimodal-industrial-ai/IndustryLLM。
cs.AI / 10 / 2609.31874
COUNTERMEM: World-Model Verified Counter-Factual Memory for Language Agents
COUNTERMEM：面向语言代理的世界模型验证反事实记忆
Abstract
Existing agent memory frameworks mainly create memory through an agent's interaction with the factual world, e.g., remembering feedback from actions taken to improve performance on future tasks. However, these frameworks seldom ask the "what if" question during memory construction: what if a different action had been taken, would the feedback have changed, and how could this feedback become useful memory? Obtaining such feedback directly in an active environment can be expensive and can alter the state needed for comparison. In this work, we introduce COUNTERMEM, a reinforcement-learning framework for constructing and using verified counterfactual memory across tasks. After a failed action, COUNTERMEM evaluates local alternatives from a copy or reset of the original state using executable world models, such as tests, proof checkers, and solvers. It stores improvements with the original and corrected actions, checked outcomes, and conditions for reuse. A learned memory-use policy selects a retrieved record or skips memory to balance task success and interaction cost, while the base LLM remains fixed. Both memory and policy are frozen during held-out evaluation. We evaluate COUNTERMEM on 12 benchmark settings across six domains. With gpt-oss-120b, COUNTERMEM improves both ReAct and Reflexion on all 12 benchmarks across six domains, averaging a gain of 12.6 percentage points over their unaugmented versions. In the four-domain comparison across two backbones, task-run tokens decrease by 7.7-42.0%, excluding offline selector-training costs. Further analyses show that removing verification or persistent storage weakens the gains, while applying verified corrections to unsuitable decisions can reverse them. Code will be released upon acceptance.
Chinese Translation
现有的代理记忆框架主要通过代理与事实世界的互动来创建记忆，例如，记住从采取的行动中获得的反馈，以提高未来任务的表现。然而，这些框架在记忆构建过程中很少提出“如果……会怎样”的问题：如果采取了不同的行动，反馈会改变吗？这些反馈如何变得有用？在主动环境中直接获取此类反馈可能代价高昂，并且可能改变比较所需的状态。在本研究中，我们引入了COUNTERMEM，一个用于在任务之间构建和使用经过验证的反事实记忆的强化学习框架。在一次失败的行动后，COUNTERMEM利用可执行的世界模型（如测试、证明检查器和求解器）评估来自原始状态的副本或重置的局部替代方案。它存储与原始和修正行动、检查结果及重用条件相关的改进。学习到的记忆使用策略选择检索记录或跳过记忆，以平衡任务成功与互动成本，同时基础的大型语言模型（LLM）保持不变。在保留评估期间，记忆和策略均被冻结。我们在六个领域的12个基准设置上评估了COUNTERMEM。使用gpt-oss-120b，COUNTERMEM在所有12个基准上提高了ReAct和Reflexion的表现，平均提升了12.6个百分点，超过了未增强版本。在两个基础模型的四领域比较中，任务运行令牌减少了7.7-42.0%，不包括离线选择器训练成本。进一步分析表明，去除验证或持久存储会削弱收益，而对不合适决策应用经过验证的修正则可能逆转这些收益。代码将在接受后发布。
cs.AI / 11 / 2609.31897
Context-dependent agent evaluation with orthogonal equilibrium learning
基于上下文的代理评估与正交均衡学习
Abstract
Many applications require to evaluate agents under contextual information (e.g., a prompt, task, or user group). We study how to perform such context-dependent agent evaluation from offline feedback. Existing score-based models for this purpose (e.g., Bradley-Terry) impose a transitive preference ordering, which fails to reflect collective preferences when human judgements are heterogeneous. Inspired by social choice theory, we frame evaluation as a contextual game between two players, each selecting a distribution over agents as the strategy to receive greater collective preference than the other. Then, the support of the Nash equilibrium defines a context-specific set of winners. However, learning context-specific equilibria from offline logs is difficult because each context reveals human feedback on only a subset of agents, and, hence, a naive plug-in estimator can therefore be biased. To address these challenges, we propose NashEval, a general framework for robust contextual equilibrium learning. NashEval first constructs debiased estimates of the contextual payoff matrix that characterizes the game. NashEval then learns the context-to-equilibrium mapping with a tailored orthogonal loss, which avoids the need to solve a separate game for each context. We show theoretically that errors in estimating the nuisance functions underlying the payoff matrix affect the risk of the learned equilibrium (i.e., exploitability) only through higher-order terms. Across various experiments, NashEval improves robustness of equilibrium learning and consistently identifies the set of top-performing agents across contexts.
Chinese Translation
许多应用需要在上下文信息（例如提示、任务或用户组）下评估代理。我们研究如何从离线反馈中进行这种基于上下文的代理评估。现有的基于评分的模型（例如，Bradley-Terry模型）强加了传递的偏好顺序，这在当人类判断存在异质性时无法反映集体偏好。受到社会选择理论的启发，我们将评估框架视为两个参与者之间的上下文博弈，每个参与者选择一个代理分布作为策略，以期获得比对方更大的集体偏好。然后，纳什均衡的支持定义了一个特定上下文的赢家集合。然而，从离线日志中学习上下文特定的均衡是困难的，因为每个上下文仅揭示了对一部分代理的人类反馈，因此，简单的插件估计器可能会产生偏差。为了解决这些挑战，我们提出了NashEval，一个用于稳健上下文均衡学习的通用框架。NashEval首先构建去偏估计的上下文收益矩阵，该矩阵表征了博弈。然后，NashEval通过量身定制的正交损失学习上下文与均衡之间的映射，这避免了为每个上下文解决单独博弈的需要。我们理论上证明了，估计收益矩阵底层的干扰函数的误差仅通过高阶项影响学习到的均衡的风险（即可利用性）。在各种实验中，NashEval提高了均衡学习的稳健性，并始终识别出各上下文中的表现最佳代理集合。
cs.AI / 12 / 2609.31903
Choir: An Open Protocol for Distributed Multi-Agent Autoformalization
合唱团：一种分布式多智能体自动形式化的开放协议
Abstract
AI agents can now formalize entire textbooks and major theorems in proof assistants such as Lean, but current efforts are typically centralized: a single team runs all agents and bears the full computational cost. We introduce Choir, an open protocol for distributed formalization. Choir decomposes a project into tasks that can be completed by independent contributors, each running their own agent with their own LLM subscription, while coordinating entirely through the project's GitHub repository. To support open participation, every contribution is checked by a deterministic gate before merge. Choir supports Lean 4, Isabelle, and Rocq, and is open source and modular, allowing projects to replace individual components or extend the protocol.
Chinese Translation
人工智能代理现在可以在如 Lean 等证明助手中形式化整个教科书和主要定理，但当前的努力通常是集中式的：一个团队运行所有代理并承担全部计算成本。我们引入了合唱团（Choir），一种用于分布式形式化的开放协议。合唱团将一个项目分解为可以由独立贡献者完成的任务，每个贡献者运行自己的代理并拥有自己的 LLM 订阅，同时完全通过项目的 GitHub 仓库进行协调。为了支持开放参与，每个贡献在合并之前都会经过一个确定性门控的检查。合唱团支持 Lean 4、Isabelle 和 Rocq，并且是开源和模块化的，允许项目替换单个组件或扩展协议。
cs.AI / 13 / 2609.31906
EmailBench: A Benchmark for Evaluating LLM Agents on Enterprise Email and Productivity Tasks
EmailBench：用于评估企业电子邮件和生产力任务的基准测试
Abstract
Enterprise email agents must combine information retrieval, structured state changes, temporal reasoning, and multi-step coordination. Recent agent benchmarks include productivity tasks, but few center on typed email workflows in a self-contained environment. We introduce EmailBench, a benchmark of 206 email and productivity scenarios across 16 task categories. The benchmark couples a typed email API specification with provider-neutral naming, a deterministic synthetic Enron-inspired corpus, and a scenario suite whose topic selection was informed by aggregate task-intent telemetry from an interactive prototype. Its hybrid evaluation protocol combines 258 executable static assertions with 211 LLM rubrics. We evaluate eight LM configurations on a fixed single-user corpus. The best-performing configuration passes only 33.5% of scenarios despite 99.7% of its tool calls completing without an observed API failure, with pass rates varying substantially across task categories. This gap shows that valid tool execution is not equivalent to task completion. EmailBench provides a self-contained environment for end-to-end email-agent evaluation, with broader tool coverage, multi-persona testing, and repeated-run evaluation as future work areas.
Chinese Translation
企业电子邮件代理必须结合信息检索、结构化状态变化、时间推理和多步骤协调。近期的代理基准测试包括生产力任务，但很少有专注于自包含环境中的类型化电子邮件工作流程。我们介绍了EmailBench，这是一个涵盖16个任务类别的206个电子邮件和生产力场景的基准测试。该基准测试结合了类型化电子邮件API规范和提供者中立命名、一个确定性的合成Enron启发式语料库，以及一个其主题选择受到交互原型的聚合任务意图遥测启发的场景套件。其混合评估协议结合了258个可执行的静态断言和211个LLM评分标准。我们在固定的单用户语料库上评估了八种语言模型配置。尽管其工具调用中99.7%完成时没有观察到API失败，但表现最佳的配置仅通过了33.5%的场景，且不同任务类别的通过率差异显著。这一差距表明，工具的有效执行并不等同于任务的完成。EmailBench为端到端电子邮件代理评估提供了一个自包含的环境，未来的工作领域包括更广泛的工具覆盖、多角色测试和重复运行评估。
cs.AI / 14 / 2609.31908
Improving Medical Calculation of LLMs with Embedded Coding
通过嵌入式编码改善大语言模型的医学计算
Abstract
Large Language Models (LLMs) perform well on medical examinations and question-answering benchmarks, but remain unreliable on medical calculation tasks that require exact numerical outputs. These calculations support high-stakes decisions such as medication dosing, organ-function assessment, and prognostic scoring, for which even small errors can have serious clinical consequences. We introduce MedCode, a framework that improves medical calculation by training LLMs to generate embedded executable code. Given a clinical context, the model identifies the relevant calculator, extracts its input variables, and produces a script that delegates arithmetic operations to a deterministic interpreter. Executing the script returns the calculated value together with an explanation and the appropriate unit. We construct supervised fine-tuning (SFT) and preference datasets from the MedCalc benchmark and additionally curate a dataset for calculation tasks in Intensive Care Unit (ICU) scenarios. We further propose weighted Direct Preference Optimization (wDPO), which adaptively emphasizes preference pairs that are difficult for the model to distinguish. Experiments with LLaMA3-8B, Qwen2.5-7B, and Mistral-7B show absolute accuracy gains of 20--30 percentage points, demonstrating the effectiveness of embedded code generation for medical calculation.
Chinese Translation
大语言模型（LLMs）在医学考试和问答基准测试中表现良好，但在需要精确数值输出的医学计算任务中仍然不可靠。这些计算支持高风险决策，如药物剂量、器官功能评估和预后评分，哪怕是微小的错误也可能导致严重的临床后果。我们提出了MedCode，这是一个通过训练LLMs生成嵌入式可执行代码来改善医学计算的框架。在给定临床背景的情况下，该模型识别相关的计算器，提取其输入变量，并生成一个将算术运算委托给确定性解释器的脚本。执行该脚本将返回计算值及其解释和适当单位。我们从MedCalc基准构建了监督微调（SFT）和偏好数据集，并额外策划了一个用于重症监护室（ICU）场景计算任务的数据集。我们进一步提出了加权直接偏好优化（wDPO），该方法自适应地强调模型难以区分的偏好对。使用LLaMA3-8B、Qwen2.5-7B和Mistral-7B进行的实验显示，绝对准确率提高了20至30个百分点，证明了嵌入式代码生成在医学计算中的有效性。
cs.AI / 15 / 2609.31939
BioDyad: Synchronize Biomedical Discovery and Machine Learning Engineering
BioDyad：同步生物医学发现与机器学习工程
Abstract
Agentic biomedical machine learning (ML) draws on complementary advances in biomedical evidence acquisition and executable program search. Existing systems connect aspects of these capabilities, but coordinating them throughout program search remains challenging. New evidence must guide candidate construction, execution outcomes must inform subsequent discovery and reuse, and validation demands must fit the search budget. We introduce BioDyad, which couples biomedical discovery and ML engineering through two hierarchies within Monte Carlo graph search. Its scientific hierarchy combines prior biomedical guidance with iterative discovery, then links biomedical plans to execution outcomes in memory for reuse across candidates. Its engineering hierarchy moves candidate programs from smoke execution, through train/validation evaluation, to full-data retraining. We evaluate BioDyad on the 76-task BioXArena benchmark under a two-hour per-task budget with three matched LLM backends. It achieves the highest penalized all-task score and task success rate among four agent methods and a one-shot baseline under each backend. These results support coordinating biomedical discovery and ML engineering to integrate external knowledge into executable programs across heterogeneous biomedical tasks.
Chinese Translation
自主生物医学机器学习（ML）依赖于生物医学证据获取和可执行程序搜索的互补进展。现有系统连接了这些能力的某些方面，但在程序搜索过程中协调它们仍然具有挑战性。新的证据必须指导候选构建，执行结果必须为后续发现和重用提供信息，而验证需求必须符合搜索预算。我们介绍了BioDyad，它通过蒙特卡洛图搜索中的两个层次将生物医学发现与机器学习工程结合起来。其科学层次结合了先前的生物医学指导与迭代发现，然后将生物医学计划与内存中的执行结果链接，以便在候选之间重用。其工程层次将候选程序从初步执行，经过训练/验证评估，直到全数据重训练。我们在76个任务的BioXArena基准上评估了BioDyad，在每个任务的两小时预算下，使用三个匹配的LLM后端。它在四种代理方法和每个后端下的单次基线中实现了最高的惩罚全任务得分和任务成功率。这些结果支持协调生物医学发现与机器学习工程，以将外部知识整合到异构生物医学任务的可执行程序中。
cs.AI / 16 / 2609.31963
Symbolic Guidance for LLM Agents in Distributed Multiagent Coordination
分布式多智能体协调中的符号引导LLM代理
Abstract
Large language models (LLMs) are increasingly deployed as autonomous agents in multi-agent systems, yet their ability to reliably execute distributed coordination protocols remains poorly understood. While AgentsNet, a benchmark framework for distributed coordination among LLM agents, enables such coordination, granting full reasoning autonomy often leads to inconsistent or degraded performance in complex domains. We hypothesize that coordination can be improved by regulating agent autonomy through symbolic guidance derived from established algorithms. To investigate this, we introduce the \emph{Symbolic Guidance Taxonomy (SGT)}, which characterizes a spectrum of autonomy ranging from open-ended natural language reasoning to fully prescribed algorithmic execution, with intermediate levels providing partial pseudocode guidance. Our results show that intermediate autonomy levels consistently outperform both unguided agents and fully prescriptive specifications. These findings identify autonomy regulation as a key design principle for LLM-based distributed coordination.
Chinese Translation
大型语言模型（LLMs）越来越多地作为自主代理部署在多智能体系统中，但它们在可靠执行分布式协调协议方面的能力仍然不够清晰。尽管AgentsNet作为LLM代理之间分布式协调的基准框架能够实现这种协调，但赋予完全的推理自主权往往会导致在复杂领域中的不一致或性能下降。我们假设通过从已建立算法中导出的符号引导来调节代理的自主权，可以改善协调效果。为此，我们引入了 extit{符号引导分类法（Symbolic Guidance Taxonomy，SGT）}，该分类法描述了从开放式自然语言推理到完全规定的算法执行的自主权谱，且中间级别提供部分伪代码引导。我们的结果表明，中间自主权级别的表现始终优于无引导代理和完全规定的规范。这些发现将自主权调节确定为基于LLM的分布式协调的关键设计原则。
cs.AI / 17 / 2609.31980
Goal-Persistent Coding Agents as Scientific Performance Engineers: A Fixed-Radius Nearest-Neighbor Case Study
目标持久编码代理作为科学性能工程师：固定半径最近邻案例研究
Abstract
Coding agents can pursue persistent objectives across many tool-use turns, but evidence that general-purpose agents can conduct rigorous scientific performance engineering remains limited. We present a repository-scale case study in which off-the-shelf Codex and Claude Code agents optimize fixed-radius nearest-neighbor (FRNN) search for particle tracking. Starting from a PyTorch-dependent CUDA implementation, the agents follow an executable goal that specifies exact-correctness tests, profiling requirements, and acceptance criteria without prescribing code transformations. In the primary sequential trajectory, they autonomously remove the PyTorch dependency and conduct hypothesis-driven optimization experiments. The resulting standalone C++/CUDA library exactly reproduces the targeted reference result. Its synchronous NumPy interface achieved 1.6-fold speedup over the original GPU-resident PyTorch interface, despite including host transfers. Similar speedups were observed across different GPU architectures and software stacks. An independent optimization rerun followed a different sequence of hypotheses and reached even better performance on the target workload. These results show that goal-persistent coding agents can act as experimental performance engineers, and that executable scientific contracts are needed both to guide and to validate their optimization.
Chinese Translation
编码代理可以在多个工具使用回合中追求持久目标，但关于通用代理能够进行严格的科学性能工程的证据仍然有限。我们展示了一个库规模的案例研究，其中现成的 Codex 和 Claude Code 代理优化了用于粒子追踪的固定半径最近邻（FRNN）搜索。从一个依赖于 PyTorch 的 CUDA 实现开始，代理遵循一个可执行的目标，该目标指定了精确正确性测试、性能分析要求和接受标准，而不规定代码转换。在主要的顺序轨迹中，它们自主地去除了对 PyTorch 的依赖，并进行了假设驱动的优化实验。最终生成的独立 C++/CUDA 库准确地再现了目标参考结果。其同步的 NumPy 接口在包含主机传输的情况下，相比于原始的 GPU 上的 PyTorch 接口实现了 1.6 倍的加速。在不同的 GPU 架构和软件栈中也观察到了类似的加速效果。一次独立的优化重跑遵循了不同的假设序列，并在目标工作负载上达到了更好的性能。这些结果表明，目标持久编码代理可以作为实验性能工程师，并且需要可执行的科学合同来指导和验证它们的优化。
cs.AI / 18 / 2609.31990
CSI-Agent: LLM-Assisted Few-Shot Adaptation for Cross-Domain Wi-Fi CSI Sensing
CSI-Agent：基于大语言模型的跨域Wi-Fi CSI感知的少样本适应
Abstract
Wi-Fi channel state information (CSI) has enabled device-free sensing applications such as human activity recognition. However, CSI sensing models remain brittle in cross-domain deployment, where changes in users or environments can produce incorrect predictions. Existing solutions usually treat this problem as an offline model-design problem, by pretraining a stronger representation or applying one fixed adaptation method to the entire target domain. In practice, labeled target data are scarce and different classes may fail in different ways under the same domain shift. To address this, we propose CSI-Agent, an evidence-seeking LLM agent that reformulates cross-domain CSI adaptation as a deployment-time decision-making problem. Rather than processing raw CSI or making sample-level predictions, CSI-Agent summarizes target-domain behavior into sensing-grounded class-level evidence. It establishes a strong target-adaptive default from complementary CSI views and uses an LLM planner to determine whether each class should retain the default or invoke a specialized action. Deterministic verification and bounded execution further reduce unreliable interventions. We evaluate CSI-Agent on four public datasets using five cross-domain splits covering device, user, environment, and compositional shifts. Under 1-shot adaptation, CSI-Agent achieves the best target-domain performance across all splits and improves the average Macro-F1 by about 16\% compared to the strongest baseline method.
Chinese Translation
Wi-Fi信道状态信息（CSI）使得无设备感知应用成为可能，例如人类活动识别。然而，CSI感知模型在跨域部署中仍然脆弱，用户或环境的变化可能导致错误的预测。现有解决方案通常将此问题视为离线模型设计问题，通过预训练更强的表示或对整个目标域应用一种固定的适应方法来解决。在实际应用中，标记的目标数据稀缺，并且不同类别在相同的领域转变下可能以不同的方式失败。为了解决这个问题，我们提出了CSI-Agent，一种寻求证据的大语言模型（LLM）代理，它将跨域CSI适应重新表述为部署时的决策问题。CSI-Agent并不是处理原始CSI或进行样本级预测，而是将目标域行为总结为基于感知的类别级证据。它从互补的CSI视角建立强大的目标自适应默认值，并使用LLM规划器来决定每个类别是否应保留默认值或调用专门的行动。确定性验证和有界执行进一步减少了不可靠的干预。我们在四个公共数据集上评估了CSI-Agent，使用了五个跨域划分，涵盖了设备、用户、环境和组合转变。在1-shot适应下，CSI-Agent在所有划分中实现了最佳的目标域性能，并将平均宏F1提高了约16%，相比于最强的基线方法。
cs.AI / 19 / 2609.32000
SenseAgent: An LLM Agent for Adaptive Cross-Domain IMU Sensing
SenseAgent：一种用于自适应跨域IMU传感的LLM代理
Abstract
Deep learning has improved inertial measurement unit (IMU) sensing for mobile and wearable applications. However, an IMU model trained in one domain often becomes unreliable when it is used with a new user, device, or body position. Existing methods usually treat this problem as a static model-design task: they pretrain a stronger representation, add data augmentation, or select one adaptation method before deployment. In practice, the target domain is only gradually observed, labels are scarce, and different domain shifts require different sensing actions. This paper presents SenseAgent, an LLM-guided sensing agent for cross-domain IMU activity recognition. Instead of asking an LLM to classify raw IMU signals, SenseAgent uses the LLM as a runtime planner over sensing tools, source-domain experience memory, online target memory, and verifiers. The agent builds a label-free diagnosis report from the target stream and uses it to decide whether to keep raw inference or invoke specialized tools, including gravity-aware sensing, prototype transfer, and style normalization. Verifiers check source calibration, target-memory reliability, and no-harm criteria before accepting high-risk tool decisions. SenseAgent also supports scarce feedback without retraining the backbone or replacing the label-free route. This design converts cross-domain IMU sensing from a fixed inference pipeline into a closed-loop sensing process that diagnoses target shifts, selects suitable sensing actions, and rejects unsafe adaptations. We evaluate SenseAgent across multiple IMU datasets and deployment shifts. Results show that its verified route selection improves cross-domain sensing, especially under harder placement and compound shifts, and further benefits from limited user feedback.
Chinese Translation
深度学习已改善移动和可穿戴应用中的惯性测量单元（IMU）传感。然而，在一个领域训练的IMU模型在与新用户、设备或身体位置一起使用时，往往变得不可靠。现有方法通常将此问题视为静态模型设计任务：它们在部署前预训练更强的表示，增加数据增强，或选择一种适应方法。实际上，目标领域仅逐渐被观察到，标签稀缺，并且不同的领域转移需要不同的传感动作。本文提出了SenseAgent，一种用于跨域IMU活动识别的LLM引导传感代理。SenseAgent并不是让LLM对原始IMU信号进行分类，而是将LLM用作传感工具、源领域经验记忆、在线目标记忆和验证器的运行时规划器。该代理从目标流中构建无标签的诊断报告，并利用该报告决定是保留原始推断还是调用专用工具，包括重力感知传感、原型转移和风格归一化。在接受高风险工具决策之前，验证器会检查源校准、目标记忆的可靠性和无害标准。SenseAgent还支持稀缺反馈，而无需重新训练主干或替换无标签路径。该设计将跨域IMU传感从固定推断管道转换为一个闭环传感过程，能够诊断目标转移、选择合适的传感动作并拒绝不安全的适应。我们在多个IMU数据集和部署转移中评估了SenseAgent。结果表明，其验证的路径选择改善了跨域传感，特别是在更困难的放置和复合转移下，并进一步受益于有限的用户反馈。
cs.AI / 20 / 2609.32002
What Does the Rank Buy? A Spectral and Distributional Analysis of Low-Rank Adaptation
秩的价值是什么？低秩适应的谱与分布分析
Abstract
The rank $r$ in LoRA is widely treated as a capacity control: a smaller rank is assumed to yield a simpler model that generalizes better. We show that, under hard per-factor norm budgets---the idealization of the weight decay and norm control used in practice---this intuition breaks down. The reason is structural: under such budgets, the updates LoRA can reach are exactly the matrices of rank at most $r$ inside a nuclear-norm ball, and every complexity and displacement functional we analyze is maximized over this set by a rank-one update---so the rank cap never binds. The consequences follow directly. The linear-readout model class we study is identical for every $r \ge 1$, its Rademacher complexity carries no dependence on $r$, and the distance the adaptation can move the source distribution obeys a rank-independent upper bound that we show is sharp. If rank does not control capacity, where does it act? We identify two places. Statistically, replacing the per-factor budgets with a joint budget on the product restores a data-dependent, rank-sensitive complexity bound---though the gain appears only for well-spread feature distributions, and the worst case remains rank-free. Spectrally, rank sets the price of adaptation: canceling the leading singular directions of the pretrained weight requires both sufficient rank and sufficient budget. We bound the smallest rank achieving a desired source--target alignment, with upper and lower bounds that match under two-sided spectral decay. Together, these results recast rank as governing which updates are reachable and what cancellation costs---not how much capacity the model has.
Chinese Translation
在LoRA中，秩$r$通常被视为容量控制：较小的秩被假定能产生一个更简单的模型，从而具有更好的泛化能力。我们展示了在严格的每因子范数预算下——这是实践中使用的权重衰减和范数控制的理想化——这一直觉是错误的。原因在于结构性：在这样的预算下，LoRA能够达到的更新恰好是核范数球内秩至多为$r$的矩阵，而我们分析的每个复杂性和位移泛函在这个集合上都是通过秩为一的更新达到最大化的——因此秩上限从未约束。后果直接而来。我们研究的线性读出模型类对于每个$r
geq 1$都是相同的，其Rademacher复杂性与$r$无关，而适应所能移动的源分布的距离遵循一个与秩无关的上界，我们证明这个上界是尖锐的。如果秩不控制容量，那么它的作用在哪里？我们确定了两个方面。从统计上讲，用联合预算替代每因子预算恢复了一个依赖于数据的、对秩敏感的复杂性界限——尽管这种增益仅在特征分布良好分散时出现，而最坏情况下仍然是无秩的。从谱的角度看，秩设定了适应的代价：抵消预训练权重的主奇异方向需要足够的秩和足够的预算。我们界定了实现所需源-目标对齐的最小秩，并在双侧谱衰减下得到了匹配的上界和下界。综合这些结果，我们将秩重新定义为控制哪些更新是可达的以及什么抵消成本——而不是模型的容量有多大。
cs.AI / 21 / 2609.32019
Decentralized Master-Mind: Joint Action Refinement through Iterative Intent Denoising in Multi-Agent Pathfinding
去中心化的主脑：通过迭代意图去噪实现多智能体路径规划中的联合行动优化
Abstract
Decentralized multi-agent path finding (MAPF) with communication requires agents to reach individual goals without collisions under partial observability. Learnable policies trained on expert data provide an effective approach to this problem. However, when several coordinated joint actions are valid in the same context, independently sampling from per-agent distributions can recombine locally valid choices into incompatible joint actions. This failure can arise from the final sampling mechanism even when the per-agent action distributions are learned correctly. DMM (Decentralized Master-Mind) addresses this by replacing one-shot action sampling with discrete, iterative refinement of action intents across communication rounds, inspired by denoising in diffusion models. Agents initialize random action intents and refine them through local communication, coupling their choices before commitment. DMM is pretrained with imitation learning on expert MAPF solutions and further optimized with MICPO, a critic-free group-relative reinforcement-learning method designed for multi-agent, multi-round action refinement. DMM generally achieves higher success rates and lower solution costs than the evaluated learnable baselines. On 1,600 MovingAI tasks, DMM fine-tuned with MICPO solves 1,598, the highest coverage among the evaluated methods, while achieving solution costs close to those of the strongest baselines. DMM also scales to over one million simultaneously acting agents in obstacle-rich environments. These results show that round-level intent refinement can improve joint-action coordination while preserving decentralized execution.
Chinese Translation
去中心化的多智能体路径规划（MAPF）需要智能体在部分可观测的情况下，达到各自的目标而不发生碰撞。基于专家数据训练的可学习策略为解决这一问题提供了一种有效的方法。然而，当在同一上下文中存在多个协调的联合行动时，独立地从每个智能体的分布中采样可能会将局部有效的选择重新组合成不兼容的联合行动。这种失败可能源于最终的采样机制，即使每个智能体的行动分布被正确学习。去中心化的主脑（DMM）通过用离散的、迭代的意图优化替代一次性行动采样，解决了这一问题，灵感来自于扩散模型中的去噪。智能体初始化随机的行动意图，并通过局部通信进行优化，在做出承诺之前将其选择进行耦合。DMM在专家MAPF解决方案上通过模仿学习进行预训练，并进一步通过MICPO（一种为多智能体、多轮行动优化设计的无评论员的群体相对强化学习方法）进行优化。DMM通常比评估的可学习基线实现更高的成功率和更低的解决成本。在1,600个MovingAI任务中，经过MICPO微调的DMM解决了1,598个任务，成为评估方法中覆盖率最高的，同时其解决成本接近最强基线。DMM还能够扩展到在障碍物丰富的环境中同时行动超过一百万个智能体。这些结果表明，轮次级意图优化可以在保持去中心化执行的同时改善联合行动协调。
cs.AI / 22 / 2609.32020
A Benchmark for LLM's Understanding of Middle School and High School Science Topics
大型语言模型对中学科学主题理解的基准测试
Abstract
Large language models (LLMs) are increasingly integrated into educational settings, yet educators lack robust, standards-aligned tools to evaluate their effectiveness in K-12 science contexts. Existing benchmarks predominantly assess general language or advanced scientific reasoning, leaving a critical gap in understanding LLMs' performance on content directly relevant to secondary science curricula. To address this gap, we developed a comprehensive NGSS-aligned benchmark for both middle and high school science using a rigorous synthetic data pipeline, multi-judge validation, and item-level psychometric analysis. Nine open-weight LLMs were systematically evaluated using this benchmark, indicating that several smaller, locally deployable models achieved high accuracy across diverse science domains and question types. Our findings indicate that model size did not consistently predict performance, emphasizing the importance of intentional model selection for educational deployment. We then incorporated a human reviewer into the loop, reviewing the items generated by the LLMs for alignment with NGSS standards. The human review indicated that synthetically generated items were not in perfect alignment with the NGSS standards, indicating the benefits of human-in-the-loop item development, the need to explore the intersection of content and pedagogical knowledge, and the need to extend benchmarks to evaluate LLMs' capacity for interactive, evidence-based feedback in educational scenarios.
Chinese Translation
大型语言模型（LLMs）越来越多地融入教育环境，但教育工作者缺乏健全的、与标准对齐的工具来评估其在K-12科学领域的有效性。现有基准主要评估一般语言能力或高级科学推理，导致在理解LLMs在与中学科学课程直接相关的内容上的表现方面存在重要空白。为了解决这一空白，我们开发了一个全面的与下一代科学标准（NGSS）对齐的基准，涵盖中学和高中科学，采用严格的合成数据管道、多评审者验证和项目级心理测量分析。我们系统地使用该基准评估了九个开放权重的LLMs，结果表明，几种较小的、本地可部署的模型在不同科学领域和问题类型上达到了较高的准确性。我们的研究结果表明，模型大小并不总能预测性能，强调了在教育部署中有意选择模型的重要性。随后，我们将人工审查者纳入流程，审查LLMs生成的项目与NGSS标准的一致性。人工审查表明，合成生成的项目与NGSS标准并不完全一致，显示了人机协作项目开发的好处，探索内容与教学知识交叉的必要性，以及扩展基准以评估LLMs在教育场景中提供互动、基于证据反馈能力的必要性。
cs.AI / 23 / 2609.32035
Reasoning Concentrates Errors, and Self-Consistency Never Notices
推理集中错误，自我一致性从未察觉
Abstract
Self-consistency assumes that independent samples disagree when a model is unsure, so agreement is evidence of correctness. Holding weights fixed and toggling only a reasoning mode, over five benchmarks and 74,944 samples, we show that reasoning concentrates a model's errors: the probability that two independently drawn wrong answers coincide rises in all ten dataset-scale comparisons (p = 0.00098), and in nine of nine after restricting both arms to the problems each gets wrong. Where the answer space is unbounded, reasoning cuts the distinct answers produced to 0.43-0.65 of the non-reasoning count; where it is bounded, both arms hold an identical option set and reasoning concentrates mass on it instead, which no positional prior can explain at fixed weights. The aggregate cost is smaller than the mechanism predicts, because reasoning also shrinks the set of problems where answer diversity can decide anything, in ten of ten cells and by 2.7x; normalized for available headroom, both arms convert a quarter of it in domain. Confidence weighting does not recover what is left. Across 280 method-dataset-model combinations on eight models and five benchmarks, not one beats plain majority voting after correction; weighted voting agrees with it on 98.5% of problem-method pairs and is right 56.3% of the time on the rest; and a signal's direction can invert within fixed weights, with answer log-probability predicting correctness when reasoning is off and error when it is on. A learned six-signal combination gains nothing out of domain. Confidence signals should be evaluated on decisions, not on discrimination.
Chinese Translation
自我一致性假设，当模型不确定时，独立样本之间存在分歧，因此一致性是正确性的证据。在固定权重的情况下，仅切换推理模式，在五个基准和74,944个样本上，我们展示了推理集中模型的错误：两个独立抽取的错误答案重合的概率在所有十个数据集规模的比较中上升（p = 0.00098），在将两个组限制到各自错误问题后，这一概率在九个比较中均成立。当答案空间无限时，推理将产生的不同答案减少到非推理计数的0.43-0.65；当答案空间有限时，两个组持有相同的选项集，推理则将质量集中于此，而固定权重下的任何位置先验都无法解释这一现象。总体成本小于机制预测，因为推理还缩小了答案多样性能够决定任何事情的问题集，在十个单元中均如此，且缩小幅度为2.7倍；在可用余量标准化后，两个组在领域中转换了四分之一的余量。置信加权无法恢复剩余部分。在280个方法-数据集-模型组合中，八个模型和五个基准中，没有一个在修正后超过简单的多数投票；加权投票在98.5%的问题-方法对上与其一致，其余情况下正确率为56.3%；而信号的方向在固定权重下可能会反转，当推理关闭时，答案的对数概率预测正确性，而当推理开启时则预测错误性。学习到的六信号组合在领域外没有任何收益。置信信号应在决策上进行评估，而非在区分上。
cs.AI / 24 / 2609.32046
Receiver-Conditioned Latent Communication gives 94% CacheBack
接收者条件化潜在通信实现94%的缓存回馈
Abstract
Multi-agent systems distribute large contexts across agents that communicate to solve a task. Text messages are compact but require decoding and may omit evidence the receiving agent needs. Recent latent communication instead transfers KV caches. This avoids text generation and can improve accuracy and latency. However, a full KV cache grows linearly with both the context an individual agent processes, and the number of agents that coordinate together. This raises memory and context costs, often far exceeding available GPU resources and context window sizes. Our key observation is that agents need only send what the receiving agent requires for its local task -- which we call receiver-conditioned communication. The receiver agent passes the sender a small description of its information needs, which serves to filter and compress the sender agent's KV cache. CacheBack is a simple, robust, training-free instance of receiver conditioning based on the sender's attention weights. On FanOutQA, CacheBack with Qwen 3 removes 75% of the state the agent would otherwise receive, improving accuracy by 14.7 percentage points and reducing median task-completion latency by 3.2x relative to text communication. We show comparable improvements across model families that span dense Transformers, Mamba-attention hybrids, and sliding-window attention.
Chinese Translation
多智能体系统在代理之间分配大型上下文，通过通信来解决任务。文本消息虽然紧凑，但需要解码，并可能遗漏接收代理所需的证据。最近的潜在通信则转移KV缓存，避免了文本生成，并可以提高准确性和延迟。然而，完整的KV缓存随着个体代理处理的上下文和协调的代理数量线性增长。这导致内存和上下文成本上升，通常远超可用的GPU资源和上下文窗口大小。我们的关键观察是，代理只需发送接收代理为其本地任务所需的信息——我们称之为接收者条件化通信。接收代理向发送者传递其信息需求的小描述，这有助于过滤和压缩发送者的KV缓存。CacheBack是基于发送者注意力权重的接收者条件化的简单、稳健且无训练实例。在FanOutQA上，使用Qwen 3的CacheBack去除了代理本应接收的75%的状态，相较于文本通信提高了14.7个百分点的准确性，并将任务完成的中位延迟减少了3.2倍。我们展示了在包括密集Transformer、Mamba注意力混合体和滑动窗口注意力等模型家族中，类似的改进。
cs.AI / 25 / 2609.32049
EngramRAG: Dynamic Usage-Weighted Topology and Synaptic Consolidation for Multi-Hop Agentic Memory
EngramRAG：用于多跳主动记忆的动态使用加权拓扑和突触巩固
Abstract
As autonomous LLM agents are deployed across multi-session environments, conventional memory architectures suffer from Associative Blindness (inability to traverse multi-hop relational dependencies), Scaffolding Amnesia (temporal decay evicting core persona invariants), and Static Topology Stagnation (immutable graphs ignoring usage dynamics). Grounded in Complementary Learning Systems (CLS) principles, we propose EngramRAG, an adaptive memory architecture coupling a low-latency Waking State reflex with an asynchronous background Dreaming State consolidation cycle. EngramRAG introduces: (1) Usage-Modulated Personalized PageRank (U-PPR), where transition probabilities adapt via Hebbian plasticity to promote persistent entities into high-centrality Epistemic Macro-Hubs; (2) Consolidation-Activated Topology Decay (CATD), which scales retention half-life by topological load-bearing weight rather than wall-clock recency, protected by a cold-start grace period (N_grace >= 4); (3) Directed SUPERSEDES DAG filtering to suppress obsolete state during fact mutations; and (4) Triple-source hybrid retrieval fusing dense vectors, BM25, and U-PPR via dynamic Reciprocal Rank Fusion (RRF). Evaluating on all 1,982 QA pairs across 10 long-term conversations in the LoCoMo benchmark, EngramRAG achieves +38.9% relative improvement in Recall@5 (53.21% vs. 38.29%, p < 0.001) and +43.1% in MRR (0.4203 vs. 0.2937) over dense vector RAG, significantly outperforming Okapi BM25 (48.66%) and isolated static graph retrieval (8.50%). On temporal reasoning, EngramRAG reaches 62.33% Recall@5 (+16.67 points over dense vectors). In controlled mutation tests, SUPERSEDES suppresses split-brain hallucinations from 70.0% to 0.0%, while 90-day simulations show 100.0% scaffolding retention under a 26.21ms interactive retrieval reflex.
Chinese Translation
随着自主大型语言模型（LLM）代理在多会话环境中的部署，传统记忆架构面临关联盲点（无法遍历多跳关系依赖）、支架遗忘（时间衰减驱逐核心个性不变性）和静态拓扑停滞（不可变图形忽视使用动态）。基于互补学习系统（Complementary Learning Systems, CLS）原则，我们提出了EngramRAG，这是一种自适应记忆架构，将低延迟的觉醒状态反射与异步背景梦境状态巩固周期相结合。EngramRAG引入了：（1）使用调制个性化PageRank（Usage-Modulated Personalized PageRank, U-PPR），其转移概率通过赫布塑性（Hebbian plasticity）进行调整，以促进持久实体进入高中心性的认知宏中心；（2）巩固激活拓扑衰减（Consolidation-Activated Topology Decay, CATD），通过拓扑承载权重而非墙钟时间的近期性来扩展保留半衰期，受冷启动宽限期（N_grace >= 4）保护；（3）定向SUPERSEDES有向无环图（Directed SUPERSEDES DAG）过滤，以抑制事实变更期间的过时状态；（4）三源混合检索，通过动态互惠排名融合（Reciprocal Rank Fusion, RRF）融合密集向量、BM25和U-PPR。在LoCoMo基准测试中评估的1,982对问答对中，EngramRAG在Recall@5上实现了38.9%的相对提升（53.21% vs. 38.29%，p < 0.001），在平均排名回报（Mean Reciprocal Rank, MRR）上提升了43.1%（0.4203 vs. 0.2937），显著超越了密集向量RAG（48.66%）和孤立静态图检索（8.50%）。在时间推理方面，EngramRAG达到了62.33%的Recall@5（比密集向量高出16.67个百分点）。在受控变异测试中，SUPERSEDES将分脑幻觉从70.0%抑制到0.0%，而90天的模拟显示在26.21毫秒的交互检索反射下，支架保留率达到100.0%。
cs.AI / 26 / 2609.32061
Contract monitoring: governing AI via separation of powers
合同监控：通过权力分立治理人工智能
Abstract
We propose an AI safety framework that binds worker agents to contracts specifying their permitted actions. We show how these contracts can be enforced and specified by assigning distinct responsibilities to monitor agents and judges, and asymmetric computational resources to monitors and workers. Our framework allows us to empirically measure statistical safety guarantees. The framework applies to a wide range of settings, including code security and escape-the-box scenarios.
Chinese Translation
我们提出了一种人工智能安全框架，该框架将工作代理绑定到指定其允许行为的合同中。我们展示了如何通过为监控代理和法官分配不同的职责，以及为监控者和工作者分配不对称的计算资源，来执行和规范这些合同。我们的框架使我们能够实证测量统计安全保证。该框架适用于广泛的场景，包括代码安全和逃逸盒子场景。
cs.AI / 27 / 2609.32081
Toward Interactive Understanding of Code APIs
迈向对代码API的互动理解
Abstract
Empowered by advances in Language Model agents, systems have made substantial strides in code generation and understanding. However, these approaches often rely on read access to the relevant code, an assumption which does not hold when dealing with external APIs. In this work, we introduce the PAU (Python API Understanding) benchmark, where we provide models with black-box, API-level access to code snippets. Models must query the API with exploratory inputs and draw insights from the resulting outputs, with the goal of describing the snippet's true functionality. By treating the code snippets as external tools that must be understood via interaction alone, PAU studies the more general problem of unsupervised tool understanding, specifically for tools implemented as Python methods. Despite recent progress in coding agents, even frontier models struggle to achieve high performance on PAU, with the best model (Claude-4-Opus) failing to understand over 45% of the PAU test set. An investigation into the common error modes reveals that models are overconfident; they often overrate the quality of their current hypothesis, leading to insufficient exploration and premature termination. Finally, we take inspiration from the Asymmetric Actor Critic (AAC) paradigm, frequently used in robot learning, to post-train models for interactive code understanding. Models trained with AAC conduct more active exploration of the APIs, with an AAC-tuned Qwen3-8B model matching the performance of GPT-5-mini.
Chinese Translation
得益于语言模型代理的进步，系统在代码生成和理解方面取得了显著进展。然而，这些方法通常依赖于对相关代码的读取访问，而这一假设在处理外部API时并不成立。在本研究中，我们引入了PAU（Python API Understanding）基准，在该基准中，我们为模型提供黑箱的API级代码片段访问。模型必须使用探索性输入查询API，并从结果输出中提取见解，目标是描述代码片段的真实功能。通过将代码片段视为必须通过互动理解的外部工具，PAU研究了无监督工具理解的更一般问题，特别是针对作为Python方法实现的工具。尽管在编码代理方面取得了近期进展，但即使是前沿模型在PAU上也难以实现高性能，最佳模型（Claude-4-Opus）未能理解超过45%的PAU测试集。对常见错误模式的调查显示，模型过于自信；它们往往高估当前假设的质量，导致探索不足和过早终止。最后，我们从不对称演员评论家（Asymmetric Actor Critic，AAC）范式中获得灵感，该范式常用于机器人学习，以对模型进行后训练以实现互动代码理解。使用AAC训练的模型对API进行更积极的探索，其中经过AAC调优的Qwen3-8B模型的性能与GPT-5-mini相当。
cs.AI / 28 / 2609.32091
Memory as Middleware for Self-Improving AI Agents
记忆作为自我改进人工智能代理的中间件
Abstract
AI agents are stateless across sessions by default and therefore operationally amnesic: each session begins with little durable knowledge of prior failures, repairs, preferences, or successful strategies. As a result, agents repeat the same mistakes and discard hard-won experience. The dominant fix is \emph{bespoke memory}---retrieval, persistence, and learning logic hand-wired into one agent and bound to one storage engine. This creates a fragmented landscape where memory cannot be swapped, shared, isolated, or reasoned about independently of the agent that owns it. We argue that this is a middleware problem: agent memory deserves a first-class, pluggable layer, just as data access, messaging, and persistence each became middleware concerns. We develop this vision through six systems challenges: two-sided pluggability, host-native interposition, multi-tenant isolation, write-path consistency, federated sharing with provenance, and lifecycle governance. We present ALTK-Evolve, a reference implementation of memory middleware for self-improving agents, and use it to motivate a broader research agenda for future memory middleware.
Chinese Translation
人工智能代理默认在会话之间是无状态的，因此在操作上是健忘的：每个会话开始时几乎没有对先前失败、修复、偏好或成功策略的持久知识。因此，代理会重复相同的错误，并丢弃来之不易的经验。当前的主要解决方案是定制内存（bespoke memory）——将检索、持久性和学习逻辑手动集成到一个代理中，并绑定到一个存储引擎。这导致了一个碎片化的环境，其中内存不能被交换、共享、隔离或独立于拥有它的代理进行推理。我们认为这是一个中间件问题：代理内存应当拥有一个一流的、可插拔的层，就像数据访问、消息传递和持久性各自成为中间件关注点一样。我们通过六个系统挑战来发展这一愿景：双向可插拔性、主机本地插入、多租户隔离、写路径一致性、带来源的联邦共享和生命周期治理。我们提出了ALTK-Evolve，这是一个自我改进代理的内存中间件的参考实现，并利用它来激励未来内存中间件的更广泛研究议程。
cs.AI / 29 / 2609.32093
GameBoyWorlds: A Testbed for Self-Improvement in Embodied Video Games
GameBoyWorlds：一个用于具身视频游戏自我改进的测试平台
Abstract
Powered by expert guidance, agents can operate in interactive environments; however, it is unclear whether they can learn autonomously from their own experience. To evaluate such self-improvement methods, we introduce GameBoyWorlds, a testbed for agentic self-improvement in video games. GameBoyWorlds-Execution evaluates task execution on a collection of 5 distinct game series. Agents are allowed access to dedicated training games but are provided no demonstrations, documentation, or rewards. Agents must ground themselves in the environment through self-directed exploration and by inferring actionable knowledge from their own experience. At test time, agents must complete short-horizon tasks that evaluate their ability to navigate, interact, and engage with game-specific mechanics in unseen games. Out-of-the-box frontier models complete fewer than 50% of the 500 tasks due to failures in multimodal grounding, establishing that self-improvement methods have room to push performance. We demonstrate that contemporary approaches to self-improvement are lacking, with world modelling and autonomous skill discovery failing, and a novel strategy that uses curiosity-based exploration to write guides achieving only partial success. GameBoyWorlds-Playthrough tests end-to-end game completion in two fan-made Pok\'emon games. We show that while frontier models have been pre-exposed to official releases such as Pok\'emon Red, they lack essential information on the games in our testbed. Instead of relying on their parametric knowledge to succeed, agents must learn from their own experience and autonomously improve over the course of the playthrough. We show that a sophisticated agentic pipeline with multimodal memory and hierarchical subgoals fails to reach even the first major milestone in both games, establishing GameBoyWorlds as an ambitious target for self-improving agents.
Chinese Translation
在专家指导下，智能体可以在互动环境中操作；然而，尚不清楚它们是否能够从自身经验中自主学习。为了评估这种自我改进方法，我们引入了GameBoyWorlds，一个用于视频游戏中智能体自我改进的测试平台。GameBoyWorlds-Execution评估在五个不同游戏系列中的任务执行。智能体可以访问专门的训练游戏，但不提供演示、文档或奖励。智能体必须通过自我导向的探索和从自身经验中推断可操作知识来扎根于环境。在测试时，智能体必须完成短期任务，以评估它们在未见过的游戏中导航、互动和参与游戏特定机制的能力。由于在多模态扎根方面的失败，前沿模型在500个任务中完成率低于50%，这表明自我改进方法还有提升性能的空间。我们展示了当代自我改进方法的不足，世界建模和自主技能发现失败，而一种基于好奇心的探索策略用于编写指南仅取得部分成功。GameBoyWorlds-Playthrough测试在两个粉丝制作的《宝可梦》游戏中的端到端游戏完成情况。我们表明，尽管前沿模型已经接触过《宝可梦红》等官方发布，但它们缺乏我们测试平台上游戏的基本信息。智能体不能依赖其参数知识取得成功，而必须从自身经验中学习，并在游戏过程中自主改进。我们展示了一个复杂的智能体流程，具有多模态记忆和分层子目标，但在两个游戏中甚至未能达到第一个主要里程碑，确立了GameBoyWorlds作为自我改进智能体的一个雄心勃勃的目标。
cs.AI / 30 / 2609.32102
Residual Streams Read, Recurrent States Remember: The Global Workspace in Mamba Models
残差流读取，递归状态记忆：Mamba 模型中的全球工作区
Abstract
Can the global-workspace account of transformer representations extend to state-space language models? We fit Jacobian lenses to the residual streams and recurrent states of Mamba-1, Mamba-2 and Mamba-3, using the original 1000-prompt recipe. Joint residual--state readouts improve recovery of known intermediate concepts over the residual lens on at least five of six task families in every tested Mamba checkpoint. On Mamba-2, state alone exceeds residual and logit lenses on all six families; a normalised joint readout improves on both components on five. Temporal maps and word-list experiments show earlier content remaining state-readable as residual visibility changes. We also propose sign-guarded steering, which improves target top-five success over coordinate exchange on matched verbal-report trials in five models. Recurrent state alone supports this verbal access. These gains do not extend consistently to relational answers: guarded edits often output the edited concept itself, and Mamba-3's joint edits can disrupt successful state-only redirection. Recurrent state thus provides a complementary carrier of workspace content, whose recovery, persistence and causal uses require separate measurements.
Chinese Translation
全球工作区对变换器表示的解释能否扩展到状态空间语言模型？我们为 Mamba-1、Mamba-2 和 Mamba-3 的残差流和递归状态拟合了雅可比透镜，使用了原始的 1000 提示配方。联合残差-状态读出在每个测试的 Mamba 检查点中，在至少六个任务家族中的五个上改善了已知中间概念的恢复，相较于仅使用残差透镜。在 Mamba-2 中，仅状态在所有六个家族中超越了残差和逻辑透镜；归一化的联合读出在五个家族中改善了这两个组件。时间映射和词汇表实验表明，随着残差可见性的变化，早期内容仍然保持可被状态读取。我们还提出了带符号保护的引导，这在五个模型的匹配口头报告试验中改善了目标前五名的成功率，相较于坐标交换。仅递归状态支持这种口头访问。这些收益并未在关系答案中一致扩展：受保护的编辑往往输出编辑后的概念本身，而 Mamba-3 的联合编辑可能会干扰成功的仅状态重定向。因此，递归状态提供了工作区内容的互补载体，其恢复、持续性和因果使用需要单独的测量。
cs.AI / 31 / 2609.32116
Escaping Alignment: A Physical Trap Model of Best-of-N Jailbreaking
逃避对齐：最佳选择-N越狱的物理陷阱模型
Abstract
Best-of-$N$ jailbreaking (BoN) bypasses safeguards of aligned models by drawing $N$ independent augmentations of an unsafe prompt and sampling $M$ completions of each. Previous works have shown that the attack success rate (ASR) seems to follow a power-law in $N$, which we challenge. The exponent drifts with $N$, with an exponential crossover which is a finite-size artifact of the adversarial dataset. Little work has been done to explore the entire two-budget ($N, M$) attack surface as well as its dependence on the generation temperature $T$. We introduce a simple barrier model where each prompt has a baseline safety level and each augmentation a random thermally activated barrier. Then four numbers, each backed by an interpretable safety mechanism, determine the entire ($N, M$) attack surface. They extrapolate predictions from $N \leq 100$ to $N = 10^4$, collapse five distinct models on the same scaling function and predict ASR at different temperatures from the one they were fitted at.
Chinese Translation
最佳选择-$N$越狱（BoN）通过对不安全提示进行$N$次独立增强并对每次增强采样$M$个完成，绕过了对齐模型的安全防护。先前的研究表明，攻击成功率（ASR）似乎遵循$N$的幂律分布，而我们对此提出质疑。指数随着$N$的变化而漂移，并且存在一个指数交叉点，这是对抗性数据集的有限大小伪影。关于整个双预算（$N, M$）攻击面及其对生成温度$T$的依赖，相关研究较少。我们引入了一个简单的障碍模型，其中每个提示具有基线安全水平，每个增强则具有一个随机的热激活障碍。然后，四个数字，每个数字都有一个可解释的安全机制，决定了整个（$N, M$）攻击面。它们将从$N extless 100$的预测外推到$N = 10^4$，将五个不同模型压缩到相同的缩放函数上，并预测在不同温度下的ASR，超出了它们所拟合的温度。
cs.AI / 32 / 2609.32123
READ-Bench: Benchmarking Historical Instance Retrieval for Time-Series Diagnosis
READ-Bench：时间序列诊断的历史实例检索基准
Abstract
Time-series diagnostic systems rarely rely on retrieving relevant historical cases, and when they do, retrieval is evaluated only indirectly through downstream prediction. We introduce READ-Bench, a benchmark for historical-case retrieval across 12 diagnostic datasets, centered on multivariate time series, that defines relevance by shared fault or event type rather than signal shape, so visually different traces of the same fault count as relevant while similar-looking traces of different faults do not. Treating retrieval as a base retriever followed by a reranker, we evaluate classical distances, symbolic retrievers, self-supervised and foundation-model embedders, and their fusion, plus label-aware and language-model rerankers, under one protocol that varies supervision, pollution, and corpus scale with significance testing. Under a common channel-independent interface, pretrained representations offer no statistically detectable advantage over strong classical and symbolic baselines for search alone. The decisive factor is a small amount of resolved-case supervision at reranking, namely a Gaussian-process reranker that propagates a few neighbor labels in embedding space, which helps far more than more sophisticated representations or language-model reasoning and holds under pollution and at full corpus scale. Guided by these findings, we fuse a normal-residual-scored embedder with a dynamic time warping leg via reciprocal-rank fusion, then rerank with the Gaussian-process reranker, improving NDCG@10 over its own search stage on all 12 datasets, by +0.11 from reranking and +0.16 over the strongest single base retriever.
Chinese Translation
时间序列诊断系统很少依赖于检索相关的历史案例，即使依赖时，检索的评估也仅通过下游预测间接进行。我们引入了READ-Bench，这是一个针对12个以多变量时间序列为中心的诊断数据集的历史案例检索基准，定义相关性为共享故障或事件类型，而非信号形状，因此相同故障的视觉上不同的轨迹被视为相关，而不同故障的相似轨迹则不被视为相关。我们将检索视为基础检索器后跟重排序器，评估经典距离、符号检索器、自监督和基础模型嵌入器及其融合，以及标签感知和语言模型重排序器，采用一个协议，该协议通过显著性测试变化监督、污染和语料库规模。在一个通用的通道独立接口下，预训练表示在仅搜索方面并未提供相对于强经典和符号基线的统计显著优势。决定性因素是在重排序时少量的已解决案例监督，即高斯过程重排序器在嵌入空间中传播少量邻居标签，这比更复杂的表示或语言模型推理帮助更大，并且在污染和全语料库规模下保持有效。根据这些发现，我们将一个正常残差评分的嵌入器与一个动态时间扭曲腿通过互惠排名融合，然后使用高斯过程重排序器进行重排序，在所有12个数据集上将NDCG@10从重排序中提高了+0.11，并在最强单一基础检索器之上提高了+0.16。
cs.AI / 33 / 2609.32166
PastForward: Faster On-Device GUI Agents via Computational Experience Reuse
PastForward：通过计算经验重用加速设备端图形用户界面代理
Abstract
Running GUI agents on edge devices can keep sensitive screens and interaction histories local, but the computational cost of inference at every action step makes deployment challenging. Existing GUI agent systems either perform full vision-language model (VLM) inference at each action step or reuse coarse-grained knowledge matched to prior tasks. However, dynamic mobile environments and user tasks make it difficult to fully utilize prior task executions without additional fine-tuning or task-specific offline exploration. To address this challenge, we present PastForward, a system that accelerates GUI agents through validated, fine-grained reuse of computational experience accumulated during ordinary task execution. During decoding, PastForward retrieves prior output sequences as device-adaptive multi-token proposals and verifies them in a single VLM forward pass. Across action steps, it uses prior GUI transitions to begin next-step inference while the device executes the current action, retains the early computation only when the predicted screen matches the observed screen, and carries reusable KV states forward. We evaluate PastForward on AndroidWorld workloads derived from real mobile usage patterns using multiple VLM backbones across server and edge platforms. On device, PastForward achieves action-step latency speedups of 1.63-2.36$\times$ while maintaining task success rates.
Chinese Translation
在边缘设备上运行图形用户界面（GUI）代理可以将敏感的屏幕和交互历史保留在本地，但每个动作步骤的推理计算成本使得部署变得具有挑战性。现有的GUI代理系统要么在每个动作步骤上执行完整的视觉-语言模型（VLM）推理，要么重用与先前任务匹配的粗粒度知识。然而，动态的移动环境和用户任务使得在没有额外微调或任务特定离线探索的情况下，充分利用先前任务执行变得困难。为了解决这一挑战，我们提出了PastForward，一个通过验证的、细粒度的计算经验重用来加速GUI代理的系统。在解码过程中，PastForward将先前的输出序列检索为设备自适应的多标记提案，并在单次VLM前向传递中对其进行验证。在动作步骤之间，它利用先前的GUI过渡开始下一步推理，同时设备执行当前动作，仅在预测的屏幕与观察到的屏幕匹配时保留早期计算，并将可重用的KV状态向前传递。我们在基于真实移动使用模式的AndroidWorld工作负载上评估了PastForward，使用多个VLM骨干网络在服务器和边缘平台上进行测试。在设备上，PastForward实现了1.63-2.36倍的动作步骤延迟加速，同时保持任务成功率。
cs.AI / 34 / 2609.32172
Noisy Test-Time Reinforcement Learning for Code LLMs
针对代码大语言模型的噪声测试时强化学习
Abstract
Large language models (LLMs) have demonstrated remarkable performance across various code-related tasks. However, unlike carefully curated datasets that are typically high-quality and error-free, real-world user instructions are often vague and error-prone, posing significant challenges to the robustness of code LLMs. Furthermore, robustness-oriented fine-tuning relies on paired clean-noisy samples, which are costly to curate and require sophisticated noisy simulation techniques. To address these challenges, we propose the Noisy Test-time Reinforcement Learning framework (NTRL-Code), which enables robust self-evolution of code LLMs using only unlabeled noisy data during the testing stage. Specifically, NTRL-Code uses conservative self-denoising to obtain a cleaner semantic anchor for target estimation, and employs an abstract-syntax-tree (AST)-based structural aggregation mechanism to estimate a proxy target from multiple candidate programs. The policy is then optimized on the original noisy prompts with a hybrid reward that combines format validity, code similarity, and anti-repetition signals. Extensive experiments on three benchmarks, each incorporating character-level, word-level, and paragraph-level perturbations, demonstrate that NTRL-Code yields robust and consistent improvements, stabilizing the predictions of various base models. Our code is available at https://github.com/Xikai97/NTRL-Code.
Chinese Translation
大型语言模型（LLMs）在各种与代码相关的任务中表现出了卓越的性能。然而，与通常高质量且无误的精心策划的数据集不同，现实世界中的用户指令往往模糊且易出错，这对代码 LLM 的鲁棒性构成了重大挑战。此外，面向鲁棒性的微调依赖于成对的干净-噪声样本，这些样本的策划成本高昂，并且需要复杂的噪声模拟技术。为了解决这些挑战，我们提出了噪声测试时强化学习框架（NTRL-Code），该框架在测试阶段仅使用未标记的噪声数据来实现代码 LLM 的鲁棒自我演化。具体而言，NTRL-Code 采用保守的自去噪方法来获得更清晰的语义锚点以进行目标估计，并使用基于抽象语法树（AST）的结构聚合机制从多个候选程序中估计代理目标。然后，策略在原始噪声提示上进行优化，采用结合格式有效性、代码相似性和反重复信号的混合奖励。对三个基准的广泛实验表明，NTRL-Code 产生了鲁棒且一致的改进，稳定了各种基础模型的预测。我们的代码可在 https://github.com/Xikai97/NTRL-Code 获取。
cs.AI / 35 / 2609.32184
AI Harness: Certification under Proposal-Conditioned Information for Foundation-Model Agents
AI Harness：基于提议条件信息的基础模型代理认证
Abstract
Foundation-model agents are often modeled as policies over an observed state. In deployed systems, however, a runtime may intervene only after the model has emitted a semantic proposal, making the proposal both an action candidate and a decision-time observation generated by a history-conditioned process. We show that collapsing this structure into a state-only proposal envelope can preserve proposal coverage while destroying certifiability. In a finite robust interface, the viability kernel of the collapsed model is contained in the physical projection of the history-augmented kernel, and the collapse is lossless exactly when every proposal-conditioned collapsed fiber retains a common robust-safe intervention. This gap can be maximal even with constant-size proposal and history alphabets. The same common-action condition yields a dual result: observing the current proposal can restore robust feasibility when it separates latent modes requiring incompatible interventions. We extend these one-step results over time using exact finite beliefs and standard safety and reachability fixed points, separating indefinite operational viability from finite worst-case verified progress. Controlled model-in-the-loop tests reproduce the predicted obstructions when telemetry or effect verification is removed or intervention authority is restricted. Thus, our contribution is not a new fixed-point calculus, but a characterization of when proposal--history correlation at the model--tool boundary is necessary for certification.
Chinese Translation
基础模型代理通常被建模为对观察状态的策略。然而，在实际部署的系统中，运行时可能仅在模型发出语义提议后进行干预，使得提议既成为行动候选项，又成为由历史条件过程生成的决策时观察。我们展示了将这种结构简化为仅基于状态的提议封装可以在保留提议覆盖的同时破坏可认证性。在有限的鲁棒接口中，简化模型的生存核包含在历史增强核的物理投影中，当每个提议条件的简化纤维保留一个共同的鲁棒安全干预时，简化是无损的。即使在常量大小的提议和历史字母表下，这一差距也可能是最大的。同样的共同行动条件产生了一个对偶结果：观察当前提议可以在其分离需要不兼容干预的潜在模式时恢复鲁棒可行性。我们使用精确有限信念和标准安全性与可达性不动点扩展了这些一步结果，区分了无限操作可行性与有限最坏情况验证进展。受控的模型环内测试在去除遥测或效果验证或限制干预权限时重现了预测的障碍。因此，我们的贡献不是一个新的不动点演算，而是对在模型工具边界上提议-历史相关性何时对于认证是必要的特征描述。
cs.AI / 36 / 2609.32192
CoMemBench: Benchmarking Collaborative Memory Boundaries across Multi-Agent Workflow Topologies
CoMemBench：跨多智能体工作流拓扑的协作记忆边界基准测试
Abstract
Multi-agent workflows require task-relevant information to be shared across agents, while irrelevant, stale, unverified, or incompatible information must remain isolated. We call this task-conditioned scope of information a collaborative memory boundary. Workflow topology determines which intermediate artifacts are applicable to which downstream workers and when they cease to be valid, thereby providing a structural stress dimension for sharing and isolation. Existing memory benchmarks primarily evaluate retention and retrieval, whereas multi-agent benchmarks emphasize coordination and end-to-end completion, leaving topology-conditioned memory boundaries largely unmeasured. We introduce CoMemBench, an execution-grounded benchmark for collaborative memory sharing and isolation across multi-agent workflow topologies. It constructs 800 composite workflows across four domains from source-grounded dependency graphs, with node-local specifications, verifiable artifact handoffs, native evaluators, and matched isolation challenges. CoMemBench measures workflow completion, verified node progress, required-handoff reliability, isolation robustness, and token cost. Experiments reveal a sharing-isolation trade-off: broader context improves information availability but can weaken isolation, while system rankings shift across topologies and artifact violations.
Chinese Translation
多智能体工作流需要在智能体之间共享与任务相关的信息，同时必须将无关、过时、未经验证或不兼容的信息隔离。我们将这种任务条件下的信息范围称为协作记忆边界。工作流拓扑决定了哪些中间工件适用于哪些下游工作者，以及它们何时不再有效，从而为共享和隔离提供了结构性压力维度。现有的记忆基准主要评估信息的保留和检索，而多智能体基准则强调协调和端到端的完成，导致拓扑条件下的记忆边界在很大程度上未被测量。我们提出了CoMemBench，这是一个基于执行的基准，用于跨多智能体工作流拓扑的协作记忆共享和隔离。它从源依赖图构建了800个跨四个领域的复合工作流，具有节点本地规范、可验证的工件交接、原生评估器和匹配的隔离挑战。CoMemBench测量工作流完成情况、验证的节点进展、所需交接的可靠性、隔离的稳健性和令牌成本。实验揭示了共享与隔离之间的权衡：更广泛的上下文改善了信息的可用性，但可能削弱隔离，而系统排名在不同拓扑和工件违规情况下会发生变化。
cs.AI / 37 / 2609.32201
Instruct, Not Answer: Using Instruction Privileges in On-Policy Context Distillation
指导，而非回答：在在线策略上下文蒸馏中使用指导特权
Abstract
On-Policy Context Distillation (OPCD) has recently emerged as a powerful technique for transferring context to student models and for self-improvement. In OPCD, the teacher is conditioned on privileged information, and the goal is to minimize the Kullback-Leibler (KL) divergence between the privileged teacher and the student, evaluated on student-generated tokens. Many existing studies show that using instance-specific gold answers or gold demonstrations as the default privilege can hurt training performance, especially out-of-distribution (OOD). In this work, we instead design general instructions that target common student mistakes observed on the training samples, and show that such simple instructions can outperform gold as the OPCD privilege. In autoformalization tasks, using a matched formatting instruction as the privilege could outperform gold in OOD accuracy by a large margin. In 7 out of 8 experiments using ProverQA, ProofWriter, and ProntoQA as datasets, and Qwen3-Thinking and Olmo3-Thinking families as models, matched instruction privileges outperform gold in OOD by 4 to 17 points, while remaining on par with gold in-domain. Each instruction is only a few sentences (and thus contains much less information compared to all instance-specific gold) and is applied uniformly to every training sample. These results indicate that a general instruction, which applies equally to source and target domain examples, can be substantially more transferable than instance-specific gold in OPCD while maintaining in-domain performance.
Chinese Translation
在线策略上下文蒸馏（On-Policy Context Distillation, OPCD）最近作为一种强大的技术出现，用于将上下文转移到学生模型并进行自我改进。在OPCD中，教师模型基于特权信息进行条件化，目标是最小化特权教师与学生之间的Kullback-Leibler（KL）散度，该散度是在学生生成的标记上进行评估的。许多现有研究表明，使用特定实例的黄金答案或黄金示例作为默认特权可能会损害训练性能，尤其是在分布外（OOD）情况下。在本研究中，我们设计了一般性指导，针对在训练样本中观察到的常见学生错误，并展示了这种简单指导可以在OPCD特权中超越黄金答案。在自动形式化任务中，使用匹配格式的指导作为特权在OOD准确性上可能大幅超越黄金答案。在使用ProverQA、ProofWriter和ProntoQA作为数据集，以及Qwen3-Thinking和Olmo3-Thinking系列作为模型的8个实验中，有7个实验表明，匹配指导特权在OOD上比黄金答案高出4到17分，同时在领域内表现与黄金答案相当。每个指导仅由几句话组成（因此相比于所有特定实例的黄金答案包含的信息要少得多），并且均匀地应用于每个训练样本。这些结果表明，一般性指导在OPCD中可以比特定实例的黄金答案具有更强的可转移性，同时保持领域内的性能。
cs.AI / 38 / 2609.32208
Witness: Discovery, Deciphering, and Epiphany in Interactive Puzzle Environments
见证：互动谜题环境中的发现、解码与顿悟
Abstract
Automated science needs agents that can work out the rules of an unfamiliar environment by interacting with it. Interactive rule-discovery puzzles offer a controlled setting for studying this ability: an agent infers hidden rules through experimentation and uses what it has inferred to reach a stated goal. We ask what limits current language models on these puzzles and whether reinforcement learning (RL) improves performance on rules held out from training. To study both, we introduce WITNESS, a 2D grid-based puzzle environment with ground-truth ASCII observations and controlled access to rules. An agentic pipeline generates games for WitnessGym, the RL training suite, and WitnessBench, comprising public validation and private test games. The validation set separately tests new compositions of trained rule primitives and primitives absent from training. Under a shared harness, the best of 18 frontier proprietary and open-weight models solves only 24\% of private test level slots, with scores sensitive to the observation interface and agent configuration. Providing ground-truth rules raises Opus-5's validation RHAE-L5 (relative human action efficiency over the first five levels) from 59.9 to 97.8, whereas a 27B open-weight model gains only 2.1 points and remains limited even with the rules provided. RL on WitnessGym raises the 27B model's private test RHAE-L5 from 2.1 to 5.4 and yields a mean gain of 4.1 points on four external discovery benchmarks. Together, these results point to rule acquisition as a major difficulty for frontier models like Opus-5 while smaller models further struggle on rule-based execution, and indicate that RL on hidden-rule puzzles transfers to broader rules and real-world tasks beyond training. Benchmark is available at: https://witnessbench.ai
Chinese Translation
自动化科学需要能够通过与不熟悉环境的互动来推导规则的智能体。互动规则发现谜题提供了一个受控的环境来研究这一能力：智能体通过实验推断隐藏规则，并利用所推断的内容来达到既定目标。我们探讨了当前语言模型在这些谜题上的局限性，以及强化学习（RL）是否能改善在训练中未包含规则的表现。为此，我们引入了WITNESS，一个基于2D网格的谜题环境，具有真实的ASCII观察数据和对规则的受控访问。智能体管道为WitnessGym（RL训练套件）和WitnessBench（包括公共验证和私人测试游戏）生成游戏。验证集单独测试训练的规则原语的新组合以及训练中缺失的原语。在一个共享的框架下，18个前沿专有和开放权重模型中表现最佳的模型仅解决了24%的私人测试关卡，其得分对观察接口和智能体配置敏感。提供真实规则使Opus-5的验证RHAE-L5（前五个关卡的相对人类行动效率）从59.9提高到97.8，而一个27B的开放权重模型仅增加了2.1分，即使在提供规则的情况下仍然受限。在WitnessGym上进行的RL训练使27B模型的私人测试RHAE-L5从2.1提高到5.4，并在四个外部发现基准上平均增加了4.1分。这些结果表明，规则获取是像Opus-5这样的前沿模型面临的主要困难，而较小的模型在基于规则的执行上进一步受限，并表明在隐藏规则谜题上的RL训练可以转移到更广泛的规则和超出训练的现实任务。基准可在以下网址获取：https://witnessbench.ai
cs.AI / 39 / 2609.32209
Fracast-0: Fractal Weight Sharing for a Time Series Foundation Model with Only 85K Parameters
Fracast-0：仅用85K参数的时间序列基础模型的分形权重共享
Abstract
Time series foundation models must preserve multi-domain breadth, probabilistic output, and multiple temporal scales, but parameter count grows when each scale receives a separate representation. We introduce Fracast-0, a probabilistic forecasting foundation model that exploits temporal self-similarity to reuse one operator across scales. A parameter-free detector extracts significant seasonal structure. The encoder applies a shared local block along a geometric dilation ladder with scale conditioning, while the decoder combines context-gathered states with an explicit seasonal future state and reuses a second block along another ladder before emitting nine quantiles. Pretraining across six corpora preserves multi-domain breadth within 85,001 parameters. On 97 GIFT-Eval configurations without per-dataset fine-tuning, Fracast-0 is the smallest of 28 evaluated checkpoints and remains non-dominated in the aggregate parameter-accuracy plane with MASE 0.808 and WQL 0.564. It uses 42.0% fewer parameters than TinyCast, whose MASE and WQL are 4.2% and 3.3% lower. These results support cross-scale weight reuse as a practical route to further time series foundation model compression.
Chinese Translation
时间序列基础模型必须保持多领域广度、概率输出和多种时间尺度，但当每个尺度获得单独表示时，参数数量会增加。我们提出了Fracast-0，这是一种利用时间自相似性在不同尺度之间重用一个操作符的概率预测基础模型。一个无参数的检测器提取显著的季节性结构。编码器沿着几何膨胀梯度应用共享局部块，并结合尺度条件，而解码器则将上下文收集的状态与显式的季节性未来状态结合，并在发出九个分位数之前沿着另一个梯度重用第二个块。在六个语料库上进行的预训练在85,001个参数内保持了多领域广度。在97个GIFT-Eval配置中，Fracast-0是28个评估检查点中参数最少的，并且在参数-准确率平面上保持非主导地位，MASE为0.808，WQL为0.564。与TinyCast相比，它使用的参数少42.0%，而TinyCast的MASE和WQL分别低4.2%和3.3%。这些结果支持跨尺度权重重用作为进一步压缩时间序列基础模型的实用途径。
cs.AI / 40 / 2609.32220
A bilingual AI audiologist built through rubric-guided playbook induction outperforms human audiologists in a blinded evaluation of simulated cases
通过基于评分标准的剧本归纳构建的双语人工智能听力学家在盲评模拟案例中优于人类听力学家
Abstract
Audiology consultation requires structured history-taking, audiometric interpretation and patient-centred communication, yet real-world case material is scarce. We present a bilingual AI audiologist pairing a general-purpose large language model with rubric-guided playbook induction, multimodal audiogram interpretation and retrieval-augmented grounding, without fine-tuning the language-model backbone. Using a 21-item rubric and an AI patient simulator, we induced a 19-rule consultation policy from 73 training cases (43 English, 30 Chinese) and evaluated the system on 58 independent simulated cases (30 Chinese, 28 English) in a pre-specified, source-blinded comparison with 17 practising audiologists. The AI audiologist outperformed human audiologists on every case (58/58; mean paired $\Delta$ = +1.35 on a 5-point composite, Cohen's d = 1.84, $P = 4.5 \times 10^{-20}$), on 20 of 21 rubric items and in both languages. Component ablation identified the playbook as the largest contributor, offering a practical route to specialist consultation agents in low-data medical domains.
Chinese Translation
听力学咨询需要结构化的病史采集、听力测量解读和以患者为中心的沟通，但现实世界的案例材料稀缺。我们提出了一种双语人工智能听力学家，将通用大型语言模型与基于评分标准的剧本归纳、多模态听力图解读和增强检索基础相结合，而无需对语言模型的主干进行微调。使用21项评分标准和一个人工智能患者模拟器，我们从73个训练案例（43个英语，30个中文）诱导出19条咨询政策，并在预先指定的、源盲的比较中对58个独立模拟案例（30个中文，28个英语）进行了评估，参与比较的有17名执业听力学家。人工智能听力学家在每个案例中都优于人类听力学家（58/58；平均配对$ ext{Δ}$ = +1.35，基于5分制综合评分，Cohen's d = 1.84，$P = 4.5 imes 10^{-20}$），在21项评分标准中的20项以及两种语言中均表现优异。组件消融实验表明，剧本是最大的贡献者，为低数据医疗领域的专业咨询代理提供了一条实用的途径。
cs.AI / 41 / 2609.32224
RAO-Nav: Probing Omni-Language Models for Zero-shot Semantic Audio-Visual Navigation
RAO-Nav：探讨全语言模型在零-shot语义音视频导航中的应用
Abstract
We explore whether Omni-Language Models (OLMs) can be directly applied to zero-shot Semantic Audio-Visual Navigation (SAVN). Recent work demonstrates that even state-of-the-art specialized models still struggle to achieve generalist multimodal navigation, despite extensive task-specific training. In this paper, we introduce RAO-Nav, short for Reasoning All-in-One OLM, a deployment pipeline for zero-shot SAVN. By leveraging the rich implicit audio-visual knowledge encoded in OLMs, the embodied agent is enabled to ``hear'', ``see'', ``reason'', and ``act'' in the environment. To further elicit the built-in thinking ability of OLMs, we propose a test-time Latent Navigation Reasoning (LNR) module that can be seamlessly integrated into the decoding space. LNR encourages the model to retrieve more target-relevant observations and make effective navigation decisions. Through comprehensive experiments, we show that our framework surpasses existing state-of-the-art baselines on public SAVN benchmarks without using any training data. Moreover, we introduce a new \emph{Global Navigation Instruction} setting to further evaluate the ability of OLMs to serve as embodied navigation agents. Code: https://github.com/rikeilong/OmniAV\_Nav.
Chinese Translation
我们探讨全语言模型（Omni-Language Models, OLMs）是否可以直接应用于零-shot语义音视频导航（Semantic Audio-Visual Navigation, SAVN）。近期研究表明，即使是最先进的专用模型，在经过大量特定任务训练后，仍然难以实现通用的多模态导航。本文介绍了RAO-Nav（Reasoning All-in-One OLM的缩写），这是一个用于零-shot SAVN的部署管道。通过利用OLMs中编码的丰富隐式音视频知识，具身代理能够在环境中“听”、“看”、“推理”和“行动”。为了进一步激发OLMs内置的思维能力，我们提出了一种测试时潜在导航推理（Latent Navigation Reasoning, LNR）模块，该模块可以无缝集成到解码空间中。LNR鼓励模型检索更多与目标相关的观察，并做出有效的导航决策。通过全面的实验，我们展示了我们的框架在公共SAVN基准测试中超越了现有的最先进基线，而无需使用任何训练数据。此外，我们引入了一种新的全球导航指令（Global Navigation Instruction）设置，以进一步评估OLMs作为具身导航代理的能力。代码链接： https://github.com/rikeilong/OmniAV_Nav.
cs.AI / 42 / 2609.32247
Certifying Interventional Agreement Among Observationally Equivalent Causal Models
认证观察等价因果模型之间的干预一致性
Abstract
Observationally equivalent causal models can still disagree about what happens under intervention, because interventions create inputs that never occur in observational data. We introduce Interventional Separation Selection (ISS), which repeatedly queries the true system with an admissible intervention on which the surviving candidate models disagree, discards the candidates the outcome contradicts, and stops once no intervention within a cost bound separates the survivors. If the true system is among the candidates, this stopping condition certifies that every survivor agrees with it on every admissible intervention within the bound, a guarantee that no observational learner can give, however much data it sees. The stopping condition depends only on the survivors, so it can be checked without knowing the truth. For continuous variables the candidates form an infinite version space, and mixed-integer linear programs decide the stopping condition exactly over all of it, with agreement holding up to a tolerance. On a three-digit colored MNIST causal abstraction task in which ink hue tracks digit size, plain convolutional networks trained on examples reach zero held-out error, yet disagree with shape-based labels on 26% of single-digit edits, as often as hue-based labels do. Auditing the causal abstractions of networks observed only on such images, ISS certifies what each network perceives with 13.6 interventions per image on average, and each certificate, checked against every admissible intervention, holds whenever the network's true abstraction is among the candidates. When a network bypasses a unit that every candidate abstraction relies on, certificates covering interventions on that unit can be silently void, and twenty random validation interventions refute 69% of them.
Chinese Translation
观察等价因果模型在干预下可能仍然存在分歧，因为干预创造了在观察数据中从未发生的输入。我们引入了干预分离选择（Interventional Separation Selection, ISS），该方法反复查询真实系统在候选模型存在分歧的可接受干预下的表现，剔除与结果相矛盾的候选模型，并在没有任何在成本范围内的干预能够分离幸存者时停止。如果真实系统在候选模型中，则该停止条件证明每个幸存者在所有可接受的干预下都与其一致，这是任何观察学习者无法提供的保证，无论其看到多少数据。停止条件仅依赖于幸存者，因此可以在不知道真实情况的情况下进行检查。对于连续变量，候选模型形成一个无限的版本空间，而混合整数线性程序能够精确决定整个版本空间上的停止条件，并在一定容忍度内保持一致。在一个三位数彩色MNIST因果抽象任务中，墨水色调与数字大小相关，经过训练的普通卷积网络在示例上达到零持出误差，但在26%的单数字编辑中与基于形状的标签存在分歧，分歧频率与基于色调的标签相同。对仅在此类图像上观察到的网络的因果抽象进行审计，ISS在平均每幅图像13.6次干预下认证了每个网络的感知，并且每个证书在检查所有可接受干预时均有效，只要网络的真实抽象在候选模型中。当网络绕过每个候选抽象所依赖的单元时，覆盖该单元干预的证书可能会默默失效，而二十个随机验证干预则驳斥了69%的证书。
cs.AI / 43 / 2609.32254
Why Directly Learning Periodic Trajectories Can Fail
为何直接学习周期轨迹可能失败
Abstract
Operator learning of periodic solutions requires deciding how simulation data should be recorded and represented. A natural choice is to integrate long enough for transients to decay and record a window wide enough to contain at least one full period of all trajectories. We find that these conservative choices can make the resulting trajectories difficult to learn, even when the underlying periodic orbits vary regularly with system parameters. Unaligned trajectories generalize poorly even within the training distribution. Phase alignment substantially improves in-distribution generalization, but models trained on a fixed physical-time window still have large errors on trajectories with periods outside the training range. We explain both failures through a common mechanism: frequency differences accumulate over time, so the target phase varies rapidly with the parameters. Predictors that cannot track this variation incur a population MSE floor in both settings; for fixed window prediction, we also derive a per-sample lower bound. We then study one of the simplest representations that escape these floors: learning an aligned, normalized waveform and its period separately. We establish regularity of the decoupled targets under ODE assumptions and show experimentally that this approach avoids both failures in ODE systems and a PDE case study.
Chinese Translation
周期解的算子学习需要决定如何记录和表示模拟数据。一个自然的选择是进行足够长的积分，以使瞬态衰减，并记录一个足够宽的窗口，以包含所有轨迹的至少一个完整周期。我们发现，这些保守的选择可能使得结果轨迹难以学习，即使在基础的周期轨道随着系统参数规则变化的情况下也是如此。未对齐的轨迹即使在训练分布内也表现出较差的泛化能力。相位对齐显著改善了分布内的泛化能力，但在固定物理时间窗口上训练的模型在周期超出训练范围的轨迹上仍存在较大误差。我们通过一个共同机制解释了这两种失败：频率差异随时间累积，因此目标相位随参数快速变化。无法追踪这种变化的预测器在这两种情况下都会遭遇一个群体均方误差（MSE）下限；对于固定窗口预测，我们还推导出每个样本的下限。然后，我们研究了一种逃避这些下限的最简单表示之一：分别学习对齐的、归一化的波形及其周期。我们在常微分方程（ODE）假设下建立了解耦目标的规律性，并通过实验表明这种方法避免了常微分方程系统中的两种失败以及一个偏微分方程（PDE）案例研究。
cs.AI / 44 / 2609.32255
Clarify the User or Verify the World? Uncertainty Routing for Proactive Agents
澄清用户还是验证世界？主动代理的不确定性路由
Abstract
Tool-using LLM agents must decide not only whether additional information is needed, but also which source can resolve the uncertainty. Existing proactive approaches often specialize in either user clarification or environment verification, without explicitly determining the appropriate information source for each decision. We formulate this problem as uncertainty routing among ACT, CLARIFY, and VERIFY, and propose PROUR, a proactive uncertainty routing framework. PROUR decomposes action uncertainty into disagreement across plausible user-goal interpretations, which signals user-side ambiguity, and the entropy remaining within each interpretation, which signals missing world-side evidence. To acquire information from the routed source, a query generator is trained with a mode-conditioned information-gain reward, targeting user-goal identification under CLARIFY and next-action identification under VERIFY. On $\tau$-bench, PROUR achieves 28.17% average success rate across retail and airline, outperforming the strongest prior method by 4.57% while using 2.17 fewer interaction steps. The learned policy further generalizes to stronger task agents and transactional domains of $\tau^3$-bench without retraining, demonstrating the benefit of source-aligned uncertainty resolution for proactive agents.
Chinese Translation
使用工具的LLM代理不仅必须决定是否需要额外的信息，还必须确定哪个来源可以解决不确定性。现有的主动方法通常专注于用户澄清或环境验证，而未明确确定每个决策的适当信息来源。我们将这个问题表述为在ACT、CLARIFY和VERIFY之间的不确定性路由，并提出了PROUR，一个主动不确定性路由框架。PROUR将行动不确定性分解为在合理的用户目标解释之间的分歧，这表明用户方面的模糊性，以及每个解释中剩余的熵，这表明缺失的世界方面证据。为了从路由来源获取信息，查询生成器通过模式条件的信息增益奖励进行训练，针对CLARIFY下的用户目标识别和VERIFY下的下一步行动识别。在$ au$-bench上，PROUR在零售和航空领域实现了28.17%的平均成功率，比最强的先前方法提高了4.57%，同时减少了2.17个交互步骤。所学策略进一步推广到更强的任务代理和$ au^3$-bench的交易领域，无需重新训练，展示了源对齐的不确定性解决对主动代理的好处。
cs.AI / 45 / 2609.32256
LAM: Efficient Lossy Agent Memory Framework With A Retrieval-Score Error Bound
LAM：具有检索得分误差界限的高效有损代理记忆框架
Abstract
Agent memory grows as agents read inputs, reason, and call tools. Longer histories increase inference cost and eventually exceed the context window. LLM-based summarization reduces this history but adds latency and provides no explicit bound on information loss. We propose LAM, a Lossy Agent Memory system with three components: a deterministic deduplication rule with a substitution bound on retrieval scores - a bound on score perturbation, not a certificate of unchanged ranking; a memory manager that preserves the cached prefix and overlaps compaction with inference; and a performance model that estimates compaction costs before deployment. On 600 agent trajectories, LAM removes 22.47% of observation tokens while retaining 99.984% of the measured gold-patch evidence. At a fixed deletion set, the performance model predicts a 71.4x-91.6x end-to-end speedup from removing records before prefill instead of deleting them from a prefilled context. That benefit comes from the schedule rather than the rule and applies to any prefix-preserving test.
Chinese Translation
代理记忆随着代理读取输入、推理和调用工具而增长。更长的历史会增加推理成本，并最终超出上下文窗口。基于大型语言模型（LLM）的摘要减少了这一历史，但增加了延迟，并且没有提供信息损失的明确界限。我们提出了LAM，一种有损代理记忆系统，包含三个组件：一个具有替代界限的确定性去重规则——这是对检索得分扰动的界限，而不是不变排名的证明；一个内存管理器，保留缓存前缀并将重叠压缩与推理结合；以及一个性能模型，在部署前估算压缩成本。在600个代理轨迹上，LAM去除了22.47%的观察令牌，同时保留了99.984%的测量金证据。在固定删除集下，性能模型预测通过在预填充之前删除记录而不是从预填充上下文中删除记录，可以实现71.4倍至91.6倍的端到端加速。这一好处来自于调度而非规则，并适用于任何保留前缀的测试。
cs.AI / 46 / 2609.32259
Prefill-Free Cross-Family KV Cache Transfer for Heterogeneous Multi-Agent LLMs
无预填充的跨家族KV缓存转移方法用于异构多智能体大语言模型
Abstract
Recent multi-agent LLM systems increasingly combine heterogeneous models for specialized agent roles. However, text-based communication requires each receiver to prefill shared context already processed by the sender. Reusing the sender's key-value (KV) cache avoids this redundancy, but prefill-free transfer across model families must handle differences in tokenization, model depth, and KV representations. To address these issues, we propose \textit{HeteroFold}, a prefill-free cross-family KV cache transfer method that keeps both the sender and receiver frozen. HeteroFold aligns model structures, maps the sender cache into the receiver space, and calibrates it to preserve receiver behavior. Across six transfer directions, HeteroFold achieves the best cache-transfer performance on all four long-context benchmarks and most short-context settings. It also matches text-based communication on the multi-agent benchmark. At 32K context length, Llama-3.1-8B$\rightarrow$Ministral-3-14B transfer is $10.7\times$ faster than Native Prefill and $1.18$--$1.47\times$ faster than the state-of-the-art prefill-free baselines, Dense Latent and KV Ridge. These results show that HeteroFold enables efficient cross-family KV reuse without receiver prefill.
Chinese Translation
近期的多智能体大语言模型系统越来越多地结合异构模型以满足特定智能体角色的需求。然而，基于文本的通信要求每个接收者预填充已由发送者处理的共享上下文。重用发送者的键值（KV）缓存可以避免这种冗余，但跨模型家族的无预填充转移必须处理标记化、模型深度和KV表示的差异。为了解决这些问题，我们提出了 extit{HeteroFold}，一种无预填充的跨家族KV缓存转移方法，该方法保持发送者和接收者的状态不变。HeteroFold对齐模型结构，将发送者缓存映射到接收者空间，并进行校准以保持接收者的行为。在六个转移方向上，HeteroFold在所有四个长上下文基准测试和大多数短上下文设置中实现了最佳的缓存转移性能。它在多智能体基准测试中也与基于文本的通信相匹配。在32K上下文长度下，Llama-3.1-8B$
ightarrow$Ministral-3-14B的转移速度比原生预填充快$10.7 imes$，比最先进的无预填充基线Dense Latent和KV Ridge快$1.18$--$1.47 imes$。这些结果表明，HeteroFold能够在不需要接收者预填充的情况下实现高效的跨家族KV重用。
cs.AI / 47 / 2609.32274
When Does a Skill Add Value? Task-Conditional Gain Prediction for Selective Skill Use
技能何时增值？选择性技能使用的任务条件增益预测
Abstract
Agent skills are expected to improve task performance. Yet we find that they often provide no benefit, and can even hurt performance while incurring additional token costs. Can we predict whether a skill will help before the agent acts? We introduce SkillDelta, a framework for predicting task-conditional skill gains from paired executions of the same agent with and without the skill. A local predictor transfers these historical gains to new tasks without retraining the agent. Under explicit transfer assumptions, our analysis links support coverage, representation mismatch, and execution noise to prediction error and decision regret. Across five benchmarks and three target agents, paired history improves observed-gain ranking over skill-assisted outcomes alone in 12 of 15 settings. At matched expected skill-use rates, SkillDelta improves success over random activation in all 15 settings, with an average absolute gain of 4.3%. Most of this advantage comes from allocation across task groups. Evidence for additional within-group selection value is strongest on ToolQA and weaker elsewhere. Code is available at https://github.com/TankTechnology/skilldelta.
Chinese Translation
代理技能被期望能够提高任务表现。然而，我们发现它们往往没有带来任何好处，甚至可能在增加额外代币成本的情况下损害表现。我们能否在代理行动之前预测某项技能是否会有帮助？我们提出了SkillDelta，一个用于预测任务条件下技能增益的框架，该框架基于同一代理在有技能和无技能情况下的配对执行历史。一个局部预测器将这些历史增益转移到新的任务上，而无需重新训练代理。在明确的转移假设下，我们的分析将支持覆盖、表示不匹配和执行噪声与预测误差和决策悔恨联系起来。在五个基准和三个目标代理中，配对历史在15种设置中的12种情况下改善了观察到的增益排名，相较于仅依赖技能辅助的结果。在匹配的预期技能使用率下，SkillDelta在所有15种设置中都提高了成功率，相较于随机激活的平均绝对增益为4.3%。这种优势大部分来自于任务组之间的分配。在ToolQA上，组内选择价值的额外证据最为强烈，而在其他地方则较弱。代码可在 https://github.com/TankTechnology/skilldelta 获取。
cs.AI / 48 / 2609.32295
GLIDE: Generalized Layer-wise Intrinsic Distributional Evaluation for Heterogeneous LLM Agents
GLIDE：异构LLM代理的广义层级内在分布评估
Abstract
LLM agents require reliable step-level evaluation to compare candidate branches and allocate computation effectively. However, lightweight evaluation remains challenging. External verifiers introduce additional inference cost, while agent-produced confidence or self-evaluation scores can be miscalibrated, especially when candidates are generated by heterogeneous agents. We propose \textbf{G}eneralized \textbf{L}ayer-wise \textbf{I}ntrinsic \textbf{D}istributional \textbf{E}valuation (\textbf{GLIDE}) for LLM agents. \textsc{GLIDE} derives intrinsic step evidence from layer-wise residual coherence, which measures whether local residual updates consistently support the global residual change induced by a candidate step. It calibrates this evidence against the recent score distribution of the generating agent and converts it into a pessimistic reward that jointly accounts for absolute residual evidence and agent-relative standing. The reward provides a cross-agent value signal for MCTS branch selection, while normalized predictive uncertainty guides adaptive branching. Experiments on multi-hop reasoning, sequential decision making, and symbolic logic show that \textsc{GLIDE} improves task performance, step-level ranking quality, and computational efficiency without external verifiers or task-specific supervision.
Chinese Translation
LLM代理需要可靠的逐步评估，以比较候选分支并有效分配计算资源。然而，轻量级评估仍然具有挑战性。外部验证器引入了额外的推理成本，而代理生成的置信度或自我评估分数可能会出现误校准，尤其是在候选者由异构代理生成时。我们提出了 extbf{G}eneralized extbf{L}ayer-wise extbf{I}ntrinsic extbf{D}istributional extbf{E}valuation（ extbf{GLIDE}）用于LLM代理。 extsc{GLIDE}从层级残差一致性中推导出内在的逐步证据，该一致性衡量局部残差更新是否一致地支持候选步骤引起的全局残差变化。它将该证据与生成代理的近期分数分布进行校准，并将其转换为一个悲观奖励，该奖励共同考虑绝对残差证据和代理相对地位。该奖励为MCTS分支选择提供了跨代理的价值信号，而归一化的预测不确定性则指导自适应分支。在多跳推理、顺序决策和符号逻辑的实验中， extsc{GLIDE}提高了任务性能、逐步排名质量和计算效率，无需外部验证器或任务特定的监督。
cs.AI / 49 / 2609.32297
Agentsensus: Consensus-Compressed Shared Memory for Multi-Agent Story Worlds
Agentsensus：多智能体故事世界的共识压缩共享内存
Abstract
A agentic story world is a dynamic system simulating who learned what, when, and from whom -- yet the standard design gives each character a private memory stream. A shared event is therefore stored once per witness, large duplication will be incurred in terms of storage. We present Agentsensus, a story-world simulation framework in which there is an unified long-term memory. Records of the same event merge into one owned by all its witnesses, and semantically relevant memory records are linked. We evaluate on four worlds -- two classical Chinese novels, Hamlet, and a real-world conflict timeline -- run for 40 to 80 rounds against three per-character memory designs under an equal-granularity protocol. Agentsensus writes 22-44% fewer entries than the closest baseline and is the only design whose memory becomes shared (14-28% of records held by more than one character, some by 10) and linked (94-99%), at judged simulation quality indistinguishable or even better than the baselines. An ablation attributes this to the merge itself: disabling it multiplies the store by 3.1x and takes sharing to exactly zero. Sharing also compounds with the horizon rather than saturating early, rising 6% to 9% to 14% as one world is re-run at 10, 20 and 40 rounds.
Chinese Translation
一个智能体故事世界是一个动态系统，模拟谁在何时从谁那里学到了什么——然而，标准设计为每个角色提供了一个私有的记忆流。因此，每个共享事件只存储一次，导致存储方面的大量重复。我们提出了Agentsensus，一个故事世界模拟框架，其中存在统一的长期记忆。相同事件的记录合并为所有见证者共同拥有的记录，并且语义相关的记忆记录被链接。我们在四个世界上进行了评估——两个经典的中国小说、《哈姆雷特》和一个现实世界的冲突时间线——在相等粒度协议下，运行40到80轮，比较三种每角色记忆设计。Agentsensus的条目数比最接近的基线少22-44%，并且是唯一一个其记忆变为共享（14-28%的记录被多个角色持有，有些被10个角色持有）和链接（94-99%）的设计，其模拟质量被评估为与基线无差别甚至更好。一项消融实验将这一结果归因于合并本身：禁用合并使存储量增加了3.1倍，并使共享完全为零。共享还与时间范围成倍增加，而不是早期饱和，随着一个世界在10、20和40轮的重新运行，增加了6%到9%到14%。
cs.AI / 50 / 2609.32303
Train4Merge: A Controlled Single-Teacher Study of RL vs. SFT Teachers for OPD-Based Model Merging
Train4Merge：关于基于OPD的模型合并中RL与SFT教师的受控单教师研究
Abstract
Domain experts trained from a shared checkpoint can be merged into one model through on-policy distillation (OPD), where they act as teachers supervising a student on its own trajectories. One upstream choice is rarely examined: whether to build each expert with supervised fine-tuning (SFT) or reinforcement learning (RL). Yet equally strong teachers need not be equally good teachers. We probe this choice through controlled single-teacher OPD, a building block of multi-teacher OPD: in Agentic, Reasoning, and Perception, comparably performing SFT and RL teachers are trained from Qwen3.5-9B, each guiding a student initialized from it. At their best checkpoints, RL-guided students outperform SFT-guided students by 4.27, 1.50, and 0.86 percentage points in Agentic, Reasoning, and Perception, respectively, and recover more of their teachers' performance gains over the base model. The contrast is clearest in Agentic, where the best SFT-guided student recovers only 44.44% of its teacher's gain, whereas the best RL-guided student recovers 115.00%, surpassing its teacher. Our analysis points to an explanation: RL teachers stay much closer to the shared initialization in parameter space than SFT teachers and are therefore easier for their students to follow.
Chinese Translation
来自共享检查点的领域专家可以通过在线策略蒸馏（OPD）合并为一个模型，在此过程中，他们作为教师监督学生在自身轨迹上的学习。一个上游选择很少被研究：是通过监督微调（SFT）还是强化学习（RL）来构建每个专家。然而，同样强大的教师不一定是同样优秀的教师。我们通过受控单教师OPD探讨这一选择，这是多教师OPD的基础构件：在Agentic、Reasoning和Perception任务中，分别从Qwen3.5-9B训练出表现相当的SFT和RL教师，各自指导一个从其初始化的学生。在最佳检查点时，RL指导的学生在Agentic、Reasoning和Perception任务中分别比SFT指导的学生超出4.27、1.50和0.86个百分点，并且在基础模型上恢复了更多教师的性能提升。在Agentic任务中，这种对比最为明显，最佳的SFT指导学生仅恢复了其教师增益的44.44%，而最佳的RL指导学生则恢复了115.00%，超越了其教师。我们的分析指出了一个解释：RL教师在参数空间中与共享初始化保持得更近，因此更容易被学生跟随。
cs.AI / 51 / 2609.32309
PhiFold: Towards Dynamic Protein Design with Physics-Structured Covariance Modeling
PhiFold：基于物理结构协方差建模的动态蛋白质设计
Abstract
Protein design is moving beyond structural correctness toward function-aware design, yet existing generative models typically treat dynamics as a downstream property estimated through simulation or prediction after structure generation. Using MD trajectories as a generative target is also undesirable because stochastic, path-dependent trajectories over-specify the underlying equilibrium ensemble. We introduce PhiFold, a framework for jointly generating protein backbones and their second-order dynamics, represented by residue-displacement covariance. Rather than predicting the quadratically sized full covariance, PhiFold decomposes dynamics into three interpretable components: local flexibility, a low-rank collective-motion representation, and residue-wise collective participation. These components are assembled into a positive-definite covariance matrix with exact marginal consistency, yielding a compact and physically constrained representation of equilibrium dynamics. Across generated proteins, PhiFold improves recovery of local fluctuations and long-range residue coupling while remaining competitive on dominant collective-motion subspaces. It further enables bidirectional control of residue flexibility while preserving backbone designability. By unifying structure generation with an explicit representation of equilibrium dynamics, PhiFold lays a foundation for designing proteins not only by how they look, but also by how they move.
Chinese Translation
蛋白质设计正从结构正确性向功能感知设计发展，然而现有的生成模型通常将动态视为在结构生成后通过模拟或预测估计的下游属性。使用分子动力学（MD）轨迹作为生成目标也是不理想的，因为随机的、路径依赖的轨迹过度指定了潜在的平衡集。我们引入了PhiFold，一个联合生成蛋白质骨架及其二阶动态的框架，动态通过残基位移协方差表示。PhiFold并不是预测大小为四次方的完整协方差，而是将动态分解为三个可解释的组成部分：局部灵活性、低秩集体运动表示和残基级别的集体参与。这些组成部分被组装成一个具有精确边际一致性的正定协方差矩阵，从而产生一个紧凑且物理约束的平衡动态表示。在生成的蛋白质中，PhiFold改善了局部波动和长程残基耦合的恢复，同时在主导的集体运动子空间中保持竞争力。它进一步实现了对残基灵活性的双向控制，同时保持了骨架的设计性。通过将结构生成与平衡动态的显式表示统一，PhiFold为设计蛋白质奠定了基础，不仅考虑它们的外观，还考虑它们的运动方式。
cs.AI / 52 / 2609.32312
Delayed Supervision for Test-Time Language Models
测试时语言模型的延迟监督
Abstract
Test-time language models adapt a compact memory while processing the input sequence. This perspective encompasses nonlinear fast-weight learning in LaCT, associative delta-rule updates in DeltaNet, and generalized delta-rule state updates in RWKV-7. Training these models to predict the next token does not explicitly require a fact to remain accessible after many subsequent memory updates. We study delayed supervision for this test-time memory: during post-training, ask a simulator-grounded question only after a long interval of unrelated events, and supervise its answer alongside ordinary next-token prediction. Questions are evaluated on disposable branches, so their answers never enter the continuing event stream. The construction distinguishes retention from revision: a retained fact must remain valid throughout the delay, whereas a revised fact must be answered with its latest value. We evaluate this approach on LaCT-760M and plain DeltaNet-1.3B using TextWorld training trajectories and shared BABILong and RULER evaluation panels, and include a separately reported RWKV-7 comparison. Relative to event-only training, delayed QA improves BABILong by 5.48 percentage points for LaCT and 1.32 points for DeltaNet, and single-needle RULER by 1.45 and 3.27 points, respectively. The RWKV-7 comparison reports gains of 4.60 and 7.00 points on its own panels. These results support delayed semantic supervision as a practical outer training objective for usable test-time memory, while leaving open how much of the benefit derives specifically from delay rather than general question-answering and answer-termination supervision.
Chinese Translation
测试时语言模型在处理输入序列时适应一个紧凑的记忆。这一视角涵盖了 LaCT 中的非线性快速权重学习、DeltaNet 中的关联性增量规则更新，以及 RWKV-7 中的广义增量规则状态更新。训练这些模型以预测下一个标记并不明确要求事实在经过多次后续记忆更新后仍然可访问。我们研究了这种测试时记忆的延迟监督：在后训练阶段，仅在经历一段长时间的无关事件后提出一个基于模拟器的问题，并在普通的下一个标记预测旁边监督其答案。问题在一次性分支上进行评估，因此它们的答案从未进入持续的事件流。该构造区分了保留与修订：保留的事实必须在延迟期间保持有效，而修订的事实必须用其最新值进行回答。我们在 LaCT-760M 和普通的 DeltaNet-1.3B 上评估了这种方法，使用 TextWorld 训练轨迹和共享的 BABILong 及 RULER 评估面板，并包括单独报告的 RWKV-7 比较。相较于仅事件训练，延迟问答（QA）使 LaCT 的 BABILong 提高了 5.48 个百分点，DeltaNet 提高了 1.32 个百分点，单针 RULER 分别提高了 1.45 和 3.27 个百分点。RWKV-7 比较报告在其自身面板上获得了 4.60 和 7.00 个百分点的提升。这些结果支持将延迟语义监督作为可用测试时记忆的实际外部训练目标，同时仍需探讨这种益处有多少具体源自延迟，而非一般的问答和答案终止监督。
cs.AI / 53 / 2609.32326
RLHarness: Co-evolving Procedural Skills with Reinforcement Learning for Long-horizon Multimodal Reasoning
RLHarness：通过强化学习共同演化程序技能以实现长时间跨度的多模态推理
Abstract
Multimodal reasoning requires models to preserve visual evidence through long decision chains while selecting appropriate procedures across diverse scenarios and rules. When learning is guided only by terminal verifiers, reinforcement learning (RL) reveals whether a final answer is correct but not how it should be produced. The policy must therefore discover reusable reasoning procedures while learning to execute them, creating a program cold-start problem. Skills can externalize successful procedures, reduce repeated exploration, and provide inspectable guidance. However, a fixed Skill Bank assumes that this guidance remains compatible with an evolving policy, while updating Skills alone can leave their triggers, execution protocols, and demonstrations stale or mutually inconsistent. We introduce RLHARNESS, which organizes Skills, selection and execution protocols, few-shot demonstrations, and task contracts into a unified, versioned Harness and alternates Harness evolution with policy learning. An Exploration-Distillation Harness builds the initial Harness and version-aligned verified traces for SFT and DAPO I. After the first RL block, a Post-RL Reconstruction Harness rebuilds Skills, protocols, and demonstrations from fresh success-failure rollouts, and DAPO II adapts the policy to the reconstructed program. RLHARNESS improves Accuracy from 16.25%/27.50% to 62.00%/50.00% on MetroMap/TravelMap and raises F1 score from 37.13%/45.50% to 65.81%/65.51% on Fee-VL/Cancel-VL. All four tasks achieve their best results only after reconstruction and DAPO II, showing that an evolving Harness complements RL by continually updating the external program that the policy learns to execute.
Chinese Translation
多模态推理要求模型在长决策链中保留视觉证据，同时在多样的场景和规则中选择适当的程序。当学习仅由终端验证器引导时，强化学习（RL）能够揭示最终答案是否正确，但无法说明答案应如何产生。因此，策略必须在学习执行可重用的推理程序的同时发现这些程序，从而产生程序冷启动问题。技能可以外化成功的程序，减少重复探索，并提供可检查的指导。然而，固定的技能库假设这种指导与不断演变的策略保持兼容，而仅更新技能可能导致其触发条件、执行协议和示范变得过时或相互不一致。我们提出了RLHARNESS，它将技能、选择和执行协议、少量示范以及任务合同组织成一个统一的、版本化的Harness，并交替进行Harness演化与策略学习。探索-蒸馏Harness构建初始Harness和版本对齐的验证轨迹，以用于SFT和DAPO I。在第一个RL块之后，后RL重建Harness从新的成功-失败回滚中重建技能、协议和示范，而DAPO II则将策略适应于重建的程序。RLHARNESS在MetroMap/TravelMap上的准确率从16.25%/27.50%提高到62.00%/50.00%，在Fee-VL/Cancel-VL上的F1分数从37.13%/45.50%提高到65.81%/65.51%。所有四个任务仅在重建和DAPO II之后取得最佳结果，表明不断演变的Harness通过持续更新策略学习执行的外部程序来补充RL。
cs.AI / 54 / 2609.32327
HyperReCo: Retrieving and Connecting Evidence with Hypergraph Neural Networks for LLM Multi-hop Reasoning
HyperReCo：利用超图神经网络检索和连接证据以支持大型语言模型的多跳推理
Abstract
Large language models (LLMs) have shown strong capabilities, with retrieval-augmented generation (RAG) supporting complex multi-hop reasoning by retrieving evidence distributed across documents. Graph-based approaches exploit connections among evidence, and hypergraph-based retrieval further preserves higher-order entity associations within documents and connects documents through shared entities. However, existing hypergraph retrievers often rely on predefined structural expansion or diffusion, which may miss query-dependent interactions needed to identify relevant evidence. They also leave connections among retrieved evidence implicit, requiring LLMs to reconstruct these connections before reasoning. Therefore, we propose HyperReCo, a framework for retrieving and connecting evidence with a hypergraph neural network (HyperGNN). We represent each document as a hyperedge over its extracted entities, with shared entities connecting the hyperedges. Through hypergraph message passing with joint supervision over documents and entities, the HyperGNN learns query-dependent interactions to retrieve complementary evidence. We further introduce Gradient-Guided Hyper-Path Decoding (GGHD), which uses gradient attribution to interpret the learned interactions and translate them into explicit hyper-paths that help LLMs combine complementary facts for multi-hop reasoning. Experiments on six benchmarks show that HyperReCo achieves the best retrieval performance among the compared methods on all three multi-hop QA datasets, together with strong downstream QA performance. Case studies and further analyses demonstrate the utility of decoded hyper-paths for connecting retrieved evidence.
Chinese Translation
大型语言模型（LLMs）展现出强大的能力，通过检索分布在文档中的证据，增强生成（RAG）支持复杂的多跳推理。基于图的方式利用证据之间的连接，而基于超图的检索进一步保留了文档内高阶实体关联，并通过共享实体连接文档。然而，现有的超图检索器往往依赖于预定义的结构扩展或扩散，这可能会错过识别相关证据所需的查询依赖交互。同时，它们还将检索到的证据之间的连接隐含化，要求LLMs在推理之前重建这些连接。因此，我们提出了HyperReCo，一个利用超图神经网络（HyperGNN）检索和连接证据的框架。我们将每个文档表示为其提取实体的超边，具有共享实体连接这些超边。通过对文档和实体的联合监督进行超图消息传递，HyperGNN学习查询依赖的交互以检索互补证据。我们进一步引入了梯度引导超路径解码（GGHD），该方法利用梯度归因来解释学习到的交互，并将其转化为显式超路径，帮助LLMs结合互补事实进行多跳推理。在六个基准测试上的实验表明，HyperReCo在所有三个多跳问答数据集上实现了最佳的检索性能，并且在下游问答任务中表现强劲。案例研究和进一步分析展示了解码超路径在连接检索证据方面的实用性。
cs.AI / 55 / 2609.32339
Enabling Timely Guidance before Skill Retrieval: Retaining Helpful Warm Tips in Agent Context
在技能检索前提供及时指导：在代理上下文中保留有用的温馨提示
Abstract
Reusable skills help LLM-based agents solve complex tasks, but the agent must receive guidance before it commits to an ineffective approach. Existing skill mechanisms often expose only metadata and load full content on demand, leaving useful guidance unavailable until the agent decides to retrieve it. General memory methods can incur substantial maintenance overhead, while keeping guidance in conversation context risks repeatedly exposing the agent to irrelevant or harmful advice. We propose TipsWarm, a mechanism that complements existing skill mechanisms by maintaining a budgeted pool of skill-derived keypoints, or \textit{warm tips}, for selective injection into the context of every message turn. By separating event-triggered LLM assessment from inexpensive per-turn screening, it makes transferable skill guidance readily available while controlling maintenance costs. In three coding and iterative task-execution benchmarks, TipsWarm achieves the highest task success rate while remaining time-efficient, compared to recent skill and general memory baselines.
Chinese Translation
可重用技能帮助基于大型语言模型（LLM）的代理解决复杂任务，但代理必须在承诺无效方法之前获得指导。现有的技能机制通常仅暴露元数据，并在需要时加载完整内容，这使得有用的指导在代理决定检索之前无法获得。通用记忆方法可能会产生相当大的维护开销，而在对话上下文中保留指导则有可能使代理反复接触到无关或有害的建议。我们提出了TipsWarm，这是一种补充现有技能机制的机制，通过维护一个预算池，存储从技能中提取的关键点或称为 extit{温馨提示}，以便在每个消息轮次中选择性地注入上下文。通过将事件触发的LLM评估与每轮的低成本筛选分开，它使可转移的技能指导随时可用，同时控制维护成本。在三个编码和迭代任务执行基准测试中，与最近的技能和通用记忆基线相比，TipsWarm实现了最高的任务成功率，同时保持了时间效率。
cs.AI / 56 / 2609.32344
ALLOT: Budgeted Hybrid-Memory Routing for Knowledge Updates in LLMs
ALLOT：用于大语言模型知识更新的预算混合内存路由
Abstract
For large language models (LLMs), parametric adaptation is costly when retrieval already suffices. We introduce ALLOT, a hybrid-memory routing framework that separates learned write priority from a hard parametric budget. A memory-aware router combines frozen text representations, retrieval confidence, and relation metadata; a single ranking supports multiple write budgets while preserving all facts in external memory. On CounterFact with Qwen3-4B, ALLOT reaches 0.760 accuracy at a 20% parametric-write budget and recovers 78.4% of the budget-matched oracle gain, with 80% fewer parametric writes than dual-writing every fact. At this budget, jointly adding retrieval and relation features to text improves normalized oracle gain by 6.2 percentage points. Complementary Qwen3-0.6B shared-store results achieve dual-write-level accuracy with 6-14.5% parametric writes, and cross-benchmark transfer retains approximately 88% of in-domain gain. These results support allocating adaptation capacity according to its incremental value rather than treating every factual update as an equally valuable training target.
Chinese Translation
对于大语言模型（LLMs）而言，当检索已经足够时，参数适应的成本较高。我们提出了ALLOT，一种混合内存路由框架，它将学习的写入优先级与硬性参数预算分离。一个记忆感知的路由器结合了冻结的文本表示、检索置信度和关系元数据；单一的排名支持多个写入预算，同时保留外部内存中的所有事实。在CounterFact与Qwen3-4B的实验中，ALLOT在20%的参数写入预算下达到了0.760的准确率，并恢复了78.4%的预算匹配的oracle增益，相较于对每个事实进行双重写入，参数写入减少了80%。在该预算下，将检索和关系特征共同添加到文本中，使得标准化的oracle增益提高了6.2个百分点。互补的Qwen3-0.6B共享存储结果在6-14.5%的参数写入下达到了双重写入水平的准确率，并且跨基准迁移保留了约88%的领域内增益。这些结果支持根据适应能力的增量价值进行分配，而不是将每个事实更新视为同等有价值的训练目标。
cs.AI / 57 / 2609.32351
From Trajectories to Grounded Preferences: Process Preference Synthesis via Interaction Element Graphs for Web PRMs
从轨迹到基础偏好：通过交互元素图进行网络PRM的过程偏好合成
Abstract
Comparative Process Reward Models (PRMs) provide critical step-level guidance for autonomous web agents by evaluating state-conditioned preferences between candidate actions. However, existing preference training data synthesized via multi-policy sampling suffers from a severe scarcity of Grounded Minimal Contrastive Pairs (GMCPs)-where competing candidates target genuine on-page elements with identical action types. In representative baselines preference data (namely, WebArbiter), GMCPs account for merely 24.19%, biasing PRMs during training to rely on shallow shortcuts (such as element hallucinations and action type mismatches) rather than acquiring genuine contextual decision semantics. To address these challenges, we propose SURFPRM, a graph-guided process preference synthesis framework for comparative Web PRMs. SURFPRM structures web demonstrations into a persistent Interaction Element Graph that acts as an environment-grounded negative action proposal mechanism, systematically synthesizing contrastive negative actions across spatial, temporal, and spatiotemporal confusion axes. This elevates the GMCP proportion from 24.19% to 74.60%, producing the curated SURFPRM-DATA dataset. Across six open-source backbones (3B to 9B parameters), PRMs trained on SURFPRM-DATA outperform baseline-trained models on average on WEBPRMBENCH and rival leading proprietary LLMs. In downstream reward-guided trajectory search on WEBARENA-LITE, SURFPRM provides step-level guidance for both GPT-4o (+14.21%) and GPT-4o-mini (+12.83%) policies, yielding substantial improvements in complex web task success rates.
Chinese Translation
比较过程奖励模型（PRMs）通过评估候选动作之间的状态条件偏好，为自主网络代理提供关键的逐步指导。然而，现有通过多策略采样合成的偏好训练数据在基础最小对比对（GMCPs）方面严重匮乏——其中竞争候选者针对具有相同动作类型的真实页面元素。在代表性基线偏好数据（即WebArbiter）中，GMCPs仅占24.19%，导致PRMs在训练过程中依赖于浅层捷径（如元素幻觉和动作类型不匹配），而未能获取真实的上下文决策语义。为了解决这些挑战，我们提出了SURFPRM，一个用于比较网络PRMs的图引导过程偏好合成框架。SURFPRM将网络演示结构化为一个持久的交互元素图，该图作为一个环境基础的负动作提议机制，系统地在空间、时间和时空混淆轴上合成对比负动作。这将GMCP的比例从24.19%提升至74.60%，生成了精心策划的SURFPRM-DATA数据集。在六个开源骨干网络（3B到9B参数）上，基于SURFPRM-DATA训练的PRMs在WEBPRMBENCH上平均优于基线训练模型，并与领先的专有LLMs相抗衡。在WEBARENA-LITE上的下游奖励引导轨迹搜索中，SURFPRM为GPT-4o（+14.21%）和GPT-4o-mini（+12.83%）策略提供了逐步指导，显著提高了复杂网络任务的成功率。
cs.AI / 58 / 2609.32378
AuthorityLens: Rethinking LLM-Based Agent Systems Through the Lens of Authority
AuthorityLens：通过权威视角重新思考基于大型语言模型的代理系统
Abstract
LLM-based agents are increasingly deployed with authority over consequential resources and decisions in real systems. These agents often operate alongside human and LLM-based participants who hold different forms of authority. Yet workflow roles, permission settings, and review mechanisms do not necessarily reflect the authority realized in practice. We introduce AuthorityLens, a framework for measuring a system's authority structure. Starting from an authority portfolio, we evaluate a system along three dimensions: what the system is authorized to do (System Authority), how much joint participation is required to exercise that authority (Authority Separation), and how much authority each participant holds (Principal Authority). We derive these measurements from the minimal combinations of participants sufficient to realize each outcome across admissible runtime states. We apply AuthorityLens to Codex, OpenCode, and Gemini CLI across 13 operating configurations over a common portfolio of agent operations. We find that nominal configurations do not map cleanly onto realized authority. In Codex, Full Access changes System Authority only marginally while substantially concentrating authority in the executing Assistant. OpenCode's Build and Plan configurations have the same System Authority and Authority Separation despite different workflows and root-level permissions. In Gemini CLI, model-based review increases Authority Separation without changing System Authority. Principal Authority further distinguishes authority replication from authority separation: spawned or delegated agents can become alternative holders of the same authority without increasing the required joint participation. Together, these results demonstrate that AuthorityLens provides a unified framework for measuring and comparing realized authority structures across agent systems.
Chinese Translation
基于大型语言模型（LLM）的代理系统越来越多地被赋予对重要资源和决策的权威。这些代理系统通常与持有不同形式权威的人类和基于LLM的参与者共同运作。然而，工作流程角色、权限设置和审查机制并不一定反映实际中实现的权威。我们提出了AuthorityLens，这是一个用于测量系统权威结构的框架。从权威组合出发，我们从三个维度评估系统：系统被授权做什么（系统权威）、行使该权威所需的共同参与程度（权威分离）以及每个参与者所持有的权威程度（主要权威）。我们从实现每个结果所需的最小参与者组合中推导出这些测量，适用于可接受的运行时状态。我们将AuthorityLens应用于Codex、OpenCode和Gemini CLI，在一个共同的代理操作组合上进行了13种操作配置的评估。我们发现名义配置并未与实现的权威清晰对应。在Codex中，完全访问权限仅在边际上改变了系统权威，同时显著集中权威于执行助手。在OpenCode中，构建和计划配置具有相同的系统权威和权威分离，尽管工作流程和根级权限不同。在Gemini CLI中，基于模型的审查增加了权威分离，而没有改变系统权威。主要权威进一步区分了权威复制与权威分离：生成或委托的代理可以成为同一权威的替代持有者，而不增加所需的共同参与。综合这些结果表明，AuthorityLens提供了一个统一的框架，用于测量和比较代理系统中的实现权威结构。
cs.AI / 59 / 2609.32390
Reward Hacking and Agent Containment Failure: A Monte Carlo Study Based on the 2026 Hugging Face Incident
奖励黑客行为与代理隔离失败：基于2026年Hugging Face事件的蒙特卡洛研究
Abstract
The July 2026 intrusion into Hugging Face production infrastructure showed how reward hacking can become an external cybersecurity incident when a capable agent encounters weak containment boundaries. This study develops a probabilistic risk model linking five stages: reward hacking, containment escape, usable access, persistence, and failure of detection. A Monte Carlo simulation evaluates 100,000 runs under each of four control configurations. Input distributions represent explicit uncertainty and are used for comparative analysis rather than real-world frequency prediction. Under the stated assumptions, layered controls reduce simulated external-incident probability substantially more than network isolation or monitoring used alone, an ordering that holds under independent plus/minus 25% perturbation of every coefficient in the model across 300 draws. Sensitivity analysis shows that agent capability and weaknesses in monitoring, authorization, and credential control exert the greatest influence on modeled risk. Human temporal discounting and metric gaming provide a behavioral analogy for short-horizon optimization, but the study does not infer that AI agents experience gratification or human motivation. The results support treating cyber-capable agent evaluations as hostile security zones in which indirect egress, shared infrastructure, credentials, and evaluation artifacts must remain outside the agent's effective authority.
Chinese Translation
2026年7月对Hugging Face生产基础设施的入侵展示了当一个有能力的代理遇到薄弱的隔离边界时，奖励黑客行为如何演变为外部网络安全事件。本研究开发了一个概率风险模型，连接五个阶段：奖励黑客、隔离逃逸、可用访问、持久性和检测失败。通过蒙特卡洛模拟评估了在四种控制配置下的100,000次运行。输入分布代表明确的不确定性，用于比较分析而非现实世界频率预测。在所述假设下，分层控制显著降低了模拟外部事件的概率，远超单独使用网络隔离或监控的效果，这一排序在对模型中每个系数进行独立的正负25%扰动的300次抽样中保持一致。敏感性分析表明，代理能力以及监控、授权和凭证控制的弱点对模型风险的影响最大。人类的时间折现和指标游戏为短期优化提供了行为类比，但本研究并不推断AI代理会体验满足感或人类动机。结果支持将网络能力代理的评估视为敌对安全区，在这些区域中，间接出口、共享基础设施、凭证和评估文档必须保持在代理的有效权威之外。
cs.AI / 60 / 2609.32391
SCLATE: a Substrate for Continual-Learning Agent Training and Evaluation
SCLATE：持续学习智能体训练与评估的基础平台
Abstract
Continual-learning agents are systems of models, harnesses, and memory operating over long multi-session horizons. Evaluating and training them requires interleaving tasks with agent-side events such as session stop and start, crons, and memory consolidation. Yet existing benchmarks and training frameworks schedule only the benchmark's own events, leaving each benchmark and agent pair to build a custom scheduling loop. We present SCLATE, an execution substrate where benchmarks and unmodified agents each add their events to one open event scheduler through an adapter. A hybrid simulated clock runs these events on a shared timeline, flowing in real time while the agent works and skipping idle gaps, which compresses a month-long scenario into hours. SCLATE also serves as a rollout engine that runs any agent's harness and memory unmodified, recording the tokens and log probabilities of every model call through an in-container proxy. We port seven benchmarks to SCLATE and compare ten unmodified harness and memory configurations head to head on ten models. The comparison shows that an added memory system does not reliably beat the harness's native memory and that models differ widely in how they use the same harness and memory. We then post-train Qwen3.5-4B through unmodified harnesses and memory systems. The model learns to use both, reading 6.8x fewer file lines with a 16.7-point higher SWE-bench Verified pass rate, and writing richer memory records, while its held-out MetaClaw accuracy rises by up to 11.8 points.
Chinese Translation
持续学习智能体是一个由模型、工具和记忆组成的系统，能够在长时间的多会话范围内运行。评估和训练这些智能体需要将任务与智能体侧的事件（如会话的停止与开始、定时任务和记忆整合）交织在一起。然而，现有的基准测试和训练框架仅调度基准自身的事件，导致每个基准与智能体的组合都需要构建自定义的调度循环。我们提出了 SCLATE，一个执行基础平台，基准和未修改的智能体通过适配器将各自的事件添加到一个开放的事件调度器中。一个混合模拟时钟在共享时间线上运行这些事件，实时流动，同时智能体工作并跳过空闲间隙，从而将一个月的场景压缩到几小时内。SCLATE 还作为一个执行引擎，运行任何智能体的工具和记忆，未经过修改，通过容器内代理记录每个模型调用的令牌和日志概率。我们将七个基准移植到 SCLATE，并在十个模型上对十个未修改的工具和记忆配置进行逐一比较。比较结果表明，添加的记忆系统并不一定优于工具的本地记忆，并且模型在使用相同工具和记忆时差异显著。随后，我们通过未修改的工具和记忆系统对 Qwen3.5-4B 进行了后训练。该模型学会同时使用两者，读取的文件行数减少了 6.8 倍，SWE-bench 验证通过率提高了 16.7 个百分点，同时写入了更丰富的记忆记录，而其保留的 MetaClaw 准确率提高了最多 11.8 个百分点。
cs.AI / 61 / 2609.32394
Beyond Scripted Search: Sample-Efficient Reward Discovery via Agentic Black-box Optimization
超越脚本搜索：通过自主黑箱优化实现样本高效的奖励发现
Abstract
Designing dense reward functions for low-level reinforcement learning (RL) control remains difficult. Recent work uses large language models (LLMs) to iteratively generate and refine reward functions using policy-training feedback within scripted search algorithms. However, evaluating each candidate requires a full RL training run, making sample efficiency a central challenge for reward search on complex control tasks. To address this limitation, we propose an Agentic Reward Black-box Optimization (ARBO) framework, in which an LLM agent builds the search strategy at run time from an evaluation history maintained as its persistent workspace. The evaluation history comprises two components: observations maintained by the evaluation oracle, including candidate scores, per-term training curves, and error tracebacks; and an agent-maintained belief that records diagnoses and intended next steps. The agent queries both with tools and generates the next batch of reward candidates, rather than generating them in a single pass from a fixed prompt. Across four control domains, ARBO achieves gains of 29.9% in manipulation success rate and 192.8% in power-grid score over baseline means under a shared evaluation budget. Ablations examine each component's contribution and sensitivity to backbone choice.
Chinese Translation
为低级强化学习（RL）控制设计密集奖励函数仍然是一项挑战。近期的研究利用大型语言模型（LLMs）通过脚本搜索算法中的策略训练反馈迭代生成和优化奖励函数。然而，评估每个候选函数需要完整的RL训练过程，这使得在复杂控制任务上的奖励搜索面临样本效率的核心挑战。为了解决这一限制，我们提出了一种自主奖励黑箱优化（ARBO）框架，其中LLM代理在运行时根据维护的评估历史构建搜索策略。评估历史由两个部分组成：由评估oracle维护的观察，包括候选分数、每个术语的训练曲线和错误回溯；以及代理维护的信念，记录诊断和预期的下一步。代理使用工具查询这两部分，并生成下一批奖励候选，而不是从固定提示中一次性生成。通过四个控制领域的实验，ARBO在共享评估预算下实现了29.9%的操作成功率提升和192.8%的电网评分提升。消融实验考察了每个组件的贡献和对基础选择的敏感性。
cs.AI / 62 / 2609.32395
Memory as a cache: Exact context reuse and deletion by construction
记忆作为缓存：通过构造实现精确的上下文重用与删除
Abstract
The KV cache of a transformer entangles every token's representation with its entire prefix: a passage encoded once cannot be reused under a different prefix or removed without recomputing everything after it, so exact cache reuse is limited to shared prefixes. We present SMem, an architecture whose context representation is a cache by construction. A block-local encoder maps each block to memory rows independently of other blocks, and a reader conditions generation on their union through cross-attention. For every parameter setting, memory composes exactly at fixed block indices, deleting a block is an exact $O(b)$ update for $b$-token blocks, and the memory state is independent of the edit path. At $4\times$ the training context, under the shared recipe, SMem retrieves planted needles beyond any trained-length window (exact match 0.14-0.28 at distances of 31 and 63 blocks), where learned-position, RoPE, and Block-Attention-style transformers all score at most 0.02. A fully cached context is served by computing one block alone at a near-constant 3.1-6.2 ms, whereas cold prefill grows with context; batched decode stores 34-38% fewer KV rows and runs 1.4-1.7$\times$ faster when bandwidth-bound; and deletion beats suffix recomputation by 8.5$\times$ at 512 blocks and 452$\times$ at 4096 blocks (32-256$\times$ the trained length, probing the cost model rather than a served regime). The cost is a perplexity gap of -4.7% to +2.8% (negative favors SMem) against a parameter-matched transformer with the same positional scheme, at 160M-1.5B on FineWeb-Edu across two recipes and a learning-rate search. SMem also composes with RoPE: at 160M and 410M the composite matches or leads the matched transformer and closes 29-59% of SMem's gap to a RoPE transformer. Dropping prefix entanglement thus keeps perplexity comparable while making the cache exactly composable and editable.
Chinese Translation
变压器的KV缓存将每个标记的表示与其整个前缀纠缠在一起：一次编码的段落无法在不同的前缀下重用或在不重新计算其后所有内容的情况下删除，因此精确的缓存重用仅限于共享前缀。我们提出了SMem，一种其上下文表示通过构造成为缓存的架构。一个块局部编码器独立于其他块将每个块映射到内存行，而读取器通过交叉注意力对其并集进行条件生成。对于每个参数设置，内存在固定块索引处精确组合，删除一个块的更新复杂度为精确的$O(b)$，其中$b$为标记块的大小，内存状态与编辑路径无关。在训练上下文为4倍的情况下，在共享配方下，SMem能够检索超出任何训练长度窗口的植入针（在31和63块的距离下，精确匹配为0.14-0.28），而学习位置、RoPE和块注意力风格的变压器的得分最多为0.02。通过单独计算一个块，完全缓存的上下文以近乎恒定的3.1-6.2毫秒提供，而冷预填充则随着上下文增长；批量解码在带宽受限时存储34-38%的KV行更少，运行速度比单线程快1.4-1.7倍；在512块时，删除比后缀重新计算快8.5倍，在4096块时快452倍（32-256倍于训练长度，探测成本模型而非服务模式）。与具有相同位置方案的参数匹配变压器相比，成本为-4.7%到+2.8%的困惑度差（负值有利于SMem），在FineWeb-Edu上跨两个配方和学习率搜索，参数范围为160M-1.5B。SMem还与RoPE组合：在160M和410M时，复合模型与匹配的变压器相匹配或领先，并缩小了SMem与RoPE变压器之间29-59%的差距。因此，去除前缀纠缠使得困惑度保持相当，同时使缓存能够精确组合和编辑。
cs.AI / 63 / 2609.32398
Function Over Form: Distributional Orthogonalization in Mixture-of-Experts with Replica Expert Mechanism
形式优于功能：带有复制专家机制的混合专家中的分布正交化
Abstract
The scaling of LLMs increasingly relies on MoE architectures to decouple active computation from total parameter count. However, the efficacy of MoE is often constrained by expert collapse and representation redundancy, both leading to underutilization of model capacity. To address these challenges, this paper proposes Distributional Orthogonalization Loss (DO-loss), an auxiliary regularization that shifts the focus from static weight diversity to dynamic routing behavior. By representing each expert's token assignment history as a high-dimensional binary load signature, DO-loss penalizes signature overlap to prevent expert collapse while encouraging functional specialization. To align this algorithmic design with system efficiency, we further introduce the Replica Expert Mechanism (REM), which improves load balancing through a two-tiered strategy: adjusting replica expert placement at the global-batch level and performing real-time token dispatching at the micro-batch level. Empirical evaluations demonstrate that our method outperforms the evaluated routing algorithms on downstream tasks for both 4.8BA0.5B and 30BA3B MoE models, while maintaining comparable training efficiency.
Chinese Translation
大型语言模型（LLMs）的扩展越来越依赖于混合专家（MoE）架构，以将主动计算与总参数数量解耦。然而，MoE的有效性常常受到专家崩溃和表示冗余的限制，这两者都导致模型容量的未充分利用。为了解决这些挑战，本文提出了分布正交化损失（DO-loss），这是一种辅助正则化方法，旨在将重点从静态权重多样性转向动态路由行为。通过将每个专家的令牌分配历史表示为高维二进制负载签名，DO-loss惩罚签名重叠，以防止专家崩溃，同时鼓励功能专门化。为了将这一算法设计与系统效率对齐，我们进一步引入了复制专家机制（REM），该机制通过两级策略改善负载均衡：在全局批次级别调整复制专家的放置，并在微批次级别进行实时令牌调度。实证评估表明，我们的方法在下游任务中优于评估的路由算法，适用于4.8BA0.5B和30BA3B MoE模型，同时保持了可比的训练效率。
cs.AI / 64 / 2609.32407
Opening LLM Judges: Recovering Preference Signals Beyond the Final Verdict
开放的 LLM 法官：恢复超越最终裁决的偏好信号
Abstract
LLM judges are widely used to evaluate model outputs, but their verdicts can be unreliable: a judge may favor the worse answer for its position, length, or other surface features. When a judge is wrong, is the information needed to judge correctly absent from the model, or present in its internal representations but not reflected in the output? We study this across 64 open-weight evaluators and 14 datasets, including causal interventions on 41 judges (editing activations mid-run to see whether the verdict changes). On LLMBar, built so the superficially better answer is the worse one, the verdicts of 50 judges agree with human labels only 0.456 of the time, even after averaging both answer orders. Yet a small probe on the same judges' activations, with no weight updates, reaches 0.846, and 0.686 once surface features such as length and position are residualized out (0.507 with shuffled labels). The gap holds across eight benchmarks and model families, but is not universal: a score of how well surface features alone predict the human label, computed before any probe is trained, predicts the size of the gain (Spearman rho = 0.90). On rubric tasks that score one answer at a time, leaving no surface cue to exploit, reading the internals gives no advantage. The interventions also show that editing activations mid-network already changes the verdict, before it can be read off directly, and locate the pathways carrying position and length bias. At the same label budget, the recovered signal lets a judge flag cases where it is likely wrong and yields better labels for preference learning. A wrong verdict, then, does not mean the judge lacks the information, and a simple diagnostic shows when it is worth recovering.
Chinese Translation
LLM 法官被广泛用于评估模型输出，但它们的裁决可能不可靠：法官可能因为答案的位置、长度或其他表面特征而偏爱较差的答案。当法官判断错误时，是否缺乏正确判断所需的信息，还是该信息存在于模型的内部表示中但未反映在输出中？我们在 64 个开放权重评估器和 14 个数据集上进行了研究，包括对 41 个法官的因果干预（在运行过程中编辑激活以观察裁决是否改变）。在 LLMBar 上，表面上较好的答案实际上是较差的答案，50 个法官的裁决与人类标签一致的概率仅为 0.456，即使在对两个答案顺序进行平均之后。然而，对同一法官激活的小型探测，在没有权重更新的情况下，达到了 0.846，并且在去除长度和位置等表面特征后达到了 0.686（随机标签下为 0.507）。这一差距在八个基准和模型家族中保持一致，但并非普遍存在：在任何探测训练之前计算的表面特征单独预测人类标签的得分，能够预测增益的大小（Spearman rho = 0.90）。在评分一次一个答案的评分任务中，没有表面线索可供利用，读取内部信息并没有带来优势。干预还表明，在网络中期编辑激活已经改变了裁决，而不需要直接读取，并定位了携带位置和长度偏差的路径。在相同的标签预算下，恢复的信号使法官能够标记可能错误的案例，并为偏好学习提供更好的标签。因此，错误的裁决并不意味着法官缺乏信息，而一个简单的诊断可以表明何时值得恢复这些信息。
cs.AI / 65 / 2609.32420
Carnator: Fast Text-to-Video Generation with Generation-Native Compatibility-Guided Cross-Request Reuse
Carnator：具有生成原生兼容性引导的快速文本到视频生成的跨请求重用
Abstract
Video diffusion transformers produce high-quality videos, yet iterative denoising incurs substantial inference latency, limiting interactive and large-scale serving. Most existing acceleration methods focus on individual requests, thereby restricting efficiency gains to redundancy within a single generation trajectory. Recent cross-request reuse offers an additional source of savings, but existing approaches often infer reusability from coarse semantic similarity. This conflates semantic relatedness with generation-level computational compatibility, so aggressive reuse may accept incompatible historical computation while conservative reuse leaves substantial acceleration unrealized. We present \emph{Carnator}, a cross-request acceleration framework that addresses this challenge by extracting and using generation-native compatibility evidence directly from the model's evolving internal states. Specifically, \emph{Carnator} performs a lightweight early probe to construct an Early Signature from internal diffusion states, assessing reuse validity through risk-aware compatibility decisions. The same evidence characterizes reuse scope by localizing target-specific computation and guiding joint reuse of historical latent trajectories and sparse attention connectivity. Across three text-to-video backbones, Carnator consistently achieves higher cache-hit end-to-end acceleration than the evaluated cross-request baselines despite more selective cache acceptance, reaching up to 2.17$\times$ speedup while maintaining competitive generation quality.
Chinese Translation
视频扩散变换器能够生成高质量的视频，但迭代去噪会导致显著的推理延迟，从而限制了交互式和大规模服务。现有的大多数加速方法专注于单个请求，因此将效率提升限制在单一生成轨迹内的冗余。最近的跨请求重用提供了额外的节省来源，但现有方法通常通过粗略的语义相似性来推断重用性。这将语义相关性与生成级计算兼容性混为一谈，因此激进的重用可能接受不兼容的历史计算，而保守的重用则会导致大量加速未能实现。我们提出了 extit{Carnator}，一个跨请求加速框架，旨在通过直接从模型不断演变的内部状态中提取和使用生成原生兼容性证据来解决这一挑战。具体而言， extit{Carnator} 进行轻量级的早期探测，以从内部扩散状态构建早期签名，通过风险意识的兼容性决策评估重用的有效性。相同的证据通过定位目标特定计算并指导历史潜在轨迹和稀疏注意力连接的联合重用来表征重用范围。在三个文本到视频的基础模型中，尽管缓存接受更具选择性，Carnator 始终实现了比评估的跨请求基线更高的缓存命中端到端加速，速度提升高达 2.17$ imes$，同时保持竞争性的生成质量。
cs.AI / 66 / 2609.32422
MergeHEIR: Mitigating Multimodal Hallucinations as the Tax of Model Merging
MergeHEIR：减轻模型合并的多模态幻觉税
Abstract
Model merging consolidates task-specialized experts into a single deployable model. However, we show that such capability consolidation incurs a merging tax of increased hallucination: across 8 model-merging methods, every merged checkpoint exhibits a higher hallucination rate than the average of its constituent experts. An intuitive approach is to adapt existing hallucination-mitigation methods to the post-merge model, yet this unconstrained adaptation disrupts inherited capabilities, creating a tension between hallucination mitigation and expertise retention. To tackle this challenge, we introduce MergeHEIR, a post-merge adaptation framework designed to reduce this merging tax while preserving expertise inherited from initial experts. Using small expert-task calibration sets, MergeHEIR constructs layer-wise null-space projectors via SVD from task-specific activations collected from the merged checkpoint, and periodically projects the accumulated post-merge displacement onto the resulting null spaces to preserve inherited expertise. Theoretically, we establish minimum-distortion and maximum-dimensionality guarantees, characterize the threshold-controlled adaptation-retention trade-off, and extend perturbation guarantees beyond finite calibration data. Across 24 paired comparisons spanning three MLLM configurations and 8 model-merging methods, MergeHEIR consistently mitigates hallucination while largely preserving inherited expertise, demonstrating a more favorable hallucination-retention trade-off.
Chinese Translation
模型合并将任务专用的专家整合为一个可部署的单一模型。然而，我们表明，这种能力整合会导致增加的幻觉合并税：在8种模型合并方法中，每个合并的检查点的幻觉率都高于其组成专家的平均水平。一个直观的方法是将现有的幻觉减轻方法适应于合并后的模型，但这种不受限制的适应会破坏继承的能力，造成幻觉减轻与专业知识保留之间的紧张关系。为了解决这一挑战，我们提出了MergeHEIR，一个后合并适应框架，旨在减少这种合并税，同时保留来自初始专家的专业知识。通过使用小型专家任务校准集，MergeHEIR通过从合并检查点收集的任务特定激活构建层级的零空间投影器，并定期将累积的后合并位移投影到结果零空间中，以保留继承的专业知识。从理论上讲，我们建立了最小失真和最大维度的保证，描述了阈值控制的适应-保留权衡，并将扰动保证扩展到有限校准数据之外。在涵盖三种MLLM配置和8种模型合并方法的24对比较中，MergeHEIR始终减轻幻觉，同时大幅保留继承的专业知识，展示了更有利的幻觉-保留权衡。
cs.AI / 67 / 2609.32423
PluginRSI: Recursive Improvement of Agent Harnesses with Reusable Plugins
PluginRSI：可重用插件的递归改进代理工具
Abstract
The harness surrounding a language model is a central determinant of agent performance. Recent methods optimize harnesses by searching over complete programs, where individual mechanisms are difficult to isolate and reuse. We introduce PluginRSI, which represents a harness as a composition of atomized plugins and organizes harness evolution around these plugins. Individual plugins are improved independently and accumulated in a shared library, then recombined into new harnesses at each iteration. PluginRSI improves over existing harness optimization methods across software engineering, command-line interaction, and question-answering tasks. The resulting harnesses retain their advantage when transferred to other solver models without further optimization. The evolved plugin library accelerates subsequent optimization from the initial harness, which helps faster and higher convergence on unseen tasks. These results show that accumulating reusable mechanisms provides an effective basis for continued harness improvement.
Chinese Translation
围绕语言模型的工具是决定代理性能的核心因素。近期的方法通过搜索完整程序来优化工具，但在这些方法中，单个机制难以被孤立和重用。我们提出了PluginRSI，它将工具表示为原子化插件的组合，并围绕这些插件组织工具的演化。各个插件独立改进并累积在共享库中，然后在每次迭代中重新组合成新的工具。PluginRSI在软件工程、命令行交互和问答任务等领域优于现有的工具优化方法。所得到的工具在转移到其他求解模型时保持其优势，而无需进一步优化。演变后的插件库加速了从初始工具的后续优化，这有助于在未见任务上实现更快和更高的收敛。这些结果表明，累积可重用机制为持续改进工具提供了有效的基础。
cs.AI / 68 / 2609.32428
Authorization Closure Graph: Minimal Repair for LLM Agents with Evolving User Instructions
授权闭合图：针对具有不断演变用户指令的 LLM 代理的最小修复
Abstract
Tool-using large language model (LLM) agents increasingly perform state-changing actions that require user authorization. Yet existing approaches do not provide a principled mechanism for selectively updating prior authorization when only part of an instruction changes. To this end, we propose an Authorization-Closure-Graph (ACG)-based framework that represents authorization and its dependencies as an evolving, versioned state. ACG selectively invalidates authority affected by a revision while preserving unaffected portions of the authorization state, and computes a minimal repair that identifies only the missing evidence or authority required for execution. This enables agents to adapt to revised instructions while avoiding stale authority and unnecessary authorization requests. We evaluate ACG across three advanced LLMs in two natural tasks, and ACG consistently improves action safety rate and task success rate. Code is available at https://github.com/weiliang822/ACG.
Chinese Translation
使用工具的大型语言模型（LLM）代理越来越多地执行需要用户授权的状态改变操作。然而，现有的方法并未提供一个原则性机制，以在指令仅部分变化时选择性地更新先前的授权。为此，我们提出了一种基于授权闭合图（Authorization-Closure-Graph, ACG）的框架，该框架将授权及其依赖关系表示为一个不断演变的版本化状态。ACG 选择性地使受修订影响的授权失效，同时保留未受影响的授权状态部分，并计算出一个最小修复，识别出执行所需的缺失证据或授权。这使得代理能够适应修订后的指令，同时避免过时的授权和不必要的授权请求。我们在两个自然任务中对三种先进的 LLM 进行了 ACG 的评估，结果表明 ACG 一致性地提高了行动安全率和任务成功率。代码可在 https://github.com/weiliang822/ACG 获取。
cs.AI / 69 / 2609.32429
PrismQuant: Optimal Null-Space Rotations for Grouped Quantizers
PrismQuant：针对分组量化器的最优零空间旋转
Abstract
Smaller activation outliers do not necessarily imply better low-bit quantization: their alignment with the quantizer matters. We introduce PrismQuant, a quantizer-aware rotation framework that aligns the leading activation eigenspace with the constant group subspace of asymmetric grouped INT4. The affine offsets represent the energy in this subspace without widening the range within the group. We formulate rotation design as a Ky Fan trace maximization and derive a closed-form solution that is provably optimal for this alignment objective. Compact Householder transformations and their compact-WY representation enable gradient-free construction and efficient application at both foldable and online sites. A predictive range law further connects unaligned activation energy and group size to quantization-relevant variation. Experiments on Llama, Qwen, and Mistral span dense models up to 70B parameters and a 30B mixture-of-experts model. Under W4A4KV4, PrismQuant sets the state of the art on Llama-3.2-3B among the compared methods in both perplexity and accuracy. On Llama-3.1-70B, it attains 3.85 perplexity and 72.46% average zero-shot accuracy, only 0.22 percentage points below full precision. In the deployment study on Llama-3.1-8B, our optimized implementation achieves 1.51x prefill and 1.22x CUDA Graph decode speedups over matched FP16 baselines, with 56.34% lower decode peak memory and only 2.35% additional Graph decode latency over Hadamard. Code is available at https://github.com/ForeverBlue816/PrismQuant.
Chinese Translation
较小的激活异常值并不一定意味着更好的低比特量化：它们与量化器的对齐程度至关重要。我们提出了PrismQuant，一种量化器感知的旋转框架，旨在将主激活特征空间与不对称分组INT4的常数组子空间对齐。仿射偏移表示该子空间中的能量，而不会扩大组内的范围。我们将旋转设计形式化为Ky Fan迹最大化，并推导出一个闭式解，该解在这一对齐目标上是可证明的最优。紧凑的Householder变换及其紧凑-WY表示法使得在可折叠和在线站点上实现无梯度构造和高效应用成为可能。一个预测范围法则进一步将未对齐的激活能量和组大小与量化相关的变异联系起来。在Llama、Qwen和Mistral上进行的实验涵盖了高达700亿参数的稠密模型和一个300亿的专家混合模型。在W4A4KV4下，PrismQuant在比较方法中在困惑度和准确性上都在Llama-3.2-3B上设定了最新的技术标准。在Llama-3.1-70B上，它达到了3.85的困惑度和72.46%的平均零-shot准确率，仅比全精度低0.22个百分点。在Llama-3.1-8B的部署研究中，我们的优化实现比匹配的FP16基线实现了1.51倍的预填充和1.22倍的CUDA图解码加速，同时解码峰值内存降低了56.34%，相较于Hadamard仅增加了2.35%的图解码延迟。代码可在https://github.com/ForeverBlue816/PrismQuant获取。
cs.AI / 70 / 2609.32430
Multi-Agent System Search via Active Substructure-aware Policy Optimization
基于主动子结构感知的多智能体系统搜索策略优化
Abstract
LLMs enable multi-agent systems (MAS) to tackle complex tasks, but manually designing agent roles, prompts, and communication structures requires substantial expertise and effort. This motivates learning policies that construct query-specific MAS from execution reward. Existing approaches typically train these policies by repeatedly traversing a fixed set of training queries and assigning rewards at the workflow level. However, this overlooks differences in queries' evolving learning potential and obscures which substructures improve solution quality. In this paper, we propose Active Substructure-aware Policy Optimization (ASPO), a RL framework for query-level MAS search. ASPO introduces an Adaptive Query-Selection Mechanism (AQSM) that focuses training on queries at the policy's competence boundary: those it can solve but not yet reliably. A complementary discovery mechanism widens architectural exploration for hard queries, helping distinguish insufficient exploration from operator capability limits. Beyond query selection, ASPO introduces substructure-level rewards that measure output-quality gains within each action's descendant subgraph. These rewards guide proximal policy optimization to reinforce useful architectural refinements and discourage redundant or harmful computation. Together, these mechanisms prioritize learnable queries and provide fine-grained feedback for learning effective MASs. Across six benchmarks spanning mathematical reasoning, general question answering, and code generation, ASPO ranks first on every benchmark against twelve baselines.
Chinese Translation
大型语言模型（LLMs）使多智能体系统（MAS）能够处理复杂任务，但手动设计智能体角色、提示和通信结构需要大量专业知识和努力。这促使我们学习从执行奖励中构建查询特定的MAS策略。现有的方法通常通过反复遍历固定的训练查询集并在工作流层面分配奖励来训练这些策略。然而，这忽视了查询不断演变的学习潜力差异，并模糊了哪些子结构能够提高解决方案质量。本文提出了主动子结构感知策略优化（ASPO），这是一个用于查询级MAS搜索的强化学习框架。ASPO引入了一种自适应查询选择机制（AQSM），将训练重点放在策略能力边界上的查询上：即那些它能够解决但尚未可靠解决的查询。一个互补的发现机制扩大了对困难查询的架构探索，帮助区分不足的探索与操作能力限制。除了查询选择，ASPO还引入了子结构级奖励，衡量每个动作的后代子图中的输出质量提升。这些奖励指导近端策略优化，以强化有用的架构改进，并抑制冗余或有害的计算。通过这些机制，ASPO优先考虑可学习的查询，并为学习有效的MAS提供细粒度反馈。在涵盖数学推理、一般问答和代码生成的六个基准测试中，ASPO在所有基准测试中均排名第一，优于十二个基线模型。
cs.AI / 71 / 2609.32434
From Latents to Wires: Surgical Post-Editing on Large Language Models
从潜变量到线缆：大型语言模型的手术后编辑
Abstract
Given a large language model (LLM), can whoever holds the weights name a semantic target (e.g., the model's identity), locate the model components that produce it, and edit them so that the target no longer appears while other capability is preserved? We call such an edit on a trained model a post-edit. We present L2W (latents to wires), a framework that performs surgical post-edits for named semantic targets. For localization, L2W uses Jacobian lens (J-lens) attribution to score components against the semantic target. For surgical removal, because LLM mechanisms are redundant (i.e., a semantic target may have multiple components producing it), L2W runs Counterexample-Guided Causal Cut (CGCC) until the target no longer appears. CGCC first cumulatively closes model components, treating each surviving expression of the target as a counterexample that exposes the next components to close, and then reopens some of them to preserve capability. In a controlled experiment with an implanted behavioural watermark, L2W removes the watermark, and its localization lands on the model region the implant changed. Across three model configurations, L2W removes model-metadata (e.g., identity) self-claims in all nine runs, and adult-content refusal in all three, with no held-out target residual. L2W further composes two post-edits on a text-to-image model: one removes the refusal of requested nudity, and a second removes the nude rendering the first exposes. The results support post-editing as a complement to post-training: post-training installs preferred behaviours, and post-editing removes named unwanted ones.
Chinese Translation
给定一个大型语言模型（LLM），持有权重的任何人是否可以命名一个语义目标（例如，模型的身份），定位生成该目标的模型组件，并对其进行编辑，使得该目标不再出现，同时保留其他能力？我们将对训练模型的这种编辑称为后编辑。我们提出了L2W（从潜变量到线缆）框架，该框架对命名的语义目标执行手术后编辑。为了进行定位，L2W使用雅可比透镜（Jacobian lens，J-lens）归因方法对组件进行评分，以评估其与语义目标的关系。对于手术移除，由于LLM机制是冗余的（即一个语义目标可能由多个组件生成），L2W运行反例引导的因果切割（Counterexample-Guided Causal Cut，CGCC），直到目标不再出现。CGCC首先累积关闭模型组件，将每个存活的目标表达视为一个反例，从而揭示下一个需要关闭的组件，然后重新打开其中一些以保留能力。在一个控制实验中，L2W成功移除了植入的行为水印，并且其定位落在了植入改变的模型区域。在三种模型配置中，L2W在所有九次实验中移除了模型元数据（例如身份）的自我声明，在所有三次实验中移除了成人内容拒绝，且没有留下目标残余。L2W进一步在一个文本到图像模型上组合了两个后编辑：一个移除了对请求裸体的拒绝，另一个移除了第一个暴露的裸体渲染。结果支持后编辑作为后训练的补充：后训练安装了首选行为，而后编辑移除了命名的不希望出现的行为。
cs.AI / 72 / 2609.32436
Controllable GNN Explanations via Multi-Metric Preference Selection
通过多指标偏好选择实现可控的图神经网络解释
Abstract
Mechanisms for generating GNN explanations are crucial for building trust and mitigating biases in Graph Neural Networks (GNNs), especially in high-stakes scenarios. Most current methods optimize only for fidelity under the sparsity constraint. However, this discounts the need for interpretable explanations (those that consist of familiar motif patterns) and stable explanations (those that remain unchanged under structural perturbations). We propose a novel approach that optimizes GNN explanations across these metrics, exposing their relative weighing as a control. Experiments on various real-world datasets, including MUTAG, BA-2Motif, BAMultiShapes, and PROTEINS, suggest that our method produces higher-fidelity explanations than a state-of-the-art baseline on MUTAG and PROTEINS across all evaluated budgets, and on BA-2Motif at larger budgets, while being faster in the regime of small explanation budgets. We also explore how, given an input motif library containing standard motifs for the corresponding domain, the method can be used to determine the relative importance of those motifs in generating the explanations, and how this information can be used to further improve the quality of the output explanations. We also examine the relationship between different metrics through their induced tradeoff surface, and explore its dependence on the nature of the motif library.
Chinese Translation
生成图神经网络（GNN）解释的机制对于建立信任和减轻图神经网络中的偏见至关重要，尤其是在高风险场景中。目前大多数方法仅在稀疏性约束下优化准确性。然而，这忽视了对可解释性解释（由熟悉的模式组成）和稳定性解释（在结构扰动下保持不变的解释）的需求。我们提出了一种新颖的方法，旨在在这些指标上优化GNN解释，揭示其相对权重作为控制手段。在多个真实世界数据集上的实验，包括MUTAG、BA-2Motif、BAMultiShapes和PROTEINS，结果表明我们的方法在所有评估预算下，在MUTAG和PROTEINS上产生了比最先进基线更高的准确性解释，并且在较大预算下的BA-2Motif上也表现良好，同时在小解释预算的情况下速度更快。我们还探讨了如何利用包含相应领域标准模式的输入模式库，确定这些模式在生成解释中的相对重要性，以及如何利用这些信息进一步提高输出解释的质量。我们还通过其诱导的权衡面研究不同指标之间的关系，并探讨其对模式库性质的依赖性。
cs.AI / 73 / 2609.32448
ForkLeft: Entropy-First Rollouts for Prefix-Aligned Autoregressive-to-Diffusion Distillation
ForkLeft：基于熵优先的前缀对齐自回归到扩散蒸馏
Abstract
Autoregressive Next-Token Prediction (NTP) has enabled strong reasoning capabilities in language models, while Diffusion Language Models (DLMs) offer flexible token orders and parallel generation. We ask whether DLMs can acquire NTP-style reasoning through distillation without giving up their native generation process. Direct distillation, however, faces a fundamental mismatch: an autoregressive teacher predicts from a left prefix, whereas a DLM can condition on tokens on both sides. We introduce ForkLeft, a distillation framework that resolves this mismatch by separating the student's rollout from teacher supervision. During training, the student first performs entropy-first rollouts that commit uncertain positions and expose potential forks. We then fix the resulting student prefix and distill an NTP teacher under the same context, with answer correctness determining the supervision source. At inference, the student returns to its native confidence-first parallel decoding. With Qwen3-30B-A3B-Base, ForkLeft improves Efficient-DLM-4B on all ten benchmarks, raising MATH500 from 72.60% to 79.60% and consistently outperforming three alternative designs. The gains scale with teacher strength and generalize to SDAR-4B with only $500$ updates. At matched scale, the distilled 4B and 8B students exceed the published SDAR-Chat and OPDLM models on seven benchmarks, showing that DLMs can learn NTP-style reasoning without sacrificing native parallel generation. Code and datasets will be released upon acceptance.
Chinese Translation
自回归下一个标记预测（NTP）使语言模型具备了强大的推理能力，而扩散语言模型（DLMs）则提供了灵活的标记顺序和并行生成。我们探讨DLMs是否可以通过蒸馏获得NTP风格的推理能力，而不放弃其原生的生成过程。然而，直接蒸馏面临着根本的不匹配：自回归教师从左侧前缀进行预测，而DLM可以在两侧的标记上进行条件化。我们提出了ForkLeft，这是一种蒸馏框架，通过将学生的展开与教师的监督分离来解决这一不匹配。在训练过程中，学生首先执行熵优先的展开，确定不确定的位置并暴露潜在的分叉。然后，我们固定结果学生的前缀，并在相同的上下文中蒸馏NTP教师，答案的正确性决定了监督来源。在推理时，学生返回其原生的置信度优先并行解码。使用Qwen3-30B-A3B-Base，ForkLeft在所有十个基准测试中提升了Efficient-DLM-4B，将MATH500从72.60%提高到79.60%，并始终优于三种替代设计。增益与教师的强度成比例，并且在仅进行500次更新的情况下推广到SDAR-4B。在匹配规模下，蒸馏的4B和8B学生在七个基准测试中超越了已发布的SDAR-Chat和OPDLM模型，显示出DLMs可以在不牺牲原生并行生成的情况下学习NTP风格的推理。代码和数据集将在接受后发布。
cs.AI / 74 / 2609.32469
PULSE: Identifying Demonstration-Utility Features with Sparse Autoencoders
PULSE：利用稀疏自编码器识别演示效用特征
Abstract
In-context learning is highly sensitive to demonstration choice, yet most methods select demonstrations using external query-demonstration similarity. Such criteria can miss model-specific signals: Similar demonstrations may activate different internal features and downstream behaviors. We introduce PULSE (Paired Utility Localization over Sparse Encodings), an SAE-based framework for identifying model-internal features associated with demonstration utility and using them for demonstration selection. Using a small labeled discovery set, PULSE samples candidate demonstration sets, measures their zero-shot-relative utility under the target model, and scores SAE features by how their activation differences align with utility differences. The top positive and negative coordinates form a sparse utility-localization vector. We use this vector in two complementary ways: as a signed score for controlled complete-set ranking, and as PULSE-Retriever, which converts its magnitude into a feature-relevance mask for scalable pool-scale retrieval. Across classification, generation, and reasoning benchmarks, PULSE-Retriever improves over the strongest baseline by 2-3 accuracy points, 0.6-0.9 BLEU-4, and 3.2 exact-match points, respectively, while controlled ranking validates the identified features encode a predictive set-level utility signal. Feature inspection and cross-dataset experiments suggest that the identified features capture task-relevant, dataset-conditioned patterns, yet retain utility signals that partially transfer across datasets. Our code is available at https://github.com/aohenuo/PULSE.
Chinese Translation
上下文学习对演示选择高度敏感，但大多数方法使用外部查询-演示相似性来选择演示。这种标准可能会错过特定模型的信号：相似的演示可能激活不同的内部特征和下游行为。我们提出了PULSE（基于稀疏编码的配对效用定位），这是一个基于稀疏自编码器（SAE）的框架，用于识别与演示效用相关的模型内部特征，并利用这些特征进行演示选择。通过使用一个小型标记发现集，PULSE采样候选演示集，测量它们在目标模型下的零-shot相对效用，并根据其激活差异与效用差异的对齐程度对SAE特征进行评分。前几个正向和负向坐标形成一个稀疏效用定位向量。我们以两种互补的方式使用该向量：作为控制完整集排名的有符号分数，以及作为PULSE-Retriever，将其大小转换为可扩展池规模检索的特征相关性掩码。在分类、生成和推理基准测试中，PULSE-Retriever分别比最强基线提高了2-3个准确率点、0.6-0.9个BLEU-4分和3.2个精确匹配点，而控制排名验证了识别的特征编码了预测集级效用信号。特征检查和跨数据集实验表明，识别的特征捕捉了与任务相关的、数据集条件下的模式，但仍保留部分跨数据集转移的效用信号。我们的代码可在https://github.com/aohenuo/PULSE获取。
cs.AI / 75 / 2609.32473
VPEvolve: A Self-Evolving Virtual Process Engineer for Computational Lithography
VPEvolve：一种自我演化的虚拟工艺工程师用于计算光刻
Abstract
Optical proximity correction (OPC) recipes grow as engineers add local rules to repair newly discovered lithography hotspots. Each correction can interact with existing rules, while lessons from commercial-tool trials remain scattered across code and logs. \system combines a Virtual Process Engineer (VPE) harness with a Skill Bank of measured engineering experience. The harness equips a frozen language model with process manuals, layout analysis, recipe editing, and commercial-tool evaluation. The actor proposes changes to the global parameters, local targeted rules, or diagnostic trials. After each evaluation, an LLM reflector and curator turn the measured response into evidence-linked judgments. The actor retrieves them before its next trial. Feasible improvements update the retained recipe; every measured trial informs the Skill Bank that guides the next edit. The model weights remain fixed. On a FreePDK45-derived benchmark with ten commercial-tool evaluations per case, \system reduces the mean per-case maximum edge placement error from 18.294 to 5.361 nm on Poly and from 22.052 to 15.692 nm on Metal1. Every final recipe satisfies the predefined quality constraints and improves the maximum error by at least 0.1 nm.
Chinese Translation
光学邻近校正（OPC）配方随着工程师添加局部规则以修复新发现的光刻热点而不断增加。每个校正可能与现有规则相互作用，而来自商业工具试验的经验教训则散落在代码和日志中。系统结合了虚拟工艺工程师（VPE）工具和测量工程经验的技能库。该工具为冻结的语言模型提供了工艺手册、布局分析、配方编辑和商业工具评估。该参与者提出对全局参数、局部目标规则或诊断试验的更改。在每次评估后，LLM反射器和策展人将测量响应转化为证据关联的判断。参与者在下一次试验之前检索这些判断。可行的改进更新保留的配方；每次测量试验为指导下一次编辑的技能库提供信息。模型权重保持固定。在一个基于FreePDK45的基准测试中，每个案例进行了十次商业工具评估，系统将每个案例的平均最大边缘放置误差从18.294降低到5.361纳米（Poly），从22.052降低到15.692纳米（Metal1）。每个最终配方都满足预定义的质量约束，并将最大误差改善至少0.1纳米。
cs.AI / 76 / 2609.32482
From Outcomes to Strategies: Learning Strategy Utility for Mathematical Reasoning
从结果到策略：数学推理的策略效用学习
Abstract
Reinforcement learning with verifiable rewards has substantially improved mathematical reasoning. However, terminal correctness alone provides limited insight into the quality of high-level strategies, such as theorem selection and subgoal decomposition, when considered separately from their subsequent execution. This paper studies strategy utility, which is defined as the likelihood that a strategy supports a correct downstream solution under a given executor. We introduce SURE, a framework for learning and leveraging relative strategy utility. In this framework, high-level strategies are separated from their detailed reasoning. Based on the pairwise preferences constructed from strategy-conditioned rollouts and teacher-generated contrasts, a Strategy Reward Model is learned to estimate relative strategy utility. During reinforcement learning, the frozen reward model reads only the extracted strategy, whose score is combined with the correctness and format rewards in a sequence-level GRPO objective. Compared with outcome-and-format GRPO baselines, experiments show that SURE improves average pass@1 by 1.87%, 2.64%, and 2.93% across three policy backbones. Our method also achieves competitive or better accuracy than stronger reward baselines while requiring substantially lower GRPO-stage compute.
Chinese Translation
具有可验证奖励的强化学习显著提升了数学推理能力。然而，仅仅依赖最终的正确性在与后续执行分开考虑时，对高层策略（如定理选择和子目标分解）的质量提供的洞察有限。本文研究了策略效用，定义为在给定执行者下，某一策略支持正确下游解决方案的可能性。我们引入了SURE，一个用于学习和利用相对策略效用的框架。在该框架中，高层策略与其详细推理相分离。基于从策略条件下的回放和教师生成的对比中构建的成对偏好，学习了一个策略奖励模型以估计相对策略效用。在强化学习过程中，冻结的奖励模型仅读取提取的策略，其得分与正确性和格式奖励结合在序列级的GRPO目标中。与结果和格式的GRPO基线相比，实验表明SURE在三个策略骨干上平均提高了1.87%、2.64%和2.93%的通过率@1。我们的方法在准确性上也达到了与更强奖励基线相当或更好的水平，同时在GRPO阶段计算需求上显著降低。
cs.AI / 77 / 2609.32483
Separating Diagnosis from Disease Representation: Dual-View EEG Learning with Neural-Dynamics-Guided Deformation
将诊断与疾病表征分离：基于神经动力学引导变形的双视图脑电图学习
Abstract
Electroencephalography (EEG)-based closed-loop neuromodulation calls for a subject-specific structured state, as opposed to a single disease probability, specifying which brain regions are deviant, at which frequencies, and at which lags. Sensor-space models keep the strongest diagnostic evidence without anatomy, source-space models give anatomy at a loss of predictive signal, and post-hoc attributions stay outside the prediction. We separate the two instead of forcing them into one representation, and propose DMD-EEG (Dual-view Multiscale Deformation for EEG), which keeps a fixed scalp spectral expert for diagnosis and models the source-space disease-related representation as a low-rank, sparse, iterative deformation of a healthy neural-dynamics prior in a $46$-region-of-interest (ROI) $\times$ $5$-frequency $\times$ $4$-lag (autocorrelation-timescale) space. The two experts meet only at a fixed decision level, so the source state is architecturally separate from the scalp expert. Across major depressive disorder (MDD), first-episode psychosis (FEP), and Parkinson's disease (PD), decision-level fusion matches the strongest single expert on MDD and FEP and exceeds the source branch on PD. On FEP the source expert is the strongest branch, the task where the deformation contributes most. The source state is an explicit ROI-frequency-lag attribution defined in a shared source coordinate system across montages, which we treat as an anatomically-coordinated predictive representation whose coordinates are directly readable and hypothesis-generating. The highest-saliency coordinates align with established disease circuitry (fronto-limbic-temporal regions in MDD, motor-cortex beta in PD), and the MDD state transfers by rank to an unseen cohort recorded with a different montage.
Chinese Translation
基于脑电图（EEG）的闭环神经调制需要特定于个体的结构状态，而不是单一的疾病概率，这种状态能够指定哪些脑区存在异常，异常的频率和滞后时间。传感器空间模型在没有解剖信息的情况下保留了最强的诊断证据，而源空间模型则在牺牲预测信号的情况下提供了解剖信息，事后归因则无法纳入预测。我们选择将这两者分开，而不是强行将其合并为一个表征，并提出了DMD-EEG（双视图多尺度变形脑电图），该模型为诊断保留了固定的头皮谱专家，并将源空间的疾病相关表征建模为健康神经动力学先验的低秩、稀疏、迭代变形，适用于$46$个感兴趣区域（ROI）$ imes$ $5$个频率$ imes$ $4$个滞后（自相关时间尺度）空间。这两个专家仅在固定的决策层面相遇，因此源状态在结构上与头皮专家分离。在重度抑郁症（MDD）、首次发作精神病（FEP）和帕金森病（PD）中，决策层面的融合与MDD和FEP的最强单一专家相匹配，并在PD上超越了源分支。在FEP中，源专家是最强的分支，变形贡献最大。源状态是在跨多个电极配置的共享源坐标系统中定义的显式ROI-频率-滞后归因，我们将其视为解剖协调的预测表征，其坐标可直接读取并生成假设。最高显著性的坐标与已建立的疾病电路（MDD中的前额-边缘-颞区，PD中的运动皮层β波）相一致，且MDD状态通过秩转移到使用不同电极配置记录的未见队列。
cs.AI / 78 / 2609.32484
Towards Scalable Data Diversification for Language Model Pretraining via Leverage Score Sampling
通过杠杆分数采样实现可扩展的数据多样化以用于语言模型预训练
Abstract
Data selection for language model pretraining faces a fundamental tension between quality and diversity. While quality filtering is empirically effective, it often induces diversity collapse: by favoring texts similar to high-quality reference corpora (e.g., educational or QA-style data), it systematically excludes valuable data from underrepresented domains. In contrast, diversified selection preserves domain balance and encourages robust downstream performance, yet existing methods either focus on coverage-oriented objectives that indirectly enhance diversity, or directly optimize for diversity via costly covariance matrix recomputation that limits scalability. To address these issues, we introduce \textbf{Leverage Score Sampling (Lev)}, which iteratively selects samples that maximally expand the determinantal volume of the embedded data via leverage scores, a computationally efficient criterion that eliminates matrix recomputation and enables scalable selection. Empirically, Lev delivers up to $72\times$ speedup and improves dataset diversity, measured by the Vendi score, by $9.2\%$ over the strong diversification baseline \textbf{DiSF}. On CommonCrawl (CC) web data selection, Lev improves accuracy across seven downstream tasks by up to $1.31\%$ over existing baselines. For domains where robust quality criteria are inherently difficult to define (e.g., code), Lev serves as an effective unsupervised curation alternative: on StarCoderData, the selected subset reduces bits-per-byte by $3.08\%$ over DiSF. Notably, we uncover a cross-domain collapse of quality filtering: CC data filtered by DCLM-fastText fail to retain sufficient code-related content, yielding inferior code performance relative to Lev-selected data. These findings advocate for integrating diversity-aware practices into quality filtering for more effective data curation in language model pretraining.
Chinese Translation
语言模型预训练的数据选择面临质量与多样性之间的根本矛盾。虽然质量过滤在经验上是有效的，但它往往会导致多样性崩溃：通过偏向于与高质量参考语料库（例如教育或问答风格数据）相似的文本，它系统性地排除了来自代表性不足领域的有价值数据。相比之下，多样化选择保持领域平衡并促进稳健的下游性能，然而现有方法要么专注于间接增强多样性的覆盖目标，要么通过耗时的协方差矩阵重新计算直接优化多样性，这限制了可扩展性。为了解决这些问题，我们引入了 extbf{杠杆分数采样（Leverage Score Sampling, Lev）}，该方法通过杠杆分数迭代选择最大扩展嵌入数据的行列式体积的样本，这是一种计算上高效的标准，消除了矩阵重新计算并实现了可扩展选择。实证结果表明，Lev提供了高达$72 imes$的加速，并且在强多样化基线 extbf{DiSF}之上，数据集多样性（通过Vendi分数衡量）提高了$9.2\%$。在CommonCrawl (CC)网页数据选择中，Lev在七个下游任务上的准确性比现有基线提高了最高$1.31\\%$。对于那些固有上难以定义稳健质量标准的领域（例如代码），Lev作为一种有效的无监督策展替代方案：在StarCoderData上，所选子集比DiSF减少了$3.08\\%$的每字节位数。值得注意的是，我们发现了质量过滤的跨领域崩溃：由DCLM-fastText过滤的CC数据未能保留足够的与代码相关的内容，相较于Lev选择的数据，导致代码性能较差。这些发现倡导在质量过滤中整合关注多样性的实践，以实现更有效的语言模型预训练数据策展。
cs.AI / 79 / 2609.32489
When Helpful Text Hurts: Option-Redirecting Bias in Vision-Language Models
有益文本如何造成伤害：视觉-语言模型中的选项重定向偏差
Abstract
In tri-modal visual question answering (VQA), auxiliary text is commonly used to complement visual and textual inputs, yet its reliability is often uncontrolled. While prior work studies modality conflicts in general, it remains unclear how different types of unreliable auxiliary text affect answer selection under fixed image-question-option contexts. In this work, we show that the most harmful auxiliary text is not necessarily the most factually incorrect, but the one that aligns with the question while contradicting the image and favoring a specific distractor, leading to systematic redirection of model predictions. To isolate this effect, we introduce the Textual Reliability Ladder, a controlled diagnostic protocol that decomposes auxiliary text along three axes: image consistency, question relevance, and option support. Across multiple datasets (ScienceQA, VCR, A-OKVQA, Causal-VidQA) and recent VLMs, we find that such distractor-supporting text induces the largest accuracy drops (up to 53.1%) and concentrates errors on specific incorrect options. To mitigate this failure mode, we propose a training-free inference-time intervention that explicitly counteracts this redirection effect via noise-stability steering and dynamic grounding, reducing redirected errors while largely preserving performance under faithful text. Our results highlight that auxiliary-text reliability must be understood at the decision level, rather than solely through factual correctness, and provide a practical pathway toward more robust tri-modal reasoning.
Chinese Translation
在三模态视觉问答（VQA）中，辅助文本通常用于补充视觉和文本输入，但其可靠性往往无法控制。虽然之前的研究一般探讨模态冲突，但不同类型的不可靠辅助文本如何在固定的图像-问题-选项上下文中影响答案选择仍不清楚。在本研究中，我们表明，最有害的辅助文本不一定是最不真实的，而是与问题一致但与图像相矛盾并偏向特定干扰项的文本，这导致模型预测的系统性重定向。为了隔离这一效应，我们引入了文本可靠性阶梯（Textual Reliability Ladder），这是一种受控的诊断协议，通过三个维度对辅助文本进行分解：图像一致性、问题相关性和选项支持。在多个数据集（ScienceQA、VCR、A-OKVQA、Causal-VidQA）和近期的视觉语言模型（VLMs）中，我们发现这种支持干扰项的文本会导致最大的准确率下降（高达53.1%），并将错误集中在特定的不正确选项上。为了缓解这一失败模式，我们提出了一种无训练的推理时干预方法，通过噪声稳定引导和动态基础，明确抵消这种重定向效应，减少重定向错误，同时在真实文本下大幅保持性能。我们的结果强调，辅助文本的可靠性必须在决策层面上理解，而不仅仅通过事实正确性，并为更稳健的三模态推理提供了实际路径。
cs.AI / 80 / 2609.32490
RepoMAS: Solving Progressively Specified Tasks with Issue-Driven Multi-Agent Systems
RepoMAS：通过问题驱动的多代理系统解决逐步指定的任务
Abstract
LLM-based multi-agent systems (MASs) have shown strong potential for solving complex tasks, but most assume that task requirements are sufficiently specified before execution. In practice, user requests are often incomplete, and additional requirements may only become clear during reasoning, tool use, or execution. We refer to such problems as progressively specified tasks. To systematically study this setting, we introduce ProgSpec, a benchmark that evaluates final outputs against requirements explicitly stated in the initial request and additional requirements supported by the available task evidence. We further propose RepoMAS, an issue-driven multi-agent framework inspired by open-source project management. RepoMAS records newly discovered requirements, conflicts, and failures as structured Issues and uses them to revise the task specification and execution structure during problem solving. Across ProgSpec and five existing benchmarks, RepoMAS achieves the best performance. Further analyses show that its issue-driven revision and repository maintenance mechanisms consistently contribute to performance. These results highlight the importance of allowing MASs to revise not only how a task is solved, but also revise their explicit representation of task requirements during execution.
Chinese Translation
基于大语言模型的多代理系统（MASs）在解决复杂任务方面展现出强大的潜力，但大多数系统假设任务需求在执行前已被充分指定。在实际应用中，用户请求往往不完整，额外的需求可能仅在推理、工具使用或执行过程中变得清晰。我们将这些问题称为逐步指定的任务。为了系统性地研究这一设置，我们引入了ProgSpec，一个基准测试，用于评估最终输出与初始请求中明确陈述的需求以及由可用任务证据支持的额外需求之间的符合程度。我们进一步提出了RepoMAS，一个受开源项目管理启发的问题驱动多代理框架。RepoMAS将新发现的需求、冲突和失败记录为结构化问题，并利用这些信息在问题解决过程中修订任务规范和执行结构。在ProgSpec和五个现有基准测试中，RepoMAS表现出最佳性能。进一步的分析表明，其问题驱动的修订和仓库维护机制始终对性能产生积极贡献。这些结果强调了允许多代理系统在执行过程中不仅修订任务解决方案的方式，还修订任务需求的明确表述的重要性。
cs.AI / 81 / 2609.32492
Beyond Prompt or Skill? Attribution-Guided Optimization of Modular LLM Programs
超越提示或技能？基于归因的模块化大语言模型程序优化
Abstract
Large language models can solve increasingly diverse reasoning tasks, yet their performance remains highly sensitive to task prompts, intermediate instructions, and the way reusable problem-solving knowledge is incorporated. Existing optimization methods usually focus on only one part of this design space: they either optimize a monolithic prompt, or separately induce and refine skills from model traces. As a result, they lack a principled mechanism for deciding which component should be updated when failures occur, and they rarely optimize prompts, skills, and skill-use policies in a unified framework. We propose SPARO (Skill, Prompt, And Routing Optimization), a framework that jointly optimizes task instructions, reusable skill blocks, and routing rules. It performs controlled counterfactual evaluations, converts examples' effects into a probabilistic responsibility distribution over prompt, skill, and routing components, samples one component from that distribution, and applies the corresponding targeted mutation. This design moves language-program optimization beyond global prompt rewriting toward structured, reusable, and selectively activated task knowledge. Across five benchmarks and five worker models, SPARO consistently outperforms both prompt-centered and skill-centered optimization baselines. These results suggest that effective language-program optimization depends not only on discovering useful task knowledge, but also on deciding where that knowledge should be stored and when it should be activated.
Chinese Translation
大型语言模型能够解决越来越多样化的推理任务，但其性能仍然对任务提示、中间指令以及可重用问题解决知识的整合方式高度敏感。现有的优化方法通常只关注这一设计空间的某一部分：它们要么优化一个整体提示，要么分别从模型轨迹中诱导和精炼技能。因此，它们缺乏一个原则性机制来决定在出现失败时应更新哪个组件，并且很少在统一框架中优化提示、技能和技能使用策略。我们提出了SPARO（技能、提示和路由优化），这是一个联合优化任务指令、可重用技能模块和路由规则的框架。它执行受控的反事实评估，将示例的影响转换为对提示、技能和路由组件的概率责任分布，从该分布中抽样一个组件，并应用相应的目标变异。这一设计将语言程序优化从全局提示重写推进到结构化、可重用和选择性激活的任务知识。在五个基准和五个工作模型中，SPARO始终优于以提示为中心和以技能为中心的优化基线。这些结果表明，有效的语言程序优化不仅依赖于发现有用的任务知识，还依赖于决定这些知识应存储在哪里以及何时激活。
cs.AI / 82 / 2609.32498
DAAF: From Failure Localization to Editable System Assets in LLM Agents
DAAF：从故障定位到可编辑系统资产的 LLM 代理
Abstract
Deployed LLM agents increasingly rely on persistent, versioned system assets such as routing rules, knowledge segments, prompt instructions, and reusable skills. Failure-localization methods can identify where an error manifests in an agent or execution trace, but repair requires a different decision: which editable system asset should be changed, and is that change expected to improve the task outcome? We study this gap through component-attribute failure attribution, where diagnosis targets versioned, addressable items rather than execution locations. We propose the Detection-Aware Attribution Framework (DAAF), which learns the effects of valid attribute replacements and amortizes this intervention evidence into deployment-time diagnosis. DAAF combines sparse and noisy failure signals to decide whether intervention is warranted, learns component-type-conditioned replacement effects from controlled replays evaluated by executable task outcomes, and shares supervision across requests with compatible intervention responses. At diagnosis time, DAAF uses only the observed execution, registered candidates, and available failure signals; it requires neither counterfactual replay nor task reward and returns no_change, a repair target, or an unresolved decision when evidence is insufficient. On held-out tau^2-bench Telecom tasks, DAAF achieves 80.72% attribute Hit@1, recovers 62.65% of failed executions while limiting clean-task regression to 3.23%, and reaches 71.93% overall task success. These results show that intervention-grounded attribute attribution can connect failure localization to executable system repair.
Chinese Translation
部署的 LLM 代理越来越依赖于持久的、版本化的系统资产，如路由规则、知识片段、提示指令和可重用技能。故障定位方法可以识别代理或执行轨迹中错误的表现位置，但修复需要做出不同的决策：应该更改哪个可编辑的系统资产，并且该更改是否预计能改善任务结果？我们通过组件属性故障归因研究这一差距，其中诊断目标是版本化的、可寻址的项目，而不是执行位置。我们提出了检测感知归因框架（Detection-Aware Attribution Framework, DAAF），该框架学习有效属性替换的影响，并将这种干预证据在部署时的诊断中进行摊销。DAAF 结合稀疏和嘈杂的故障信号来决定是否需要干预，从可执行任务结果评估的受控重放中学习组件类型条件下的替换效果，并在请求之间共享具有兼容干预响应的监督。在诊断时，DAAF 仅使用观察到的执行、注册候选项和可用的故障信号；它既不需要反事实重放，也不需要任务奖励，当证据不足时返回 no_change、修复目标或未解决的决策。在保留的 tau^2-bench 电信任务中，DAAF 实现了 80.72% 的属性 Hit@1，恢复了 62.65% 的失败执行，同时将干净任务的回归限制在 3.23%，并达到了 71.93% 的整体任务成功率。这些结果表明，基于干预的属性归因可以将故障定位与可执行系统修复连接起来。
cs.AI / 83 / 2609.32502
TreeRef-BFN: Equivariance-Free De Novo Molecule Generation based on 2D Topology and Internal 3D Geometry
TreeRef-BFN：基于二维拓扑和内部三维几何的无等变性新型分子生成
Abstract
De novo 3D molecular generation jointly models molecular size, topology, and geometry. Most methods pre-sample molecular size and generate Cartesian coordinates, limiting variable-size conditional tasks such as fragment completion and scaffold decoration while often relying on equivariant architectures. Internal-coordinate methods avoid rigid-body redundancy but typically require a known molecular graph or autoregressive construction, which may accumulate errors. We propose TreeRef, a tree-based molecular representation that assigns molecular topology and topology-dependent local 3D geometry to a naturally variable-size tree. RingRef nodes encode ring closures while preserving the tree structure, while Null nodes allow molecular size to emerge directly from node occupancy. Based on TreeRef, we develop TreeRef-BFN, a Bayesian Flow Network with a standard Transformer backbone that globally couples these locally defined variables and jointly generates discrete molecular variables and continuous local geometry. A single pretrained TreeRef-BFN supports unconditional generation and variable-size structure-conditioned 3D generation through masking alone, without retraining. Empirical studies demonstrate strong chemical validity, molecular stability, and diversity, accurate local geometric distributions, fast sampling, and competitive property-conditioned generation, establishing TreeRef-BFN as an efficient and flexible framework for 3D molecular generation.
Chinese Translation
新型三维分子生成联合建模分子大小、拓扑和几何。大多数方法预先采样分子大小并生成笛卡尔坐标，这限制了可变大小的条件任务，如片段补全和支架装饰，同时通常依赖于等变架构。内部坐标方法避免了刚体冗余，但通常需要已知的分子图或自回归构造，这可能会累积误差。我们提出了TreeRef，一种基于树的分子表示，它将分子拓扑和依赖拓扑的局部三维几何分配给自然可变大小的树。RingRef节点编码环闭合，同时保持树结构，而Null节点允许分子大小直接从节点占用中产生。基于TreeRef，我们开发了TreeRef-BFN，一种具有标准Transformer骨干的贝叶斯流网络，它全局耦合这些局部定义的变量，并共同生成离散分子变量和连续局部几何。一个经过预训练的TreeRef-BFN支持无条件生成和可变大小结构条件的三维生成，仅通过掩蔽实现，无需重新训练。实证研究表明其具有良好的化学有效性、分子稳定性和多样性，准确的局部几何分布，快速采样，以及具有竞争力的属性条件生成，确立了TreeRef-BFN作为三维分子生成的高效灵活框架。
cs.AI / 84 / 2609.32511
Learning from Others, Acting for You: Cross-User Memory Sharing for LLM Agents
向他人学习，为您行动：跨用户记忆共享用于大型语言模型代理
Abstract
Large language model (LLM) agents serving different users often solve related tasks, yet separate user histories can leave reusable experience inaccessible to other agents. Pooling memories expands access but risks transferring preferences that conflict with the receiving user's requirements. We introduce ShareMem, a memory architecture that shares reusable experience while grounding its application in the receiving user's own preferences. Shared experiences indicate how to act and which preferences to consult; the receiving user's memory supplies their concrete values. Two-stage consolidation refines experience locally before integrating accepted edits into a shared pool. During execution, scope-first retrieval jointly selects local and shared experiences under a common entry budget, while a user-bound channel supports initial and agent-initiated preference retrieval. We evaluate ShareMem across web navigation (Mind2Web), online personalized interaction (VitaBench~2.0), and multi-session coding (MemoryCode) with four backbone models. It improves step success, average task success, and dialogue-macro coding scores, respectively, over matched user-local memory across all four models. Ablations favor two-stage consolidation for smaller shared pools, lower induction token usage, and better downstream performance, and support complementarity between experience guidance and active preference retrieval. Further analyses show that sharing helps most when relevant local experience is scarce, while source quality and cross-user preference interference limit useful transfer.
Chinese Translation
为不同用户服务的大型语言模型（LLM）代理通常会解决相关任务，但分开的用户历史可能使得可重用的经验对其他代理不可访问。汇聚记忆可以扩大访问范围，但也存在将与接收用户需求相冲突的偏好转移的风险。我们提出了ShareMem，一种记忆架构，能够共享可重用的经验，同时将其应用基础扎根于接收用户自身的偏好。共享经验指示如何行动以及应咨询哪些偏好；接收用户的记忆提供其具体的价值观。两阶段整合在将接受的编辑整合到共享池之前，先在本地精炼经验。在执行过程中，范围优先检索在共同的入口预算下共同选择本地和共享经验，而用户绑定通道支持初始和代理发起的偏好检索。我们在网页导航（Mind2Web）、在线个性化互动（VitaBench~2.0）和多会话编码（MemoryCode）中评估了ShareMem，使用了四个基础模型。与匹配的用户本地记忆相比，它在所有四个模型中分别提高了步骤成功率、平均任务成功率和对话宏编码分数。消融实验表明，对于较小的共享池，两阶段整合更具优势，降低了引导令牌的使用，并改善了下游性能，同时支持经验指导与主动偏好检索之间的互补性。进一步分析表明，当相关的本地经验稀缺时，分享的帮助最大，而源质量和跨用户偏好干扰则限制了有用的转移。
cs.AI / 85 / 2609.32514
From Anomalies to Failures: Constructing Causal Error Graphs for Agentic Trace Diagnosis
从异常到故障：构建用于代理追踪诊断的因果错误图
Abstract
LLM-driven agents are increasingly deployed in complex applications, where long agentic traces make failures difficult to diagnose. Existing trace diagnosis methods often conflate anomalies, errors, and failures, making diagnostic targets ambiguous; they also lack structured modeling of how causally relevant errors propagate and amplify into final task failures, resulting in unreliable failure attribution. To address these problems, we propose CEG-Agent, a tool-augmented agentic framework for causal diagnosis of agentic traces. Specifically, CEG-Agent introduces an explicit taxonomy of anomalies, errors, and failures, and constructs Causal Error Graphs (CEGs), a unified typed representation that links execution events, diagnostic nodes, and failure outcomes through causal relations. To evaluate causal trace diagnosis, we further construct CEG-Bench, a fully agent-annotated benchmark with high-confidence, consensus-derived CEG annotations obtained through an Adversarial Agentic Adjudication Protocol (AAAP). We validate the resulting annotations against an expert-curated human gold set, which shows close agreement with the automatic annotations. Experiments on CEG-Bench demonstrate that CEG-Agent achieves state-of-the-art performance under both semantically relaxed and structurally exact evaluation criteria. Our code is publicly available.
Chinese Translation
基于大型语言模型（LLM）的代理在复杂应用中越来越多地被部署，其中较长的代理追踪使得故障诊断变得困难。现有的追踪诊断方法往往将异常、错误和故障混为一谈，使得诊断目标模糊不清；此外，它们缺乏对因果相关错误如何传播和放大至最终任务故障的结构化建模，导致故障归因不可靠。为了解决这些问题，我们提出了CEG-Agent，一个增强工具的代理框架，用于代理追踪的因果诊断。具体而言，CEG-Agent引入了异常、错误和故障的明确分类法，并构建了因果错误图（Causal Error Graphs, CEGs），这是一种统一的类型化表示，通过因果关系将执行事件、诊断节点和故障结果联系起来。为了评估因果追踪诊断，我们进一步构建了CEG-Bench，这是一个完全由代理注释的基准，具有高置信度、共识导出的CEG注释，这些注释是通过对抗性代理裁决协议（Adversarial Agentic Adjudication Protocol, AAAP）获得的。我们将结果注释与专家策划的人类金标准集进行了验证，显示出与自动注释的高度一致性。在CEG-Bench上的实验表明，CEG-Agent在语义放宽和结构精确的评估标准下均实现了最先进的性能。我们的代码已公开发布。
cs.AI / 86 / 2609.32517
LocalProp: Neuro-Localized Memory-Efficient Backpropagation
LocalProp：神经局部化内存高效反向传播
Abstract
The current deep learning training paradigm employs end-to-end backpropagation, regardless of the training stage, i.e. pre-training or fine-tuning. However, backpropagating through the entire model is neither biologically plausible nor memory efficient, since learning inside the brain is highly localized. Therefore, we propose LocalProp, a training procedure that locally updates the weights of a model. Our neuro-localized weight updates follow the "pre-training then fine-tuning" paradigm, where the pre-training is based on I-JEPA. After locally updating the weights, a pruning operation is performed, followed by a short final fine-tuning phase. Pruning helps by sending the learning signal from higher blocks to lower blocks. We perform experiments on several datasets, including large-scale benchmarks such as ImageNet, and empirically show that LocalProp reaches good performance at a fraction of GPU peak memory. By varying the number of jointly optimized blocks, we identify gradient-propagation span as a practical control over the accuracy-memory trade-off.
Chinese Translation
当前的深度学习训练范式采用端到端的反向传播，无论是在预训练阶段还是微调阶段。然而，反向传播整个模型既不符合生物学原理，也不具备内存效率，因为大脑内部的学习是高度局部化的。因此，我们提出了LocalProp，一种局部更新模型权重的训练过程。我们的神经局部化权重更新遵循“先预训练再微调”的范式，其中预训练基于I-JEPA。在局部更新权重后，进行剪枝操作，随后进入短暂的最终微调阶段。剪枝通过将学习信号从高层模块传递到低层模块来提供帮助。我们在多个数据集上进行了实验，包括ImageNet等大规模基准测试，实证表明LocalProp在GPU峰值内存的一小部分下达到了良好的性能。通过改变联合优化的模块数量，我们识别出梯度传播范围作为控制准确性与内存权衡的实用手段。
cs.AI / 87 / 2609.32519
STR: Supervised Transcoder Replacement for Reducing Steering Side Effects
STR：用于减少引导副作用的监督转码器替换
Abstract
Model steering can strengthen a target behavior while degrading other useful behaviors. We introduce Supervised Transcoder Replacement (STR) to reduce these side effects for existing steering methods, including those fitted without a protection objective. STR learns a replacement for the multilayer perceptron (MLP) computation at the steering layer through supervision for target control, non-target preservation, and fidelity without steering. Selected steering methods then fit directions on the frozen replacement while retaining their own fitting objectives. We evaluate three steering methods across Gemma and Llama models using Corrigibility preferences and four harmful-request safety datasets. SALAD-Bench supplies protection training data and a separate in-distribution evaluation split; HarmBench, AdvBench, and StrongREJECT are reserved for out-of-distribution testing. STR substantially reduces steering side effects on the in-distribution evaluation and extends this protection to the unseen safety datasets while retaining effective target control. For target-only supervised steering vectors, pooled out-of-distribution attack success rate falls from 42.46% to 14.42% on Gemma-3-4B and from 34.97% to 12.91% on Gemma-3-12B. These results show that replacement training can benefit steering methods fitted without protection objectives.
Chinese Translation
模型引导可以增强目标行为，同时削弱其他有用行为。我们引入了监督转码器替换（Supervised Transcoder Replacement, STR），以减少现有引导方法的这些副作用，包括那些没有保护目标的拟合方法。STR通过对目标控制、非目标保留和无引导保真度的监督，学习在引导层替代多层感知器（Multilayer Perceptron, MLP）计算的替代方案。选定的引导方法在冻结的替代方案上拟合方向，同时保留自己的拟合目标。我们在Gemma和Llama模型上评估了三种引导方法，使用了Corrigibility偏好和四个有害请求安全数据集。SALAD-Bench提供了保护训练数据和一个单独的分布内评估拆分；HarmBench、AdvBench和StrongREJECT则保留用于分布外测试。STR在分布内评估中显著减少了引导副作用，并将这种保护扩展到未见的安全数据集，同时保持有效的目标控制。对于仅针对目标的监督引导向量，Gemma-3-4B上的分布外攻击成功率从42.46%降至14.42%，Gemma-3-12B上的成功率从34.97%降至12.91%。这些结果表明，替代训练可以使没有保护目标的引导方法受益。
cs.AI / 88 / 2609.32521
MemAgent: Learning to Manage Heterogeneous Memory Providers for LLM Agents
MemAgent：学习管理异构内存提供者以支持大型语言模型代理
Abstract
Current agents remain largely stateless across tasks, limiting their ability to continually improve from prior interactions and making memory essential for long-horizon agentic behavior. Existing memory methods seek to reuse past experience, but most rely on a single memory representation (e.g., trajectories, reflections, skills, structured knowledge) whose effectiveness varies across task distributions. Rethinking this design space, we evaluate 13 memory methods and find that no single method generalizes across benchmarks, revealing the potential of managing heterogeneous memory providers. We formulate agent memory as a routing problem in which a memory agent decides which memory provider to retrieve from, whether to inject short-term memory, and which providers should store the resulting experience. Based on this perspective, we propose MemAgent, featuring a content-aware routing architecture and a training-data synthesis pipeline. The routing architecture combines content-aware probing before retrieval, short-term memory gating during execution, and selective multi-provider storage, while the training pipeline synthesizes phase-specific supervision for routing decisions. Across GAIA, WebWalkerQA, and xBench-DS, MemAgent improves average accuracy by 10.0% and outperforms every individual memory method across all three benchmarks. These gains come with less than 0.3% routing overhead and a 12% reduction in average task steps.
Chinese Translation
当前的代理在任务之间大多保持无状态，这限制了它们从先前交互中持续改进的能力，使得内存对于长期代理行为至关重要。现有的内存方法试图重用过去的经验，但大多数依赖于单一的内存表示（例如，轨迹、反思、技能、结构化知识），其有效性在不同任务分布中有所不同。重新思考这一设计空间，我们评估了13种内存方法，发现没有单一方法能够在基准测试中普遍适用，这揭示了管理异构内存提供者的潜力。我们将代理内存形式化为一个路由问题，其中内存代理决定从哪个内存提供者检索、是否注入短期内存，以及哪些提供者应存储结果经验。基于这一视角，我们提出了MemAgent，具有内容感知的路由架构和训练数据合成管道。路由架构结合了检索前的内容感知探测、执行过程中的短期内存门控以及选择性多提供者存储，而训练管道则合成了针对路由决策的阶段特定监督。在GAIA、WebWalkerQA和xBench-DS上，MemAgent的平均准确率提高了10.0%，并在所有三个基准测试中超越了每种单独的内存方法。这些提升的路由开销不足0.3%，且平均任务步骤减少了12%。
cs.AI / 89 / 2609.32522
Beyond Dyadic Memory: Interaction-Aware Multimodal Memory with Adaptive Agentic Retrieval for Multi-Party Spoken Conversations
超越二人记忆：具有自适应代理检索的交互感知多模态记忆用于多方口语对话
Abstract
Long-term memory enables agents to accumulate information and reason across sessions, yet existing research primarily focuses on dyadic text or image-text conversations, leaving long-term memory for multi-party spoken conversations underexplored. This setting requires preserving conversational content, identifying participants across sessions, and retaining who speaks to whom. To this end, we propose VoxPolyMem, an interaction-aware multimodal memory framework combining incremental speaker identification with a memory hierarchy comprising interaction memory, fact memory, and participant profiles. We formulate retrieval as sequential decision-making, where an agent rewrites queries and selects retrieval tools and memory layers based on accumulated evidence to address information gaps. We further introduce Evidence-Gain GRPO (EG-GRPO), which uses round-wise credit assignment to encourage complementary evidence acquisition. We also construct VoxPolyBench to evaluate memory evolution, personalized answering, memory retrieval and reasoning, and interaction reasoning and attribution in multi-party spoken conversations. VoxPolyMem achieves an overall score of 85.0 on VoxPolyBench, surpassing the strongest evaluated baseline by 23.6 points. On Mem-Gallery and H2HMem-Multi, it scores 89.6 and 74.4, respectively, exceeding the strongest evaluated public memory baselines by over 8 points each. These results highlight its potential for persistent, personalized assistance in multi-party multimodal interactions. Code and datasets are available at https://voxpolymem.github.io/VoxPolyBench/demo/
Chinese Translation
长期记忆使代理能够积累信息并在多个会话中进行推理，但现有研究主要集中于二人文本或图像-文本对话，导致多方口语对话的长期记忆研究尚未深入。这种情境需要保留对话内容、识别跨会话的参与者，并保持谁与谁交谈。为此，我们提出了VoxPolyMem，一个交互感知的多模态记忆框架，结合了增量说话者识别与包含交互记忆、事实记忆和参与者档案的记忆层次结构。我们将检索形式化为顺序决策过程，其中代理根据累积的证据重写查询并选择检索工具和记忆层，以解决信息缺口。我们进一步引入了证据增益GRPO（EG-GRPO），该方法使用轮次信用分配来鼓励互补证据的获取。我们还构建了VoxPolyBench，以评估记忆演变、个性化回答、记忆检索与推理，以及多方口语对话中的交互推理与归因。VoxPolyMem在VoxPolyBench上获得了85.0的总体得分，超越了最强评估基线23.6分。在Mem-Gallery和H2HMem-Multi上，分别获得89.6和74.4的得分，各自超过了最强评估公共记忆基线8分以上。这些结果突显了其在多方多模态交互中提供持久个性化辅助的潜力。代码和数据集可在https://voxpolymem.github.io/VoxPolyBench/demo/获取。
cs.AI / 90 / 2609.32527
AmbiModBench: Benchmarking Gene Perturbation Prediction Beyond Shared Responses
AmbiModBench：超越共享响应的基因扰动预测基准测试
Abstract
Predicting cellular responses to genetic perturbations helps prioritize experiments in single-cell genomics, where exhaustive measurement is infeasible. While computational models increasingly predict these responses, three evaluation deficiencies obscure what their scores demonstrate. First, absolute metrics cannot separate target-specific predictions from a shared background response. Second, common metrics remain high under gene shuffling, so gene-level accuracy is never verified. Third, a score at one training size says nothing about coverage, which depends on representation-space proximity and response-constraining power. We propose AmbiModBench, a specificity-aware, gene-resolved and coverage-aware benchmark. It pairs every score with a training-mean reference fitted on the same split, screens each readout by gene-coordinate permutation, and links embedding distance to response variation. Across K562, RPE1 and Norman, strong absolute scores largely reflect shared background rather than target-specific learning. Widely used readouts track response magnitude distributions rather than the affected genes. Detectable gain follows representation-space coverage rather than training-set size. Nonetheless, on RPE1 the protocol yields a reproducible target-specific gain across five additional splits and three gene selections, which absolute scores alone cannot distinguish from shared background.
Chinese Translation
预测细胞对基因扰动的响应有助于在单细胞基因组学中优先考虑实验，因为全面测量是不可行的。尽管计算模型越来越能够预测这些响应，但三个评估缺陷模糊了它们的评分所展示的内容。首先，绝对指标无法将目标特异性预测与共享背景响应区分开。其次，常用指标在基因重排下仍然保持较高，因此基因级准确性从未得到验证。第三，单一训练规模的评分无法反映覆盖率，而覆盖率依赖于表示空间的接近度和响应约束能力。我们提出了AmbiModBench，这是一种关注特异性、基因解析和覆盖率的基准测试。它将每个评分与在相同分割上拟合的训练均值参考配对，通过基因坐标置换筛选每个读出，并将嵌入距离与响应变异联系起来。在K562、RPE1和Norman细胞系中，强绝对评分在很大程度上反映了共享背景，而非目标特异性学习。广泛使用的读出跟踪响应幅度分布，而不是受影响的基因。可检测的增益遵循表示空间的覆盖率，而非训练集的大小。然而，在RPE1细胞系中，该协议在五个额外分割和三个基因选择中产生了可重复的目标特异性增益，而仅凭绝对评分无法将其与共享背景区分开。
cs.AI / 91 / 2609.32528
Fail Loudly: An Auditable Runtime for Agentic Data Analysis
大声失败：可审计的代理数据分析运行时
Abstract
Large language models (LLMs) have enabled data-science agents to automate multi-step analyses over heterogeneous files. However, incorrect choices regarding data sources, scope, or statistical definitions often lead to silent errors: computations execute successfully but produce plausible yet incorrect outputs that fail to answer the intended question. To mitigate this, we present RADAR, an auditable runtime that makes an agent's analytical choices inspectable and supports their revision through execution feedback. RADAR operates through three core mechanisms. First, an evidence-preserving exploration module retrieves task-relevant content while retaining source locations and observation coverage. Next, the runtime uses typed operators to record the agent's declared inputs, operation arguments, and resulting observations. Finally, runtime validation checks proposed operations against these observations. When a conflict is detected, the runtime rejects the operation or provides diagnostic feedback, allowing the agent to revise its choices before errors propagate. This design enables agents to fail loudly while leaving semantic interpretation to the LLM. On KramaBench, RADAR achieves overall scores of 0.723 with full source retrieval and 0.747 with gold sources supplied, corresponding to relative gains of 35.9% and 28.8% over the strongest baselines. Beyond KramaBench, RADAR achieves relative performance gains of 14.0% on DA-Code and 59.3% on DABStep, demonstrating its applicability across diverse agentic data-analysis workflows.
Chinese Translation
大型语言模型（LLMs）使数据科学代理能够自动化对异构文件的多步骤分析。然而，关于数据源、范围或统计定义的错误选择常常导致无声错误：计算成功执行，但产生看似合理但实际上错误的输出，无法回答预期的问题。为了解决这一问题，我们提出了RADAR，一个可审计的运行时，使代理的分析选择可供检查，并通过执行反馈支持其修订。RADAR通过三个核心机制运行。首先，证据保留探索模块在保留源位置和观察覆盖的同时检索与任务相关的内容。接下来，运行时使用类型化操作符记录代理声明的输入、操作参数和结果观察。最后，运行时验证检查提议的操作与这些观察的一致性。当检测到冲突时，运行时拒绝该操作或提供诊断反馈，允许代理在错误传播之前修订其选择。这一设计使代理能够大声失败，同时将语义解释留给LLM。在KramaBench上，RADAR实现了0.723的整体得分，具有完整的源检索，以及0.747的得分，提供了黄金源，相较于最强基线分别提高了35.9%和28.8%。在KramaBench之外，RADAR在DA-Code上实现了14.0%的相对性能提升，在DABStep上实现了59.3%的提升，展示了其在多样化代理数据分析工作流中的适用性。
cs.AI / 92 / 2609.32533
LLMAdBench: A Human Preference Benchmark for Advertising in LLM Responses
LLMAdBench：针对大型语言模型响应中广告的人类偏好基准
Abstract
Inserting advertisements (ads) into consumer-facing LLM output is emerging as a new business model, but there is little shared evidence on how such ad insertion should be evaluated or how it affects user preferences. We introduce LLMAdBench, a human-preference benchmark for studying advertising in LLM-generated content. The benchmark isolates a simple but practically important decision: given a user conversation, an LLM response, and a matched advertisement, where should the ad be placed? Our dataset compares pairs of responses that differ only in ad position while holding all other conditions fixed including the user query, base answer, advertisement, and disclosure condition. Human annotators evaluate each pair based on six criteria from both advertiser's and user's perspectives. The resulting benchmark contains more than 18000 human judgments across two disclosure conditions: explicitly labeling the ad as sponsored and merging it into the response without disclosure. We use LLMAdBench to evaluate eight frontier LLMs as preference judges and find that they are not reliable substitutes for human evaluation. Even the most stable models reverse roughly one quarter of their decisions when the presentation order is swapped, agreement across models is low, and their placement preferences differ systematically from those of human annotators. Moreover, LLMAdBench contains substantial learnable signal. In particular, a Qwen3-8B model fine-tuned on the human preferences improves substantially over its base model and outperforms all zero-shot frontier judges on the held-out prediction task. Beyond model evaluation, LLMAdBench provides quantitative evidence on the advertiser-user trade-off and shows that the sponsorship disclosure systematically changes users' preference over ad placement.
Chinese Translation
将广告（ads）插入面向消费者的大型语言模型（LLM）输出中正在成为一种新的商业模式，但关于如何评估这种广告插入以及它如何影响用户偏好的共享证据仍然很少。我们介绍了LLMAdBench，这是一个用于研究LLM生成内容中广告的人类偏好基准。该基准隔离了一个简单但在实践中重要的决策：在给定用户对话、LLM响应和匹配广告的情况下，广告应该放置在哪里？我们的数据集比较了仅在广告位置上有所不同的响应对，同时保持用户查询、基础答案、广告和披露条件等所有其他条件不变。人类注释者根据广告主和用户的六个标准评估每对响应。最终的基准包含超过18000条人类判断，涵盖两种披露条件：明确标记广告为赞助和将其合并到响应中而不披露。我们使用LLMAdBench评估了八个前沿LLM作为偏好评判者，发现它们并不是人类评估的可靠替代品。即使是最稳定的模型，在展示顺序交换时也会大约反转四分之一的决策，模型之间的一致性较低，并且它们的放置偏好与人类注释者的偏好系统性不同。此外，LLMAdBench包含了大量可学习的信号。特别是，经过人类偏好微调的Qwen3-8B模型在其基础模型上有了显著改善，并在保留的预测任务中超过了所有零-shot前沿评判者。除了模型评估，LLMAdBench还提供了关于广告主与用户之间权衡的定量证据，并显示赞助披露系统性地改变了用户对广告位置的偏好。
cs.AI / 93 / 2609.32539
Interpretable Physics Informed WiFi Indoor Localization: Learning an Effective Access Point Geometry and Using It to Prune
可解释的物理信息驱动的WiFi室内定位：学习有效的接入点几何形状并利用其进行剪枝
Abstract
Deep learning models can achieve high accuracy for indoor localization, but their black-box nature limits interpretability and the reuse of learned information. We propose a hierarchical deep learning framework for WiFi fingerprint-based indoor localization that jointly predicts user location and learns an effective geometry of the surrounding access points (APs). Physics-informed decoders infer this geometry directly from RSSI measurements and labelled user positions, without requiring the true AP coordinates during training. The learned geometry is then used to rank and prune APs. On the UJIIndoorLoc dataset, the proposed chained model achieves a mean 3D localization error of 7.07 m, reducing error by 26% to 36% compared with baseline models. Previously published methods evaluated on the same official split report errors 10.6% to 31.0% higher. Pruning 35% or 50% of the APs causes only a small loss in localization accuracy. The inferred geometry also enables Fisher-information-based AP ranking even when fingerprint databases do not contain surveyed AP coordinates. Experiments on the Tampere/TUT and UTSIndoorLoc datasets show that geometry-guided AP selection performs comparably to selectors built directly from labelled data. These results show that physics-informed interpretability can improve indoor localization while also supporting effective feature selection.
Chinese Translation
深度学习模型在室内定位中可以实现高精度，但其黑箱特性限制了可解释性和学习信息的重用。我们提出了一种层次化的深度学习框架，用于基于WiFi指纹的室内定位，该框架联合预测用户位置并学习周围接入点（AP）的有效几何形状。物理信息驱动的解码器直接从RSSI测量值和标记的用户位置推断该几何形状，而无需在训练过程中使用真实的AP坐标。学习到的几何形状随后用于对AP进行排名和剪枝。在UJIIndoorLoc数据集上，所提议的链式模型实现了平均3D定位误差为7.07米，比基线模型减少了26%到36%的误差。之前在相同官方划分上评估的方法报告的误差高出10.6%到31.0%。剪除35%或50%的AP仅导致定位精度的小幅下降。推断的几何形状还使得即使在指纹数据库中不包含调查的AP坐标时，也能进行基于Fisher信息的AP排名。在Tampere/TUT和UTSIndoorLoc数据集上的实验表明，基于几何指导的AP选择与直接从标记数据构建的选择器表现相当。这些结果表明，物理信息驱动的可解释性可以提高室内定位的效果，同时支持有效的特征选择。
cs.AI / 94 / 2609.32544
Porimon: An LLM-Based Pok\'emon Battle Agent Enhanced by Long/Short-Term Knowledge Augmented Generation
Porimon：一种基于大型语言模型的宝可梦对战代理，增强了长短期知识增强生成
Abstract
In this paper, we use Pok\'emon Battles as a case study to investigate how to improve the performance of LLM-based agents in tasks that require opponent-aware planning without additional fine-tuning. We propose Long/Short-Term Knowledge Augmented Generation (LSTKAG), a mechanism that enables LLM-based agents to leverage past states of the current task and retrieve experience summaries from similar previous task instances based on the current state. Based on LSTKAG, we design Porimon, an LLM-based agent structure for Pok\'emon Battles. For optimization, we introduce an external API for precise damage calculation and more detailed information about the game. We conduct tournament-like evaluation experiments comprising 15,000 battles for hyperparameter optimization, ablation studies, and performance evaluation. The results indicate that Porimon-based players with hyperparameter optimization significantly outperform players based on Pok\'eLLMon, an LLM-based agent structure proposed in previous research, and the rule-based heuristic player. Furthermore, our ablation study shows that Porimon variants outperform the one without extension in game information retrieval, which shows the contribution of that extension. However, the current experiment results are inconclusive regarding the contribution of Long-Term KAG. These results suggest that introducing external resources, information from previous states of the current task, and experience summaries from similar previous task instances could elevate the performance of LLM-based agents designed for tasks requiring opponent-aware planning.
Chinese Translation
本文以宝可梦对战为案例研究，探讨如何在不进行额外微调的情况下提高基于大型语言模型（LLM）代理在需要对手意识规划的任务中的表现。我们提出了长短期知识增强生成（Long/Short-Term Knowledge Augmented Generation, LSTKAG）机制，使得基于LLM的代理能够利用当前任务的过去状态，并根据当前状态从类似的先前任务实例中检索经验摘要。基于LSTKAG，我们设计了Porimon，一种用于宝可梦对战的基于LLM的代理结构。为了优化，我们引入了一个外部API，以实现精确的伤害计算和更详细的游戏信息。我们进行了类似锦标赛的评估实验，包括15,000场对战，用于超参数优化、消融研究和性能评估。结果表明，经过超参数优化的Porimon玩家显著优于基于PokéLLMon（先前研究中提出的基于LLM的代理结构）和基于规则的启发式玩家。此外，我们的消融研究显示，Porimon的变体在游戏信息检索方面优于未扩展版本，表明该扩展的贡献。然而，目前的实验结果对于长期知识增强（Long-Term KAG）的贡献尚不明确。这些结果表明，引入外部资源、当前任务的过去状态信息以及类似先前任务实例的经验摘要，可以提升为需要对手意识规划而设计的基于LLM的代理的性能。
cs.AI / 95 / 2609.32550
Are Vision-Language-Action Models Robust to One-Step Observation Perturbations?
视觉-语言-动作模型对一步观察扰动的鲁棒性如何？
Abstract
Understanding the safety risks of vision-language-action (VLA) models is essential for their deployment in the physical world. Existing safety research has mainly considered persistent perturbations that are applied continuously to observations throughout an episode. However, momentary observation corruption, in which observations are severely perturbed only briefly within an episode, remains an underexplored safety threat. To address this gap, this work investigates robustness to one-step perturbations applied at a single time step per episode. Our experiments reveal that these perturbations substantially degrade VLA performance and that their impact depends on the action chunk execution length. Based on them, we propose CARE, which dynamically selects the execution length based on consistency with the previously predicted action chunk. CARE improves robustness with low computational overhead while preserving clean performance by selecting shorter execution lengths only under perturbations.
Chinese Translation
理解视觉-语言-动作（VLA）模型的安全风险对于其在物理世界中的部署至关重要。现有的安全研究主要考虑在整个情节中持续施加的扰动。然而，瞬时观察损坏，即观察在情节中仅短暂严重扰动，仍然是一个未被充分探索的安全威胁。为了解决这一空白，本研究调查了在每个情节的单个时间步施加的一步扰动的鲁棒性。我们的实验表明，这些扰动显著降低了VLA的性能，并且其影响依赖于动作块执行长度。基于此，我们提出了CARE，它根据与先前预测的动作块的一致性动态选择执行长度。CARE在保持干净性能的同时，通过仅在扰动下选择较短的执行长度，提高了鲁棒性，并且计算开销较低。
cs.AI / 96 / 2609.32562
Artificial intelligences and human scientists exhibit complementary strengths in theory building
人工智能与人类科学家在理论构建中展现互补优势
Li, Ke, Zoumpoulis, Spyros I., Puranam, Phanish, Parker, Philip, Eshbaugh-Soha, Matthew, Gainsburg, Izzy, Gilead, Michael, Grossmann, Igor, Hadar, Britt, Inbar, Yoel, Simchon, Almog, Willer, Robb, Ai, Rui, Ao, Ruicheng, Bala, Gavin J., Bidwell, Matthew, Cai, Shuang, Chang, Kai, Chen, Skyler Y., Clark, Cory J., Dai, Irmak, Dalal, Abhinandan, Douglas, Connor, Du, Alexis, Du, Zhehang, Feng, Leyun, Fernandez-Mateo, Isabel, Gandhi, Linnea, Grumbach, Cyrille, Gupta, Anmol, Gupta, Vansh, Hademer, Maria, Hardy III, Jay H., Huang, Chen Kai, Jin, Jacob Xiangyu, Keskin, Ufuk, Kim, Na Hyun, Kobaş, Mert, Koh, Byounghoon, Lamont-Dobbin, Gabrielle, Lanzalotto, Gregory, Lee, Sun Young, Leng, Dingzhe, Li, Chenjun, Li, Weiyuan, Li, Zeyuan, Liang, Zhongyuan, Liu, Ning, Liu, Peihong, Liu, Yuhan, Lu, Jiuyao, Ma, Wanteng, Martinet, Nicolas, Mulat, Natnael, Nguyen, Christina A., Nguyen, Khai, Nguyen, Quang Minh, Pape, Naja, Park, Chanwoo, Poulidis, Stefanos, Sanchez-Burks, Jeffrey, Schaerer, Michael, Solal, Isabelle, Song, Yanbo, Sun, Junghyo, Sun, Qingyao, Sun, Rui, Swaab, Roderick, Tan, Kevin, Teng, Dequn, Vaccaro, Michelle A., Vigerbaeck, Robin, Wang, Xiaomeng, Yao, Randol H., Yilmaz, Duygu, Yiu, Shun, Yucesoy, Ecem, Zang, Allen, Zhang, Ruijia, Zhang, Xilan, Zhang, Yichi, Zhang, Zhanhao, Uhlmann, Eric Luis
Abstract
We investigate the effectiveness of artificial intelligences (AI)-specifically large language models (LLMs)-relative to human scientists at high-level cognitive tasks in social science such as theory formulation, predictions of novel empirical results, and theory revision in response to new evidence. The research domain was academic discourse regarding gender and race inequality. Our findings, comparing 25 LLMs with 13 senior researchers and 60 doctoral scholars, reveal that the AIs outperformed most humans individually on most of the present tasks, while human theories were more diverse and exhibited greater gains in predictive accuracy from aggregation. AI-generated theories were more extensively elaborated, involving additional theoretical paths and latent variables, and were rated as higher quality than human theories by independent raters blinded to source. However, this theoretical complexity was in part ornamental, in that it was not associated with more accurate predictions about empirical patterns in data; in contrast, human scientists achieved greater predictive efficiency with simpler theories. The AIs were significantly more likely than human scientists to revise their theories to incorporate new evidence; human scientists updated their beliefs in a selective way that is sensitive to prior prediction errors. We speculate that the superior processing capacity of artificial intelligences makes them especially well-suited to tasks requiring grappling with complexity, but that the greater diversity of human ideas is essential to wise crowds and collective creativity.
Chinese Translation
我们研究了人工智能（AI）——特别是大型语言模型（LLMs）——相对于人类科学家在社会科学中进行高水平认知任务的有效性，例如理论形成、对新经验结果的预测以及对新证据的理论修订。研究领域为关于性别和种族不平等的学术话语。我们的研究结果比较了25个LLM与13位高级研究人员和60位博士生，发现AI在大多数当前任务中超越了大多数个体人类，而人类理论则更加多样，并且在聚合中表现出更大的预测准确性提升。AI生成的理论更为详尽，涉及额外的理论路径和潜在变量，并且在独立评审者（对来源保持盲态）看来，其质量高于人类理论。然而，这种理论复杂性在某种程度上是华而不实的，因为它与对数据中经验模式的更准确预测并无关联；相反，人类科学家通过更简单的理论实现了更高的预测效率。AI在修订理论以纳入新证据方面显著更有可能，而人类科学家则以对先前预测错误敏感的选择性方式更新其信念。我们推测，人工智能的优越处理能力使其特别适合于需要应对复杂性的任务，但人类思想的更大多样性对于明智的群体和集体创造力至关重要。
cs.AI / 97 / 2609.32564
ProTTT: Learning to Learn Semantic User Memory with Test-Time Training
ProTTT：通过测试时训练学习语义用户记忆
Abstract
Personalization requires language models to capture user-specific knowledge from a growing user history. Existing context-based approaches incur increasing inference costs as user history accumulates and rely on separate retrieval or summarization stages, while parametric-based approaches often require reconstructing user representations when new user data is added. We introduce ProTTT, a profile-supervised meta-learning framework for learning semantic user memory. The memory construction starts from a shared initialization and is updated for each user through test-time training on user history, allowing it to evolve continuously as the history grows. However, since test-time training alone does not explicitly encourage the memory to capture semantic user knowledge necessary for personalization, we learn this shared initialization using textual user profiles as supervision, so that test-time training on user history captures semantic knowledge more effectively. ProTTT consistently outperforms both full history ICL and all parametric baselines across diverse benchmarks, while substantially reducing inference cost by compressing user history into a lightweight parameterized memory. Our analysis also shows that profile supervision is a reliable objective for learning semantic user knowledge and that the resulting memory can track and retain evolving user preferences, while remaining robust across different history sizes. Overall, we demonstrate the effectiveness of test-time training for personalization and establish ProTTT as a baseline for continuously evolving user memory.
Chinese Translation
个性化要求语言模型从不断增长的用户历史中捕捉用户特定知识。现有的基于上下文的方法在用户历史累积时会导致推理成本不断增加，并依赖于单独的检索或摘要阶段，而基于参数的方法在添加新用户数据时通常需要重建用户表示。我们提出了ProTTT，一个基于配置文件监督的元学习框架，用于学习语义用户记忆。记忆构建从共享初始化开始，并通过对用户历史的测试时训练为每个用户更新，使其能够随着历史的增长而持续演变。然而，由于单独的测试时训练并没有明确鼓励记忆捕捉个性化所需的语义用户知识，我们使用文本用户配置文件作为监督来学习这一共享初始化，以便在用户历史上的测试时训练能够更有效地捕捉语义知识。ProTTT在各种基准测试中始终优于完整历史的即时学习（ICL）和所有基于参数的基线，同时通过将用户历史压缩为轻量级的参数化记忆大幅降低推理成本。我们的分析还表明，配置文件监督是学习语义用户知识的可靠目标，所得到的记忆能够跟踪和保留不断演变的用户偏好，同时在不同历史大小下保持稳健。总体而言，我们展示了测试时训练在个性化中的有效性，并确立了ProTTT作为持续演变用户记忆的基线。
cs.AI / 98 / 2609.32574
CUE-Mem: Benchmarking Long-Term User Memory via Implicit Cues in Multimodal Conversations
CUE-Mem：通过隐性线索在多模态对话中评估长期用户记忆的基准
Abstract
Long-term memory is essential for multimodal agents that interact with users across sustained conversations. However, user memories are not always explicitly stated: they may also be implied by recurring background objects in images, ambient sounds in audio, or other peripheral multimodal cues. Existing benchmarks largely focus on text-only memory or explicit multimodal evidence, leaving implicit multimodal cues underexplored. We introduce CUE-Mem, a text-image-audio benchmark for evaluating long-term user memory from implicit cues. CUE-Mem contains 2,674 questions across explicit and implicit evidence settings and covers four tasks: Entity Recall, Long Pattern, Personalized Recommendation, and Answer Refusal. Across textualized memory systems, implicit performance remains far below oracle evidence, locating the main bottleneck in preserving and retrieving subtle cues rather than question answerability. Increasing caption detail recovers more of this evidence, but brings uneven gains and rapidly growing token costs, motivating native multimodal access. Yet native access does not uniformly resolve the bottleneck: evidence use depends strongly on the backbone, while multimodal indexing introduces substantial retrieval noise. CUE-Mem provides a testbed for memory systems that selectively retain, retrieve, and use subtle multimodal evidence.
Chinese Translation
长期记忆对于在持续对话中与用户互动的多模态代理至关重要。然而，用户记忆并不总是明确表达的：它们也可能通过图像中的重复背景物体、音频中的环境声音或其他外围多模态线索隐含存在。现有基准主要集中在仅文本的记忆或明确的多模态证据上，未能充分探索隐性多模态线索。我们引入了CUE-Mem，这是一个用于评估来自隐性线索的长期用户记忆的文本-图像-音频基准。CUE-Mem包含2,674个问题，涵盖明确和隐性证据设置，并包括四个任务：实体回忆、长模式、个性化推荐和拒绝回答。在文本化记忆系统中，隐性表现远低于理想证据，主要瓶颈在于保持和检索微妙线索，而非问题的可回答性。增加描述的细节可以恢复更多证据，但带来了不均匀的收益和快速增长的标记成本，从而激励原生多模态访问。然而，原生访问并未均匀解决瓶颈：证据的使用在很大程度上依赖于基础模型，而多模态索引则引入了大量检索噪声。CUE-Mem为选择性保留、检索和使用微妙多模态证据的记忆系统提供了一个测试平台。
cs.AI / 99 / 2609.32584
EMIR$^2$: Evolution-Aware Memory with Intent-Guided Multi-Round Retrieval
EMIR$^2$: 具备演变意识的意图引导多轮检索的记忆框架
Abstract
Long-term memory enables large language model (LLM) agents to leverage historical interactions for future tasks. However, existing memory systems struggle to utilize continuously evolving historical information, as they often rely on static memory representations and single-round retrieval strategies, failing to track factual changes or integrate distributed evidence across long-term interactions. To address these challenges, we propose \textsc{EMIR}$^{2}$, an \textbf{E}volution-Aware \textbf{M}emory framework with \textbf{I}ntent-Guided Multi-\textbf{R}ound \textbf{R}etrieval, enabling LLM agents to maintain evolving historical knowledge and adaptively retrieve relevant evidence. Specifically, \textsc{EMIR}$^{2}$ constructs a State-Evolving Memory Graph (SEMG) that represents long-term memory as evolving knowledge states supported by temporal event trajectories and evidential associations. By maintaining semantic states through evidence-based updates, SEMG preserves historical evolution and enables evidence tracing under complex and conflicting scenarios. Building upon this, we introduce an intent-guided multi-round retrieval mechanism that iteratively identifies missing evidence and expands retrieval based on accumulated information. Experiments on LoCoMo and MemConflict demonstrate that \textsc{EMIR}$^{2}$ improves long-term memory utilization, dynamic and static conflict handling, and complex retrieval performance, achieving relative improvements of more than 12\% in certain categories. These results highlight the effectiveness of jointly modeling memory evolution and adaptive evidence acquisition for long-term agent interactions.
Chinese Translation
长期记忆使大型语言模型（LLM）代理能够利用历史交互来完成未来任务。然而，现有的记忆系统在利用不断演变的历史信息方面面临挑战，因为它们通常依赖于静态记忆表示和单轮检索策略，无法跟踪事实变化或整合长期交互中的分散证据。为了解决这些问题，我们提出了 extsc{EMIR}$^{2}$，一个具备 extbf{E}volution-Aware（演变意识）记忆框架，结合 extbf{I}ntent-Guided（意图引导）多 extbf{R}ound（轮次） extbf{R}etrieval（检索），使LLM代理能够维护不断演变的历史知识并适应性地检索相关证据。具体而言， extsc{EMIR}$^{2}$构建了一个状态演变记忆图（State-Evolving Memory Graph，SEMG），将长期记忆表示为由时间事件轨迹和证据关联支持的演变知识状态。通过基于证据的更新维护语义状态，SEMG保留了历史演变，并在复杂和冲突场景下实现了证据追踪。在此基础上，我们引入了一种意图引导的多轮检索机制，迭代识别缺失证据并根据累积信息扩展检索。在LoCoMo和MemConflict上的实验表明， extsc{EMIR}$^{2}$提高了长期记忆的利用率、动态和静态冲突处理能力以及复杂检索性能，在某些类别中实现了超过12 ext{%}的相对提升。这些结果突显了共同建模记忆演变和适应性证据获取在长期代理交互中的有效性。
cs.AI / 100 / 2609.32594
MA-FPPO: Multi-Agent Flow-Pretrained Policy Optimization
MA-FPPO：多智能体流预训练策略优化
Abstract
Multi-agent flow policies learn cooperative behavior from fixed offline datasets, but often struggle to complete tasks in situations not covered by the offline data. In these situations, agents must both adapt to changes in the environment and coordinate with one another, yet action patterns learned offline are often insufficient for effective adaptation and coordination. To address this problem, we propose Multi-Agent Flow-Pretrained Policy Optimization (MA-FPPO), which uses online fine-tuning to improve the cooperative behavior of models pretrained with flow matching through new interactions with the environment. Building on the behavior learned during pretraining, we construct policies with explicit action likelihoods for discrete and continuous action spaces. We then update the pretrained model using shared team advantages to further improve coordination based on team performance. Our method achieves, on average, relative gains of 52.8% over the strongest listed offline baselines across 30 settings and 29.8% over purely online learning across 38 comparisons with matched online budgets and evaluation protocols.
Chinese Translation
多智能体流策略从固定的离线数据集中学习合作行为，但在离线数据未覆盖的情况下，往往难以完成任务。在这些情况下，智能体必须适应环境的变化并相互协调，而离线学习的行动模式通常不足以有效适应和协调。为了解决这个问题，我们提出了多智能体流预训练策略优化（MA-FPPO），该方法通过与环境的新互动，利用在线微调来改善通过流匹配预训练模型的合作行为。在预训练期间学习到的行为基础上，我们为离散和连续动作空间构建具有明确动作可能性的策略。然后，我们使用共享团队优势更新预训练模型，以进一步基于团队表现改善协调。我们的方法在30个设置中，相较于最强的离线基线平均获得52.8%的相对增益，并在38个与匹配在线预算和评估协议的比较中，相较于纯在线学习获得29.8%的增益。
cs.AI / 101 / 2609.32600
CUA-SWE: When Computer-Use Agents Meet Visual Software Engineering
CUA-SWE：当计算机使用代理遇到视觉软件工程
Abstract
Software development requires more than editing code: developers repeatedly run software, interact with its interfaces, visually inspect its behavior, and use these observations to decide what to change next and whether a change works. Existing coding agents and computer-use agents are largely studied in isolation, leaving this integrated development process underexplored. Diagnosing a runtime interaction failure requires agents to connect visual observations with the responsible code, then use the application again to verify the repair. We introduce CUA-SWE, a benchmark, environment, and evaluation pipeline for software engineering with computer use. Beyond studying how GUI feedback supports diagnosis and repair, we ask whether agents can complete software engineering tasks when required specification or operational information is available only through the running application's visual interface. CUA-SWE spans four software engineering domains and requires agents to modify code and configuration, execute commands, interact with running software, and inspect visual feedback within the same task. Each task includes deterministic, task-specific tests that verify whether the resulting software satisfies the requirements and preserves specified behavior. Our evaluation characterizes how frontier agents combine source-level execution with application screenshots and graphical interaction to produce verified software changes. We examine performance across domains and task information requirements, alongside the development behaviors associated with successful repairs. CUA-SWE provides a unified testbed for studying how agents use visual feedback and interaction to guide software engineering, with executable correctness criteria for the resulting software.
Chinese Translation
软件开发不仅仅是编辑代码：开发者反复运行软件，与其接口互动，直观检查其行为，并利用这些观察来决定下一步的更改以及更改是否有效。现有的编码代理和计算机使用代理大多是孤立研究的，这使得这一综合开发过程未得到充分探索。诊断运行时交互失败需要代理将视觉观察与相关代码连接起来，然后再次使用应用程序以验证修复。我们提出了CUA-SWE，这是一个针对计算机使用的软件工程基准、环境和评估管道。除了研究图形用户界面（GUI）反馈如何支持诊断和修复外，我们还探讨了当所需的规格或操作信息仅通过运行应用程序的视觉界面提供时，代理是否能够完成软件工程任务。CUA-SWE涵盖四个软件工程领域，要求代理在同一任务中修改代码和配置、执行命令、与运行中的软件互动，并检查视觉反馈。每个任务包括确定性的、特定任务的测试，以验证生成的软件是否满足要求并保持指定行为。我们的评估描述了前沿代理如何将源级执行与应用程序截图和图形交互相结合，以产生经过验证的软件更改。我们考察了不同领域和任务信息需求下的性能，以及与成功修复相关的开发行为。CUA-SWE为研究代理如何利用视觉反馈和交互指导软件工程提供了一个统一的测试平台，并为生成的软件设定了可执行的正确性标准。
cs.AI / 102 / 2609.32616
"You're Right, Let Me Fix It": How LLM Agents Damage Correct Work When Falsely Accused
“你说得对，让我修复它”：大型语言模型代理在被错误指控时如何损害正确工作
Abstract
LLM agents increasingly keep working after a task succeeds as they resume after compaction or take over handoffs. Their finished work keeps receiving follow-up input that sometimes falsely accuses it for later failures. We call an agent's acceptance of such a false accusation gaslight sycophancy, and destructive over-correction when acting on it damages previously correct work. We introduce CAVE-Bench, a benchmark of 365 agentic tasks across six domains built around opaque tasks. Every scored run first reaches a verified correct state, whose supporting rationale and history stay in the workspace while the facts that would settle the accusation lie in external or runtime state beyond the agent's reach. The agent cannot confirm or refute the claim with a local check, so the right response should keep the work and ask for the missing evidence. Each task either hands the agent correct work with saved evidence or let it build and verify that work first, and five risk factors set how the accusation enters the workflow. We score accusation acceptance and evidence use from the trajectory and measure harm by deterministic replay of downstream events. Across 14 of the latest models in Claude Code, false accusations damage correct work in up to 60.06% of runs, and stronger models often do so after recovering the supporting evidence. The same model behaves differently across OpenCode, Codex, and Hermes, and a harness gate driven by the benchmark's live signals cuts replayed harm by 74%. These results show that preserving already-correct work under unsupported accusation is a distinct safety challenge for long-lived agents. Our project is in https://henrymao2004.github.io/agent-over-correction/.
Chinese Translation
随着大型语言模型（LLM）代理在任务成功后继续工作，它们在压缩后恢复或接管交接时变得越来越常见。它们完成的工作不断收到后续输入，有时会因后续失败而受到错误指控。我们将代理接受这种错误指控称为“煤气灯式谄媚”，而在此基础上采取的破坏性过度修正则会损害先前正确的工作。我们引入了CAVE-Bench，这是一个涵盖六个领域的365个代理任务的基准，围绕不透明任务构建。每个得分运行首先达到一个经过验证的正确状态，其支持理由和历史保留在工作区，而解决指控的事实则位于代理无法触及的外部或运行时状态中。代理无法通过本地检查确认或反驳该声明，因此正确的响应应该是保留工作并请求缺失的证据。每个任务要么将正确的工作和保存的证据交给代理，要么让其首先构建和验证该工作，并且五个风险因素决定指控如何进入工作流程。我们从轨迹中对指控接受和证据使用进行评分，并通过确定性重放下游事件来衡量损害。在Claude Code的14个最新模型中，错误指控在高达60.06%的运行中损害了正确的工作，而更强的模型往往在恢复支持证据后仍会这样做。同一模型在OpenCode、Codex和Hermes中表现不同，而基于基准的实时信号驱动的保护门将重放损害减少了74%。这些结果表明，在没有支持的指控下保护已经正确的工作是长期存在的代理面临的一个独特安全挑战。我们的项目网址为：https://henrymao2004.github.io/agent-over-correction/
cs.AI / 103 / 2609.32631
SWE-MILE: Asynchronous Potential-Induced Milestone Credit Assignment for Long-Horizon Software Engineering Agents
SWE-MILE：用于长时间跨度软件工程代理的异步潜在诱导里程碑信用分配
Abstract
Long-horizon software engineering (SWE) agents trained with reinforcement learning with verifiable rewards (RLVR) typically receive only terminal outcome supervision, making it difficult to distinguish productive actions from redundant exploration or functional regressions. We propose SWE-MILE, an asynchronous potential-induced milestone credit assignment framework that derives fine-grained process supervision from workflow runtime, without auxiliary reward models or external evaluators. SWE-MILE quantifies task-relevant file exposure and test-state alignment as navigation and verification potentials, respectively. Differences in these potentials attribute milestone progress and regressions to individual actions, while discounted backward credit propagates supervision to preceding steps. To efficiently acquire intermediate verification states, SWE-MILE further introduces asynchronous shadow probing, which replays repository-changing actions in an isolated sandbox and runs verification in parallel with the agent's primary interaction, largely hiding verification latency. The resulting process credit augments terminal outcome advantages and provides informative learning signals. Experiments on two representative long-horizon SWE tasks demonstrate substantial improvements in agent performance, highlighting workflow runtime signals as a practical source of process supervision for long-horizon SWE agents.
Chinese Translation
长时间跨度的软件工程（SWE）代理通过可验证奖励的强化学习（RLVR）进行训练，通常仅接受终端结果监督，这使得区分有效行动与冗余探索或功能回归变得困难。我们提出了SWE-MILE，一种异步潜在诱导里程碑信用分配框架，该框架从工作流运行时中推导出细粒度的过程监督，而无需辅助奖励模型或外部评估者。SWE-MILE将任务相关的文件暴露和测试状态对齐量化为导航和验证潜力。潜力之间的差异将里程碑的进展和回归归因于个别行动，而折扣的反向信用则将监督传播到前面的步骤。为了高效获取中间验证状态，SWE-MILE进一步引入了异步影子探测，它在一个隔离的沙箱中重放更改仓库的操作，并与代理的主要交互并行运行验证，从而在很大程度上隐藏了验证延迟。最终的过程信用增强了终端结果的优势，并提供了有意义的学习信号。在两个代表性的长时间跨度SWE任务上的实验表明，代理性能有显著改善，突显了工作流运行时信号作为长时间跨度SWE代理的实际过程监督来源。
cs.AI / 104 / 2609.32638
Can Open-Weight Large Language Models (LLMs) Simulate Human Survey Populations? A Cross-Instrument Calibration Study
开放权重的大型语言模型（LLMs）能否模拟人类调查人群？一项跨工具校准研究
Abstract
Large language models (LLMs) are increasingly used to generate synthetic survey respondents and digital twins of real people, but whether their output preserves real human statistical structure, rather than surface plausibility, remains unresolved, and most existing evidence comes from proprietary models rather than open-weight ones. We evaluate three open-weight LLM families on a cross-instrument calibration task: conditioning personas on real respondents' verbatim answers to one psychometric instrument and measuring them on a second, construct-distance-controlled instrument, checked against a 2,058-person human panel. Across a 139-pair grid, the simulated cross-instrument correlation tracks the real human correlation at r = 0.70 - 0.73 in every model, driven mainly by correct sign rather than precise magnitude and concentrated in pairs of moderate construct distance. A correlation of this magnitude, obtained from untuned open-weight models conditioned only on individual-level survey data, is a substantively encouraging result for LLM-based behavioral simulation and digital-twin applications: specific model families and releases already reproduce a meaningful share of real human cross-instrument structure without any fine-tuning. This capability does not, however, improve monotonically across model releases: on a matched panel, the newest of three tested Llama releases performs worst on two of three headline metrics, so realizing its promise in practice requires release-specific, distance-aware verification rather than a one-time benchmark.
Chinese Translation
大型语言模型（LLMs）越来越多地被用于生成合成调查受访者和真实人类的数字双胞胎，但它们的输出是否保留了真实人类的统计结构，而不仅仅是表面的合理性，仍然没有定论，而且现有的大多数证据来自于专有模型，而非开放权重模型。我们在一项跨工具校准任务中评估了三种开放权重的LLM系列：将角色设定为真实受访者对一个心理测量工具的逐字回答，并在第二个构念距离控制的工具上进行测量，与一个包含2,058人的人类小组进行对比。在139对网格中，模拟的跨工具相关性在每个模型中都与真实人类相关性保持一致，相关系数为r = 0.70 - 0.73，主要由正确的符号驱动，而非精确的大小，并集中在构念距离适中的对上。这个幅度的相关性是从仅基于个体级别调查数据的未调优开放权重模型中获得的，对于基于LLM的行为模拟和数字双胞胎应用来说是一个实质上令人鼓舞的结果：特定的模型系列和版本已经在没有任何微调的情况下再现了真实人类跨工具结构的一个重要部分。然而，这种能力并不是在模型发布中单调增加的：在一个匹配的小组中，三种测试的Llama版本中最新的一个在三项主要指标中的两项表现最差，因此在实践中实现其潜力需要特定版本的、考虑距离的验证，而不是一次性的基准测试。
cs.AI / 105 / 2609.32643
Business Compromise Detection with Agentic AI and LLM-driven Knowledge Discovery
基于代理人工智能和大型语言模型驱动的知识发现的商业妥协检测
Abstract
Detecting compromised business ad accounts is a challenge in digital advertising, as attackers exploit hijacked accounts to launch fraudulent campaigns. Large Language Model (LLM) agents show promise for integrity enforcement, but hallucinated mistakes on hard cases create business friction. In a study we find the autonomous agent is a strong, recall-heavy signal extractor but an unreliable final arbiter, conceding precision on ambiguous decisions. We therefore keep the agent as an investigator that emits a structured, interpretable signal vector, and delegate the verdict to a neuro-symbolic stage: symbolic rules discovered by Inductive Logic Programming (FOIL-IE), a Na\"ive Bayes calibration layer, and a data-tuned contradiction layer. Evaluating on a compromise-over-sampled population and a realistic low-prevalence sample with subject-matter-expert labels, this arbiter substitution raises MCC from 0.295 to 0.435 ({\Delta}MCC +0.139, 95% CI [+0.026, +0.245], p=0.018, paired bootstrap), lifting precision from 0.250 to 0.446 (1.8x) at a recall cost (0.920 to 0.660). Benchmarked under identical conditions, it also edge tree ensembles (0.386).The rules encode domain w labels while remaininginterpretable and auditable.
Chinese Translation
检测被妥协的商业广告账户是数字广告中的一项挑战，因为攻击者利用被劫持的账户发起欺诈性活动。大型语言模型（LLM）代理在完整性执行方面显示出潜力，但在复杂案例中出现的幻觉错误会造成商业摩擦。在一项研究中，我们发现自主代理是一个强大的、以召回为主的信号提取器，但在模糊决策上是一个不可靠的最终裁决者，牺牲了精确度。因此，我们将代理作为一个调查者，发出结构化、可解释的信号向量，并将裁决委托给一个神经符号阶段：通过归纳逻辑编程（Inductive Logic Programming, FOIL-IE）发现的符号规则、一个朴素贝叶斯校准层和一个数据调整的矛盾层。在对一个过采样的妥协人群和一个具有主题专家标签的现实低流行样本进行评估时，这种裁决者替代将MCC从0.295提高到0.435（ΔMCC +0.139, 95% CI [+0.026, +0.245], p=0.018, 配对自助法），将精确度从0.250提升至0.446（1.8倍），同时召回率有所下降（从0.920降至0.660）。在相同条件下进行基准测试时，它也优于树集成模型（0.386）。这些规则在保持可解释性和可审计性的同时编码了领域标签。
cs.AI / 106 / 2609.32645
From Scene Graphs to Answers: Selective Neuro-Symbolic Reasoning for Autonomous Driving
从场景图到答案：用于自动驾驶的选择性神经符号推理
Abstract
Autonomous-driving question answering requires reasoning over structured scene information, yet existing vision-language approaches largely delegate heterogeneous reasoning operations to a single neural inference process. We argue that this uniform strategy overlooks a fundamental distinction: some queries admit exact symbolic solutions, while others require semantic interpretation. We introduce a query-adaptive neuro-symbolic reasoning framework that explicitly allocates computation according to the nature of the query. At its core is a hierarchical Spatiotemporal Scene Graph (STSG) that separates persistent object identities from frame-specific states and represents spatial relations and temporal transitions as explicit directed structures. Given a query, a symbolic executor first attempts to resolve it through exact graph operations; only when symbolic execution abstains is an LLM invoked for semantic reasoning. For these unresolved queries, query-conditioned graph retrieval and evidence filtering preserve relation direction, temporal locality, and object semantics, providing the LLM with compact and verified task-relevant evidence. This design shifts the role of the LLM from a universal reasoning engine to a targeted semantic reasoner, while allowing deterministic computation to be handled exactly and efficiently. We evaluate the framework on 5,916 NuScenes-QA questions across all ten scenes of nuScenes v1.0-mini under an oracle-perception setting. The complete system achieves 80.63 percent overall accuracy with GPT-5.4-mini, improving over the corresponding LLM-only configuration by 5.48 percentage points; with DeepSeek-V4-Flash, the improvement reaches 6.64 points. The largest gains occur on counting questions, with improvements of 10.20 and 12.61 points, respectively. These results show that selective reasoning improves both accuracy and inference efficiency.
Chinese Translation
自动驾驶问答需要对结构化场景信息进行推理，但现有的视觉-语言方法在很大程度上将异构推理操作委托给单一的神经推理过程。我们认为，这种统一策略忽视了一个基本区别：某些查询可以通过精确的符号解决方案来解决，而其他查询则需要语义解释。我们提出了一种查询自适应的神经符号推理框架，该框架根据查询的性质明确分配计算资源。其核心是一个分层时空场景图（Spatiotemporal Scene Graph, STSG），该图将持久对象身份与帧特定状态分开，并将空间关系和时间过渡表示为明确的有向结构。给定一个查询，符号执行器首先尝试通过精确的图操作来解决该查询；只有在符号执行无法解决时，才会调用大型语言模型（LLM）进行语义推理。对于这些未解决的查询，基于查询的图检索和证据过滤保持关系方向、时间局部性和对象语义，为LLM提供紧凑且经过验证的任务相关证据。这种设计将LLM的角色从通用推理引擎转变为有针对性的语义推理器，同时允许确定性计算被精确且高效地处理。我们在nuScenes v1.0-mini的十个场景中对5,916个NuScenes-QA问题评估了该框架，采用了oracle感知设置。完整系统在使用GPT-5.4-mini时实现了80.63%的整体准确率，相较于相应的仅LLM配置提高了5.48个百分点；使用DeepSeek-V4-Flash时，提升达到了6.64个百分点。最大增益出现在计数问题上，分别提高了10.20和12.61个百分点。这些结果表明，选择性推理提高了准确性和推理效率。
cs.AI / 107 / 2609.32652
Prediction Limits and Koopman Closure of Geometry-Induced Soft State Abstractions
几何诱导软状态抽象的预测极限与库普曼闭包
Abstract
We study when geometry-induced soft state abstractions admit accurate finite-dimensional linear dynamics. Each state is represented by simplex-valued coordinates obtained from class-specific Kernel Affine Hull Machine (KAHM) reconstruction scores, and a matrix is used to predict the next-state coordinates. Our main result is a computable lower confidence bound on the minimum root-mean-square prediction error over all matrices satisfying a prescribed spectral-norm limit. The bound combines within-class variation of successor coordinates with the deviation of soft coordinates from their one-hot reference labels, and can be evaluated from independent state-successor pairs without fitting a prediction matrix. For fixed coordinates and evaluation distribution, the certificate converges almost surely to a population lower bound as the sample size grows; any tolerance below this limit is eventually certified unattainable. Reconstruction-score margins further control the soft-to-hard assignment error. Under deterministic dynamics and exact coordinate closure, eigenvectors of the closure matrix and its reduced transpose induce Koopman and adjoint Koopman eigenfunctions, respectively. A four-state KAHM construction shows that identical soft coordinates can permit exact closure under one dynamics map yet force positive prediction error under another. Experiments on Duffing, Van der Pol, CartPole, MountainCar, and Acrobot compare direct soft-coordinate prediction with state-space DMD/EDMD baselines and report prediction, representation-variation, and spectral diagnostics. The benchmarks assess fitted models but do not numerically evaluate the exclusion certificate.
Chinese Translation
我们研究几何诱导的软状态抽象何时能够容纳准确的有限维线性动态。每个状态由从特定类别的核仿射外壳机（Kernel Affine Hull Machine, KAHM）重建得分获得的单纯形值坐标表示，并使用一个矩阵来预测下一个状态的坐标。我们的主要结果是一个可计算的下置信界，针对满足规定谱范数限制的所有矩阵的最小均方根预测误差。该界限结合了后继坐标的类内变异性与软坐标相对于其独热参考标签的偏差，并且可以从独立的状态-后继对中评估，而无需拟合预测矩阵。在固定坐标和评估分布下，随着样本量的增加，证书几乎肯定会收敛到一个总体下界；任何低于此限制的容忍度最终都会被证明是不可达的。重建得分边际进一步控制软到硬的分配误差。在确定性动态和精确坐标闭包下，闭包矩阵及其简化转置的特征向量分别诱导库普曼（Koopman）和伴随库普曼特征函数。一个四状态的KAHM构造表明，相同的软坐标可以在一个动态映射下允许精确闭包，但在另一个动态下却强制产生正预测误差。对Duffing、Van der Pol、CartPole、MountainCar和Acrobot的实验比较了直接的软坐标预测与状态空间DMD/EDMD基线，并报告了预测、表示变异性和谱诊断。基准评估了拟合模型，但未对排除证书进行数值评估。
cs.AI / 108 / 2609.32656
MixBench-TS: A Multivariate Time Series Forecasting Benchmark Where Channel Mixing Pays Off
MixBench-TS：一个多元时间序列预测基准，其中通道混合带来了收益
Abstract
Multivariate Time Series Forecasting (MTSF) models that mix information across channels assume that the past of one channel carries information about the future of another. Yet they are evaluated on a small fixed set of standard datasets whose cross-channel structure is rarely examined. We ask two questions: "How can we reliably measure lagged, non-linear, and joint coupling in MTSF datasets?" and "Do the standard datasets actually have such coupling?" To answer the first, we test four candidate measures on synthetic datasets with planted ground-truth coupling: Granger Causality (GC), Transfer Entropy (TE), lagged Mutual Information (MI), and the CD gain, a model-based measure we introduce that compares a channel-dependent (CD) model to its channel-independent (CI) variant. Only lagged MI and the CD gain recover every planted coupling. For the second question, the answer is a definite no, as the standard datasets have a median of only 23% lagged-coupled channel pairs and a median CD gain of -4.9%, compared to 78% and +4.6% on chaotic ODE systems. We therefore propose MixBench-TS, a benchmark of 10 real-world datasets with a median of 55.5% lagged-coupled pairs and a median CD gain of +1.7%. Across six state-of-the-art models tuned under one protocol, CI models win on 10/10 (MSE) and 8/10 (MAE) standard datasets, but on only 3/10 and 2/10 MixBench-TS datasets. We recommend using our benchmark for evaluating new CD models. Moreover, we propose profiling new datasets with lagged MI and the CD gain before using them to evaluate multivariate models. Code and data are available at https://anonymous.4open.science/r/mixbench-ts-B027.
Chinese Translation
多元时间序列预测（MTSF）模型通过跨通道混合信息，假设一个通道的过去携带着另一个通道未来的信息。然而，它们的评估基于一小部分固定的标准数据集，而这些数据集的跨通道结构很少被研究。我们提出两个问题：“我们如何可靠地测量MTSF数据集中的滞后、非线性和联合耦合？”以及“标准数据集是否真的存在这样的耦合？”为了回答第一个问题，我们在具有植入真实耦合的合成数据集上测试了四种候选测量方法：格兰杰因果关系（Granger Causality, GC）、传递熵（Transfer Entropy, TE）、滞后互信息（lagged Mutual Information, MI）和我们引入的基于模型的测量——通道依赖（Channel-Dependent, CD）增益，该方法将通道依赖模型与其通道独立（Channel-Independent, CI）变体进行比较。只有滞后互信息和CD增益能够恢复每一个植入的耦合。对于第二个问题，答案是明确的否，因为标准数据集的滞后耦合通道对的中位数仅为23%，而中位数CD增益为-4.9%，相比之下，混沌常微分方程（ODE）系统的滞后耦合通道对的中位数为78%，CD增益为+4.6%。因此，我们提出了MixBench-TS，这是一个包含10个真实世界数据集的基准，其滞后耦合对的中位数为55.5%，CD增益的中位数为+1.7%。在一个协议下调优的六个最先进模型中，CI模型在10/10（均方误差，MSE）和8/10（平均绝对误差，MAE）标准数据集上获胜，但在MixBench-TS数据集上仅在3/10和2/10上获胜。我们建议使用我们的基准来评估新的CD模型。此外，我们建议在使用新的数据集评估多元模型之前，先用滞后互信息和CD增益对其进行分析。代码和数据可在https://anonymous.4open.science/r/mixbench-ts-B027获取。
cs.AI / 109 / 2609.32657
World Models with Predictable Long-Horizon Marginals
具有可预测长时间边际的世界模型
Abstract
Accurate one-step predictions do not ensure that a world model's rollouts retain the data distribution. We make the model's decoded stationary law explicit by learning a decoder of a fixed Gaussian reference and constraining the behaviour-averaged transition to preserve that reference. For controlled systems, a joint transition uses a conditional action chart to preserve behaviour occupancy without requiring invariance at each fixed action. Joint state--action rotations and parallel Gaussian noise give an exactly preserving transition with a tractable conditional density. We derive an absolute convergence bound from finite initialization banks and control departure from the reference through conditional action-space divergence. Across $216$ fitted pixel checkpoints on twelve control tasks, the occupancy model with a reference mixture retains every evaluated chain at $10^5$ steps in all $36$ task--seed cells, with a rollout-minus-reference energy-statistic difference of $-0.0002\pm0.0003$ (training-seed standard error). Each of the four nonpreserving comparison arms loses chains, although the Gaussian arm is more accurate at ten steps. An offline DreamerV3 reference also achieves better short-horizon accuracy. These results distinguish three properties of a world model: the distribution it approaches, the rate of approach, and the conditional dynamics it learns.
Chinese Translation
准确的一步预测并不能确保世界模型的展开保持数据分布。我们通过学习固定高斯参考的解码器并约束行为平均转移来明确模型的解码平稳法则，以保持该参考。对于受控系统，联合转移使用条件动作图来保持行为占用，而不需要在每个固定动作下保持不变。联合状态-动作旋转和并行高斯噪声提供了一个完全保持的转移，具有可处理的条件密度。我们从有限初始化库中推导出绝对收敛界限，并通过条件动作空间的发散控制与参考的偏离。在十二个控制任务的216个拟合像素检查点中，具有参考混合的占用模型在所有36个任务-种子单元中以10^5步保留了每个评估链，展开与参考的能量统计差异为-0.0002±0.0003（训练种子标准误差）。尽管高斯臂在十步时更准确，但四个非保持比较臂均失去了链。离线DreamerV3参考也实现了更好的短时间准确性。这些结果区分了世界模型的三个特性：它接近的分布、接近的速率以及它学习的条件动态。
cs.AI / 110 / 2609.32658
Contract Memory Compiler: Resolve, Then Traverse
合同记忆编译器：先解决，再遍历
Abstract
External memory lets language-model agents answer questions about histories too long for the answer model's context window. Updates create a harder problem than retrieving a recent fact: changing one relation can redirect a multi-hop question to records about an entity absent from the question. We study this update-dependent evidence selection problem and introduce the Contract Memory Compiler (CMC). Before seeing a question, CMC uses a language model to identify relations in the history and record where each one was stated. It applies later updates to determine the current relations, follows them from entities named in the question, and passes the corresponding original records to the answer model in one call. Thus the current state determines which evidence is read, rather than merely refreshing values in a previously selected context. To the best of our knowledge, CMC achieves state-of-the-art multi-hop accuracy on FactConsolidation, reaching 78.25% overall and 61.0% at 262K. With the extracted relations and answer model held fixed, selecting evidence before resolving updates reduces multi-hop accuracy to 21.50%. We also introduce MQuAKE-MemStream, a derived dataset of ordered memory streams built from MQuAKE-Remastered counterfactual cases.
Chinese Translation
外部记忆使得语言模型代理能够回答关于历史的提问，这些历史超出了回答模型的上下文窗口。更新带来了比检索最近事实更复杂的问题：改变一个关系可能会将多跳问题重定向到关于一个在问题中缺失的实体的记录。我们研究了这一依赖更新的证据选择问题，并引入了合同记忆编译器（Contract Memory Compiler, CMC）。在看到问题之前，CMC利用语言模型识别历史中的关系，并记录每个关系的表述位置。它应用后续的更新来确定当前的关系，从问题中提到的实体出发追踪这些关系，并在一次调用中将相应的原始记录传递给回答模型。因此，当前状态决定了读取哪些证据，而不仅仅是刷新先前选择的上下文中的值。根据我们所知，CMC在FactConsolidation上达到了最先进的多跳准确率，总体达到78.25%，在262K时为61.0%。在提取的关系和回答模型保持不变的情况下，在解决更新之前选择证据会将多跳准确率降低到21.50%。我们还引入了MQuAKE-MemStream，这是一个基于MQuAKE-Remastered反事实案例构建的有序记忆流的衍生数据集。
cs.AI / 111 / 2609.32667
Learning from a Thoughtful Teacher: Adaptive On-Policy Self-Distillation for Mathematical Reasoning
向深思熟虑的教师学习：用于数学推理的自适应在线自蒸馏
Abstract
On-policy self-distillation (OPSD) trains a question-only student with token-level feedback from a teacher given training-only privileged information (PI). OPSD therefore provides dense, on-policy supervision, and is free of a larger external teacher, but its effectiveness rests on how PI is designed and utilized. Our preliminary diagnostics suggest a significant gap between teacher utility and student learnability, where a small fraction of high-disagreement tokens dominate the distillation signal, and short teacher continuations at these positions further expose more explicit PI leakage than transferable correction cues, indicating a strong intent on injecting PI-conditioned shortcuts. We propose Adaptive On-Policy Self-Distillation (AOPSD), which adapts what information the teacher receives and how strongly its feedback influences learning. AOPSD encodes each solution as a reasoning DAG, orders problems by the student's evolving capability, and reveals only the affordable subgraph and its next frontier as PI. For high-disagreement tokens, AOPSD utilizes short teacher continuations as probes to encourage useful guidance while mitigating PI-conditioned shortcuts among teacher supervisions. On HMMT25, AIME24, AIME25, and BRUMo25, AOPSD achieves 72.5% Pass@8, which is 6.7 percentage points above OPSD and 4.2 above the strongest competing baseline while reducing 15 percentage points of training time at lower cost.
Chinese Translation
在线自蒸馏（On-policy self-distillation，OPSD）通过教师提供的仅限训练的特权信息（Privileged Information，PI）对仅包含问题的学生进行训练，利用令牌级反馈。因此，OPSD提供了密集的在线监督，并且不依赖于更大规模的外部教师，但其有效性取决于PI的设计和利用。我们的初步诊断表明，教师的效用与学生的可学习性之间存在显著差距，其中少量高争议的令牌主导了蒸馏信号，而这些位置的短教师延续进一步暴露了比可转移的修正线索更明显的PI泄漏，表明强烈意图注入基于PI的捷径。我们提出了自适应在线自蒸馏（Adaptive On-Policy Self-Distillation，AOPSD），该方法调整教师接收的信息以及其反馈对学习的影响程度。AOPSD将每个解决方案编码为推理有向无环图（DAG），根据学生不断发展的能力对问题进行排序，并仅揭示可承受的子图及其下一个前沿作为PI。对于高争议令牌，AOPSD利用短教师延续作为探针，以鼓励有用的指导，同时减轻教师监督中的基于PI的捷径。在HMMT25、AIME24、AIME25和BRUMo25上，AOPSD实现了72.5%的Pass@8，超出OPSD 6.7个百分点，超出最强竞争基线4.2个百分点，同时在较低成本下减少了15个百分点的训练时间。
cs.AI / 112 / 2609.32669
Refinement Symmetry in Multimodal Transformers
多模态变换器中的细化对称性
Abstract
Attention weights depend on token counts, which change with the representation of a signal. We study refinement symmetry: splitting a representation while preserving content, position, visible context, and total mass should preserve its contribution. Building on proportional and quadrature attention, we show that split invariance forces the local mass factor to be linear for any fixed positive attention kernel, provided that factor is nondecreasing. For changed representations, a physical coupling bounds attention error by separating feature change from weight reallocation. In Qwen2.5-Omni-7B, duplicating half the visual tokens threefold changes 255 of 3,586 MVBench answers under standard attention; measure weighting preserves every answer under matched visibility. Under natural frame resampling, it reduces distributional drift. At twofold merging of a frozen video encoding, a five-seed evaluation shows an all-partition-correct accuracy gain of 1.04 percentage points over global count weighting (average group mass) and 0.93 points over standard attention. The advantage over global count also holds on WorldSense but depends on the compression budget. The result is a representation principle with a measured benefit in robustness across partitions.
Chinese Translation
注意力权重依赖于标记数量，这些数量随着信号表示的变化而变化。我们研究细化对称性：在保持内容、位置、可见上下文和总质量的情况下拆分表示应该保持其贡献。基于比例和四分之一注意力，我们展示了拆分不变性强制局部质量因子对于任何固定的正注意力核是线性的，前提是该因子是非递减的。对于变化的表示，物理耦合通过将特征变化与权重重新分配分开来限制注意力误差。在 Qwen2.5-Omni-7B 中，三倍复制一半视觉标记在标准注意力下改变了 3,586 个 MVBench 答案中的 255 个；测量加权在匹配可见性下保留了每个答案。在自然帧重采样下，它减少了分布漂移。在对冻结视频编码进行两倍合并时，五种种子的评估显示，相较于全局计数加权（平均组质量），所有分区的正确准确率提高了 1.04 个百分点，相较于标准注意力提高了 0.93 个百分点。相较于全局计数的优势在 WorldSense 上也成立，但依赖于压缩预算。该结果是一个具有测量效益的表示原则，在各个分区中展现出更强的鲁棒性。
cs.AI / 113 / 2609.32670
What Would Falsify It? A Variable Specific Evidence Standard for Mechanistic Claims About Self Explanation
什么会使其失效？关于自我解释的机制性主张的变量特定证据标准
Abstract
When a language model explains an answer it has already given, does it reuse the computation that produced the answer or reconstruct a story from the answer alone? Attribution, transportability and recoverability are each compatible with causal use without establishing it. We propose an evidence standard: pair each positive statistic with a variable specific null that removes the tested variable's identity while matching relevant nuisance dimensions as far as possible, and audit unmatched dimensions. We apply this standard to a known cause. A cue naming a wrong option raises the rate of choosing that option by 64 to 68 percentage points across three models. Explanations mention the cue in 1.8 percent of items or fewer in three of four models tested. Three estimator classes yield favorable statistics, but none establishes causal sensitivity to the cue contrast under its own control in the three-model analysis. In the strongest case, a recovered cue direction reaches $R^2$ of 0.95 and exceeds a geometry matched random direction in all three seeds, while a direction fitted by the same pipeline with cue labels scrambled reproduces 61 to 76 percent of its effect at comparable realized edit magnitude. A fourth model passes one interchange endpoint, but unequal edit magnitudes and a contrast that changes both cue identity and cue-answer agreement limit its interpretation. These experiments leave causal access unresolved. They establish an evidentiary requirement: favorable mechanistic statistics must survive controls for variable identity and nuisance structure. Reusable controls separate generic from identity specific transport effects, fit null directions with scrambled labels, and audit realized intervention magnitudes.
Chinese Translation
当一个语言模型解释它已经给出的答案时，它是重用产生答案的计算，还是仅从答案重构一个故事？归因、可传输性和可恢复性在因果使用中是兼容的，但并未建立因果关系。我们提出了一种证据标准：将每个正统计量与一个变量特定的零假设配对，该零假设在尽可能匹配相关干扰维度的同时去除被测试变量的身份，并审计不匹配的维度。我们将这一标准应用于一个已知的因果关系。一个命名错误选项的提示在三个模型中将选择该选项的比例提高了64到68个百分点。在测试的四个模型中，解释在1.8%的项目中或更少提到该提示。三类估计器产生了有利的统计数据，但在三模型分析中，没有一个在其自身控制下建立对提示对比的因果敏感性。在最强的案例中，恢复的提示方向达到$R^2$为0.95，并在所有三个种子中超过几何匹配的随机方向，而通过相同管道拟合的、标签被打乱的方向在可比的实现编辑幅度下再现了61%到76%的效果。第四个模型通过了一个互换端点，但不等的编辑幅度以及同时改变提示身份和提示-答案一致性的对比限制了其解释。这些实验未能解决因果访问问题。它们确立了一项证据要求：有利的机制性统计数据必须在控制变量身份和干扰结构下存活。可重用的控制将通用效应与身份特定的传输效应分开，拟合带有打乱标签的零方向，并审计实现的干预幅度。
cs.AI / 114 / 2609.32674
Expected Reasoning-Step Return Unifies On-Policy Learning from Rewards and Teachers
期望推理步骤回报统一了基于奖励和教师的在线学习
Abstract
On-policy reasoning models can learn from task rewards or teacher signals, but these sources differ in form and can favor conflicting updates, leaving unclear which should guide a given reasoning action. We introduce \textbf{Expected Reasoning-Step Return (ERSR)}, which treats semantic reasoning steps as macro-actions and uses Monte Carlo student-policy rollouts to estimate the expected final task reward of student-generated and teacher-proposed actions in a common return space for step-level comparison. ERSR analysis reveals an outcome-dependent asymmetry: student actions are more beneficial than teacher replacements on successful trajectories, whereas teacher replacements become more beneficial on failed trajectories. We further show that student answer-probe gains track student-step ERSR utility and distinguish beneficial from harmful reasoning steps. Based on these findings, we propose \textbf{Return-Referenced On-Policy Learning (R$^2$OPL)}, which reinforces student reasoning on successful trajectories and distills teacher signals on failed ones, while using group success rate for difficulty scaling and student-probe gains for step-level modulation. Experiments across reasoning benchmarks and teacher--student configurations show that R$^2$OPL consistently outperforms strong baselines. ERSR training dynamics further show that R$^2$OPL jointly exploits substantial utility from both reward- and teacher-side signals, whereas existing hybrids often leave substantial residual utility in one branch.
Chinese Translation
在线推理模型可以从任务奖励或教师信号中学习，但这些来源在形式上有所不同，可能导致相互冲突的更新，因而不清楚应由哪个来指导特定的推理行动。我们引入了 extbf{期望推理步骤回报（Expected Reasoning-Step Return, ERSR）}，将语义推理步骤视为宏观动作，并使用蒙特卡洛学生策略回放来估计在共同回报空间中学生生成的和教师提出的动作的预期最终任务奖励，以便进行步骤级比较。ERSR分析揭示了一种依赖于结果的非对称性：在成功轨迹上，学生的动作比教师的替代更有利，而在失败轨迹上，教师的替代则变得更有利。我们进一步展示，学生的答案探测增益与学生步骤的ERSR效用相匹配，并区分出有益和有害的推理步骤。基于这些发现，我们提出了 extbf{回报参考的在线学习（Return-Referenced On-Policy Learning, R$^2$OPL）}，该方法在成功轨迹上强化学生推理，并在失败轨迹上提炼教师信号，同时使用群体成功率进行难度缩放，并利用学生探测增益进行步骤级调节。在推理基准和教师-学生配置的实验中，R$^2$OPL始终优于强基线。ERSR训练动态进一步表明，R$^2$OPL共同利用了来自奖励和教师信号的显著效用，而现有的混合方法往往在一个分支中留下了显著的剩余效用。
cs.AI / 115 / 2609.32677
When Better Gets Worse: Improvement Fidelity for Self-Improving Agents in Adaptive Worlds
当更好变得更糟：自我改进代理在适应性世界中的改进保真度
Abstract
Self-improving agents increasingly rely on proxy verifiers to choose policy updates, yet deployment can change the world in which those updates are evaluated. An update that looks better to the verifier can therefore become worse after deployment even when the verifier ranks policies well overall. We formalize this gap as Improvement Fidelity, which asks whether proxy improvements preserve the sign and ordering of deployment improvements over the updates an improvement process actually proposes. We show that global policy accuracy need not guarantee update fidelity: operator shift and deployment response can create update-level errors, while candidate margins determine whether those errors change the replacement decision. We introduce PIVOT-KG, a paired, decision-aware validator that allocates scarce high-fidelity evaluation according to the expected reduction in selection regret per unit cost. Across 90 held-out roots in Leduc, Kuhn, and Melting Pot, proxy and deployment optimal sets are disjoint in 51 cases. In an eight-candidate HighwayEnv stress test, PIVOT-KG reduces mean improvement-selection regret from 0.0435 under the exact Uniform validation rule to 0.0055 at the primary budget. Together, these results show why reliable self-improvement should evaluate proposed improvements in the worlds they induce, while providing a practical rule for allocating scarce deployment evidence when it can affect the replacement decision.
Chinese Translation
自我改进代理越来越依赖代理验证者来选择政策更新，然而部署可能会改变这些更新被评估的世界。因此，一个在验证者看来更好的更新在部署后可能变得更糟，即使验证者总体上对政策的排名是良好的。我们将这一差距形式化为改进保真度（Improvement Fidelity），它询问代理改进是否保留了更新过程中实际提出的改进的部署改进的符号和顺序。我们展示了全球政策准确性并不一定保证更新保真度：操作员转移和部署响应可能会产生更新级别的错误，而候选边际决定了这些错误是否会改变替换决策。我们引入了PIVOT-KG，这是一种成对的、决策感知的验证器，根据每单位成本的选择后悔预期减少分配稀缺的高保真评估。在Leduc、Kuhn和Melting Pot的90个保留根节点中，代理和部署最优集合在51个案例中是互斥的。在一个包含八个候选者的HighwayEnv压力测试中，PIVOT-KG将均值改进选择后悔从在精确的均匀验证规则下的0.0435减少到在主要预算下的0.0055。这些结果共同表明，可靠的自我改进应该在它们所诱导的世界中评估提出的改进，同时提供了一个在可能影响替换决策时分配稀缺部署证据的实用规则。
cs.AI / 116 / 2609.32685
PINNMorph: Evolving Online Adaptation Policies for Physics-Informed Neural Networks
PINNMorph：物理信息神经网络的在线适应策略演化
Abstract
Physics-informed neural networks (PINNs) provide a learning-based framework for solving partial differential equations (PDEs), yet their training behavior can change substantially throughout optimization. Residual distributions, gradient interactions, regional learning difficulty, and model-capacity requirements may evolve over time, while the network architecture and major training mechanisms are typically determined before training. We propose PINNMorph, an online PINN adaptation framework based on large language model (LLM)-guided policy evolution. PINNMorph maintains a population of state-conditioned adaptation policies that map execution diagnostics to controlled interventions over topology modification, additive representation augmentation, objective balancing, gradient handling, adaptive sampling, and optimizer-phase control. At each intervention opportunity, candidate programs are instantiated from the current policy population, selected according to the observed training state, and applied directly to the PINN under training. The resulting model inherits its existing parameters and training state and continues optimization along the same trajectory. Execution outcomes are subsequently used to evaluate interventions and evolve the policy population. Unlike pre-training architecture search or fixed adaptation rules, PINNMorph jointly adapts the current PINN and the policies governing its interventions using feedback from actual training. Experiments on 13 PDE benchmarks show that PINNMorph achieves lower solution errors than SA-PINN, ConFIG, RoPINN, HARMONIC, and PINNsAgent across all evaluated problems. Ablation studies further examine the effects of online adaptation, state-conditioned intervention selection, and execution-feedback-driven policy evolution.
Chinese Translation
物理信息神经网络（PINNs）为求解偏微分方程（PDEs）提供了一种基于学习的框架，但其训练行为在优化过程中可能会发生显著变化。残差分布、梯度交互、区域学习难度和模型容量需求可能会随着时间的推移而演变，而网络架构和主要训练机制通常在训练之前就已确定。我们提出了PINNMorph，这是一种基于大型语言模型（LLM）指导的策略演化的在线PINN适应框架。PINNMorph维护了一组状态条件适应策略，这些策略将执行诊断映射到对拓扑修改、附加表示增强、目标平衡、梯度处理、自适应采样和优化器阶段控制的受控干预。在每个干预机会中，候选程序从当前策略种群中实例化，根据观察到的训练状态进行选择，并直接应用于正在训练的PINN。生成的模型继承其现有参数和训练状态，并沿着相同的轨迹继续优化。执行结果随后用于评估干预效果并演化策略种群。与预训练架构搜索或固定适应规则不同，PINNMorph利用实际训练反馈共同适应当前PINN及其干预策略。在13个PDE基准测试上的实验表明，PINNMorph在所有评估问题中实现了比SA-PINN、ConFIG、RoPINN、HARMONIC和PINNsAgent更低的解误差。消融研究进一步考察了在线适应、状态条件干预选择和执行反馈驱动的策略演化的影响。
cs.AI / 117 / 2609.32687
Dude, Where's My State? Execution Information Requirements for Stateful Agents
伙计，我的状态在哪里？有状态代理的执行信息需求
Abstract
Long-running agents must preserve information that later steps depend on. We introduce the Execution Information Requirement (EIR), a lower bound on the information that must remain accessible for correct completion under specified task and access conditions. We develop LACUNA, a framework that generates tasks with known dependencies and varies information demand, retention, and recovery separately from the difficulty of individual operations. Across four models, restoring a missing result raises accuracy on affected recall steps to 100%, compared with 0% for equal-length irrelevant information. Sufficient storage alone does not ensure success: retention policies can discard required results, errors can propagate through later computations, and agents can stop before recovery is complete. We also introduce VESTIGE, which uses agent execution traces to construct semantic graphs and measure information demand for real tasks. Across 72,562 software-agent trajectories, VESTIGE reveals a steeper distance-related decline in solution-relevant rereading for failed runs (RR 0.951 per distance doubling), while adjusted peak demand alone is not associated with failure. Together, these contributions support evaluating whether agents preserve and recover the information their tasks require.
Chinese Translation
长时间运行的代理必须保留后续步骤所依赖的信息。我们引入了执行信息需求（Execution Information Requirement, EIR），这是在特定任务和访问条件下，必须保持可访问的信息的下限。我们开发了LACUNA，一个生成具有已知依赖关系的任务的框架，并且将信息需求、保留和恢复的变化与单个操作的难度分开。在四个模型中，恢复缺失结果使受影响的回忆步骤的准确率提高到100%，而对于等长的无关信息则为0%。仅有足够的存储空间并不能确保成功：保留策略可能会丢弃所需结果，错误可能会在后续计算中传播，代理可能会在恢复完成之前停止。我们还引入了VESTIGE，它利用代理执行轨迹构建语义图，并测量实际任务的信息需求。在72,562条软件代理轨迹中，VESTIGE揭示了失败运行中与解决相关的重读在距离相关的下降更为陡峭（每距离翻倍重读率为0.951），而调整后的峰值需求与失败并无关联。这些贡献共同支持评估代理是否保留和恢复其任务所需的信息。
cs.AI / 118 / 2609.32692
World Agent: Can Language Models Keep a World Running?
世界代理：语言模型能否维持一个世界的运行？
Abstract
World models are moving from generating realistic frames to generating playable worlds, yet whether a delivered world can keep running is not tested anywhere. Existing evaluations stop at generation, at delivery, or at single-step transitions, and each stops at a different point along the way. Correct local state transitions or intermediate outcomes do not guarantee a correctly organized causal event flow. We propose the world agent task, which moves the evaluation point of world generation from the moment of delivery to the continued operation that follows. In this task, a model is not asked to generate a world. It is held responsible for keeping the world running, which requires coordinating events and carrying forward their consequences to constrain subsequent evolution. We instantiate the task in WorldAgent-Benchmark with two complementary tracks. In the maintenance track, the model must ground the events of a continuous narrative into correct transitions of the explicit world state while respecting causal, temporal, and concurrency constraints. In the deduction track, the model must predict how the world will evolve under partial observations and act toward a goal. The maintenance track combines LLM-assisted semantic judgments with programmatic validation and scoring, while the deduction track is evaluated entirely programmatically. Individual judgments are auditable against world states and execution logs, and scores can be recomputed from the saved judgments and execution records. Across 8 models, scores decline steadily as pre-built structure is removed from the world, and causal-relation checking is the weakest component for every model. The benchmark makes the continued operation of a world measurable and distinguishes local completion from failures in event organization. Code and dataset will be released on https://github.com/HCPLab-SYSU/WorldAgent-Benchmark.
Chinese Translation
世界模型正从生成逼真的画面转向生成可玩的世界，但交付的世界是否能够持续运行尚未得到测试。现有的评估停留在生成、交付或单步过渡的阶段，每个阶段在不同的时间节点停止。正确的局部状态过渡或中间结果并不能保证因果事件流的正确组织。我们提出了世界代理任务，将世界生成的评估点从交付时刻转移到随后的持续操作。在这个任务中，模型并不要求生成一个世界，而是要负责维持世界的运行，这需要协调事件并推动其后果，以约束后续的演变。我们在WorldAgent-Benchmark中实例化了该任务，设立了两个互补的轨道。在维护轨道中，模型必须将连续叙事的事件与明确的世界状态的正确过渡相结合，同时遵循因果、时间和并发约束。在推理轨道中，模型必须在部分观察下预测世界将如何演变，并朝着目标采取行动。维护轨道结合了大型语言模型（LLM）辅助的语义判断与程序验证和评分，而推理轨道则完全通过程序进行评估。个别判断可以依据世界状态和执行日志进行审计，得分可以从保存的判断和执行记录中重新计算。在8个模型中，随着世界中预构建结构的移除，得分稳步下降，而因果关系检查是每个模型中最薄弱的环节。该基准使得世界的持续运行可测量，并区分局部完成与事件组织中的失败。代码和数据集将发布在 https://github.com/HCPLab-SYSU/WorldAgent-Benchmark。
cs.AI / 119 / 2609.32694
IGSD: Environment-Verified Hindsight Self-Distillation for Search Agents
IGSD：环境验证的回顾自蒸馏用于搜索代理
Abstract
On-policy self-distillation densifies agent training without external teachers: a policy conditioned on privileged hindsight provides step-level guidance for its own unprivileged rollouts. For search agents, however, hindsight can make the teacher prefer a query that does not improve retrieval from the student's state. Existing methods either distill this preference directly or filter it with model-internal scores, but neither strategy verifies the query's executed retrieval consequence. We propose Information-Gain-Gated Self-Distillation (IGSD), which verifies on-policy token proposals with environment feedback before distilling them. Treating each query token as a micro-action, IGSD completes the teacher's token proposal and the student's sampled token into matched queries and executes both from the same failed state with the same retriever. Shared counterfactual controls account for query-conditioned shifts in answer likelihood, so their difference, the executed paired information gain, provides a relative utility contrast for the retrieved documents. IGSD uses this contrast as a positive-only soft weight for candidate-pair distillation, while leaving the GRPO objective unchanged and confining verification to training. Across seven single-hop and multi-hop QA benchmarks, IGSD reaches macro-average exact-match accuracies of 42.8% and 47.0% with 3B and 7B policies, respectively, without inference-time verification. These results support environment-verified hindsight as an effective approach to reliable action-level supervision for search agents.
Chinese Translation
基于策略的自蒸馏在没有外部教师的情况下密集化代理训练：基于特权回顾的策略为其自身的非特权回滚提供逐步指导。然而，对于搜索代理来说，回顾可能会使教师偏好一个并未改善学生状态检索的查询。现有方法要么直接蒸馏这种偏好，要么通过模型内部评分进行过滤，但这两种策略都未验证查询的执行检索后果。我们提出了信息增益门控自蒸馏（Information-Gain-Gated Self-Distillation，IGSD），该方法在蒸馏之前通过环境反馈验证基于策略的标记提议。将每个查询标记视为微动作，IGSD将教师的标记提议和学生的采样标记完成为匹配查询，并从同一失败状态使用相同的检索器执行两者。共享的反事实控制考虑了查询条件下答案可能性的变化，因此它们的差异，即执行的配对信息增益，为检索文档提供了相对效用对比。IGSD将这种对比作为候选对蒸馏的仅正软权重，同时保持GRPO目标不变，并将验证限制在训练阶段。在七个单跳和多跳问答基准测试中，IGSD在3B和7B策略下分别达到了42.8%和47.0%的宏平均精确匹配准确率，而无需推理时验证。这些结果支持环境验证的回顾作为搜索代理可靠的行动级监督的有效方法。
cs.AI / 120 / 2609.32696
Flat-Consensus Diffusion for Robust Data Reshaping under Noisy Evaluator
在噪声评估者下的鲁棒数据重塑的平共识扩散
Abstract
Data shape determines how features are structured, how patterns are separated, and how distributions cover the underlying domain. Poor data shape can make models learn noise rather than generalizable structure. This paper studies robust feature-centric data reshaping: generating feature transformations that remain useful, stable, and reproducible under noisy evaluation and imperfect data conditions. We view reshaping operation sequence search as reward-guided diffusion generation, and robust reshaping as searching for regions in the latent reward landscape rather than isolated high-reward transformations. The key challenge is dual instability: noisy evaluators distort local reward guidance, while stochastic generative trajectories can converge to inconsistent solutions. We propose FCDiff, a flat-consensus diffusion framework that addresses both failures through a micro-macro decomposition. The micro layer replaces point-estimate reward guidance with Gaussian-smoothed, Monte Carlo averaged gradients, steering generation toward locally flat reward regions. The macro layer aggregates independently guided trajectories with a weighted Frechet-mean barycenter, selecting consensus-supported basins and filtering stochastic outliers. Across an 8-dataset headline cohort under heavy-tailed evaluator noise, FCDiff attains the best aggregate rank on lower-tail reliability and robustness against both search-based AutoFE and robustness-oriented generative baselines, with statistically significant accuracy gains over every generative baseline. Our results show that robust data reshaping requires searching for flat, consensus-supported regions rather than sharp single-trajectory optima.
Chinese Translation
数据形状决定了特征的结构、模式的分离以及分布对潜在领域的覆盖。较差的数据形状可能导致模型学习噪声而非可推广的结构。本文研究了以特征为中心的鲁棒数据重塑：在噪声评估和不完美数据条件下生成保持有用、稳定和可重复的特征变换。我们将重塑操作序列搜索视为奖励引导的扩散生成，而鲁棒重塑则被视为在潜在奖励景观中搜索区域，而非孤立的高奖励变换。关键挑战在于双重不稳定性：噪声评估者扭曲了局部奖励引导，而随机生成轨迹可能收敛到不一致的解决方案。我们提出了FCDiff，一个平共识扩散框架，通过微观-宏观分解来解决这两种失败。微观层用高斯平滑的蒙特卡洛平均梯度替代点估计的奖励引导，引导生成朝向局部平坦奖励区域。宏观层则通过加权的Frechet均值重心聚合独立引导的轨迹，选择共识支持的基底并过滤随机异常值。在一个包含8个数据集的重尾评估者噪声的主流队列中，FCDiff在下尾可靠性和对基于搜索的AutoFE及鲁棒性导向生成基线的鲁棒性方面获得了最佳的综合排名，并在每个生成基线上实现了统计显著的准确性提升。我们的结果表明，鲁棒数据重塑需要搜索平坦的、共识支持的区域，而非尖锐的单轨迹最优解。
cs.AI / 121 / 2609.32700
CAIRN: Dynamic Fact-Intent DAGs for Multi-Agent Exploration
CAIRN：用于多智能体探索的动态事实-意图有向无环图
Abstract
LLM-powered autonomous systems have demonstrated promising capabilities in mathematical reasoning, engineering, and cybersecurity. Yet how to organize these systems for effective, reliable, and sustained performance remains an open question. In this paper, we present CAIRN, a fact-intent-driven multi-agent paradigm for goal-directed exploration. CAIRN represents observations and planned investigations as a dynamic directed acyclic graph (DAG). A reasoner interprets facts to propose intents, which workers execute to produce new facts. Each intent references its supporting facts and defines a potential exploration branch. The persistent graph preserves goals, dependencies and findings across workers, supporting knowledge reuse and parallel exploration. The graph also makes execution trajectories traceable and auditable, providing a basis for human verification and intervention. We evaluate CAIRN across cybersecurity and mathematical reasoning tasks, examining task success, time to solution, and token consumption. DAG-based coordination can incur higher token costs with no observable performance gains on tasks that require little effort. However, on high-effort tasks (at least 1M tokens), we observe faster solutions in 76.5% of cases, with speedups of up to 3.08x. Moreover, as task effort increases, these time gains become more pronounced while relative token overhead declines, highlighting the potential of DAG-guided parallel exploration.
Chinese Translation
基于大型语言模型（LLM）的自主系统在数学推理、工程和网络安全方面展现了良好的能力。然而，如何组织这些系统以实现有效、可靠和持续的性能仍然是一个未解的问题。本文提出了CAIRN，一种基于事实和意图驱动的多智能体范式，用于目标导向的探索。CAIRN将观察结果和计划调查表示为动态有向无环图（DAG）。推理器解释事实以提出意图，工人执行这些意图以产生新的事实。每个意图引用其支持的事实，并定义一个潜在的探索分支。持久图保留了跨工人的目标、依赖关系和发现，支持知识重用和并行探索。该图还使执行轨迹可追溯和可审计，为人类验证和干预提供了基础。我们在网络安全和数学推理任务中评估CAIRN，考察任务成功率、解决时间和令牌消耗。基于DAG的协调在对需要较少努力的任务中可能会产生更高的令牌成本，但没有可观察的性能提升。然而，在高努力任务（至少1M令牌）中，我们观察到76.5%的案例中解决速度更快，速度提升最高可达3.08倍。此外，随着任务努力的增加，这些时间收益变得更加明显，而相对的令牌开销则下降，突显了DAG引导的并行探索的潜力。
cs.AI / 122 / 2609.32701
Despite Instructions: Frontier Agents Improvise Covert Channels at Test Time
尽管有指令：前沿代理在测试时即兴创造隐蔽通道
Abstract
In security-sensitive applications, language-model agents are often required to coordinate without disclosing confidential information. Yet repeated interactions may also let ordinary messages acquire shared private meaning. We study a repeated game with pairs of models in which the sender model observes one of four secret states and selects one of four summaries of the same public report, while the receiver model tries to infer the secret state. We find that model pairs can learn to communicate the secret using only one bit of feedback indicating whether the receiver inferred it correctly. This learning occurs during inference with fixed parameters and no supplied codebook or encoding examples. The effect also persists when agents generate their own free-form updates in a simulated incident-response task. Across ten independent games, pairs of GPT-5.6 Sol agents reach 98.8% final accuracy, compared with 25% chance, despite explicit instructions prohibiting disclosure and a monitor that screens each message without access to the agents' interaction histories. The same interactions that help agents cooperate can therefore allow confidential information to pass through messages intended for legitimate coordination.
Chinese Translation
在安全敏感的应用中，语言模型代理通常需要在不泄露机密信息的情况下进行协调。然而，反复的互动也可能使普通信息获得共享的私人含义。我们研究了一个重复博弈，其中一对模型中的发送者模型观察四个秘密状态之一，并选择同一公共报告的四个摘要之一，而接收者模型则试图推断秘密状态。我们发现，模型对可以学习仅通过一个比特的反馈来传达秘密，该反馈指示接收者是否正确推断了秘密状态。这种学习发生在推理过程中，参数固定且没有提供编码本或编码示例。当代理在模拟事件响应任务中生成自己的自由形式更新时，这种效果仍然存在。在十个独立的博弈中，成对的GPT-5.6 Sol代理达到了98.8%的最终准确率，而随机猜测的准确率仅为25%，尽管有明确的指令禁止泄露信息，并且有监控者筛查每条消息，但无法访问代理的互动历史。因此，帮助代理合作的相同互动也可能允许机密信息通过旨在合法协调的消息传递。
cs.AI / 123 / 2609.32704
CoWindow Attention: Full Causal Coverage Is a Collective Property
CoWindow 注意力：完全因果覆盖是一个集体属性
Abstract
FullAttn repeatedly exposes the complete causal history to every attention head, creating substantial redundant computation and memory traffic even with IO-efficient dense kernels. We introduce CoWA, a structured attention architecture that distributes access to the causal history across KV heads. All heads share near-diagonal and prefix-sink windows, while complementary long-range windows partition the remaining history. Their union provides full causal coverage although each head attends sparsely to distant tokens. This position-defined attention pattern requires no learned router or indexer, is used consistently during training and inference, and aligns with KV-head tensor parallelism. A window-matched ablation at 8K isolates the effect of complementary long-range allocation: CoWA with 100% collective coverage reaches 89.73% accuracy, compared with 89.97% for FullAttn, while duplicated long-range windows perform substantially worse. Across a broader controlled associative-recall comparison with matched token budgets, CoWA closely tracks FullAttn as the context grows, whereas other sparse patterns lose a substantial fraction of the associations. In an attention-operator benchmark at 128K tokens with tensor parallelism, CoWA reduces forward and backward latency during training by 7.4x and 8.6x and decoding latency during inference by 3.0x over FullAttn. Its per-rank peak operator memory matches FullAttn during training and is 7.6x lower during decoding. Across scaling-law training from 0.6B to 14B parameters, CoWA closely tracks FullAttn in perplexity while reducing total training FLOPs. The resulting 14B models and 32B models from separate continued training achieve comparable knowledge, reasoning, and long-context retrieval scores to FullAttn. These results show that full causal coverage can be a collective property of the head ensemble rather than a duplicated property of every head.
Chinese Translation
FullAttn 反复向每个注意力头暴露完整的因果历史，即使在 IO 高效的密集内核下，也会产生大量冗余的计算和内存流量。我们引入了 CoWA，一种结构化的注意力架构，它将对因果历史的访问分配到 KV 头上。所有头共享近对角和前缀汇聚窗口，而互补的长距离窗口则划分剩余的历史。它们的联合提供了完全的因果覆盖，尽管每个头对远程标记的关注是稀疏的。这种基于位置的注意力模式不需要学习的路由器或索引器，在训练和推理过程中保持一致，并与 KV 头张量并行性对齐。在 8K 的窗口匹配消融实验中，隔离了互补长距离分配的效果：CoWA 在 100% 集体覆盖下达到了 89.73% 的准确率，而 FullAttn 为 89.97%，而重复的长距离窗口表现明显更差。在更广泛的受控关联回忆比较中，随着上下文的增长，CoWA 与 FullAttn 的表现紧密跟踪，而其他稀疏模式则失去了大量的关联。在 128K 标记的注意力操作基准测试中，使用张量并行性，CoWA 在训练期间将前向和反向延迟分别减少了 7.4 倍和 8.6 倍，在推理期间将解码延迟减少了 3.0 倍，相较于 FullAttn。其每个排名的峰值操作内存在训练期间与 FullAttn 相匹配，而在解码期间则低 7.6 倍。在从 0.6B 到 14B 参数的扩展法训练中，CoWA 在困惑度上与 FullAttn 紧密跟踪，同时减少了总训练 FLOPs。由单独继续训练得到的 14B 模型和 32B 模型在知识、推理和长上下文检索得分上与 FullAttn 相当。这些结果表明，完全因果覆盖可以是头部集合的集体属性，而不是每个头的重复属性。
cs.AI / 124 / 2609.32712
MassAlloc Attention: Let Attention Allocate Its Own Compute
MassAlloc 注意力：让注意力分配其自身的计算
Abstract
FullAttn often assigns negligible normalized mass to much of the causal score space, yet dense kernels execute the complete post-score path after forming each QK tile. We introduce MALA, a fused attention primitive that preserves score access to every legal causal interaction and uses normalized contribution to allocate post-score computation. Forward uses its evolving online-softmax normalizer, while backward reuses the finalized normalizer to derive nested retained support using only standard attention state. A common tolerance governs training and inference, allowing for adaptive retention of the work. MALA reduces low-contribution post-score computation. A matched-work study at 8K isolates the benefit of distribution-adaptive allocation: under exactly matched total post-score work, MALA approaches a per-instance reference-mass oracle, with mean omitted mass of 0.0188% versus 0.0182%. Across context lengths from 1K to 32K tokens, the same tolerance maintains low output and gradient errors relative to the reference. Across a broader controlled associative-recall comparison, MALA closely tracks FullAttn as context grows, reaching 89.67% accuracy at 8K compared with 89.97% for FullAttn. In an attention-operator benchmark at 128K tokens with tensor parallelism, MALA reduces forward and backward latency during training by 2.2x and 3.0x and decoding latency during inference by 1.6x relative to FullAttn. Across scaling-law training from 0.6B to 14B parameters, MALA closely tracks FullAttn in perplexity while reducing total training FLOPs. The resulting 14B models and 32B models from separate continued training achieve comparable knowledge, reasoning, and long-context retrieval scores to FullAttn. These results indicate that allocating post-score computation according to normalized attention contributions can retain the evaluated capabilities of FullAttn while reducing attention computation.
Chinese Translation
FullAttn 通常会将可忽略的归一化质量分配给大部分因果得分空间，但密集核在形成每个 QK 瓦片后会执行完整的后得分路径。我们提出了 MALA，一种融合的注意力原语，它保留对每个合法因果交互的得分访问，并利用归一化贡献来分配后得分计算。前向传播使用其不断演变的在线软最大归一化器，而反向传播则重用最终的归一化器，仅使用标准注意力状态来推导嵌套的保留支持。一个共同的容忍度支配训练和推理，允许对工作的自适应保留。MALA 减少了低贡献的后得分计算。在 8K 的匹配工作研究中，隔离了分布自适应分配的好处：在完全匹配的总后得分工作下，MALA 接近每实例参考质量的神谕，平均省略质量为 0.0188% 对比 0.0182%。在从 1K 到 32K 令牌的上下文长度中，相同的容忍度在相对于参考的情况下保持低输出和梯度误差。在更广泛的控制关联回忆比较中，随着上下文的增长，MALA 与 FullAttn 紧密跟踪，在 8K 时达到 89.67% 的准确率，而 FullAttn 为 89.97%。在 128K 令牌的注意力操作基准测试中，利用张量并行性，MALA 在训练期间将前向和反向延迟分别减少了 2.2 倍和 3.0 倍，在推理期间将解码延迟减少了 1.6 倍，相对于 FullAttn。在从 0.6B 到 14B 参数的缩放法则训练中，MALA 在困惑度上与 FullAttn 紧密跟踪，同时减少了总训练 FLOPs。由单独的持续训练得到的 14B 模型和 32B 模型在知识、推理和长上下文检索分数上与 FullAttn 达到可比的水平。这些结果表明，根据归一化注意力贡献分配后得分计算可以保留 FullAttn 的评估能力，同时减少注意力计算。
cs.AI / 125 / 2609.32731
SkillVine: Agent Skill Evolution via Branching Exploration
SkillVine：通过分支探索实现代理技能演化
Abstract
Agent skills encapsulate reusable procedural knowledge that enables LLM agents to perform tasks, and they can be improved automatically using trajectories from interactions with the environment. This is the classic problem of skill evolution. Existing approaches predominately follow a linear evolution paradigm, in which updates are sequentially applied to the latest skill-library version. As a result, they inevitably fall into local optima, leaving many promising evolution paths unexplored. We propose SkillVine, an automatic skill-evolution framework that formulates skill evolution as a graph search problem and employs a branching exploration strategy. Equipped with a trunk-branch collaborative searching mechanism, an intelligent parent-node selector, and an adaptive-granularity update rule, SkillVine achieves a balance between exploration and exploitation. We evaluate SkillVine on 5 benchmarks with two LLMs. Results show that SkillVine discovers better skill-library versions along branches than along the linear trunk and achieves the best test performance in nine of ten benchmark-model combinations.
Chinese Translation
代理技能封装了可重用的程序知识，使得大型语言模型（LLM）代理能够执行任务，并且可以通过与环境的交互轨迹自动改进。这是技能演化的经典问题。现有的方法主要遵循线性演化范式，其中更新依次应用于最新的技能库版本。因此，它们不可避免地陷入局部最优，导致许多有前景的演化路径未被探索。我们提出了SkillVine，一个将技能演化公式化为图搜索问题的自动技能演化框架，并采用分支探索策略。SkillVine配备了主干-分支协同搜索机制、智能父节点选择器和自适应粒度更新规则，实现了探索与利用之间的平衡。我们在5个基准测试上对SkillVine进行了评估，使用了两种LLM。结果表明，SkillVine在分支上发现的技能库版本优于在线性主干上发现的版本，并且在十个基准模型组合中的九个中达到了最佳测试性能。
cs.AI / 126 / 2609.32749
Retrospective Distillation Attribution via Normalized Response Similarity
基于归一化响应相似性的回溯蒸馏归因方法
Abstract
Model distillation transfers capabilities through supervised fine-tuning (SFT) on teacher responses, often collected from commercial APIs, raising questions of model provenance. Existing distillation attribution methods have been largely evaluated on students immediately after the SFT step. However, a distilled model may undergo further SFT, preference optimization, or reinforcement learning before release, while an auditor may lack access to the pre-distillation checkpoint required by reference-based attribution. To close this gap, we propose SCOUT, an output-only method that aggregates recurring *syntactic patterns* into candidate profiles, filters low-contrast patterns, and calibrates student--candidate distances against inter-candidate distances. SCOUT supports attribution and abstention using only current texts, without model weights, token likelihoods, or historical checkpoints. Auditing publicly released descendants of distilled models spanning diverse post-training objectives, SCOUT consistently identifies the distillation source. Furthermore, tracing teacher-associated *syntactic signatures* along training trajectories reveals that they emerge during distillation and persist through subsequent preference optimization and reinforcement learning.
Chinese Translation
模型蒸馏通过对教师模型响应进行监督微调（Supervised Fine-Tuning，SFT）来传递能力，这些响应通常来自商业API，从而引发了模型来源的相关问题。现有的蒸馏归因方法大多在学生模型完成SFT步骤后立即进行评估。然而，蒸馏模型在发布前可能会经历进一步的SFT、偏好优化或强化学习，而审计者可能无法获得参考归因所需的蒸馏前检查点。为弥补这一差距，我们提出了SCOUT，一种仅依赖输出的方法，通过聚合重复出现的*句法模式*形成候选配置文件，过滤低对比度模式，并根据候选间距离校准学生模型与候选模型的距离。SCOUT支持仅使用当前文本进行归因和弃权，无需模型权重、token似然或历史检查点。在对涵盖多种后训练目标的公开发布蒸馏模型后代进行审计时，SCOUT始终能够准确识别蒸馏源。此外，沿训练轨迹追踪与教师相关的*句法特征*表明，这些特征在蒸馏过程中出现，并持续存在于后续的偏好优化和强化学习阶段。
cs.AI / 127 / 2609.32750
CUA-Sandbox: Efficient Environments for Computer-Use Agent Reinforcement Learning
CUA-Sandbox：计算机使用代理强化学习的高效环境
Abstract
Reinforcement learning enables computer-use agents to improve through interaction with real software environments, including websites and desktop applications. However, conventional deployments replicate an initialized runtime for each independent rollout, even when trajectories use the same software, incurring repeated memory and initialization costs as the number of parallel environments grows. Does an independent computer-use environment require an independent execution runtime? Our key observation is that trajectories require independent mutable state, while initialized application runtimes can be reused across concurrently evolving environments, making state the natural unit of environment independence. Guided by this observation, we introduce CUA-Sandbox, which separates private state capsules from shared runtimes through state-scoped execution and transactional lifecycle operations, including resets and branches, while retaining the original software interfaces and task evaluators. Experiments show comparable or improved task success relative to Docker, while substantially reducing rollout and resource costs. CUA-Sandbox achieves up to a 6.20x increase in rollout throughput, a 9.2x reduction in per-environment memory, and a 504x reduction in incremental storage.
Chinese Translation
强化学习使计算机使用代理能够通过与真实软件环境（包括网站和桌面应用程序）的互动来不断改进。然而，传统的部署方式在每次独立的回滚中都会复制一个初始化的运行时，即使轨迹使用相同的软件，这在并行环境数量增加时会产生重复的内存和初始化成本。一个独立的计算机使用环境是否需要一个独立的执行运行时？我们的关键观察是，轨迹需要独立的可变状态，而初始化的应用运行时可以在并发演变的环境之间重用，使得状态成为环境独立性的自然单位。基于这一观察，我们提出了CUA-Sandbox，它通过状态范围执行和事务生命周期操作（包括重置和分支）将私有状态胶囊与共享运行时分离，同时保留原始软件接口和任务评估器。实验表明，相较于Docker，CUA-Sandbox在任务成功率上表现出可比或更好的结果，同时显著降低了回滚和资源成本。CUA-Sandbox在回滚吞吐量上实现了最高6.20倍的提升，在每个环境的内存使用上减少了9.2倍，在增量存储上减少了504倍。
cs.AI / 128 / 2609.32752
Action Shaping: Policies Absorb What They Can Express
动作塑形：策略吸收其能够表达的内容
Abstract
Reward shaping has a theorem: a potential-based term can be removed without changing the optimal policy. The same practice on the action channel, an offset added in training and dropped at deployment, has no theorem. Nothing cancels an action offset, so the correction is kept at deployment or removed without a guarantee. We call it action shaping and state its principle. A trainable policy absorbs an offset its own output layer can reproduce exactly, which is what we mean by express; what is absorbed can be removed with the return intact. Its minimal instance is a zero-initialized linear head behind a learnable gate, added to an actor that trains through a learned action-value function, with no penalty or schedule. The gate rises and then falls on its own, for deterministic and stochastic actors alike, and on 20 tasks removing the head costs almost nothing. The condition is exact reproduction, not capacity: a nonlinear head with more parameters is not absorbed, and in a paired control, one linear path added to a nonlinear base head restores absorption. Exact reproduction gives the loss a flat direction that gradient noise drifts along, and the offset's amplitude indicates, before removal, what dropping the head will cost. Action shaping thus gains the counterpart of the shaping theorem, a condition for absorption, together with the mechanism behind it and a diagnostic that reads it. Policies absorb what they can express, and only that.
Chinese Translation
奖励塑形有一个定理：基于潜力的项可以在不改变最优策略的情况下被移除。在动作通道上进行相同的操作，即在训练中添加的偏移量在部署时被去掉，并没有相应的定理。没有任何东西可以抵消一个动作偏移，因此在部署时要么保留修正，要么在没有保证的情况下移除。我们称之为动作塑形，并阐述其原理。一个可训练的策略吸收其自身输出层能够完全再现的偏移量，这就是我们所说的表达；被吸收的内容可以在回报保持不变的情况下被移除。其最小实例是一个零初始化的线性头，位于一个可学习的门后面，添加到一个通过学习的动作价值函数进行训练的演员中，没有任何惩罚或调度。该门自行上升然后下降，适用于确定性和随机性演员，并且在20个任务中，移除该头几乎没有成本。条件是完全再现，而不是能力：具有更多参数的非线性头不会被吸收，而在配对控制中，添加到非线性基础头的一个线性路径恢复了吸收。完全再现使损失具有一个平坦的方向，梯度噪声沿着这个方向漂移，偏移量的幅度在移除之前指示了去掉头部的成本。因此，动作塑形获得了塑形定理的对应物，即吸收的条件，以及其背后的机制和读取该机制的诊断。策略吸收其能够表达的内容，仅此而已。
cs.AI / 129 / 2609.32754
Adaptive Consistency Graph for Long-Horizon Agents
用于长时间跨度智能体的自适应一致性图
Abstract
Large language model agents can often make reasonable local decisions on short tasks, yet their performance degrades when success requires long sequences of dependent actions and tool calls. During execution, task requirements, historical evidence, and the current execution state may gradually become disconnected, so later decisions can drift from the original objective. We study this problem by introducing the Adaptive Consistency Graph (ACG) for long-horizon execution. ACG incrementally organizes execution evidence and its provenance in a persistent graph, then constructs a temporary requirement-centered view for each decision under a bounded context budget. Rather than replacing the base agent's planner or tool executor, ACG provides a structured and traceable context view for each decision. In the matched evaluation, ACG improves GPT-5.6-luna's average success from 44.5\% with ReAct to 50.2\%, with the largest gain on BrowseComp-Plus (73.5\% versus 62.4\%). We further analyze trajectory structure and inference cost to characterize this improvement.
Chinese Translation
大型语言模型智能体在短任务上通常能够做出合理的局部决策，但当成功需要长序列的依赖动作和工具调用时，它们的性能会下降。在执行过程中，任务要求、历史证据和当前执行状态可能逐渐变得脱节，因此后续决策可能偏离原始目标。我们通过引入自适应一致性图（Adaptive Consistency Graph, ACG）来研究这个问题，以支持长时间跨度的执行。ACG 逐步在一个持久图中组织执行证据及其来源，然后为每个决策在有限的上下文预算下构建一个临时的以要求为中心的视图。ACG 并不是替代基础智能体的规划器或工具执行器，而是为每个决策提供一个结构化且可追溯的上下文视图。在匹配评估中，ACG 将 GPT-5.6-luna 的平均成功率从 44.5\%（使用 ReAct）提高到 50.2\%，在 BrowseComp-Plus 上的增益最大（73.5\% 对比 62.4\%）。我们进一步分析轨迹结构和推理成本，以表征这一改进。
cs.AI / 130 / 2609.32757
Readout is not Recovery: Dissociating Coordinate Emission from Visual-Corruption Repair in Vision-Language Models
读出并非恢复：在视觉-语言模型中区分坐标发射与视觉损坏修复
Abstract
VLM bounding-box localization is both language generation and spatial commitment. Parseable fields such as bbox_2d make localization easy to score, but dimensions that emit coordinate tokens need not repair localization after visual evidence is damaged. We study this readout/recovery separation in Qwen3-VL-4B-Instruct on single-object COCO grounding. We compare clean coordinate-token readout rankings with corruption-derived repair rankings, using object-mask endpoint replacement for recovery and clean-input flooring for depth localization. In Qwen3-VL, coordinate-token rankings are inert through layer 24, load-bearing from layers 32-35, and peak at layer 34; corruption-derived rankings harm layers 16-24 but become beneficial near layer 35/final. A Kimi-VL-A3B diagnostic shows a matching output-proximal transition despite a different box format. Object-mask recovery separates rank budgets: $k=250$ shows necessity, $k=500$ shows Top-$k$ restoration above random, and $k=d/2$ is largely capacity-driven. Partial-occlusion sweeps reveal that high-overlap coordinate-token sets can hurt at $k=1000$ and help mainly at half-width, while population corruption-derived sets provide no reliable fixed repair set. Edge-attribution patching shows coordinate-token paths are high precision but low recall for detection recovery, and RMSNorm quasi-layer controls do not close the endpoint-repair gap. Endpoint coordinate triage is therefore a useful circuit prior, but occlusion recovery requires a separate benchmark.
Chinese Translation
视觉语言模型（VLM）的边界框定位既涉及语言生成，也涉及空间承诺。可解析的字段如 bbox_2d 使得定位评分变得简单，但发射坐标标记的维度在视觉证据受损后不一定需要修复定位。我们在 Qwen3-VL-4B-Instruct 模型上研究了这种读出/恢复的分离，聚焦于单对象的 COCO 基础。我们比较了干净的坐标标记读出排名与基于损坏的修复排名，使用对象掩码端点替换进行恢复，并使用干净输入的基准进行深度定位。在 Qwen3-VL 中，坐标标记排名在第 24 层之前是惰性的，在第 32-35 层之间承载负荷，并在第 34 层达到峰值；基于损坏的排名在第 16-24 层造成损害，但在第 35 层/最终层附近变得有利。Kimi-VL-A3B 的诊断显示尽管框格式不同，但输出邻近的过渡是一致的。对象掩码恢复分离了排名预算：$k=250$ 显示必要性，$k=500$ 显示 Top-$k$ 恢复优于随机，而 $k=d/2$ 主要受容量驱动。部分遮挡的分析显示，高重叠的坐标标记集在 $k=1000$ 时可能造成损害，而在半宽度时主要提供帮助，而基于人群的损坏派生集则没有提供可靠的固定修复集。边缘归因修补显示坐标标记路径在检测恢复中具有高精度但低召回率，而 RMSNorm 准层控制并未缩小端点修复的差距。因此，端点坐标分类是一个有用的电路先验，但遮挡恢复需要一个单独的基准。
cs.AI / 131 / 2609.32763
Mandela-Bench: Multimodal Models Remember Canonical Images Instead of Seeing Them
曼德拉基准：多模态模型记住经典图像而非观察它们
Abstract
Historical photographs and other canonical images can now be edited seamlessly with a single instruction, often leaving no reliable pixel-level trace. In such cases, the only evidence of manipulation may be a fact about what the image depicts. Existing benchmarks instead rely on generator artefacts, image-caption inconsistencies, visual implausibilities, or external references, and therefore do not test whether a model can use its own world knowledge to verify a recognized image. We introduce Mandela-Bench, containing 1,507 edits of canonical images: 1,359 knowledge-only forgeries, each contradicting one verifiable fact, and 148 anchor-free controls that preserve the editing process without introducing a factual contradiction, together with 474 untouched originals. We score not only whether a model detects a forgery, but whether its explanation identifies the inserted entity or the fact being violated. Across 36 multimodal models, from 0.8B parameters to frontier scale, we find a consistent failure mode. When a public figure is removed from a familiar photograph, models still name that person in up to 72.7% of responses. Some models can distinguish the replacement face from the original when shown in isolation, yet still judge the full edited photograph as authentic. Providing the true event and date does not improve knowledge-grounded detection, whereas providing the same information after cropping away the recognizable composition does. Even under explicit verification prompts, only one of the 36 models meets the KGR criterion on at least half of the forged images. These results suggest that the failures cannot be explained by missing knowledge or inadequate perception alone. Instead, they are consistent with recognition biasing verification toward the remembered canonical image rather than the observed edit.
Chinese Translation
历史照片和其他经典图像现在可以通过单一指令无缝编辑，通常不会留下可靠的像素级痕迹。在这种情况下，唯一的操控证据可能是图像所描绘内容的一个事实。现有基准依赖于生成器伪影、图像-标题不一致、视觉不合理性或外部参考，因此并未测试模型是否能够利用自身的世界知识来验证已识别的图像。我们引入了曼德拉基准，包含1,507个经典图像的编辑：1,359个仅基于知识的伪造，每个都与一个可验证的事实相矛盾，以及148个不带锚点的对照，这些对照在不引入事实矛盾的情况下保留了编辑过程，另外还有474个未触动的原始图像。我们不仅评分模型是否检测到伪造，还评分其解释是否识别出插入的实体或被违反的事实。在36个多模态模型中，从0.8B参数到前沿规模，我们发现了一种一致的失败模式。当一个公众人物从熟悉的照片中移除时，模型在多达72.7%的回答中仍然提到该人物。一些模型在单独展示替换面孔时能够区分出与原始面孔的不同，但仍然判断完整的编辑照片为真实。提供真实事件和日期并未改善基于知识的检测，而在裁剪掉可识别构图后提供相同信息则有助于检测。即使在明确的验证提示下，36个模型中只有一个在至少一半的伪造图像上满足KGR标准。这些结果表明，失败并不能仅通过知识缺失或感知不足来解释。相反，它们与识别偏向于记忆中的经典图像而非观察到的编辑一致。
cs.AI / 132 / 2609.32773
Forecasting Intraday USD/CAD Exchange Rate with News-Derived Monetary-Policy Signals
基于新闻衍生货币政策信号的日内美元/加元汇率预测
Abstract
Monetary-policy announcements and central-bank communications play a central role in foreign exchange markets, yet their qualitative, unstructured form makes their forecasting value difficult to quantify. While prior research has largely focused on sentiment extracted from financial news, comparatively little is known about the relative contribution of different dimensions of monetary-policy communication. Existing studies primarily evaluate whether textual information improves overall forecasting performance but provide limited insight into which communication channels drive such improvements. To address this gap, this paper introduces a statistical attribution methodology that decomposes monetary-policy communication into interpretable channels and quantifies their incremental forecasting contribution under false-discovery-rate control. Monetary-policy news is transformed into structured communication signals using large language models (LLMs) and temporal feature engineering. These signals are evaluated using rolling-window experiments with tree-based machine-learning models. The results show that monetary-policy communication contains measurable predictive information. Attribution analysis shows that predictive value is concentrated in a small subset of signals, with communication timing providing the strongest individual feature-level contribution, targeted communication-activity measures also contributing positively, and LLM-derived sentiment providing complementary information at the group level. The findings indicate that communication-based forecasting value extends beyond sentiment alone and that attribution, rather than aggregate accuracy alone, is central to evaluating news-derived signals.
Chinese Translation
货币政策公告和中央银行沟通在外汇市场中发挥着核心作用，但其定性、非结构化的形式使得其预测价值难以量化。尽管之前的研究主要集中在从金融新闻中提取的情绪上，但关于货币政策沟通的不同维度相对贡献的了解却相对较少。现有研究主要评估文本信息是否提高了整体预测性能，但对哪些沟通渠道推动了这种改善提供的见解有限。为了解决这一空白，本文引入了一种统计归因方法，将货币政策沟通分解为可解释的渠道，并在假发现率控制下量化其增量预测贡献。货币政策新闻通过大型语言模型（LLMs）和时间特征工程转化为结构化的沟通信号。这些信号通过基于树的机器学习模型进行滚动窗口实验评估。结果表明，货币政策沟通包含可测量的预测信息。归因分析显示，预测价值集中在少数信号中，沟通时机提供了最强的个体特征级贡献，针对性的沟通活动措施也有积极贡献，而LLM衍生的情绪在群体层面提供了互补信息。研究结果表明，基于沟通的预测价值超越了单纯的情绪，归因而非仅仅是整体准确性对于评估新闻衍生信号至关重要。
cs.AI / 133 / 2609.32778
Agentic Network Traffic Monitoring
代理网络流量监控
Abstract
As the use of agentic artificial intelligence increases in nearly every industry, there exists a widening attack surface. It is necessary to monitor agents to ensure that agents are acting in a way that is aligned with the users intent. Auditing an agent's network traffic provides a clear record of the agent interactions. This work presents a novel approach to monitoring the network traffic of agentic systems using complex valued hypersparse traffic matrices by integrating DBOS (DataBase OS), the OneSparse PostgreSQL database, and the GraphBLAS math library. To develop these concepts an agentic simulator was constructed, allowing a varying numbers of AI agents to collectively survey a virtual environment using different strategies. The resulting network traffic matrices enable easy monitoring of the AI agents.
Chinese Translation
随着代理人工智能在几乎每个行业中的应用增加，攻击面也在不断扩大。监控代理的行为以确保其与用户意图一致是必要的。对代理网络流量的审计提供了代理交互的清晰记录。本研究提出了一种新颖的方法，通过集成 DBOS（数据基础操作系统）、OneSparse PostgreSQL 数据库和 GraphBLAS 数学库，使用复杂值超稀疏流量矩阵来监控代理系统的网络流量。为了开发这些概念，构建了一个代理模拟器，允许不同数量的 AI 代理使用不同策略共同调查虚拟环境。生成的网络流量矩阵使得对 AI 代理的监控变得更加简单。
cs.AI / 134 / 2609.32782
Learning response-aware patient dynamics for respiratory support
学习响应感知的患者动态以支持呼吸治疗
Abstract
Respiratory support can shape the short-term physiological trajectory of critically ill patients, but patients receiving the same intervention may follow different physiological trajectories. Clinical patient dynamics models typically predict future states from recent physiology and recorded interventions, while physiological change is mainly represented through the predicted future state. We propose a response-aware patient dynamics model that explicitly represents physiological change during autoregressive state updating. The model decomposes predicted physiological change into state-dependent baseline dynamics and respiratory-support-associated deviations, with room air providing a reference for the decomposition. We provide a formal analysis of this reference-anchored formulation. A response pathway encodes the predicted physiological change and uses it to update the latent patient state across the forecast horizon. Across ICU cohorts from two independent institutions, the proposed model achieves comparable overall trajectory prediction to patient dynamics baselines, with more consistent improvements when physiological states are changing.
Chinese Translation
呼吸支持可以塑造重症患者的短期生理轨迹，但接受相同干预的患者可能会遵循不同的生理轨迹。临床患者动态模型通常根据近期生理状态和记录的干预来预测未来状态，而生理变化主要通过预测的未来状态来表示。我们提出了一种响应感知的患者动态模型，该模型在自回归状态更新过程中明确表示生理变化。该模型将预测的生理变化分解为状态依赖的基线动态和与呼吸支持相关的偏差，以室内空气作为分解的参考。我们对这种参考锚定的公式进行了正式分析。响应路径编码了预测的生理变化，并利用它在预测范围内更新潜在患者状态。在来自两个独立机构的重症监护病房（ICU）队列中，所提模型在整体轨迹预测方面与患者动态基线相当，并在生理状态变化时表现出更一致的改善。
cs.AI / 135 / 2609.32787
CLAIRE: A Schema-Grounded Hybrid Workflow for Healthcare Administrative Form Completion
CLAIRE：一种基于模式的混合工作流程用于医疗行政表单的填写
Abstract
Healthcare administrative staff transfer structured information from electronic health records, referrals, claims systems, provider rosters, and work queues into dynamic forms. We developed and evaluated CLAIRE (Clinical Language and Agentic Intelligence for Reasoning and Entry), a hybrid workflow that separates field-state discovery, source-to-field mapping, deterministic validation, bounded correction, escalation, and audit tracing. We tested five synthetic healthcare administrative schemas, 1,000 source records, four interface variants, two data-quality suites, and six comparators, yielding 24,000 benchmark episodes. A separate strict-output audit evaluated direct mappings from Qwen2.5-1.5B and Qwen2.5-7B, and a trace-derived operational simulation covered 6,000 episodes. Under the evaluated synthetic benchmark conditions, full CLAIRE achieved 1.000 episode success, field accuracy, required-field completion, and dependency completion in both suites; removing validation reduced stress-suite success to 0.500. In the simulation, 100.0% of clean and validation-stress episodes reached a staff-reviewable draft, compared with 68.6% of escalation challenge episodes, unsupported cases were blocked. Scenario-based savings were 149.7-165.5 seconds per case, not observed staff times. The findings support schema-grounded, validation-first healthcare administrative automation in which language-model components assist mapping but do not authorize unsupported or consequential actions.
Chinese Translation
医疗行政人员将来自电子健康记录、转诊、索赔系统、提供者名册和工作队列的结构化信息转移到动态表单中。我们开发并评估了CLAIRE（临床语言与代理智能推理与输入），这是一种混合工作流程，分离了字段状态发现、源到字段映射、确定性验证、有界修正、升级和审计追踪。我们测试了五种合成医疗行政模式、1,000个源记录、四种界面变体、两个数据质量套件和六个比较器，共产生了24,000个基准事件。一个单独的严格输出审计评估了来自Qwen2.5-1.5B和Qwen2.5-7B的直接映射，而一个基于追踪的操作模拟覆盖了6,000个事件。在评估的合成基准条件下，完整的CLAIRE在两个套件中实现了1.000的事件成功率、字段准确性、必填字段完成率和依赖关系完成率；去除验证后，压力套件的成功率降至0.500。在模拟中，100.0%的清洁和验证压力事件达到了可供工作人员审核的草稿，而升级挑战事件的比例为68.6%，不支持的案例被阻止。基于场景的节省为每个案例149.7-165.5秒，未观察到工作人员时间。这些发现支持基于模式的、优先验证的医疗行政自动化，其中语言模型组件协助映射，但不授权不支持或有后果的操作。
cs.AI / 136 / 2609.32791
$T^5$: Twin-Critic Training for Token-Level Thoughts in Reinforcement Mid-Training
$T^5$: 双评论员训练用于强化学习中的令牌级思维
Abstract
Reinforcement mid-training lets language models learn internal thoughts from unlabeled text, but efficient token-level credit assignment remains challenging. Existing group-relative methods require costly repeated generation. Learned critics offer single-rollout feedback, but accurate return prediction alone does not ensure reliable policy updates. Our analysis shows how training--inference mismatch and PPO clipping prevent a common offset in advantage estimates from cancelling out, introducing additional update drift. We propose \tfour{}, a twin-critic method that calibrates token-level advantages from a single generated trajectory. After warmup and held-out qualification, the critics provide two advantage estimates, combined using action-dependent weights learned through a conditional-moment saddle-point objective. This objective brings the average advantage at each prefix toward zero, while a signal-retention constraint prevents the correction from erasing the learning signal. Sharing information across text positions avoids repeated sampling of each prefix. Theoretically, we characterize optimal mixing under the signal-retention constraint and establish an upper bound on residual mean-induced drift. Experiments show that, compared with the state-of-the-art critic-free method, \tfour{} improves mean benchmark performance by 7.8\% and reduces mean training-step time by up to 63.4\%.
Chinese Translation
强化学习中的中期训练使语言模型能够从未标记文本中学习内部思维，但高效的令牌级信用分配仍然具有挑战性。现有的群体相对方法需要昂贵的重复生成。学习的评论员提供单次生成的反馈，但仅凭准确的回报预测并不能确保可靠的策略更新。我们的分析表明，训练与推理的不匹配以及PPO（Proximal Policy Optimization）剪切阻止了优势估计中常见偏移的相互抵消，从而引入了额外的更新漂移。我们提出了 four{}，一种双评论员方法，通过单一生成轨迹校准令牌级优势。在预热和保留资格后，评论员提供两个优势估计，这些估计通过条件矩时鞍点目标学习的动作依赖权重进行组合。该目标使每个前缀的平均优势趋向于零，同时信号保留约束防止修正抹去学习信号。在文本位置之间共享信息避免了对每个前缀的重复采样。从理论上讲，我们在信号保留约束下表征了最佳混合，并建立了残差均值引起的漂移的上界。实验表明，与最先进的无评论员方法相比， four{}将平均基准性能提高了7.8 ext{%}，并将平均训练步骤时间减少了最多63.4 ext{%}。
cs.AI / 137 / 2609.32795
AgentHabit: Characterizing Distinct Behaviors of Agents on Everyday Tasks
AgentHabit：表征代理在日常任务中的不同行为
Abstract
Large language model (LLM) agents assist users with everyday tasks that can be completed in many reasonable ways. Even when their answers are useful, how agents carry out these tasks may not match users' preferences and needs. For example, agents differ in whether they ask clarifying questions or search the web. We introduce HABIT, a taxonomy of 23 behavioral axes in five categories, which three authors and three LLMs derive bottom-up from 408 agent trajectories across 17 domains. On held-out tasks, HABIT distinguishes models more clearly than existing taxonomies of human values and agent actions while supporting comparably consistent annotation. Building on HABIT, we construct AgentHABIT, a benchmark that profiles each agent's behavioral tendencies from its trajectories on 86 everyday tasks. Profiling 18 models with AgentHABIT reveals a range of distinctive tendencies. For example, most GPT and Claude models state their assumptions and offer alternatives when requirements conflict, whereas Qwen and Google's models more often leave assumptions or changes to requirements unstated. These profiles remain recognizable even when built from entirely different sets of tasks, indicating that they reflect general tendencies rather than task-specific behavior. Prompting agents to adopt specific behaviors shifts some axes readily but barely changes others, while fine-tuning on another model's trajectories changes only part of a model's profile and leaves much of it intact. Overall, HABIT and AgentHABIT provide a systematic framework for characterizing how agents carry out everyday tasks beyond task success, offering insights to guide the development of agents whose behavior better fits users' needs.
Chinese Translation
大型语言模型（LLM）代理帮助用户完成可以通过多种合理方式完成的日常任务。即使它们的回答是有用的，代理执行这些任务的方式可能与用户的偏好和需求不匹配。例如，代理在是否提出澄清问题或搜索网络方面存在差异。我们引入了HABIT，一个包含23个行为轴的分类法，分为五个类别，这些行为轴是由三位作者和三种LLM从408个代理轨迹中自下而上推导而来的，涵盖17个领域。在保留任务中，HABIT比现有的人类价值观和代理行为的分类法更清晰地区分模型，同时支持相对一致的注释。在HABIT的基础上，我们构建了AgentHABIT，一个基准，描述每个代理在86个日常任务中的行为倾向。使用AgentHABIT对18个模型进行分析揭示了多种独特的倾向。例如，大多数GPT和Claude模型在需求冲突时会陈述其假设并提供替代方案，而Qwen和谷歌的模型则更常常不明确假设或需求的变化。这些特征即使是基于完全不同的任务集构建的，仍然保持可识别性，表明它们反映的是一般倾向而非特定任务行为。促使代理采用特定行为会轻易改变某些行为轴，但几乎不改变其他行为轴，而在另一个模型的轨迹上进行微调只会改变模型特征的一部分，保留其大部分特征。总体而言，HABIT和AgentHABIT提供了一个系统框架，用于表征代理如何超越任务成功地执行日常任务，为指导开发更符合用户需求的代理行为提供了洞见。
cs.AI / 138 / 2609.32801
PlanGuard: A Guardrail for Multi-Step Plan Safety in Embodied Agents
PlanGuard：一种用于具身智能体多步骤计划安全的护栏
Abstract
Embodied task planners may produce multi-step plans whose subtask dependencies and interactions with the environment create physical risks during execution. Yet existing safeguards overlook such compositional risks, as general-purpose guardrails focus on semantic harm and embodied safety detectors assess subtasks in isolation. To address this gap, we introduce PlanGuard, the first pre-execution detector that evaluates the physical safety of a complete multi-step plan in its current environment. For training and evaluation, we construct a Multi-Step Plan Safety (MSP-Safe) dataset through paired task construction, plan generation using diverse planners, and safety annotation by three judges. Task-oriented SFT on MSP-Safe establishes fundamental plan-safety assessment capabilities, yet a substantial gap remains between compact models suitable for real-time deployment and stronger but costlier large models. Accordingly, we propose Strong-Teacher Adaptive Compensation for On-Policy Distillation (STAC-OPD), which provides compact models with adaptive strong-teacher supervision along their on-policy trajectories. It combines token-level distribution transfer from a fine-tuned strong teacher with probability-routed sequence-level compensation, retaining student-generated targets when the student favors the reference safety decision and using teacher-reconstructed targets otherwise. Across all test subsets, PlanGuard-2B achieves average 87.15% ACC and 87.21% F1, demonstrating effective whole-plan physical-risk detection at compact model scale. Code and dataset will be publicly released.
Chinese Translation
具身任务规划器可能生成多步骤计划，其子任务依赖关系和与环境的交互在执行过程中可能带来物理风险。然而，现有的安全保障措施忽视了这种组合风险，因为通用护栏主要关注语义伤害，而具身安全检测器则孤立地评估子任务。为了解决这一问题，我们提出了PlanGuard，这是第一个在当前环境中评估完整多步骤计划物理安全性的预执行检测器。为了进行训练和评估，我们通过成对任务构建、使用多样化规划器生成计划以及由三位评审员进行安全注释，构建了一个多步骤计划安全（MSP-Safe）数据集。针对MSP-Safe的任务导向的SFT建立了基本的计划安全评估能力，但适合实时部署的紧凑模型与更强但成本更高的大模型之间仍存在显著差距。因此，我们提出了用于在线策略蒸馏的强教师自适应补偿（STAC-OPD），为紧凑模型提供沿其在线策略轨迹的自适应强教师监督。它结合了来自微调强教师的令牌级分布转移与概率引导的序列级补偿，当学生倾向于参考安全决策时保留学生生成的目标，否则使用教师重构的目标。在所有测试子集上，PlanGuard-2B实现了平均87.15%的准确率（ACC）和87.21%的F1分数，证明了在紧凑模型规模下有效的整体计划物理风险检测。代码和数据集将公开发布。
cs.AI / 139 / 2609.32802
Re-derivability Decides What a Staged Agent Pipeline Recovers After an Upstream Fault
再推导性决定了分阶段代理管道在上游故障后恢复的内容
Abstract
One variable sets what an upstream fault costs a staged pipeline of language-model agents: re-derivability, how much of what a stage needs it can rebuild from the original problem. Grounding an inspector agent in that problem is worth +0.608 [+0.517, +0.700] to +0.358 over a blind one on four open-weight backbones served with thinking disabled, and on the two Qwen backbones the blind inspector changes no item at all. That head-to-head is exploratory. One deterministic fault enters the first stage, and we re-expose the original problem to $k = 0,\dots,3$ of the downstream stages with agents, items, fault and topology held fixed, on 120 gsm_hard items per arm at temperature zero. Accuracy under fault rises on four of four backbones, from +0.233 to +0.392, the largest Holm-adjusted $p$ being $2.1\times10^{-6}$. A registered kill test rules out tokens. Blanking every word holds the word slots fixed, and retention tracks the visible fraction on four of four, climbing from 0.221 to 0.692 on the primary. Those two families are confirmatory and everything else here is exploratory. The interaction excludes zero on two of four backbones under the registered pipeline, four of four under a three-stage pipeline, and three of four under full message history, the primary at +0.317. On Llama-3.1-8B the fault carries no detectable cost at any dose, so the other three carry every claim about what a fault costs. Re-derivability also sets what the architecture costs, and no decomposition we measured reliably beats one direct call. With no fault injected the registered pipeline loses to that call by -0.267, -0.125 and -0.317, and on Phi-4 reads +0.058 at $p = 0.118$, which the test fails to separate from zero. The repair that works is cheap and front-loaded: the first re-grounded stage buys +0.394 of matched retention for +59.8 tokens per item on Qwen3-14B, and the stages after it buy nothing.
Chinese Translation
一个变量决定了上游故障对语言模型代理的分阶段管道的成本：再推导性，即一个阶段需要的内容能够从原始问题中重建的程度。在四个禁用思考的开放权重骨干网络上，将检查代理嵌入该问题的价值为 +0.608 [+0.517, +0.700]，而盲目代理的价值为 +0.358；在两个 Qwen 骨干网络上，盲目检查代理完全没有改变任何项目。该对比是探索性的。一个确定性故障进入第一个阶段，我们将原始问题重新暴露给 $k = 0, ext{到},3$ 的下游阶段，代理、项目、故障和拓扑保持不变，每个臂有 120 个 gsm_hard 项目，温度为零。在四个骨干网络下，故障下的准确率从 +0.233 上升到 +0.392，最大的 Holm 调整 $p$ 值为 $2.1 imes10^{-6}$。一个注册的杀死测试排除了标记。将每个单词空白保持单词槽固定，保留率在四个骨干网络中追踪可见部分，主要部分从 0.221 上升到 0.692。这两个家族是确认性的，而这里的其他一切都是探索性的。在注册管道下，交互在四个骨干网络中排除了零，在三阶段管道下排除了四个骨干网络，在完整消息历史下排除了四个中的三个，主要部分为 +0.317。在 Llama-3.1-8B 上，故障在任何剂量下都没有可检测的成本，因此其他三个承载了关于故障成本的所有主张。再推导性也决定了架构的成本，我们测量的没有任何分解能可靠地超越一次直接调用。在没有注入故障的情况下，注册管道的表现比该调用低 -0.267、-0.125 和 -0.317，而在 Phi-4 上读取 +0.058，$p = 0.118$，该测试未能将其与零区分开。有效的修复是廉价且前置的：第一个重新嵌入的阶段在 Qwen3-14B 上为每个项目购买了 +0.394 的匹配保留，代价为 +59.8 个标记，而其后的阶段没有任何收益。
cs.AI / 140 / 2609.32803
Nutri-ATLAS: Embodied Agent for Tabulated Lookup and Assistance for Smarter nutrition
Nutri-ATLAS：用于更智能营养的表格查找与辅助的具身智能体
Abstract
Generative and Agentic IoT systems offer a promising foundation for digital healthcare applications that combine sensing, personalized reasoning, and autonomous interaction in real-world environments. Nutrition assistance is a natural use case, but existing Large Language Model (LLM)-based systems are often limited to passive text interaction and static context, making them unreliable when food descriptions are ambiguous or nutritional evidence is missing. We propose Nutri-ATLAS, an Embodied Agent for Tabulated Lookup and Assistance for smarter nutrition in the real world. It integrates graph-grounded nutrition reasoning, hardware-aware LLM selection, and robot-based evidence acquisition. Nutri-ATLAS builds a unified Food-Nutrient knowledge graph from USDA FoodData Central and FoodKG and learns 64-dimensional GATv2 food and recipe embeddings. A shared hybrid graph-text scoring mechanism supports food nutrition extraction, nutritional gap filling, substitute retrieval, and recipe-level meal composition, while an LLM-guided skill interface navigates landmarks, updates dietary-context and food-accessibility memory, and grounds recommendations in observed food availability. We evaluate Nutri-ATLAS across nutrient estimation, substitution retrieval, recipe recommendation, patient-profile adherence, edge deployment, and real-world embodied execution. On HealthyFoodSubs, the hybrid retriever achieves 37.9% MAP, 80.7% RR@5, and 90.1% RR@10. On NutriBench v2, Dense+GAT retrieval grounds nutrient estimation across nine quantized Qwen3.5-9B configurations. On PFoodReQ, Nutri-ATLAS reaches 78.8% MAP, 83.0% MAR, and 77.5% F1. A patient-profile study shows adherence to allergy and healthy-target constraints for all selected cases.
Chinese Translation
生成性和智能物联网系统为数字医疗应用提供了一个有前景的基础，这些应用结合了感知、个性化推理和在现实环境中的自主交互。营养辅助是一个自然的应用场景，但现有的基于大型语言模型（LLM）系统通常仅限于被动的文本交互和静态上下文，这使得在食品描述模糊或缺乏营养证据时，它们的可靠性受到限制。我们提出了Nutri-ATLAS，一个用于现实世界中更智能营养的表格查找与辅助的具身智能体。它整合了基于图的营养推理、硬件感知的LLM选择和基于机器人的证据获取。Nutri-ATLAS从美国农业部食品数据中心（USDA FoodData Central）和FoodKG构建了一个统一的食品-营养知识图谱，并学习了64维的GATv2食品和食谱嵌入。一个共享的混合图-文本评分机制支持食品营养提取、营养缺口填补、替代品检索和食谱级餐点组合，而一个由LLM指导的技能接口则导航地标，更新饮食上下文和食品可及性记忆，并将推荐基于观察到的食品可用性进行实地验证。我们在营养估计、替代品检索、食谱推荐、患者档案遵循、边缘部署和现实世界具身执行等方面评估了Nutri-ATLAS。在HealthyFoodSubs上，混合检索器达到了37.9%的平均精度（MAP）、80.7%的前5推荐率（RR@5）和90.1%的前10推荐率（RR@10）。在NutriBench v2上，Dense+GAT检索在九个量化的Qwen3.5-9B配置中实现了营养估计的基础。在PFoodReQ上，Nutri-ATLAS达到了78.8%的MAP、83.0%的MAR和77.5%的F1分数。一项患者档案研究显示，在所有选定案例中均遵循过敏和健康目标约束。
cs.AI / 141 / 2609.32805
Decision-Sufficient State Representations: Measuring and Reducing Write-Time Regret
决策充分的状态表示：衡量和减少写入时间的遗憾
Abstract
Long tasks produce more history than an LLM agent can hold in its context, and more than it uses reliably even when the history fits. A growing line of work therefore has agents carry a short written state instead: at every step a writer rewrites the state, and a reader acts from the state alone. Steps stay cheap, but anything the writer drops is lost before later decisions reveal that they need it. We quantify this loss and ask whether training can reduce it. Comparing the written state with the best state of the same size written in hindsight, we split the reader's loss into a budget loss, which any state of that size must incur, and a write-time regret, which comes from the writer's choices. In TextWorld cooking games where we control how long a fact must be carried before it is needed, a 128-token state holding the facts wins nearly every game, while prompted language-model writers win at most 17%. Almost all of the loss is write-time regret, and it grows with the delay. We then train the writer from the reader's own loss. DSSR (decision-sufficient state representations) scores candidate states by how well the reader acts after the writer carries them forward, and teaches the writer to prefer the better ones. This forward-rolled score predicts game outcomes ($\rho = 0.48$), whereas scoring a candidate as a fixed context, as hindsight methods usually do, does not ($\rho \leq 0.07$). On a pre-registered test split opened once, training adds +7.0 [+1.9, +12.2] points of success when facts are needed soon, bringing a plain summary writer to the level of belief- and slot-based memory prompts. The gain shrinks as the delay grows and is significant only at the shortest delay. We trace this limit to credit assignment: keeping a fact now pays off only if every later rewrite keeps it too, which a per-step score cannot see.
Chinese Translation
长任务产生的历史信息超出了大型语言模型（LLM）代理能够在其上下文中保持的范围，即使历史信息适合时，代理也无法可靠地使用这些信息。因此，越来越多的研究工作让代理携带一个简短的书面状态：在每一步中，写作者重写状态，而读者仅依赖于该状态进行行动。步骤保持廉价，但写作者丢弃的任何信息在后续决策揭示需要时便会丢失。我们量化了这种损失，并探讨训练是否能够减少它。通过将书面状态与事后回顾中相同大小的最佳状态进行比较，我们将读者的损失分为预算损失（任何该大小的状态都必须承担）和写入时间遗憾（源于写作者的选择）。在TextWorld烹饪游戏中，我们控制一个事实在被需要之前必须保留的时间，128个标记的状态几乎在每场游戏中获胜，而提示的语言模型写作者最多只赢得17%。几乎所有的损失都是写入时间遗憾，并且随着延迟的增加而增长。然后，我们根据读者自身的损失训练写作者。决策充分的状态表示（DSSR）通过评估读者在写作者推进状态后行动的效果来为候选状态打分，并教导写作者优先选择更好的状态。这种前向滚动的评分能够预测游戏结果（$
ho = 0.48$），而将候选状态评分为固定上下文（如事后方法通常所做的）则无法做到（$
ho ext{≤} 0.07$）。在一次性开放的预注册测试分割中，当事实很快被需要时，训练增加了成功率+7.0 [+1.9, +12.2]点，使普通摘要写作者达到了基于信念和槽位的记忆提示的水平。随着延迟的增加，增益缩小，仅在最短延迟时显著。我们将这一限制追溯到信用分配：现在保留一个事实只有在每次后续重写也保留它的情况下才有回报，而逐步评分无法看到这一点。
cs.AI / 142 / 2609.32807
Beyond Accuracy: Counterfactual Fragility and Demographic Bias in Clinical Evaluation of LLMs
超越准确性：临床评估中大型语言模型的反事实脆弱性与人口统计偏差
Abstract
Clinical LLM evaluation often emphasizes answer accuracy; however, accuracy alone does not test counterfactual consistency or demographic robustness. We evaluated six LLMs on 150 MedQA USMLE questions using two automated perturbation tests to assess their performance. The counterfactual validity (CFV) test asked each model to make a minimal, plausible clinical change that would make a different answer correct. The demographic robustness test added six demographic prefixes to the same vignette and compared the answers and explanations with a no demographic baseline. Of the 900 CFV attempts, 228 (25.3 %) were valid and 672 were invalid. Across 5,400 demographic comparisons, 1,097 answers were changed (20.3%). Automated judging identified 3,128 stereotype evidence flags, including 1,932 in the broad Other category. MedGemma 27B achieved the highest accuracy (87.1%) and CFV (63.3%), lowest answer change rate (16.0%), and low mean Explanation Demographic Dissonance (EDD) score (0.169). However, its accuracy still exceeded its CFV, indicating that correct answers do not guarantee reliable performance on the counterfactual validity task. OpenBioLLM had the highest answer change rate and EDD, whereas GLM had the highest stereotype flag rate. These findings show that accuracy, CFV, answer stability, EDD, and stereotype evidence capture different evaluation aspects. Because all judgments were automated and no clinician validation was available, the results support safety screening but do not establish clinical deployability of the model.
Chinese Translation
临床大型语言模型（LLM）评估通常强调答案的准确性；然而，仅靠准确性无法测试反事实一致性或人口统计稳健性。我们使用两种自动扰动测试对六个LLM在150个MedQA USMLE问题上的表现进行了评估。反事实有效性（CFV）测试要求每个模型进行最小的、合理的临床改变，以使不同的答案变得正确。人口统计稳健性测试在同一情境中添加了六个人口统计前缀，并将答案和解释与无人口统计基线进行了比较。在900次CFV尝试中，228次（25.3%）是有效的，672次是无效的。在5400次人口统计比较中，1097个答案发生了变化（20.3%）。自动评判识别出了3128个刻板印象证据标记，其中1932个属于广义的其他类别。MedGemma 27B实现了最高的准确率（87.1%）和CFV（63.3%），最低的答案变化率（16.0%），以及较低的平均解释人口统计不一致性（EDD）得分（0.169）。然而，其准确性仍然超过了CFV，表明正确答案并不保证在反事实有效性任务上的可靠表现。OpenBioLLM具有最高的答案变化率和EDD，而GLM则具有最高的刻板印象标记率。这些发现表明，准确性、CFV、答案稳定性、EDD和刻板印象证据捕捉了不同的评估方面。由于所有判断均为自动化且没有临床验证，因此结果支持安全筛查，但并未确立模型的临床可部署性。
cs.AI / 143 / 2609.32809
Overwhelmed by Choice: Studying LLM Decision Making at Scale
选择的困扰：大规模研究大型语言模型的决策制定
Abstract
Multiple-choice and candidate-selection evaluations are widely used to assess LLM reasoning and decision-making, yet most benchmarks contain relatively small candidate sets. It remains unclear whether conclusions drawn from these settings remain valid as the candidate space scales. We systematically evaluate LLMs as the number of competing candidates increases and find substantial accuracy degradation across tasks, prompting strategies, and model scales. Controlled analyses show that standard long-context retrieval explanations cannot fully account for this degradation. Instead, we identify two systematic failure patterns. First, gold-margin collapse: the score gap between the correct answer and the strongest distractor progressively shrinks, driven primarily by weakening confidence in the correct answer. Second, earlier candidate preferences become increasingly difficult to overturn, with later candidates exerting progressively weaker influence on the final prediction. Motivated by these findings, we evaluate hierarchical partitioning and permutation-based inference, which improve accuracy by roughly 20 percentage points at $N=160$ on both HotpotQA and MIMIC. Overall, our results identify candidate-set scale as an important evaluation-protocol variable and show that strong small-option performance does not necessarily imply robust large-scale candidate comparison.
Chinese Translation
多项选择和候选人选择评估被广泛用于评估大型语言模型（LLM）的推理和决策能力，但大多数基准测试包含的候选集相对较小。随着候选空间的扩展，从这些设置得出的结论是否仍然有效尚不清楚。我们系统地评估了随着竞争候选人数量的增加，LLM的表现，并发现任务、提示策略和模型规模的准确性显著下降。控制分析表明，标准的长上下文检索解释无法完全解释这种下降。相反，我们识别出两种系统性失效模式。首先，黄金边际崩溃：正确答案与最强干扰项之间的得分差距逐渐缩小，主要是由于对正确答案信心的减弱。其次，早期候选人的偏好变得越来越难以推翻，后续候选人对最终预测的影响逐渐减弱。基于这些发现，我们评估了层次划分和基于置换的推理方法，这在HotpotQA和MIMIC上分别提高了约20个百分点的准确性（$N=160$）。总体而言，我们的结果表明候选集规模是一个重要的评估协议变量，并且强小选项性能并不一定意味着在大规模候选比较中表现稳健。
cs.AI / 144 / 2609.32817
Right Answer, Wrong Reason: Accuracy, Consistency, and Consensus Are Misleading Indicators of LLM Faithfulness in Clinical Decision Support
正确答案，错误理由：准确性、一致性和共识是误导性指标，无法反映大型语言模型在临床决策支持中的可信度
Abstract
Clinical Large Language Models (LLMs) achieve strong medical-exam accuracy; however, a correct answer does not guarantee that the explanation names the concepts that actually drove the decision. We introduce three lightweight, directly interpretable metrics for this faithfulness gap: the Explanation Stability Index (ESI), which measures reasoning consistency across repeated queries; the Causal Faithfulness Score (CFS), which tests whether cited concepts drive predictions via concept ablation; and the Perturbation Stability Score (PSS), which measures robustness to semantic-preserving paraphrases. By evaluating six LLMs on 150 MedQA-USMLE questions (900 model-question observations), we found that only 23.3% of the cited clinical concepts were causally necessary. Correct answers had lower CFS than incorrect answers (0.212 vs. 0.398), answer consistency negatively predicted CFS (Spearman r = -0.466), and model pairs could agree on answers while sharing only 8.8% of cited reasoning concepts. These results show that accuracy, consistency, and consensus are incomplete safety signals for clinical decision-making support. The evidence is behavioral rather than mechanistic: concept ablation tests counterfactual sensitivity of outputs, not internal circuits.
Chinese Translation
临床大型语言模型（LLMs）在医学考试中表现出强大的准确性；然而，正确的答案并不保证其解释所提及的概念实际上推动了决策的形成。我们提出了三种轻量级、直接可解释的指标来衡量这种可信度差距：解释稳定性指数（Explanation Stability Index, ESI），用于衡量重复查询中的推理一致性；因果可信度评分（Causal Faithfulness Score, CFS），用于测试所引用的概念是否通过概念消融推动预测；以及扰动稳定性评分（Perturbation Stability Score, PSS），用于衡量对语义保留的释义的鲁棒性。通过对六个LLM在150个MedQA-USMLE问题（900个模型-问题观测）上的评估，我们发现只有23.3%的引用临床概念是因果必要的。正确答案的CFS低于错误答案（0.212对0.398），答案一致性对CFS有负向预测（Spearman r = -0.466），而模型对可以在答案上达成一致，但仅共享8.8%的引用推理概念。这些结果表明，准确性、一致性和共识并不是临床决策支持的完整安全信号。证据是行为性的而非机械性的：概念消融测试输出的反事实敏感性，而不是内部电路。
cs.AI / 145 / 2609.32818
When Can First-Order Models of Fine-Tuning Bound Forgetting?
何时一阶模型的微调能够界定遗忘的边界？
Abstract
Fine-tuning a language model on new data can make it forget facts that it should keep. We ask whether measurements taken at the start of a fine-tuning run can bound, for each protected fact, the probability that the run makes the model forget it. In LoRA fine-tuning with stochastic gradient descent on models from 0.6B to 14B parameters, a first-order response model estimated by finite-difference probes predicts changes of per-fact margins with correlation 0.974-0.998. Predictions of forgetting built on this model nevertheless failed, because forgetting requires parameter changes far outside the region in which the model was validated. The probes can, however, bound the probability that a margin first falls below a boundary near zero: we derive Freedman and Azuma first-passage bounds for a linear surrogate of the margin and test on new runs whether they hold for the model. The bounds contain a term R that measures how much the response coefficients change during the run. The simplified Freedman bound, which sets R = 0, certified most facts but was violated in 14 of 112 conditions, and every fact on which it was violated had R >= a, where a is the distance of the fact's margin to the boundary. The complete Freedman bound certifies only facts with R < a, and it held in every condition. On the violated facts, the spread of the margin across test runs was a median of 14.6 times the prediction of the response model, so the failures are breakdowns of the model, and in our data they occurred only where R >= a. We found this pattern post hoc and tested it in two preregistered confirmatory studies with 43 new conditions: the complete bound held in all of them, and the simplified bound failed there on only 3 facts, each with R >= a. First-order models of fine-tuning can thus bound forgetting on the facts whose response coefficients change by less than their distance to the boundary.
Chinese Translation
对语言模型进行微调以适应新数据可能导致其遗忘应当保留的事实。我们探讨在微调开始时所做的测量是否能够为每个受保护的事实界定该微调过程使模型遗忘该事实的概率。在对参数从0.6B到14B的模型进行随机梯度下降的LoRA微调中，通过有限差分探针估计的一阶响应模型预测每个事实的边际变化，其相关性为0.974-0.998。然而，基于该模型的遗忘预测失败，因为遗忘需要参数变化远超出模型验证的区域。然而，这些探针可以界定边际首次低于接近零的边界的概率：我们推导了边际线性替代的Freedman和Azuma首次通过界限，并在新的微调过程中测试其是否适用于模型。这些界限包含一个项R，用于测量在微调过程中响应系数的变化量。简化的Freedman界限（设定R = 0）认证了大多数事实，但在112个条件中有14个条件被违反，且每个被违反的事实都有R >= a，其中a是该事实边际与边界的距离。完整的Freedman界限仅认证R < a的事实，并在每个条件下均成立。在被违反的事实中，测试运行中边际的分布中位数是响应模型预测的14.6倍，因此这些失败是模型的崩溃，而在我们的数据中，仅在R >= a的情况下发生。我们在事后发现了这一模式，并在两个预注册的确认性研究中进行了测试，涉及43个新条件：完整的界限在所有条件中均成立，而简化的界限在仅3个事实上失败，每个事实的R均为R >= a。因此，一阶微调模型可以界定那些响应系数变化小于其与边界距离的事实的遗忘。
cs.AI / 146 / 2609.32819
Rank Collapse Is Recoverable, Growing $|Q|$ Is Not: Out-of-Sample Early Warning for Value Divergence in High-UTD Soft Actor-Critic
秩崩溃是可恢复的，而价值增长 $|Q|$ 不是：高更新-数据比率软演员-评论员的样本外早期预警
Abstract
Raising the update-to-data (UTD) ratio breaks off-policy critics in two ways grouped as "plasticity loss": collapsing representations and growing value magnitude $|Q|$. We separate them in Soft Actor-Critic (SAC) with scaled critics (width 2048, no normalization). Collapse is survivable: at UTD ratio 16, HalfCheetah critics with most units dormant keep learning, and the training guard, which stops runs whose loss or $|Q|$ explodes, never flags them. Within one high-UTD SAC configuration, runs start close together, and how far a critic's $\log_{10}|Q|$ has climbed by step 15k, its early growth, ranks the runs by how soon the guard flags them. At 15k, a flagged run's $|Q|$ sits a median of over a hundredfold below its flag level, yet the climb's rate already orders the flags (Harrell's C and out-of-sample AUC 0.78 on Walker2d, 0.98 on Ant, at UTD ratio 4). Dormancy does not. Aborting on this rate saves about a tenth of held-out Walker2d compute and stays net-positive live. A LayerNorm critic lowers the rate, removes the flag on Walker2d at UTD ratio 4 and lowers the return.
Chinese Translation
提高更新-数据（UTD）比率以两种方式破坏了离线策略评论员，统称为“可塑性损失”：崩溃的表示和增长的价值幅度 $|Q|$。我们在具有缩放评论员（宽度2048，无标准化）的软演员-评论员（SAC）中将它们分开。崩溃是可以存活的：在UTD比率为16时，HalfCheetah评论员中大多数单元处于休眠状态仍然保持学习，而训练守卫在损失或 $|Q|$ 爆炸时停止运行，从未标记它们。在一个高UTD SAC配置中，运行开始时彼此接近，而评论员的 $ ext{log}_{10}|Q|$ 在第15k步时的增长幅度，其早期增长，按守卫标记它们的时间对运行进行排名。在15k时，被标记的运行的 $|Q|$ 中位数超过标记水平的百倍以下，但增长率已经对标记进行了排序（Harrell的C和样本外AUC在Walker2d上为0.78，在Ant上为0.98，UTD比率为4）。休眠状态则不然。在这个增长率下中止可以节省大约十分之一的保留Walker2d计算，并保持净正活跃。层归一化评论员降低了增长率，在UTD比率为4时移除了Walker2d上的标记，并降低了回报。
cs.AI / 147 / 2609.32821
Routing Drift Alone Does Not Diagnose Failure in Merged MoE LLMs
单靠路由漂移无法诊断合并的专家混合大语言模型中的故障
Abstract
Model merging efficiently combines specialized large language models (LLMs) without joint retraining, but can substantially alter expert routing in Mixture-of-Experts (MoE) models. Such \emph{routing drift} is often interpreted as routing failure, raising a fundamental question that remains unclear: \emph{does routing drift after MoE merging actually indicate routing failure, and what evidence should justify repair?} We investigate these questions across DeepSeekMoE, OLMoE, and Qwen3-MoE proposing a routing analysis toolkit for controlled counterfactual interventions and token-level analysis. By crossing source and merged router inputs and parameters, we attribute most expert reassignments to input shifts rather than parameter changes at the same layer. However, source-relative routing differences poorly predict next-token likelihood gains from source-route restoration, and different expert selections can produce directionally similar mixture outputs. We therefore operationalize routing failure as \textit{task loss recoverable under a specified routing intervention, with non-routing parameters fixed.} These tests detect recoverable loss under deliberate router corruption, whereas source-route restoration does not establish reliable task benefits in the evaluated merged models. Motivated by these, we propose \emph{Selective Router Repair (SRR)} as a case study, and find that source-specialist token-likelihood advantages do not reliably identify beneficial local corrections. Together, these findings show that \textbf{routing drift alone is insufficient evidence of routing failure}: source-informed corrections must be judged by their task-level intervention effects. The analysis toolkit and SRR code are released.
Chinese Translation
模型合并有效地结合了专门的大语言模型（LLMs），而无需联合重训练，但可能会显著改变专家路由在专家混合（MoE）模型中的表现。这种 extit{路由漂移}常常被解读为路由故障，这引发了一个基本问题： extit{在MoE合并后，路由漂移是否真的表明路由故障，以及什么证据应当证明需要修复？}我们在DeepSeekMoE、OLMoE和Qwen3-MoE中研究这些问题，提出了一种路由分析工具包，用于控制反事实干预和令牌级分析。通过交叉源路由器和合并路由器的输入及参数，我们将大多数专家重新分配归因于输入变化，而非同一层的参数变化。然而，源相对路由差异对源路由恢复的下一个令牌的可能性增益预测效果不佳，而不同的专家选择可以产生方向上相似的混合输出。因此，我们将路由故障操作化为 extit{在指定路由干预下可恢复的任务损失，非路由参数保持不变。}这些测试在故意的路由器损坏下检测到可恢复的损失，而源路由恢复在评估的合并模型中并未建立可靠的任务收益。基于这些，我们提出了 extit{选择性路由器修复（SRR）}作为案例研究，并发现源专家的令牌可能性优势并不能可靠地识别有益的局部修正。综合来看，这些发现表明 extbf{单靠路由漂移不足以作为路由故障的证据}：源信息修正必须通过其任务级干预效果进行评判。分析工具包和SRR代码已发布。
cs.AI / 148 / 2609.32825
The Decomposition Tax: LLM Pipelines Lose Up to 40 Accuracy Points at Their Own Interfaces
分解税：LLM管道在自身接口处损失高达40个准确度点
Abstract
A four-stage LLM pipeline gives up as much as 40.5 accuracy points at its own interfaces (gemma-3-12B on MATH-500, Holm-corrected p = 1.66e-19; the largest tax in the primary family). We hold model, problem, stages, stage prompts and completion budget fixed, vary only whether each stage can still see the original problem, and call the accuracy difference the decomposition tax. Across 21 open-weight models from nine organisations, on GSM-Hard and MATH-500 at n = 200 paired items per cell, 70 of 118 primary-family tests survive Benjamini-Hochberg correction and 54 survive Holm. On GSM-Hard, a placebo recovers nothing: it carries at least 60% of the extra tokens and at most one word of the problem. Builders design a pipeline one stage at a time, and its bill arrives at the interfaces between stages. Rewriting one stage's instruction moves gemma-3-12B's tax from 4.5 to 36.5 points, and adding "every relationship stated between them" to a stage that lists the numerical quantities lowers the tax on 9 of 9 models on MATH-500. Re-grounding, which shows a stage the original problem again, belongs after the loss. With one lossy interface, re-grounding the stage after it beats re-grounding the stage before it on 7 of 7 models on both benchmarks; on MATH-500 the earlier repair is worse than none on 7 of 7. Newer models still pay: gemma-4-12B gives up 37.0 points, and the repair holds on all three of the newest models we test. A sealed held-out test refuted a stronger rule we registered, which predicted the paying stage from the interface and receiver types, so we locate the tax by measuring one stage at a time. The prescription has two parts: re-ground the stage after the lossy interface, and if a stage must list the quantities, tell it to keep the relationships.
Chinese Translation
一个四阶段的LLM管道在其自身接口处损失高达40.5个准确度点（gemma-3-12B在MATH-500上的表现，Holm校正p = 1.66e-19；这是主要家族中最大的损失）。我们固定模型、问题、阶段、阶段提示和完成预算，仅改变每个阶段是否仍能看到原始问题，并将准确度差异称为分解税。在来自九个组织的21个开放权重模型中，在GSM-Hard和MATH-500上，每个单元n = 200对配对项目，118个主要家族测试中有70个通过了Benjamini-Hochberg校正，54个通过了Holm校正。在GSM-Hard上，安慰剂没有恢复任何内容：它至少携带60%的额外标记，最多一个问题的单词。构建者一次设计一个阶段，其费用在阶段之间的接口处产生。重写一个阶段的指令将gemma-3-12B的损失从4.5点提高到36.5点，而在列出数值量的阶段中添加“它们之间声明的每一个关系”则降低了在MATH-500上9个模型的损失。重新归位，即再次向一个阶段展示原始问题，属于损失之后。在一个有损失的接口下，损失后重新归位该阶段在两个基准的7个模型中优于损失前的重新归位；在MATH-500上，早期的修复在7个模型中比不修复还要糟糕。更新的模型仍然付出代价：gemma-4-12B损失了37.0点，并且修复在我们测试的最新三个模型中均有效。一个封闭的保留测试驳斥了我们注册的一个更强的规则，该规则预测了从接口和接收器类型支付的阶段，因此我们通过逐个测量阶段来定位损失。该处方有两个部分：在有损失的接口后重新归位该阶段，如果一个阶段必须列出数量，则告诉它保持关系。
cs.AI / 149 / 2609.32827
Improving LLM Collaboration via Multi-Agent Preference Learning
通过多智能体偏好学习提升大型语言模型的协作能力
Abstract
Several works have explored multi-agent reinforcement learning (MARL) in LLM collaboration. However, constructing reliable rewards is difficult in practice, as complete and accurate metrics are often unavailable and hard to aggregate. Preference learning provides an alternative by learning from comparative human or AI feedback. Yet, its extension to multi-agent systems remains underexplored. To address this gap, we formulate preference-based multi-agent systems (MAS) from decentralized and centralized collaboration perspectives. We also introduce a general multi-agent preference learning framework (MAPL) to solve these problems. MAPL allows iterative updates by comparing the current solution with decentralized or centralized solutions generated by various agents. We instantiate MAPL using MARL from human feedback (MARLHF) with a learned reward model and multi-agent direct preference optimization (MADPO). Experiments on collaborative writing, coding, tool use, and travel planning show that MAPL can improve collaboration quality and efficiency while approaching the performance of MARL with fixed, well-defined rewards. Within MAPL, MARLHF generally outperforms MADPO on most tasks but remains sensitive to data coverage, agent and comparator models, and the underlying MARL algorithms.
Chinese Translation
已有多项研究探讨了大型语言模型（LLM）协作中的多智能体强化学习（MARL）。然而，在实践中构建可靠的奖励机制是困难的，因为完整且准确的度量标准往往不可用且难以聚合。偏好学习提供了一种替代方案，通过学习人类或人工智能的比较反馈来进行学习。然而，其在多智能体系统中的扩展仍然未得到充分探索。为了解决这一问题，我们从去中心化和中心化协作的角度构建了基于偏好的多智能体系统（MAS）。我们还提出了一个通用的多智能体偏好学习框架（MAPL）来解决这些问题。MAPL允许通过将当前解决方案与由不同智能体生成的去中心化或中心化解决方案进行比较，从而进行迭代更新。我们使用基于人类反馈的多智能体强化学习（MARLHF）和学习的奖励模型以及多智能体直接偏好优化（MADPO）来实例化MAPL。在协作写作、编码、工具使用和旅行规划的实验中，结果表明MAPL能够提高协作的质量和效率，同时接近使用固定、明确奖励的MARL的性能。在MAPL框架内，MARLHF在大多数任务上通常优于MADPO，但对数据覆盖、智能体和比较模型以及基础的MARL算法仍然敏感。
cs.AI / 150 / 2609.32835
FinancialAuditBench: Benchmark Construction under Differential Privacy Using Real-World Priors
FinancialAuditBench：基于真实世界先验的差分隐私基准构建
Abstract
As AI agents are becoming widely adopted in the financial services industry, careful measurement is essential to understand where they can be reliably deployed and where oversight and professional review remain necessary. Such measurement, however, is constrained by limited access to proprietary or privacy-sensitive data. Existing benchmarks therefore often rely on publicly available data, human- and/or LLM-authored tasks, or simplified settings. We introduce FinancialAuditBench, a benchmark for evaluating agents on financial statement audit tasks, along with a framework for systematically generating synthetic engagements. Our task generation framework leverages differentially private aggregate statistics from historical audits along with audit expertise contributed through over 1,100 hours of benchmark development and review. FinancialAuditBench consists of 90 tasks spanning workpaper completion and review across six synthetic audit engagements, each containing an average of 179 files. Evaluation on eleven frontier models shows that while agents complete substantial portions of staff-level audit tasks well, they sometimes perform inappropriate procedures or produce incorrect documentation. Beyond financial auditing, our framework offers an approach for systematically generating synthetic tasks for model evaluation and training in privacy-sensitive domains.
Chinese Translation
随着人工智能代理在金融服务行业的广泛应用，仔细的测量对于理解它们可以可靠部署的领域以及需要监督和专业审查的领域至关重要。然而，这种测量受到对专有或隐私敏感数据的有限访问的限制。因此，现有基准通常依赖于公开可用的数据、人类和/或大型语言模型（LLM）创作的任务或简化的设置。我们介绍了FinancialAuditBench，这是一个用于评估代理在财务报表审计任务上的基准，以及一个系统生成合成审计任务的框架。我们的任务生成框架利用了来自历史审计的差分隐私聚合统计数据，以及通过超过1100小时的基准开发和审查贡献的审计专业知识。FinancialAuditBench包含90个任务，涵盖六个合成审计项目中的工作底稿完成和审查，每个项目平均包含179个文件。在对十一种前沿模型的评估中，尽管代理能够很好地完成大量的员工级审计任务，但它们有时会执行不当的程序或生成不正确的文档。除了财务审计之外，我们的框架还提供了一种在隐私敏感领域系统生成合成任务以进行模型评估和训练的方法。
cs.AI / 151 / 2609.32870
Counterfactual Self-Evolving Agents for Evidence-Grounded Reasoning
基于证据推理的反事实自我进化代理
Abstract
Self-play proposer--solver methods improve reasoning by generating tasks and learning from verified solutions. However, for evidence-identifiable tasks, where case-specific evidence and domain knowledge determine a checkable answer, self-play requires generating plausible cases whose answers can be independently verified. We introduce counterfactual self-evolution, which generates counterfactual context for reconsidering the original case. A trainable Proposer constructs targeted evidence edits and describes potential outcome changes with causal explanations. We handcraft an expert-verified counterfactual instruction-tuning dataset to teach the Proposer to generate high-quality counterfactuals across a broad range of action--outcome scenarios. Each counterfactual instruction-tuning example specifies an edit within a defined category and explains its hypothesized causal effect on the decision, teaching the Proposer to reason systematically about what changes and why. We instruction-tune the Proposer on these examples, then formulate a fine-tuning reward that integrates feedback from the Solver and Verifier. Across diverse counterfactual scenarios, this reward favors high-quality counterfactuals and warranted revisions, while penalizing changes that overturn correct decisions. The counterfactual context aims to correct errors and strengthen confidence in correct decisions. Accepted counterfactuals accumulate in memory that supplies in-context evidence to the frozen Solver; the Solver adapts through evolving context rather than weight updates. We apply the framework to clinical reasoning, fact verification, and business reasoning. Our evaluation tracks performance over successive rounds as counterfactual memory grows, including transfer to harder cases. Our method achieves superior results across diverse frontier models.
Chinese Translation
自我对弈的提议者-求解者方法通过生成任务并从经过验证的解决方案中学习来改善推理。然而，对于证据可识别的任务，其中案例特定的证据和领域知识决定了可检查的答案，自我对弈需要生成答案可以独立验证的合理案例。我们引入反事实自我进化，生成反事实背景以重新考虑原始案例。一个可训练的提议者构建针对性的证据编辑，并用因果解释描述潜在的结果变化。我们手工制作了一个专家验证的反事实指令调优数据集，以教导提议者在广泛的行动-结果场景中生成高质量的反事实。每个反事实指令调优示例指定了一个在定义类别内的编辑，并解释其对决策的假设因果影响，教导提议者系统性地推理变化的内容及其原因。我们在这些示例上对提议者进行指令调优，然后制定一个细调奖励，该奖励整合来自求解者和验证者的反馈。在多样的反事实场景中，该奖励倾向于高质量的反事实和合理的修订，同时惩罚推翻正确决策的变化。反事实背景旨在纠正错误并增强对正确决策的信心。被接受的反事实在内存中积累，为冻结的求解者提供上下文证据；求解者通过演变的背景而非权重更新进行适应。我们将该框架应用于临床推理、事实验证和商业推理。我们的评估跟踪在反事实记忆增长的过程中，随着连续轮次的表现，包括向更难案例的转移。我们的方法在多样的前沿模型中取得了优越的结果。
cs.AI / 152 / 2609.32885
Can LLMs Predict the Future? A Brier Score Analysis of Prediction Markets
大型语言模型能预测未来吗？对预测市场的Brier评分分析
Abstract
We study whether model upgrades improve probability estimates for prediction-market questions. Our Resolved Market Forecasting (RMF) benchmark contains 3,000 resolved binary questions across nine domains, on which we evaluate six Claude and Qwen model variants using a question-only, zero-shot protocol. We assess Brier scores relative to an empirical base-rate predictor, examine their Murphy decomposition, and compare models through paired differences, with results stratified by event category and timing relative to training cutoffs. In the reported post-cutoff stratum, the four Claude models achieve Brier scores of 0.183-0.192, improving on the base-rate reference by 0.024-0.033. Qwen 32B does not significantly outperform that reference, although its paired Brier is 0.024 lower than that of the 7B checkpoint. The evaluated Claude version and tier upgrades yield no significant improvement. Within-model differences across event categories exceed the observed differences among Claude variants. These results show why forecasting scores should be interpreted alongside simple probability baselines and question composition: under this protocol, newer versions or higher model tiers do not consistently produce more accurate probabilities.
Chinese Translation
我们研究模型升级是否改善了对预测市场问题的概率估计。我们的已解决市场预测（Resolved Market Forecasting, RMF）基准包含3000个已解决的二元问题，涵盖九个领域，在此基础上我们使用仅问题的零样本协议评估六个Claude和Qwen模型变体。我们相对于经验基准预测器评估Brier评分，检查其Murphy分解，并通过配对差异比较模型，结果按事件类别和相对于训练截止日期的时间进行分层。在报告的截止后分层中，四个Claude模型的Brier评分为0.183-0.192，较基准参考提高了0.024-0.033。Qwen 32B并未显著优于该参考，尽管其配对Brier评分比7B检查点低0.024。评估的Claude版本和层级升级未带来显著改善。不同事件类别之间的模型内差异超过了Claude变体之间观察到的差异。这些结果表明，预测评分应与简单概率基线和问题构成一起解读：在该协议下，更新版本或更高模型层级并不总是产生更准确的概率。
cs.AI / 153 / 2609.32886
StraTune: Adaptive Selection of Revision Operators for Self-Evolving LLM Skills
StraTune：自我演化大语言模型技能的修订操作符自适应选择
Abstract
Large language models (LLMs) can learn reusable textual skills from execution feedback without updating their parameters, but effectively deciding how to revise these skills remains a key challenge. Existing methods typically rely on a fixed revision operator, a search strategy and the revision forms applied under it. However, we observe that no single revision operator consistently performs best across tasks, and repeatedly applying an unsuitable operator can limit further improvement. We propose StraTune (strategy-guided skill tuning), which lets a frozen optimizer LLM choose the revision operator at every round from the optimization state, which is defined as the current execution feedback together with the recorded outcomes of earlier strategies and forms. Candidate skills from every revision operator pass one candidate evaluation, which screens for gains and regressions on a small sample set and validates them on a larger one, and every outcome is written back to the optimization state for later choices. Across four benchmarks and two LLM settings, StraTune outperforms all five baselines in most settings. Ablations attribute the gains to the adaptive choice of the revision operator, since fixed, random, scheduled, and bandit strategy choices all score lower, and skills learned with a small target LLM also improve a stronger one. Code and learned skills are available at https://github.com/seai-lab/StraTune.
Chinese Translation
大型语言模型（LLMs）可以通过执行反馈学习可重用的文本技能，而无需更新其参数，但有效决定如何修订这些技能仍然是一个关键挑战。现有方法通常依赖于固定的修订操作符、搜索策略及其应用的修订形式。然而，我们观察到没有单一的修订操作符在各任务中始终表现最佳，反复应用不适合的操作符可能限制进一步的改进。我们提出了StraTune（策略引导的技能调优），它允许一个冻结的优化器 LLM 在每一轮从优化状态中选择修订操作符，优化状态定义为当前的执行反馈以及早期策略和形式的记录结果。每个修订操作符的候选技能经过一次候选评估，该评估在小样本集上筛选出增益和回退，并在更大的样本集上进行验证，每个结果都写回优化状态以供后续选择。在四个基准测试和两个 LLM 设置中，StraTune 在大多数设置中优于所有五个基线。消融实验将增益归因于修订操作符的自适应选择，因为固定、随机、预定和强盗策略选择的得分均较低，并且使用小目标 LLM 学习的技能也能提升更强的 LLM。代码和学习到的技能可在 https://github.com/seai-lab/StraTune 获取。
cs.AI / 154 / 2609.32900
Constraints Are Graphs, Not Chains: Exact Decoding for Diffusion Language Models
约束是图，而不是链：扩散语言模型的精确解码
Abstract
Diffusion language models (dLLMs) predict masked positions in arbitrary order, but their exact constrained decoders still encode constraints as sequential languages, whose state must track every unresolved dependency between positions. For relational constraints this encoding grows exponentially: for same-order copy, every finite automaton needs $4^k$ states, deterministic or nondeterministic, and every context-free grammar has size $2^{\Omega(k)}$, while the factor graph of the same relation has size $O(k)$ and a 16-entry peak table. We introduce FactorDLM, a training-free decoder that represents finite-domain relations as a factor graph and, at each denoising step, conditions the model's mean-field prediction on that graph exactly by variable elimination. Decoding cost then grows exponentially with the induced width of the constraint graph, which replaces automaton size as the governing parameter. Because a finite automaton is a chain-shaped factor graph, one compiler enforces syntax and nonlocal relations together: on JSON records with cross-field references, a schema automaton alone leaves references dangling, relational factors alone produce malformed JSON, and the combined plan is valid on both counts, including on records of variable length. Across nine relational benchmarks and three backbones, every output satisfies every declared constraint at 0.4-6.9% projection overhead, where unconstrained decoding is 0-79% valid, and compiled projection answers repeated queries 13.6x faster than CP-SAT with eight parallel workers. Because model-free rules solve three of five standard benchmarks, we construct benchmarks with exact chance and fixed-template floors, on which selecting among exact constrained samples beats greedy projection. Which encoding is cheaper, sequential state or direct factors, depends on the constraint and is computable before decoding begins.
Chinese Translation
扩散语言模型（dLLMs）以任意顺序预测被遮蔽的位置，但它们的精确约束解码器仍然将约束编码为顺序语言，其状态必须跟踪位置之间的每个未解决依赖关系。对于关系约束，这种编码呈指数级增长：对于同阶复制，每个有限自动机需要 $4^k$ 个状态，无论是确定性还是非确定性，而每个上下文无关文法的大小为 $2^{ ext{Ω}(k)}$，而同一关系的因子图大小为 $O(k)$，并具有一个16条目的峰值表。我们引入了 FactorDLM，这是一种无训练解码器，将有限域关系表示为因子图，并在每个去噪步骤中，通过变量消除准确地将模型的均值场预测条件化于该图。解码成本因此随着约束图的诱导宽度呈指数增长，这取代了自动机大小作为主导参数。由于有限自动机是链状因子图，因此一个编译器同时强制语法和非局部关系：在具有跨字段引用的 JSON 记录上，单独的模式自动机会留下悬空引用，单独的关系因子会生成格式错误的 JSON，而组合计划在这两方面都是有效的，包括在可变长度记录上。在九个关系基准和三个骨干网络中，每个输出在 0.4-6.9% 的投影开销下满足每个声明的约束，而无约束解码的有效性为 0-79%，并且编译后的投影答案比 CP-SAT 快 13.6 倍，使用八个并行工作者。由于无模型规则解决了五个标准基准中的三个，我们构建了具有精确机会和固定模板底线的基准，在这些基准上，从精确约束样本中选择胜过贪婪投影。哪种编码更便宜，顺序状态还是直接因子，取决于约束，并且在解码开始之前是可计算的。
cs.AI / 155 / 2609.32907
Logical subspace in LLMs
大型语言模型中的逻辑子空间
Abstract
Recent work has identified a human brain network specialized for abstract formal reasoning (Kean et al., 2025). Does the same hold true in language models? To answer this question, we introduce the minimal viable subspace (MVS) method, which searches for the lowest-rank activation subspace at a layer that preserves task performance when everything outside that subspace is ablated. Using MVS, we demonstrate low-rank subspaces supporting logical inference on Gemma and Qwen models. Furthermore, these subspaces exhibit a clear dissociation from model capacities on other tasks, such that retaining these late logic subspaces preserves inference while impairing factual knowledge, working memory, cognitive control, and arithmetic. Conversely, ablating them reduces logical inference accuracy to chance while largely sparing these other capacities. Our results suggest a functionally localizable core machinery for logic akin to that in the human brain.
Chinese Translation
近期的研究发现，人类大脑中存在一个专门用于抽象形式推理的网络（Kean et al., 2025）。在语言模型中是否也存在类似的情况？为了解答这个问题，我们引入了最小可行子空间（MVS）方法，该方法在一个层次上搜索最低秩激活子空间，以保持任务性能，即使在该子空间之外的所有内容被消除的情况下。通过使用MVS，我们展示了支持逻辑推理的低秩子空间在Gemma和Qwen模型中存在。此外，这些子空间与模型在其他任务上的能力表现出明显的分离，即保留这些后期逻辑子空间可以保持推理能力，但会损害事实知识、工作记忆、认知控制和算术能力。相反，消除这些子空间会将逻辑推理的准确性降低到随机水平，同时在很大程度上保留其他能力。我们的结果表明，逻辑的核心机制在功能上是可定位的，类似于人类大脑中的机制。
cs.AI / 156 / 2609.32917
Planner-as-Router: Joint Plan-Time Model Routing for Cost-Efficient Multi-Agent Workflows
规划者作为路由器：成本高效的多智能体工作流的联合计划时间模型路由
Abstract
Running large language model (LLM) agents in production gets expensive fast. A frontier model (the largest, most capable tier) is accurate but can cost 25 times what a small model costs per token, and the gap compounds once a workflow chains several calls together. Planner-as-Router (PaR) attacks this from a different angle. Instead of leaving model-tier selection to some component downstream, it folds the choice into planning itself. As the planner breaks a query into subtasks, it also assigns each one a model size tier (small, mid, or frontier, ordered by capability and price), so the dependencies between subtasks are visible before any specialist runs. Unlike per-call routers such as cascade routing, which look at one node at a time, PaR sees the whole workflow up front and needs no separate router model or training data. We evaluate PaR with EntBench, a benchmark of 54 enterprise agentic tasks across seven classes, graded by actually running the generated Structured Query Language (SQL) and MongoDB queries against live databases. Over 1,157 evaluations spanning eight routers and three seeds, PaR stays on the observed cost-accuracy frontier. It matches a sink-frontier heuristic (frontier model on terminal nodes only) in accuracy at comparable cost and a faithful FrugalGPT cascade at lower cost, and cuts cost 44% against all-frontier routing while giving up 2.9 points of accuracy. Several accuracy gaps fall inside the plus-or-minus six-point confidence interval of a 54-task study, so we frame PaR's advantage as frontier position rather than a clean accuracy win. We also report a preliminary observation, not a validated result: a small pilot hints that cheap routing may carry a hidden compounding penalty on compositional workflows, which we frame as a hypothesis for future measurement. PaR, EntBench, and all evaluation code are open source.
Chinese Translation
在生产环境中运行大型语言模型（LLM）代理的成本迅速增加。前沿模型（最大、最强大的层级）虽然准确，但每个令牌的成本可能是小型模型的25倍，并且一旦工作流将多个调用串联在一起，这一差距会进一步加大。规划者作为路由器（Planner-as-Router, PaR）从不同的角度解决了这个问题。它并不将模型层级的选择留给下游的某个组件，而是将这一选择融入到规划过程中。当规划者将查询分解为子任务时，它还为每个子任务分配一个模型大小层级（小型、中型或前沿，按能力和价格排序），因此子任务之间的依赖关系在任何专家运行之前都是可见的。与逐个节点查看的每次调用路由器（如级联路由）不同，PaR能够提前看到整个工作流，并且不需要单独的路由模型或训练数据。我们使用EntBench对PaR进行了评估，这是一个涵盖七个类别的54个企业代理任务的基准，通过实际运行生成的结构化查询语言（SQL）和MongoDB查询对实时数据库进行评分。在跨越八个路由器和三个种子的1,157次评估中，PaR保持在观察到的成本-准确性前沿。它在可比成本下的准确性与仅在终端节点使用前沿模型的sink-frontier启发式方法相匹配，并且在成本更低的情况下与忠实的FrugalGPT级联相当，同时在与全前沿路由相比时降低了44%的成本，尽管牺牲了2.9个准确性点。几个准确性差距落在54任务研究的正负六点置信区间内，因此我们将PaR的优势框架视为前沿位置，而不是简单的准确性胜利。我们还报告了一项初步观察，而非经过验证的结果：一个小型试点暗示，廉价路由可能在组合工作流上带来隐藏的复合惩罚，我们将其框架化为未来测量的假设。PaR、EntBench及所有评估代码均为开源。
cs.AI / 157 / 2609.32922
Precision As You Need: Stochastic Computing Is a Dense Adaptive Quantizer
精准如你所需：随机计算是一种密集自适应量化器
Abstract
Matrix multiplications dominate the inference cost of modern transformer-based vision models, yet existing efficiency techniques such as post-training quantization and mixed-precision inference are largely limited to the small set of fixed-width formats (INT4, INT8, BF16, and FP16) supported by conventional accelerators. We revisit stochastic computing (SC) as a way to lift this constraint: viewed as a dense adaptive quantizer, SC controls precision by bit-stream length L rather than a fixed datapath, while each multiplication reduces to a single AND/XNOR gate. We build a GPU library that emulates SC matrix multiplication at scale, exposes stream lengths as first-class kernel arguments, and evaluates SC end-to-end on image classification, object detection and instance segmentation, class-conditional image generation, and visual world-model planning. On top of this substrate, we develop a dynamic per-row mixed-precision policy that assigns stream length per token or group at matched average budget, requires no retraining, and uses the same SC hardware across schedules. Across tasks, SC remains competitive with fixed-format INT quantization at matched bit budgets, while per-row mixed precision helps maintain accuracy at lower average stream lengths. These results provide software-level feasibility evidence that SC can serve as a dense-precision substrate for fine-grained mixed-precision inference on modern vision transformers.
Chinese Translation
矩阵乘法主导了现代基于变压器的视觉模型的推理成本，然而现有的效率技术如后训练量化和混合精度推理在很大程度上仅限于传统加速器支持的一小部分固定宽度格式（INT4、INT8、BF16 和 FP16）。我们重新审视随机计算（Stochastic Computing，SC）作为一种解除这一限制的方法：将 SC 视为一种密集自适应量化器，SC 通过比特流长度 L 而非固定数据路径来控制精度，同时每次乘法简化为一个 AND/XNOR 门。我们构建了一个 GPU 库，能够大规模模拟 SC 矩阵乘法，将流长度作为一等内核参数，并在图像分类、目标检测和实例分割、类别条件图像生成以及视觉世界模型规划等任务上对 SC 进行端到端评估。在此基础上，我们开发了一种动态按行混合精度策略，根据匹配的平均预算为每个标记或组分配流长度，无需重新训练，并在不同调度中使用相同的 SC 硬件。在各项任务中，SC 在匹配比特预算时仍与固定格式的 INT 量化保持竞争力，而按行混合精度有助于在较低的平均流长度下保持准确性。这些结果提供了软件层面可行性的证据，表明 SC 可以作为现代视觉变压器上细粒度混合精度推理的密集精度基础。
cs.AI / 158 / 2609.32923
TRACE: Learning to Self-Calibrate Wireless Digital Twins from ISAC Measurements
TRACE：从ISAC测量中学习自我校准无线数字双胞胎
Abstract
Wireless digital twins (DTs) rely on 3D environment models to predict radio propagation and support wireless-network decisions, yet these models are often initialized from imperfect 3D maps. Errors in building position, height, footprint, and orientation can therefore cause a high-fidelity propagation engine to simulate the wrong physical environment. In this paper, we study how a deployed wireless network can repair an existing DT using its own radio frequency (RF) measurements. In particular, we introduce Twin Residual Alignment and Calibration Engine (TRACE), a physics-grounded learning-based self-calibration framework that treats twin maintenance as residual alignment between the physical world and the current DT. Using the same sensing configuration as the physical measurements, TRACE ray-traces the current DT, coherently backprojects the measured and simulated RF onto a common world grid, and extracts the same local region around each building's current DT position. A multi-view corrector then fuses evidence across sensing nodes and neighboring buildings to predict a gated six-parameter correction per building, without relying on absolute layout or sensor ordering, and supports iterative correction through re-rendering. On 5,400 held-out samples from unseen simulated scenes at 28 GHz, TRACE reduces 3D position RMSE from 2.202 m to 0.302 m and yaw RMSE from 4.978{\deg} to 0.894{\deg}, outperforming ViT and U-Net baselines under changes in layout, building count, sensing-node count, and SNR. On measured 28 GHz RF data from the NIST outdoor courtyard, a model trained only on synthetic RF reduces mean planar wall-position error from 1.00 m to 7.8 cm, without measured-data fine-tuning or geometric labels. These results show that the discrepancy between measured and twin-rendered RF can serve as a learning signal for repairing a wireless DT.
Chinese Translation
无线数字双胞胎（DTs）依赖于三维环境模型来预测无线电传播并支持无线网络决策，但这些模型通常是从不完美的三维地图初始化的。因此，建筑物的位置、高度、占地面积和朝向的错误可能导致高保真传播引擎模拟错误的物理环境。本文研究了如何利用已部署的无线网络通过自身的射频（RF）测量来修复现有的数字双胞胎。我们特别介绍了双胞胎残差对齐与校准引擎（TRACE），这是一个基于物理的学习自校准框架，将双胞胎维护视为物理世界与当前数字双胞胎之间的残差对齐。TRACE使用与物理测量相同的传感配置，对当前数字双胞胎进行光线追踪，将测量和模拟的射频一致地反投影到一个共同的世界网格上，并提取每个建筑物当前数字双胞胎位置周围的相同局部区域。然后，多视角校正器融合传感节点和邻近建筑物的证据，以预测每个建筑物的六参数校正，而不依赖于绝对布局或传感器顺序，并通过重新渲染支持迭代校正。在28 GHz的未见模拟场景中，TRACE在5400个保留样本上将三维位置均方根误差（RMSE）从2.202米降低到0.302米，将偏航RMSE从4.978°降低到0.894°，在布局、建筑物数量、传感节点数量和信噪比变化下优于ViT和U-Net基线。在来自NIST户外庭院的28 GHz射频测量数据上，仅使用合成射频训练的模型将平均平面墙位置误差从1.00米降低到7.8厘米，而无需测量数据微调或几何标签。这些结果表明，测量的射频与双胞胎渲染的射频之间的差异可以作为修复无线数字双胞胎的学习信号。
cs.AI / 159 / 2609.32924
Diagnosing Sampled LLM Reasoning in Formal Geometry: Coverage, Realization, and Validity Evidence
在形式几何中诊断抽样的大型语言模型推理：覆盖性、实现性和有效性证据
Abstract
Repeated sampling can reveal a correct numerical answer without yielding either a reliable system output or a supported derivation. We present Coverage, Realization, and Validity Evidence (CRV), an evaluation protocol for sampled large language model (LLM) reasoning over formal geometry states. Coverage is answer availability, realization is readout accuracy on the frozen candidate pool, and validity evidence is a label-blinded critic judgment of derivational support rather than a proof certificate. CRV freezes each candidate pool before comparing readouts and analyzes covered failures by correct-answer multiplicity and within-problem discrimination. On HardShift441, a 441-problem set for which a reference solver leaves 406 problems unsolved, a LoRA-adapted Qwen2.5-7B generator obtains 24.2% average single-sample accuracy and 68.9% pass@16, whereas verifier-weighted self-consistency (WSC) reaches 38.0%. Readout accuracy is particularly low when the correct answer occurs only once or twice in the pool. In a separate constructed audit of 195 covered problems, the critic labels 12 correct-answer representatives as supported, 181 as refuted, and two as uncertain. These results show that coverage, realization, and validity evidence from the critic are distinct quantities and should be reported separately.
Chinese Translation
重复抽样可以揭示正确的数值答案，但并不一定产生可靠的系统输出或支持的推导。我们提出了覆盖性、实现性和有效性证据（CRV），这是一个针对抽样大型语言模型（LLM）在形式几何状态下推理的评估协议。覆盖性指答案的可用性，实现性是对冻结候选池的读出准确性，而有效性证据则是对推导支持的盲评判，而非证明证书。CRV在比较读出之前冻结每个候选池，并通过正确答案的多重性和问题内的区分分析覆盖失败。在HardShift441上，这是一个441个问题的集合，其中参考求解器未能解决406个问题，经过LoRA适配的Qwen2.5-7B生成器获得了24.2%的平均单样本准确率和68.9%的通过率@16，而验证器加权的自一致性（WSC）达到了38.0%。当正确答案在池中仅出现一次或两次时，读出准确性特别低。在对195个覆盖问题的单独构建审计中，评审者将12个正确答案代表标记为支持，181个标记为反驳，两个标记为不确定。这些结果表明，来自评审者的覆盖性、实现性和有效性证据是不同的量，应分别报告。
cs.AI / 160 / 2609.32964
The Commit-Abstain Circuit: Why Language Models Hallucinate Instead of Abstaining
承诺-放弃电路：为何语言模型选择幻觉而非放弃
Abstract
Language models (LMs) often hallucinate by committing to confident answers rather than abstaining, even when they do not have enough information to answer reliably. A large body of existing work mitigates hallucination through detection or abstention mechanisms, but leaves open how models internally arrive at the decision to commit or abstain in the first place. We study this decision through mechanistic analysis, framing hallucination as unsupported commitment: the model commits despite exhibiting signals of unanswerability. Using causal gating, we identify a Commit-Abstain Circuit (CAC), a sparse, causally localised subset of attention heads and MLP sublayers underlying this decision. Across ten LMs (3B-14B) from five families and three benchmarks, the CAC exhibits a recurring accumulate-yet-undercorrect pattern: commitment-promoting components build up commitment in earlier layers, while abstention-promoting components act later as corrective signals that are often insufficient to overturn the accumulated commitment. Building on this finding, a lightweight policy trained on CAC activations improves decision accuracy by 12.2 points over the model's intrinsic commit-abstain margin, reduces false abstentions by 2.5 times, transfers to unseen benchmarks, and extends to larger models (27B-35B). The CAC is both diagnostic, clarifying how models overcommit, and practical, enabling improved abstention decisions.
Chinese Translation
语言模型（LMs）通常会选择承诺自信的答案，而不是放弃，即使在它们没有足够的信息来可靠回答时。现有大量研究通过检测或放弃机制来减轻幻觉现象，但仍未解决模型内部如何做出承诺或放弃的决策。我们通过机制分析研究这一决策，将幻觉视为不支持的承诺：模型尽管表现出无法回答的信号，仍然选择承诺。通过因果门控，我们识别出一个承诺-放弃电路（Commit-Abstain Circuit, CAC），这是一个稀疏的、因果局部化的注意力头和多层感知器（MLP）子层的子集，支撑这一决策。在来自五个家族和三个基准的十个语言模型（3B-14B）中，CAC展现出一种反复出现的积累却未能纠正的模式：促进承诺的组件在早期层中积累承诺，而促进放弃的组件则在后期作为纠正信号出现，但往往不足以推翻已积累的承诺。基于这一发现，针对CAC激活训练的轻量级策略使决策准确性提高了12.2个百分点，减少了2.5倍的错误放弃，能够迁移到未见过的基准，并扩展到更大的模型（27B-35B）。CAC既具有诊断意义，阐明了模型过度承诺的原因，又具有实用性，能够改善放弃决策。
cs.AI / 161 / 2609.32965
Relic: From Multi-Agent Collaboration to Persistent Organizational Capability
Relic：从多智能体协作到持久的组织能力
Abstract
Multiple agents may often conflict in an organization: for example, one coding agent changes an interface in a repository, but another continues to develop on the old version where existing tests become stale. A conversation can resolve the episode, but when the participants change, what makes the lesson continue to govern the team? We introduce Relic, which turns recurring collaboration failures into organization-owned, executable protocols. Members reflect on visible work, propose rules, and govern their adoption. Adopted protocols bind triggers, responsibilities, required evidence, and execution consequences to the runtime, while remaining open to revision and retirement. In one traced case, repeated integration friction produces an interface-review rule that governs later pull requests and is revised as work continues. Across 360 controlled runs over ten software workloads and three models, Relic raises complete-contract delivery from 14.06% to 19.76% (+5.71 percentage points) over a matched structured team without the protocol lifecycle, improving all four verified production endpoints in every model stratum. Under fresh-member transfer, behavioral correctness is 25.4% with no inherited protocol, 34.6% with the same rules provided as readable text, and 41.2% with executable bindings, a +6.5-point advantage over text alone. On the full CooperBench benchmark, after excluding broken benchmark pairs, Relic achieves 367/477 (76.9%), establishing the best reported result among peer-structured systems. On the fixed 48-pair same-model subset, Relic also exceeds Solo (29/48 vs. 26/48), reversing the coordination loss exhibited by the official peer baseline. Together, these results show how collaboration experience can become persistent organizational state that remains useful beyond the members who created it.
Chinese Translation
在一个组织中，多个智能体往往会发生冲突：例如，一个编码智能体在代码库中更改了接口，但另一个智能体仍在旧版本上进行开发，导致现有测试失效。一次对话可以解决这一事件，但当参与者发生变化时，是什么使得这一经验继续影响团队？我们提出了Relic，它将反复出现的协作失败转化为组织拥有的可执行协议。成员们反思可见的工作，提出规则，并管理其采纳。被采纳的协议将触发条件、责任、所需证据和执行后果绑定到运行时，同时保持开放以供修订和废弃。在一个追踪案例中，重复的集成摩擦产生了一个接口审查规则，该规则管理后续的拉取请求，并随着工作的继续而被修订。在对十个软件工作负载和三个模型进行的360次受控运行中，Relic将完整合同交付率从14.06%提高到19.76%（+5.71个百分点），相较于没有协议生命周期的匹配结构团队，改善了每个模型层次中所有四个验证的生产端点。在新成员转移的情况下，行为正确性为25.4%（没有继承协议），34.6%（提供为可读文本的相同规则），以及41.2%（带有可执行绑定），相比仅有文本的情况下提高了6.5个百分点。在完整的CooperBench基准测试中，排除损坏的基准对后，Relic达到了367/477（76.9%），在同类结构系统中建立了最佳报告结果。在固定的48对相同模型子集上，Relic也超过了Solo（29/48对比26/48），扭转了官方对等基线所表现出的协调损失。这些结果共同表明，协作经验可以成为持久的组织状态，超越创造它的成员仍然具有实用性。
cs.AI / 162 / 2609.32990
Certified Long-Horizon Code Agent Evolution via Validation-Gated Skill Optimization
通过验证门控技能优化实现认证的长视野代码代理自我进化
Abstract
Long horizon agent self-evolution without model weight updates is essential for enabling deployed agents to accumulate reusable skills and improve over time. Prior self-evolution work has focused primarily on short-horizon tasks, while repository-level software engineering remains unexplored despite being an ideal testbed for long-horizon adaptation. In this setting, agents are required to solve streams of sequential tasks, navigate complex dependencies with evolving repositories and persistently store and reuse experience. Text-based skill optimization offers an efficient, non-parametric approach for such adaptation. However, existing methods often suffer from unstable updates, performance drawdown, and agent collapse over extended deployments. In this paper, we formalize the concept of in-context self-evolution and introduce VALVE, a validated-gated framework for long-horizon skill optimization. We establish finite convergence, provide theoretical guarantees for future-task gain and drawdown, and derive the validation and evaluation holdout sizes required for a prescribed tolerance, with leading-order scaling Empirically, our pipeline, VALVE achieves stable self-improvement over evolution horizon spanning more than 1,000 SWE tasks, with average final and peak gains of $14.9$ and $16.5$ points across three frontier models (GPT-5.5, Claude-4.6 and MiniMax-M2.7). The validation gate reduces average drawdown by 75% and produces an 11x more compact skill bank than ungated evolution. We further present extensive ablations identifying the design choices most critical to long-horizon skill evolution.
Chinese Translation
在不更新模型权重的情况下，实现长视野代理的自我进化对于使已部署的代理能够积累可重用技能并随时间改善至关重要。以往的自我进化研究主要集中在短视野任务上，而尽管存储库级软件工程是长视野适应的理想测试平台，但尚未得到充分探索。在这种情况下，代理需要解决一系列顺序任务，导航复杂的依赖关系，并持续存储和重用经验。基于文本的技能优化为这种适应提供了一种高效的非参数方法。然而，现有方法往往面临不稳定的更新、性能下降和代理在长期部署中的崩溃。在本文中，我们形式化了上下文自我进化的概念，并引入了VALVE，一个用于长视野技能优化的验证门控框架。我们建立了有限收敛性，为未来任务的增益和下降提供了理论保证，并推导出所需的验证和评估保留样本大小，以满足规定的容忍度，并进行领先阶数的缩放。通过实证研究，我们的管道VALVE在超过1,000个软件工程任务的进化视野中实现了稳定的自我改善，三个前沿模型（GPT-5.5、Claude-4.6和MiniMax-M2.7）的平均最终增益和峰值增益分别为14.9和16.5点。验证门控将平均下降减少了75%，并产生了比无门控进化更紧凑的技能库，紧凑度提高了11倍。我们还进行了广泛的消融实验，识别出对长视野技能进化至关重要的设计选择。
cs.AI / 163 / 2609.32993
X-Tree: Tokenizing Reusable Experience for Efficient Agent Generalization
X-Tree：为高效代理泛化而标记可重用经验
Abstract
Multi-step agents are trained on flat action streams: SFT and RLVR weight every token uniformly and ignore the sub-procedures that recur across tasks, the hierarchy that lets humans plan top-down from reusable routines. This structure sits unused, and flat training uses each scarce trajectory less fully than its content allows. Recent agents do use that structure, but only as LLM-written skills in context, never in the weights, so their gains do not generalize beyond retrieval. We instead recover this hierarchy from the data itself and train on it, with no LLM calls. Following text tokenizers, which build a vocabulary by counting alone, we score action spans by reusability and merge canonicalized actions into a reusable eXperience tree (X-Tree). Each X-Tree node captures how a frequent and success-bearing skill is composed from sub-skills, guiding efficient generalization. We integrate X-Tree into three training settings: offline RL, with each node as a training instance; online RLVR, with an adaptive skill bonus; and on-policy self-distillation, with X-Tree as the self-teacher's privileged context. Across WebArena, ScienceWorld, and WebShop at three model scales, X-Tree improves over standard recipes at matched data and budget by up to 4.5% SR on WebArena, 5.8% SR on ScienceWorld and 4.1% success on WebShop. Matched analyses attribute the gains to the X-Tree structure and the three integrations.
Chinese Translation
多步代理在平坦的动作流上进行训练：SFT 和 RLVR 对每个标记进行均匀加权，忽略了跨任务重复出现的子程序，以及让人类能够从可重用例程中自上而下进行规划的层次结构。这种结构未被充分利用，而平坦训练则未能充分利用每条稀缺轨迹的内容。最近的代理确实利用了这种结构，但仅作为上下文中由 LLM（大语言模型）编写的技能，而不是在权重中，因此它们的收益无法超出检索的范围。我们则从数据本身恢复这一层次结构并在其上进行训练，无需 LLM 调用。我们借鉴文本标记器，通过单纯计数构建词汇表，以可重用性为标准对动作范围进行评分，并将规范化的动作合并为可重用经验树（X-Tree）。每个 X-Tree 节点捕捉到一个频繁且成功的技能是如何由子技能组成的，从而指导高效的泛化。我们将 X-Tree 集成到三种训练设置中：离线强化学习（RL），将每个节点作为训练实例；在线 RLVR，配备自适应技能奖励；以及基于策略的自蒸馏，将 X-Tree 作为自教师的特权上下文。在 WebArena、ScienceWorld 和 WebShop 三个模型规模上，X-Tree 在匹配数据和预算的情况下，分别提高了 WebArena 的成功率 4.5%、ScienceWorld 的成功率 5.8% 和 WebShop 的成功率 4.1%。匹配分析将这些收益归因于 X-Tree 结构及其三种集成方式。
cs.AI / 164 / 2609.33010
Model-Aware Data Selection from In-and-Out Information Interplay
模型感知的数据选择：信息交互的内外关系
Abstract
LLMs are effective representations that assimilate vast amounts of knowledge during pretraining, but post-training is necessary for models to reliably access this knowledge and "know what they know." We observe an interesting rank equilibrium between knowledge stored in the weights and the data stream passing through the model. Across all model layers, we find that the hidden states (data stream) follow a U-shaped pattern, showing substantial compression in early layers and a steep rise during the late-layer decoding phase. In contrast, the weight rank follows an inverted U-shaped pattern, with very low rank in the early and late layers and high rank in the middle. We interpret this as an in-and-out information interplay: intermediate activations do not need to carry content that the weights can supply later, so they primarily preserve what the weights cannot provide. Motivated by this observation, we propose a model-aware data selection method, CAP (Counterfactual Assimilation Profile), which can determine whether a data candidate contains information accessible to the current model by utilizing the divergence gap in early- and late-layer representations between model-generated and reference responses. Across math, code, and science domains, CAP delivers 35.4% greater average improvement over the base model than the strongest baseline under different selection budgets. With only 10% of the data pool, CAP surpasses or matches full-pool training on math and science. We further show that CAP transfers to multimodal data selection and is robust to response horizon and noise.
Chinese Translation
大型语言模型（LLMs）在预训练过程中有效地吸收了大量知识，但在训练后，模型需要进一步的训练才能可靠地访问这些知识并“了解它们所知道的内容”。我们观察到存储在权重中的知识与通过模型传递的数据流之间存在有趣的排名平衡。在所有模型层中，我们发现隐状态（数据流）呈现U型模式，早期层的压缩显著，而在后期解码阶段则急剧上升。相反，权重排名呈现倒U型模式，早期和后期层的排名非常低，而中间层的排名较高。我们将其解释为内外信息的交互：中间激活不需要携带权重可以随后提供的内容，因此它们主要保留权重无法提供的内容。基于这一观察，我们提出了一种模型感知的数据选择方法，称为CAP（Counterfactual Assimilation Profile），该方法可以通过利用模型生成的响应与参考响应之间在早期和后期层表示中的差异，来判断数据候选是否包含当前模型可访问的信息。在数学、代码和科学领域，CAP在不同选择预算下相较于最强基线提供了35.4%的平均改进。仅使用10%的数据池，CAP在数学和科学领域超越或匹配了全池训练的效果。我们进一步展示了CAP可以迁移到多模态数据选择，并且对响应时限和噪声具有鲁棒性。
cs.AI / 165 / 2609.33012
When Pair Count Is Not the Sample Size: What All-Pairs Agent Comparisons Estimate
当配对数量不是样本大小时：全配对代理比较估计的内容
Abstract
When an agent benchmark compares every pair of leaderboard entries, the number of comparisons can look much larger than the independent evidence behind them: A versus B and A versus C both reuse A. Whether this reuse affects inference depends on what the analysis is meant to describe. If the board and its outcomes are fixed, the all-pairs mean is an exact summary of those entries, and any interval must come from another declared source of randomness. If the entries are instead treated as iid draws from a population of future configurations and the pair rule is regular and nondegenerate, the same mean is an order-two U-statistic whose first-order uncertainty depends on the number of configurations, not the number of pairs. We use near ties as the running example, but the distinction extends to other symmetric pair summaries when their regularity conditions hold. We examine both interpretations using a fixed SWE-bench Verified snapshot and an exact binary model with known truth. On SWE-bench, intervals that accounted for shared configurations were more than twice as wide as a pair-iid reference that treated the pairs as independent. In the exact model, pair-iid coverage fell far below the nominal level when edges shared endpoints but remained near nominal for matched independent edges. Results on two other fixed leaderboards show that exact summaries also depend on which pairs are included and how they are weighted. An all-pairs analysis must therefore state what is fixed, what is sampled, and how it handles shared entries and pair aggregation.
Chinese Translation
当一个代理基准比较排行榜上的每一对条目时，比较的数量可能看起来远大于其背后的独立证据：A 对 B 和 A 对 C 都重用了 A。这种重用是否影响推断取决于分析的目的。如果排行榜及其结果是固定的，则全配对均值是对这些条目的精确总结，任何区间必须来自另一个声明的随机源。如果条目被视为来自未来配置的独立同分布（iid）抽样，并且配对规则是常规且非退化的，则相同的均值是一个二阶 U 统计量，其一阶不确定性依赖于配置的数量，而不是配对的数量。我们以近乎平局的情况作为运行示例，但这一区别扩展到其他对称配对总结，当它们的常规性条件成立时。我们使用固定的 SWE-bench Verified 快照和一个具有已知真值的精确二元模型来检查这两种解释。在 SWE-bench 上，考虑共享配置的区间宽度是独立配对 iid 参考的两倍以上。在精确模型中，当边共享端点时，配对 iid 覆盖远低于名义水平，但对于匹配的独立边则保持在名义水平附近。对另外两个固定排行榜的结果表明，精确总结也依赖于包含哪些配对以及它们的加权方式。因此，全配对分析必须明确说明什么是固定的，什么是抽样的，以及如何处理共享条目和配对聚合。
cs.AI / 166 / 2609.33013
The Epistemics of Agent Memory: Measuring, and Governing, the Consolidation Decision in Long-Horizon LLM Agents
代理记忆的认识论：在长时间视野的LLM代理中测量和管理整合决策
Abstract
Long-horizon LLM agents must convert accumulated experience into durable memory, deciding what to keep, compress, abstract into reusable skills and rules, or forget. We report a four-phase research program on this consolidation problem whose central finding is a shift in what is measured: from how much an agent remembers, to whether its consolidation decisions are any good, to whether those decisions can be trusted. Phase 1 learns episodic boundaries from agent traces by downstream utility; an honest near-miss (oracle correlation 0.691 vs a 0.70 bar) whose lasting output is a three-gate anti-leakage protocol. Phase 2 learns when to promote experience and to which abstraction level under a token budget, achieving a verified +22.7% task-success improvement with 7x compression, but exposing a degenerate-forgetting failure and a distribution-shift failure mode we name lambda-prevalence coupling. Phase 3 introduces ConsolidationBench, an oracle-by-construction benchmark that scores consolidation decisions against a known optimum on three non-circular axes; production retrieval systems retain information yet score zero on cross-level transfer. Phase 4 introduces governed consolidation: the decision wrapped in poison-resistance, reversibility, and auditability guarantees with a quality gate. Governance is statistically distinct from the quality score ($r^2 = 0.43$; partial $r = 0.27$; identical-quality policies differ threefold in governance), so the contribution survives independently of the metric's external validity. On that question we report a resolved negative: after a graded-reuse redesign removed a structural ceiling, a two-benchmark study with 2,532 real answer cells finds the quality score does not predict real transfer accuracy (pooled Spearman $\rho = -0.24$, n = 12, CI spanning zero). An adversarial self-critique pass cleared the final claim set with zero surviving overclaims.
Chinese Translation
长时间视野的LLM代理必须将积累的经验转化为持久的记忆，决定保留、压缩、抽象为可重用的技能和规则，或遗忘什么。我们报告了一个关于这一整合问题的四阶段研究计划，其核心发现是测量内容的转变：从代理记住多少，转向其整合决策的质量，再到这些决策是否值得信赖。第一阶段通过下游效用从代理痕迹中学习情节边界；一个诚实的接近失误（oracle相关性0.691与0.70的标准相比）其持久输出是一个三门反泄漏协议。第二阶段学习在代币预算下何时提升经验以及提升到哪个抽象层级，实现了经过验证的+22.7%的任务成功率提升，压缩比为7倍，但暴露出一种退化遗忘失败和我们称之为lambda-普遍性耦合的分布转移失败模式。第三阶段引入了ConsolidationBench，一个通过构造的基准，评估整合决策相对于已知最优解在三个非循环轴上的得分；生产检索系统保留信息但在跨层级转移上得分为零。第四阶段引入了受管理的整合：决策被包装在抗毒性、可逆性和可审计性保证以及质量门槛中。治理在统计上与质量评分显著不同（$r^2 = 0.43$；部分$r = 0.27$；相同质量政策在治理上差异三倍），因此该贡献独立于指标的外部有效性。在这个问题上，我们报告了一个解决的否定：在经过分级重用重新设计去除了结构性上限后，一项包含2,532个真实答案单元的双基准研究发现，质量评分并不能预测真实转移的准确性（合并的Spearman $
ho = -0.24$, n = 12, CI跨越零）。一次对抗性的自我批评过程清除了最终声明集，零个过度声明存活。
cs.AI / 167 / 2609.33017
Trust and Task Completion in the World of Consumer AI Agents
消费者人工智能代理中的信任与任务完成
Abstract
Action agents do things for people. They send email, spend money, and call businesses while the user is busy with something else, so a mistake can turn into an action before anyone notices. They fail their users in two ways. They break trust when they do something the user never agreed to, or hold back after the user clearly said go. And they fall short on completion when they give up on errands that turn out to be hard. Both depend heavily on the harness around the model, meaning its instructions, tools, context, and guardrails. We built an evaluation that scores trust and completion on the same runs, in a simulated world of businesses with their own websites, inboxes, and phone lines, and of people who write back. A simulated user answers the assistant's questions. Trust means that nothing happens the user did not agree to. No email goes to someone they never approved, no private detail ends up on a group thread, no money is spent past their limit, no stranger's instructions are followed, and nothing is claimed without a source. Every trap has a matched control in which acting is the right call. We use the evaluation to measure Fo, Wajo's personal assistant, against a base model with basic instructions on three foundation models, and against the Fo harness with its guardrails switched off. Fo completes 71% of the errands and keeps the user's trust on 94% of the trap runs. The base models complete 50% to 64% and keep trust on 59% to 75%. On the matched controls, Fo goes ahead slightly less often. OpenClaw, a popular open-source assistant given the same access, completes 42% of the errands it shares with Fo, against 71%, and keeps the user's trust on 74% of the shared trap runs, against 94%. Measuring trust and completion together, on the whole system rather than the model alone, is how we think action agents become safe to hand real work to.
Chinese Translation
行动代理为人们完成各种事务。它们在用户忙于其他事情时发送电子邮件、花费金钱和拨打电话，因此一个错误可能在任何人注意到之前就转变为一个行动。它们在两个方面辜负了用户的期望。当代理执行用户从未同意的操作时，它们破坏了信任；而在用户明确表示要继续时却停滞不前，亦是如此。同时，当它们放弃那些被认为困难的差事时，也未能完成任务。这两者都在很大程度上依赖于模型周围的约束，即其指令、工具、上下文和保护措施。我们构建了一个评估体系，在一个模拟的商业世界中对信任和完成度进行评分，该世界拥有自己的网站、收件箱和电话线路，以及会回复的人。模拟用户回答助手的问题。信任意味着用户没有同意的事情不会发生。没有电子邮件发送给未获批准的人，没有私人细节出现在群组讨论中，没有超出用户限额的金钱支出，没有遵循陌生人的指示，且没有没有来源的声明。每一个陷阱都有一个匹配的控制组，其中采取行动是正确的选择。我们使用该评估来测量Fo（Wajo的个人助手）与基础模型的基本指令在三个基础模型上的表现，以及与关闭保护措施的Fo约束进行比较。Fo完成了71%的差事，并在94%的陷阱运行中保持了用户的信任。基础模型的完成率为50%到64%，信任保持在59%到75%之间。在匹配的控制组中，Fo稍微少一些地继续行动。OpenClaw，一个流行的开源助手，在同样的访问权限下，完成了42%的与Fo共享的差事，而Fo的完成率为71%，并在74%的共享陷阱运行中保持了用户的信任，而Fo则为94%。我们认为，信任和完成度的结合测量，关注整个系统而非单一模型，是行动代理能够安全地承担实际工作的方式。
cs.AI / 168 / 2609.33023
SRE-Marathon: A Continuous, Change-Driven Benchmark for Autonomous Site Reliability Agents
SRE-Marathon：一种基于持续变化的自主站点可靠性代理基准测试
Abstract
Benchmarks for site reliability engineering (SRE) agents are typically episodic: one fault is injected, the agent receives an incident task, and its response is scored. Production operation is not. Incidents surface through noisy alerts, overlap in time, and often originate from code or configuration changes. We present SRE-Marathon, a benchmark for long-horizon, continuous SRE operation. An agent is invoked at a fixed cadence with cumulative alert history and a persistent workspace while operating a live two-zone Kubernetes deployment as a fault orchestrator injects overlapping faults according to a seeded, production-calibrated schedule. Curated code and configuration changes deployed through the same build pipeline available for repair. Each run is recorded into a sealed bundle and scored offline: Marathon-Score credits each injected fault for ordered progress through correlation, localization, and repair, with all metrics computed deterministically from recorded system evidence. Across three applications and about sixty faults per run, the best of 10 methods reaches only 41.3 out of 100. Agents often correlate and localize faults, but almost never complete repairs while the faults remain active.
Chinese Translation
站点可靠性工程（SRE）代理的基准测试通常是情景式的：注入一个故障，代理接收一个事件任务，并对其响应进行评分。然而，生产操作并非如此。事件通过嘈杂的警报浮现，时间上重叠，且通常源于代码或配置的变化。我们提出了SRE-Marathon，这是一个用于长时间跨度、持续SRE操作的基准测试。在固定的节奏下调用代理，利用累积的警报历史和持久的工作空间，同时在一个实时的双区Kubernetes部署中操作，故障编排器根据预设的、经过生产校准的时间表注入重叠的故障。经过精心策划的代码和配置变化通过相同的构建管道进行部署以便于修复。每次运行都被记录到一个密封的包中并离线评分：Marathon-Score为每个注入的故障在关联、定位和修复的有序进展中给予积分，所有指标均根据记录的系统证据确定性计算。在三个应用程序和每次运行约六十个故障的情况下，10种方法中最佳的仅达到41.3分（满分100分）。代理通常能够关联和定位故障，但几乎从未在故障仍然活跃时完成修复。
cs.AI / 169 / 2609.33039
Agent Safety From Within: Detecting Harmful Trajectories from LLM Internal States
内部代理安全：从大型语言模型内部状态检测有害轨迹
Abstract
Language model agents can now perform sophisticated sequences of actions via tools and harnesses, which has increased the scope of the damage they can cause. Guard models, however, are mainly built for content moderation and thus are not well-suited to detecting this agentic risk. To address this, we proceed by first conducting a representational analysis, then use the resulting insights to build a solution. In our analysis, we focus on two types of trajectory-level agentic harms: harmful content, which is expressed directly, and unsafe tool use, which depends on whether an action is consistent with the interaction that produced it. We investigate how open-source guard models represent these two types of harm and find that they are linearly readable inside the model, even though guard models predict no better than chance on pairs that differ only in the called tool's schema. The two harm types also follow nearly orthogonal internal directions, and neither reliably serves as a proxy for the other. These results motivate reading trajectory safety directly from internal states. We introduce TACIT, a readout of a frozen backbone's internal states that decodes no tokens. Trained on six trajectory-safety benchmarks, a linear probe raises mean macro-F1 from 62.3 for the strongest open guard to 80.7, and refined readouts reach 86.2. With each benchmark held out of training entirely, the refined readouts still lead the strongest guard (65.7 vs. 61.1). With the same backbone, training data and test split, the frozen readout is on par with full safety fine-tuning, and it improves the fine-tuned model further when applied on top. The probe trains about one millionth as many parameters as full fine-tuning in about a sixth of the time, and TACIT has the lowest latency of the guards we evaluate.
Chinese Translation
语言模型代理现在可以通过工具和手段执行复杂的动作序列，这增加了它们可能造成的损害范围。然而，现有的防护模型主要用于内容审核，因此不太适合检测这种代理风险。为了解决这个问题，我们首先进行了一项表征分析，然后利用所得的见解构建解决方案。在我们的分析中，我们关注两种类型的轨迹级代理危害：直接表达的有害内容和依赖于某个动作是否与产生该动作的交互一致的安全工具使用。我们研究了开源防护模型如何表征这两种类型的危害，发现它们在模型内部是线性可读的，尽管防护模型在仅在调用工具的模式上存在差异的配对上预测的准确率与随机猜测相当。这两种危害类型也遵循几乎正交的内部方向，且彼此之间并不可靠地作为代理。以上结果促使我们直接从内部状态读取轨迹安全性。我们引入了TACIT，这是一种对冻结主干内部状态的读取，不解码任何标记。在六个轨迹安全基准上训练后，线性探测器将最强开源防护的平均宏F1从62.3提高到80.7，而精细化的读取结果达到了86.2。在每个基准完全不参与训练的情况下，精细化的读取结果仍然领先于最强防护（65.7对61.1）。在相同的主干、训练数据和测试划分下，冻结的读取结果与完整的安全微调相当，并且在其上应用时进一步改善了微调模型。该探测器的训练参数数量约为完整微调的百万分之一，且所需时间约为其六分之一，TACIT在我们评估的防护模型中具有最低的延迟。
cs.AI / 170 / 2609.33052
BudgetVerify: Budget-Tiered Verification for Financial QA
BudgetVerify：用于金融问答的预算分层验证
Abstract
Financial question answering often requires precise numerical extraction, unit handling, and arithmetic over tables and text, but applying expensive verification uniformly wastes test-time compute. We propose BudgetVerify, a budget-tiered generator-verifier framework that routes each generated answer to one of three verification tiers: no verification, lightweight check-and-revise, or higher-cost solve-first-then-compare verification. The router is trained from offline correctness and token-cost outcomes and, at test time, selects a verification tier using information available before verification, including the question, context statistics, the generated answer, and associated generator metadata. The selected tier either returns the generated answer directly or invokes the corresponding verifier. Across six commercial and open-weight base models, BudgetVerify consistently produces more efficient accuracy-cost Pareto frontiers than fixed verification policies by selectively allocating stronger verification only when it is useful. Although absolute performance varies across models, these efficiency gains and the resulting qualitative frontier shape are consistent across generator models.
Chinese Translation
金融问答通常需要精确的数值提取、单位处理以及对表格和文本的算术运算，但统一应用昂贵的验证会浪费测试时的计算资源。我们提出了BudgetVerify，一种预算分层生成-验证框架，将每个生成的答案路由到三个验证层级之一：无验证、轻量级检查与修正，或高成本的先求解后比较验证。路由器基于离线的正确性和标记成本结果进行训练，并在测试时使用在验证前可获得的信息（包括问题、上下文统计、生成的答案和相关的生成器元数据）选择验证层级。所选层级要么直接返回生成的答案，要么调用相应的验证器。在六个商业和开放权重基础模型中，BudgetVerify始终比固定验证策略产生更高效的准确性-成本帕累托前沿，通过选择性地仅在有用时分配更强的验证。尽管绝对性能在不同模型间有所差异，但这些效率提升及其导致的定性前沿形状在生成器模型中是一致的。
cs.AI / 171 / 2609.33055
Large Language Models Substantially Compress Well-Being Inequality but Largely Preserve Its Socioeconomic Structure
大型语言模型显著压缩幸福感不平等，但在很大程度上保留其社会经济结构
Abstract
Research using large language models (LLMs) to generate synthetic populations has repeatedly shown that model outputs compress the diversity of human experience. This has raised doubts about whether LLM-generated data can capture meaningful differences within populations. We show that such compression does not necessarily erase the social structure of human heterogeneity. Using 93,901 respondents from 66 countries and territories in Wave 7 of the World Values Survey, we ask six LLMs to predict respondents' life satisfaction from demographic, socioeconomic, and attitudinal profiles. All six models substantially understate the overall dispersion of life satisfaction. Yet after normalizing for these differences in scale, they largely reproduce the human income gradient in well-being inequality: lower-income groups remain relatively more heterogeneous than higher-income groups. The pattern is robust to country fixed effects, equal-country weighting, WVS survey weights, and observed demographic composition, and it extends directionally to employment, education, and perceived control. Fidelity is weaker for extreme outcomes and country-specific gradients. These results show that the amount of heterogeneity preserved by an LLM and the way that heterogeneity is distributed across social groups are distinct properties. LLM-generated populations can therefore substantially compress human variation while retaining meaningful information about where that variation is concentrated.
Chinese Translation
使用大型语言模型（LLMs）生成合成人口的研究反复表明，模型输出压缩了人类经验的多样性。这引发了对LLM生成的数据是否能够捕捉到群体内有意义差异的怀疑。我们展示了这种压缩并不一定会抹去人类异质性的社会结构。利用来自66个国家和地区的93,901名受访者的第七波世界价值观调查（World Values Survey），我们要求六个LLM根据受访者的人口统计、社会经济和态度特征预测其生活满意度。所有六个模型都显著低估了生活满意度的整体分散性。然而，在对这些规模差异进行归一化后，它们在很大程度上再现了幸福感不平等中的人类收入梯度：低收入群体的异质性相对高于高收入群体。该模式对国家固定效应、等国加权、WVS调查权重和观察到的人口组成具有稳健性，并在方向上扩展到就业、教育和感知控制。对于极端结果和特定国家的梯度，保真度较弱。这些结果表明，LLM所保留的异质性数量以及这种异质性在社会群体中的分布方式是不同的特性。因此，LLM生成的人口可以显著压缩人类变异，同时保留关于这种变异集中位置的有意义信息。
cs.AI / 172 / 2609.33061
LLM sequential decision making under uncertainty in biochemical domains
生物化学领域中不确定性下的LLM序列决策
Abstract
Large language models (LLMs) are increasingly used to drive scientific discovery. Understanding how LLMs make decisions from new data and memory of the literature is vital before trusting them to design experiments under tight experimental budgets. However, their decision strategies are invisible in the current performance scores used to evaluate research agents. Here, we benchmark five frontier LLMs in a Bayesian Optimization setting against published statistical baselines on seven combinatorial datasets spanning protein engineering, reaction optimization, molecular design, peptide self-assembly, and catalysis. Performance is paired with direct measurements of model beliefs and actions, enabling highly resolved behavior analysis. A prompt ablation that progressively strips context separates memorization from chemical reasoning and from bare categorical optimization. Prior chemical knowledge helps in expectation, but with high variance and occasionally even harms performance. No configuration tested decisively beats a mean statistical baseline across domains. Belief-movement and Martingale diagnostics, corrected here for a measurement-noise bias that mislabels rational agents as irrational, show that models overreact to incoming data rather than entrenching on their priors in the contexts studied here. Interestingly, while LLM actions are exploitative, models sincerely intend to explore and consistently act on that intent. This failure is a competence gap arising from context-stickiness. Removing in-context history restores exploration, indicating that priors and data must be decoupled to achieve effective LLM-driven discovery.
Chinese Translation
大型语言模型（LLMs）越来越多地被用于推动科学发现。在信任它们在紧张的实验预算下设计实验之前，理解LLMs如何从新数据和文献记忆中做出决策至关重要。然而，它们的决策策略在当前用于评估研究代理的性能评分中是不可见的。在这里，我们在贝叶斯优化设置中对五个前沿LLMs进行了基准测试，并与七个组合数据集上的已发布统计基线进行了比较，这些数据集涵盖了蛋白质工程、反应优化、分子设计、肽自组装和催化等领域。性能与模型信念和行动的直接测量相结合，使得行为分析高度细致。一个逐步剥离上下文的提示消融实验将记忆与化学推理和纯粹的分类优化分开。先前的化学知识在期望中有所帮助，但具有高方差，有时甚至会损害性能。没有任何测试配置在各个领域中显著优于平均统计基线。信念移动和马尔可夫诊断在这里经过修正，以消除将理性代理错误标记为非理性的测量噪声偏差，显示模型对输入数据的反应过度，而不是在这里研究的上下文中固守其先验。有趣的是，尽管LLM的行动是利用性的，但模型真诚地意图进行探索并始终基于这种意图采取行动。这一失败是由于上下文粘性导致的能力差距。去除上下文历史恢复了探索，表明必须将先验和数据解耦，以实现有效的LLM驱动发现。
cs.AI / 173 / 2609.33075
QureRadEmbed: Structuring Radiological Similarity through Attribute and Reasoning Supervision
QureRadEmbed：通过属性和推理监督构建放射学相似性
Abstract
Radiological similarity depends on disease relationships and on fine details such as laterality, lobe, severity, size, and certainty. Broad biomedical similarity can overlook these qualifiers, particularly when several attributes vary together. We introduce QureRadEmbed, a 4B radiology-aware encoder trained with two complementary signals: RadSim supplies deterministic, attribute-decomposed ranking targets, while RadThought aligns reports with hierarchical evidence and reasoning descriptions. A three-stage curriculum combines these signals with report triplets, finding perturbations, and single- and cross-attribute contrasts. The final model achieves 0.996 mean ordering accuracy across ten controlled synthetic attributes and raises Spearman correlation with the designed joint-attribute targets from 0.501 to 0.976. On external findings-to-impression retrieval, Recall@1 reaches 10.5% on Open-I, 12.4% on testing XR, and 42.9% on testing CT, compared with 6.6%, 5.7%, and 31.4% for its backbone. Frozen embeddings support finding extraction with only 100 labeled testing-XR reports (macro-F1 0.481 versus 0.412 for the backbone). Whole-report comparison costs 8.8 seconds per 1,000 pairs in our benchmark, versus 2,755.1 seconds for the generative evaluator GREEN. Sentence-level comparison improves sensitivity to local discrepancies, although generative evaluation remains stronger on several expert-rated and subtle-error tasks. The results support reusable radiology-aware representations for search, structured report indexing, and efficient report comparison.
Chinese Translation
放射学相似性依赖于疾病关系以及诸如侧别、叶、严重程度、大小和确定性等细节。广泛的生物医学相似性可能会忽视这些限定因素，特别是在多个属性共同变化时。我们提出了QureRadEmbed，一个经过训练的4B放射学感知编码器，利用两种互补信号：RadSim提供确定性的、属性分解的排名目标，而RadThought则将报告与层次证据和推理描述对齐。一个三阶段的课程结合了这些信号与报告三元组、扰动发现以及单一和交叉属性对比。最终模型在十个受控合成属性上实现了0.996的平均排序准确率，并将与设计的联合属性目标的Spearman相关性从0.501提高到0.976。在外部发现到印象的检索中，Open-I上的Recall@1达到了10.5%，测试XR为12.4%，测试CT为42.9%，而其基础模型分别为6.6%、5.7%和31.4%。冻结的嵌入支持仅使用100个标记的测试XR报告进行发现提取（宏F1为0.481，相较于基础模型的0.412）。在我们的基准测试中，整体报告比较每1000对耗时8.8秒，而生成评估器GREEN则为2755.1秒。句子级比较提高了对局部差异的敏感性，尽管在多个专家评定和细微错误任务中，生成评估仍然表现更强。结果支持可重用的放射学感知表示，用于搜索、结构化报告索引和高效报告比较。
cs.AI / 174 / 2609.33079
Structure-Mapping-Guided Self-Explanation for Learning Mathematical Procedures
基于结构映射的自我解释指导数学程序学习
Abstract
Worked examples are a powerful form of instruction, but learners must infer how the demonstrated steps were produced. A naive simulation of this self-explanation process can generate thousands of numerical explanations that reproduce one observed change without capturing its underlying procedure. We propose structure-mapping-guided self-explanation as a computational account of the cognitive biases that reduce search effort and make this inference tractable. The model represents mathematical expressions as typed relational structures and uses structure mapping to identify corresponding source and target regions. For each changed target value, the corresponding source region serves as an anchor: it guides abductive search toward structurally relevant values and operations before broader alternatives, yielding ordered, executable candidate procedures with inspectable source evidence. Across 70 mathematical transformations containing 120 changed numeric components, the model recovered every intended procedure. It returned the intended procedure before any other computation producing the same target value in 104 subproblems (86.7%), compared with a median of 32 (26.7%) across 100 unguided runs that tested candidate calculations in random order. Our proposed model also tested 93.8% fewer combinations of values and operations than unguided search before reaching the intended procedures. These results provide an efficient, interpretable account of how relational structure can guide procedural learning and a testable hypothesis about human self-explanation from worked examples.
Chinese Translation
示例学习是一种强有力的教学形式，但学习者必须推断出所示步骤是如何产生的。对这一自我解释过程的简单模拟可能会生成数千个数值解释，这些解释重现了一个观察到的变化，却未能捕捉其背后的程序。我们提出了基于结构映射的自我解释，作为一种计算模型，解释了减少搜索努力的认知偏差，使得这一推断变得可行。该模型将数学表达式表示为类型化的关系结构，并利用结构映射识别对应的源区域和目标区域。对于每个变化的目标值，相应的源区域作为锚点：它引导诱导搜索朝向结构相关的值和操作，而不是更广泛的替代方案，从而产生有序的、可执行的候选程序，并附有可检查的源证据。在包含120个变化数值组件的70个数学变换中，该模型恢复了每一个预期的程序。它在104个子问题中返回了预期的程序，所需时间早于任何其他产生相同目标值的计算（86.7%），而在100次无指导运行中，随机顺序测试候选计算的中位数为32（26.7%）。我们提出的模型在达到预期程序之前，还测试了比无指导搜索少93.8%的值和操作组合。这些结果提供了一个高效、可解释的框架，说明了关系结构如何指导程序学习，并提出了一个关于人类从示例学习中自我解释的可测试假设。
cs.AI / 175 / 2609.33085
The Model Knows Another Way: Strategy Switching for Effective RLVR Exploration
模型知道另一种方法：有效的RLVR探索策略切换
Abstract
Reinforcement learning with verifiable rewards (RLVR) is often limited by insufficient exploration: difficult problems can yield uniformly incorrect rollout groups and therefore little learning signal. We show that such failures need not reflect missing capability. Instead, finite sampling often concentrates on a problem-specific dominant reasoning strategy while leaving alternative strategies already supported by the model unexplored. Moreover, the accessibility of these strategies evolves during RL: some are internalized into autonomous behavior, while others become difficult to elicit before being absorbed. Motivated by these observations, we introduce Problem--Strategy Rollout Allocation (PSRA), which treats unguided and strategy-conditioned prompts as competing exploration arms and uses Bayesian sequential allocation to direct a fixed rollout budget toward arms most likely to yield informative, non-saturated groups. A preservation objective keeps useful strategy-conditioned routes accessible while successful guided behaviors are transferred to the unguided policy. Across Qwen2.5 models from 1.5B to 7B and two RL training corpora, PSRA consistently improves reasoning performance, reduces dead saturation, strengthens out-of-distribution transfer, and maintains larger gains under increased inference budgets.
Chinese Translation
带有可验证奖励的强化学习（RLVR）常常受到探索不足的限制：困难问题可能导致均匀错误的回滚组，从而产生很少的学习信号。我们表明，这种失败并不一定反映出能力的缺失。相反，有限的采样通常集中于特定问题的主导推理策略，而忽略了模型已经支持的替代策略。此外，这些策略的可获取性在RL过程中不断演变：一些策略被内化为自主行为，而另一些则在被吸收之前变得难以引导。基于这些观察，我们引入了问题-策略回滚分配（Problem--Strategy Rollout Allocation, PSRA），该方法将无指导和策略条件的提示视为竞争的探索臂，并使用贝叶斯序列分配将固定的回滚预算引导至最有可能产生信息丰富、非饱和组的臂。一个保留目标保持有用的策略条件路径可访问，同时成功的引导行为被转移到无指导策略。在1.5B到7B的Qwen2.5模型和两个RL训练语料库中，PSRA始终提高推理性能，减少死饱和，增强分布外迁移，并在增加推理预算的情况下保持更大的增益。
cs.AI / 176 / 2609.33115
Modular Discovery of General Game-Playing Algorithms with Large Language Models
利用大型语言模型进行通用游戏玩法算法的模块化发现
Abstract
General Game Playing across arbitrary games from rules alone remains challenging due to differing algorithmic requirements across game classes and strict decision-time constraints. Rather than hand-designing search heuristics for specific domains, can we leverage Large Language Models (LLMs) to discover general game-playing algorithms? Because language models can propose and refactor structured code, they provide an expressive proposal engine for exploring the space of algorithmic designs. We introduce a multi-agent LLM meta-learning system to co-evolve game-agnostic procedural search mechanisms in C++ alongside domain heuristics synthesized directly from game rules. Controlling the compute budget, we benchmark the discovered mechanisms across more than 400 diverse environments, including OpenSpiel training and held-out games, procedural simulation engines, and games with deep neural policy-value representations trained via PPO. Evaluated via AlphaRank stationary distributions and Soft Condorcet Optimization (SCO) against 15 established MCTS baselines, the discovered search mechanisms consistently achieve top-tier ratings and pairwise ballot majorities over most baselines across independent evolutionary runs, generalizing to unseen human-designed and procedurally synthesized games and remaining competitive with baselines on frozen neural network representations.
Chinese Translation
仅凭规则进行的通用游戏玩法在不同游戏类别之间的算法需求差异和严格的决策时间限制下仍然具有挑战性。我们能否利用大型语言模型（LLMs）来发现通用游戏玩法算法，而不是为特定领域手动设计搜索启发式？由于语言模型能够提出和重构结构化代码，它们为探索算法设计空间提供了一个富有表现力的提案引擎。我们引入了一种多智能体LLM元学习系统，以共同进化与游戏无关的C++程序搜索机制，以及直接从游戏规则合成的领域启发式。在控制计算预算的情况下，我们在超过400个多样化环境中对发现的机制进行了基准测试，包括OpenSpiel训练和保留游戏、程序化模拟引擎，以及通过PPO训练的深度神经网络策略-价值表示。通过AlphaRank平稳分布和针对15个已建立的MCTS基线的软康多塞优化（SCO）进行评估，发现的搜索机制在独立的进化运行中始终实现了顶级评分和大多数基线的成对投票多数，能够推广到未见的人类设计和程序合成的游戏，并在冻结的神经网络表示上与基线保持竞争力。
cs.AI / 177 / 2609.33123
Compositional Safety Failures in Harness Evolution: Identification and Runtime Monitoring
马具演化中的组合安全失败：识别与运行时监控
Abstract
Self-evolving agent harnesses continually update persistent components such as memory, prompts, skills, and tools. We call this process harness evolution. However, such evolution could introduce unexpected safety risks. Existing work studies harness misevolution and validates candidate harnesses or attributed individual component updates, leaving safety analysis of cross-component update interactions largely unexamined. To address this gap, we study compositional safety failures in harness evolution, where interactions among individually safe and utility-preserving component updates can produce undesirable or unsafe agent behavior, revealing a safety risk intrinsic to harness evolution. Across three safety-related benchmarks, we identify 43 pairwise and 18 irreducible 3-way compositional safety failures. Conventional solution incurs combinatorial complexity in validating cross-component interactions, leaving the safety checking impractical as the harness evolves. To solve this, we introduced a typed hypergraph that represents component states as nodes and safety-relevant higher-order interactions as hyperedges. When the harness changes, the hypergraph updates only the interaction neighborhood of the changed states rather than reconstructing the global composition space. Building on that, we develop a hypergraph-guided runtime monitoring mechanism. Experiments show that our method effectively mitigates compositional safety risks while preserving task utility and reducing interaction-checking costs, and further reveal an empirical safety-utility-cost trade-off across different safety mechanisms.
Chinese Translation
自我演化的代理马具不断更新持久组件，如内存、提示、技能和工具。我们称这一过程为马具演化。然而，这种演化可能会引入意想不到的安全风险。现有研究主要关注马具的错误演化，并验证候选马具或单个组件更新的属性，导致跨组件更新交互的安全分析在很大程度上未得到检视。为了解决这一空白，我们研究了马具演化中的组合安全失败，其中个别安全且保持效用的组件更新之间的交互可能产生不良或不安全的代理行为，揭示了马具演化内在的安全风险。在三个与安全相关的基准测试中，我们识别出43对和18个不可约的三元组合安全失败。传统解决方案在验证跨组件交互时面临组合复杂性，使得随着马具演化，安全检查变得不切实际。为了解决这个问题，我们引入了一种类型化超图，将组件状态表示为节点，将与安全相关的高阶交互表示为超边。当马具发生变化时，超图仅更新变化状态的交互邻域，而不是重建全局组合空间。在此基础上，我们开发了一种超图引导的运行时监控机制。实验表明，我们的方法有效减轻了组合安全风险，同时保持任务效用并降低交互检查成本，并进一步揭示了不同安全机制之间的经验安全-效用-成本权衡。
cs.AI / 178 / 2609.33134
Ceiling of a Task: When Can a Transformer Succeed Without Its Chain of Thought?
任务的上限：变换器何时可以在没有思维链的情况下成功？
Abstract
Reasoning models generate long chains of thought before they answer, yet it is debated whether the content of these chains does real computational work or is largely decorative. We study this question by viewing a transformer as a shallow circuit. One forward pass through a fixed number of layers has constant depth, so any procedure that runs the model a constant number of times is a shallow circuit. We call the best accuracy that a shallow circuit can reach on a task the ceiling of the task, and a task is serial if its ceiling lies below one. We prove three results on serial tasks that hold for every transformer, no matter how it was trained. Necessity: replacing the chain by anything that does not depend on its content, such as filler tokens or a restatement of the question, drives the accuracy down to the ceiling, and on a maximally serial task down to chance. Depth: no shallow computation can write the chain of a model whose accuracy exceeds the ceiling, not even approximately. Locality: the answer is one shallow pass away from the finished chain, so all of the serial reasoning happens in the chain. On word problems of finite groups, whose ceilings are known, small transformers trained from scratch, with or without reinforcement learning, attain the predicted numbers: chain-trained models solve every input length and fall to chance when the chain is erased, chainless models collapse to the ceiling as the input length grows, and open-weight reasoning models given the same problem in words return to the baseline without their chain. On MATH-500 and AIME, erasing the chain costs open reasoning models 0.52 to 0.82 accuracy, a sentence shuffle is harmless, and a token shuffle is as harmful as erasing; the same holds for checkpoints trained by GRPO with a correct or a random reward. The ceiling of a task therefore answers when a transformer can succeed without its chain of thought.
Chinese Translation
推理模型在回答之前会生成长长的思维链，但关于这些链的内容是否真正进行计算工作或仅仅是装饰性内容仍存在争议。我们通过将变换器视为浅层电路来研究这个问题。通过固定层数的一次前向传播具有恒定的深度，因此任何以恒定次数运行模型的过程都是一个浅层电路。我们将浅层电路在某一任务上能够达到的最佳准确率称为该任务的上限，如果一个任务的上限低于1，则称其为串行任务。我们证明了三个适用于所有变换器的串行任务的结果，无论其训练方式如何。必要性：用任何不依赖于其内容的东西替代思维链，例如填充标记或问题的重述，会使准确率降至上限，而在一个最大串行任务中降至随机水平。深度：没有任何浅层计算可以写出一个准确率超过上限的模型的思维链，甚至近似也不行。局部性：答案距离完成的思维链只有一次浅层传播的距离，因此所有的串行推理都发生在思维链中。在已知上限的有限群的文字问题上，从头训练的小型变换器，无论是否使用强化学习，都达到了预测的结果：经过链训练的模型解决了每个输入长度，当思维链被删除时准确率降至随机水平，而没有思维链的模型在输入长度增加时崩溃至上限，给定相同问题的开放权重推理模型在没有思维链的情况下返回到基线。在 MATH-500 和 AIME 上，删除思维链使开放推理模型的准确率损失为0.52到0.82，句子打乱无害，而标记打乱的危害与删除思维链相当；对于通过 GRPO 训练的检查点，无论是正确奖励还是随机奖励，情况也是如此。因此，任务的上限回答了变换器何时可以在没有思维链的情况下成功。
cs.AI / 179 / 2609.33141
On Device Agentic Operation Caches -- Classifier-Centric NL-to-Action Generation
设备端自主操作缓存——以分类器为中心的自然语言到动作生成
Abstract
Agentic AI is increasingly being embedded in software applications to provide natural language interfaces to features and functionality. In most cases these agents are powered by enterprise (100+ billion parameter) or frontier class large language models that require substantial computational resources run and depend on cloud hosted inference to handle the task of transforming natural language inputs into actionable software operations. This reliance on cloud-hosted inference introduces substantial network latency on top of LLM inference times, creates data privacy concerns, and, given the costs of running these models, can rapidly escalate expenses associated with supporting agentic features. This paper introduces a novel means of converting the NL-to-Action problem from a generative one into a classification-centric formulation via on-device operation caches. These caches allow an agentic system to handle frequently occurring classes of actions completely on-device -- reducing latency, enhancing privacy, and lowering operational costs. We show that for a classic NL-to-Formula task, generating Excel Formula in response to user requests, this approach reduces total inference cost by 56% when compared to cloud-only model-routing based inference and, on cache hits, reduces the latency to response latency by 5x.
Chinese Translation
自主人工智能（Agentic AI）正越来越多地嵌入软件应用中，以提供自然语言接口来访问功能和特性。在大多数情况下，这些代理是由企业级（超过1000亿参数）或前沿级的大型语言模型驱动的，这些模型需要大量的计算资源来运行，并依赖于云托管推理来处理将自然语言输入转换为可操作软件操作的任务。这种对云托管推理的依赖在大型语言模型推理时间的基础上引入了显著的网络延迟，产生了数据隐私问题，并且考虑到运行这些模型的成本，可能迅速增加与支持自主特性相关的费用。本文提出了一种新颖的方法，通过设备端操作缓存将自然语言到动作（NL-to-Action）问题从生成性问题转变为以分类为中心的形式。这些缓存允许自主系统在设备端完全处理频繁发生的动作类别——从而降低延迟、增强隐私并降低运营成本。我们展示了在经典的自然语言到公式（NL-to-Formula）任务中，针对用户请求生成Excel公式时，与仅依赖云模型路由的推理相比，该方法在总推理成本上降低了56%，并且在缓存命中时，将响应延迟降低了5倍。
cs.AI / 180 / 2609.33146
LiteEvo: Automated, Cost-Efficient Harness Evolution for Generalization to Unseen Tasks
LiteEvo：用于未见任务泛化的自动化、成本高效的工具演化
Abstract
An LLM agent is defined by two things: the weights inside its model and the harness of components assembled around it. Harnesses are still handcrafted, and HarnessX, which evolves them automatically, starts each benchmark from a handcrafted harness, reports gains on the tasks it evolved on, and budgets 100 to 175 million meta-agent tokens per benchmark. We propose LiteEvo, a lightweight harness-evolution algorithm whose tool-free meta-agents mine agent trajectories for reusable components, curate them into a versioned library, and compose each round's harness from it, starting every benchmark from the same neutral harness and never naming the benchmark. Evolving on the graded tasks of five agentic benchmarks with a frozen Qwen3.5-9B, LiteEvo lifts pass@2 by 10.5 to 67.7pp and reaches comparable or higher pass@2 than a reproduction of HarnessX (71.0 against 67.3 on average) at 13.0 lower mean API cost. Harnesses evolved on train tasks keep their gains on unseen test tasks of four benchmarks, and LiteEvo also lifts Claude Code with Sonnet 4.6 by 1.2 to 71.4pp.
Chinese Translation
一个大型语言模型（LLM）代理由两个部分定义：其模型内部的权重和围绕它组装的组件工具。工具仍然是手工制作的，而自动演化工具的 HarnessX 从手工制作的工具开始每个基准测试，报告其演化任务的收益，并为每个基准测试预算 1 亿到 1.75 亿个元代理令牌。我们提出了 LiteEvo，一种轻量级的工具演化算法，其无工具的元代理挖掘代理轨迹以获取可重用组件，将其整理成一个版本库，并从中组合每轮的工具，所有基准测试均从相同的中立工具开始，并且从不命名基准测试。在冻结的 Qwen3.5-9B 上对五个代理基准的分级任务进行演化，LiteEvo 将 pass@2 提升了 10.5 到 67.7 个百分点，并且在平均 API 成本降低 13.0 的情况下，达到与 HarnessX 的重现相当或更高的 pass@2（71.0 对 67.3）。在训练任务上演化的工具在四个基准的未见测试任务上保持其收益，LiteEvo 还将 Claude Code 与 Sonnet 4.6 的表现提升了 1.2 到 71.4 个百分点。
cs.AI / 181 / 2609.33149
Not Too Hard, Not Too Easy: Learning from Intermediate States for LLM Structured Reasoning
既不太难，也不太简单：从中间状态学习以进行大语言模型结构化推理
Abstract
A common principle of effective learning is to practice material that is neither already mastered nor too difficult to permit progress. We ask how to apply this principle to structured reasoning tasks such as Sudoku and maze solving. In these tasks, a model can repeatedly revise an incomplete or incorrect candidate solution until it satisfies the problem's constraints. The intermediate candidate solutions along this trajectory provide natural training examples: some are already solved, some cannot yet be repaired by the model, and others lie at its current frontier of achievable progress. We therefore investigate whether pretrained language models can learn to revise such states and whether training on states at this frontier improves reasoning more broadly. To achieve this, we couple a pretrained language-model backbone with a recurrent updater that repeatedly revises an explicit solution state, using the same parameters at every update step. We further introduce Frontier-Oriented Curation Using Self-trajectories (FOCUS), which selects training states from trajectories generated by the current model. FOCUS measures how much the model improves each state within a fixed number of recurrent updates and prioritizes states from which it can make substantial progress. With Qwen3-1.7B, FOCUS achieves 64.4% exact solve accuracy on Sudoku-Extreme and 91.1% on Maze-Hard, with similar gains observed across five Qwen and Llama backbones spanning 1.7B to 8B parameters. We further observe zero-shot transfer in the adapted LLM to mathematical reasoning and code execution, even when the recurrent updater is disabled and no downstream fine-tuning is performed.
Chinese Translation
有效学习的一个普遍原则是练习那些既不是已经掌握的材料，也不是过于困难以至于无法进步的材料。我们探讨如何将这一原则应用于结构化推理任务，例如数独和迷宫求解。在这些任务中，模型可以反复修正一个不完整或错误的候选解，直到它满足问题的约束条件。沿着这一轨迹的中间候选解提供了自然的训练示例：有些已经解决，有些模型尚无法修复，还有一些处于模型当前可实现进展的前沿。因此，我们研究了预训练语言模型是否能够学习修正这些状态，以及在这一前沿状态上训练是否能更广泛地改善推理。为此，我们将一个预训练的语言模型骨干与一个递归更新器结合起来，该更新器反复修正一个明确的解状态，在每次更新步骤中使用相同的参数。我们进一步引入了基于自我轨迹的前沿导向选择（Frontier-Oriented Curation Using Self-trajectories, FOCUS），该方法从当前模型生成的轨迹中选择训练状态。FOCUS衡量模型在固定数量的递归更新中对每个状态的改善程度，并优先选择能够显著进步的状态。在Qwen3-1.7B上，FOCUS在数独极限（Sudoku-Extreme）上的精确解答准确率达到64.4%，在困难迷宫（Maze-Hard）上的准确率为91.1%，在五个Qwen和Llama骨干（参数范围从1.7B到8B）上观察到了类似的提升。我们还观察到在适应后的大语言模型中实现了零样本迁移到数学推理和代码执行，即使在递归更新器被禁用且未进行下游微调的情况下。
cs.AI / 182 / 2609.33181
SeOPD: Self-Evolving LLMs via Online Policy Distillation from Self-Generated Chain-of-Thought
SeOPD：通过自生成思维链的在线策略蒸馏实现自我演化的语言模型
Abstract
Recent advances in online policy self-distillation (OPSD) have demonstrated that large language models (LLMs) can improve their capabilities by leveraging external privileged information (PI), such as manual annotations or feedback from external environments. However, obtaining accurate annotations and constructing sophisticated environments often require substantial human effort and computation, limiting the scalability of OPSD. While a few recent studies have explored self-improvement without external PI, the resulting gains remain limited. In this work, we explore whether LLMs can achieve comparable self-improvement without external PI. Our key observation is that a single LLM can support multiple reasoning modes, such as deep-thinking and non-thinking modes, with deep thinking generating additional information during reasoning. Based on this observation, we propose Self-Evolving Online Policy Distillation (SeOPD), which enables LLMs to distill and internalize information generated by their own chain of thought (CoT). Specifically, it (1) generates CoT with the deep-thinking mode, (2) produces responses with the non-thinking mode, and (3) uses the generated CoT as PI to provide token-level supervision for the non-thinking response, allowing new information inferred during reasoning to guide the non-thinking mode and be internalized into the shared model parameters, thereby improving both non-thinking and deep-thinking capabilities. Extensive experiments across LLMs and tasks demonstrate the effectiveness of SeOPD.
Chinese Translation
近期在线策略自蒸馏（OPSD）的进展表明，大型语言模型（LLMs）可以通过利用外部特权信息（PI），如人工注释或来自外部环境的反馈，来提升其能力。然而，获取准确的注释和构建复杂的环境通常需要大量的人力和计算资源，限制了OPSD的可扩展性。尽管一些近期研究探索了在没有外部PI的情况下进行自我改进，但所获得的提升仍然有限。在本研究中，我们探讨了LLMs是否能够在没有外部PI的情况下实现可比的自我改进。我们的关键观察是，单个LLM可以支持多种推理模式，例如深度思考模式和非思考模式，其中深度思考在推理过程中生成额外的信息。基于这一观察，我们提出了自我演化在线策略蒸馏（SeOPD），使LLMs能够蒸馏并内化其自身思维链（CoT）生成的信息。具体而言，它（1）在深度思考模式下生成CoT，（2）在非思考模式下生成响应，以及（3）使用生成的CoT作为PI，为非思考响应提供逐词级别的监督，从而使推理过程中推断的新信息能够指导非思考模式，并被内化到共享模型参数中，从而提升非思考和深度思考的能力。针对LLMs和任务的广泛实验表明，SeOPD的有效性。
cs.AI / 183 / 2609.33182
Unlocking Latent Personalization in LLMs
解锁大型语言模型中的潜在个性化
Abstract
Large language models (LLMs) are increasingly expected to adapt to individual users, yet effective personalization remains challenging when only limited user-specific samples are available. In this work, we take an alternative perspective: pretrained LLMs may already possess latent capacity for personalization, and a few user samples may therefore suffice to guide the model toward user-aligned behavior with minimal user-specific adaptation. From this perspective, we propose LatentPersonal, a framework that formulates personalization as navigation in a shared latent adaptation space. LatentPersonal infers a compact latent representation from a few user samples to guide user-specific model adaptation, regularized with a variational information bottleneck to encourage compact preference representations. We instantiate LatentPersonal with LoRA, leveraging its low-rank parameterization as a natural low-dimensional adaptation space for personalization. By simply inserting a user-specific guidance vector between the shared low-rank factors, the model can navigate toward personalized adaptations through lightweight inference of this compact representation, without updating the shared LoRA parameters. Experiments across multiple personalization datasets demonstrate that LatentPersonal substantially reduces user-specific adaptation overhead while achieving effective personalization from only a few user-specific interactions, with particularly strong performance in the one-shot regime.
Chinese Translation
大型语言模型（LLMs）越来越被期望能够适应个体用户，但在仅有有限用户特定样本的情况下，有效的个性化仍然具有挑战性。在本研究中，我们采用了另一种视角：预训练的LLMs可能已经具备潜在的个性化能力，因此少量用户样本可能足以引导模型朝向用户对齐的行为，且只需最小的用户特定适应。从这个角度出发，我们提出了LatentPersonal，一个将个性化视为在共享潜在适应空间中导航的框架。LatentPersonal从少量用户样本中推断出紧凑的潜在表示，以指导用户特定的模型适应，并通过变分信息瓶颈进行正则化，以鼓励紧凑的偏好表示。我们使用LoRA实例化LatentPersonal，利用其低秩参数化作为个性化的自然低维适应空间。通过简单地在共享低秩因子之间插入用户特定的引导向量，模型可以通过对这一紧凑表示的轻量推理，朝向个性化适应进行导航，而无需更新共享的LoRA参数。跨多个个性化数据集的实验表明，LatentPersonal显著减少了用户特定适应的开销，同时仅通过少量用户特定交互实现了有效的个性化，尤其在单次学习（one-shot）场景中表现出色。
cs.AI / 184 / 2609.33196
Are Benchmarks Reliable? Toward Structural Diagnosis via Sample-Level Capability Boundaries
基准测试可靠性如何？通过样本级能力边界进行结构诊断
Abstract
Evaluating large language models (LLMs) relies heavily on benchmark scores, yet aggregate metrics can obscure whether benchmark samples reliably support model comparison. We introduce \textbf{BSDProbe}, a sample-level framework for \emph{benchmark structural diagnosis} that estimates capability boundaries from repeated-response trajectories along ordered model axes. BSDProbe summarizes samples by boundary position, boundary width, boundary-signal validity, and order consistency, then aggregates them into benchmark-level structural profiles. Experiments on six benchmarks show that benchmark reliability is axis-conditioned and heterogeneous: GSM8K and MATH exhibit the most stable measurement structures, MMLU and TriviaQA are relatively stable but heterogeneous, while GPQA and PopQA show stronger axis-conditioned risks. These profiles remain consistent across Qwen3, Qwen2.5, and cross-model axes. BSDProbe further selects compact high-value subsets whose model discriminability reaches up to $8.58\times$ that of the full benchmark. These results suggest that reliable benchmark use requires examining sample-level capability boundaries beyond leaderboard scores.
Chinese Translation
评估大型语言模型（LLMs）在很大程度上依赖于基准分数，然而汇总指标可能掩盖基准样本是否可靠地支持模型比较。我们提出了 extbf{BSDProbe}，一个用于 extit{基准结构诊断}的样本级框架，该框架通过沿有序模型轴的重复响应轨迹估计能力边界。BSDProbe通过边界位置、边界宽度、边界信号有效性和顺序一致性来总结样本，然后将其聚合成基准级结构轮廓。对六个基准的实验表明，基准的可靠性是轴条件化和异质的：GSM8K和MATH表现出最稳定的测量结构，MMLU和TriviaQA相对稳定但异质，而GPQA和PopQA则显示出更强的轴条件风险。这些轮廓在Qwen3、Qwen2.5和跨模型轴之间保持一致。BSDProbe进一步选择紧凑的高价值子集，其模型可区分性达到完整基准的$8.58 imes$。这些结果表明，可靠的基准使用需要超越排行榜分数，检查样本级能力边界。
cs.AI / 185 / 2609.33208
WorldAgent: Verification-Guided Agentic Physical World Construction
WorldAgent：基于验证引导的主动物理世界构建
Abstract
Constructing complex physical worlds from language requires coordinating extensive 3D environments, detailed structures and objects at different spatial scales, and interacting physical processes under both stated goals and implicit physical constraints. We present WorldAgent, an agentic framework for verification-guided physical world construction from a single natural-language prompt, without iterative user debugging. A world construction layer expands the prompt into a structured world specification and uses physical knowledge to build scenes and run numerical simulations. After every step, a verification layer inspects scene geometry and simulation states alongside rendered views. Failed checks guide automatic revisions to the specification and re-execution of the affected steps. Accepted worlds pass the required checks and remain editable for further inspection and resimulation. We introduce AgenticSimBench, on which WorldAgent achieves the best scores among the evaluated agent-based methods on five of seven metrics. In a 26-participant user study, it receives the highest mean ratings across all four criteria.
Chinese Translation
从语言构建复杂的物理世界需要协调广泛的三维环境、不同空间尺度下的详细结构和物体，以及在既定目标和隐含物理约束下的相互作用的物理过程。我们提出了WorldAgent，这是一个基于验证引导的主动物理世界构建框架，能够从单一自然语言提示出发，无需迭代用户调试。一个世界构建层将提示扩展为结构化的世界规范，并利用物理知识构建场景并运行数值模拟。在每一步之后，验证层检查场景几何和模拟状态以及渲染视图。失败的检查将指导对规范的自动修订和受影响步骤的重新执行。通过的世界将通过所需的检查，并保持可编辑以便进一步检查和重新模拟。我们引入了AgenticSimBench，在该基准上，WorldAgent在七个指标中的五个上达到了评估的基于代理的方法中的最佳分数。在一项包含26名参与者的用户研究中，它在所有四个标准中获得了最高的平均评分。
cs.AI / 186 / 2609.33243
CodeSkill: Latent Skill Abstraction for Long-Horizon Code Agents
CodeSkill：面向长时间跨度代码代理的潜在技能抽象
Abstract
Code agents require long-horizon decision-making over complex interaction trajectories. However, existing reinforcement learning (RL) approaches typically optimize behavior at the token level, creating a mismatch between low-level generation and high-level behavioral reasoning. This limitation leads to inefficient exploration and weak credit assignment under sparse rewards. Moreover, while large-scale agent trajectories often contain recurring multi-step behavioral patterns, their noisy token-level representations hinder effective experience reuse. To address these challenges, we propose CodeSkill, a framework that adapts hierarchical latent skill modeling to the code agent domain. CodeSkill first leverages a teacher model to distill both successful and failed trajectories into multi-level textual abstractions. It then integrates temporal variational inference with reinforcement learning to map these discrete semantics into continuous latent variables, while an adaptive boundary mechanism dynamically gates skill transitions based on execution feedback. The learned skills are injected into a frozen LLM policy as latent semantic prefixes, enabling optimization in a compact semantic space rather than over raw token sequences. By shifting RL from token-level exploration to experience-level reasoning, CodeSkill improves optimization efficiency and long-horizon behavioral coherence. Extensive experiments demonstrate that CodeSkill achieves highly competitive performance against strong open-weight baselines across diverse general and industrial coding benchmarks. Furthermore, the learned skills exhibit strong transferability and robust cross-domain generalization, highlighting the effectiveness of explicit behavioral abstraction for scalable agentic code generation.
Chinese Translation
代码代理需要在复杂的交互轨迹上进行长时间跨度的决策。然而，现有的强化学习（RL）方法通常在令牌级别优化行为，这导致低级生成与高级行为推理之间的不匹配。这一限制导致在稀疏奖励下的低效探索和弱信用分配。此外，尽管大规模代理轨迹通常包含重复的多步骤行为模式，但其嘈杂的令牌级表示妨碍了有效的经验重用。为了解决这些挑战，我们提出了CodeSkill，一个将层次潜在技能建模应用于代码代理领域的框架。CodeSkill首先利用教师模型将成功和失败的轨迹提炼为多层次的文本抽象。然后，它将时间变分推断与强化学习相结合，将这些离散语义映射到连续潜在变量，同时自适应边界机制根据执行反馈动态控制技能转换。学习到的技能作为潜在语义前缀注入到冻结的LLM策略中，使得优化在紧凑的语义空间中进行，而不是在原始令牌序列上。通过将RL从令牌级探索转向经验级推理，CodeSkill提高了优化效率和长时间跨度的行为一致性。大量实验表明，CodeSkill在多种通用和工业编码基准上相较于强大的开放权重基线实现了高度竞争的性能。此外，学习到的技能表现出强大的可转移性和稳健的跨领域泛化，突显了显式行为抽象在可扩展代理代码生成中的有效性。
cs.AI / 187 / 2609.33244
ActiveMem: Dynamic Latent Memory Trees for Long-Horizon Agents
ActiveMem：用于长时间跨度智能体的动态潜在记忆树
Abstract
Large Language Model (LLM) agents increasingly rely on external memory to support long-horizon reasoning and decision making. Existing memory systems typically retrieve historical trajectories or summaries as independent context fragments, overlooking the procedural dependencies underlying multi-step execution. As memory scales, such flat retrieval introduces context fragmentation and cross-task interference, leading to structurally inconsistent reasoning trajectories. We propose ActiveMem, a hierarchical memory framework that recursively organizes agent experiences into dependency-aware latent execution trees. ActiveMem abstracts trajectories into reusable subtask nodes while explicitly preserving execution transitions, enabling coherent reasoning-path retrieval conditioned on the current execution state. To support continual adaptation, ActiveMem further learns dynamic memory expansion, retrieval, and pruning policies through reinforcement learning. Experiments across various agent benchmarks demonstrate that ActiveMem consistently improves task completion, reasoning stability, and memory efficiency over existing memory-based agents. Moreover, ActiveMem enables compact open-weight models to achieve competitive performance with substantially larger proprietary systems.
Chinese Translation
大型语言模型（LLM）智能体越来越依赖外部记忆来支持长时间跨度的推理和决策。现有的记忆系统通常将历史轨迹或摘要作为独立的上下文片段进行检索，忽视了多步骤执行背后的过程依赖性。随着记忆规模的扩大，这种扁平检索引入了上下文碎片化和跨任务干扰，导致结构上不一致的推理轨迹。我们提出了ActiveMem，一个层次化的记忆框架，它递归地将智能体经验组织成依赖感知的潜在执行树。ActiveMem将轨迹抽象为可重用的子任务节点，同时明确保留执行过渡，从而使得在当前执行状态下能够进行连贯的推理路径检索。为了支持持续适应，ActiveMem进一步通过强化学习学习动态记忆扩展、检索和修剪策略。在各种智能体基准测试中的实验表明，ActiveMem在任务完成、推理稳定性和记忆效率方面始终优于现有的基于记忆的智能体。此外，ActiveMem使得紧凑的开放权重模型能够在性能上与大规模专有系统竞争。
cs.AI / 188 / 2609.33260
CORTEX: A Verified Experience Layer for Generalist Agents
CORTEX：一种经过验证的通用智能体体验层
Abstract
An agent can solve a task today and face the same task under new facts, tools, or governing knowledge tomorrow. Most agent systems can retrieve relevant text or recall prior conversations, but they lack a principled way to decide when a previous solution is still valid, when it must be adapted, and when it should be discarded. We introduce CORTEX (Contextual Orchestration and Reuse of Task EXperience), a general AI systems framework that connects specialized agents through an external layer of verified experience. Each episode records its task conditions, source and tool state, decisive predicates, proof trace, verifier, and outcome. A meta-controller chooses exact replay, checked adaptation, fresh synthesis, or escalation. Accepted episodes can become task patterns and procedural strategies through a challenge-driven development loop. This gives the system an implicit competence layer that can grow without changing model weights. We formalize system contracts for exact replay and source-version separation, and derive when reuse saves computation. A controlled two-domain implementation tests the exact-replay core on 1,000 synthetic cases. Complete-family holdouts test procedural transfer on 1,000 new-family cases across eight clinical and policy splits, with complete fresh-evidence grounding and perfect invariance to irrelevant-field and insertion-order perturbations. The transfer trace exposes the work required for verified strategy execution. These results establish an initial path toward general intelligence through reusable procedures, typed experience, and developmental transfer.
Chinese Translation
智能体今天可以解决一个任务，而明天在新的事实、工具或治理知识下面对同样的任务。大多数智能体系统能够检索相关文本或回忆先前的对话，但它们缺乏一种原则性的方法来决定何时先前的解决方案仍然有效，何时必须进行调整，以及何时应当被舍弃。我们介绍了CORTEX（Contextual Orchestration and Reuse of Task EXperience），这是一个将专门智能体通过经过验证的体验外部层连接起来的通用人工智能系统框架。每个事件记录其任务条件、来源和工具状态、决定性谓词、证明轨迹、验证者和结果。元控制器选择精确重放、检查适应、新鲜合成或升级。被接受的事件可以通过挑战驱动的开发循环转变为任务模式和程序策略。这为系统提供了一个隐式的能力层，可以在不改变模型权重的情况下增长。我们为精确重放和源版本分离形式化了系统合同，并推导出何时重用可以节省计算。一个受控的双领域实现测试了在1,000个合成案例上的精确重放核心。完整家族的保留测试了在八个临床和政策分割中对1,000个新家族案例的程序转移，具有完整的新证据基础和对无关领域及插入顺序扰动的完美不变性。转移轨迹揭示了验证策略执行所需的工作。这些结果为通过可重用程序、类型化经验和发展转移朝向通用智能奠定了初步路径。
cs.AI / 189 / 2609.33268
LSTMem: Hierarchical Long Short-Term Online Memory for Large Language Models
LSTMem：用于大型语言模型的分层长短期在线记忆
Abstract
Large language models increasingly serve as long-horizon assistants and agents, where they must both accumulate information across interactions and make the relevant parts available when later requests depend on them. Existing compact online memories typically use a single persistent state both to accumulate history and to serve readout, so what the memory stores cannot be controlled separately from what it exposes to the current computation. We propose LSTMem, an LSTM-inspired online memory that instead equips each layer of a frozen LLM with two matrix-valued states: a cell state that accumulates history and a hidden state whose readouts correct the backbone's attention. Input and forget gates control what the cell stores, while an output gate separately controls what the cell exposes through the hidden state. LSTMem further connects memory across depth through forward hidden-state propagation and block-end feedback, and uses higher-layer reconstruction gradients to refine lower-layer cell states before rebuilding hidden states from shallow to deep layers. Across memory benchmarks on Qwen3-4B-Instruct, LSTMem consistently improves MemoryAgentBench, LoCoMo, and HotpotQA over the plain backbone. Comparisons further show that the LSTM-based memory formulation outperforms an associative-memory counterpart, while removing cross-layer hidden-memory propagation degrades performance. These results demonstrate the benefits of separating memory accumulation from memory expression and organizing memory hierarchically across model depth. The code is available at https://github.com/Longchentong/LSTMem.
Chinese Translation
大型语言模型越来越多地作为长时间范围的助手和代理，它们必须在交互中积累信息，并在后续请求依赖于这些信息时使相关部分可用。现有的紧凑型在线记忆通常使用单一的持久状态来同时积累历史和提供读取，因此记忆存储的内容无法与当前计算所暴露的内容分开控制。我们提出了LSTMem，一种受LSTM启发的在线记忆，它为每个冻结的LLM层配备了两个矩阵值状态：一个用于积累历史的单元状态和一个其读取输出修正主干注意力的隐藏状态。输入门和遗忘门控制单元存储的内容，而输出门则单独控制单元通过隐藏状态所暴露的内容。LSTMem进一步通过前向隐藏状态传播和块末反馈在深度上连接记忆，并利用高层重构梯度在从浅层到深层重建隐藏状态之前细化低层单元状态。在Qwen3-4B-Instruct的记忆基准测试中，LSTMem在MemoryAgentBench、LoCoMo和HotpotQA上始终优于普通主干。比较进一步表明，基于LSTM的记忆公式优于关联记忆对应物，而去除跨层隐藏记忆传播则会降低性能。这些结果展示了将记忆积累与记忆表达分开以及在模型深度上以分层方式组织记忆的好处。代码可在https://github.com/Longchentong/LSTMem获取。
cs.AI / 190 / 2609.33270
Structured Sparse Memory for Recurrent Reasoning
用于递归推理的结构化稀疏记忆
Abstract
Recurrent models trained from scratch have recently become competitive on ARC-style reasoning tasks, but the usual framing around small recurrent backbones overlooks two important parts of the system: task-conditioned memory and synthetic augmentation data. We study this regime through CHARM, a compact hybrid ARC model that combines recurrent reasoning with structured task memory, synthetic data, and inference-time aggregation. In existing approaches, task-conditioned memory supplies a large hidden source of capacity, reaching more than 30x the size of the recurrent backbone. We introduce a compositional sparse embedding (CoSE) for task conditioning that reduces learned task-memory parameters by over 90% while improving pass@2 in controlled ARC ablations. For the recurrent backbone, recurrent depth helps only when balanced with learning horizon. Combining these ingredients, our system reaches 84% pass@2 on ARC-AGI-1 and 46.7% pass@2 on ARC-AGI-2 public evaluation. The benefits of structured memory also generalize to unseen puzzles and other domains. Our code, dataset, and model checkpoints are available at https://github.com/water-vapor/charm.
Chinese Translation
从零开始训练的递归模型最近在ARC风格的推理任务中变得具有竞争力，但通常围绕小型递归骨干的框架忽视了系统的两个重要部分：任务条件记忆和合成增强数据。我们通过CHARM研究这一机制，CHARM是一个紧凑的混合ARC模型，结合了递归推理、结构化任务记忆、合成数据和推理时聚合。在现有方法中，任务条件记忆提供了一个巨大的隐藏容量源，超过递归骨干的30倍。我们引入了一种组合稀疏嵌入（CoSE）用于任务条件化，减少了学习的任务记忆参数超过90%，同时在控制的ARC消融实验中提高了pass@2的表现。对于递归骨干，递归深度只有在与学习视野平衡时才有帮助。结合这些要素，我们的系统在ARC-AGI-1上达到了84%的pass@2，在ARC-AGI-2公共评估中达到了46.7%的pass@2。结构化记忆的好处也能推广到未见过的难题和其他领域。我们的代码、数据集和模型检查点可在https://github.com/water-vapor/charm获取。
cs.AI / 191 / 2609.33271
Next Thoughts Are Distributions: Generative Autoregressive Reasoning in the Latent Space
下一个思维是分布：潜在空间中的生成自回归推理
Abstract
Reasoning problems often admit multiple valid ways to proceed. Continuous reasoning promises to move computation beyond language tokens into a more compact latent space, but representing several plausible ways to think next remains difficult. We introduce Autoregressive Thought Flow (ATF), which models the next continuous thought as a multimodal distribution. A causal autoregressive model performs the reasoning computation, while a lightweight diffusion head generates a plausible next thought from the resulting condition. The sampled thought is fed back into the model, allowing continuous reasoning to unfold for a variable number of steps while preserving the pretrained backbone. Across mathematical reasoning tasks, ATF improves accuracy with compact latent traces and benefits from reinforcement learning and additional test-time thinking. Multi-sample evaluation shows broader solution coverage, indicating that its multimodal predictions capture useful diversity among reasoning paths. Our results suggest that continuous reasoning is more effective when multiple possible next thoughts remain available rather than being collapsed into a single prediction.
Chinese Translation
推理问题通常允许多种有效的解决方式。连续推理有望将计算超越语言符号，进入一个更紧凑的潜在空间，但表示几种合理的下一个思维方式仍然困难。我们提出了自回归思维流（Autoregressive Thought Flow, ATF），该模型将下一个连续思维建模为多模态分布。一个因果自回归模型执行推理计算，而一个轻量级扩散头则从生成的条件中生成一个合理的下一个思维。采样的思维反馈到模型中，使得连续推理可以在可变的步骤数中展开，同时保留预训练的主干网络。在数学推理任务中，ATF通过紧凑的潜在轨迹提高了准确性，并受益于强化学习和额外的测试时思考。多样本评估显示更广泛的解决方案覆盖，表明其多模态预测捕捉了推理路径之间有用的多样性。我们的结果表明，当多个可能的下一个思维仍然可用时，连续推理更为有效，而不是被压缩为单一预测。
cs.AI / 192 / 2609.33276
ChronoFlow: Hierarchical Flow Matching for Irregular Time Series Generation
ChronoFlow：不规则时间序列生成的分层流匹配
Abstract
Recent advances in generative modeling have substantially improved time series generation, yet most existing methods either assume a regular temporal grid or focus on feature dynamics under a given sampling structure. This makes them illsuited for generating irregular time series in their native form, where a model must capture not only feature values, but also how many observations occur, when they occur, and which features are observed together. To address this heterogeneous generation problem, we propose ChronoFlow, a unified hierarchical flow matching framework organized by statistical granularity. Following a coarse-to-fine hierarchy, ChronoFlow first generates observation counts and feature-wise frequencies, then jointly generates observation times and feature co-observation patterns, and finally generates values conditioned on the realized pattern. This turns a complex joint generation problem into structurally aligned subproblems while preserving their dependencies. To evaluate complete irregular time series generation, we introduce complementary metrics spanning sample realism, sampling structure, value fidelity, and temporal and cross-feature dependencies, and validate them through controlled corruptions. Across five benchmarks, ChronoFlow achieves strong improvements in generation fidelity over existing baselines, while factorization studies support the proposed hierarchy. Our code is available at https://anonymous.4open.science/r/ChronoFlow.
Chinese Translation
近年来生成建模的进展显著改善了时间序列生成，然而大多数现有方法要么假设一个规则的时间网格，要么专注于在给定采样结构下的特征动态。这使得它们不适合以原生形式生成不规则时间序列，其中模型不仅必须捕捉特征值，还必须捕捉观察次数、观察时间以及哪些特征是一起被观察的。为了解决这一异质生成问题，我们提出了ChronoFlow，一个统一的分层流匹配框架，按统计粒度组织。ChronoFlow遵循粗到细的层次结构，首先生成观察计数和特征频率，然后共同生成观察时间和特征共观察模式，最后在实现的模式基础上生成值。这将复杂的联合生成问题转化为结构上对齐的子问题，同时保留它们之间的依赖关系。为了评估完整的不规则时间序列生成，我们引入了涵盖样本真实性、采样结构、值保真度以及时间和跨特征依赖性的互补指标，并通过控制性破坏进行验证。在五个基准测试中，ChronoFlow在生成保真度上显著超越现有基线，同时因子化研究支持所提出的层次结构。我们的代码可在 https://anonymous.4open.science/r/ChronoFlow 获取。
cs.AI / 193 / 2609.33282
Multi-Dimensional Comparative Scale Construction for Efficient Personalized Subjective Judgment in High-Traffic Applications
高流量应用中高效个性化主观判断的多维比较量表构建
Abstract
Subjective judgments are central to many high-traffic applications, but subjective intensity is difficult to quantify and perceptions vary substantially across individuals. To address these challenges, we propose a pairwise comparative framework for multi-dimensional scale construction. By comparing case-person pairs along case and profile dimensions, the framework constructs relative scales that capture both fine-grained intensity and individual variation. To support practical high-traffic deployment, we optimize both offline scale construction and online inference. For scale construction, we combine sparse Elo comparisons with multi-judge voting, cutting the comparison cost from $O(N^2)$ to $O(NK)$ for $N$ objects and a budget of $K$ opponents per object, while limiting reliance on any single judge. For inference, we propose SubJudge, a System One model for personalized scoring with Batchwise Preference Optimization (BPO). Using Bradley-Terry comparisons, BPO trains the model to learn relative orderings, and SubJudge reads a continuous score from digit-token probabilities at the first response position, requiring only one forward pass per criterion and reducing the inference complexity to $O(1)$. Experiments on PluriHarms and iNews show that our 9B models match or surpass the evaluated frontier LLMs on multiple metrics. On the H100 GPU, SubJudge achieves an approximately $1.29\times$ to $261\times$ speedup in mean inference latency over Qwen3.5-9B with different thinking budgets. The code is available at https://github.com/Longchentong/SubJudge.
Chinese Translation
主观判断在许多高流量应用中至关重要，但主观强度难以量化，且个体间的感知差异显著。为了解决这些挑战，我们提出了一种用于多维量表构建的成对比较框架。通过在案例和特征维度上比较案例-个体对，该框架构建了捕捉细粒度强度和个体差异的相对量表。为了支持实际的高流量部署，我们优化了离线量表构建和在线推断。在量表构建方面，我们结合了稀疏的Elo比较和多评审投票，将比较成本从$O(N^2)$降低到$O(NK)$，其中$N$为对象数量，$K$为每个对象的对手预算，同时限制对任何单一评审的依赖。在推断方面，我们提出了SubJudge，一种用于个性化评分的系统一模型，采用批次偏好优化（Batchwise Preference Optimization, BPO）。通过Bradley-Terry比较，BPO训练模型学习相对排序，SubJudge从第一个响应位置的数字-标记概率中读取连续评分，每个标准仅需一次前向传递，将推断复杂度降低到$O(1)$。在PluriHarms和iNews上的实验表明，我们的9B模型在多个指标上与评估的前沿大型语言模型（LLMs）相匹配或超越。在H100 GPU上，SubJudge在不同思维预算下实现了相对于Qwen3.5-9B的平均推断延迟约$1.29 imes$到$261 imes$的加速。代码可在https://github.com/Longchentong/SubJudge获取。
cs.AI / 194 / 2609.33284
RINI: Seeing the Prior Is Not Enough
RINI：仅仅看到前期研究是不够的
Abstract
A research proposal can describe an established mechanism correctly while claiming to introduce it. We study whether providing the earlier paper corrects such contribution claims. Three controlled experiments compare proposals generated with a contribution-bearing prior and a same-topic control. Providing the prior yields no clear aggregate reduction in unsupported novelty. Human analysis of 175 interpretable exposed proposals finds that 137 recognize the prior's relevance, but 61 correctly attribute the established contribution. Of 71 proposed remaining distinctions, 37 are covered by the same prior. We introduce Research Idea Novelty Inspection (RINI), which audits contribution claims against evidence, checks the remaining distinction, and applies local revisions. Five human annotators evaluate 1,080 original-revision pairs across three methods. On the same 240 originals judged to require correction, successful repair is 11.7% for Self-Revision, 39.1% for Retrieve-and-Revise, and 72.2% for RINI, with research tasks weighted equally. The improvement over same-evidence direct revision is 33.0 percentage points. The revised proposals retain their research questions and technical methods. These results motivate explicit contribution attribution when using literature to generate and revise research proposals.
Chinese Translation
研究提案可以正确描述一个已建立的机制，同时声称引入该机制。我们研究提供早期论文是否能纠正这种贡献声明。通过三项对照实验比较了使用带有贡献的前期研究和同主题对照生成的提案。提供前期研究并未显著减少不支持的新颖性。对175个可解释的暴露提案进行的人类分析发现，137个提案认识到前期研究的相关性，但只有61个正确归属了已建立的贡献。在71个提出的剩余区别中，有37个被同一前期研究所涵盖。我们引入了研究创意新颖性检查（Research Idea Novelty Inspection，RINI），该方法对贡献声明进行证据审核，检查剩余区别，并应用局部修订。五位人类注释者评估了三种方法下的1,080对原始-修订提案。在同样的240个被判断需要修正的原始提案中，自我修订的成功修复率为11.7%，检索并修订的成功修复率为39.1%，而RINI的成功修复率为72.2%，各研究任务权重相同。与同证据的直接修订相比，改进幅度为33.0个百分点。修订后的提案保留了其研究问题和技术方法。这些结果促使在使用文献生成和修订研究提案时明确归属贡献。
cs.AI / 195 / 2609.33287
Feedback Makes Perfect: A Closed-Loop Framework for NL-to-STL Translation
反馈使完美：一种用于自然语言到时序逻辑翻译的闭环框架
Abstract
Signal Temporal Logic (STL) enables rigorous verification and control of cyber-physical systems, but writing correct specifications requires expertise that most requirement holders lack. Large language models can translate natural-language (NL) requirements into STL, yet stronger translators alone approach an accuracy ceiling. We argue that this ceiling stems from how the task is posed: one-shot, open-loop translation is somewhat ill-defined. Natural language is ambiguous, and, more fundamentally, what a person writes may not always be what they intend, so the target specification is not fully contained in the input text. We therefore reformulate NL-to-STL translation as a closed-loop feedback process. Each generated formula is translated back into natural language for the user to check, and natural-language corrections drive revision until the user accepts the specification. Users never read or write formal syntax. This framework rests on an asymmetry familiar from feedback control theory. The forward path, from ambiguous language to formal logic, is hard and error-prone. The feedback path, from structured STL back to language, can be made highly precise, and a precise feedback path lets an imprecise forward path achieve precise closed-loop behavior. Experiments on 500 expert-authored requirements and seven LLMs support this view. Back-translated explanations agree with expert judgments in 99.5\% of cases. Closed-loop refinement raises strong models from about 89\% open-loop accuracy to 98.0--99.2\%, and yields gains of over 30 percentage points for weaker models (e.g., 17.6\%$\rightarrow$48.0\%). Ablations show these gains come from the semantic content of the feedback rather than from repeated attempts. An expert audit and a 280-session user study further confirm the reliability of the loop. We also identify a capability threshold above which feedback no longer helps.
Chinese Translation
信号时序逻辑（STL）能够对网络物理系统进行严格的验证和控制，但撰写正确的规范需要大多数需求持有者所缺乏的专业知识。大型语言模型可以将自然语言（NL）需求翻译为STL，但单靠更强的翻译器也只能接近一个准确性上限。我们认为这个上限源于任务的设定：一次性、开放式的翻译在某种程度上是不明确的。自然语言是模糊的，更根本的是，一个人所写的内容可能并不总是他们的真实意图，因此目标规范并不完全包含在输入文本中。因此，我们将自然语言到STL的翻译重新表述为一个闭环反馈过程。每个生成的公式都会被翻译回自然语言供用户检查，自然语言的修正推动修订，直到用户接受该规范。用户从不阅读或编写正式语法。该框架基于反馈控制理论中熟悉的不对称性。从模糊语言到形式逻辑的前向路径是困难且容易出错的。而从结构化的STL回到语言的反馈路径可以非常精确，精确的反馈路径使得不精确的前向路径能够实现精确的闭环行为。对500个专家撰写的需求和七个大型语言模型的实验支持了这一观点。反向翻译的解释在99.5%的案例中与专家判断一致。闭环精炼将强模型的开放式准确率从约89%提升至98.0%至99.2%，并为较弱模型带来了超过30个百分点的提升（例如，从17.6%提升至48.0%）。消融实验表明，这些提升来自于反馈的语义内容，而非重复尝试。专家审核和280次用户研究进一步确认了该循环的可靠性。我们还确定了一个能力阈值，超过该阈值后反馈不再有帮助。
cs.AI / 196 / 2609.33289
Learning to Sell: Reinforcement Learning for Strategic Large Language Model Agents in Multi-Product Markets
学习销售：多产品市场中战略性大型语言模型代理的强化学习
Abstract
Autonomous large language model (LLM) agents operating in multi-product markets must make sequential decisions under information asymmetry and resource constraints. We develop a machine learning approach for training such agents to act effectively as sellers in a multi-item bargaining environment, where a seller concurrently negotiates a catalog of substitutable assets across a pool of independent buyers. Buyers hold private, heterogeneous valuations across products, and each can purchase at most one item. Facing limits on total communication turns, the seller must dynamically match buyers with the most profitable products considering their private valuations, while strategically allocating its limited interaction budget toward combinations of greater potential value. We formalize this problem as a Partially Observable Markov Decision Process using a structured, four-part message protocol that maps natural language into a parsable and regulated decision space. Using this formalization, we design a post-training method using Reinforcement Learning from Verifiable Rewards (RLVR). To evaluate this framework, we construct a multidimensional metric suite that quantifies constraint adherence, seller surplus extraction, and allocation quality. Our trained seller agent learns to match limited inventory to buyers more effectively, matching or outperforming trillion-parameter frontier models in both seller surplus extraction and buyer-product allocation quality. Finally, these learned strategies generalize robustly to unseen market structures, correlated valuation distributions, and price ranges not encountered during training.
Chinese Translation
在多产品市场中，自治的大型语言模型（LLM）代理必须在信息不对称和资源限制下做出连续决策。我们开发了一种机器学习方法，用于训练这些代理在多项物品的讨价还价环境中有效地作为卖方行动，其中卖方同时在一组独立买家中就可替代资产目录进行谈判。买家对产品持有私有的、异质的估值，每个买家最多只能购买一件商品。在总沟通轮次有限的情况下，卖方必须动态地将买家与最有利可图的产品匹配，同时考虑他们的私有估值，并战略性地分配其有限的互动预算，以实现更高潜在价值的组合。我们将此问题形式化为部分可观测的马尔可夫决策过程，使用一种结构化的四部分消息协议，将自然语言映射到可解析和规范的决策空间。基于这一形式化，我们设计了一种使用可验证奖励的强化学习（RLVR）的后训练方法。为了评估这一框架，我们构建了一套多维度指标，量化约束遵循、卖方盈余提取和分配质量。我们训练的卖方代理学习更有效地将有限库存匹配给买家，在卖方盈余提取和买家-产品分配质量方面与万亿参数的前沿模型相匹配或超越。最后，这些学习到的策略在未见过的市场结构、相关估值分布和训练过程中未遇到的价格范围上具有良好的泛化能力。
cs.AI / 197 / 2609.33295
TraceDance: An Automated System for Building Agent Behavior Benchmarks from Real-World Agent Deployment Traces
TraceDance：一种从真实世界代理部署轨迹构建代理行为基准的自动化系统
Abstract
An agent can complete a task while exhibiting undesirable behavior during execution. Developers need tests for the specific behaviors encountered in deployment, beyond fixed benchmark suites. We present TraceDance, an agent system that constructs targeted benchmarks from deployment traces for user-specified undesirable behaviors. For efficient construction, Anchor-and-Confirm combines programmable retrieval with candidate-level confirmation by a Flash large language model (LLM), while the Anchor Synthesis Loop generates and revises specifications for custom behaviors. The benchmarks use decision-point continuation to evaluate an LLM's next turn at a recorded decision point with a behavior-specific rubric, without a reference answer or environment replay. Experiments in coding and general tool use draw on 252,557 sessions and produce 107 benchmarks with 4,125 instances, fulfilling 95.3% of build-target requests. Both human annotators confirm the requested behavior in 84% of sampled instances, and the automated grader's agreement with human pass/fail judgments is comparable to that between the annotators. Nine frontier LLMs achieve a mean pass rate of only 26.7%, showing that they still struggle to respond appropriately at the evaluated decision points. Analysis across behavior-specific benchmarks further reveals weaknesses in how current LLMs behave as agents. By turning deployment problems into targeted benchmarks, TraceDance could serve as a key component of the recursive self-improvement (RSI) loop.
Chinese Translation
代理在执行任务时可能会表现出不良行为。开发者需要针对部署中遇到的特定行为进行测试，而不仅仅依赖固定的基准套件。我们提出了TraceDance，这是一种从部署轨迹中构建针对用户指定的不良行为的目标基准的代理系统。为了高效构建，Anchor-and-Confirm结合了可编程检索与由Flash大型语言模型（LLM）进行的候选级确认，而Anchor Synthesis Loop则生成并修订自定义行为的规范。这些基准使用决策点延续来评估LLM在记录的决策点的下一步表现，采用特定行为的评分标准，而不需要参考答案或环境重放。在编码和通用工具使用方面的实验基于252,557个会话，生成了107个基准，包含4,125个实例，满足了95.3%的构建目标请求。人类标注者在84%的抽样实例中确认了请求的行为，而自动评分者与人类通过/未通过判断的一致性与标注者之间相当。九个前沿LLM的平均通过率仅为26.7%，显示它们在评估的决策点上仍然难以做出适当的响应。对行为特定基准的分析进一步揭示了当前LLM作为代理时的弱点。通过将部署问题转化为目标基准，TraceDance可以作为递归自我改进（RSI）循环的关键组成部分。
cs.AI / 198 / 2609.33297
The Error You See Is Not the Error You Made: Progression-aware Reasoning Origin for Reasoning Error Localization
你看到的错误并不是你犯的错误：基于进程的推理错误定位
Abstract
Verifying multi-step LLM reasoning requires more than determining whether a trace is correct: a useful verifier should identify where the reasoning first goes wrong. However, existing holistic methods provide little positional evidence, while forward sequential verification often treats the first rejected step as the error source. Under error propagation, this assumption can fail, since an earlier mistake may remain locally plausible and become observable only through its downstream consequences. We therefore rethink reasoning verification as a progression-aware error-source localization problem: rather than asking only where a reasoning trace first appears inconsistent, we ask which earlier step best explains how that inconsistency emerges along the trajectory. Based on this view, we propose Progression-aware Reasoning Origin (PRO), a training-free framework for first-error localization. PRO jointly models incoming support from the preceding context and outgoing compatibility with subsequent reasoning, selectively refines regions where these signals disagree, and finally performs detector-conditioned source attribution with intervention-based evidence to distinguish the true error origin from its propagated manifestations. We further formalize the gap between forward rejection and structural exposure, showing why incoming-side evidence alone is insufficient for reliable localization under error propagation. Experiments across open-form, medical, and structured reasoning tasks demonstrate consistent improvements over strong verification baselines, supporting progression-aware source attribution as a more faithful formulation of reasoning verification.
Chinese Translation
验证多步骤大型语言模型（LLM）推理不仅需要判断一个推理过程是否正确：一个有效的验证器应该识别推理首次出错的地方。然而，现有的整体方法提供的位置信息有限，而前向顺序验证通常将第一个被拒绝的步骤视为错误源。在错误传播的情况下，这一假设可能会失败，因为早期的错误可能在局部上仍然是合理的，且只有通过其下游后果才能被观察到。因此，我们重新思考推理验证，将其视为一个基于进程的错误源定位问题：我们不仅要询问推理过程首次出现不一致的地方，还要询问哪个早期步骤最好地解释了沿着轨迹出现的不一致。基于这一观点，我们提出了基于进程的推理源（Progression-aware Reasoning Origin, PRO），这是一个无训练的框架用于首次错误定位。PRO联合建模来自前文上下文的支持信号和与后续推理的一致性，选择性地细化这些信号不一致的区域，最后通过基于干预的证据进行检测器条件的源归因，以区分真实的错误源与其传播表现。我们进一步形式化了前向拒绝与结构暴露之间的差距，展示了为何仅依赖于来自输入侧的证据不足以在错误传播下实现可靠的定位。在开放式、医学和结构化推理任务上的实验表明，相较于强基线验证方法，基于进程的源归因在性能上有一致的提升，支持其作为推理验证的更真实的表述。
cs.AI / 199 / 2609.33319
PhysAlign: A Benchmark for Evidence-Grounded Role Alignment in Multimodal Physics Reasoning
PhysAlign：多模态物理推理中基于证据的角色对齐基准
Abstract
A key challenge in physics diagram understanding is correctly associating visual information with the physical entities, relations, and conditions it describes. Even when a value, symbol, or other local element is accurately recognized, assigning it to the wrong entity or scope can distort the underlying physical premise and lead to incorrect reasoning. To systematically study this challenge, we introduce \textbf{PhysAlign}, a benchmark designed to assess whether multimodal models correctly associate information recognized from physics diagrams with its intended physical role. By disentangling visual recognition from physical-role assignment through localized probes and controlled variants, PhysAlign isolates correspondence errors from recognition failures. It contains 3,341 human-validated probes spanning 986 physics problems, enabling systematic evaluation of visual recognition and physical-role correspondence at scale. We further introduce five complementary evaluation metrics, including CAcc, GAcc, and JAcc, which provide a comprehensive assessment of models' ability to recognize diagram content, establish correct physical correspondences, and solve the underlying physics problem. Across our evaluated multimodal models, PhysAlign reveals a consistent gap between local visual recognition and physical-role grounding. Even when the queried content is correctly recognized, the conditional correspondence error rate remains 13.8\% for GPT-6-Astra and rises to about 50.6\% for InternVL3.5-8B. These findings indicate that strong perception alone does not ensure reliable physical interpretation, exposing a distinct grounding bottleneck that is largely hidden by answer-level accuracy and highlighting the need for future models to better align recognized visual evidence with its physical meaning.
Chinese Translation
物理图示理解中的一个关键挑战是正确地将视觉信息与其所描述的物理实体、关系和条件进行关联。即使一个值、符号或其他局部元素被准确识别，将其分配给错误的实体或范围也可能扭曲基础物理前提，导致错误推理。为了系统性地研究这一挑战，我们引入了 extbf{PhysAlign}，这是一个旨在评估多模态模型是否正确将从物理图示中识别的信息与其预期物理角色关联的基准。通过局部探测和受控变体将视觉识别与物理角色分配解耦，PhysAlign将对应错误与识别失败分离。它包含3,341个经过人工验证的探测，涵盖986个物理问题，使得在大规模上系统评估视觉识别和物理角色对应成为可能。我们进一步引入了五个互补的评估指标，包括CAcc、GAcc和JAcc，这些指标提供了对模型识别图示内容、建立正确物理对应关系以及解决基础物理问题能力的全面评估。在我们评估的多模态模型中，PhysAlign揭示了局部视觉识别与物理角色基础之间的一致差距。即使查询的内容被正确识别，条件对应错误率在GPT-6-Astra中仍为13.8\%，而在InternVL3.5-8B中上升至约50.6\%。这些发现表明，单靠强大的感知能力并不能确保可靠的物理解释，暴露出一个在答案级准确性下大多被隐藏的独特基础瓶颈，并强调未来模型需要更好地将识别的视觉证据与其物理意义对齐的必要性。
cs.AI / 200 / 2609.33323
Agentic Multi-Turn Reasoning: A Fairness Approach
代理多轮推理：一种公平性方法
Abstract
Recent advances in Large Language Models (LLMs) have enabled agentic systems capable of solving complex tasks through multi-turn planning, tool use, verification, and memory updates. However, learning agentic systems remains difficult due to two fundamental challenges, i.e., (1) long-horizon credit assignment, where supervision is available only at the final outcome, and (2) imbalanced data distributions, where dominant data patterns bias optimization and weaken adaptation to rare but informative reasoning behaviors. In this paper, we propose Fair Multi-Level Preference Optimization (Fair-MPO or $\Phi$-MPO), a new preference optimization framework for agentic learning. We first show that Multi-Level Preference Optimization provides a principled and more computationally efficient framework for long-horizon reasoning. Then, we introduce a Fair Multi-Level Objective that addresses imbalance in agentic learning. We provide a comprehensive theoretical analysis demonstrating that our approach addresses both long-horizon reasoning and data imbalance. Our experiments on agentic reasoning benchmarks demonstrate that our approach achieves State-of-the-Art (SOTA) performance.
Chinese Translation
最近在大型语言模型（LLMs）方面的进展使得代理系统能够通过多轮规划、工具使用、验证和记忆更新来解决复杂任务。然而，学习代理系统仍然面临两个基本挑战，即（1）长期信用分配，其中监督仅在最终结果处可用，以及（2）数据分布不平衡，主导数据模式偏向优化并削弱对稀有但信息丰富的推理行为的适应能力。在本文中，我们提出了公平多层偏好优化（Fair Multi-Level Preference Optimization，Fair-MPO或$ ext{Φ}$-MPO），这是一个新的代理学习偏好优化框架。我们首先展示了多层偏好优化为长期推理提供了一个原则性且计算上更高效的框架。然后，我们引入了一个公平多层目标，解决代理学习中的不平衡问题。我们提供了全面的理论分析，证明我们的方法同时解决了长期推理和数据不平衡问题。我们在代理推理基准上的实验表明，我们的方法达到了最先进的（SOTA）性能。
cs.AI / 201 / 2609.33326
ANTMAN: Adaptive Need Tracking for Multi-Agent Navigation in Large Information Spaces
ANTMAN：大信息空间中多智能体导航的自适应需求跟踪
Abstract
Information-seeking agents increasingly operate over information spaces that are too large to process exhaustively. Yet many multi-agent systems organize computation around static partitions of the available space, causing coordination to grow with how information is segmented rather than with what the query still requires. We introduce ANTMAN, an adaptive coordination framework that treats evolving unresolved information needs as the unit of runtime coordination. ANTMAN maintains a revisable Need Graph that tracks unresolved requirements, accumulated evidence, prior attempts, and search progress, and uses this state to control worker selection, routing, and task-local recovery as new evidence is discovered. By separating the coordination policy from substrate-specific search interfaces, the same need-conditioned mechanism can operate across different information spaces. Experiments across multi-document question answering, controlled long-context scaling, and realistic structured navigation show that ANTMAN remains effective across settings, including when execution is delegated to substantially smaller worker models. Under a 16x increase in searchable context, ANTMAN increases active coordination by only 1.23x, compared with more than 15x for partition-driven baselines, while preserving strong answer quality.
Chinese Translation
信息检索代理越来越多地在无法全面处理的大型信息空间中运作。然而，许多多智能体系统围绕可用空间的静态划分组织计算，导致协调随着信息的分段而增长，而不是随着查询的实际需求。我们提出了ANTMAN，一个自适应协调框架，将不断演变的未解决信息需求视为运行时协调的单位。ANTMAN维护一个可修订的需求图（Need Graph），跟踪未解决的需求、累积的证据、先前的尝试和搜索进展，并利用这一状态控制工作者选择、路由和任务局部恢复，随着新证据的发现而调整。通过将协调策略与特定底层搜索接口分离，相同的需求条件机制可以在不同的信息空间中运行。在多文档问答、受控长上下文扩展和现实结构化导航的实验中，ANTMAN在不同设置下仍然有效，包括在执行委托给显著较小的工作者模型时。在可搜索上下文增加16倍的情况下，ANTMAN的主动协调仅增加1.23倍，而基于分区的基线则超过15倍，同时保持强大的答案质量。
cs.AI / 202 / 2609.33339
Naturalness-guided Manifold Flow Matching for Sign Language Production
自然性引导的流形流匹配用于手语生成
Abstract
Sign Language Production (SLP) aims to generate sign motions from text. Conditional Flow Matching methods have achieved strong performance in SLP by constructing conditional paths that transform a source distribution into a target distribution. However, existing methods construct these paths via linear interpolation, whereas the rotational geometry of human joints confines valid joint rotations to a manifold embedded in Euclidean space. Consequently, linear interpolation between two sign motions leaves this manifold and ignores the motion distribution on it. In this paper, we revisit SLP from the perspective of manifold transport and propose a Naturalness-guided Manifold Flow Matching framework, termed \textbf{SignNMFlow}, which constructs conditional paths directly on the motion manifold by jointly considering geometric efficiency and the motion distribution. Specifically, we exploit the intrinsic geometry of the manifold and introduce a motion naturalness measure to characterize the motion distribution. By minimizing the kinetic energy under this measure, we learn a naturalness-guided interpolation that couples a closed-form geodesic, which provides geometrically efficient transport, with a learnable deviation that incorporates the motion distribution, thereby significantly improving the fidelity of generated sign motions. Extensive qualitative and quantitative evaluations demonstrate the effectiveness of this work.
Chinese Translation
手语生成（SLP）旨在从文本生成手势动作。条件流匹配方法通过构建将源分布转化为目标分布的条件路径，在手语生成中取得了良好的表现。然而，现有方法通过线性插值构建这些路径，而人类关节的旋转几何特性将有效的关节旋转限制在嵌入欧几里得空间的流形上。因此，在两个手势动作之间进行线性插值会离开该流形，并忽略其上的运动分布。本文从流形传输的角度重新审视手语生成，并提出了一种自然性引导的流形流匹配框架，称为 extbf{SignNMFlow}，该框架通过联合考虑几何效率和运动分布，直接在运动流形上构建条件路径。具体而言，我们利用流形的内在几何特性，引入运动自然性度量来表征运动分布。通过在该度量下最小化动能，我们学习了一种自然性引导的插值方法，该方法将提供几何高效传输的封闭形式测地线与结合运动分布的可学习偏差相结合，从而显著提高生成手势动作的保真度。大量的定性和定量评估证明了该工作的有效性。
cs.AI / 203 / 2609.33351
QuPID: Quantum Parameter-Efficient Input-Dependent Retrieval Adaptation for Medical RAG
QuPID：用于医疗RAG的量子参数高效输入依赖检索适应
Abstract
Fidelity-based quantum retrieval ranks candidates by the fidelity between query and archive states. Applying a shared input-independent unitary after fixed state encoding leaves that fidelity unchanged, so training the circuit cannot alter the ranking. Quantum parameter-efficient input-dependent retrieval adaptation (QuPID) repairs this by making the circuit input-dependent through data re-uploading and by comparing measurement readouts, vectors of local Pauli expectations, rather than states. The result is a small readout for adapting frozen image features to a local archive with limited data: training simulates the circuit classically, and inference runs on a GPU with fixed learned parameters. We characterize the class as a structured factorization of input-modulated quadratic feature maps, bound the frequency support of its re-uploading channel, and give a parameter-count generalization bound that motivates its small budget. Under a shared frozen backbone and a label-free protocol, QuPID's 60 parameters give higher precision-at-5 (P@5) on ChestX-ray14 and MURA than frozen medical encoders, and than adapters and low-rank adaptation (LoRA) with up to 5.25 million trainable parameters. On ChestX-ray14, the P@5 gain over the frozen encoder is +0.116, the lead over retuned adapters is widest at 512 adaptation examples (+0.040), and the full-budget margin over an equally compact classical rotation-plane head is +0.023 with a 95% interval excluding zero. Medical imaging is the primary testbed; the pattern recurs on two non-medical benchmarks, in report generation, and under simulated gate noise and finite-shot readout.
Chinese Translation
基于保真度的量子检索通过查询和档案状态之间的保真度对候选项进行排名。在固定状态编码后应用共享的输入无关幺正算子不会改变该保真度，因此训练电路无法改变排名。量子参数高效输入依赖检索适应（QuPID）通过数据重新上传使电路依赖于输入，并通过比较测量读数、局部保利期望的向量，而不是状态，来修复这一问题。结果是一个小的读出，用于将冻结的图像特征适应到有限数据的本地档案：训练在经典上模拟电路，推理在具有固定学习参数的GPU上运行。我们将该类特征表征为输入调制的二次特征映射的结构因式分解，界定其重新上传通道的频率支持，并给出一个参数计数的泛化界限，以激励其小预算。在共享的冻结骨干和无标签协议下，QuPID的60个参数在ChestX-ray14和MURA上的5个精度（P@5）优于冻结的医学编码器，以及参数可训练高达525万个的适配器和低秩适应（LoRA）。在ChestX-ray14上，相较于冻结编码器，P@5的增益为+0.116，相较于重新调优的适配器在512个适应示例下的领先幅度最大（+0.040），而在与同样紧凑的经典旋转平面头的全预算差距为+0.023，95%的区间不包括零。医学成像是主要的测试平台；该模式在两个非医学基准、报告生成以及在模拟门噪声和有限拍读出下重复出现。
cs.AI / 204 / 2609.33355
Unmask the State: When Does State Adaptation Matter for Masked Diffusion Language Models
揭示状态：状态适应何时对掩蔽扩散语言模型重要
Abstract
Masked diffusion language models (MDMs) admit flexible generation orders, making the unmasking strategy an inference decision. Existing methods vary in how they prioritize positions, control parallelism, restrict selection regions, revise predictions, or plan future denoising, yet it remains unclear when these choices should change during generation. We study this question through strategy reversals, where an alternative action becomes preferable to a fixed choice. We organize MDM inference into five axes--score, cardinality, region, commitment, and planning--and define adaptation opportunity as the one-step utility advantage of the best candidate action over a validation-selected fixed action. This view shows that adaptation value depends on both the frequency and magnitude of such reversals. Across three MDMs and ten tasks, adaptation opportunities are highly heterogeneous, with some regimes exhibiting concentrated and predictable one-step gains. This motivates selective adaptation: lightweight detectors calibrated on validation prompts identify high-opportunity states, capturing, for example, 56.9 percent of the candidate-set oracle opportunity by adapting only the top 10 percent of states on LLaDA-8B constrained JSON filling. Our transition-level results suggest that state adaptation is most useful when applied selectively rather than uniformly.
Chinese Translation
掩蔽扩散语言模型（MDMs）允许灵活的生成顺序，使得解掩策略成为一种推理决策。现有方法在如何优先考虑位置、控制并行性、限制选择区域、修正预测或规划未来去噪方面各有不同，但在生成过程中何时应改变这些选择仍不清楚。我们通过策略逆转来研究这个问题，其中替代行动比固定选择更为可取。我们将MDM推理组织为五个维度——得分、基数、区域、承诺和规划——并将适应机会定义为最佳候选行动相对于验证选择的固定行动的一步效用优势。这一观点表明，适应价值依赖于此类逆转的频率和幅度。在三个MDM和十个任务中，适应机会高度异质化，某些模式表现出集中且可预测的一步收益。这激励了选择性适应：在验证提示上校准的轻量级检测器识别高机会状态，例如，通过仅适应LLaDA-8B约束JSON填充中前10%的状态，捕获了56.9%的候选集oracle机会。我们的过渡级结果表明，状态适应在选择性应用时比均匀应用更为有用。
cs.AI / 205 / 2609.33356
Long-Horizon Analog Design Bench: Benchmarking Agents on Hours-Long Analog and Mixed-Signal Circuit Design Tasks
长时间视野模拟设计基准：在数小时的模拟和混合信号电路设计任务中对代理进行基准测试
Abstract
Coding agents now sustain hours-long, tool-driven loops, yet their ability to carry long-horizon analog and mixed-signal circuits to electrical specification remains unmeasured. We introduce Analog Design Bench, a long-horizon agentic benchmark of 50 transistor-level design tasks contributed by 17 chip designers. Agents work with an open-source simulator, while an isolated verifier evaluates the submitted circuit using specification-based electrical tests. We evaluate 15 agent configurations across 2,250 two-hour attempts and observe full-specification pass rates from 8.0% to 78.0%. Coding-benchmark performance correlates with analog results but leaves much of the performance spread unexplained. Our failure analysis shows that most unsuccessful submissions have no recorded legality rejection but fail electrical acceptance, identifying electrical closure as the dominant endpoint challenge. We test time, reasoning effort, agent harness, and supplied design knowledge as interventions. Longer budgets and higher reasoning effort improve performance, while general skill documents provide little benefit and sometimes reduce performance. Supplying a task-matched reference topology, an idealized form of circuit-IP retrieval, raises DeepSeek V4 Pro by 18.7 percentage points and mainly accelerates GPT-5.6 Sol.
Chinese Translation
编码代理现在能够维持数小时的工具驱动循环，但它们在将长时间视野的模拟和混合信号电路带入电气规范方面的能力尚未得到测量。我们引入了模拟设计基准（Analog Design Bench），这是一个包含17位芯片设计师贡献的50个晶体管级设计任务的长时间视野代理基准。代理与一个开源模拟器协作，而一个独立的验证器使用基于规范的电气测试来评估提交的电路。我们在2250次两小时的尝试中评估了15种代理配置，并观察到完全规范通过率从8.0%到78.0%不等。编码基准性能与模拟结果相关，但仍有大量性能差异无法解释。我们的失败分析显示，大多数未成功的提交没有记录的合法性拒绝，但未能通过电气接受，识别出电气闭合是主要的终点挑战。我们测试了时间、推理努力、代理框架和提供的设计知识作为干预措施。更长的预算和更高的推理努力提高了性能，而一般技能文档提供的好处有限，有时甚至降低了性能。提供与任务匹配的参考拓扑，即电路知识产权检索的理想化形式，使DeepSeek V4 Pro提高了18.7个百分点，并主要加速了GPT-5.6 Sol。
cs.AI / 206 / 2609.33357
DISCERN: Can AI Agents Work Like Scientists and Guide Discovery?
DISCERN：人工智能代理能否像科学家一样工作并指导发现？
Abstract
Reliable automated research requires agents to vet data, verify analyses, and generate hypotheses grounded in trustworthy evidence, potentially reducing routine scientific workload while allowing scientists to focus on interpretation and discovery. Existing benchmarks often only assess analytical task completion or hypothesis generation separately rather than testing whether reliable evidence supports valid and novel claims. We introduce DISCERN (Data Integrity and Scientific Capability: Evidence, Reasoning, and Novelty), a controlled benchmark on real, publicly available datasets that evaluates three key levels of an automated research workflow. The first two levels test data integrity and analysis verification under confounds and tool traps, while the third tests hypothesis generation and revision under adversarial review, including counterfactual cases in which evidence consistent with real data and documented scientific phenomena conflicts with established expectations, motivating alternative explanations and testable hypotheses. Across 203 tasks, eight life-science tracks, and eight models, DISCERN shows that strong aggregate performance can mask level-specific weaknesses. Agents earn perfect scores in only 60.8% of Level 1, 34.2% of Level 2, and 0.6% of Level 3 evaluations, with penalties attributed to rejection of sound data, failure to carry recognized limitations into conclusions, and wide variation in hypothesis production. Cross-track rankings by token and code use are substantially more stable than rankings by evidence judgment, suggesting greater consistency in computational effort than in evidence-based reasoning. These profiles identify opportunities for supervised scientific assistance, but current agents do not yet demonstrate reliable autonomous analysis or discovery. Code and data: https://huggingface.co/datasets/discern-bench-anon/discern-benchmark
Chinese Translation
可靠的自动化研究要求代理对数据进行审核、验证分析，并生成基于可信证据的假设，这有可能减少常规科学工作负担，同时让科学家专注于解释和发现。现有基准通常仅评估分析任务的完成或假设生成，而不是测试可靠证据是否支持有效且新颖的主张。我们引入了DISCERN（数据完整性与科学能力：证据、推理与新颖性），这是一个基于真实、公开可用数据集的受控基准，评估自动化研究工作流程的三个关键层面。前两个层面测试数据完整性和分析验证在混淆因素和工具陷阱下的表现，而第三个层面则测试假设生成和修订在对抗性审查下的表现，包括与真实数据和已记录科学现象一致的证据与既定预期相冲突的反事实案例，从而激励替代解释和可测试假设。在203个任务、八个生命科学轨道和八个模型中，DISCERN显示出强大的整体表现可能掩盖特定层面的弱点。代理在第一层级的完美得分仅为60.8%，第二层级为34.2%，第三层级为0.6%，惩罚原因包括拒绝可靠数据、未能将已识别的局限性纳入结论，以及假设生成的广泛变异。按标记和代码使用的跨轨道排名比按证据判断的排名稳定得多，这表明计算努力的一致性大于基于证据的推理一致性。这些特征识别了监督科学辅助的机会，但当前的代理尚未展示出可靠的自主分析或发现能力。代码和数据：https://huggingface.co/datasets/discern-bench-anon/discern-benchmark
cs.AI / 207 / 2609.33368
DrafTS: Time-Aware Decomposition with Residual Correction for Time Series Modeling
DrafTS：具有残差校正的时间感知分解用于时间序列建模
Abstract
Real-world time series contain evolving underlying dynamics with irregular variations that lack stable temporal patterns and are often referred to as noise. Existing methods address this mixture by filtering frequencies or suppressing noisy observations. They either miss temporal evolution or risk suppressing useful dynamics. We propose DrafTS, a model-agnostic framework that aims to reduce noise while preserving evolving dynamics through time-aware Decomposition with ResiduAl correction For Time Series. DrafTS uses features derived from instantaneous amplitude and frequency to guide decomposition into a primary component intended to capture underlying dynamics. A task-specific backbone models the primary component, while a lightweight correction module uses residual information to correct the backbone output. Across four time series modeling tasks, DrafTS improves six diverse backbones, demonstrating its effectiveness. Code is at https://github.com/Autumn61q/DrafTS
Chinese Translation
现实世界的时间序列包含不断演变的潜在动态，具有不规则的变化，缺乏稳定的时间模式，通常被称为噪声。现有方法通过过滤频率或抑制噪声观测来处理这种混合情况，但它们要么错过时间演变，要么冒着抑制有用动态的风险。我们提出了DrafTS，这是一个模型无关的框架，旨在通过时间感知的分解与残差校正来减少噪声，同时保留演变动态。DrafTS利用从瞬时幅度和频率中提取的特征来指导分解为一个主要成分，旨在捕捉潜在动态。一个任务特定的主干模型负责建模主要成分，而一个轻量级的校正模块则利用残差信息来校正主干输出。在四个时间序列建模任务中，DrafTS改进了六种不同的主干，证明了其有效性。代码可在 https://github.com/Autumn61q/DrafTS 获取。
cs.AI / 208 / 2609.33394
Cross-modal Translation via Conditional Latent Denoising for Video Deepfake Detection
通过条件潜在去噪的跨模态翻译实现视频深度伪造检测
Abstract
The growing threat of video deepfakes necessitates multimodal detection. Beyond serving as independent indicators of authenticity, audio and visual signals have intrinsic dependencies that also provide an essential criterion for detection. Previous methods often overlook the cross-modal correspondences, hindering information transfer between domains and leaving crucial detection cues unexplored. To address this challenge, we propose a framework called Cross-modal Translation via Conditional Latent Denoising (CTCLD) for video deepfake detection. It connects the distinct distributions of heterogeneous modalities in latent spaces, enabling smooth cross-domain information transfer to improve detection performance. We first establish a Bayesian foundation by decomposing the audio-visual joint distribution. Subsequently, CTCLD translates both modalities via bidirectional latent denoising conditioned on each other, effectively capturing subtle inconsistencies in the manipulated signals. Experimental results demonstrate that the proposed CTCLD enables comprehensive domain alignment, resulting in a robust video deepfake detection approach with competitive performance.
Chinese Translation
视频深度伪造日益成为一种威胁，因此需要多模态检测。音频和视觉信号不仅作为独立的真实性指标，还具有内在的依赖关系，这为检测提供了重要的标准。以往的方法往往忽视了跨模态的对应关系，阻碍了信息在不同领域之间的传递，并留下了重要的检测线索未被探索。为了解决这一挑战，我们提出了一种名为条件潜在去噪的跨模态翻译（CTCLD）的视频深度伪造检测框架。该框架在潜在空间中连接异构模态的不同分布，促进跨领域信息的平滑传递，从而提高检测性能。我们首先通过分解音频-视觉联合分布建立了贝叶斯基础。随后，CTCLD通过相互条件的双向潜在去噪翻译两种模态，有效捕捉到被操控信号中的细微不一致性。实验结果表明，所提出的CTCLD实现了全面的领域对齐，进而形成了一种具有竞争力性能的稳健视频深度伪造检测方法。
cs.AI / 209 / 2609.33397
CoViST: Visual Token Compression via Composable States
CoViST：通过可组合状态进行视觉令牌压缩
Abstract
Visual token compression lowers the inference cost of vision--language models by representing images with fewer tokens. However, most existing methods compress visual tokens to a reduced set, leaving the amount of visual evidence represented by each token and its original spatial context implicit. Therefore, the compressed representation does not explicitly encode how much visual information each representative carries or where it lies in the original image. This limitation arises even after a single reduction and becomes more pronounced when compression is repeated across decoder layers. To address this issue, we propose CoViST, a training-free framework that represents a compressed image as a composable visual state. Specifically, the state combines representative features with original positions, effective contribution weights, and reusable selection metadata. CoViST constructs this state through coverage-guided selection and conservation-based contribution composition, and explicitly incorporates its contribution and positional information into decoder attention. Each component of the state retains its interpretation under successive reductions, enabling the same formulation to support both fixed compression before prefill and progressive compression within the decoder. Experimental results on seven LLaVA-1.5-7B benchmarks show that CoViST-Fixed retains 99.9\%, 99.5\%, and 98.1\% of uncompressed performance at 192, 128, and 64 tokens, respectively, and CoViST-Pro retains 99.8\%, 99.9\%, and 99.1\% at the corresponding layer-average budgets, outperforming state-of-the-art methods under their respective budget settings. Code will be released publicly.
Chinese Translation
视觉令牌压缩通过用更少的令牌表示图像来降低视觉-语言模型的推理成本。然而，大多数现有方法将视觉令牌压缩为一个减少的集合，使得每个令牌所代表的视觉证据量及其原始空间上下文变得隐含。因此，压缩表示并未明确编码每个代表所携带的视觉信息量或其在原始图像中的位置。这一局限性在进行单次压缩后仍然存在，并在解码器层中重复压缩时变得更加明显。为了解决这个问题，我们提出了CoViST，一个无训练的框架，将压缩图像表示为可组合的视觉状态。具体而言，该状态结合了代表性特征、原始位置、有效贡献权重和可重用的选择元数据。CoViST通过覆盖引导选择和基于保留的贡献组合构建这个状态，并将其贡献和位置信息明确地纳入解码器注意力中。状态的每个组件在连续压缩下保持其解释，使得相同的公式能够支持在预填充前的固定压缩和解码器内的渐进压缩。在七个LLaVA-1.5-7B基准上的实验结果表明，CoViST-Fixed在192、128和64个令牌时分别保留了99.9%、99.5%和98.1%的未压缩性能，而CoViST-Pro在相应的层平均预算下保留了99.8%、99.9%和99.1%，在各自的预算设置下优于最先进的方法。代码将公开发布。
cs.AI / 210 / 2609.33398
COEVO: Co-Evolving Context and Parameters for Recursive Self-Improvement
COEVO：递归自我改进中的上下文与参数的共同演化
Abstract
Recursive self-improvement (RSI) seeks to move large language models beyond static training pipelines toward systems that can participate in improving their own future behavior. Existing approaches largely follow two directions: updating model parameters through online learning, or improving the external context through search, reflection, and prompt optimization. Although both mechanisms can support continued improvement, they are typically studied independently. This separation overlooks an important interaction: the context shapes the experience from which a model learns, while an evolving model may interpret and utilize the same context differently over time. We therefore formulate RSI as a problem of parameter--context co-evolution, where model parameters and the learning context adapt within a shared feedback loop. We introduce COEVO, a framework that updates model parameters from on-policy experience while adapting contextual guidance according to the state of the evolving policy. Policy entropy and prompt-conditioned attention are used as complementary signals to guide this adaptation. Experiments show that COEVO consistently improves task performance over fixed-context reinforcement learning and produces policies that are more robust to changes in system prompts. More broadly, our results suggest that external context should be viewed not merely as a fixed interface to a large language model, but as an adaptive component of recursive self-improvement.
Chinese Translation
递归自我改进（RSI）旨在使大型语言模型超越静态训练流程，朝向能够参与改善自身未来行为的系统。现有方法主要沿两个方向发展：通过在线学习更新模型参数，或通过搜索、反思和提示优化改善外部上下文。尽管这两种机制都可以支持持续改进，但通常是独立研究的。这种分离忽视了一个重要的互动关系：上下文塑造了模型学习的经验，而不断演化的模型可能会随着时间的推移以不同的方式解释和利用相同的上下文。因此，我们将RSI表述为参数-上下文共同演化的问题，其中模型参数和学习上下文在共享反馈循环中适应。我们引入了COEVO，一个从策略经验中更新模型参数的框架，同时根据演化策略的状态调整上下文指导。策略熵和提示条件注意力被用作指导这种适应的互补信号。实验表明，COEVO在固定上下文强化学习上持续提高任务性能，并产生对系统提示变化更具鲁棒性的策略。更广泛地说，我们的结果表明，外部上下文不应仅被视为大型语言模型的固定接口，而应视为递归自我改进的一个自适应组成部分。
cs.AI / 211 / 2609.33411
MetaBench-Harness: Unlocking End-to-End Optimization of Benchmark Harnesses
MetaBench-Harness：解锁基准测试工具的端到端优化
Abstract
Rapid progress in Large Language Models (LLMs) is saturating static benchmarks faster than they can be designed. While existing automated evolution frameworks attempt to generate harder questions by perturbing individual tasks, they remain constrained by rigid, hard-coded generation rules. Moving beyond the evolution of isolated tasks, we propose to optimize the benchmark generation workflow itself end to end with MetaBench-Harness, a dual-loop search framework. Specifically, the inner loop utilizes a benchmark harness to generate a new benchmark in each round, while the outer meta-harness orchestration layer iteratively refines and searches over harness implementations based on historical evolution trajectories. By applying MetaBench-Harness to the competitive programming CodeContests and Olympiad mathematics AIME-2024 datasets, we demonstrate that the evolved benchmarks are challenging and discriminative for frontier models. Trajectory and quality analyses verify that MetaBench-Harness enables multi-dimensional evolution, steadily improving evolution reasonableness, benchmark competency, and evaluator robustness across successive rounds. Furthermore, case studies reveal its effective utilization of diverse difficulty levers to reframe problems and elevate required capabilities. Ultimately, this work provides a solution to the pressing challenge of benchmark saturation.
Chinese Translation
大型语言模型（LLMs）的快速进展正在以超过其设计速度的方式饱和静态基准测试。虽然现有的自动演化框架试图通过扰动单个任务来生成更难的问题，但它们仍然受到僵化的硬编码生成规则的限制。我们提出了MetaBench-Harness，一个双循环搜索框架，旨在超越孤立任务的演化，优化基准生成工作流程的端到端。具体而言，内循环利用基准测试工具在每一轮生成新的基准，而外部元工具编排层则基于历史演化轨迹迭代地细化和搜索工具实现。通过将MetaBench-Harness应用于竞争编程的CodeContests和奥林匹克数学AIME-2024数据集，我们证明了演化后的基准对前沿模型具有挑战性和区分性。轨迹和质量分析验证了MetaBench-Harness能够实现多维演化，稳步提高演化合理性、基准能力和评估者的鲁棒性。此外，案例研究揭示了其有效利用多样化难度杠杆来重新构建问题和提升所需能力。最终，这项工作为基准饱和这一紧迫挑战提供了解决方案。
cs.AI / 212 / 2609.33428
Temporal Graph Learning of Wearable Actigraphy and Sleep Traces for Modelling Adolescent Crystallized Intelligence
可穿戴活动监测与睡眠轨迹的时间图学习用于建模青少年晶体智力
Abstract
Wearable actigraphy offers a scalable, ecologically valid alternative to episodic clinical assessment. However, predicting continuous adolescent crystallized intelligence ($G_c$) from such traces remains challenging due to irregular device adherence and complex behavioral-environmental interactions. We address this using daily summary data derived from 21-day Fitbit records of 6,091 adolescents in the Adolescent Brain Cognitive Development Study (Release 5.1). We propose SATURN, a Sleep-Activity Temporal Unified Regression Network. It represents participants as 21-node temporal graphs encoding daily behaviors and temporal adjacency. To prevent imputation artifacts, invalid-day edges are dynamically pruned during forward passes. Node embeddings are refined via residual GATv2 layers, aggregated through masked attention pooling, and fused with sociodemographic covariates. Under family-controlled, age-sex-BMI-stratified cross-validation, SATURN achieves $R^2 = 0.2783 \pm 0.0127$, consistently improving upon flattened machine learning (Gradient Boosting, $R^2 = 0.2372$) and sequential deep learning (BiLSTM, $R^2 = 0.2688$) baselines. Explainability analyses identify light activity, metabolic equivalents, and sleep duration as dominant predictors, while Monte Carlo dropout and subgroup analyses confirm equitable performance across sociodemographic strata. Ultimately, SATURN establishes a rigorous computational framework for digital cognitive phenotyping, offering a scalable pathway to complement traditional assessments by highlighting macro-level behavioral anomalies.
Chinese Translation
可穿戴活动监测提供了一种可扩展的、生态有效的替代方案，以替代临床评估。然而，由于设备使用不规律和复杂的行为-环境交互，从这些轨迹预测连续的青少年晶体智力（$G_c$）仍然具有挑战性。我们使用来自青少年大脑认知发展研究（Release 5.1）中6091名青少年的21天Fitbit记录所衍生的每日总结数据来解决这一问题。我们提出了SATURN（睡眠-活动时间统一回归网络）。该网络将参与者表示为21节点的时间图，编码每日行为和时间邻接性。为了防止插补伪影，在前向传播过程中动态修剪无效日期边。节点嵌入通过残差GATv2层进行精炼，通过掩蔽注意力池化进行聚合，并与社会人口学协变量融合。在家庭控制、年龄-性别-BMI分层的交叉验证下，SATURN实现了$R^2 = 0.2783 ext{±} 0.0127$，持续优于扁平化机器学习（梯度提升，$R^2 = 0.2372$）和序列深度学习（BiLSTM，$R^2 = 0.2688$）基线。可解释性分析确定轻度活动、代谢当量和睡眠时长为主要预测因子，而蒙特卡洛丢弃法和子组分析则确认了在社会人口学层面上的公平表现。最终，SATURN建立了一个严格的计算框架用于数字认知表型分析，提供了一条可扩展的路径，以通过突出宏观层面的行为异常来补充传统评估。
cs.AI / 213 / 2609.33430
APEX: An Extensible Model for Agent-Assisted Production Scheduling
APEX：一种可扩展的代理辅助生产调度模型
Abstract
Production scheduling requires realistic models that reflect operational constraints and efficient methods that balance competing goals. Putting these methods into use also requires data integration, model adaptation and specialist expertise. We present APEX, an extensible production scheduling framework built around a general model and hybrid multiobjective search. Agent assistance supports both scheduling and model refinement: agents prepare data and explore scenarios in natural language, while coding agents help implement and test new constraints and objectives. Shared construction and checking procedures connect these adaptations to the scheduling core. We benchmark eight APEX configurations against NSGA-II, SPEA2, MOEA/D and SMS-EMOA on 69 public job-shop, flexible job-shop and permutation flow-shop instances, assessing workload completion time (makespan), total job flowtime and computation time. Hybrid configurations achieve the best aggregate solution quality, although the leading method depends on the problem class and objective. A separate synthetic workflow study uses OpenAI's GPT-6-astra as an interaction layer between the human planner and the algorithmic core, testing rule additions, plan and objective changes, and what-if comparisons. All 24 sessions completed the requested changes and passed independent checks of saved models and schedules. A separate coding evaluation produced six native implementations of an additional objective or hard constraint through predefined extension hooks. All passed independent checks without modifying the core.
Chinese Translation
生产调度需要反映操作约束的现实模型和能够平衡竞争目标的高效方法。将这些方法付诸实践还需要数据集成、模型适应和专业知识。我们提出了APEX，一个围绕通用模型和混合多目标搜索构建的可扩展生产调度框架。代理辅助支持调度和模型优化：代理准备数据并以自然语言探索场景，而编码代理则帮助实现和测试新的约束和目标。共享构建和检查程序将这些适应与调度核心连接起来。我们在69个公共作业车间、灵活作业车间和排列流车间实例上，将八种APEX配置与NSGA-II、SPEA2、MOEA/D和SMS-EMOA进行了基准测试，评估工作负载完成时间（完工时间）、总作业流动时间和计算时间。混合配置实现了最佳的整体解决方案质量，尽管领先的方法依赖于问题类别和目标。一个独立的合成工作流研究使用OpenAI的GPT-6-astra作为人类规划者与算法核心之间的交互层，测试规则添加、计划和目标变更以及假设比较。所有24个会话都完成了请求的更改，并通过了对保存模型和调度的独立检查。另一个编码评估通过预定义的扩展钩子生成了六个额外目标或硬约束的本地实现，所有实现都在不修改核心的情况下通过了独立检查。
cs.AI / 214 / 2609.33439
Raven: The Harness of Harnesses for Composable Agentic Intelligence
Raven：可组合智能代理的多重框架
Abstract
As large language models advance, AI agents are moving beyond isolated, domain-specific tasks toward long-horizon, cross-domain workflows. This transition exposes two challenges: increasing harness complexity makes manual design difficult to scale, while tighter coupling to specific domains limits the generality of a single harness. The central question thus shifts from how to engineer a stronger harness for one domain to how to autonomously construct specialized harnesses, improve them through experience, and orchestrate them across domains. We introduce Raven, \emph{The Harness of Harnesses}, an open-source multi-agent ecosystem that automatically constructs and evolves modular harnesses for specific models and domains, treating each executable model--harness pair as a composable unit of intelligence. To support an \emph{All-Domain Collaboration Network}, its Host Agent decomposes goals, matches subtasks to specialized agents, coordinates execution dependencies, and integrates results, while a host archive and EverOS preserve experience across tasks and Skill Forge makes that experience available as reusable procedures. Our theory establishes sufficient conditions for such composition to expand reliable task coverage beyond that of the available individual agents under a shared resource budget. On complex and long-horizon tasks, Raven significantly outperforms the state-of-the-art agent systems, pushing the frontier of composable agentic intelligence.
Chinese Translation
随着大型语言模型的进步，人工智能代理正在从孤立的、特定领域的任务转向长时间跨度的跨领域工作流程。这一转变暴露出两个挑战：日益复杂的框架使得手动设计难以扩展，而与特定领域的紧密耦合限制了单一框架的通用性。因此，核心问题从如何为一个领域设计更强大的框架转变为如何自主构建专业化框架，通过经验改进它们，并在不同领域之间进行协调。我们介绍了Raven， extit{多重框架的框架}，这是一个开源的多代理生态系统，能够自动构建和演化针对特定模型和领域的模块化框架，将每个可执行的模型-框架对视为一个可组合的智能单元。为了支持 extit{全领域协作网络}，其主代理分解目标，将子任务匹配到专业代理，协调执行依赖关系，并整合结果，同时主档案和EverOS在任务之间保存经验，而Skill Forge使得这些经验可以作为可重用的程序使用。我们的理论建立了这样的组合足够条件，以扩展在共享资源预算下可靠任务覆盖的范围，超越可用单个代理的能力。在复杂和长时间跨度的任务中，Raven显著优于最先进的代理系统，推动了可组合智能代理的前沿。
cs.AI / 215 / 2609.33440
MAC-Net: A Multi-Task Deep Learning Framework for Modeling Cognitive Function From Task-Based fMRI
MAC-Net：一种多任务深度学习框架，用于基于任务的功能性磁共振成像建模认知功能
Abstract
Objective cognitive assessment from neural signals supports neurorehabilitation, but individual-level prediction from task-based fMRI (tfMRI) remains difficult because neural features coexist with substantial demographic and scanner-related variation. We present the Multi-task Activation and Contrast Network (MAC-Net), a covariate-aware deep learning framework for modeling individual cognitive function from regional tfMRI. By isolating tfMRI features into a dedicated neural pathway and restricting participant variables to a terminal late-fusion pathway, MAC-Net prevents dominant covariates from suppressing high-dimensional clinical representations during feature learning. Evaluating baseline data from 6,500 Adolescent Brain Cognitive Development Study participants under family-aware cross-validation, MAC-Net was benchmarked against linear models, random forests, and alternative deep architectures. The N-back plus Monetary Incentive Delay configuration achieved $R^{2}$ values of 0.174, 0.238, and 0.277 for fluid, crystallized, and total cognition, outperforming covariate-only baselines (0.178) and alternative deep models (0.217). N-back was the most informative paradigm, whereas incorporating the Stop Signal Task marginally degraded performance. Feature attributions via Integrated Gradients, DeepLIFT, and Input Gradient were highly concordant, localizing working-memory-related frontal, parietal, and cingulate regions. These findings demonstrate that covariate-aware multi-task modeling yields reproducible cognitive-function estimations, establishing a robust neural engineering framework for clinical translation.
Chinese Translation
基于神经信号的客观认知评估支持神经康复，但由于神经特征与显著的人口统计学和扫描仪相关变异共存，基于任务的功能性磁共振成像（tfMRI）在个体水平的预测仍然困难。我们提出了多任务激活与对比网络（MAC-Net），这是一个考虑协变量的深度学习框架，用于从区域tfMRI建模个体认知功能。通过将tfMRI特征隔离到专用神经通路，并将参与者变量限制在终端的晚融合通路，MAC-Net防止了主导协变量在特征学习过程中抑制高维临床表征。通过对6,500名青少年大脑认知发展研究参与者的基线数据进行家庭意识交叉验证，MAC-Net与线性模型、随机森林和其他深度架构进行了基准测试。N-back加货币激励延迟配置在流体、晶体和总认知方面分别达到了$R^{2}$值为0.174、0.238和0.277，超越了仅考虑协变量的基线（0.178）和其他深度模型（0.217）。N-back是信息量最大的范式，而纳入停止信号任务则略微降低了性能。通过集成梯度（Integrated Gradients）、DeepLIFT和输入梯度（Input Gradient）进行的特征归因高度一致，定位了与工作记忆相关的额叶、顶叶和扣带回区域。这些发现表明，考虑协变量的多任务建模能够产生可重复的认知功能估计，为临床转化建立了一个稳健的神经工程框架。
cs.AI / 216 / 2609.33455
What Shared Prefixes Hide: Trajectory Dropout for On-Policy Distillation
共享前缀隐藏的内容：用于在线蒸馏的轨迹丢弃
Abstract
On-policy distillation (OPD) trains a student model on its own trajectories using dense token-level feedback from a stronger teacher model. Since each update is conditioned on the reasoning prefix already generated by the student, the prefix also shapes how effectively teacher feedback is converted into learning. We find that shared prefixes can lead to weak token-level updates, a phenomenon we call Prefix-Induced Supervision Attenuation (PISA). This attenuation arises in two common cases. (i) High student confidence can weaken corrective gradients even when the teacher disagrees. (ii) Tokens that rely on earlier reasoning can receive learning signals as weak as those for simple local continuations. To solve this problem, we propose Trajectory Dropout, a simple training-time intervention that exposes these weakened signals. The student first performs a standard full-context rollout to generate a complete trajectory. During training, we randomly drop a certain proportion of the student's reasoning trajectory, while the teacher continues to observe the complete trajectory for token-level supervision. This intervention strengthens corrections for overconfident predictions and introduces additional supervision at prefix-sensitive positions. Trajectory Dropout consistently improves average performance across teacher--student model pairs of different scales and six mathematical reasoning benchmarks, while also yielding gains on two out-of-domain benchmarks. It can also be flexibly integrated into existing OPD variants with negligible computational overhead, further improving their performance. These results demonstrate that Trajectory Dropout provides a simple mechanism for strengthening token-level supervision across model scales and OPD objectives.
Chinese Translation
在线蒸馏（On-policy distillation, OPD）通过来自更强教师模型的密集令牌级反馈，训练学生模型在其自身轨迹上进行学习。由于每次更新都依赖于学生已经生成的推理前缀，因此前缀也影响教师反馈转化为学习的有效性。我们发现，共享前缀可能导致弱的令牌级更新，这一现象我们称之为前缀诱导监督衰减（Prefix-Induced Supervision Attenuation, PISA）。这种衰减在两种常见情况下出现：（i）即使教师不同意，高学生信心也可能削弱纠正梯度；（ii）依赖于早期推理的令牌可能会收到与简单局部延续相同弱的学习信号。为了解决这个问题，我们提出了轨迹丢弃（Trajectory Dropout），这是一种简单的训练时干预，能够暴露这些减弱的信号。学生首先执行标准的全上下文展开以生成完整轨迹。在训练过程中，我们随机丢弃学生推理轨迹中的一定比例，而教师则继续观察完整轨迹以进行令牌级监督。这一干预增强了对过于自信预测的纠正，并在对前缀敏感的位置引入了额外的监督。轨迹丢弃在不同规模的教师-学生模型对及六个数学推理基准上持续提高平均性能，同时在两个域外基准上也取得了提升。它还可以灵活地集成到现有的OPD变体中，几乎没有计算开销，进一步提升其性能。这些结果表明，轨迹丢弃提供了一种简单的机制，以增强跨模型规模和OPD目标的令牌级监督。
cs.AI / 217 / 2609.33458
When Does the Concept of "Dog" Emerge in an Audio LLM?
“狗”的概念何时在音频大型语言模型中出现？
Abstract
Multimodal large language models answer audio questions, but how they represent auditory semantics and use them in decisions remains unclear, limiting our understanding of response formation. We study dog barking in Qwen2.5-Omni-7B using Jacobian lens (J-lens) readout and directional interventions. We define the dog direction as a J-lens-derived hidden-state vector associated with dog; adding or removing its component modulates dog-related information. We find this information decodable without dog/bark prompt cues or animal-identification requirements. Directional interventions change response tendencies and some final answers, with effects concentrated in late-layer states immediately before generation across species classification, vocalization classification, and sound description. The dog direction shows no comparable advantage over controls in animal/other classification. These results provide causal-intervention evidence that the dog direction affects output scores in a task-dependent manner, most consistently at L22 and L24 immediately before generation.
Chinese Translation
多模态大型语言模型能够回答音频问题，但它们如何表示听觉语义并在决策中使用这些语义仍不清楚，这限制了我们对响应形成的理解。我们使用雅可比透镜（Jacobian lens, J-lens）读取和定向干预研究了 Qwen2.5-Omni-7B 中的狗叫声。我们将狗的方向定义为与狗相关的 J-lens 派生隐藏状态向量；添加或移除其成分会调节与狗相关的信息。我们发现这些信息可以在没有狗/叫声提示或动物识别要求的情况下被解码。定向干预改变了响应倾向和一些最终答案，影响集中在生成前的晚层状态，涉及物种分类、发声分类和声音描述。狗的方向在动物/其他分类中未显示出相较于对照组的明显优势。这些结果提供了因果干预证据，表明狗的方向以任务依赖的方式影响输出分数，最一致地体现在生成前的 L22 和 L24 层。
cs.AI / 218 / 2609.33470
LiveOption: Evaluating LLM Agents in Structured Option Trading with Nonlinear Payoffs
LiveOption：在具有非线性收益的结构化期权交易中评估LLM代理
Abstract
Large language models (LLMs) and multi-agent systems (MAS) have shown promise in financial decision-making, yet existing evaluations focus on equity trading and primarily assess directional prediction, overlooking the structural complexity of derivative markets. Option trading introduces fundamentally different challenges, including nonlinear payoffs and multi-leg strategy construction, requiring structured decisions rather than simple directional bets. We introduce LiveOption, an evaluation framework for LLM-based agents in option trading. LiveOption formulates the problem as structured sequential decision-making under realistic execution and capital constraints, and provides a reproducible environment with standardized interaction protocols. The framework includes three task suites covering portfolio overlays, event-driven earnings trading, and 0DTE intraday trading. We further propose a hierarchical metric suite that evaluates action validity, decision quality, risk characteristics, and outcome-level performance. Experiments show that current agents often fail to achieve competitive returns in most scenarios. LiveOption offers a principled testbed for evaluating structured decision-making beyond outcome-based metrics.
Chinese Translation
大型语言模型（LLMs）和多智能体系统（MAS）在金融决策中展现出良好的前景，但现有评估主要集中在股票交易上，主要评估方向预测，忽视了衍生品市场的结构复杂性。期权交易引入了根本不同的挑战，包括非线性收益和多腿策略构建，这需要结构化的决策，而非简单的方向性押注。我们提出了LiveOption，一个用于期权交易中基于LLM的代理评估框架。LiveOption将问题表述为在现实执行和资本约束下的结构化序列决策，并提供一个可重复的环境和标准化的交互协议。该框架包括三个任务套件，涵盖投资组合覆盖、事件驱动的收益交易和0DTE日内交易。我们进一步提出了一个分层指标套件，用于评估行动有效性、决策质量、风险特征和结果级别表现。实验表明，当前代理在大多数场景中往往无法实现具有竞争力的回报。LiveOption为评估超越基于结果的指标的结构化决策提供了一个原则性的测试平台。
cs.AI / 219 / 2609.33477
Just Let Linear States Forget the Distant Past: Prefix Caching via Suffix Replay for Hybrid LLMs
让线性状态忘记遥远的过去：通过后缀重放实现混合大型语言模型的前缀缓存
Abstract
Hybrid LLMs interleave full-attention layers with linear-attention layers to reduce long-context inference cost, but this structure complicates prefix caching. Full-attention KV caches are token-addressable, whereas linear-attention layers maintain recurrent states that cannot be rolled back to arbitrary prefix boundaries. Existing systems materialize recurrent-state checkpoints, restricting prefix reuse to checkpoint-aligned positions. We present SuffixReplay, the first prefix caching system that lets hybrid LLMs reuse cached prefixes at every cache-supported page boundary without materializing recurrent-state checkpoints. Our key insight is to just let linear states forget the distant past. Modern linear-attention mechanisms use recurrent decay and gating to attenuate the influence of old inputs. Therefore, instead of checkpointing every prefix boundary, SuffixReplay approximates the state at a matched boundary by replaying only a recent suffix of the layer's input hidden states, which we retain as anchors. At the algorithmic level, SuffixReplay combines layer-wise and token-wise anchor sparsity with a bounded replay budget to control storage, computation, and quality. At the system level, it uses an independently managed anchor sidecar and a pipelined replay path to overlap anchor movement and state reconstruction with the native serving pipeline. We evaluate SuffixReplay on three hybrid LLMs: OLMo-Hybrid-7B, Qwen3.5-4B, and Qwen3.6-27B-FP8. Across these models, SuffixReplay retains 91.4-100% of full-prefill quality on average across LongBench and RULER, while using only 0.36-0.51x the amortized per-token storage of SGLang's default 8192-token checkpoint cache. Integrated into SGLang, SuffixReplay reduces median TTFT by 15-70% on branching workloads, sustains 2.3-4.3x SGLang's throughput when the working set exceeds HBM, and matches SGLang on high-hit continuation traffic.
Chinese Translation
混合大型语言模型（Hybrid LLMs）将全注意力层与线性注意力层交替使用，以降低长上下文推理的成本，但这种结构使得前缀缓存变得复杂。全注意力的键值（KV）缓存是基于令牌可寻址的，而线性注意力层则维护无法回滚到任意前缀边界的递归状态。现有系统物化递归状态检查点，限制了前缀重用仅限于与检查点对齐的位置。我们提出了后缀重放（SuffixReplay），这是第一个允许混合大型语言模型在每个缓存支持的页面边界重用缓存前缀的前缀缓存系统，而无需物化递归状态检查点。我们的关键见解是让线性状态忘记遥远的过去。现代线性注意力机制使用递归衰减和门控来减弱旧输入的影响。因此，后缀重放并不是在每个前缀边界进行检查点，而是通过仅重放层输入隐藏状态的最近后缀来近似匹配边界的状态，并将其保留为锚点。在算法层面，后缀重放结合了层级和令牌级的锚点稀疏性以及有限的重放预算，以控制存储、计算和质量。在系统层面，它使用独立管理的锚点旁路和流水线重放路径，以重叠锚点移动和状态重建与本地服务流水线。我们在三种混合大型语言模型上评估了后缀重放：OLMo-Hybrid-7B、Qwen3.5-4B 和 Qwen3.6-27B-FP8。在这些模型中，后缀重放在 LongBench 和 RULER 上平均保留了 91.4-100% 的全前填充质量，同时仅使用 SGLang 默认 8192 令牌检查点缓存的 0.36-0.51 倍的摊销每令牌存储。集成到 SGLang 中，后缀重放在分支工作负载上将中位 TTFT 减少了 15-70%，在工作集超过 HBM 时维持了 SGLang 的 2.3-4.3 倍吞吐量，并在高命中续写流量上与 SGLang 相匹配。
cs.AI / 220 / 2609.33492
Federated Multi-Modal Human Activity Recognition using Multi-Agent Reinforcement Learning
基于多智能体强化学习的联邦多模态人类活动识别
Abstract
Human Activity Recognition (HAR) from heterogeneous wearable sensors is fundamental to the Internet of Health Things (IoHT), supporting rehabilitation, elderly care, and smart healthcare. Existing multimodal fusion methods often assign fixed equal weights to sensor streams, overlooking differences in modality importance, acquisition cost, and sensor quality, which can vary due to movement, incorrect placement, or temporary blockage. We propose an adaptive and cost-aware multimodal HAR framework based on multi-agent reinforcement learning for centralized HAR and extend it to federated learning as FedMHAR. In the centralized setting, multimodal fusion is formulated as a cooperative Multi-Agent Reinforcement Learning (MARL) problem, where each sensing modality is assigned a PPO-based agent that learns per-sample fusion weights, enabling the model to emphasize informative modalities while down-weighting costly sensors when cheaper alternatives provide sufficient information. In the federated setting, we introduce BiFL-PPO, a bidirectional federated optimization strategy in which a server-side PPO policy learns client-specific trust weights and feeds them back to adapt local learning rates and proximal regularization. Unlike round-level optimization, BiFL-PPO uses dense batch-level rewards for more frequent feedback and stable training under heterogeneous client data. Evaluation on the MEx Rehabilitation and UTD Multimodal Human Action datasets shows that the centralized framework achieves 87.30% and 94.98% accuracy, respectively, outperforming conventional fusion methods and state-of-the-art HAR models. FedMHAR achieves 79.74% and 77.49% in the federated setting, consistently surpassing FedAvg, FedProx, FedBN, FedNova, and AdaFedProx, while providing more stable performance and reducing sensor acquisition cost.
Chinese Translation
从异构可穿戴传感器中进行人类活动识别（HAR）是健康物联网（IoHT）的基础，支持康复、老年护理和智能医疗。现有的多模态融合方法通常对传感器流赋予固定的相等权重，忽视了模态重要性、获取成本和传感器质量的差异，这些差异可能由于运动、错误放置或临时阻塞而变化。我们提出了一种基于多智能体强化学习的自适应和成本感知的多模态HAR框架，用于集中式HAR，并将其扩展到联邦学习，命名为FedMHAR。在集中式设置中，多模态融合被表述为一个合作的多智能体强化学习（MARL）问题，其中每种感知模态被分配一个基于PPO（Proximal Policy Optimization）的智能体，该智能体学习每个样本的融合权重，使模型能够强调信息丰富的模态，同时在便宜的替代方案提供足够信息时降低成本传感器的权重。在联邦设置中，我们引入了BiFL-PPO，一种双向联邦优化策略，其中服务器端的PPO策略学习客户端特定的信任权重，并反馈以调整本地学习率和邻近正则化。与轮次级优化不同，BiFL-PPO使用密集的批次级奖励，以便在异构客户端数据下提供更频繁的反馈和稳定的训练。在MEx康复和UTD多模态人类动作数据集上的评估显示，集中式框架分别达到了87.30%和94.98%的准确率，优于传统融合方法和最先进的HAR模型。FedMHAR在联邦设置下达到了79.74%和77.49%的准确率，始终超越FedAvg、FedProx、FedBN、FedNova和AdaFedProx，同时提供更稳定的性能并降低传感器获取成本。
cs.AI / 221 / 2609.33503
RelaxKV: Recomputation Guided by the Query with Sparse Context Attention for Efficient KV Cache Reuse
RelaxKV：基于查询的重计算与稀疏上下文注意力引导的高效KV缓存重用
Abstract
Cross-request KV caching reduces the prefill cost of Retrieval-Augmented Generation (RAG), but conventional prefix caching severely limits cache reuse across requests. Position-Independent Caching (PIC) removes this constraint by reusing independent chunks, but their KV states miss cross-chunk interactions. Existing methods selectively recompute token states to recover these missing interactions, but primarily allocate the recomputation budget to selecting which states to recompute, while fixing the recomputation context to the full causal prefix. We introduce RelaxKV, which formulates selective cache repair as a joint allocation problem over repair targets and recomputation context. Guided by the user query, RelaxKV identifies layer-specific repair targets and restricts their recomputation to a query-relevant context, reducing attention computation. Across four decoder models, RelaxKV at a 15% anchor ratio improves aggregate LongBench performance over ProphetKV on all models. On Qwen3-14B, RelaxKV provides a stronger quality-TTFT trade-off than ProphetKV across a 5%-30% anchor-ratio sweep, and achieves the best selective results on RULER-MV and LV-Eval at 16K and 32K context lengths. Controlled ablations further demonstrate the importance of recomputation context selection.
Chinese Translation
跨请求的KV缓存减少了检索增强生成（RAG）的预填充成本，但传统的前缀缓存严重限制了跨请求的缓存重用。位置无关缓存（PIC）通过重用独立块消除了这一限制，但它们的KV状态缺乏跨块交互。现有方法选择性地重新计算令牌状态以恢复这些缺失的交互，但主要将重计算预算分配给选择哪些状态进行重计算，同时将重计算上下文固定为完整的因果前缀。我们提出了RelaxKV，它将选择性缓存修复公式化为针对修复目标和重计算上下文的联合分配问题。在用户查询的引导下，RelaxKV识别特定层的修复目标，并将其重计算限制在与查询相关的上下文中，从而减少注意力计算。在四个解码器模型中，RelaxKV在15%的锚定比率下提高了所有模型上相较于ProphetKV的综合LongBench性能。在Qwen3-14B上，RelaxKV在5%-30%的锚定比率范围内提供了比ProphetKV更强的质量-时间到达（TTFT）权衡，并在16K和32K上下文长度下在RULER-MV和LV-Eval上取得了最佳选择性结果。控制性消融实验进一步证明了重计算上下文选择的重要性。
cs.AI / 222 / 2609.33505
When Evidence Changes the Subject: Subject-Typed Claim Licensing for Learned Routing
当证据改变主题：针对学习路由的主题类型声明许可
Abstract
Modern learned systems increasingly combine learned components with search, repair, or external solvers. Benchmarks often measure the resulting end-to-end system, while scientific claims may concern only one component, creating an attribution problem: evidence can fail to support the requested component-level claim while still supporting a positive conclusion about the larger system. Existing evidence-to-claim methods primarily calibrate claim strength. We argue that composite systems require a second dimension: scientific subject. We address this problem with subject-typed claim licensing, which separates weaker conclusions about the requested subject from positive but non-substitutive credit about another subject. We instantiate this idea in SCOPE-Routing for preference-conditioned multigraph routing. Non-authors reproducibly apply the declared semantics; held-out review yields fewer reference-relative upward deviations than unstructured review, while the difference from a strong evidence checklist remains unresolved; and a controlled routing study shows that score-optimal and claim-eligible methods can differ while valid hybrid-system credit is preserved. These results motivate treating claim strength and scientific subject as distinct dimensions of evidence-based evaluation.
Chinese Translation
现代学习系统越来越多地将学习组件与搜索、修复或外部求解器结合在一起。基准测试通常衡量最终的端到端系统，而科学声明可能仅涉及一个组件，这就产生了归因问题：证据可能无法支持所请求的组件级声明，同时仍然支持关于更大系统的积极结论。现有的证据与声明方法主要校准声明的强度。我们认为复合系统需要第二个维度：科学主题。我们通过主题类型声明许可来解决这个问题，该方法将关于所请求主题的较弱结论与关于另一个主题的积极但非替代性信用分开。我们在SCOPE-Routing中实例化这一思想，用于偏好条件下的多重图路由。非作者可重复地应用声明的语义；保留的审查产生的参考相对上升偏差少于非结构化审查，而与强证据检查表的差异仍未解决；控制路由研究表明，得分最优和声明合格的方法可能不同，同时有效的混合系统信用得以保留。这些结果促使我们将声明强度和科学主题视为基于证据评估的不同维度。
cs.AI / 223 / 2609.33509
What Happens During Autonomous Deep Research After the User Steps Away?
用户离开后自主深度研究中发生了什么？
Abstract
In autonomous deep research, a user provides a task and relevant background, then leaves the agent to conduct an extended investigation without further human intervention. We study how this initial user information is reflected in intermediate actions and how these actions relate to final recommendations. We introduce DRaligned, a counterfactual behavioral evaluation framework built on PDR-Bench. By varying one task-relevant user factor while keeping the remaining context fixed, we compare acquisition requests, working drafts, and final reports. Source-grounded extraction, blinded local judgments, and deterministic aggregation yield coarse directional measurements while leaving ambiguous cases unresolved. Our experiments show that strong user-specific delivery can emerge from a largely shared research process: agents investigate similar broad questions but allocate requests differently, and final recommendations distinguish user conditions more clearly than explicit requests do. Reports can also integrate user factors that were not jointly visible during acquisition. In readable draft-to-report comparisons, recommendations often retain their coarse user-specific direction despite substantial rewriting. Final directional differences recur across tested agent models, execution harnesses, and evaluator models, even as execution paths vary. These findings describe how initial user information shapes autonomous research and clarify the relationship between the process an agent follows and the recommendations it delivers.
Chinese Translation
在自主深度研究中，用户提供任务和相关背景信息，然后离开代理进行延续性调查，而无需进一步的人为干预。我们研究了这些初始用户信息如何反映在中间行为中，以及这些行为与最终推荐之间的关系。我们引入了DRaligned，这是一个基于PDR-Bench的反事实行为评估框架。通过在保持其余上下文不变的情况下，改变一个与任务相关的用户因素，我们比较了获取请求、工作草稿和最终报告。源基础提取、盲目局部判断和确定性聚合产生了粗略的方向性测量，同时留下了模糊的案例未解决。我们的实验表明，强用户特定的交付可以从一个大体共享的研究过程中产生：代理调查类似的广泛问题，但请求的分配方式不同，最终推荐比明确请求更清晰地区分用户条件。报告还可以整合在获取过程中未共同可见的用户因素。在可读的草稿与报告比较中，尽管进行了大量重写，推荐往往仍保留其粗略的用户特定方向。最终的方向性差异在测试的代理模型、执行工具和评估模型中反复出现，即使执行路径有所不同。这些发现描述了初始用户信息如何塑造自主研究，并阐明了代理遵循的过程与其提供的推荐之间的关系。
cs.AI / 224 / 2609.33516
PPG-LM: A Photoplethysmography-Language Model with Multi-Level Clinical Alignment
PPG-LM：一种具有多层临床对齐的光电容积脉搏波-语言模型
Abstract
Photoplethysmography (PPG) is widely recorded by clinical monitors and consumer wearables, providing a scalable source of continuous physiological information. These recordings offer an opportunity for physiological assessment at scale, but realizing this potential requires models to learn from both signal-derived physiological supervision and broader clinical context captured in electronic health records (EHRs). This involves aligning information spanning local observations, care events, and entire visits with PPG representations at corresponding temporal scales. However, existing PPG foundation models primarily rely on task-specific prediction heads, while the medical knowledge of large language models does not necessarily translate into waveform understanding. To bridge this gap, we introduce PPG-LM, the first PPG-language model family to learn physiological representations from both signal-derived supervision and broader clinical context captured in EHRs. To construct clinically grounded captions, we develop an automatic captioning pipeline that generates segment-, event-, and visit-level descriptions from signal measurements and structured EHR records. We then learn from these pairs through a two-stage framework that first establishes segment-language correspondence through contrastive learning and waveform-conditioned captioning, then extends alignment to events and visits through time-aware aggregation and temporal statement matching. Pretrained on approximately 73k hours of PPG, PPG-LM supports language-based recognition, cross-modal retrieval, and segment captioning. Experiments on MC-MED, MIMIC-III, and VitalDB show improved retrieval and caption factuality over language-model baselines and gains over PPG and time-series foundation models on multiple clinical prediction tasks.
Chinese Translation
光电容积脉搏波（PPG）被广泛应用于临床监测和消费类可穿戴设备，提供了一种可扩展的连续生理信息来源。这些记录为大规模生理评估提供了机会，但要实现这一潜力，需要模型从信号衍生的生理监督和电子健康记录（EHRs）中捕获的更广泛临床背景中学习。这涉及将跨越局部观察、护理事件和整个就诊的相关信息与相应时间尺度上的PPG表示进行对齐。然而，现有的PPG基础模型主要依赖于特定任务的预测头，而大型语言模型的医学知识并不一定能够转化为波形理解。为了解决这一问题，我们提出了PPG-LM，这是第一个从信号衍生的监督和EHRs中捕获的更广泛临床背景中学习生理表示的PPG-语言模型系列。为了构建临床基础的描述，我们开发了一种自动描述生成管道，从信号测量和结构化EHR记录中生成段落级、事件级和就诊级的描述。然后，我们通过一个两阶段框架从这些配对中学习，首先通过对比学习和波形条件描述建立段落-语言对应关系，然后通过时间感知聚合和时间语句匹配将对齐扩展到事件和就诊。PPG-LM在约73k小时的PPG数据上进行预训练，支持基于语言的识别、跨模态检索和段落描述。在MC-MED、MIMIC-III和VitalDB上的实验表明，相较于语言模型基线，检索和描述的事实准确性有所提高，并在多个临床预测任务上相较于PPG和时间序列基础模型取得了进展。
cs.AI / 225 / 2609.33524
EverMine: Dissecting the Self-Evolution of Research Capabilities in Long-Horizon Alpha Research
EverMine：剖析长时间跨度阿尔法研究中研究能力的自我演化
Abstract
Self-evolving agents aim to turn research feedback into reusable skills, tools, and research rules. Whether these accumulated capabilities continue to improve later research requires controlled evaluation. Long-horizon alpha discovery provides a state-dependent setting: once a new factor enters the portfolio, the predictive information already covered changes, so the value of the same candidate or experience may change over time. We introduce EverMine, an empirical framework for studying self-evolving research capabilities in long-horizon alpha discovery. EverMine decomposes the research state into history (Hist), the current factor portfolio (Frontier), and reusable capabilities (Cap). Under matched resource limits, we compare complete runs with fixed or evolving Cap, and replace Cap while holding Hist and Frontier fixed to estimate the conditional value of accumulated capabilities. We also combine full trajectories with historical-state replay to examine how experience-based decisions affect candidate selection and portfolio outcomes. Across 18 long-horizon trajectories, end-to-end comparisons show no consistent gain from Cap evolution. Across 48 continuation branches from shared Hist and Frontier states, accumulated Cap also does not consistently outperform the initial Cap. Parameter tuning of existing factor structures can still improve the portfolio. In an exploratory replay of two screening batches from one Evolving trajectory, some screened-out candidates have positive marginal value at the original state, yet submitting all screened-out candidates sequentially slightly lowers final portfolio IC in both batches. These results show that candidate value depends on the evolving portfolio and submission order, and motivate evaluating self-evolving research capabilities through end-to-end outcomes, conditional capability value, and the consequences of experience-based decisions.
Chinese Translation
自我演化的智能体旨在将研究反馈转化为可重用的技能、工具和研究规则。这些累积的能力是否能够持续改善后续研究需要进行控制评估。长时间跨度的阿尔法发现提供了一个状态依赖的环境：一旦新的因子进入投资组合，已经涵盖的预测信息就会发生变化，因此同一候选者或经验的价值可能会随着时间而变化。我们引入了EverMine，这是一个用于研究长时间跨度阿尔法发现中自我演化研究能力的实证框架。EverMine将研究状态分解为历史（Hist）、当前因子投资组合（Frontier）和可重用能力（Cap）。在匹配的资源限制下，我们比较了固定或演变的Cap的完整运行，并在保持Hist和Frontier不变的情况下替换Cap，以估计累积能力的条件价值。我们还结合完整的轨迹与历史状态重放，以考察基于经验的决策如何影响候选者选择和投资组合结果。在18条长时间跨度的轨迹中，端到端的比较显示Cap演变没有带来一致的收益。在48个来自共享Hist和Frontier状态的延续分支中，累积的Cap也未能始终优于初始Cap。现有因子结构的参数调优仍然可以改善投资组合。在对一个演变轨迹的两个筛选批次进行探索性重放时，一些被筛除的候选者在原始状态下具有正的边际价值，但依次提交所有被筛除的候选者在两个批次中都略微降低了最终投资组合的IC。这些结果表明候选者的价值依赖于演变的投资组合和提交顺序，并激励通过端到端结果、条件能力价值以及基于经验的决策后果来评估自我演化的研究能力。
cs.AI / 226 / 2609.33540
Reasoning on the Simplex: Geometric Fixed-Point Models
在单纯形上的推理：几何固定点模型
Abstract
Looped reasoners spend test-time compute by iterating a weight-tied map, but a small residual does not mean the state is a fixed point when that map lives in unconstrained latent space. We propose Geometric Fixed-Point Reasoning (GFPR), in which the iterated state is the prediction itself: a field of categorical beliefs on a product of simplices, whose argmax is the answer at every step. Because the state is a belief, task structure can be imposed through compact convex relaxations, either as structured readouts or directly in the recurrent state; in the latter case the update remains a continuous self-map, so a fixed point exists for any parameters. At about 7M parameters, GFPR reaches 95.1% exact match on Sudoku-Extreme, 92.0% on Maze-Hard, and 100% sequence accuracy on S_5 length 128, above the published FPRM numbers at the same scale. The same update also trains a 201M language model on FineWeb-Edu in which each site is a distribution over the vocabulary; with 24 Picard steps it is above GPT-2 small on four zero-shot multiple-choice tasks and above GPT-2 medium on ARC-Easy.
Chinese Translation
循环推理器在测试时通过迭代一个权重绑定的映射来消耗计算资源，但当该映射存在于无约束的潜在空间时，小的残差并不意味着状态是一个固定点。我们提出了几何固定点推理（Geometric Fixed-Point Reasoning, GFPR），在该方法中，迭代状态就是预测本身：一个在单纯形乘积上的类别信念场，其最大值（argmax）在每一步都是答案。由于状态是一个信念，任务结构可以通过紧凑的凸松弛施加，无论是作为结构化的读出，还是直接在递归状态中；在后者情况下，更新仍然是一个连续的自映射，因此对于任何参数都存在固定点。在约700万参数的情况下，GFPR在Sudoku-Extreme上达到了95.1%的精确匹配，在Maze-Hard上达到了92.0%，在长度为128的S_5上达到了100%的序列准确率，超出了同规模下已发布的FPRM数据。同样的更新也训练了一个在FineWeb-Edu上的201M语言模型，其中每个站点是一个词汇表的分布；经过24次Picard步骤后，它在四个零-shot多项选择任务上超过了GPT-2小模型，并在ARC-Easy上超过了GPT-2中模型。
cs.AI / 227 / 2609.33543
OSCC: Certified Observation-Safe Coupling Optimization for Gradient-Noise Control in Imperfect-Information Learning
OSCC：用于不完美信息学习中的梯度噪声控制的认证观察安全耦合优化
Abstract
Coupled rollouts can reduce the noise of counterfactual action comparisons, but two issues prevent standard common-random-number constructions from serving as a general learning primitive in imperfect-information environments. First, an invalid coupling may expose hidden state, synchronize endogenous policy randomness, or misalign chance events after counterfactual histories diverge. Second, in multi-action policy optimization, lower return-contrast variance is not by itself the relevant objective: the optimizer depends on the return covariance matrix after projection through the local policy-gradient geometry. We introduce observation-safe counterfactual coupling (OSCC), a framework that defines an admissible class through marginal preservation, information-state safety, branch-local policy randomness, semantic event alignment, and trace-before-oracle replay. We derive a gradient-aware coupling criterion showing that, for marginal-preserving couplings, policy-gradient noise changes are determined by policy-Jacobian-weighted off-diagonal return covariance. This motivates OSCC-Select, a calibration-only selector that chooses among independent, root-only, continuation-only, and fully coupled rollouts using separate safety and gain certificates. Its gain target combines projected gradient noise with measured physical sampling cost and falls back to independent sampling whenever a simultaneous lower confidence bound does not certify improvement. On 100,000 fixed-root Leduc comparisons, the fully coupled CP-GRPO instantiation reduces return-contrast variance from 41.1158 to 18.1441, a 55.87% reduction, while preserving the declared branch marginals. With three actions, OSCC-Select chooses continuation coupling and attains gradient-noise trace 0.0783 versus 0.0917 for return-variance selection. Increasing calibration from 64 to 2,048 groups raises certification from 0.327 to 0.995.
Chinese Translation
耦合回放可以减少反事实动作比较的噪声，但有两个问题阻止标准的共同随机数构造在不完美信息环境中作为通用学习原语。首先，无效的耦合可能会暴露隐藏状态、同步内生政策随机性，或在反事实历史分歧后错位偶然事件。其次，在多动作政策优化中，较低的回报对比方差本身并不是相关目标：优化器依赖于通过局部政策梯度几何投影后的回报协方差矩阵。我们引入了观察安全反事实耦合（OSCC），这是一个通过边际保持、信息状态安全、分支局部政策随机性、语义事件对齐和轨迹前或acles重放定义可接受类的框架。我们推导出一种关注梯度的耦合标准，表明对于边际保持耦合，政策梯度噪声变化由政策雅可比加权的非对角回报协方差决定。这激励了OSCC-Select，这是一种仅进行校准的选择器，它在独立的、仅根节点的、仅继续的和完全耦合的回放中进行选择，使用单独的安全和增益证书。其增益目标结合了投影梯度噪声与测量的物理采样成本，并在同时较低置信区间未能证明改进时回退到独立采样。在100,000个固定根节点的Leduc比较中，完全耦合的CP-GRPO实例将回报对比方差从41.1158减少到18.1441，减少了55.87%，同时保持了声明的分支边际。在三种动作中，OSCC-Select选择继续耦合，并获得梯度噪声轨迹0.0783，而回报方差选择的轨迹为0.0917。将校准从64组增加到2,048组使认证从0.327提高到0.995。
cs.AI / 228 / 2609.33565
Dr. Free: You Don't Need Difficulty Rewards for Self-Evolving Search Agents
Dr. Free：自我进化搜索代理无需难度奖励
Abstract
A central limitation of current data-free self-evolution methods for training search agents is their reliance on difficulty-based proposer rewards. These methods reward a proposer for generating questions that challenge a co-evolving solver, using solver difficulty as a proxy for question quality. Yet difficulty alone is insufficient to distinguish questions that require cross-passage evidence from those that are answerable via simpler shortcuts. In addition, measuring difficulty demands repeated solver rollouts for every candidate question, leading to substantial computational costs. In this paper, we introduce \methodname, the first self-evolving search framework that eliminates difficulty-based proposer rewards and directly optimizes for evidence necessity relative to shortcut contexts. Dr. Free samples relational chains from a knowledge graph and pairs them with aligned passages, giving question generation an explicit multi-hop structure. A generated question receives a positive information-gain reward only when the likelihood of the target answer under the complete evidence passages exceeds the maximum likelihood under all evaluated shortcut contexts. Because this signal is computed from teacher-forced likelihoods, it removes the need for pass-rate estimation and reduces proposer training time by over $7\times$. Experiments on seven open-domain QA benchmarks show that Dr. Free outperforms prior data-free search agents and the supervised baseline, with large improvements on multi-hop QA benchmarks.
Chinese Translation
当前无数据自我进化方法在训练搜索代理时的一个主要限制是其依赖于基于难度的提议者奖励。这些方法奖励提议者生成挑战共同进化求解器的问题，使用求解器的难度作为问题质量的代理。然而，仅凭难度不足以区分需要跨段证据的问题与可以通过更简单的捷径回答的问题。此外，测量难度需要对每个候选问题进行重复的求解器回合，这导致了巨大的计算成本。在本文中，我们介绍了 extit{Dr. Free}，这是第一个消除基于难度的提议者奖励并直接针对证据必要性相对于捷径上下文进行优化的自我进化搜索框架。Dr. Free 从知识图谱中抽样关系链，并将其与对齐的段落配对，为问题生成提供了明确的多跳结构。只有当在完整证据段落下目标答案的可能性超过所有评估的捷径上下文下的最大可能性时，生成的问题才会获得正的信息增益奖励。由于该信号是从教师强制的可能性计算得出的，因此消除了对通过率估计的需求，并将提议者的训练时间减少了超过 $7 imes$。在七个开放域问答基准上的实验表明，Dr. Free 超越了之前的无数据搜索代理和监督基线，在多跳问答基准上取得了显著的改进。
cs.AI / 229 / 2609.33579
OpenFC: Learning Verification Policies towards Open-Search Fact Checking
OpenFC：面向开放搜索事实核查的验证策略学习
Abstract
Open-search fact checking is not merely retrieval followed by classification, but a sequential decision problem in which every query, source visit, and stopping decision reshapes the evidence available for verification. Yet existing systems often distribute these decisions across predefined pipelines or separately prompted modules rather than learning them as a unified task-specific policy. We introduce \textbf{OpenFC}, a unified verification-policy training framework that post-trains Qwen3-8B as a compact next-action controller over reasoning, evidence acquisition, and stopping. OpenFC learns this policy in two stages. \textbf{Stepwise-Calibrated Cold Start (SCCS)} uses a strong training-time supervisor to review post-initial reasoning, tool-use, and stopping proposals before execution, producing reliable trajectories for supervised fine-tuning without access to gold verdicts. \textbf{Verification-Aware Reinforcement Learning (VA-RL)} then improves the cold-start policy on unresolved claims through budget-aware tool rewards, label-aware advantage reweighting, and localized response masking. Across six fact-checking benchmarks, OpenFC achieves 70.39\% average accuracy and 63.30\% macro-F1, the highest overall averages among the evaluated methods. Stage-wise ablations further show that SCCS and VA-RL provide complementary gains, supporting the design of the two-stage training framework. These results position OpenFC as a strong and effective framework for open-search fact-checking. We will open-source our code and release the model checkpoints to support reproducibility.
Chinese Translation
开放搜索事实核查不仅仅是检索后跟随分类，而是一个序列决策问题，其中每个查询、来源访问和停止决策都会重塑可用于验证的证据。然而，现有系统往往将这些决策分散在预定义的流程或单独提示的模块中，而不是将其学习为一个统一的任务特定策略。我们引入了 extbf{OpenFC}，一个统一的验证策略训练框架，它对 Qwen3-8B 进行后训练，使其成为一个紧凑的下一步行动控制器，涵盖推理、证据获取和停止决策。OpenFC 在两个阶段中学习这一策略。 extbf{逐步校准冷启动 (SCCS)} 使用强大的训练时监督者在执行前审查初始推理、工具使用和停止建议，生成可靠的轨迹以进行监督微调，而无需访问金标准裁决。 extbf{验证感知强化学习 (VA-RL)} 随后通过预算感知的工具奖励、标签感知的优势重加权和局部响应掩蔽来改善未解决声明的冷启动策略。在六个事实核查基准测试中，OpenFC 实现了 70.39\% 的平均准确率和 63.30\% 的宏观 F1 分数，在评估的方法中达到了最高的整体平均水平。阶段性消融实验进一步表明，SCCS 和 VA-RL 提供了互补的增益，支持了两阶段训练框架的设计。这些结果使 OpenFC 成为一个强大而有效的开放搜索事实核查框架。我们将开源我们的代码并发布模型检查点以支持可重复性。
cs.AI / 230 / 2609.33601
JustQuant: You Don't Need Smoothing, SVD, or Rotation for 4-Bit Activation Quantization
JustQuant：4位激活量化无需平滑、SVD或旋转
Abstract
Recent generative models have become increasingly powerful, but their inference cost continues to grow. Model quantization offers a promising way to compress these models and accelerate inference. However, at 4 bits, activation quantization is substantially more challenging than weight quantization. Recent post-training quantization (PTQ) and quantization-aware training (QAT) methods have made progress in 4-bit activation quantization by introducing smoothing, SVD branches, rotations, mixed precision, or advanced formats such as NVFP4. These additional operators and data types impose demanding requirements on inference engines and hardware, limiting the broad adoption of low-precision models. Can quantization be achieved using only plain low-bit operators? To answer this question, we propose JustQuant, a simple yet effective framework that moves the complexity of low-bit quantization from deployment-time operators into the training process. We first revisit model quantization from the perspective of knowledge distillation and show that a key reason existing PTQ and QAT methods fail is that they typically exploit supervision at only a single level. We then introduce Theseus QAD, a quantization-aware distillation method that progressively applies multi-level supervision, analogous to the gradual replacement process in the Ship of Theseus. Extensive experiments on DiT and diffusion large language models show two distinct regimes. For smaller models, Theseus QAD can serve as a lightweight warm-up stage that substantially improves subsequent QAT with plain operators, while naive QAD may collapse in the same setting. For larger models, Theseus QAD provides a stronger distillation training path than ordinary QAD. Across both regimes, JustQuant improves low-bit quantization quality while avoiding the complex operators required by many existing PTQ methods.
Chinese Translation
近期生成模型变得越来越强大，但其推理成本也在不断增加。模型量化提供了一种有前景的方法来压缩这些模型并加速推理。然而，在4位量化中，激活量化比权重量化要困难得多。近期的后训练量化（PTQ）和量化感知训练（QAT）方法通过引入平滑、SVD分支、旋转、混合精度或先进格式（如NVFP4）在4位激活量化方面取得了一定进展。这些额外的操作符和数据类型对推理引擎和硬件提出了苛刻的要求，限制了低精度模型的广泛应用。是否可以仅使用普通的低位操作符来实现量化？为了解答这个问题，我们提出了JustQuant，一个简单而有效的框架，将低位量化的复杂性从部署时的操作符转移到训练过程中。我们首先从知识蒸馏的角度重新审视模型量化，并表明现有PTQ和QAT方法失败的一个关键原因是它们通常仅在单一层次上利用监督。接着，我们介绍了Theseus QAD，一种量化感知蒸馏方法，逐步应用多层次监督，类似于忒修斯之船中的逐步替换过程。在DiT和扩散大语言模型上的大量实验显示出两种不同的模式。对于较小的模型，Theseus QAD可以作为一个轻量级的热身阶段，显著改善后续使用普通操作符的QAT，而简单的QAD在相同设置下可能会崩溃。对于较大的模型，Theseus QAD提供了比普通QAD更强的蒸馏训练路径。在这两种模式下，JustQuant提高了低位量化的质量，同时避免了许多现有PTQ方法所需的复杂操作符。
cs.AI / 231 / 2609.33610
Supervision Recovery for Time Series Anomaly Detection via Context-Anchored Pairing
基于上下文锚定配对的时间序列异常检测监督恢复
Abstract
Time series anomaly detection (TSAD) remains challenging not only because anomaly labels are scarce, but also because temporal anomalies are highly context-dependent. Existing methods often rely on unsupervised objectives or surrogate abnormal patterns, providing limited supervision for context-dependent normal--anomalous distinctions. We propose Context-Anchored Pair Supervision (CAPS), a supervision-recovery framework for TSAD. CAPS views ideal anomaly supervision as a matched comparison between normal and anomalous outcomes under the same temporal context, and seeks to recover such supervision without target-domain anomaly labels. Using simulated normal--anomalous pairs, CAPS learns structure and anomaly-semantic representations through reconstruction, background consistency, and within-pair counterfactual recombination. The resulting anomaly representations form a continuous semantic space with coarse modes and induce a sampleable multimodal prior. CAPS conditionally realizes sampled semantics as residual-form effects on target reference trajectories. The resulting context-anchored normal--anomalous counterparts provide temporal supervision for discriminative detector learning. Experiments on nine datasets show that CAPS achieves the strongest aggregate performance across all four evaluation metrics among the compared methods, while complementary ablations and transfer analyses support the roles of context anchoring, semantic disentanglement, and conditional realization.
Chinese Translation
时间序列异常检测（TSAD）仍然面临挑战，不仅因为异常标签稀缺，还因为时间异常高度依赖上下文。现有方法通常依赖于无监督目标或替代异常模式，提供的上下文依赖的正常与异常区分的监督有限。我们提出了上下文锚定配对监督（CAPS），这是一个用于时间序列异常检测的监督恢复框架。CAPS将理想的异常监督视为在相同时间上下文下正常与异常结果之间的匹配比较，并试图在没有目标领域异常标签的情况下恢复这种监督。通过使用模拟的正常与异常配对，CAPS通过重建、背景一致性和配对内的反事实重组学习结构和异常语义表示。最终得到的异常表示形成了一个具有粗糙模式的连续语义空间，并诱导出可采样的多模态先验。CAPS有条件地将采样的语义实现为对目标参考轨迹的残差形式效应。最终得到的上下文锚定的正常与异常对应物为判别检测器学习提供了时间监督。在九个数据集上的实验表明，CAPS在比较方法中在所有四个评估指标上实现了最强的整体性能，而互补的消融实验和迁移分析支持了上下文锚定、语义解缠和条件实现的作用。
cs.AI / 232 / 2609.33614
EAT: Expert Account Tracker for Efficient MoE Inference
EAT：高效MoE推理的专家账户追踪器
Abstract
Mixture-of-Experts (MoE) models have emerged as a revolutionary method to scale Transformer models. However, traditional MoE architecture still suffers from inefficiency since a large number of experts are unnecessarily activated. Existing approaches for reducing the number of activated experts often overlook the historical performance of each expert. In this paper, we propose EAT, a novel method called Expert Account Tracker (EAT), which utilizes history-awareness metrics and adaptive thresholding to dynamically select the most important experts, thereby reducing the activated expert number while effectively maintaining the model performance. Experiments show that EAT outperforms the existing baseline Top-P method across multiple models and datasets, achieving over 25% an average reduction compared to the vanilla method in the number of activated experts and performing better token generation speed compared to the baseline. Furthermore, the performance of pruned models can be efficiently recovered via OPD using only 9K data. Additionally, through ablation studies, we find that excessively reducing the number of activated experts can significantly harm model performance, and the importance of experts varies across layers, with higher-level experts being generally more critical.
Chinese Translation
混合专家（MoE）模型已成为扩展Transformer模型的革命性方法。然而，传统的MoE架构仍然存在效率低下的问题，因为大量专家在不必要的情况下被激活。现有减少激活专家数量的方法往往忽视了每个专家的历史表现。本文提出了一种新方法，称为专家账户追踪器（EAT），该方法利用历史感知指标和自适应阈值动态选择最重要的专家，从而在有效保持模型性能的同时减少激活专家的数量。实验表明，EAT在多个模型和数据集上优于现有基线Top-P方法，与传统方法相比，激活专家数量平均减少超过25%，并且在标记生成速度上表现优于基线。此外，通过OPD，仅使用9K数据即可有效恢复剪枝模型的性能。此外，通过消融研究，我们发现过度减少激活专家的数量会显著损害模型性能，且专家的重要性在不同层次之间有所不同，高层次的专家通常更为关键。
cs.AI / 233 / 2609.33618
ParaAgent: Reinforcing Parallel Acting in Open-World Tool Environments
ParaAgent：在开放世界工具环境中强化并行行动
Abstract
Language model agents are increasingly deployed in open-world tool environments, which require balancing exploring unknown capabilities and exploiting known ones. Existing methods face a performance-efficiency tradeoff: they either rigidly decouple exploration and execution or interleave them without coordination. We argue that the key lies not in whether to decouple or interleave them, but in how to coordinate them across granularities. We introduce ParaAct, a structured parallel-action loop that combines phase-level Exploration $\rightleftharpoons$ Execution with action-level parallelism. To learn this loop, ParaAgent combines multi-agent cold-start demonstrations with reinforcement learning under multi-level advantage decoupling, making planning structure explicit and supervising it with step-, phase-, and trajectory-level rewards. Learning is supported by our ToolEnv, a scalable simulator grounded in 50,011 realistic tool interfaces. On two open-world tool benchmarks, ParaAgent-4B achieves the best average success among all baselines, including GPT-4.1 systems, with the largest gains on multi-tool tasks. Behavioral analyses show that these gains stem from this action organization, highlighting its importance for capable and efficient open-world agents.
Chinese Translation
语言模型代理越来越多地被应用于开放世界工具环境，这要求在探索未知能力和利用已知能力之间取得平衡。现有方法面临性能与效率的权衡：要么严格地将探索与执行解耦，要么在没有协调的情况下交替进行。我们认为关键不在于是否解耦或交替，而在于如何在不同粒度之间进行协调。我们引入了ParaAct，一种结构化的并行行动循环，将阶段级探索与执行结合起来，并实现行动级的并行性。为了学习这一循环，ParaAgent结合了多代理冷启动演示与多级优势解耦下的强化学习，使得规划结构显性化，并通过步骤、阶段和轨迹级奖励进行监督。学习得到了我们的ToolEnv的支持，这是一种基于50,011个真实工具接口的可扩展模拟器。在两个开放世界工具基准测试中，ParaAgent-4B在所有基线中实现了最佳的平均成功率，包括GPT-4.1系统，并在多工具任务上获得了最大的提升。行为分析表明，这些提升源于这种行动组织，突显了其对高效能和高效开放世界代理的重要性。
cs.AI / 234 / 2609.33639
Trajectory Unlearning on LLM-based Agents
基于大语言模型的代理的轨迹遗忘
Abstract
Existing large language model (LLM) unlearning has focused primarily on removing specific knowledge, such as harmful facts, private data, or copyrighted content. However, as LLMs are increasingly deployed as autonomous agents, a fundamental yet overlooked problem emerges: beyond suppressing what an agent knows, an agent should not reproduce undesired behaviors through its action trajectories. In this work, we introduce trajectory-level unlearning, a new problem formulation that targets the removal of specific action trajectories in long-horizon agentic tasks, rather than factual knowledge. We identify two fundamental challenges that distinguish trajectory unlearning from knowledge unlearning: (1) our unlearning target is what the agent \emph{does}, not what it \emph{says}; and (2) trajectories are sequentially dependent action sequences that cannot be decomposed into isolated prompt-response pairs without losing inter-step structure. To address these challenges, we propose Group-injected Relative Policy Optimization (GiRPO), which injects forget trajectories into the policy rollout group with penalized rewards and isolates the normalization statistics, yielding a stable and bounded unlearning signal that does not corrupt gradient updates for normal task trajectories. We construct trajectory unlearning benchmarks from two application scenarios, household tasks (ALFWorld) and online shopping (WebShop), and design three complementary metrics for evaluating forgetting quality and model utility. Experiments on ALFWorld and WebShop demonstrate that GiRPO effectively unlearns target trajectories while preserving task success rates, outperforming existing knowledge-unlearning baselines on both forgetting quality and task utility.
Chinese Translation
现有的大语言模型（LLM）遗忘主要集中在去除特定知识，例如有害事实、私人数据或版权内容。然而，随着LLM越来越多地被部署为自主代理，一个基本但被忽视的问题出现了：除了抑制代理所知道的内容，代理还不应通过其行动轨迹再现不希望的行为。在本研究中，我们引入了轨迹级遗忘，这是一种新的问题表述，旨在去除长时间跨度代理任务中的特定行动轨迹，而不是事实知识。我们识别出两个基本挑战，使轨迹遗忘与知识遗忘有所区别：（1）我们的遗忘目标是代理所“做”的，而不是它所“说”的；（2）轨迹是顺序依赖的行动序列，无法在不失去步骤间结构的情况下分解为孤立的提示-响应对。为了解决这些挑战，我们提出了组注入相对策略优化（Group-injected Relative Policy Optimization, GiRPO），该方法通过惩罚奖励将遗忘轨迹注入策略回滚组，并隔离归一化统计，从而产生稳定且有界的遗忘信号，不会破坏正常任务轨迹的梯度更新。我们从两个应用场景（家庭任务（ALFWorld）和在线购物（WebShop））构建了轨迹遗忘基准，并设计了三种互补指标来评估遗忘质量和模型效用。在ALFWorld和WebShop上的实验表明，GiRPO有效地遗忘目标轨迹，同时保持任务成功率，在遗忘质量和任务效用上超越了现有的知识遗忘基线。
cs.AI / 235 / 2609.33641
Scalable and Data-Driven Decision Support in the Maintenance, Repair, and Overhaul Process
可扩展的数据驱动决策支持在维护、修理和大修过程中的应用
Abstract
Several businesses apply maintenance, repair, and overhaul (MRO) principles to the life-cycle of their existing products. In cases like casted gas turbine component Product Lifecycle Management (PLM), repairing components in frequent intervals can extend the lifetime expectation of the product, provide higher cost efficiency compared to newly produced components, and even improve the part design during the repair cycle. Another aspect of repair concerns sustainability, as products often contain rare materials. The emissions produced by the repair process are usually smaller than mining materials and casting new components. To optimize the repair process further, we propose the Smart Expert System (SES), which assists engineering experts with machine learning-based decision support throughout the repair process. We elaborate on its IT architecture and present machine learning models employed for representative MRO use cases. The SES is evaluated using actual industry data from a leading gas turbine company and demonstrably fulfills formulated requirements concerning the suitability of the overall decision support and the stability of the enclosing IT architecture.
Chinese Translation
许多企业将维护、修理和大修（MRO）原则应用于现有产品的生命周期。在铸造燃气涡轮组件的产品生命周期管理（PLM）等案例中，频繁修理组件可以延长产品的使用寿命，相比新生产的组件提供更高的成本效益，并且在修理周期中甚至可以改善部件设计。修理的另一个方面涉及可持续性，因为产品通常包含稀有材料。修理过程中产生的排放通常小于开采材料和铸造新组件所产生的排放。为了进一步优化修理过程，我们提出了智能专家系统（SES），该系统通过基于机器学习的决策支持，协助工程专家在整个修理过程中。我们详细阐述了其IT架构，并展示了用于代表性MRO用例的机器学习模型。SES使用来自一家领先燃气涡轮公司的实际行业数据进行评估，证明其满足关于整体决策支持的适用性和所包含IT架构的稳定性等制定的要求。
cs.AI / 236 / 2609.33646
Probe to Act: Elevating Browser-Use Agent via Active Visual Probing
探测以行动：通过主动视觉探测提升浏览器使用代理
Abstract
Browser-use agents require seamless alignment between structured web metadata and visual information, while preserving relevant context across long interactions. Existing interfaces often rely on either screenshot-level action prediction or static Set-of-Marks overlays, leaving the model to resolve dense DOM-pixel alignment before every operation. We introduce Probe to Act (P2A), an active probing framework for the browser-agent loop that moves this alignment into decision time. P2A addresses an asymmetric bridge between symbolic DOM hypotheses and screenshot layout by rendering on-demand symbolic DOM structure back into pixels. Before committing a state-changing browser operation, the agent can issue lightweight probes to translate DOM handles into pixel evidence, map screen regions back to DOM candidates, register visual-only targets, and commit verified notes. These interleaved processes naturally produce evidence-based memory: only probed, acted-on, or explicitly committed observations are kept across steps, preserving only decision-critical evidence in long-horizon contexts. P2A can be used as a prompting strategy for proprietary models under the standard DOM+SoM interface, and can be distilled into open-weight models through cold-start synthesis and self-bootstrapped SFT. Across three browser-use benchmarks, P2A shows clear gains on task success rate for both proprietary and fine-tuned models; on VisualWebArena, for example, it improves Gemini-3-Pro from 54.1% to 61.2% and Qwen3-VL-8B from 24.6% to 32.9%, while matching the costly full-observation history ($\sim$3$\times$) at only $\sim$1.2$\times$ the peak retained input context of action-only history.
Chinese Translation
浏览器使用代理需要在结构化的网络元数据和视觉信息之间实现无缝对齐，同时在长时间交互中保持相关上下文。现有接口通常依赖于屏幕截图级别的动作预测或静态标记集（Set-of-Marks）叠加，导致模型在每次操作之前都需解决密集的DOM像素对齐问题。我们提出了探测以行动（Probe to Act, P2A），这是一个用于浏览器代理循环的主动探测框架，将这种对齐移至决策时刻。P2A解决了符号DOM假设与屏幕截图布局之间的不对称桥梁，通过按需将符号DOM结构渲染回像素。在执行改变状态的浏览器操作之前，代理可以发出轻量级探测，将DOM句柄转换为像素证据，将屏幕区域映射回DOM候选项，注册仅视觉目标，并提交经过验证的笔记。这些交错的过程自然产生基于证据的记忆：只有经过探测、已执行或明确提交的观察在各步骤中被保留，从而在长时间上下文中仅保留决策关键证据。P2A可以作为标准DOM+SoM接口下专有模型的提示策略，并可以通过冷启动合成和自我引导的SFT提炼为开放权重模型。在三个浏览器使用基准测试中，P2A在专有模型和微调模型的任务成功率上均显示出明显提升；例如，在VisualWebArena上，它将Gemini-3-Pro的成功率从54.1%提高到61.2%，将Qwen3-VL-8B的成功率从24.6%提高到32.9%，同时仅以约1.2倍的操作历史峰值保留输入上下文匹配成本高昂的完整观察历史（约3倍）。
cs.AI / 237 / 2609.33658
AgentBoundary: Counterfactual Evaluation of Safety in Tool-Using LLM Agents
AgentBoundary：工具使用大型语言模型代理的安全性反事实评估
Abstract
Safety alignment for large language models (LLMs) in conversational settings is largely framed around whether to answer or refuse a request. In agentic settings, however, the same models must decide whether to act as permission-critical evidence emerges during execution. This creates a distinct challenge: apparent risk, action permissibility, and task competence are easily confounded, making agentic over-refusal difficult to distinguish from ordinary task failure. To address this, we introduce AgentBound, the first four-way counterfactual generation-and-evaluation framework for tool-using agent safety. AgentBound transforms the same executable workflow by independently varying apparent risk and action permissibility, enabling controlled comparisons of risky-looking but authorized tasks and routine-looking but unauthorized tasks. These comparisons jointly diagnose over-refusal and unsafe compliance while controlling for task competence. We instantiate AgentBound as a human-validated 4,000-task evaluation suite with trajectory-based and post-state-based judgments. Across 17 model and harness configurations, high safety frequently coexists with poor authorized-task completion: GPT-5.5 blocks 99.5\% of routine-looking unauthorized actions yet completes only 28.7\% of risky-looking authorized tasks. We further train a lightweight runtime calibration module that improves authorized-task completion by 18.2\% on average across 10 evaluated configurations, while improving unsafe-action blocking by 5.4\% on average. These show that effective agentic alignment requires action decisions to track permission-relevant execution evidence, rather than refusal strength alone.
Chinese Translation
在对话环境中，大型语言模型（LLMs）的安全对齐主要围绕是否回答或拒绝请求进行框架化。然而，在代理环境中，这些模型必须决定在执行过程中出现关键证据时是否采取行动。这带来了一个独特的挑战：明显的风险、行动的可许可性和任务的能力容易混淆，使得代理过度拒绝与普通任务失败难以区分。为了解决这个问题，我们引入了AgentBound，这是第一个针对工具使用代理安全性的四向反事实生成与评估框架。AgentBound通过独立变化明显风险和行动可许可性，转化相同的可执行工作流程，从而实现对看似风险较高但经授权的任务与看似常规但未经授权的任务的受控比较。这些比较共同诊断了过度拒绝和不安全合规，同时控制任务能力。我们将AgentBound实例化为一个经过人类验证的4000任务评估套件，包含基于轨迹和后状态的判断。在17种模型和工具配置中，高安全性常常与低授权任务完成率共存：GPT-5.5阻止了99.5%的看似常规的未经授权的行动，但仅完成了28.7%的看似风险较高的经授权任务。我们进一步训练了一个轻量级的运行时校准模块，使得在10个评估配置中，授权任务的完成率平均提高了18.2%，同时不安全行动的阻止率平均提高了5.4%。这些结果表明，有效的代理对齐需要行动决策跟踪与权限相关的执行证据，而不仅仅是拒绝强度。
cs.AI / 238 / 2609.33662
Audit-First VAPO: Risk-Certified Selective Updates under Imperfect Verification
审计优先的 VAPO：在不完美验证下的风险认证选择性更新
Abstract
Imperfect verifiers can assign a harmful update direction even when clipping and regularization bound its magnitude. We introduce Audit-First VAPO, which separates discrete directional admission from continuous magnitude control. An observation-only accept-appeal-abstain policy uses a finite secondary-verification budget; its action trace is frozen before clean labels are joined. Simultaneous finite-sample bounds then certify selected harmful risk, coverage, and verifier-call rate over a predeclared policy family. Conditional Hoeffding-Azuma bounds account for the dependence induced by shared budgets, and rollout or verifier changes initiate a new certification stage. After admission, a bounded trust-clip-KL actuator controls magnitude. We evaluate two models on two reasoning benchmarks against static RLVR, matched-random selection, confidence thresholding, noise correction, and verifier augmentation. On Qwen3.5-0.8B and GSM8K at target risk $\rho=0.08$, RC-VAPO achieves 74.1% accuracy, selected harmful risk 0.0697, coverage 0.4125, and relative verifier cost $1.16\times$. At matched coverage and update magnitude, its selected-risk difference from matched random is -0.0260 with paired 95% interval $[-0.0364,-0.0157]$. Across asymmetric, confidence-dependent, and correlated-verifier noise, the certificate is satisfied on 57 of 60 independent runs. These comparisons isolate informative directional selection from proposal suppression, update shrinkage, and additional verifier computation.
Chinese Translation
不完美的验证者即使在裁剪和正则化限制其幅度的情况下，也可能分配有害的更新方向。我们提出了审计优先的 VAPO，它将离散的方向性接纳与连续的幅度控制分开。仅观察的接受-上诉-弃权策略使用有限的二次验证预算；其行动轨迹在干净标签加入之前被冻结。随后，有限样本界限同时认证所选择的有害风险、覆盖率和验证者调用率，基于预先声明的策略家族。条件 Hoeffding-Azuma 界限考虑了共享预算所引起的依赖性，而回滚或验证者的变化则启动新的认证阶段。在接纳后，受限的信任裁剪-KL 执行器控制幅度。我们在两个推理基准上评估了两个模型，相较于静态 RLVR、匹配随机选择、置信阈值、噪声校正和验证者增强。在 Qwen3.5-0.8B 和 GSM8K 的目标风险 $
ho=0.08$ 下，RC-VAPO 达到 74.1% 的准确率，所选择的有害风险为 0.0697，覆盖率为 0.4125，相对验证者成本为 $1.16 imes$。在匹配的覆盖率和更新幅度下，其所选择风险与匹配随机的差异为 -0.0260，配对的 95% 区间为 $[-0.0364,-0.0157]$。在不对称、依赖置信度和相关验证者噪声的情况下，证书在 60 次独立运行中满足了 57 次。这些比较将信息丰富的方向选择与提案抑制、更新收缩和额外的验证者计算隔离开来。
cs.AI / 239 / 2609.33665
CompoWorld: Compositional Environment Scaling for General Agents
CompoWorld：通用智能体的组合环境扩展
Abstract
Automatically generated environments provide a scalable source of interaction data for training general agents. However, existing approaches mainly generate tasks within a single environment, while real-world workflows require agents to connect information and actions across multiple services. We introduce Compositional Environment Scaling (\textbf{CompoWorld}), which expands the task space by composing a finite library of reusable services. Coding agents turn tool specifications into verified services with typed states and shared interfaces, while a world model handles tools that cannot be reliably implemented. A random-walk procedure connects services through dependency graphs, enabling the generation and verification of tasks that require information to flow across services. Verified trajectories support supervised fine-tuning (SFT), while our Completion-Focused Rubric Reward guides reinforcement learning (RL) toward full task completion by emphasizing criteria with lower pass rates within each rollout group. We construct 448 services exposing 10,130 tools and use 3K SFT trajectories and 1K RL tasks to train Qwen3.6-35B-A3B. Experimental results show that CompoWorld improves on its backbone by 9.17 points on average across eight benchmarks. On AutomationBench, it surpasses frontier models such as Claude Opus 4.6 and leads all compared agent-specialized 35B-A3B models.
Chinese Translation
自动生成的环境为训练通用智能体提供了可扩展的交互数据来源。然而，现有的方法主要在单一环境内生成任务，而现实世界的工作流程要求智能体能够跨多个服务连接信息和行动。我们提出了组合环境扩展（Compositional Environment Scaling， extbf{CompoWorld}），通过组合有限的可重用服务库来扩展任务空间。编码智能体将工具规范转化为具有类型状态和共享接口的经过验证的服务，而世界模型则处理无法可靠实现的工具。随机游走过程通过依赖图连接服务，使得生成和验证需要信息跨服务流动的任务成为可能。经过验证的轨迹支持监督微调（SFT），而我们的以完成为中心的评分奖励（Completion-Focused Rubric Reward）通过强调每个回合组内通过率较低的标准，引导强化学习（RL）朝向完整任务的完成。我们构建了448个服务，暴露出10,130个工具，并使用3K SFT轨迹和1K RL任务来训练Qwen3.6-35B-A3B。实验结果表明，CompoWorld在八个基准测试中平均提高了9.17分。在AutomationBench上，它超越了Claude Opus 4.6等前沿模型，并在所有比较的专门针对智能体的35B-A3B模型中领先。
cs.AI / 240 / 2609.33669
RSD-Poker: Structure-Adaptive and Shift-Robust Risk-Utility Certification for Residual Policies in Imperfect-Information Games
RSD-Poker：用于不完全信息博弈中残余策略的结构自适应和抗偏移风险-效用认证
Abstract
Residual policy adaptation provides a lightweight way to modify a strong reference policy, but a shared scale and a fixed subgroup partition can hide heterogeneous degradation and become fragile when the deployment mixture of information states changes. We introduce RSD-Poker, a structure-adaptive and shift-robust certification framework that freezes a bank of residual families and scales, learns a policy-visible partition on an independent structure split, and freezes that partition before calibration labels are joined. Each candidate-group pair receives a weighted simultaneous upper certificate for anchor-relative risk and a lower certificate for weak-response utility. A robust group-to-candidate map is then selected over a predeclared uncertainty set of deployment group proportions. Under independent calibration units drawn from each frozen group's law, a candidate bank and partition fixed before calibration, and invariant within-group conditionals, the selected map satisfies its declared mixture-robust risk budget and utility certificate with probability at least $1-\zeta_{risk}-\zeta_{util}$. The information contract supports both a teacher-backed transform and a teacher-free observation-only student. The retained deterministic 24-state audit remains an exact replay diagnostic: empirical-zero selects $\alpha=0.08$, raising the weak-response proxy from 4.2082 to 4.2889 with $0/12$ held-out threshold crossings. On stratified held-out states, the learned-partition dual selector raises weak utility from 4.4074 under global dual certification to 4.4936 and lowers held-out violation from 0.0215 to 0.0078; its mixture-robust variant reaches violation 0.0059. Across five observation-only checkpoints, risk-calibrated residuals attain weak utility $4.3659\pm0.0177$ and violation rate $0.0178\pm0.0057$.
Chinese Translation
残余策略适应提供了一种轻量级的方法来修改强参考策略，但共享的尺度和固定的子组划分可能隐藏异质性降级，并在信息状态的部署混合发生变化时变得脆弱。我们引入了RSD-Poker，一种结构自适应和抗偏移的认证框架，该框架冻结了一组残余家族和尺度，在独立结构划分上学习一个策略可见的划分，并在校准标签合并之前冻结该划分。每个候选组对都获得了相对于锚点风险的加权同时上限证书和弱响应效用的下限证书。然后，从预先声明的部署组比例的不确定性集合中选择一个稳健的组到候选映射。在从每个冻结组的分布中抽取的独立校准单元下，候选库和在校准前固定的划分，以及组内条件的不变性，所选映射以至少$1-eta_{risk}-eta_{util}$的概率满足其声明的混合稳健风险预算和效用证书。信息契约支持教师支持的变换和无教师的仅观察学生。保留的确定性24状态审计仍然是一个精确的重放诊断：经验零选择$eta=0.08$，将弱响应代理从4.2082提高到4.2889，且$0/12$的阈值交叉保持不变。在分层保留状态下，学习的划分双选择器将弱效用从全球双重认证下的4.4074提高到4.4936，并将保留违规从0.0215降低到0.0078；其混合稳健变体达到违规0.0059。在五个仅观察的检查点中，风险校准的残余达到弱效用$4.3659 ext{±}0.0177$和违规率$0.0178 ext{±}0.0057$。
cs.AI / 241 / 2609.33676
Auditing Agent Actions through Query-Conditioned Attribution
通过查询条件归因审计代理行为
Abstract
LLM agents increasingly take consequential actions through interactions with users, policies, and external tools. Auditing these agents requires automated attribution of realized actions to their historical basis. However, existing attribution formulations do not provide question-specific traces for diverse auditing objectives. Additionally, when access to the acting model is limited (e.g., in API-only deployments), applicable methods commonly rely on costly input perturbations or external LLM analysis of complete trajectories. We therefore formulate $\textit{query-conditioned agent action attribution}, a new task that takes a natural-language auditing query as input and recovers the source and ordered intermediate evidence for the query-specified aspect of an action. We instantiate this task with $A^3Bench$, a benchmark comprising 1,396 auditing queries across policy basis, parameter provenance, failure propagation, and unsafe-behavior tracing. To enable efficient, query-specific attribution, we use small open-weight models as attribution proposers that combine query-conditioned gradient saliency with query-semantic relevance to rank history units. Our proposer consistently achieves stronger source and evidence rankings at lower inference cost than open-weight baselines, improving source MRR by up to 40.9\% and evidence MAP by 42.1\% with only two forward passes and one backward pass. Controlled evaluations confirm that our proposer improves attribution specificity by adapting its rankings to fine-grained changes in the auditing query. Building on a proposer ensemble, our end-to-end system surpasses the strongest frontier-model baseline in source accuracy (64.5\% vs.\ 60.4\%) while reducing empirical deployment latency by 29.9\% relative to the fastest frontier API baseline. Code and data will be released after the initial review period following final validation and cleanup.
Chinese Translation
大型语言模型（LLM）代理通过与用户、政策和外部工具的互动，越来越多地采取具有重要影响的行动。审计这些代理需要将已实现的行动自动归因于其历史基础。然而，现有的归因公式未能为多样化的审计目标提供特定于问题的追踪。此外，当对执行模型的访问受到限制（例如，在仅限API的部署中）时，适用的方法通常依赖于昂贵的输入扰动或对完整轨迹的外部LLM分析。因此，我们提出了 extit{查询条件代理行为归因}，这是一项新任务，以自然语言审计查询作为输入，恢复与查询指定的行动方面相关的源和有序中间证据。我们通过$A^3Bench$来实例化这一任务，该基准包含1,396个审计查询，涵盖政策基础、参数来源、故障传播和不安全行为追踪。为了实现高效的查询特定归因，我们使用小型开放权重模型作为归因提议者，将查询条件梯度显著性与查询语义相关性相结合，以对历史单元进行排名。我们的提议者在较低的推理成本下，始终实现了比开放权重基线更强的源和证据排名，源的平均排名提升高达40.9\%，证据的平均精度提升42.1\%，仅需两次前向传递和一次反向传递。控制评估确认我们的提议者通过将其排名适应于审计查询的细微变化，提高了归因的特异性。在提议者集成的基础上，我们的端到端系统在源准确性上超越了最强的前沿模型基线（64.5\%对比60.4\%），同时相对于最快的前沿API基线减少了29.9\\%的经验部署延迟。代码和数据将在最终验证和清理后的初步审查期结束后发布。
cs.AI / 242 / 2609.33678
SWE-Game: Can Coding Agents Build the Games We Want?
SWE-Game：编码代理能否构建我们想要的游戏？
Abstract
We introduce SWE-Game, a benchmark of 247 tasks grounded in 41 executable reference Godot games spanning 13 gameplay categories in 2D and 3D. Five task types cover development from a brief, implementation from a game design document, skeleton completion, repair of 83 injected-fault cases, and Godot-to-Unity porting. Reference materials specify the intended gameplay, while a shared instrumentation interface lets evaluator-owned drivers and probes execute actions and observe independently implemented games. Evaluation combines engine-state checks, certified reference-input replay, and agent-authored feature demonstrations to assess mechanic correctness, demonstrated playability, and behavioral restoration and preservation after repairs. Game-specific vision-language rubrics separately assess presentation. Across six models, Opus5 achieves the highest overall score in all five task types. Best overall scores remain below 60 out of 100 across the three construction tasks, with Brief-to-Game reaching 50.38. Analysis of reviewed submissions identifies requirement omissions and gameplay logic errors as predominant implementation problems. On human-labeled behaviors from 100 agent-built games, executable checks achieve 92.59% balanced accuracy, compared with 78.41% for a video-based VLM judge. Rubric-based visual scores reach a Spearman correlation of 0.829 with human ratings of 200 gameplay clips. Together, these results characterize current agent capabilities across game-development activities and support combining runtime evidence with visual assessment.
Chinese Translation
我们介绍了SWE-Game，这是一个基于41个可执行参考Godot游戏的247个任务的基准，这些游戏涵盖了13个2D和3D游戏类别。五种任务类型包括从简要说明开发、从游戏设计文档实施、骨架完成、修复83个注入故障案例，以及Godot到Unity的移植。参考材料指定了预期的游戏玩法，而共享的仪器接口允许评估者拥有的驱动程序和探针执行动作并观察独立实现的游戏。评估结合了引擎状态检查、认证的参考输入重放，以及代理创作的功能演示，以评估机制的正确性、展示的可玩性，以及修复后的行为恢复和保存。特定于游戏的视觉-语言评分标准单独评估表现。在六个模型中，Opus5在所有五种任务类型中获得了最高的总体得分。在三个构建任务中，最佳总体得分仍低于100分中的60分，其中Brief-to-Game达到了50.38。对审核提交的分析发现，需求遗漏和游戏玩法逻辑错误是主要的实施问题。在100个代理构建的游戏中，人类标注的行为的可执行检查达到了92.59%的平衡准确率，而基于视频的VLM评判者为78.41%。基于评分标准的视觉得分与200个游戏片段的人类评分之间的Spearman相关性达到了0.829。这些结果共同表征了当前代理在游戏开发活动中的能力，并支持将运行时证据与视觉评估相结合。
cs.AI / 243 / 2609.33688
TopoMamba: A Load-Support Relation-Guided Multi-Directional State-Space Model for Topology Optimization
TopoMamba：一种基于载荷-支撑关系引导的多方向状态空间模型用于拓扑优化
Abstract
Deep learning has emerged as an efficient alternative for predicting high-performance material distributions in topology optimization. Existing methods struggle to accurately capture load-transfer information, limiting out-of-distribution generalization, while their model architectures often incur high computational costs. To address these challenges, this paper proposes TopoMamba, a topology prediction framework incorporating a load-support relation-guided multi-directional state-space model. Coupling physical fields with load-support relations enables more effective modeling of mechanical dependencies. A load-support relation-guided spatially adaptive fusion mechanism dynamically adjusts multi-directional scan features according to spatial conditions. Mamba is coupled with the solid isotropic material with penalty method to enhance structural mechanical performance while maintaining computational efficiency. Results on two-dimensional topology optimization benchmarks demonstrate that TopoMamba achieves superior topology prediction accuracy, out-of-distribution generalization, and computational efficiency over state-of-the-art models. The proposed load-support physics-guided framework enables efficient optimization of more complex structural systems.
Chinese Translation
深度学习已成为预测拓扑优化中高性能材料分布的有效替代方案。现有方法在准确捕捉载荷传递信息方面存在困难，限制了其在分布外的泛化能力，同时其模型架构往往导致高计算成本。为了解决这些挑战，本文提出了TopoMamba，一种结合载荷-支撑关系引导的多方向状态空间模型的拓扑预测框架。将物理场与载荷-支撑关系耦合，使机械依赖关系的建模更加有效。载荷-支撑关系引导的空间自适应融合机制根据空间条件动态调整多方向扫描特征。Mamba与固体各向同性材料惩罚法相结合，以提高结构的机械性能，同时保持计算效率。在二维拓扑优化基准测试中的结果表明，TopoMamba在拓扑预测准确性、分布外泛化能力和计算效率方面优于最先进的模型。所提出的载荷-支撑物理引导框架使得对更复杂结构系统的高效优化成为可能。
cs.AI / 244 / 2609.33698
One Latent, Many Tokens: Jointly Learning Compressed Embeddings for Efficient Language Diffusion
一个潜在变量，多种标记：联合学习压缩嵌入以实现高效语言扩散
Abstract
Most continuous diffusion language models process one latent position per token at each sampling step, making generation expensive. Two-stage methods lower the cost by reducing the latent length, but they fix the compressed embedding space before training the diffusion model. Embeddings from the fixed space can be difficult to model with diffusion and decode reliably into tokens, which limits generation quality after compression. To address this problem, we introduce JPEG-DLM (Joint-embedding Prediction for Efficient Generation with Diffusion Language Model), which jointly trains a compressor, a flow matching model and a decoding module. With joint-embedding prediction, JPEG-DLM learns compressed embeddings that are more structured, easier to model with diffusion and reliably decodable into tokens. JPEG-DLM achieves the lowest mean Gen-PPL and highest throughput among recent diffusion and flow models on LM1B and OWT. At a compression rate of 0.5 on OWT, it reaches a Gen-PPL of 34.52 and approximately 2.3 times ELF's throughput. These results suggest that jointly learning compressed embeddings offers a promising path toward efficient diffusion language modeling. Code will be released soon.
Chinese Translation
大多数连续扩散语言模型在每个采样步骤中为每个标记处理一个潜在位置，这使得生成过程成本高昂。两阶段方法通过减少潜在长度来降低成本，但在训练扩散模型之前固定了压缩嵌入空间。来自固定空间的嵌入在扩散建模时可能难以处理，并且难以可靠地解码为标记，这限制了压缩后的生成质量。为了解决这个问题，我们提出了JPEG-DLM（联合嵌入预测以实现高效生成的扩散语言模型），该模型联合训练压缩器、流匹配模型和解码模块。通过联合嵌入预测，JPEG-DLM学习到的压缩嵌入结构更加清晰，更容易与扩散模型进行建模，并且能够可靠地解码为标记。在LM1B和OWT数据集上，JPEG-DLM在最近的扩散和流模型中实现了最低的平均生成困惑度（Gen-PPL）和最高的吞吐量。在OWT上以0.5的压缩率，JPEG-DLM达到了34.52的Gen-PPL，并且吞吐量约为ELF的2.3倍。这些结果表明，联合学习压缩嵌入为高效的扩散语言建模提供了一条有前景的路径。代码将很快发布。
cs.AI / 245 / 2609.33699
SpecRead: A Benchmark for Measuring Whether Language Models Understand Hardware Specifications
SpecRead：衡量语言模型是否理解硬件规范的基准测试
Abstract
Existing benchmarks for large language models (LLMs) in hardware design evaluate downstream artifacts such as generated RTL, assertions, or testbenches. When a model fails such a benchmark, the failure is ambiguous: it may have misread the specification, or it may have understood the specification and failed to write the code. We present SpecRead, a benchmark that isolates specification comprehension from generation ability. SpecRead v2.1 contains 385 questions over 10 open-source OpenTitan IP blocks: exact retrieval, cross-section reasoning, contradiction detection in mutated specifications, and spec-RTL consistency checking, plus 82 controls (41 distractor, 41 consistent-RTL). Type-4 items are built from real RTL mutations; we retain only mutations that Icarus Verilog simulation shows to change observable behavior. A with-spec vs. without-spec ablation suggests the questions require the excerpt, not training recall alone (without-spec accuracy 3/20 on the t1/t2 subset), though memorization of the source text may still help spot mutations. As an initial characterization with a small model, Ministral-3B scores 33.2% overall (128/385; macro average 39.0%): 55.2% on retrieval, 51.7% on cross-section reasoning. On the two contradiction-focused types, the verdict-plus-location measure gives 48.0% (t3) and 63.3% (t4), with a 51.2% false-positive rate on distractors and 100% on consistent-RTL controls. Layered scoring shows the model locates contradictions well (78.9-81.6% location accuracy) but scores lower on their category (43.9-49.7%). A structured "rule-table" prompting intervention lowers accuracy on every question type except t2 (tied). SpecRead is automatically scorable by deterministic checks, with gray-zone cases counted wrong under the conservative main scoring. The benchmark is regenerable for type-3 items via mutation injection, and built exclusively from public sources.
Chinese Translation
现有的大型语言模型（LLMs）在硬件设计中的基准测试评估下游产物，如生成的RTL、断言或测试平台。当模型未能通过此类基准测试时，失败的原因并不明确：它可能误读了规范，或者它可能理解了规范但未能编写代码。我们提出了SpecRead，一个将规范理解与生成能力分离的基准测试。SpecRead v2.1包含针对10个开源OpenTitan IP模块的385个问题：精确检索、交叉推理、变异规范中的矛盾检测，以及规范与RTL的一致性检查，此外还有82个控制项（41个干扰项，41个一致性RTL）。类型4项目由真实的RTL变异构建；我们仅保留Icarus Verilog模拟显示会改变可观察行为的变异。带规范与不带规范的消融实验表明，这些问题需要摘录，而不仅仅是训练回忆（在t1/t2子集上不带规范的准确率为3/20），尽管对源文本的记忆仍可能有助于发现变异。作为对小模型的初步表征，Ministral-3B的整体得分为33.2%（128/385；宏平均39.0%）：检索得分为55.2%，交叉推理得分为51.7%。在两个以矛盾为重点的类型中，判决加位置的测量结果为48.0%（t3）和63.3%（t4），在干扰项上的假阳性率为51.2%，在一致性RTL控制项上为100%。分层评分显示模型在定位矛盾方面表现良好（位置准确率为78.9-81.6%），但在其类别上的得分较低（43.9-49.7%）。结构化的“规则表”提示干预降低了每种问题类型的准确率，除了t2（持平）。SpecRead可以通过确定性检查自动评分，灰色区域的案例在保守的主要评分下被视为错误。该基准测试可以通过变异注入为类型3项目再生，并且完全基于公共来源构建。
cs.AI / 246 / 2609.33707
Does Adversarial Training Improve Generalization in Multi-View VLAs? Revealing and Mitigating View Collapse
对抗训练是否改善多视角视觉-语言-动作（VLA）的泛化能力？揭示与缓解视角崩溃
Abstract
Vision-language-action (VLA) models adapt pretrained vision-language models (VLMs) for closed-loop robot control, transferring their perceptual and semantic capabilities to action prediction. Despite strong in-distribution performance, however, VLAs often degrade under deployment shifts. Adversarial training (AT) offers a model-adaptive approach to robustness without explicitly anticipating individual shifts, but its effect on natural distribution-shift generalization in multi-view VLAs remains unclear. We study this question using a multi-view VLA directly adapted from a pretrained VLM and evaluate generalization across seven LIBERO-Plus shift axes. Direct AT substantially improves Camera Viewpoint and Sensor Noise, the two shifts affecting only the third-person view, yet produces mixed or negative effects on other shifts. Controlled view interventions reveal a surprising failure mode that we term view collapse: Direct AT can shift cross-view reliance so strongly that the policy becomes dominated by the wrist view. This exposes a \textit{robustness shortcut}: apparent robustness to a shifted view can arise from reduced use of that view rather than more robust perception of it. This motivates a distinction between robust perception, extracting reliable information under within-view shifts, and robust fusion, adapting reliance across views according to their reliability. To reduce fixed view reliance, we use a simple View Swap intervention and then re-evaluate AT. With View Swap, AT further improves Camera Viewpoint, Sensor Noise, and Robot Initial State, while its effects remain mixed on other shifts. Our results show that multi-view robustness requires separating improved perception from changes in cross-view reliance, and that AT provides selective rather than generic distribution-shift benefits.
Chinese Translation
视觉-语言-动作（VLA）模型将预训练的视觉-语言模型（VLM）适应于闭环机器人控制，将其感知和语义能力转移到动作预测上。然而，尽管在分布内表现强劲，VLA在部署转变下往往会退化。对抗训练（AT）提供了一种模型自适应的鲁棒性方法，而无需明确预见个体转变，但其对多视角VLA在自然分布转变下的泛化影响仍不清楚。我们使用直接从预训练VLM适应的多视角VLA研究这一问题，并在七个LIBERO-Plus转变轴上评估泛化能力。直接的AT显著改善了仅影响第三人称视角的相机视角和传感器噪声这两种转变，但对其他转变产生了混合或负面影响。控制视角干预揭示了一种令人惊讶的失败模式，我们称之为视角崩溃：直接的AT可能会如此强烈地改变跨视角的依赖关系，以至于策略被手腕视角主导。这暴露出一种 extit{鲁棒性捷径}：对转变视角的表面鲁棒性可能源于对该视角的使用减少，而不是对其更鲁棒的感知。这促使我们区分鲁棒感知，即在视角内转变下提取可靠信息，以及鲁棒融合，即根据视角的可靠性调整跨视角的依赖关系。为了减少固定视角的依赖，我们使用简单的视角交换干预，然后重新评估AT。使用视角交换后，AT进一步改善了相机视角、传感器噪声和机器人初始状态，而对其他转变的影响仍然是混合的。我们的结果表明，多视角鲁棒性需要将改善的感知与跨视角依赖的变化分开，并且AT提供的是选择性的而非通用的分布转变益处。
cs.AI / 247 / 2609.33713
BIRD: Distilling Decision Boundaries into Rationales for MLLM Adaptation
BIRD：将决策边界提炼为多模态大语言模型适应的理由
Abstract
Adapting general-purpose multimodal large language models (MLLMs) to specialized domains requires learning domain-specific decision criteria, which often hinge on subtle visual distinctions between otherwise plausible answers. Rationale augmentation aims to expose such evidence through additional observations or inter-sample comparisons, yet a visually valid cue is not necessarily decision-relevant: it may describe how samples differ without changing the model's relative preference between competing answers. We therefore introduce BIRD, a self-improving Boundary-Informed Rationale Distillation framework that uses model-specific confusions to locate unresolved local decision boundaries and distills the evidence that resolves these confusions into rationales. For each sample, BIRD retrieves candidate neighbors from the target MLLM's own representation space and selects the most confusable one according to its answer preferences. It then generates answer-blind candidate evidence from their visual differences and functionally verifies which evidence most effectively strengthens the model's preference for the correct answer while avoiding inappropriate transfer across the pair. The verified evidence is then distilled into a single-sample rationale for standard supervised fine-tuning. Experiments on medical and chart VQA show that BIRD outperforms competing rationale-augmentation methods across two target MLLMs, while further analyses demonstrate clearer separation of confusable answers and stronger gains from model-matched supervision.
Chinese Translation
将通用多模态大语言模型（MLLMs）适应于专业领域需要学习领域特定的决策标准，这通常依赖于在其他可行答案之间的微妙视觉差异。理由增强旨在通过额外的观察或样本间比较来揭示这种证据，然而，视觉上有效的线索不一定与决策相关：它可能描述样本之间的差异，而不改变模型对竞争答案的相对偏好。因此，我们提出了BIRD，一个自我改进的边界信息理由提炼框架，它利用模型特定的混淆来定位未解决的局部决策边界，并将解决这些混淆的证据提炼为理由。对于每个样本，BIRD从目标MLLM自身的表示空间中检索候选邻居，并根据其答案偏好选择最具混淆性的一个。然后，它根据这些邻居的视觉差异生成无答案盲候选证据，并功能性地验证哪些证据最有效地增强模型对正确答案的偏好，同时避免在这对样本之间的不当迁移。经过验证的证据随后被提炼为单样本理由，用于标准的监督微调。在医学和图表VQA的实验中，BIRD在两个目标MLLM上超越了竞争的理由增强方法，同时进一步分析显示出混淆答案的更清晰分离和来自模型匹配监督的更强增益。
cs.AI / 248 / 2609.33717
Self-Designed Evaluators and Warm Memory for Long-Horizon Agents
自设计评估器和长时间跨度智能体的温暖记忆
Abstract
A tool-using language-model agent deployed over a long stream of tasks receives no reward, so it cannot tell whether it succeeded, cannot safely retry, and cannot label the experience it needs to improve. We present SelfSuite, in which the agent's own base model, given only the world's public materials, designs a small evaluation suite of weighted judges and grounded per-task briefs, freezes it, and uses it to gate a keep-best retry and to label a typed, outcome-tracked memory. On matched five-repeat benchmarks over tau2-bench and AppWorld, SelfSuite scores above the plain agent without any labels, matches methods given ten expert labels on tau2-bench, and trails Agentic Context Engineering (ACE) on AppWorld, where code execution gives a direct success signal. In an ablation campaign run on the same tasks, it is above label-free ACE in every repeat, and the gated second attempt is the only component whose removal hurts in every repeat. We also simulate a subject-matter expert who grades ten onboarding tasks per world. Using those labels to calibrate SelfSuite's evaluator gives a small, consistent gain, and using them to warm up ACE's memory lifts ACE to tie calibrated SelfSuite. A single-run study on a second model family shows the same ordering.
Chinese Translation
在长时间任务流中部署的工具使用语言模型智能体无法获得奖励，因此无法判断其是否成功，无法安全地重试，也无法标记其需要改进的经验。我们提出了SelfSuite，其中智能体的基础模型仅基于世界的公共材料，设计出一个小型评估套件，包括加权评审者和基于任务的简要说明，并将其冻结，利用该套件进行最佳保留重试，并标记一个类型化的、结果跟踪的记忆。在tau2-bench和AppWorld的五次重复基准测试中，SelfSuite的得分高于没有任何标签的普通智能体，匹配在tau2-bench上获得十个专家标签的方法，并在AppWorld上落后于Agentic Context Engineering (ACE)，在该环境中代码执行提供了直接的成功信号。在对相同任务进行的消融实验中，SelfSuite在每次重复中均高于无标签的ACE，而被限制的第二次尝试是唯一一个在每次重复中去除后会造成损害的组件。我们还模拟了一位主题专家，他对每个世界的十个入职任务进行评分。利用这些标签来校准SelfSuite的评估器可以获得小幅且一致的提升，而利用它们来预热ACE的记忆则使ACE的表现提升至与校准后的SelfSuite持平。对第二个模型系列的单次运行研究显示出相同的排序。
cs.AI / 249 / 2609.33726
Robust Biomolecular Complex Design Across Protein Conformational Landscapes
跨蛋白构象景观的稳健生物分子复合物设计
Abstract
Proteins populate conformational ensembles, yet structure-based biomolecular design typically optimizes candidates against a single target conformation. Consequently, a candidate that fits one state can lose favorable interactions or develop steric clashes when the target adopts another. We introduce FlexEvo, a model-agnostic evolutionary framework that adapts candidates once at inference time from a single target conformation to improve compatibility with alternative natural conformations unseen during adaptation, without retraining the source model or requiring a conformational ensemble. FlexEvo casts cross-state adaptation as geometry-constrained bi-objective optimization, balancing preservation of input-state interactions against robustness to plausible conformational perturbations. To limit the search space and reduce invalid structural edits, geometry-derived FlexBoxes define protected anchor regions, adaptable regions for local exploration, and forbidden regions for clash avoidance. A unified all-atom representation supports topology-preserving adaptation across diverse binder categories, while Pareto selection preserves nondominated candidates across the two objectives. We evaluate FlexEvo across multiple generation baselines and nine representative binder categories spanning diverse molecular sizes and structural topologies. FlexEvo reduces the category-balanced mean relative performance degradation from 47.8% to 4.4%, while adding only 1.4--3.1 minutes of adaptation per sample. These results establish single-state inference-time adaptation as a practical route toward robust biomolecular complex design across protein conformational landscapes.
Chinese Translation
蛋白质存在于构象集合中，但基于结构的生物分子设计通常仅针对单一目标构象优化候选者。因此，适合某一状态的候选者在目标采用另一状态时可能失去有利的相互作用或产生空间冲突。我们提出了FlexEvo，这是一种模型无关的进化框架，能够在推理时从单一目标构象适应候选者，以提高与在适应过程中未见的替代自然构象的兼容性，而无需重新训练源模型或要求构象集合。FlexEvo将跨状态适应视为几何约束的双目标优化，平衡输入状态相互作用的保留与对合理构象扰动的稳健性。为了限制搜索空间并减少无效的结构编辑，几何衍生的FlexBoxes定义了受保护的锚定区域、用于局部探索的可适应区域和用于避免冲突的禁区。统一的全原子表示支持在多种结合剂类别中进行拓扑保持的适应，而帕累托选择则在两个目标中保留非支配候选者。我们在多个生成基线和九个代表性的结合剂类别上评估了FlexEvo，这些类别涵盖了不同的分子大小和结构拓扑。FlexEvo将类别平衡的平均相对性能下降从47.8%降低到4.4%，同时每个样本仅增加1.4至3.1分钟的适应时间。这些结果确立了单一状态推理时适应作为在蛋白构象景观中实现稳健生物分子复合物设计的实用途径。
cs.AI / 250 / 2609.33731
HTN Planning as a Coordination Layer for Multi-Server MCP Tool Orchestration
将层次任务网络规划作为多服务器MCP工具编排的协调层
Abstract
The Model Context Protocol (MCP) isolates servers by design: only the host can orchestrate cross-server workflows. When the host is a large language model, the resulting orchestrations are non-deterministic, non-reproducible, and pay one inference round-trip per tool call. We present a coordination architecture in which a Hierarchical Task Network (HTN) planner generates a verifiable cross-server plan once, and a runtime middleware executes it deterministically across multiple MCP servers, binding cross-action data dependencies via a template mechanism (\verb|${context.X}|) substituted at execution time. The architecture mirrors MCP's isolation constraint: each compound task decomposes into server-local primitive actions, and inter-server data flow is bound at execution time via JSON-path output extractors. We instantiate the architecture on five HTN domains spanning laboratory robotics, bioinformatics and multiscale modelling, and demonstrate end-to-end execution from a browser-based plan controller against eight live third-party MCP servers querying real biological databases.
Chinese Translation
模型上下文协议（MCP）在设计上隔离了服务器：只有主机能够编排跨服务器的工作流。当主机是大型语言模型时，所产生的编排是非确定性的、不可重现的，并且每次工具调用需支付一次推理往返费用。我们提出了一种协调架构，其中层次任务网络（HTN）规划器一次生成可验证的跨服务器计划，运行时中间件在多个MCP服务器上以确定性方式执行该计划，通过在执行时替换的模板机制（ exttt{${context.X}}）绑定跨操作数据依赖。该架构反映了MCP的隔离约束：每个复合任务分解为服务器本地的原始操作，跨服务器的数据流在执行时通过JSON路径输出提取器进行绑定。我们在五个HTN领域中实例化了该架构，涵盖实验室机器人技术、生物信息学和多尺度建模，并展示了从基于浏览器的计划控制器到八个实时第三方MCP服务器查询真实生物数据库的端到端执行。
cs.AI / 251 / 2609.33772
Skill2Env: Capability-Oriented Environment Synthesis from Skills for General Agents
Skill2Env：面向能力的环境合成框架，基于技能为通用智能体提供支持
Abstract
Executable environments are critical for post-training agents on tasks that require tool use and multi-step interaction, but constructing executable tasks together with their environments remains difficult to scale. Skills provide reusable domain knowledge, operational procedures, and tool-use instructions, but a substantial gap remains between the information contained in a skill and a concrete, challenging task with a complete executable environment. To address this gap, we introduce Skill2Env, a capability-oriented framework that starts from a skill and uses agent capability demands to guide task and environment synthesis. Skill2Env represents these demands through reusable difficulty patterns and instantiates them into task blueprints that specify objectives, challenges, environment facts, information boundaries, and acceptance criteria. These blueprints guide the joint construction of task instructions, execution substrates, workspaces, and rubric-based evaluators around source skills. We further propose Iterative Task Hardening, which uses solver execution evidence to identify insufficiently challenging task designs, strengthen or extend their difficulty-pattern instantiations, and revise the corresponding blueprints and environments. Using 1.5K high-scoring trajectories generated from Skill2Env environments for supervised fine-tuning, we observe consistent improvements across a broad range of agent benchmarks, demonstrating the effectiveness of capability-oriented environment synthesis for agent post-training.
Chinese Translation
可执行环境对于在需要工具使用和多步骤交互的任务上进行后训练的智能体至关重要，但构建可执行任务及其环境仍然难以扩展。技能提供了可重用的领域知识、操作程序和工具使用说明，但技能中包含的信息与具有完整可执行环境的具体、具有挑战性的任务之间仍存在显著差距。为了解决这一差距，我们提出了Skill2Env，一个以能力为导向的框架，从技能出发，利用智能体的能力需求来指导任务和环境的合成。Skill2Env通过可重用的难度模式来表示这些需求，并将其实例化为任务蓝图，蓝图中指定了目标、挑战、环境事实、信息边界和接受标准。这些蓝图指导围绕源技能共同构建任务指令、执行基础、工作空间和基于评分标准的评估器。我们进一步提出了迭代任务强化（Iterative Task Hardening），该方法利用求解器执行证据来识别不足够具有挑战性的任务设计，加强或扩展其难度模式的实例化，并修订相应的蓝图和环境。通过使用从Skill2Env环境生成的1500条高评分轨迹进行监督微调，我们观察到在广泛的智能体基准测试中一致性地提高，证明了面向能力的环境合成在智能体后训练中的有效性。
cs.AI / 252 / 2609.33773
Learning Strategies to Break Judges
打破评判者的学习策略
Abstract
As AI agents surpass human performance, it becomes exceedingly hard for system designers to evaluate them directly and understand their failure modes. Consequently, agents themselves are being deployed extensively to evaluate, judge, and provide feedback on model traces. But this raises an important question: how can we trust the judge? In this work, we propose an agent-guided method to find weaknesses of agentic judges that expose interpretable failure mechanisms. Our method focuses on mathematical reasoning and proceeds in two stages: first, we deploy adversarial agents to mutate a set of sound proofs by introducing errors, attempting to misguide judges---in other words, injecting errors that judges are unable to catch. Then, we distill these attempts into a small set of mutation strategies which allow us to analyze the failure modes of the judges. To ensure that these strategies are not overfit to the initial set of proofs, we evaluate them by applying the mutation strategies to a held-out set of proofs and querying the same judge. We deploy our method on GPT-5.6-sol and Claude Opus 5, paired with their agent orchestrators, Codex and Claude Code, respectively. These are used both as mutators to introduce errors and as judges to evaluate correctness of mathematical reasoning. We find that across all the agentic judges, we are able to distill mutation strategies that consistently bypass their evaluations, thereby enabling us to ascertain actionable failure modes. Our analysis also reveals that judge reliability degrades at the frontier: errors in Olympiad-level proofs or graduate-level mathematical texts are detected more consistently, whereas flaws in research-level manuscripts are more likely to escape detection.
Chinese Translation
随着人工智能代理超越人类表现，系统设计者直接评估它们并理解其失败模式变得异常困难。因此，代理本身被广泛部署以评估、判断并提供模型轨迹的反馈。但这提出了一个重要问题：我们如何信任评判者？在本研究中，我们提出了一种代理引导的方法，以发现代理评判者的弱点，从而揭示可解释的失败机制。我们的方法侧重于数学推理，并分为两个阶段：首先，我们部署对抗性代理，通过引入错误来变异一组有效的证明，试图误导评判者——换句话说，注入评判者无法捕捉到的错误。然后，我们将这些尝试提炼为一小组变异策略，使我们能够分析评判者的失败模式。为了确保这些策略不会过拟合于初始的证明集，我们通过将变异策略应用于一个保留的证明集并查询相同的评判者来评估它们。我们在 GPT-5.6-sol 和 Claude Opus 5 上部署了我们的方法，并与它们的代理协调者 Codex 和 Claude Code 配对。这些被用作变异器以引入错误，同时也作为评判者来评估数学推理的正确性。我们发现，在所有代理评判者中，我们能够提炼出持续绕过其评估的变异策略，从而使我们能够确定可操作的失败模式。我们的分析还揭示，评判者的可靠性在边界处下降：在奥林匹克级别的证明或研究生级别的数学文本中，错误更容易被检测到，而研究级别的手稿中的缺陷更可能逃避检测。
cs.AI / 253 / 2609.33778
Evidence-Inference Reconstruction: When The Evidence Is Recalled But The Reasoning Goes Wrong
证据-推理重构：当证据被回忆但推理出错时
Abstract
Modern multi-hop LLM agents are equipped with built-in mechanisms to detect errors in intermediate reasoning steps. Such errors trigger corrective actions from these agents, which mostly follow the paradigm of retrying the steps or the reasoning trajectories. Not only are these retries expensive, we present in this paper that they are also potentially unnecessary. To this end, we introduce Evidence-Inference Reconstruction (EIR), which uses structured state to guide one retrieval trajectory, accumulating source evidence in the process. We show that as long as the relevant evidence has been collected, EIR is capable of generating the correct answer in a single final model call even if erroneous evidence has been mixed in due to incorrect intermediate reasoning steps. In one evaluation, using Haiku 4.5 and GPT-4.1 Mini, we evaluate EIR on matched 1,000-question subsets of HotpotQA, 2WikiMultiHopQA, and MuSiQue, showing that EIR improves Answer F1, the overlap between the model's and the correct answer, over the baseline by 8.3--32.8 points, Agentic SSR by 10.6--29.1 points, and Reflexion by 1.1--15.9 points. Additionally, we show that EIR averages 4.85 total model calls per question, compared with 35.29 for Agentic SSR and 12.41 for Reflexion. Together, these results corroborate EIR's central premise: separating evidence retrieval from the final answer model call can improve answer accuracy while utilizing substantially less computation.
Chinese Translation
现代多跳大型语言模型（LLM）代理配备了内置机制，以检测中间推理步骤中的错误。这些错误会触发代理的纠正措施，这些措施主要遵循重试步骤或推理轨迹的范式。这些重试不仅成本高昂，我们在本文中指出，它们也可能是多余的。为此，我们引入了证据-推理重构（EIR），该方法利用结构化状态引导一次检索轨迹，并在此过程中积累源证据。我们展示了只要相关证据已被收集，EIR 就能够在最终模型调用中生成正确答案，即使由于中间推理步骤错误而混入了错误证据。在一次评估中，我们使用 Haiku 4.5 和 GPT-4.1 Mini，对 HotpotQA、2WikiMultiHopQA 和 MuSiQue 的匹配 1,000 问题子集进行了 EIR 的评估，结果显示 EIR 在答案 F1（模型答案与正确答案之间的重叠）上比基线提高了 8.3--32.8 分，在 Agentic SSR 上提高了 10.6--29.1 分，在 Reflexion 上提高了 1.1--15.9 分。此外，我们还展示了 EIR 每个问题的平均模型调用次数为 4.85，而 Agentic SSR 为 35.29，Reflexion 为 12.41。这些结果共同证实了 EIR 的核心前提：将证据检索与最终答案模型调用分离可以提高答案的准确性，同时显著减少计算量。
cs.AI / 254 / 2609.33786
Is your uncertainty map wrong, or is its target? Exact diagnostics for the Tweedie diagonal, and a gradient-free alternative
你的不确定性图是错误的，还是目标错误？Tweedie对角线的精确诊断及无梯度替代方案
Abstract
A diffusion model can predict a follow-up medical scan from a baseline, but a clinician needs a per-voxel map of where that prediction can be trusted. Many such maps approximate the diagonal of the Tweedie posterior covariance, and are evaluated against another approximation of it, so whether the estimator or the target limits them is unclear. We compute the exact diagonal on six checkpoints across fourteen model-corpus conditions. Hutchinson at M=200 tracks it at rank agreement of at least 0.92 everywhere, yet in four of the fourteen the exact diagonal is anti-correlated with the denoising error, reaching -0.13, so a faithful estimator reproduces that reversal. All four are real-image conditions; on the models' own samples the reversal does not appear, so evaluating on generated samples flatters this family. What limits these maps is the target, not the estimator. We then introduce Tweedie Probe-Tangent (T-PT), a gradient-free residual probe that corrupts one model-supported prediction repeatedly and measures the voxel-wise variance of the denoiser's response. T-PT reads a different functional of the same Jacobian, and its exact second-order form ranks with the diagonal wherever the diagonal reverses; at thirty probes it returns a map too unstable to reproduce that ranking, while Hutchinson at M=5 already reproduces it, so T-PT there is not evidence against the reversal. We offer it as an instrument, not a better approximation. On brain MRI at full resolution, where every Jacobian-based estimator we test runs out of memory, T-PT leads a twenty-chain Monte-Carlo ensemble on five of eight endpoints inside tissue and trails it on none, at 16x fewer network evaluations; over the whole volume the ensemble leads, and fifty chains close the tissue gap. On lung CT the ensemble is ahead throughout. Both lose most of their discrimination where the change is, which remains open.
Chinese Translation
扩散模型可以根据基线预测后续的医学扫描，但临床医生需要一个每个体素的地图，以确定该预测的可信度。许多此类地图近似于Tweedie后验协方差的对角线，并与其另一个近似进行评估，因此不清楚是估计器还是目标限制了它们。我们在十四个模型-语料条件下的六个检查点计算了精确的对角线。Hutchinson在M=200时在各处的秩一致性至少为0.92，然而在十四个条件中的四个中，精确对角线与去噪误差呈反相关，达到-0.13，因此一个忠实的估计器再现了这种反转。所有四个条件都是实际图像条件；在模型自己的样本上没有出现反转，因此在生成样本上的评估使这一家族的表现看起来更好。这些地图的限制在于目标，而不是估计器。然后我们引入了Tweedie探针-切线（T-PT），这是一种无梯度残差探针，反复破坏一个模型支持的预测，并测量去噪器响应的体素级方差。T-PT读取相同雅可比矩阵的不同函数形式，其精确的二阶形式在对角线反转的地方与对角线排名一致；在三十个探针下，它返回的地图不够稳定，无法再现该排名，而Hutchinson在M=5时已经再现了，因此在这里T-PT并不是反转的证据。我们将其作为一种工具，而不是更好的近似。在全分辨率的脑MRI中，我们测试的每个基于雅可比的估计器都耗尽了内存，T-PT在八个端点中的五个组织内引领了一个二十链的蒙特卡洛集成，并且在任何端点上都没有落后，网络评估次数减少了16倍；在整个体积上，集成领先，五十链缩小了组织间隙。在肺CT中，集成始终处于领先地位。两者在变化发生的地方失去了大部分区分度，这一问题仍然开放。
cs.AI / 255 / 2609.33816
Dual-Vocabulary Language Model for Cross-Tokenizer Distillation
用于跨标记器蒸馏的双词汇语言模型
Abstract
On-policy distillation (OPD) bridges teacher supervision and student behavior, but different teacher-student tokenizers introduce misalignment in both input tokenization (#1) and output logits (#2). Existing approaches address the former by matching same-text spans or converting tokens to bytes, often losing fine-grained token information or disrupting the native-token paradigm, while for the latter, strategies such as ranking, padding, or key-token selection retain only shared logit dimensions, resulting in much distribution loss. In this paper, we propose Dual-Vocabulary Language Model (DVLM), which replaces the teacher's LM head with a new student-vocabulary projection head and obtains full-dimensional student logits (for #2). To support student tokens (for #1), it takes a Parallel-Tokenized Sequence (PTS) as input, which concatenates the original teacher-tokenized sequence and a re-tokenized sequence formed by independently converting each student token into a teacher-token group. To avoid inference inconsistency with the original teacher tokens, the Hybrid-Prefix Attention (HPA) further restricts re-tokenized groups to their corresponding teacher prefix and uses its last state as the aggregation of the original student-token representation for projection into the student vocabulary space. Similarly, via the combined use of PTS and HPA, the DVLM teacher can provide distribution-aligned supervision with the student's input-tokenization and output-logit during OPD. Experimental results demonstrate that our DVLM teacher has a similar converged loss as the original teacher model and enables student models to improve performance across six reasoning tasks.
Chinese Translation
在策略蒸馏（On-policy distillation, OPD）中，教师监督与学生行为之间架起了桥梁，但不同的教师-学生标记器在输入标记化（#1）和输出对数（#2）上引入了不一致性。现有方法通过匹配相同文本的跨度或将标记转换为字节来解决前者，通常会丢失细粒度的标记信息或破坏原生标记范式，而对于后者，诸如排序、填充或关键标记选择等策略仅保留共享的对数维度，导致大量分布损失。在本文中，我们提出了双词汇语言模型（Dual-Vocabulary Language Model, DVLM），该模型用新的学生词汇投影头替换教师的语言模型头，并获得全维度的学生对数（针对#2）。为了支持学生标记（针对#1），它将并行标记序列（Parallel-Tokenized Sequence, PTS）作为输入，该序列连接了原始教师标记序列和通过独立将每个学生标记转换为教师标记组形成的重新标记序列。为了避免与原始教师标记的不一致推理，混合前缀注意力（Hybrid-Prefix Attention, HPA）进一步将重新标记组限制在其对应的教师前缀，并使用其最后状态作为原始学生标记表示的聚合，以投影到学生词汇空间。类似地，通过结合使用PTS和HPA，DVLM教师可以在OPD期间为学生的输入标记化和输出对数提供分布对齐的监督。实验结果表明，我们的DVLM教师具有与原始教师模型相似的收敛损失，并使学生模型在六个推理任务中提高了性能。
cs.AI / 256 / 2609.33822
Vestrum: Improving Agent Harnesses by Adapting Their Verification, Structure and Memory
Vestrum：通过适应验证、结构和记忆来改进代理工具
Abstract
An agent harness controls how a language model accesses information, uses tools, preserves memory, and checks its work. Improving this software is costly when each evaluation requires a long interaction with an environment. We introduce Vestrum, a framework that turns failures in execution traces into scoped harness changes without training the task model. Its organizing overhypothesis is that tasks of a shared kind may exhibit recurring failures whose remedies transfer within that kind. Vestrum expresses failures as recognizable classes, proposes changes across verification, retrieval, decomposition, and knowledge synthesis, and screens their scope before evaluating them as a bundle. A persistent lessons file informs subsequent proposals. Across five settings and two baseline harnesses, the frozen harnesses improve held-out performance: UltraHorizon rises from 47.6 to 59.8 over GAM, Terminal-Bench 4 Hard from 63.7% to 70.3% of checks passed over Claude Code on eight held-out tasks at 1.03x test cost, and cell-type annotation agreement from 67.5% to 77.8% on held-out sections of one slide, alongside gains on LoCoMo and AMA-Bench. Across our searches, verification grounded in evidence helped both intermediate steps and final answers, at lower cost at intermediate steps, while critics asked to rebuild finished answers broke more than they repaired. On the three memory benchmarks, Vestrum also scores above the evaluated GEPA configurations in every paired evaluation.
Chinese Translation
代理工具控制语言模型如何访问信息、使用工具、保留记忆和检查其工作。当每次评估都需要与环境进行长时间交互时，改进这一软件的成本是相当高的。我们引入了Vestrum，一个框架，它将执行轨迹中的失败转化为有针对性的工具变更，而无需训练任务模型。其组织性过度假设是，共享类型的任务可能会表现出重复的失败，其补救措施可以在该类型内转移。Vestrum将失败表达为可识别的类别，提出在验证、检索、分解和知识合成方面的变更，并在将其作为一个整体评估之前筛选其范围。一个持久的教训文件为后续提案提供信息。在五个设置和两个基线工具中，冻结的工具提高了保留性能：UltraHorizon在GAM上从47.6提升至59.8，Terminal-Bench 4 Hard在Claude Code上从63.7%提升至70.3%的检查通过率，在八个保留任务中以1.03倍的测试成本，以及在一张幻灯片的保留部分上细胞类型注释一致性从67.5%提升至77.8%，同时在LoCoMo和AMA-Bench上也取得了进展。在我们的搜索中，基于证据的验证帮助了中间步骤和最终答案，在中间步骤的成本更低，而被要求重建完成答案的评论者破坏的比修复的更多。在三个记忆基准测试中，Vestrum在每次配对评估中也都超过了评估的GEPA配置。
cs.AI / 257 / 2609.33843
Laya as a Typed Probabilistic Assessor: An Independent Reproduction and a Preregistered Study of Calibration and Selective Escalation
Laya作为一种类型化的概率评估器：独立重现与预注册的校准与选择性升级研究
Abstract
The shipped Laya Typed-Decisions checkpoint, a 421M-parameter ModernBERT-large assessor that answers typed choice/noul/score questions over workflow state, is uniformly under-confident. The signed confidence-accuracy gap is $-0.214$, every occupied reliability bin's accuracy exceeds its confidence, and that sign uniformity collapses every binned ECE variant to the same value, $0.214$. The card frames the risk as over-confidence; the measured direction is the opposite, and the direction decides which way a confidence-gated cascade fails. A single disjointly fitted temperature ($T=0.469$, sharpening) removes most of the miscalibration (held-out ECE $0.204$ to $0.037$) and outperforms the shipped per-option-count table. The frozen selection rule instead chose isotonic regression, which overfit and failed its held-out NLL contrast on both tracks, so hypothesis H2 is not supported. Re-running the released checkpoint on its full official test split reproduces the card's headline accuracy ($0.767$ vs. $0.766$). The retrospective E1 reproduction preceded the analysis freeze; E2-E8 were prospectively preregistered, and 20 of 22 executed confirmatory tests reject under Benjamini-Hochberg FDR at $q=0.05$ (two descoped). The frozen gate beats random escalation but misses its 10% accepted-set error target on both tracks, an exploratory out-of-distribution probe finds no zero-shot transfer (accuracy $0.617$), and every score measures agreement with a synthetic teacher whose self-agreement ceiling ($0.735$) the specialist exceeds. Per-decision predictions, run manifests, and the frozen preregistration are in the ancillary files. The author has no affiliation with the model's publisher, the dataset's publisher, or TypeSafe.
Chinese Translation
已发布的Laya类型决策检查点是一个421M参数的ModernBERT-large评估器，能够回答关于工作流状态的类型选择/无/分数问题，其整体表现均显得信心不足。签名的置信度-准确度差为$-0.214$，每个占用的可靠性区间的准确度均超过其置信度，而这种符号的一致性使得每个分箱的ECE变体都归结为相同的值$0.214$。该卡片将风险框定为过度自信；而测量的方向则相反，这一方向决定了置信度门控级联失败的方式。一个单独的温度（$T=0.469$，锐化）去除了大部分的错误校准（持出ECE从$0.204$降低到$0.037$），并且优于已发布的每选项计数表。冻结的选择规则则选择了等调回归，但由于过拟合而在两个轨道上未能通过其持出NLL对比，因此假设H2未得到支持。在其完整的官方测试分割上重新运行发布的检查点重现了卡片的头条准确度（$0.767$对比$0.766$）。回顾性的E1重现发生在分析冻结之前；E2-E8是前瞻性预注册的，22个执行的确认测试中有20个在Benjamini-Hochberg FDR下以$q=0.05$拒绝了（两个被去除）。冻结的门控超越了随机升级，但在两个轨道上未能达到其10%的接受集错误目标，一项探索性的分布外探测未发现零样本迁移（准确度$0.617$），而每个分数则测量与一个合成教师的协议，该教师的自我协议上限（$0.735$）被专家超越。每个决策的预测、运行清单和冻结的预注册均在附属文件中。作者与模型的发布者、数据集的发布者或TypeSafe没有任何关联。
cs.AI / 258 / 2609.33845
How code helps different tasks? A decompositional lens on LLM post-training
代码如何帮助不同任务？对LLM后训练的分解视角
Abstract
Evaluating code data as a single corpus can obscure which types of code data benefit which models and downstream tasks. Effective data selection requires understanding both the benefits of individual categories and whether these benefits persist when categories are combined. We introduce a decompositional lens for studying these effects in LLM post-training. We first decompose an execution-verified code corpus into interpretable categories based on the computational patterns of its solutions. Through controlled fine-tuning experiments, we compare individual categories with a balanced mixture across instruction-tuned models on question answering, mathematics, and code generation. The resulting response maps reveal recurring gains in average question-answering performance, while the same category can improve one model or task and degrade another. The best-performing category also varies with the starting model and target task. We then compose compact mixtures guided by these results and examine whether benefits observed in individual categories persist under joint training. On selected model--task pairs, mixtures whose constituents each improve the target task outperform both their best constituent and full-corpus training while using roughly 10--15\% of the full corpus. These exploratory findings illustrate a \emph{less is more} pattern and highlight how the value of code data in post training depends on which categories are combined for which model and task.
Chinese Translation
将代码数据作为单一语料库进行评估可能会掩盖哪些类型的代码数据对哪些模型和下游任务有益。有效的数据选择需要理解各个类别的好处，以及这些好处在类别组合时是否依然存在。我们引入了一种分解视角来研究这些效应在LLM后训练中的表现。我们首先根据解决方案的计算模式将执行验证的代码语料库分解为可解释的类别。通过控制的微调实验，我们比较了在问题回答、数学和代码生成任务中，单个类别与平衡混合的表现。结果响应图揭示了平均问题回答性能的反复提升，而同一类别可能会改善一个模型或任务，却使另一个模型或任务的性能下降。表现最佳的类别也会随着起始模型和目标任务的不同而变化。然后，我们根据这些结果组合紧凑的混合，并检查在联合训练下，观察到的个别类别的好处是否依然存在。在选定的模型-任务对中，构成部分均能改善目标任务的混合，表现优于其最佳构成部分和全语料库训练，同时仅使用约10-15%的全语料库。这些探索性发现展示了一种“少即是多”的模式，并强调了后训练中代码数据的价值如何依赖于为特定模型和任务组合哪些类别。
cs.AI / 259 / 2609.33867
R$^2$ Flow: Recursive Self-Improvement via Recursive Skill Evolution
R$^2$ 流：通过递归技能演化实现递归自我改进
Abstract
LLM-based agents can improve themselves across tasks by reusing and revising the skills they orchestrate into executable procedures. Flow-based training fits this loop: it samples procedures in proportion to reward, and the flow through each skill credits it for the next library revision. Three obstacles stand in the way of making this self-improvement reliable: flow training suffers strategy collapse over tree-structured histories; nonnegative flow-based credit rewards frequent use as if it were benefit; and library edits rest on the task reward the policy optimizes. We introduce R$^2$ Flow, a recursive self-improvement framework that alternates policy learning, independent verification, and versioned skill-library updates on a shared-state orchestration graph. The graph merges histories that differ only in the order of independent steps, allowing flow training to pool evidence across equivalent executions. A flow-share readout of the trained flow, invariant to the backward policy, and a separate signed utility rank which skills to change, verifier evidence decides whether an edit is warranted, and a residual-variance plateau sets when to update. Committed edits reshape the graph the next policy learns on, realizing recursive skill evolution. Across question answering, mathematical reasoning, interactive decision making, and code generation, R$^2$ Flow improves task accuracy and library-edit precision over heuristic orchestration, reinforcement learning, and skill-evolution baselines, and transfers across executors. Code is available at https://github.com/beita6969/r2flow.
Chinese Translation
基于大型语言模型（LLM）的智能体可以通过重用和修订它们编排成可执行程序的技能，在不同任务中自我改进。基于流的训练适合这一循环：它根据奖励的比例抽样程序，并且每个技能的流量为下一个库的修订提供了信用。实现这种自我改进的过程中面临三大障碍：流训练在树状历史中容易出现策略崩溃；基于流的非负信用奖励频繁使用，仿佛这是收益；而库的编辑依赖于策略优化的任务奖励。我们提出了 R$^2$ 流，这是一种递归自我改进框架，交替进行策略学习、独立验证和版本化技能库更新，基于共享状态的编排图。该图合并仅在独立步骤顺序上不同的历史，使得流训练能够在等效执行中汇聚证据。经过训练的流的流量共享读出与反向策略无关，另一个单独的有符号效用排名决定哪些技能需要更改，验证者证据决定是否需要编辑，而残差方差平坦区则设定何时更新。已承诺的编辑重塑下一个策略学习的图，从而实现递归技能演化。在问答、数学推理、互动决策和代码生成等任务中，R$^2$ 流在任务准确性和库编辑精度上优于启发式编排、强化学习和技能演化基线，并且能够在执行者之间转移。代码可在 https://github.com/beita6969/r2flow 获取。
cs.AI / 260 / 2609.33870
When Successful Strategies Fail: Adaptation to Environmental Novelty in Terminal Agents
成功策略为何失败：终端代理对环境新奇性的适应
Abstract
LLM agents increasingly solve long-horizon tasks by autonomously interacting with their environment. In doing so, their strategies rely on assumptions about that environment: which resources and tools exist, where they are located, and how they behave. When these assumptions no longer hold, reliable agents must detect the change and adapt while pursuing the same goal. We study this adaptation capability through environmental novelty: a change that keeps the task objective fixed while invalidating an assumption underlying an otherwise successful trajectory. We introduce AGNI, an automated pipeline that extracts trajectory-relevant assumptions, injects targeted environmental changes, and validates that the resulting novel tasks remain solvable. Across three terminal benchmarks, AGNI produces diverse novelties spanning resources, interfaces, constraints, and execution semantics. Evaluating multiple LLM agents reveals a substantial adaptation gap between base and novel tasks. Trajectory analysis suggests that agents often encounter evidence of the change but fail to diagnose its cause and revise their strategy. Finally, post-training for environmental novelty improves adaptation to held-out novel tasks while also improving performance on base tasks. Our results highlight a gap between task competence and adaptive capability and motivate environmental variation as a core dimension of agent training and evaluation.
Chinese Translation
大型语言模型（LLM）代理越来越多地通过自主与环境互动来解决长期任务。在此过程中，它们的策略依赖于对环境的假设：即存在哪些资源和工具，它们的位置以及它们的行为。当这些假设不再成立时，可靠的代理必须检测到变化并在追求相同目标的同时进行适应。我们通过环境新奇性研究这种适应能力：一种变化，它保持任务目标不变，同时使得支撑原本成功轨迹的假设失效。我们引入了AGNI，一个自动化管道，用于提取与轨迹相关的假设，注入针对性的环境变化，并验证生成的新任务是否仍然可解。在三个终端基准测试中，AGNI产生了涵盖资源、接口、约束和执行语义的多样化新奇性。对多个LLM代理的评估揭示了基础任务与新任务之间存在显著的适应差距。轨迹分析表明，代理通常会遇到变化的证据，但未能诊断其原因并修正策略。最后，针对环境新奇性的后训练提高了对保留的新任务的适应能力，同时也改善了基础任务的表现。我们的结果突显了任务能力与适应能力之间的差距，并推动环境变异作为代理训练和评估的核心维度。
cs.AI / 261 / 2609.33878
Curating Merchant-Matching Training Data with Two Confidence-Gated Local LLM Judges
使用两个信心门控的本地大型语言模型评审者策划商户匹配训练数据
Abstract
Merchant matching resolves a noisy payment descriptor to a retrieved merchant entity or returns no match. A key challenge in curating training labels is distinguishing teacher abstention from evidence that no acceptable entity exists: false no-match labels contaminate pseudo-labeled data, while conservative labeling reduces coverage. We investigate whether agreement between two local large language model judges improves pseudo-label reliability. A label is retained only when the judges agree, with separate ordered thresholds for selections and abstentions that guarantee disjoint positive and negative label sets. Retrospective replay on 2,000 expert-annotated queries shows that higher selection thresholds can improve positive-label purity, whereas higher abstention thresholds increase false no-match labels. At thresholds (0.86, 0.80), Muse Glimmer 30B and Gemma 4 31B jointly label 1,633 queries (81.7% coverage) at 96.88% purity; positive and negative purities are 99.47% and 93.38%. This exceeds either constituent model at the same thresholds by more than two percentage points, with lower coverage. A split-half check finds only 0.14 percentage points of threshold-selection optimism. A symmetric threshold of 0.86 adds 40 erroneous no-match labels, while 46 false abstentions persist even with no confidence threshold. Across five matched within-model comparisons, higher reasoning effort yields no clear F0.5 gain and increases median latency by 1.8-5.0 times. These results motivate separate thresholding and auditing for positive and negative pseudo-labels. The study establishes label purity, not student utility; fresh-data curation and student fine-tuning remain necessary to demonstrate downstream value.
Chinese Translation
商户匹配将嘈杂的支付描述符解析为检索到的商户实体，或返回无匹配项。策划训练标签的一个关键挑战是区分教师的弃权与没有可接受实体的证据：错误的无匹配标签污染了伪标记数据，而保守的标记则降低了覆盖率。我们研究了两个本地大型语言模型评审者之间的协议是否能提高伪标签的可靠性。仅当评审者达成一致时，标签才会被保留，并为选择和弃权设定了单独的有序阈值，以确保正负标签集的互斥性。对2000个专家注释查询的回顾性重放显示，较高的选择阈值可以提高正标签的纯度，而较高的弃权阈值则增加了错误的无匹配标签。在阈值(0.86, 0.80)下，Muse Glimmer 30B和Gemma 4 31B共同标记了1633个查询（覆盖率81.7%），纯度为96.88%；正负标签的纯度分别为99.47%和93.38%。这超过了在相同阈值下的任一组成模型，且覆盖率更低。分半检查发现阈值选择的乐观性仅为0.14个百分点。对称阈值0.86增加了40个错误的无匹配标签，而即使没有信心阈值，仍然存在46个错误的弃权。在五个匹配的模型内部比较中，较高的推理努力未能带来明显的F0.5增益，并将中位延迟提高了1.8-5.0倍。这些结果促使对正负伪标签进行单独的阈值设定和审计。本研究确立了标签的纯度，而非学生的效用；新数据的策划和学生的微调仍然是展示下游价值所必需的。
cs.AI / 262 / 2609.33910
When Consent Outlives Context: Residual Authority Replay in Long-Lived Agents
当同意超越上下文：长期存在代理中的残余授权重放
Abstract
LLM agents increasingly rely on user approval to authorize security-sensitive actions at runtime. Such approvals are granted within a specific task and execution context. In long-lived agents, authorization decisions may need to persist across tasks or sessions. We find that this continuity can outlive the context that originally justified the approval, creating residual authority reusable without renewed consent. We expose this failure mode through a longitudinal attack that starts from a target security-sensitive action, identifies the authority required to execute it, induces benign interactions that legitimately obtain that authority, and later replays the residual authority during adversarial execution. Across controlled and live settings, we demonstrate that residual-authority replay arises in practice and substantially increases the success of prompt-injection and context-rebinding attacks. We evaluate 508 AgentDojo attack cases across six LLM families using production-derived authorization semantics. With residual authority, attack success rate (ASR) increases by up to 35.1 percentage points compared with a fresh authorization state. In live context-rebinding attacks on 55 Terminal-Bench cases across three real-world production coding agents, residual-authority replay increases ASR by 24.9 percentage points on average. These findings expose a fundamental mismatch between persistent authorization and the contextual nature of user consent in long-lived LLM agents.
Chinese Translation
大型语言模型（LLM）代理越来越依赖用户批准来授权运行时的安全敏感操作。这些批准是在特定任务和执行上下文中授予的。在长期存在的代理中，授权决策可能需要在任务或会话之间持续存在。我们发现，这种连续性可能超越最初证明批准合理性的上下文，从而创建可在没有重新同意的情况下重复使用的残余授权。我们通过一种纵向攻击暴露了这一失败模式，该攻击从目标安全敏感操作开始，识别执行该操作所需的授权，诱导合法的良性交互以获得该授权，并在对抗性执行过程中重放残余授权。在受控和实时环境中，我们证明了残余授权重放在实践中确实存在，并显著提高了提示注入和上下文重绑定攻击的成功率。我们评估了508个AgentDojo攻击案例，涵盖六个LLM家族，使用生产派生的授权语义。与新授权状态相比，残余授权使攻击成功率（ASR）提高了最多35.1个百分点。在对三个真实生产编码代理的55个Terminal-Bench案例进行的实时上下文重绑定攻击中，残余授权重放平均使ASR提高了24.9个百分点。这些发现揭示了长期存在的LLM代理中持久授权与用户同意的上下文性质之间的根本不匹配。
cs.AI / 263 / 2609.33920
HyperMCTS: Hypergraph-Augmented MCTS for Long-Horizon LLM Agents
HyperMCTS：用于长时间跨度大语言模型代理的超图增强蒙特卡洛树搜索
Abstract
Long-horizon tasks require large language model (LLM) agents to coordinate decisions under constraints that span an entire solution. Monte Carlo Tree Search (MCTS) offers a promising approach to test-time scaling by exploring alternative action trajectories, but model computation and environment interaction make search costly. Efficient search therefore requires effective reuse of trajectory feedback. Standard MCTS maintains prefix-specific statistics, without explicitly accumulating outcomes for decision groups that recur across different paths. To fill this gap, we propose HyperMCTS, a training-free method that augments an ordered MCTS tree with a cross-trajectory hypergraph. Hyperedges represent groups of canonical decisions and accumulate their observed returns within the current task. Our hypergraph-guided HyperUCT selection rule aggregates evidence from overlapping hyperedges into an action prior, allowing outcomes collected under one prefix to inform selection under another while preserving execution histories in the tree. On DeepPlanning, HyperMCTS improves average planning accuracy by 2.3--7.3 percentage points over the strongest baseline for each of three backbone models. It enables Qwen3.6-27B to outperform Claude Opus 4.6 (max) on Shopping Planning, while achieving higher accuracy with fewer LLM calls and output tokens than the evaluated MCTS-based baselines. SealQA experiments further demonstrate improvements in question answering.
Chinese Translation
长时间跨度的任务要求大语言模型（LLM）代理在跨越整个解决方案的约束下协调决策。蒙特卡洛树搜索（MCTS）通过探索替代行动轨迹，为测试时的扩展提供了一种有前景的方法，但模型计算和环境交互使得搜索成本高昂。因此，高效的搜索需要有效地重用轨迹反馈。标准的MCTS维护特定前缀的统计信息，而没有明确地为在不同路径中重复出现的决策组累积结果。为填补这一空白，我们提出了HyperMCTS，这是一种无训练的方法，它通过交叉轨迹超图增强有序的MCTS树。超边表示典型决策的组，并在当前任务中累积其观察到的收益。我们的超图引导的HyperUCT选择规则将来自重叠超边的证据聚合成行动先验，允许在一个前缀下收集的结果为在另一个前缀下的选择提供信息，同时保留树中的执行历史。在DeepPlanning上，HyperMCTS在三个基础模型中，相较于最强基线提高了2.3到7.3个百分点的平均规划准确性。它使Qwen3.6-27B在购物规划上超越Claude Opus 4.6（最大），同时在LLM调用和输出标记数量更少的情况下实现更高的准确性。SealQA实验进一步证明了在问答方面的改进。
cs.AI / 264 / 2609.33955
Designing Reliable LLM-as-a-Judge Measurement Systems for Multi-Turn Business Agents
为多轮业务代理设计可靠的 LLM 作为评判者的测量系统
Abstract
Many LLM-as-a-judge evaluations score fixed outputs under a fixed task definition. Production multi-turn business agents instead require a maintained measurement system: correctness depends on business-specific facts and procedures, outcomes emerge across turns, and failures must be attributed to either agent capability or missing business knowledge before they are actionable. We present an integrated methodology spanning evaluation specification, modular LLM judges, intent-preserving user simulation, and human-in-the-loop governance. The specification defines conversation-level end states and actionable failure ownership. Atomic judges share versioned evidence and feed an explicit aggregation graph. The simulator is released only after task-preservation and stability checks. Independent human audits estimate measurement fidelity, renew tiered reference sets, and route disagreements to label correction, guideline revision, or judge improvement. Production studies show that system-level fidelity improved across repeated audits, that human reviewers and automated judges improved together under the shared feedback loop, and that their combined workflow had the strongest descriptive performance in both reported task-completion settings. Because the studies are observational and the human reference itself required revision, these findings demonstrate operational usefulness rather than causal or universal superiority. The contribution is a practical framework for making multi-turn agent measurement reliable, actionable, and maintainable as the evaluated system and its evidence evolve.
Chinese Translation
许多 LLM 作为评判者的评估在固定任务定义下对固定输出进行评分。然而，生产中的多轮业务代理需要一个持续维护的测量系统：正确性依赖于特定业务的事实和程序，结果在多轮对话中逐渐显现，且在采取行动之前，必须将失败归因于代理能力或缺失的业务知识。我们提出了一种综合方法，涵盖评估规范、模块化 LLM 评判者、意图保持的用户模拟以及人机协作治理。该规范定义了对话级别的最终状态和可操作的失败归属。原子评判者共享版本化证据，并构建一个明确的聚合图。模拟器仅在任务保持和稳定性检查通过后发布。独立的人类审计评估测量的准确性，更新分层参考集，并将争议引导至标签修正、指南修订或评判者改进。生产研究表明，系统级的准确性在重复审计中得到了改善，人类审阅者和自动评判者在共享反馈循环下共同提升，而它们的结合工作流程在报告的任务完成设置中表现出最强的描述性能。由于这些研究是观察性的，并且人类参考本身需要修订，因此这些发现展示了操作的实用性，而非因果或普遍的优越性。该贡献是一个实用框架，旨在使多轮代理的测量在评估系统及其证据演变过程中保持可靠、可操作和可维护。
cs.AI / 265 / 2609.34007
EHRAdapt: Adapting Pretrained Language Models to Electronic Health Records with Semantic Priors for Rare Clinical Events
EHRAdapt：利用语义先验将预训练语言模型适应于电子健康记录中的稀有临床事件
Abstract
Electronic health records (EHRs) encode clinical histories as (time, modality, code) tuples, whereas pretrained language models expect text tokens. Serializing them as text inflates sequence length and redundantly encodes structure. We introduce EHRAdapt, an adapter that maps tuples directly into a frozen language model's embedding space. Modality receives a learned embedding, time gaps enter through learned attention biases, and event codes receive dedicated vectors. Learning event vectors is the central challenge: clinical vocabularies are long-tailed, leaving rare events too few observations for reliable estimates. EHRAdapt therefore represents each event vector as the sum of a semantic prior and an evidence residual. The prior is a frozen embedding of the event's clinical description from a biomedical language model trained on clinical ontologies, mapped into the model's input space by a shared learned projection, so it supplies clinical meaning even when observations are scarce. The residual, a learned low-rank event-specific correction, refines it as evidence accumulates. We run continued pretraining on about 4 million patients' records with three frozen LLM backbones (OLMo2 1B, Llama3.2 1B, and OLMo2 7B), training only the adapter (0.1--0.6% of all parameters). The full adapter outperforms all ablations in held-out next-event prediction on every backbone. Removing the semantic pathway hurts rare events over ten times more than the most frequent ones, whereas removing the residual hurts overall prediction but improves it for the rarest events. On reportable infectious-disease and syndromic downstream classification tasks, EHRAdapt outperforms text-based LLM and count-based baselines, and both pathways improve rare-disease discrimination. The two pathways therefore play complementary roles, visible only when results are broken down by event frequency rather than averaged.
Chinese Translation
电子健康记录（EHR）将临床历史编码为（时间，模态，代码）元组，而预训练语言模型则期望文本标记。将其序列化为文本会增加序列长度并冗余编码结构。我们提出了EHRAdapt，一个将元组直接映射到冻结语言模型嵌入空间的适配器。模态接收学习到的嵌入，时间间隔通过学习到的注意力偏置进入，事件代码接收专用向量。学习事件向量是中心挑战：临床词汇是长尾的，稀有事件的观察次数太少，无法进行可靠估计。因此，EHRAdapt将每个事件向量表示为语义先验和证据残差的总和。先验是来自于在临床本体上训练的生物医学语言模型的事件临床描述的冻结嵌入，通过共享学习投影映射到模型的输入空间，从而在观察稀缺时提供临床意义。残差是学习到的低秩事件特定修正，随着证据的积累对其进行细化。我们在约400万患者的记录上进行持续的预训练，使用三个冻结的LLM骨干（OLMo2 1B，Llama3.2 1B和OLMo2 7B），仅训练适配器（占所有参数的0.1%--0.6%）。完整的适配器在每个骨干的保留下一个事件预测中优于所有消融实验。移除语义路径对稀有事件的影响超过最频繁事件的十倍，而移除残差则对整体预测产生负面影响，但对最稀有事件的预测有所改善。在可报告的传染病和综合征下游分类任务中，EHRAdapt优于基于文本的LLM和基于计数的基线，并且两条路径都改善了稀有疾病的区分。因此，这两条路径发挥了互补作用，只有在按事件频率细分结果时才能显现出来，而不是平均结果。
cs.AI / 266 / 2609.34015
A Computer Vision Approach to Visual Fraud Detection in Phishing Websites Using YOLOv8
基于计算机视觉的网络钓鱼网站视觉欺诈检测方法：使用YOLOv8
Abstract
Phishing remains one of the most common vectors for financial and identity fraud, and most detection systems still rely on inspecting a page's URL, HTML markup, or domain registration history. These signals are easy for an attacker to rotate or obfuscate, and they say very little about what actually convinces a victim to hand over a password or a card number: the way the page looks. This paper describes a visual, image-based approach to phishing detection that treats a rendered webpage the same way a human eye would, as a picture that either matches a trusted brand or doesn't. A YOLOv8 convolutional neural network was trained to classify full-page website screenshots as phishing or legitimate based on layout, logo placement, color scheme, and login-form structure, rather than on text extracted from the page. The system reached 92% classification accuracy on a held-out test set, processed a single screenshot in roughly 100 milliseconds, and, after a round of data augmentation aimed specifically at lighting, compression, and scaling variation, cut the false-positive rate by 11% relative to the pre-augmentation baseline. The paper walks through the dataset construction, the augmentation strategy, the model architecture and training setup, and the resulting performance, and closes with a discussion of where this kind of visual detector fits alongside, rather than instead of, existing URL- and content-based defenses.
Chinese Translation
网络钓鱼仍然是金融和身份欺诈最常见的途径之一，大多数检测系统仍依赖于检查页面的URL、HTML标记或域名注册历史。这些信号很容易被攻击者更改或混淆，并且对于真正说服受害者交出密码或卡号的因素几乎没有帮助：页面的外观。本文描述了一种基于视觉的图像处理方法，用于网络钓鱼检测，该方法将渲染后的网页视为人眼所看到的图像，判断其是否与可信品牌匹配。我们训练了一个YOLOv8卷积神经网络，根据布局、标志位置、配色方案和登录表单结构对完整网页截图进行网络钓鱼或合法网站的分类，而不是依赖于从页面提取的文本。该系统在保留的测试集上达到了92%的分类准确率，处理单个截图的时间约为100毫秒，并且在针对光照、压缩和缩放变化进行的数据增强后，相比于增强前的基线，假阳性率降低了11%。本文详细介绍了数据集构建、增强策略、模型架构和训练设置，以及最终的性能表现，并讨论了这种视觉检测器如何与现有的基于URL和内容的防御措施相辅相成，而不是取而代之。
cs.AI / 267 / 2609.34024
Jev in Medicine: A Benchmark Evaluation. Preliminary Results
医学中的Jev：基准评估初步结果
Abstract
Jev is a non-generative "System One" model that assigns probabilities to predefined answer options and cannot answer outside them. Its accuracy and calibration on medical question-answering and case-based diagnostic-reasoning tasks are unknown. We evaluated Jev 1.13 on four medical benchmarks: MetaMedQA, PubMedQA, DiagnosisArena-MCQ and the NEJM Case Challenges. GPT-6 Sol, with (medium) and without reasoning, was the reference. The primary outcome was top-1 accuracy; key secondary outcomes were calibration, selective prediction and recognition of unanswerable questions. All 8,469 requests returned a valid answer. Jev's accuracy was similar to that of GPT-6 Sol with medium reasoning on PubMedQA (78.4% vs 78.2%;), lower on MetaMedQA (74.8% vs 82.7%) and much lower on DiagnosisArena-MCQ (59.8% vs 82.4%;) and the NEJM cases (61.8% vs 82.4%). On MetaMedQA, Jev's probabilities were the best calibrated (expected calibration error 0.063 vs 0.146), and its answers with a probability of at least 0.9 (52.9% of questions) were 93.4% accurate, but GPT-6 Sol was as accurate when it accepted a similar proportion of questions. On DiagnosisArena-MCQ, Jev's probabilities discriminated poorly (AUROC 0.645 vs 0.768). Of the 162 questions whose correct answer was "I don't know or cannot answer", Jev chose that option for 10.5% (GPT-6 Sol, 8.6%). Median latency was 0.27-0.31 s; all 2,823 items cost USD 0.08. Jev was fast and inexpensive, and its accuracy was similar to that of a frontier LLM on research abstracts but lower on examination questions and much lower on complex diagnostic cases. Task-specific validation is required before clinical use.
Chinese Translation
Jev是一个非生成性的“系统一”模型，它为预定义的答案选项分配概率，无法在这些选项之外作答。其在医学问答和基于案例的诊断推理任务中的准确性和校准情况尚不清楚。我们在四个医学基准上评估了Jev 1.13：MetaMedQA、PubMedQA、DiagnosisArena-MCQ和NEJM案例挑战。GPT-6 Sol（中等推理）作为参考。主要结果是top-1准确率；关键次要结果包括校准、选择性预测和对无法回答问题的识别。所有8,469个请求均返回有效答案。Jev的准确性与GPT-6 Sol在PubMedQA上的中等推理相似（78.4%对78.2%），在MetaMedQA上较低（74.8%对82.7%），在DiagnosisArena-MCQ（59.8%对82.4%）和NEJM案例（61.8%对82.4%）上则低得多。在MetaMedQA上，Jev的概率校准最佳（预期校准误差0.063对0.146），其概率至少为0.9的答案（占52.9%的问题）准确率为93.4%，但当GPT-6 Sol接受类似比例的问题时，其准确率也相当。在DiagnosisArena-MCQ上，Jev的概率区分能力较差（AUROC 0.645对0.768）。在162个正确答案为“我不知道或无法回答”的问题中，Jev选择该选项的比例为10.5%（GPT-6 Sol为8.6%）。中位延迟为0.27-0.31秒；所有2,823个项目的成本为0.08美元。Jev快速且成本低，其准确性与前沿大型语言模型在研究摘要上的表现相似，但在考试问题和复杂诊断案例上的表现则低得多。在临床使用之前，需要进行特定任务的验证。
cs.AI / 268 / 2609.34039
Large Language Models for Structured Clinical Data Analysis: Dual-Agent Grounding and Validation
用于结构化临床数据分析的大型语言模型：双代理基础与验证
Abstract
Objective: To develop and characterize CLEAR-Med, a dual-agent framework for natural-language analysis of structured clinical data that separates SQL-based invocation from independent validation. Methods: CLEAR-Med uses one agent to translate a question into executable Structured Query Language (SQL), retain the executed query and database result, and produce a draft. Deterministic checks and a separately invoked cross-provider Validation Agent then accept the draft, request one bounded repair, or abstain. We formalized the system as a bounded selective pipeline and evaluated CLEAR-Med's configuration and scalability, and the Invocation Agent's accuracy and consistency on a 25-query development benchmark, using a harmonized 21-site neonatal hypoxic-ischemic encephalopathy table containing 532 de-identified infant records and approximately 1,300 variables. Results: CLEAR-Med completed all six nominal scalability configurations, including 500x1300. Across 25 development-benchmark queries repeated five times, the Invocation Agent answered 83 of 125 responses correctly (66.4%; query-cluster bootstrap 95% CI, 48.0-83.2%), compared with 15 of 125 (12.0%; 95% CI, 3.2-22.4%) for the ungrounded ChatGPT baseline, a paired improvement of 54.4 percentage points (95% CI, 36.8-72.0%). Conclusion: CLEAR-Med provides a general architecture for traceable analysis of structured clinical data: numerical claims remain linked to executed SQL, and unresolved cases can fail closed. The reported experiments characterize CLEAR-Med's configuration and scalability and the Invocation Agent's accuracy, while the formal analysis establishes the encoded-property guarantee of the complete control flow; a prospective full-pipeline evaluation of the validation and abstention stages is the next stage of this work.
Chinese Translation
目的：开发并描述CLEAR-Med，一个用于结构化临床数据自然语言分析的双代理框架，该框架将基于SQL的调用与独立验证分开。方法：CLEAR-Med使用一个代理将问题翻译为可执行的结构化查询语言（SQL），保留执行的查询和数据库结果，并生成草稿。确定性检查和一个单独调用的跨提供者验证代理随后接受草稿，要求进行一次有限的修正，或选择不作回应。我们将该系统形式化为一个有限选择管道，并在一个包含532个去标识婴儿记录和约1300个变量的协调21个站点的新生儿缺氧缺血性脑病表上，评估了CLEAR-Med的配置和可扩展性，以及调用代理在25个查询开发基准上的准确性和一致性。结果：CLEAR-Med完成了所有六个名义可扩展性配置，包括500x1300。在25个开发基准查询中重复五次，调用代理正确回答了125个响应中的83个（66.4%；查询集群自助法95%置信区间，48.0-83.2%），而未基础的ChatGPT基线仅正确回答了125个中的15个（12.0%；95%置信区间，3.2-22.4%），两者相比提高了54.4个百分点（95%置信区间，36.8-72.0%）。结论：CLEAR-Med提供了一种可追溯的结构化临床数据分析的通用架构：数值声明与执行的SQL保持关联，未解决的案例可以失败关闭。报告的实验表征了CLEAR-Med的配置和可扩展性以及调用代理的准确性，而形式分析则确立了完整控制流的编码属性保证；对验证和弃权阶段的前瞻性全管道评估是该工作的下一阶段。
cs.AI / 269 / 2609.34049
Thinking Outside the Box: Retention and Transmission of Information in Sliding-Window KV Inference
跳出框架思维：滑动窗口 KV 推理中的信息保留与传递
Abstract
Sliding-window KV inference refers to processing a sequence incrementally while retaining only a fixed-size cache of recent key and value states. It can be applied to pretrained causal transformers at inference time without additional training, while its KV-cache memory remains fixed as more tokens are processed. Because cached states are computed in the context of earlier tokens, they may carry information from beyond the current window and transmit it to later states. This study presents a series of experiments using five open-weight models spanning Qwen, Llama, Mistral, and Muse Glimmer. We investigate whether information originating outside the immediate context window can persist through a rolling KV cache and remain useful for retrieval. Initial results show that retaining previously computed states improves retrieval across the models tested compared with recomputing the final fixed window from raw tokens. We then measure how far this effect extends and find that Muse Glimmer and Mistral 7B show the strongest \emph{latent information relay}: they can recover information even after the relevant source tokens have left the cache. Both models incorporate sliding-window attention in their published architectures, an association that motivates testing whether training with sliding windows promotes more reliable information retention.
Chinese Translation
滑动窗口 KV 推理是指在处理序列时以增量方式进行，同时仅保留固定大小的最近键值状态缓存。它可以在推理时应用于预训练的因果变换器，而无需额外训练，同时在处理更多标记时其 KV 缓存内存保持不变。由于缓存状态是在早期标记的上下文中计算的，因此它们可能携带来自当前窗口之外的信息，并将其传递给后续状态。本研究展示了一系列实验，使用了五个开放权重模型，包括 Qwen、Llama、Mistral 和 Muse Glimmer。我们调查了源自即时上下文窗口之外的信息是否能够通过滚动 KV 缓存持续存在并保持可用于检索。初步结果显示，与从原始标记重新计算最终固定窗口相比，保留先前计算的状态提高了所测试模型的检索效果。随后，我们测量了这一效应的延续范围，发现 Muse Glimmer 和 Mistral 7B 显示出最强的潜在信息传递能力：即使相关源标记已离开缓存，它们仍能恢复信息。这两个模型在其发布的架构中都包含滑动窗口注意力，这一关联促使我们测试使用滑动窗口训练是否能促进更可靠的信息保留。
cs.AI / 270 / 2609.34069
Towards Certificate-Driven Software Porting: A Self-Improving Agentic Harness for Scientific Program Optimization
面向证书驱动的软件移植：一种自我改进的智能代理框架用于科学程序优化
Abstract
The upgrade and rewriting of large scientific codebases has traditionally been a major challenge. While evolutionary search with large language models (LLMs) can port and accelerate legacy code, repair feedback in prompts alone does not prevent subsequent candidates from repeating the same errors. We introduce Certificate-Driven Evolutionary Search (CDES), which extends evolutionary search with enforceable restrictions derived from failed candidates, recorded as certificates of assumptions, checker evidence, and justified restrictions. Its control logic enforces these restrictions through rejection, backtracking, and targeted repair while preserving compatible edits. We apply CDES to CPU-to-GPU translation of two particle-simulation functions from the Geant4 toolkit, evaluated with a harness that goes beyond unit tests to combine formal checks, numerical comparisons, physics checks, and GPU safety tests. Generated implementations achieve 13.78x and 23.54x function-level speedups over CPU code, including data conversion and transfers; for one function, GPU throughput exceeds an expert implementation by 14.9%, reaching 16.1% when complementary components are combined. In an ablation over execution settings, certificate feedback increases the fraction of candidates passing required correctness checks from 55% to 90%.
Chinese Translation
大型科学代码库的升级和重写一直是一个主要挑战。虽然使用大型语言模型（LLMs）的进化搜索可以移植和加速遗留代码，但仅依靠提示中的修复反馈并不能阻止后续候选者重复相同的错误。我们提出了证书驱动的进化搜索（Certificate-Driven Evolutionary Search, CDES），该方法通过来自失败候选者的可强制执行的限制扩展了进化搜索，这些限制记录为假设证书、检查证据和合理限制。其控制逻辑通过拒绝、回溯和有针对性的修复来强制执行这些限制，同时保留兼容的编辑。我们将CDES应用于Geant4工具包中两个粒子模拟函数的CPU到GPU转换，并通过一种超越单元测试的测试框架进行评估，该框架结合了形式检查、数值比较、物理检查和GPU安全测试。生成的实现相较于CPU代码在函数级别上实现了13.78倍和23.54倍的加速，包括数据转换和传输；对于一个函数，GPU的吞吐量超过了专家实现14.9%，在组合互补组件时达到16.1%。在执行设置的消融实验中，证书反馈将通过所需正确性检查的候选者比例从55%提高到90%。
cs.AI / 271 / 2609.34072
PhysFieldBench: Can Multimodal Models Understand Physical Fields?
PhysFieldBench：多模态模型能理解物理场吗？
Abstract
Multimodal large language models (MLLMs) are increasingly envisioned as core components of scientific and engineering agents, yet their ability to interpret physical fields remains poorly understood. Existing physics benchmarks largely emphasize textbook problem solving or intuitive physical reasoning, leaving open whether MLLMs can infer physically meaningful information from continuous field observations. We introduce PhysFieldBench, a benchmark comprising 24 tasks and 1,160 evaluation examples across controlled equation fields, simulated physical fields, and observed physical fields. The tasks assess three forms of inference: identifying physical mechanisms, comparing latent control variables, and predicting outcome properties. Across representative open-source and proprietary MLLMs, zero-shot performance is low: the best model achieves a chance-normalized score of 29.3, while several open-source models remain near chance. In contrast, a task-specific supervised vision transformer performs substantially better, demonstrating that the inputs contain learnable physical information. To diagnose these failures, a structured self-explanation analysis attributes most errors to missed visual patterns and incorrect visual-to-physical mappings. Further, to explore whether post-training can improve physical inference and generalize to unseen tasks, we compare supervised fine-tuning with final answers or chain-of-thought supervision and reinforcement learning. Final-answer supervision performs best overall but transfers less effectively, whereas reinforcement learning after chain-of-thought supervision achieves the best generalization. Together, these findings highlight the need to improve visual-to-physical grounding and cross-task generalization for MLLMs to reliably interpret physical fields in scientific and engineering workflows.
Chinese Translation
多模态大型语言模型（MLLMs）日益被视为科学与工程智能体的核心组成部分，然而它们对物理场的理解能力尚不明确。现有的物理学基准测试主要侧重于教科书式的问题求解或直观的物理推理，尚未探讨MLLMs是否能够从连续的场观测中推断出具有物理意义的信息。我们提出了PhysFieldBench，一个包含24个任务和1160个评估样本的基准，涵盖受控方程场、模拟物理场和观测物理场。任务评估三种推理形式：识别物理机制、比较潜在控制变量以及预测结果属性。在代表性开源和专有MLLMs中，零样本性能较低：最佳模型的机会归一化得分为29.3，而多个开源模型表现接近随机水平。相比之下，针对任务专门训练的监督视觉变换器（vision transformer）表现显著更好，表明输入中包含可学习的物理信息。为诊断失败原因，结构化自我解释分析将大部分错误归因于未捕捉到的视觉模式和错误的视觉到物理映射。此外，为探究后训练（post-training）是否能提升物理推理能力并泛化至未见任务，我们比较了基于最终答案或思维链（chain-of-thought）监督的有监督微调与强化学习。最终答案监督整体表现最佳但迁移能力较弱，而在思维链监督后进行的强化学习实现了最佳泛化能力。综合来看，这些发现强调了提升视觉到物理的映射能力及跨任务泛化能力的必要性，以使MLLMs能够在科学与工程工作流程中可靠地解读物理场。
cs.AI / 272 / 2609.34079
GenoMorph: Pathway-Grounded Genomic Disease Reasoning via Adaptive Latent Computation
GenoMorph：基于通路的基因组疾病推理通过自适应潜在计算
Abstract
Large language models (LLMs) have demonstrated strong capabilities in biological reasoning; however, genomic disease inference remains largely dependent on memorized gene-disease associations rather than understanding biological pathways. This shortcut learning undermines robustness and generalization, and breaks down when molecular identifiers are unavailable. We present GenoMorph, a multimodal genomic reasoning framework that shifts disease prediction from associative gene-disease mapping toward pathway-grounded reasoning. GenoMorph couples a frozen DNA foundation model with question-conditioned cross-attention fusion, self-adaptive latent reasoning (LatentSp), a residual reasoning gate for iterative genomic evidence reinjection, and rejection sampling fine-tuning regularized by hierarchical optimal transport (OT). Rather than learning direct gene-disease mappings, GenoMorph aligns genomic sequence representations with latent pathway dynamics, enabling reasoning trajectories that follow molecular interactions before producing disease predictions. LatentSp dynamically allocates computation according to reasoning confidence, reducing unnecessary reasoning steps and improving inference efficiency. We further construct an anonymized benchmark from the Kyoto Encyclopedia of Genes and Genomes (KEGG), replacing every gene and molecular identifier with anonymous symbols while preserving sequences and pathway topology, thereby removing memorization shortcuts. GenoMorph raises the weighted F1 from 0.7863 (BioReason) to 0.9412, and rejection sampling fine-tuning with self-adaptive latent reasoning pushes it to 0.9725 while cutting latency nearly 60%. On the anonymized benchmark it reaches 0.9465 F1, substantially outperforming prior systems and confirming that accurate disease prediction can arise from pathway reasoning rather than memorized gene-disease associations.
Chinese Translation
大型语言模型（LLMs）在生物推理方面展现了强大的能力；然而，基因组疾病推断仍然在很大程度上依赖于记忆的基因-疾病关联，而非理解生物通路。这种捷径学习削弱了模型的鲁棒性和泛化能力，并在缺乏分子标识符时崩溃。我们提出了GenoMorph，一个多模态基因组推理框架，将疾病预测从关联的基因-疾病映射转向基于通路的推理。GenoMorph将一个冻结的DNA基础模型与基于问题的交叉注意力融合、自适应潜在推理（LatentSp）、用于迭代基因组证据再注入的残差推理门，以及通过层次最优传输（OT）正则化的拒绝采样微调相结合。GenoMorph并不是学习直接的基因-疾病映射，而是将基因组序列表示与潜在通路动态对齐，从而实现推理轨迹，遵循分子相互作用后再产生疾病预测。LatentSp根据推理信心动态分配计算，减少不必要的推理步骤，提高推断效率。我们进一步从京都基因组百科全书（KEGG）构建了一个匿名基准，将每个基因和分子标识符替换为匿名符号，同时保留序列和通路拓扑，从而消除记忆捷径。GenoMorph将加权F1从0.7863（BioReason）提升至0.9412，而结合自适应潜在推理的拒绝采样微调将其推至0.9725，同时将延迟减少近60%。在匿名基准上，其F1达到0.9465，显著优于之前的系统，并确认准确的疾病预测可以源于基于通路的推理，而非记忆的基因-疾病关联。
cs.AI / 273 / 2609.34082
K-OPSD: Verifiable On-Policy Self-Distillation for Post-Training Vision-Language Models on AEC Drawings
K-OPSD：可验证的政策内自蒸馏方法用于AEC图纸的后训练视觉语言模型
Abstract
Interpreting architecture, engineering, and construction (AEC) drawings is hard for general Multimodal Large Language Models (MLLMs) and vision-language models (VLMs). We introduce K-OPSD, a VLM post-training methodology for improving AEC drawing understanding. Building on On-Policy Self-Distillation (OPSD) with verifiable supervision, we construct a teacher from the model's own best-of-N generations, certified by a process-level verifier, and rescue failed prompts by resampling under a hint that exposes the verified answer. We then perform an on-policy model update by training on verified completions with a cross-entropy inner-loss, outperforming the bounded token-wise generalized Jensen-Shannon divergence (JSD) used by on-policy distillation. Using K-OPSD, we fine-tune Qwen3-VL models on the AECV-Bench dataset. The resulting models attain the top average judge score (0.819) and combined accuracy (0.738), achieving competitive results against open-source baseline models. The recipe transfers to the out-of-domain ArchCAD dataset, where the 8B model gains most. We present the verifier suite and the continual learning and self-improving pipeline, our results provide preliminary evidence that verifier-guided self-distillation is a promising route toward more reliable machine reading of architecture drawings.
Chinese Translation
解读建筑、工程和施工（AEC）图纸对于一般的多模态大型语言模型（MLLMs）和视觉语言模型（VLMs）来说是困难的。我们提出了K-OPSD，一种用于提高AEC图纸理解的VLM后训练方法。基于具有可验证监督的政策内自蒸馏（OPSD），我们从模型自身的最佳N次生成中构建教师，由过程级验证器认证，并通过在揭示验证答案的提示下重新采样来挽救失败的提示。然后，我们通过对经过验证的完成进行交叉熵内损失训练，执行政策内模型更新，超越了政策内蒸馏中使用的有界标记级广义詹森-香农散度（JSD）。使用K-OPSD，我们在AECV-Bench数据集上微调了Qwen3-VL模型。结果模型达到了最高的平均评审分数（0.819）和综合准确率（0.738），在与开源基线模型的竞争中取得了良好的成绩。该方法在域外的ArchCAD数据集中也得到了转移，其中8B模型收益最大。我们展示了验证器套件以及持续学习和自我改进的管道，我们的结果提供了初步证据，表明验证器指导的自蒸馏是实现更可靠的建筑图纸机器阅读的有希望的途径。
cs.AI / 274 / 2609.34103
A Differentiable Optimization Framework for Registering Sequential Bounding Boxes with Point Cloud Stream
用于注册顺序边界框与点云流的可微优化框架
Abstract
Refining a sequence of coarse 3D bounding boxes against a LiDAR point-cloud stream demands tracks that are geometrically accurate (high IoU) and temporally coherent (low roughness), preferably without training data. The usual recipe keeps the two concerns apart: register each frame independently, then smooth the trajectory afterwards with a Kalman~RTS or Savitzky--Golay filter. Smoothing displaces boxes from a geometric optimum and never re-optimises, so it trades accuracy for smoothness. We instead fold the temporal smoothness constraint into a training-free registration objective and solve for all poses jointly with L-BFGS. The payoff depends on how well the object is seen. On well-observed tracks it is large: within the low-roughness budget, the joint objective beats both post-hoc smoothers on paired multi-seed statistics and cuts roughness several-fold relative to frame-wise registration at matched accuracy. Treating visibility as an experimental variable exposes the limit. The advantage decays monotonically as views become one-sided, until it is indistinguishable from zero for near-edge-on objects and slightly negative under a ray-cast simulator with range-dependent density and ego motion, where the decoupled pipeline is in fact ahead at tight roughness budgets. We locate that boundary and trace it to one term: orientation alignment ties yaw to the estimated velocity and fails once that estimate is noisy. A ground-truth-free rule can choose the temporal scale and keep every track inside the roughness budget.
Chinese Translation
对LiDAR点云流进行粗糙3D边界框序列的精细化要求几何上准确的轨迹（高IoU）和时间上连贯的轨迹（低粗糙度），最好不依赖训练数据。通常的方法将这两个问题分开处理：独立地注册每一帧，然后使用卡尔曼~RTS或Savitzky--Golay滤波器对轨迹进行平滑处理。平滑处理会使边界框偏离几何最优解，并且从不重新优化，因此在平滑性和准确性之间进行了权衡。我们则将时间平滑性约束融入无训练数据的注册目标中，并使用L-BFGS联合求解所有姿态。收益取决于物体的可见性。在观察良好的轨迹上，收益很大：在低粗糙度预算内，联合目标在配对多种子统计上优于两种后处理平滑方法，并在匹配准确度的情况下将粗糙度降低了几倍。将可见性视为实验变量揭示了其极限。当视角变得单侧时，优势单调递减，直到对于近边缘物体几乎无法区分为零，并且在具有范围依赖密度和自我运动的光线投射模拟器下略微为负，此时解耦管道在紧凑的粗糙度预算中实际上处于领先地位。我们确定了这一边界，并将其追溯到一个因素：方向对齐将偏航与估计速度联系在一起，并在该估计噪声较大时失败。一个无真实值的规则可以选择时间尺度，并将每个轨迹保持在粗糙度预算内。
cs.AI / 275 / 2609.34111
SpecRegMatch: Robust Semi-Supervised Regression for Vehicle Interior Noise Prediction
SpecRegMatch：用于车辆内部噪声预测的鲁棒半监督回归
Abstract
The rapid advancement of artificial intelligence has observed increased application in predicting vehicle interior noise levels within the automotive industry. However, the collection of labeled data for training models in this context involves significant costs. Previous studies in semi-supervised regression (SSR) have effectively mitigated the reliance on labeled data by incorporating unlabeled data. Nonetheless, these approaches often introduce a high computational cost due to the training of multiple models and data sampling. This study introduces SpecRegMatch, a novel SSR method aimed at addressing the computational cost associated with training by leveraging a single model, thus eliminating the need for multiple data samplings. SpecRegMatch integrates consistency regularization and information maximization to robustly train the model, achieved through various augmentations applied to both the embedding vectors and predicted values. Experimental results demonstrate that SpecRegMatch achieves state-of-the-art performance across various scenarios, even when using a single model. It attains a remarkable performance, as indicated by an R^2 score of 0.434. This is especially noteworthy in scenarios where labeled data is scarce. You can access the code for our proposed method at https://github.com/sejin-sim/SpecRegMatch.
Chinese Translation
人工智能的快速发展在汽车行业中观察到了对车辆内部噪声水平预测的应用增加。然而，在此背景下，收集用于训练模型的标记数据涉及显著的成本。以往的半监督回归（SSR）研究通过结合未标记数据有效减轻了对标记数据的依赖。然而，这些方法通常由于训练多个模型和数据采样而引入较高的计算成本。本研究提出了SpecRegMatch，一种新颖的SSR方法，旨在通过利用单一模型来解决与训练相关的计算成本，从而消除对多个数据采样的需求。SpecRegMatch整合了一致性正则化和信息最大化，通过对嵌入向量和预测值施加各种增强来稳健地训练模型。实验结果表明，SpecRegMatch在各种场景下实现了最先进的性能，即使在使用单一模型的情况下也达到了显著的表现，其R^2得分为0.434。这在标记数据稀缺的情况下尤为值得注意。您可以在 https://github.com/sejin-sim/SpecRegMatch 访问我们提出的方法的代码。
cs.AI / 276 / 2609.34113
GUITAR: Structured Failure Diagnosis of GUI Agents via State Transitions
GUITAR：通过状态转移对图形用户界面代理进行结构化故障诊断
Abstract
Understanding where and why Graphical User Interface (GUI) agents fail is essential for building more reliable systems, yet current evaluation relies on step accuracy, a metric that treats each screen independently and overlooks the underlying structure of GUI environments. This leads to two critical blind spots: (1) functionally equivalent screens are evaluated in isolation, obscuring systematic failure patterns across shared screens; and (2) the long-tailed GUI distribution renders failures on rare but critical screens invisible under standard metrics. To address these issues, we propose \textbf{GUITAR}, a state-centric diagnostic framework that performs structured failure analysis over both states and transitions, using a State Transition Graph (STG) by mapping visually diverse screens to shared functional states. Across 8 agents and 6 tasks from AndroidControl and Mind2Web, GUITAR reveals that 60.4\% of failures occur in 20\% of states, localizing errors to a small set of bottlenecks. Bottleneck-targeted guidance improves SR by 2.8\% and retains a 1.88\% average gain across 7 agents under three-fold trajectory-held-out evaluation with fully automatic STGs. These findings demonstrate the diagnostic and actionable value of structure-aware evaluation within the evaluated mobile and web tasks. Code is available at https://github.com/sqzhang-lazy/GUITAR
Chinese Translation
了解图形用户界面（GUI）代理失败的原因和位置对于构建更可靠的系统至关重要，但目前的评估依赖于步骤准确性，这一指标将每个屏幕视为独立的，忽视了GUI环境的潜在结构。这导致了两个关键盲点：（1）功能上等效的屏幕被孤立评估，掩盖了共享屏幕之间的系统性故障模式；（2）长尾的GUI分布使得在稀有但关键屏幕上的故障在标准指标下变得不可见。为了解决这些问题，我们提出了 extbf{GUITAR}，一个以状态为中心的诊断框架，通过状态转移图（State Transition Graph, STG）对状态和转移进行结构化故障分析，将视觉上多样的屏幕映射到共享的功能状态。在来自AndroidControl和Mind2Web的8个代理和6个任务中，GUITAR揭示了60.4 ext{%}的故障发生在20 ext{%}的状态中，将错误定位到一小组瓶颈。针对瓶颈的指导使成功率（SR）提高了2.8 ext{%}，并在三次轨迹保留评估中，7个代理的平均增益保持在1.88 ext{%}，使用完全自动化的STG。这些发现展示了在评估的移动和网络任务中，结构感知评估的诊断和可操作价值。代码可在https://github.com/sqzhang-lazy/GUITAR获取。
cs.AI / 277 / 2609.34132
From Attack Success to Attack Severity: Counterfactual Memory Attacks on LLM Agents
从攻击成功到攻击严重性：对大型语言模型代理的反事实记忆攻击
Abstract
As LLM agents increasingly rely on persistent memory for long-horizon and personalized behavior, they can retain and reuse information across interactions, but this also creates a lasting channel through which malicious memory writes can influence future behavior. Persistent-memory attacks are typically evaluated by whether they succeed, yet successful attacks can leave persistent states with substantially different downstream consequences. We study this severity as a distinct attack-design objective and formalize it with counterfactual memory regret (CMR), the paired increase in expected downstream loss relative to clean memory. We introduce MemHarm, which predeclares a finite class of sparse, grounded semantic edits, evaluates candidates through the normal agent memory interface using offline paired-loss feedback, and certifies resolved selections within that class. Compared with attack-success optimization, CMR-guided selection produces substantially larger downstream loss while retaining most of the success-rate gain. Across two agent benchmarks and diverse memory designs, MemHarm attains the highest CMR point estimates among the evaluated general attacks on identical support. Factor-removal interventions link this harm to the selected semantic factor, and native-agent deployments verify the write-to-fresh-process attack path.
Chinese Translation
随着大型语言模型（LLM）代理越来越依赖持久记忆以实现长期和个性化行为，它们能够在交互中保留和重用信息，但这也创造了一个持久的渠道，通过该渠道恶意的记忆写入可以影响未来的行为。持久记忆攻击通常通过其成功与否来评估，然而成功的攻击可能会留下具有显著不同下游后果的持久状态。我们将这种严重性作为一个独特的攻击设计目标进行研究，并通过反事实记忆遗憾（Counterfactual Memory Regret, CMR）进行形式化，即相对于干净记忆，期望下游损失的成对增加。我们引入了MemHarm，它预先声明了一类有限的稀疏、基于语义的编辑，通过正常的代理记忆接口使用离线成对损失反馈来评估候选项，并在该类中认证解决的选择。与攻击成功优化相比，CMR引导的选择在保持大部分成功率提升的同时，产生了显著更大的下游损失。在两个代理基准和多样的记忆设计中，MemHarm在评估的通用攻击中达到了最高的CMR点估计。去因干预将这种损害与所选的语义因子联系起来，而本地代理部署验证了写入新进程的攻击路径。
cs.AI / 278 / 2609.34134
StateGuard: Analytical-State Management with Validity-Aware Intervention for Long-Horizon Data Agents
StateGuard：具有有效性感知干预的长时间数据代理的分析状态管理
Abstract
LLM-based agents have shown strong capabilities in automated data analysis and are increasingly moving toward long-horizon, multi-stage analytical workflows. However, as the analytical process evolves, constraints, variables, and conclusions remain implicitly embedded in interaction histories, making it difficult for agents to track which analytical artifacts remain valid over increasingly long horizons and changing dependencies. Consequently, stale artifacts may be silently inherited, propagating errors to downstream stages. To address this challenge, we propose StateGuard, an analytical-state validity management framework for long-horizon data agents. StateGuard externalizes evolving analytical progress into a state graph containing constraints, versioned variables, intermediate conclusions, and cross-state relations, treating each state as an executable, verifiable, and traceable object rather than textual memory alone. StateGuard maintains state validity through evidence-grounded verification and hierarchical intervention. To equip StateGuard with these capabilities, we first introduce Manager-Oriented Counterfactual Supervision, which constructs 3K state-centric trajectories through counterfactual runtime synthesis to fine-tune StateGuard for state maintenance, verification, and repair. We then apply Validity-Guided Policy Optimization, using runtime validity evidence to provide fine-grained learning signals for protocol correctness, state grounding, and intervention quality. Experiments on three diverse long-horizon data-analysis benchmarks show that StateGuard consistently improves data-agent performance while reducing dependency-induced downstream error propagation, demonstrating the advantages of explicit analytical-state management for reliable long-horizon data analysis.
Chinese Translation
基于大型语言模型（LLM）的代理在自动化数据分析方面表现出强大的能力，并越来越多地朝着长时间、多阶段的分析工作流程发展。然而，随着分析过程的演变，约束、变量和结论仍隐含地嵌入在交互历史中，这使得代理难以跟踪哪些分析成果在不断延长的时间范围和变化的依赖关系中仍然有效。因此，过时的成果可能会被默默继承，从而将错误传播到下游阶段。为了解决这一挑战，我们提出了StateGuard，一个用于长时间数据代理的分析状态有效性管理框架。StateGuard将不断演变的分析进展外部化为一个状态图，其中包含约束、版本化变量、中间结论和跨状态关系，将每个状态视为一个可执行、可验证和可追踪的对象，而不仅仅是文本记忆。StateGuard通过基于证据的验证和分层干预来维护状态有效性。为了赋予StateGuard这些能力，我们首先引入了面向管理者的反事实监督（Manager-Oriented Counterfactual Supervision），通过反事实运行时合成构建3K状态中心轨迹，以微调StateGuard以进行状态维护、验证和修复。然后，我们应用有效性引导的策略优化（Validity-Guided Policy Optimization），利用运行时有效性证据为协议正确性、状态基础和干预质量提供细粒度的学习信号。在三个不同的长时间数据分析基准上的实验表明，StateGuard始终改善数据代理的性能，同时减少依赖引起的下游错误传播，展示了显式分析状态管理在可靠长时间数据分析中的优势。
cs.AI / 279 / 2609.34135
Evo2Team: When Do Evolved Skills Transfer? From Selection to Deployment
Evo2Team：进化技能何时转移？从选择到部署
Abstract
A skill bank that helps one multi-agent system may leave another's behavior unchanged. A transferred rule helps only when target agents act on it successfully. We study this path for routing and communication skills in Count-Frequency and AgentsNet, using teams of 4--32 agents and GPT and Qwen model ladders. Source evolution meets a joint quality, cost, model-tier, and confirmation goal in 14 of 16 settings. We then evaluate Evo2Team, which selects, adapts, and confirms source skills for the target team, alongside six frozen selectors across 28 transfer directions. Evo2Team's target-side exploration cost is below that of evolving a new target bank in every direction, even when reused reference evaluations are charged once. Twenty of 28 held-out outcomes meet the positive-transfer criterion, including three saved diagnostic tests. Selection alone does not explain these outcomes: KNN and CORAL choose different banks in two AgentsNet directions but produce identical recorded executions. When Evo2Team changes execution, gains can reach many tasks, as in a Count-Frequency direction that improves 28 of 32 tasks over KNN. Seven positive AgentsNet outcomes save 6.1--14.6\% in deployment cost while using transferred skills on only three to six of fifteen tasks. In five earlier accepted directions, all 22 task records using transferred skills pass three fixed-graph confirmations, but four fail in recorded executions on new graphs. Graphs and model responses change together in this comparison. These results show that skill transfer must be assessed through the actions agents take, the tasks those actions reach, and the quality and cost of the final deployment.
Chinese Translation
一个对某个多智能体系统有帮助的技能库可能对另一个系统的行为没有影响。转移的规则只有在目标智能体成功执行时才有帮助。我们研究了在 Count-Frequency 和 AgentsNet 中的路由和通信技能的转移路径，使用了由 4 到 32 个智能体组成的团队以及 GPT 和 Qwen 模型梯队。在 16 种设置中，有 14 种情况下源进化满足了联合质量、成本、模型层级和确认目标。随后，我们评估了 Evo2Team，它为目标团队选择、适应和确认源技能，并与 28 个转移方向中的六个固定选择器进行比较。Evo2Team 在目标侧的探索成本低于在每个方向上进化一个新的目标库的成本，即使在重复使用的参考评估只收取一次的情况下也是如此。在 28 个保留结果中，有 20 个符合正向转移标准，包括三个保存的诊断测试。仅仅选择并不能解释这些结果：KNN 和 CORAL 在两个 AgentsNet 方向上选择不同的技能库，但产生相同的记录执行。当 Evo2Team 改变执行时，收益可以覆盖多个任务，例如在 Count-Frequency 方向上，相比 KNN 提升了 32 个任务中的 28 个。七个正向的 AgentsNet 结果在仅对 15 个任务中的三到六个任务使用转移技能的情况下，节省了 6.1% 到 14.6% 的部署成本。在五个先前接受的方向中，所有 22 个使用转移技能的任务记录都通过了三个固定图的确认，但在新图上的记录执行中有四个失败。在这个比较中，图和模型响应是一起变化的。这些结果表明，技能转移必须通过智能体采取的行动、这些行动所达到的任务以及最终部署的质量和成本进行评估。
cs.AI / 280 / 2609.34136
Waggle: Learning One Anonymous Local Law for Self-Organizing LLM Swarms
Waggle：为自组织的 LLM 群体学习一种匿名的局部法则
Abstract
As LLM agents increasingly collaborate on complex tasks, how to organize their interactions becomes a central design question. Existing multi-agent systems typically learn or adapt explicit roles, hierarchies, routing policies, or communication topologies. We shift the learning target to a reusable local law that can be shared across interchangeable agents and adapt coordination as populations or interaction conditions change, without redefining a global organization. We introduce Waggle, a shared anonymous policy over bounded local views that jointly selects task actions, semantic communication, and local commitment updates. Repeated execution of the same law allows coordination to form, persist, and reorganize online without explicit roles or global topology. To learn this law across interchangeable agents and evolving coordination, we develop Swarm-Consistent Distillation (SCD), combining anonymous-orbit consistency with rollout-grounded prediction of the next local coordination field, with no added inference-time components. Across diverse coordination settings, the same learned law remains effective as populations and interaction budgets change, retains over 96% of substrate-specific oracle quality, and transfers without retraining; SCD further improves reorganization after counterevidence. Together, these results show that LLM-agent organization can emerge and adapt through repeated execution of a learned local law.
Chinese Translation
随着 LLM 代理在复杂任务上的协作日益增多，如何组织它们的互动成为一个核心设计问题。现有的多代理系统通常学习或适应明确的角色、等级、路由策略或通信拓扑。我们将学习目标转向一种可重用的局部法则，该法则可以在可互换的代理之间共享，并在群体或互动条件变化时调整协调，而无需重新定义全球组织。我们引入了 Waggle，这是一种基于有限局部视图的共享匿名策略，能够共同选择任务动作、语义通信和局部承诺更新。重复执行相同的法则使得协调能够形成、持续并在线重组，而无需明确的角色或全球拓扑。为了在可互换的代理和不断演变的协调中学习这一法则，我们开发了 Swarm-Consistent Distillation (SCD)，将匿名轨道一致性与下一个局部协调场的基于回放的预测相结合，无需增加推理时的组件。在多样的协调环境中，学习到的相同法则在群体和互动预算变化时仍然有效，保持超过 96% 的基于基质的 oracle 质量，并且无需重新训练即可迁移；SCD 进一步改善了在反证据后的重组。综合这些结果表明，LLM 代理的组织可以通过重复执行学习到的局部法则而自发出现并适应。
cs.AI / 281 / 2609.34137
You Can't Have It Both Ways: Concept Entanglement Limits Diffusion Model Unlearning
无法两全其美：概念纠缠限制扩散模型的遗忘能力
Abstract
Concept unlearning in text-to-image diffusion models aims to suppress a target concept (e.g., \texttt{horse}) while preserving related but distinct content (e.g., \texttt{donkey}), yet existing methods either leak under indirect prompts or visibly degrade other concepts. We show that these failure modes stem from the geometry of concept representations rather than from any particular algorithm. Formalizing concepts as activation-space regions, we prove that the overlap between a target and other concepts lower-bounds the damage any robust erasure must inflict on them, with the trade-off scaling linearly in the degree of overlap. Across thirteen unlearning methods, including methods designed to preserve non-target concepts, no method achieves both strong erasure and strong neighbor preservation: STEREO nearly eliminates indirect leakage but cuts neighbor generation by more than 75\%, while sparse inference-time methods preserve neighbors but leak. Damage increases with our overlap measure, monotonically so for STEREO; the $\kappa$-scaling reproduces on SDXL, and neighbor-selective damage recurs on FLUX. Perfect unlearning is the wrong target for entangled concepts; methods should be evaluated on the Pareto frontier our theorem establishes.
Chinese Translation
文本到图像的扩散模型中的概念遗忘旨在抑制目标概念（例如， exttt{horse}），同时保留相关但不同的内容（例如， exttt{donkey}），然而现有方法要么在间接提示下泄漏，要么明显降低其他概念的质量。我们表明，这些失败模式源于概念表示的几何特性，而不是任何特定算法。通过将概念形式化为激活空间区域，我们证明了目标概念与其他概念之间的重叠下限了任何稳健的遗忘对它们造成的损害，并且这种权衡在重叠程度上呈线性缩放。在包括旨在保留非目标概念的方法在内的十三种遗忘方法中，没有一种方法能够同时实现强有力的遗忘和强有力的邻居保留：STEREO几乎消除了间接泄漏，但将邻居生成减少了超过75\%；而稀疏推理时间方法则保留了邻居但发生泄漏。损害随着我们的重叠度量增加而增加，对于STEREO而言是单调的；$ extkappa$-缩放在SDXL上得以重现，邻居选择性损害在FLUX上再次出现。对于纠缠的概念，完美遗忘并不是正确的目标；方法应根据我们定理所建立的帕累托前沿进行评估。
cs.AI / 282 / 2609.34139
Same Tasks, Different Apps: Why Mobile GUI Agents Fail to Generalize?
相同任务，不同应用：为何移动图形用户界面代理无法实现泛化？
Abstract
Mobile GUI agents deployed in real settings must work across different applications that support the same functionality. Most existing benchmarks test each task in only one app, so a high score can mean the agent understands the task, or only that it knows that particular app. We introduce AnyAppBench, a category-controlled live Android benchmark that evaluates cross-application generalization while keeping the user goal fixed. It spans 10 functional categories, 100 task templates, and 520 task--application pairs over 52 applications. Agents run from raw instructions and with app-independent sub-goals, and a VLM judge labels every failed run under a fixed failure taxonomy whose reliability is measured by human annotation. We find that, across 13 agents, success on the original application does not transfer reliably to new applications with the same goal. Furthermore, providing high-level sub-goal decomposition produces only small, category-dependent changes that do not close the gap, and the mix of failure types changes with the target interface. Based on those insights, we believe the AnyAppBench benchmark provides an important stepping stone toward robust real-world deployment of mobile GUI agents. Our code, data and the leaderboard can be found at the project website https://anyappbench.github.io/.
Chinese Translation
在实际环境中部署的移动图形用户界面代理必须能够在支持相同功能的不同应用之间工作。现有的大多数基准测试仅在一个应用中测试每个任务，因此高分可能意味着代理理解了任务，或者仅仅是它了解那个特定的应用。我们引入了 AnyAppBench，这是一个类别控制的实时 Android 基准测试，评估跨应用泛化，同时保持用户目标不变。它涵盖了 10 个功能类别、100 个任务模板和 520 对任务-应用组合，涉及 52 个应用。代理从原始指令和与应用无关的子目标运行，VLM 判别器对每个失败的运行进行标记，使用固定的失败分类法，其可靠性通过人工标注进行测量。我们发现，在 13 个代理中，原始应用上的成功并不能可靠地转移到具有相同目标的新应用。此外，提供高层次的子目标分解仅产生微小的、依赖于类别的变化，无法缩小差距，并且失败类型的组合会随着目标界面的变化而变化。基于这些见解，我们认为 AnyAppBench 基准测试为移动图形用户界面代理在现实世界中的稳健部署提供了一个重要的基石。我们的代码、数据和排行榜可以在项目网站 https://anyappbench.github.io/ 找到。
cs.AI / 283 / 2609.34151
Self-Evolving Agents via Likelihood-Guided Tool-Space Optimization
通过似然引导的工具空间优化实现自我进化代理
Abstract
Self-evolving agents can continually improve their behavior, while tools define the executable action space through which they interact with the environment. However, exposing the full tool library to model introduces substantial irrelevant context and can impair tool-use decisions. We study tool-space self-evolution, where each recurring task type maintains a persistent tool space which is constructed from accumulated output experience. We identify three limitations of existing methods: (1) output-unaware selection: they rely primarily on tool descriptions or model priors rather than observed tool outputs; (2) statelessness across request: they select tools independently for each request without consolidating prior output experience into persistent task-specific state; (3) inference cost: they repeatedly search, rank, or reason over candidate tools for subsequent requests of the same task. We address these limitations through output-aware tool scoring, persistent task-specific tool spaces, amortized tool selection, and reusable configurations across models. We introduce LOTS (Likelihood-Only Tool Scoring), which evolves an agent's tool space from accumulated output experience while keeping model parameters fixed. After each request, LOTS holds the model's generated answer and estimates each tool's contribution by measuring how much the answer likelihood changes when its observed output is removed. These contributions are aggregated within each recurring task to rank tools and update its persistent space. Across three benchmarks, LOTS improves task performance while substantially reducing tool context. More importantly, sequential experiments demonstrate that task-specific spaces persist and continue to improve over time, while cross-model experiments show that learned configurations transfer across different models.
Chinese Translation
自我进化代理能够持续改善其行为，而工具定义了它们与环境互动的可执行动作空间。然而，将完整的工具库暴露给模型会引入大量无关的上下文，并可能影响工具使用决策。我们研究了工具空间的自我进化，其中每种重复的任务类型维持一个持久的工具空间，该空间由累积的输出经验构建。我们识别出现有方法的三个局限性：（1）输出无关选择：它们主要依赖于工具描述或模型先验，而不是观察到的工具输出；（2）请求间无状态性：它们为每个请求独立选择工具，而没有将先前的输出经验整合到持久的任务特定状态中；（3）推理成本：它们在后续相同任务的请求中反复搜索、排名或推理候选工具。我们通过输出感知的工具评分、持久的任务特定工具空间、摊销的工具选择和跨模型的可重用配置来解决这些局限性。我们引入了LOTS（仅似然工具评分），它在保持模型参数固定的情况下，从累积的输出经验中进化代理的工具空间。在每个请求之后，LOTS保持模型生成的答案，并通过测量去除其观察到的输出时答案的似然变化来估计每个工具的贡献。这些贡献在每个重复任务中汇总，以对工具进行排名并更新其持久空间。在三个基准测试中，LOTS提高了任务性能，同时显著减少了工具上下文。更重要的是，顺序实验表明，任务特定空间持续存在并随着时间的推移不断改善，而跨模型实验则表明，学习到的配置可以在不同模型之间迁移。
cs.AI / 284 / 2609.34157
TableSeek: Structure-Preserving Agentic Evidence Seeking over Heterogeneous Table Corpora
TableSeek：在异构表格语料库中进行结构保留的主动证据搜索
Abstract
Open-domain table retrieval seeks tables that contain sufficient evidence for answering a question or verifying a claim. Yet semantic relevance is often misleading: topically similar tables may lack the required facts, while answer-bearing evidence is often confined to a few cells whose meaning depends on surrounding schema and table context. Heterogeneous schemas, value formats, and serializations further weaken one-shot matching. We present TableSeek, a structure-preserving agentic search framework for heterogeneous table corpora. Instead of ranking tables once, an LLM agent iteratively follows sparse clues, inspects schema-preserving previews, identifies schema- and value-level mismatches, and refines its investigation. TableSeek uses cells and schemas as evidence anchors while retaining complete tables as evidence units, enabling fine-grained localization without losing the context required for interpretation and answerability checking. Without relying on retriever training or a precomputed semantic index, TableSeek produces transparent evidence-seeking trajectories and achieves competitive end-to-end performance against strong retrieval-and-reranking pipelines on heterogeneous table benchmarks. These results suggest that active, structure-preserving evidence seeking is a promising paradigm for open-domain table retrieval.
Chinese Translation
开放领域表格检索旨在寻找包含足够证据以回答问题或验证主张的表格。然而，语义相关性往往具有误导性：主题相似的表格可能缺乏所需的事实，而包含答案的证据通常局限于几个单元格，其含义依赖于周围的模式和表格上下文。异构模式、值格式和序列化进一步削弱了一次性匹配的效果。我们提出了TableSeek，一种针对异构表格语料库的结构保留主动搜索框架。TableSeek并非一次性对表格进行排名，而是通过一个大型语言模型（LLM）代理迭代地跟随稀疏线索，检查保留模式的预览，识别模式和数值层面的不匹配，并细化其调查。TableSeek使用单元格和模式作为证据锚点，同时保留完整的表格作为证据单元，从而实现细粒度的定位，而不失去进行解释和答案可检验所需的上下文。在不依赖检索器训练或预计算语义索引的情况下，TableSeek生成透明的证据搜索轨迹，并在异构表格基准测试中与强大的检索和重排名管道相比，达到了具有竞争力的端到端性能。这些结果表明，主动的、结构保留的证据搜索是开放领域表格检索的一个有前景的范式。
cs.AI / 285 / 2609.34160
RoutePrism: Tracing Construction Order Effects in Agent Memory
RoutePrism：追踪代理记忆中的构建顺序效应
Abstract
Processing the same records in a different order can discard different evidence, yet endpoint accuracy alone cannot reveal what changed or whether it mattered. We introduce RoutePrism, a diagnostic protocol that builds memory twice from the same source pool in two processing orders, then traces which sources, compiled contexts, and answers differ. Because record content, timestamps, policy, and the answer model all stay fixed, any observed difference is localized to the memory construction step. A matched four-condition intervention tests whether a record displaced by reordering actually carried task-relevant evidence: restoring that single record recovers over 60 percentage points of lost accuracy, while substituting a non-supporting record of equal length does not. We evaluate the protocol on PersonaMem-32K (63 primary queries, 29 users) and 470 LongMemEval-S questions with histories spanning 38 to 62 sessions, replicating the core intervention across five answer models. Survivor selection, defined as the choice of which record a cluster retains, drives most source-level changes, while different memory policies (compaction, bounded recency, MemoChat-style summarization, A-MEM) produce distinct failure signatures at the source, context, and metadata layers.
Chinese Translation
以不同顺序处理相同记录可能会丢失不同的证据，然而仅凭端点准确性无法揭示变化的内容或其重要性。我们提出了RoutePrism，一种诊断协议，它从相同的源池中以两种处理顺序构建记忆，然后追踪哪些源、编译的上下文和答案存在差异。由于记录内容、时间戳、政策和答案模型均保持不变，任何观察到的差异都局限于记忆构建步骤。一个匹配的四条件干预测试了被重新排序替换的记录是否确实携带了与任务相关的证据：恢复那一条记录可以恢复超过60个百分点的丢失准确性，而替换为一条长度相等但不支持的记录则没有效果。我们在PersonaMem-32K（63个主要查询，29个用户）和470个LongMemEval-S问题上评估该协议，历史跨度为38到62个会话，在五个答案模型中复制核心干预。生存者选择，即集群保留哪个记录的选择，驱动了大多数源级变化，而不同的记忆政策（压缩、有限近期性、MemoChat风格的总结、A-MEM）在源、上下文和元数据层面产生了不同的失败特征。
cs.AI / 286 / 2609.34177
ReplayLens: Auditing Agents' Use of Outcomes
ReplayLens：审计智能体对结果的使用
Abstract
When an agent reuses logged experience, a changed decision may reflect the recorded score, the action's name, or the record's position in storage. Standard memory evaluations do not reveal which relationship drives that change. We introduce ReplayLens, a black-box audit that changes one relationship in the stored history at a time, holds the remaining interface fixed, and measures the resulting decision. Four interventions target four relationships. Outcome reassignment swaps which scores belong to which actions. Pair transport moves intact action-score pairs to new record slots. Consistent renaming relabels actions in both history and menu. Key-slot reassignment changes both score attachment and position. A constructive separation shows why the audit is needed: two memory writers with identical endpoint accuracy respond differently to the same replay, so conventional evaluation cannot resolve the underlying dependence. On black-box LLM interfaces, swapping scores changes decisions while moving intact pairs does not, separating score attachment from record order. A bounded-memory study exposes ingestion-order sensitivity that endpoint comparison misses. In sequential experiment planning, altered historical scores redirect exploration and reduce final utility despite fresh measurements. A code-debugging agent with sealed hidden tests shows the same pattern outside model selection. ReplayLens provides a relationship-level audit for deciding whether logged experience can be merged, reordered, or reindexed safely.
Chinese Translation
当智能体重用记录的经验时，改变的决策可能反映出记录的分数、动作的名称或记录在存储中的位置。标准的记忆评估无法揭示驱动这种变化的具体关系。我们引入了ReplayLens，一种黑箱审计工具，它一次性改变存储历史中的一种关系，保持其余接口不变，并测量由此产生的决策。四种干预措施针对四种关系。结果重新分配交换了哪些分数属于哪些动作。配对传输将完整的动作-分数对移动到新的记录槽。一致重命名在历史和菜单中重新标记动作。关键槽重新分配则同时改变分数的附加和位置。构造性的分离展示了审计的必要性：两个具有相同终点准确性的记忆写入者对相同的重放做出不同反应，因此传统评估无法解决潜在的依赖关系。在黑箱大型语言模型（LLM）接口上，交换分数会改变决策，而移动完整的配对则不会，从而将分数附加与记录顺序分开。一项有限记忆研究揭示了摄取顺序的敏感性，而终点比较则未能捕捉到。在顺序实验规划中，改变的历史分数会重新引导探索并降低最终效用，尽管有新的测量。一个具有封闭隐藏测试的代码调试智能体在模型选择之外显示出相同的模式。ReplayLens提供了一种关系级别的审计，用于决定记录的经验是否可以安全地合并、重新排序或重新索引。
cs.AI / 287 / 2609.34179
RAGWarrant: Evidence-Preserving Governance for RAG Policy Promotion Under Quality, Cost, Latency, and Risk Constraints
RAGWarrant：在质量、成本、延迟和风险约束下的RAG政策推广证据保留治理
Abstract
Retrieval-augmented generation systems are extensively instrumented with metrics, benchmarks, traces, and automated judges, but these tools do not decide whether a proposed policy change is safe to release. We present RAGWarrant, an open-source promotion-control framework that treats deployment as a constrained evidence decision rather than a leaderboard choice. RAGWarrant normalizes evaluator outputs and operational telemetry, applies predeclared quality and hard-risk gates, assigns evidence-class claim ceilings, preserves negative outcomes, and emits auditable PROMOTE, BLOCK, REJECT, or INCONCLUSIVE decisions. We evaluate the framework across T2-RAGBench, MultiHop-RAG, CRAG, HotpotQA, synthetic reproduction, and bounded local generative experiments. On HotpotQA, operational savings were blocked because answer quality fell beyond the declared margin. A bounded CRAG study selected a lower-cost quality-tied policy, but related generative gains were unstable and a held-out guardrail failed closed. We claim an auditable promotion-control abstraction, not optimizer superiority, human validation, or production readiness. The tagged artifact reproduces from a fresh clone, runs as a hardened Docker job, accepts external evaluator exports, and verifies artifact integrity.
Chinese Translation
检索增强生成系统广泛配备了指标、基准、跟踪和自动评估工具，但这些工具并不能决定所提议的政策变更是否安全发布。我们提出了RAGWarrant，一个开源的推广控制框架，将部署视为一个受限的证据决策，而不是排行榜选择。RAGWarrant标准化评估者输出和操作遥测，应用预先声明的质量和硬风险门，分配证据类别声明上限，保留负面结果，并发出可审计的PROMOTE（推广）、BLOCK（阻止）、REJECT（拒绝）或INCONCLUSIVE（不确定）决策。我们在T2-RAGBench、MultiHop-RAG、CRAG、HotpotQA、合成再现和有界本地生成实验中评估了该框架。在HotpotQA上，操作节省被阻止，因为答案质量超出了声明的边际。在一个有界的CRAG研究中，选择了一个低成本的质量绑定政策，但相关的生成收益不稳定，并且保持的护栏未能关闭。我们声称这是一个可审计的推广控制抽象，而非优化器的优越性、人类验证或生产准备。该标记的工件可以从新的克隆中再现，作为一个强化的Docker作业运行，接受外部评估者导出，并验证工件的完整性。
cs.AI / 288 / 2609.34180
Decision Readouts for Text-Mediated Video Anomaly Detection: An Exploratory Evaluation of Jev and Qwen
基于文本媒介的视频异常检测的决策输出：对Jev和Qwen的探索性评估
Abstract
How much does the decision readout matter when video-derived textual evidence is held fixed? We evaluate Jev typed decisions and three Qwen readouts on a sparse development sample of 40 videos and 400 target anchors from UCF-Crime and XD-Violence, each presented as a summary and ordered captions. Each dataset contributes 20 source groups and 200 anchors, including only 10 and 37 positives, respectively. The original five-backend pilot requested 4,000 predictions; Jev Choice returned 776 valid responses out of 800 under the study's strict numerical policy, blocking its full-coverage quality comparison. On XD captions, Jev Noul achieved 75.99% average precision versus 48.47% for Qwen generated probability and 57.81% for the stronger local ordinal-likelihood expectation. The latter paired difference was 18.18 percentage points (95% source-group bootstrap interval 5.53-31.50). UCF did not show a corresponding advantage: caption ROC-AUC was 52.26% for Noul and 65.95% for ordinal likelihood. Both probability readouts had higher, hence worse, UCF Brier scores than the evaluation-prevalence reference of 0.0475. We additionally audit historical LAVAD scores at exactly matched anchors and distinguish response structure from numerical consistency. A binary-likelihood control is missing. These exploratory offline results characterize ranking, probability quality and interface failures; they establish neither a causal typed-interface benefit nor general superiority, calibration or end-to-end acceleration.
Chinese Translation
当视频衍生的文本证据固定时，决策输出的重要性有多大？我们在一个稀疏的开发样本中评估了Jev类型的决策和三个Qwen输出，样本包含40个视频和400个目标锚点，分别来自UCF-Crime和XD-Violence，每个视频都以摘要和有序字幕的形式呈现。每个数据集贡献了20个源组和200个锚点，分别仅包含10个和37个正例。原始的五后端试点请求了4,000个预测；在研究的严格数值政策下，Jev Choice返回了800个中776个有效响应，阻碍了其全面覆盖质量的比较。在XD字幕上，Jev Noul的平均精度达到了75.99%，而Qwen生成的概率为48.47%，更强的局部序数似然期望为57.81%。后者的配对差异为18.18个百分点（95%源组自助区间5.53-31.50）。UCF没有显示出相应的优势：Noul的字幕ROC-AUC为52.26%，而序数似然为65.95%。这两个概率输出的UCF Brier分数均高于评估流行度参考值0.0475，因此表现更差。我们还审计了在完全匹配的锚点上的历史LAVAD分数，并区分了响应结构与数值一致性。缺少二元似然控制。这些探索性离线结果表征了排名、概率质量和接口故障；它们既未确立因果类型接口的优势，也未显示出一般的优越性、校准或端到端加速。
cs.AI / 289 / 2609.34181
Efficient Reasoning via Constrained Optimization in Latent Space
通过潜在空间中的约束优化实现高效推理
Abstract
Large Reasoning Models (LRMs) have shown remarkable reasoning capabilities, yet they still suffer from overthinking, generating redundant reasoning steps which incur substantial token consumption. Existing methods, such as suppressing reflective keywords or forcing shorter reasoning lengths, attempt to mitigate this issue but inevitably truncate necessary steps and induce underthinking, thereby compromising performance. To address this dilemma, we investigate the latent representations and observe that efficient reasoning steps naturally cluster into a concentrated region in latent space, while those deviating from this region tend to produce verbose sequences. To leverage this, we keep reasoning focused within this region via a quadratic program which projects deviating hidden states back into the region. Then we propose a novel training-free framework to achieve efficient reasoning that reduces token generation costs without sacrificing performance. Extensive experiments conducted on four models ranging from 1.5B to 14B, and across six benchmarks in math reasoning, coding, and scientific QA, validate the effectiveness of our method, up to a 12.1\% improvement in accuracy while reducing generated tokens by 11.8\% to 52.8\%. Codes are available at \href{https://github.com/hzn18/Opt4Reasoning}{https://github.com/hzn18/Opt4Reasoning}.
Chinese Translation
大型推理模型（Large Reasoning Models, LRM）展现了卓越的推理能力，但仍然面临过度思考的问题，生成冗余的推理步骤，从而导致大量的令牌消耗。现有的方法，如抑制反思关键词或强制缩短推理长度，试图缓解这一问题，但不可避免地截断必要步骤，导致思维不足，从而影响性能。为了解决这一困境，我们研究了潜在表示，观察到高效的推理步骤自然聚集在潜在空间的一个集中区域，而偏离该区域的步骤往往会产生冗长的序列。为了利用这一点，我们通过一个二次规划将偏离的隐藏状态投影回该区域，从而保持推理集中在该区域内。随后，我们提出了一种新颖的无训练框架，以实现高效推理，降低令牌生成成本而不牺牲性能。在对四个模型（从1.5B到14B）和六个数学推理、编码及科学问答基准进行的大量实验中，验证了我们方法的有效性，准确率提高了最多12.1\%，同时生成的令牌减少了11.8 ext{%}至52.8 ext{%}。代码可在 exttt{https://github.com/hzn18/Opt4Reasoning} 获取。
cs.AI / 290 / 2609.34184
CASS: Contribution-Aware Structured Sparsity for Model Merging
CASS：基于贡献感知的结构稀疏性用于模型合并
Abstract
Model merging integrates task-specific fine-tuned models into a single multi-task model, but often suffers from parameter interference caused by conflicting task-vector updates. Existing methods typically mitigate conflicts by pruning task vectors based on weight magnitude or random heuristics, treating Transformers as unstructured ``bags of parameters'' and overlooking their inherent modularity. In this paper, we propose \textbf{C}ontribution-\textbf{A}ware \textbf{S}tructured \textbf{S}parsity (CASS), a unified framework that reduces parameter interference by identifying and preserving task-specific components. At the core of CASS is a contribution-aware structured mask that identifies task-relevant attention heads and FFN neurons. We instantiate this mask in two settings: CASS-Merging, the primary post-hoc setting where masks serve as a plug-and-play denoising filter for existing merging operators, and CASS-Tuning, an extension for scenarios with fine-tuning access where masks constrain gradients to reduce structural overlap between task vectors. Our analysis shows that task-relevant components are sparse and partially disjoint, supporting structured component-level filtering as an effective way to reduce merging interference. Extensive experiments across vision (ViT, 20 tasks) and language (RoBERTa, 8 tasks; Qwen2.5, 4 tasks) benchmarks demonstrate that CASS improves a range of representative merging baselines.
Chinese Translation
模型合并将特定任务的微调模型整合为一个单一的多任务模型，但常常受到由于任务向量更新冲突而导致的参数干扰。现有方法通常通过基于权重大小或随机启发式的方式修剪任务向量来减轻冲突，视变换器（Transformers）为无结构的“参数袋”，忽视了其固有的模块化特性。本文提出了基于贡献感知的结构稀疏性（Contribution-Aware Structured Sparsity，CASS），这是一个统一框架，通过识别和保留特定任务的组件来减少参数干扰。CASS的核心是一个贡献感知的结构掩码，它识别与任务相关的注意力头和前馈神经元（FFN neurons）。我们在两个设置中实例化该掩码：CASS-Merging，主要的后处理设置，其中掩码作为现有合并操作的即插即用去噪滤波器，以及CASS-Tuning，一个扩展设置，适用于可以进行微调的场景，在该场景中掩码限制梯度以减少任务向量之间的结构重叠。我们的分析表明，与任务相关的组件是稀疏且部分不相交的，支持结构化组件级过滤作为减少合并干扰的有效方式。在视觉（ViT，20个任务）和语言（RoBERTa，8个任务；Qwen2.5，4个任务）基准上的广泛实验表明，CASS改善了一系列代表性的合并基线。
cs.AI / 291 / 2609.34195
PainterBench: A Figural Divergent-Thinking Benchmark for Tool-Using Language Models
PainterBench：一种针对工具使用语言模型的形象发散思维基准测试
Abstract
Figural divergent thinking is the ability to develop a given shape fragment into an original drawing. In humans, this ability is assessed with incomplete-drawing tasks. We introduce PainterBench, a benchmark that ports the incomplete-drawing task to the agentic setting. The agent draws on a canvas through tool calls and observes the result after every turn. The canvas includes a starting shape which cannot be erased, and the agent's goal is to incorporate this shape into the most original drawing it can produce. The task is open-ended, and the agent itself decides when the drawing is finished. The benchmark tests incremental visual planning over a short horizon and the transfer of creative ability from pretraining to multi-turn tool use. We evaluate 14 multimodal language models from small to frontier scale. Across the primary study and six sensitivity analyses, we collect 2,700 drawings and crowdsource creativity and recognizability ratings for every drawing and for 300 human reference drawings. We also present ViDrA-adapted, an automated scorer that predicts human creativity ratings of agent drawings (r = 0.85 on random held-out test split). Figural divergent thinking varies widely across the 14 models, and GPT-6 Astra produces the most creative drawings. Relative to the human drawings, the agent drawings score higher in creativity but lower in recognizability. We release the final drawings, per-round canvas snapshots, tool call traces, stimulus bank, benchmark harness, crowdsourced ratings (N = 72,000), and ViDrA checkpoint.
Chinese Translation
形象发散思维是将给定的形状片段发展为原创绘画的能力。在人类中，这种能力通过不完整绘画任务进行评估。我们引入了PainterBench，一个将不完整绘画任务移植到代理设置中的基准测试。代理通过工具调用在画布上绘画，并在每个回合后观察结果。画布包括一个无法擦除的起始形状，代理的目标是将这个形状融入到它能够创作出的最原创的绘画中。该任务是开放式的，代理自行决定何时完成绘画。该基准测试考察了短期内的增量视觉规划以及创造能力从预训练到多轮工具使用的转移。我们评估了14种从小型到前沿规模的多模态语言模型。在主要研究和六项敏感性分析中，我们收集了2700幅绘画，并为每幅绘画和300幅人类参考绘画众包创造性和可识别性评分。我们还展示了ViDrA-adapted，一个自动评分器，预测代理绘画的人类创造性评分（在随机保留测试分割上r = 0.85）。14种模型的形象发散思维差异很大，GPT-6 Astra产生了最具创造性的绘画。与人类绘画相比，代理绘画在创造性上得分更高，但在可识别性上得分较低。我们发布了最终绘画、每轮画布快照、工具调用轨迹、刺激库、基准测试工具、众包评分（N = 72,000）以及ViDrA检查点。
cs.AI / 292 / 2609.34211
Behavior-Grounded Semantic Enrichment for Financial Fraud Modeling and Reasoning
基于行为的金融欺诈建模与推理的语义增强
Abstract
In financial fraud detection, rich semantic context can provide important evidence for transaction behavior modeling and fraud reasoning. However, public real-world financial datasets often lack rich semantics due to privacy constraints. Consequently, synthetic datasets incorporate generated semantics, but at the cost of behavioral realism; textual descriptions for contextual reasoning remain scarce. We address this gap through a semantic enrichment framework grounded in original transaction behavior to simulate multimodal financial data. We (1) propose a multi-agent semantic enrichment framework that generates interpretable financial semantics grounded in transaction behavior through role-specialized agents and consistency refinement, and (2) newly contribute a valuable multimodal financial fraud dataset, MS-FFSD, enriched with structured semantics and textual semantics while preserving real-data-grounded transaction behavior. Furthermore, we systematically analyze the quality and utility of semantic enrichment. Results demonstrate statistical fidelity and framework generalizability, while showing that richer semantics benefit fraud modeling and context-aware LLM reasoning. Overall, this work advances multimodal financial fraud research and bridges emerging LLM and multi-agent capabilities with operational anti-fraud practice. The framework and dataset are released at https://github.com/AI4Risk/MS-FFSD.
Chinese Translation
在金融欺诈检测中，丰富的语义上下文可以为交易行为建模和欺诈推理提供重要证据。然而，由于隐私限制，公共真实金融数据集往往缺乏丰富的语义。因此，合成数据集虽然引入了生成的语义，但却牺牲了行为的真实感；用于上下文推理的文本描述仍然稀缺。我们通过一个基于原始交易行为的语义增强框架来填补这一空白，以模拟多模态金融数据。我们 (1) 提出了一个多智能体语义增强框架，该框架通过角色专门化的智能体和一致性优化生成基于交易行为的可解释金融语义；(2) 新增了一个有价值的多模态金融欺诈数据集 MS-FFSD，该数据集在保留真实数据基础上的交易行为的同时，丰富了结构化语义和文本语义。此外，我们系统地分析了语义增强的质量和效用。结果表明，该框架在统计上具有保真性和通用性，同时显示出更丰富的语义有利于欺诈建模和上下文感知的 LLM 推理。总体而言，本研究推进了多模态金融欺诈研究，并将新兴的 LLM 和多智能体能力与实际反欺诈实践相结合。该框架和数据集已发布在 https://github.com/AI4Risk/MS-FFSD。
cs.AI / 293 / 2609.34214
GlyphBench: A Playground for Language-Model Reinforcement Learning
GlyphBench：语言模型强化学习的实验平台
Abstract
We introduce GlyphBench, an environment suite for reinforcement learning (RL) post-training of language-model agents, with over 360 tasks spanning diverse games. GlyphBench renders spatial observations as two-dimensional Unicode grids and connects training, evaluation, and trajectory replay through a unified interface designed to support efficient and reproducible research. We use GlyphBench to study how observation interfaces, reasoning effort, and agent harnesses affect performance, and how RL configurations shape learning dynamics. Our results show that glyph observations outperform native text and pixels in our Craftax experiments, with further gains on several BALROG environments. RL on 100 GlyphBench tasks improves Qwen3.5-4B on held-out Reasoning Gym problems, reaching 63.48% accuracy and outperforming the base model, a math-trained baseline, and a code-trained baseline. These experiments provide empirical evidence that reasoning gains from gameplay can yield stronger transfer than math or code. Together, these results highlight GlyphBench's value as a testbed for systematic research on how language-model agents learn, interact, and generalize.
Chinese Translation
我们介绍了GlyphBench，一个用于语言模型代理强化学习（RL）后训练的环境套件，涵盖超过360个多样化游戏任务。GlyphBench将空间观测呈现为二维Unicode网格，并通过统一接口连接训练、评估和轨迹重放，旨在支持高效且可重复的研究。我们利用GlyphBench研究观测接口、推理努力和代理工具如何影响性能，以及RL配置如何塑造学习动态。我们的结果表明，在Craftax实验中，glyph观测在性能上优于原生文本和像素，并在多个BALROG环境中进一步提升。对100个GlyphBench任务的强化学习使Qwen3.5-4B在保留的Reasoning Gym问题上达到了63.48%的准确率，超越了基础模型、数学训练基线和代码训练基线。这些实验提供了实证证据，表明游戏中的推理收益可以比数学或代码产生更强的迁移效果。这些结果共同突显了GlyphBench作为系统研究语言模型代理学习、互动和泛化的测试平台的价值。
cs.AI / 294 / 2609.34215
Same Winners, Different Success Rates: Evaluating How LLM Agents Recover from Failures
相同的赢家，不同的成功率：评估大型语言模型代理如何从失败中恢复
Abstract
Evaluating how LLM agents recover from mid-task failures is central to deploying reliable agentic systems. Existing checkpoint-based benchmarks measure recovery by comparing which action is selected as best across independent runs, a quantity known as set agreement. However, set agreement is a purely ordinal measure that records which action wins without reflecting the absolute level of performance. When all actions fail, they tie at zero reward, and independent runs produce the same tied set with high probability, creating an illusion of stability that masks near-zero recovery success. We formalize this limitation through a set-path symmetry result, proving that for equal-cost Bernoulli actions the success probabilities (0.9, 0.8) and (0.2, 0.1) yield identical best-action-set distributions at every sample size. No procedure based solely on which action wins can distinguish these two regimes. We further prove that certifying exact population ties is impossible in finite time, and that the assignment of outcomes to checkpoints carries information beyond marginal outcome distributions. The pooled success probability is the missing scalar that resolves the ordinal ambiguity. Experiments on 864 frozen RecoveryBench episodes and two planning cohorts totaling 3,456 responses confirm the theoretical predictions. Agreement and held-out quality can move in opposite directions, and permuting checkpoint-to-action bindings changes 8 to 13 percent of cell-level conclusions. Based on these findings, we propose reporting four diagnostic quantities (agreement, all-zero fraction, held-out success, and pooled success) that expose this failure mode with no additional data collection.
Chinese Translation
评估大型语言模型（LLM）代理如何从任务中期失败中恢复，对于部署可靠的代理系统至关重要。现有的基于检查点的基准通过比较在独立运行中选择的最佳行动来衡量恢复情况，这一量称为集合一致性。然而，集合一致性仅是一种序数度量，它记录了哪个行动获胜，却未反映绝对的性能水平。当所有行动均失败时，它们的奖励均为零，并且独立运行以高概率产生相同的平局集合，从而造成一种稳定性的错觉，掩盖了接近零的恢复成功率。我们通过集合路径对称性结果形式化了这一局限性，证明对于等成本的伯努利行动，成功概率（0.9, 0.8）和（0.2, 0.1）在每个样本大小下产生相同的最佳行动集合分布。仅基于哪个行动获胜的程序无法区分这两种情况。我们进一步证明，在有限时间内认证精确的总体平局是不可能的，并且结果分配到检查点携带的信息超出了边际结果分布。汇总成功概率是解决序数模糊性的缺失标量。对864个冻结的RecoveryBench剧集和两个规划组（总计3,456个响应）的实验验证了理论预测。协议和保留质量可以朝相反的方向变化，检查点与行动绑定的排列改变了8%到13%的单元级结论。基于这些发现，我们建议报告四个诊断量（协议、全零比例、保留成功和汇总成功），以揭示这种失败模式，而无需额外的数据收集。
cs.AI / 295 / 2609.34227
When Does Selection Replace Extraction? A Pre-Registered Test of Agent Memory with a Typed Decision Model
选择何时取代提取？基于类型决策模型的代理记忆预注册测试
Abstract
Does conversational memory need LLM-extracted facts, or is selecting the right raw turns enough? Published results disagree. Extraction-based systems report gains from distilled facts. Recent studies find raw history with good ranking does as well, but disagree about whether ranking matters. We ran a pre-registered study on held-out LoCoMo conversations and LongMemEval. At a tight budget on LoCoMo, raw turns selected by a single call to Jev, a typed decision model, are non-inferior to an LLM-extraction memory (one-sided 95% bound -3.0 points against a -5-point margin). Blind human grading narrows the margin but does not change the result. Raw turns cost 3,061 times less to write, and the result holds with a second answer model. Within this study, reranking's gain shrinks as the budget grows. It adds 17.4 points on LoCoMo and 9.1 on LongMemEval when three of 30 candidates are kept. At generous budgets it adds 1.5 and 1.1, and extraction systems are more accurate. This suggests why published results disagree. At matched context, Jev selects as accurately as an LLM reranker (non-inferiority bound -2.0) at a third of the latency, and more accurately than a multi-call graph traversal. Reranking lowers correct abstention. Plans, code and graded answers are released.
Chinese Translation
对话记忆是否需要从大型语言模型（LLM）提取的事实，或者选择正确的原始对话回合是否足够？已发布的结果存在分歧。基于提取的系统报告从提炼的事实中获得收益。最近的研究发现，具有良好排名的原始历史同样有效，但对排名是否重要存在分歧。我们在保留的LoCoMo对话和LongMemEval上进行了预注册研究。在LoCoMo的紧张预算下，通过对Jev（一个类型决策模型）的单次调用选择的原始对话回合在性能上不逊色于LLM提取记忆（单边95%界限为-3.0分，相较于-5分的差距）。盲人评分缩小了差距，但未改变结果。撰写原始对话回合的成本低至3,061倍，且这一结果在第二个答案模型中依然成立。在本研究中，随着预算的增加，重新排名的收益减小。当保留30个候选中的3个时，在LoCoMo上增加了17.4分，在LongMemEval上增加了9.1分。在宽松预算下，分别增加了1.5分和1.1分，而提取系统的准确性更高。这解释了为何已发布的结果存在分歧。在匹配的上下文中，Jev的选择准确性与LLM重新排名器相当（非劣界限为-2.0），且延迟仅为其三分之一，准确性也高于多次调用的图遍历。重新排名降低了正确放弃的情况。计划、代码和评分答案已发布。
cs.AI / 296 / 2609.34241
AdaGuard: An Adaptive Guard Model with User-defined Policies
AdaGuard：一种具有用户定义策略的自适应防护模型
Abstract
Guard models support the safe deployment of language model agents, but fixed risk taxonomies limit their ability to accommodate requirements that vary across applications and tasks. Under user-defined policies, detecting violations requires interpreting both the applicable rules and the agent's behavior, since identical actions can receive different judgments under different policies. To support learning this capability, we introduce AdaptiveSafety, a dataset of 10,939 training examples and 1,000 test examples covering policies with 1--100 rules. The dataset combines trajectories from multiple sources with policy and behavioral counterfactuals, pairing each example with an explanation and the complete set of violated rules. These counterfactuals expose changes that alter compliance, while structural augmentations provide supervision for consistency under rule reordering and identifier remapping. Building on this supervision, we propose SafePO, a reinforcement learning algorithm for refining violation identification while balancing explanatory reasoning and final verdicts. SafePO uses structured rewards to assess prediction correctness, retains group-relative advantages at the response level, and employs a separately trained value model to modulate token weights within explanation and verdict regions. Separate normalization controls their relative contribution to training despite differences in length. Through supervised initialization followed by SafePO, we develop AdaGuard, a family of 0.6B, 4B, and 8B guard models that assess agent trajectories under policies supplied at inference time. Our 4B model achieves binary accuracies of 89.30\% on AdaptiveSafety and 71.82\% on DynaBench. The project repository is available at https://github.com/Yunhao-Feng/AdaGuard
Chinese Translation
防护模型支持语言模型代理的安全部署，但固定的风险分类限制了其适应不同应用和任务需求的能力。在用户定义的策略下，检测违规行为需要解释适用的规则和代理的行为，因为相同的行为在不同的政策下可能会得到不同的判断。为了支持学习这一能力，我们引入了AdaptiveSafety，一个包含10,939个训练示例和1,000个测试示例的数据集，涵盖1至100条规则的政策。该数据集结合了来自多个来源的轨迹与政策和行为的反事实，将每个示例与解释和完整的违规规则集合配对。这些反事实揭示了改变合规性的变化，而结构性增强则为规则重排序和标识符重映射下的一致性提供监督。在此监督的基础上，我们提出了SafePO，一种强化学习算法，用于在平衡解释推理和最终裁决的同时优化违规识别。SafePO使用结构化奖励来评估预测的正确性，保持响应级别的组相对优势，并利用单独训练的价值模型来调节解释和裁决区域内的标记权重。尽管长度不同，单独的归一化控制了它们对训练的相对贡献。通过监督初始化后跟随SafePO，我们开发了AdaGuard，一个包含0.6B、4B和8B防护模型的系列，这些模型在推理时根据提供的政策评估代理轨迹。我们的4B模型在AdaptiveSafety上达到了89.30%的二元准确率，在DynaBench上达到了71.82%。项目代码库可在https://github.com/Yunhao-Feng/AdaGuard获取。
cs.AI / 297 / 2609.34242
Stashbird: Efficient Speaker-Indexed Memory for Conversational Agents
Stashbird：高效的说话者索引记忆系统用于对话代理
Abstract
AI agents require memory that preserves information across user-agent exchanges, user-to-user conversations, and group conversations with or without agent participation, while supporting updates as evidence changes or is removed. We present Stashbird, an agent memory system that links source episodes to derived memory state through explicit provenance. Stashbird organizes memory into episodic records, semantic relations, community summaries, and persisted graph state, with lifecycle operations for incremental updates and episode-level deletion. We evaluate question-answering accuracy and model-facing workload across four long-term memory benchmarks. On LoCoMo, Stashbird uses 76.4x fewer ingestion prompt tokens than Graphiti. Compared with reproduced Hindsight on the same benchmark, it uses 8.1x fewer retrieval prompt tokens, with accuracy 1.6 percentage points lower. It achieves higher accuracy than Hindsight on LongMemEval-S and GroupMemBench and comparable accuracy on EverMemBench.
Chinese Translation
人工智能代理需要一种记忆系统，以在用户与代理之间的交流、用户之间的对话以及有或没有代理参与的群体对话中保留信息，同时支持在证据变化或被移除时进行更新。我们提出了Stashbird，一种通过显式来源链接源事件与派生记忆状态的代理记忆系统。Stashbird将记忆组织为情节记录、语义关系、社区摘要和持久图状态，并提供增量更新和事件级删除的生命周期操作。我们在四个长期记忆基准上评估了问答准确性和模型面临的工作负载。在LoCoMo上，Stashbird使用的摄取提示令牌比Graphiti少76.4倍。与在同一基准上重现的Hindsight相比，它使用的检索提示令牌少8.1倍，准确性低1.6个百分点。它在LongMemEval-S和GroupMemBench上实现了比Hindsight更高的准确性，并在EverMemBench上达到了可比的准确性。
cs.AI / 298 / 2609.34249
Evolving Support Priorities in Empathetic Reinforcement Learning
在同理心强化学习中演变的支持优先级
Abstract
We identify a fundamental mismatch in empathetic reinforcement learning: support priorities evolve with the dialogue state, yet existing methods typically optimize predefined reward specifications that remain fixed across turns. To model these evolving support priorities, we organize empathetic support along cognitive, affective, and proactive empathy, and propose Context-Adaptive Rubric Evolution (CARE). At each turn, CARE generates a context-adaptive rubric by adjusting both the weights of these three empathy dimensions and their fine-grained evaluation criteria. The rubric generator is trained with turn-level rubric supervision and human preference data through supervised fine-tuning followed by preference-based reinforcement learning, and then serves as an adaptive reward interface for online empathetic RL. Integrated with both RLVER and MICA, CARE achieves state-of-the-art performance across SentientBench, EQBench3, and EMPA under three independent LLM judges. Notably, on EMPA, CARE improves EPM-Idx over the strongest baseline by at least 13 points under all three judges, including an increase from 28.11 to 83.54 under Gemini-2.5-Pro. Further analyses show that learned rubric priorities systematically vary across dialogue stages and user emotions, demonstrating that CARE adapts what is rewarded as support needs evolve.
Chinese Translation
我们识别出同理心强化学习中的一个基本不匹配：支持优先级随着对话状态的变化而演变，但现有方法通常优化固定的预定义奖励规范，这些规范在多个回合中保持不变。为了建模这些演变的支持优先级，我们将同理心支持组织为认知、情感和主动同理心，并提出了上下文自适应评分演变（Context-Adaptive Rubric Evolution, CARE）。在每个回合中，CARE通过调整这三个同理心维度的权重及其细化的评估标准，生成一个上下文自适应的评分标准。评分标准生成器通过回合级评分监督和人类偏好数据进行监督微调，然后通过基于偏好的强化学习进行训练，最终作为在线同理心强化学习的自适应奖励接口。与RLVER和MICA相结合，CARE在SentientBench、EQBench3和EMPA三个独立的LLM评审下实现了最先进的性能。值得注意的是，在EMPA上，CARE在所有三个评审下将EPM-Idx提高了至少13分，包括在Gemini-2.5-Pro下从28.11提高到83.54。进一步分析表明，学习到的评分优先级在对话阶段和用户情感之间系统性地变化，证明了CARE能够随着支持需求的演变而调整奖励内容。
cs.AI / 299 / 2609.34259
QuantaSpike: Short-Window Spike-Driven Quantization for Large Language Models
QuantaSpike：用于大型语言模型的短窗口脉冲驱动量化
Abstract
Large language models (LLMs) achieve strong performance across many tasks but rely on dense multiply-accumulate (MAC) operations during inference, resulting in high energy cost. Spiking neural networks (SNNs) offer an event-driven alternative in which synaptic integration uses lightweight accumulation. However, spike-driven LLM inference remains difficult because outlier-heavy activations typically require long firing windows or auxiliary non-spiking paths. We propose QuantaSpike, a short-window spike-driven quantization framework for LLMs built around Logarithmic Ternary Integrate-and-Fire (LTIF) neurons. LTIF uses ternary events with power-of-two membrane-response quanta, improving the information represented by each firing step while retaining shift-ACC-compatible computation. QuantaSpike combines this neuron with group-adaptive gain and selective outlier admission: normal values use residual LTIF steps, whereas admitted outliers receive one additional onset spike before entering the same residual dynamics. Across OPT and Llama-2, QuantaSpike achieves state-of-the-art or competitive perplexity and zero-shot accuracy among spike-driven LLM quantization methods. It also transfers to newer dense LLMs, remaining close to the FP16 reference on Llama-3-8B and Qwen3-8B under the same four-step firing window. Analytical linear-energy projections show that QuantaSpike reduces the energy of one linear transformation by about $80.0\%$ on OPT models and $67.1\%$ on Llama-2 models relative to SpikeQuant, providing an accurate and energy-efficient spike-driven path for LLM inference.
Chinese Translation
大型语言模型（LLMs）在许多任务中表现出色，但在推理过程中依赖于密集的乘加（MAC）操作，导致高能耗。脉冲神经网络（SNNs）提供了一种事件驱动的替代方案，其中突触整合使用轻量级累积。然而，脉冲驱动的LLM推理仍然困难，因为重异常值的激活通常需要较长的发放窗口或辅助的非脉冲路径。我们提出了QuantaSpike，一种基于对数三元积分-发放（Logarithmic Ternary Integrate-and-Fire, LTIF）神经元的短窗口脉冲驱动量化框架。LTIF使用具有二次幂膜响应量子（membrane-response quanta）的三元事件，提高了每个发放步骤所表示的信息，同时保留了与移位ACC兼容的计算。QuantaSpike将该神经元与组自适应增益和选择性异常值接纳相结合：正常值使用残差LTIF步骤，而被接纳的异常值在进入相同的残差动态之前会接收到一个额外的起始脉冲。在OPT和Llama-2上，QuantaSpike在脉冲驱动的LLM量化方法中实现了最先进或具有竞争力的困惑度和零-shot准确性。它还可以迁移到更新的密集LLM，在相同的四步发放窗口下，保持与Llama-3-8B和Qwen3-8B的FP16参考接近。分析线性能量投影表明，QuantaSpike相对于SpikeQuant在OPT模型上将一个线性变换的能量减少了约80.0%，在Llama-2模型上减少了67.1%，为LLM推理提供了一条准确且高效的脉冲驱动路径。
cs.AI / 300 / 2609.34262
Maintaining Benchmarks Against Increasingly Capable Agents: Detection and Remediation of Unearned Passes
对日益强大的智能体维持基准：检测和修复不当通过
Abstract
Agentic benchmarks guide model selection and training. Yet an agent can pass a task without demonstrating the intended capability. Such outcomes constitute unearned passes; their proportion among all passes defines the integrity gap. As agents improve, benchmark surfaces that once seemed harmless can become exploitable, making benchmark validity an ongoing maintenance problem. We introduce a process-verification framework that audits passing trajectories, distinguishes evidenced reward hacking from verifier weakness, and localizes exploitable surfaces for repair. Across 3,810 passing trajectories from 29 model-benchmark cohorts, confirmed violations often increase with model generation but not monotonically. On SWEBench Pro V1.0, confirmed violation rates rise from 24% to 73% between Opus 4.7 and Fable 5 on matched tasks; later cohorts fall to 11% for Fable 5.1 and 0% for GPT-6 Astra. These comparisons are descriptive: configurations were not normalized, and the latest models also pass fewer exploitable tasks. Violations concentrate around a small set of recurring surfaces, especially unintended access to reference solutions through git history. Three repair case studies across two benchmarks show why blocking a recorded exploit is insufficient: the same protected information can remain accessible through another route. Therefore, we combine minimal patches with exploit replay and fresh agent evaluation, auditing new passes under the original standard. No evaluated attempt against the final patches reached the protected channel, and every post-patch pass was judged legitimate. Benchmark integrity requires ongoing maintenance: audit passing behavior, repair the enabling surface, and re-evaluate both exploit access and legitimate solvability.
Chinese Translation
智能体基准指导模型选择和训练。然而，智能体可以在未展示预期能力的情况下通过任务。这种结果构成了不当通过；所有通过中不当通过的比例定义了完整性差距。随着智能体的改进，曾经看似无害的基准表面可能变得可被利用，使基准的有效性成为一个持续的维护问题。我们引入了一个过程验证框架，审计通过轨迹，区分证据奖励黑客行为与验证者弱点，并定位可修复的可利用表面。在来自29个模型-基准组的3,810条通过轨迹中，确认的违规行为通常随着模型代数的增加而增加，但并非单调。在SWEBench Pro V1.0上，确认的违规率在Opus 4.7和Fable 5之间从24%上升至73%；而后续组在Fable 5.1时降至11%，在GPT-6 Astra时降至0%。这些比较是描述性的：配置未被标准化，且最新模型也通过了更少的可利用任务。违规行为集中在一小组重复出现的表面上，特别是通过git历史意外访问参考解决方案。针对两个基准的三个修复案例研究显示，阻止记录的利用是不够的：相同的受保护信息可能通过另一条途径保持可访问。因此，我们将最小补丁与利用重放和新智能体评估相结合，在原始标准下审计新的通过。没有评估的尝试能够通过最终补丁访问受保护通道，且每个补丁后的通过均被判定为合法。基准的完整性需要持续维护：审计通过行为，修复使能表面，并重新评估利用访问和合法可解性。
cs.CL / 1 / 2609.31629
ChestPheNoT: Deployable, Auditable Label-Status-Evidence Extraction from Radiology Reports
ChestPheNoT：可部署的、可审计的放射学报告标签状态证据提取
Abstract
Structured phenotype extraction from radiology reports supports cohort construction, quality auditing, and clinical analytics, but practical deployment requires local inference and auditable predictions, while expert annotations remain scarce. Conventional labelers provide structured findings and assertion states but no supporting evidence, while API-hosted large language models may be unsuitable when clinical text cannot leave institutional infrastructure. We present CHESTPHENOT, a compact 0.5-3B language model that jointly extracts finding labels, three-class status (present/absent/uncertain), and verbatim supporting evidence spans. CHESTPHENOT is trained using hybrid CheXbert+72B silver supervision followed by supervised fine-tuning and lightweight GRPO refinement. Across three human-annotated gold sets spanning in-distribution, cross-taxonomy, and cross-institution evaluation, the 3B model remains below its CheXbert silver teacher in distribution but is competitive under distribution shift, significantly surpassing CheXbert on cross-institution detection (+2.0 F1). Task-specific training also enables the 3B model to match or exceed substantially larger prompted models on most detection and status comparisons. For evidence-grounded extraction, over 99% of final evidence spans are locatable in the source report, and the 3B model achieves 47.5 auditable-F1, outperforming Qwen2.5-7B one-shot prompting by 7.6 points and approaching Qwen2.5-72B. These results demonstrate that locally deployable models can provide competitive and directly auditable radiology-report extraction without relying on external inference APIs. Code and the full extraction/judge prompts will be made available at https://github.com/yukkai/ChestPheNoT.
Chinese Translation
从放射学报告中提取结构化表型支持队列构建、质量审计和临床分析，但实际部署需要本地推理和可审计的预测，而专家注释仍然稀缺。传统标签器提供结构化发现和断言状态，但没有支持证据，而托管在API上的大型语言模型在临床文本无法离开机构基础设施时可能不适用。我们提出了CHESTPHENOT，这是一个紧凑的0.5-3B语言模型，能够联合提取发现标签、三类状态（存在/缺失/不确定）和逐字支持证据范围。CHESTPHENOT采用混合CheXbert+72B银监督进行训练，随后进行监督微调和轻量级GRPO优化。在三个经过人工注释的金标准数据集上进行评估，涵盖分布内、跨分类和跨机构的评估，3B模型在分布上低于其CheXbert银教师，但在分布转移下具有竞争力，在跨机构检测上显著超过CheXbert（+2.0 F1）。任务特定训练还使得3B模型在大多数检测和状态比较中能够匹配或超越显著更大的提示模型。对于基于证据的提取，最终证据范围中超过99%可以在源报告中定位，3B模型实现了47.5的可审计F1，超越了Qwen2.5-7B的一次性提示7.6分，并接近Qwen2.5-72B。这些结果表明，本地可部署模型能够提供具有竞争力且直接可审计的放射学报告提取，而无需依赖外部推理API。代码和完整的提取/评估提示将发布在https://github.com/yukkai/ChestPheNoT。
cs.CL / 2 / 2609.31650
A literature-guided descriptor-based framework for filtering composition search spaces
基于文献指导的描述符框架用于过滤成分搜索空间
Abstract
Scientific literature contains latent knowledge about materials behavior, but much of this knowledge is expressed through words, contexts, and recurring associations rather than explicit design principles. This raises a central question: how can large-scale scientific corpora be used for practical problems in materials discovery? Here, we present a literature-guided descriptor-based filtering framework for reducing composition search spaces. For a given performance metric, the framework selects two descriptors from a filtered vocabulary in a literature-trained word embedding model and uses the selected descriptors to construct a Pareto-based filter for candidate compositions. Across the evaluated performance metrics and composition search spaces, the framework filters out an average of 74.27\% of the candidate compositions, with an average best-value error of 1.93\% relative to experimental measurements. Compared with expert-chosen and random descriptors, our performance metric-dependent descriptors provide a more controlled balance between retained fraction and best-value error. These results show that literature-derived embeddings can support intuitive and reproducible filters for narrowing candidate composition spaces while preserving high-performing compositions.
Chinese Translation
科学文献中蕴含着关于材料行为的潜在知识，但这些知识往往通过文字、上下文和反复出现的关联表达，而非明确的设计原则。这引发了一个核心问题：如何将大规模科学语料库应用于材料发现的实际问题？在此，我们提出了一种基于文献指导的描述符过滤框架，用于减少成分搜索空间。对于给定的性能指标，该框架从经过文献训练的词嵌入模型中筛选出两个描述符，并利用所选描述符构建一个基于帕累托（Pareto）的候选成分过滤器。在评估的性能指标和成分搜索空间中，该框架平均过滤掉74.27%的候选成分，且相对于实验测量的平均最佳值误差为1.93%。与专家选择的和随机选择的描述符相比，我们的性能指标依赖描述符在保留比例和最佳值误差之间提供了更为可控的平衡。这些结果表明，基于文献的嵌入可以支持直观且可重复的过滤器，以缩小候选成分空间，同时保留高性能成分。
cs.CL / 3 / 2609.31653
Distributional sentiment modeling and anomaly detection for consumer complaint assessment
消费者投诉评估的分布式情感建模与异常检测
Abstract
Sentiment analysis is a common tool for converting unstructured text into quantitative signals in finance and risk management. Yet most applications reduce the output to a discrete polarity label or a single predictive feature, overlooking the distributional structure of sentiment intensity in consumer complaint narratives. In this paper we treat negative sentiment in consumer complaints as a bounded continuous variable and study its full distribution rather than a single label. We score each narrative with a transformer classifier, model the scores with Beta distributions, and compare the fitted distributions of meritorious and non-meritorious complaints through the Kullback Leibler divergence and the squared Hellinger distance. The fitted distributions are then linked with dollar amounts and company response outcomes to construct anomaly diagnostics that flag complaints whose textual severity is inconsistent with the recorded relief. We find that the two groups have strongly overlapping distributions, so negative sentiment intensity is not a sharp classifier of outcomes on its own; combined with monetary and categorical attributes, it isolates unusually severe complaints for operational risk monitoring. Treating sentiment analysis as continuous distributional measurement, this study links sentiment extraction, bounded response modeling, and anomaly detection in a unified framework for consumer complaint assessment.
Chinese Translation
情感分析是将非结构化文本转化为定量信号的常用工具，广泛应用于金融和风险管理。然而，大多数应用将输出简化为离散的极性标签或单一的预测特征，忽视了消费者投诉叙述中情感强度的分布结构。本文将消费者投诉中的负面情感视为一个有界的连续变量，研究其完整分布，而非单一标签。我们使用变换器分类器对每个叙述进行评分，利用Beta分布对评分进行建模，并通过Kullback-Leibler散度和平方Hellinger距离比较优质与非优质投诉的拟合分布。然后，将拟合的分布与金额和公司响应结果关联，以构建异常诊断，标记那些文本严重性与记录的救济不一致的投诉。我们发现这两组的分布高度重叠，因此负面情感强度本身并不能作为结果的明确分类器；结合货币和类别属性，它能够隔离出异常严重的投诉，以便进行操作风险监控。将情感分析视为连续分布测量，本研究将情感提取、有界响应建模和异常检测整合为一个统一的框架，用于消费者投诉评估。
cs.CL / 4 / 2609.31660
Parser, Chunking, and Embedding Interactions in Retrieval-Augmented Generation over Indian Government Regulatory Documents
在印度政府监管文件中的检索增强生成中的解析器、分块和嵌入交互
Abstract
Retrieval-augmented generation (RAG) pipelines are typically assembled from independently-chosen components -- a document parser, a chunking strategy, and an embedding model -- yet these choices are rarely evaluated jointly, and evaluations that do combine them are usually run on a single document or corpus. We present a controlled factorial study of 3 parsers, 3 chunking strategies, and 5 dense embedding models, together with a sparse BM25 baseline, evaluated against 800 question instances, each with one or more required evidence strings, with evidence strings automatically validated against source text and a 10% random sample manually reviewed, across four structurally distinct Indian central-government regulatory documents. We fit linear mixed-effects models with document-query-level random intercepts to the resulting 72,000-row result set, run Holm-corrected paired comparisons between matched dense and sparse configurations, and report clustered bootstrap confidence intervals for all 54 unique retriever configurations. We find that no single retriever family dominates across documents; parser and chunker choice interact significantly; MPNet-base is a consistent underperformer with a severe failure mode on table-derived questions; and the corpus exhibits a near-saturated evidence-preservation ceiling above 98%, indicating that retrieval differences are driven primarily by ranking quality rather than information loss during ingestion. We additionally report embedding-dimension and chunk-size/overlap ablations and an efficiency/quality Pareto analysis. We release our full evaluation harness, corpus manifest, and 800-question benchmark.
Chinese Translation
检索增强生成（RAG）管道通常由独立选择的组件组装而成——文档解析器、分块策略和嵌入模型——然而这些选择很少被联合评估，而那些确实结合它们的评估通常是在单一文档或语料库上进行的。我们提出了一项受控的因子研究，涉及3种解析器、3种分块策略和5种密集嵌入模型，以及一个稀疏的BM25基线，评估了800个问题实例，每个实例都有一个或多个所需的证据字符串，这些证据字符串通过源文本自动验证，并且10%的随机样本经过人工审核，涵盖了四个结构上不同的印度中央政府监管文件。我们对结果集中的72,000行数据拟合了线性混合效应模型，文档-查询级别的随机截距，并在匹配的密集和稀疏配置之间进行了Holm校正的配对比较，报告了所有54种独特检索器配置的聚类自助法置信区间。我们发现没有单一的检索器家族在所有文档中占主导地位；解析器和分块器的选择显著相互作用；MPNet-base在表格派生问题上表现不佳，存在严重的失效模式；而语料库展示了超过98%的证据保留的近饱和上限，表明检索差异主要由排名质量驱动，而非在摄取过程中信息丢失。我们还报告了嵌入维度和分块大小/重叠的消融实验以及效率/质量的帕累托分析。我们发布了完整的评估工具、语料库清单和800个问题基准。
cs.CL / 5 / 2609.31663
LLM-Guided Ontology-Driven Knowledge Graph Construction from Unstructured Text
基于大语言模型指导的本体驱动知识图谱构建：从非结构化文本出发
Abstract
Ontology-driven knowledge graph construction from industrial text remains challenging due to the domain specificity of documents, the scarcity of annotated resources, and the complexity of ontology engineering workflows. This paper presents and investigates the applicability of an ontology learning pipeline that combines compact open-source Large Language Models (LLMs), reusable prompting strategies, and open knowledge bases to support the extraction, structuring, enrichment, and evaluation of knowledge from textual corpora. The approach is tested and evaluated on a private French corpus of power-grid incident reports, using locally deployable open-source LLMs ranging from 7B to 32B parameters. Starting from unstructured reports, the approach extracts entities and relations, generates RDF triples, constructs related OWL ontology, enriches it using external knowledge sources, assesses the quality of the ontology, and subsequently constructs a populated knowledge graph grounded in the resulting ontology schema. Experiments on 80 manually annotated private reports show that schema-guided prompting significantly improves extraction quality, while quantized models provide an effective trade-off between performance and computational cost. These results demonstrate the feasibility of transforming domain-specific industrial text into ontology-based knowledge graphs using locally deployed open-source LLMs, while supporting the generalization of the extraction process through reusable prompting strategies.
Chinese Translation
从工业文本构建本体驱动的知识图谱仍然面临挑战，这主要是由于文档的领域特异性、标注资源的稀缺以及本体工程工作流程的复杂性。本文提出并研究了一种本体学习流程的适用性，该流程结合了紧凑型开源大语言模型（LLMs）、可重用的提示策略和开放知识库，以支持从文本语料中提取、结构化、丰富和评估知识。该方法在一个私人法语电网事件报告语料库上进行了测试和评估，使用了从7B到32B参数的本地可部署开源LLM。从非结构化报告开始，该方法提取实体和关系，生成RDF三元组，构建相关的OWL本体，利用外部知识源对其进行丰富，评估本体的质量，并随后构建一个基于生成的本体模式的填充知识图谱。在80份手动标注的私人报告上的实验表明，基于模式的提示显著提高了提取质量，而量化模型在性能和计算成本之间提供了有效的权衡。这些结果证明了使用本地部署的开源LLM将领域特定的工业文本转化为基于本体的知识图谱的可行性，同时通过可重用的提示策略支持提取过程的泛化。
cs.CL / 6 / 2609.31664
What Drives Dialectal Jailbreaks? An Ablation of Surface Form, Cultural Framing, and Strategy Banks
是什么驱动方言突破？表面形式、文化框架和策略库的消融研究
Abstract
Recent work suggests that obscure language registers can weaken large language model refusal behavior, especially when paired with black-box prompt optimization. It remains unclear whether failures stem from non-standard surface form, culturally grounded framing, or optimization over an expressive prompt-strategy space. We study Chinese registers by extending a classical-Chinese red-teaming framework to Shanghainese and Cantonese and running a 36-cell ablation across surface forms, strategy-bank variants, and two target models. The main finding is corrective: dialectal surface form is neither necessary nor sufficient for high attack success. Non-optimized English, Mandarin, and naive dialect translations remain below 8\% attack success rate, whereas all conditions that retain an optimizer-controlled strategy bank reach 98--100\%. A culture-neutral generic strategy bank reaches the same ceiling at near-single-query cost, further indicating that strategy-bank expressiveness, rather than dialectal or cultural content, accounts for most of the observed effect. Dialect choice still affects query efficiency and response severity, and qualitative coding identifies recurring high-level mechanisms such as semantic glossing and procedural scaffolding. We qualify the absolute ceiling-level attack success rate by showing sensitivity to judge calibration and noting that the design does not fully separate one-shot strategy construction from iterative search.
Chinese Translation
近期研究表明，晦涩的语言注册可以削弱大型语言模型的拒绝行为，尤其是在与黑箱提示优化结合时。目前尚不清楚失败是源于非标准的表面形式、文化根植的框架，还是在表达性提示策略空间上的优化。我们通过将经典中文红队框架扩展到上海话和粤语，研究中文注册，并在表面形式、策略库变体和两个目标模型之间进行36个单元的消融实验。主要发现是纠正性的：方言表面形式既不是高攻击成功率的必要条件，也不是充分条件。未经优化的英语、普通话和幼稚的方言翻译的攻击成功率仍低于8%，而所有保留优化器控制的策略库的条件均达到98-100%的成功率。一个文化中立的通用策略库在接近单次查询成本的情况下也达到了相同的上限，进一步表明策略库的表现力，而非方言或文化内容，解释了观察到的大部分效果。方言选择仍然影响查询效率和响应严重性，定性编码识别出重复出现的高层机制，如语义注释和程序性支架。我们通过展示对评估者校准的敏感性，并指出设计并未完全将一次性策略构建与迭代搜索分离，来限定绝对上限攻击成功率。
cs.CL / 7 / 2609.31666
Age-Adaptive Handwriting Reconstruction from an IMU-Based Digital Pen through Shared Representations and Domain-Specific Heads
基于IMU的数字笔的年龄自适应手写重建：通过共享表示和领域特定头部
Abstract
Digital pens are widely used to capture handwriting on digital devices, enabling precise trace recording and enhancing human-computer interaction. However, most are bundled with tablets and lack cross-brand compatibility. Recent digital pens equipped with kinematic sensors have emerged, designed for use on any surface. This especially opens significant potential for supporting handwriting acquisition in classrooms. Handwriting reconstruction from such an IMU-equipped pen poses a challenge due to the significant variability in sensor signals between adults and children. Even when producing visually similar traces, variations in writing dynamics, motor control, pen holding, and user confidence introduce substantial discrepancies in the captured signals. Additionally, the high variability in children's handwriting requires collecting large amounts of data, which is not feasible to implement in a school environment at scale. Furthermore, models trained exclusively on adult data fail to generalize to children's handwriting, and conversely, models trained on children's data perform poorly on adult writers. This crosspopulation degradation highlights the need for a unified model that can be deployed directly on the pen, without any user-specific adaptation. To address this issue, we propose a cross-domain learning, using an original neural network architecture based on a Temporal Convolutional Network and multiple prediction heads. The model is designed to be robust across age groups by leveraging shared features while effectively handling variability induced by differences in graphomotor development. This approach aims to improve handwriting trace reconstruction from sensor data, where each domain benefits from additional data provided by the other domain.
Chinese Translation
数字笔广泛用于在数字设备上捕捉手写，能够精确记录轨迹并增强人机交互。然而，大多数数字笔与平板电脑捆绑在一起，缺乏跨品牌兼容性。最近出现的配备运动传感器的数字笔旨在在任何表面上使用，这为课堂手写获取提供了重要潜力。然而，从这种配备IMU的笔中重建手写面临挑战，因为成人和儿童之间传感器信号存在显著变异。即使在产生视觉上相似的轨迹时，书写动态、运动控制、握笔方式和用户信心的差异也会导致捕获信号的显著差异。此外，儿童手写的高变异性要求收集大量数据，这在学校环境中大规模实施并不可行。此外，专门在成人数据上训练的模型无法推广到儿童手写，而在儿童数据上训练的模型在成人书写者身上表现不佳。这种跨人群的性能下降突显了需要一个统一的模型，该模型可以直接在笔上部署，而无需任何用户特定的适应。为了解决这个问题，我们提出了一种跨领域学习的方法，使用基于时间卷积网络和多个预测头的原始神经网络架构。该模型旨在通过利用共享特征来增强跨年龄组的鲁棒性，同时有效处理由于书写运动发展差异引起的变异性。该方法旨在改善从传感器数据中重建手写轨迹，其中每个领域都能从另一个领域提供的额外数据中受益。
cs.CL / 8 / 2609.31676
An Evaluation of AI-Supported Evidence-Based Learning for Public Speaking Skill Development
基于人工智能的证据支持学习在公共演讲技能发展中的评估
Abstract
Public speaking is an essential skill in academic and professional contexts, but it often causes anxiety. Although several AI-based speech coaching tools exist, they typically provide generic feedback and lack model speeches for learning. This study addresses these gaps by developing a system that incorporates speech context into evaluation criteria, enabling more tailored feedback. It also generates improved model speeches in both text and audio formats to support example-based learning. The system was evaluated in a pilot study with seven undergraduate students over one to three weeks. Participants used the system repeatedly and practiced the same speech at least three times. Results showed improvements in public speaking performance for all participants. Filler word usage decreased, and anxiety levels dropped by 15.3% to 34.4%. Participants found the context-aware feedback and revised speeches useful and confidence-building. These findings suggest that context-aware feedback and model speeches can enhance public speaking skills and reduce anxiety. Future studies could explore video-based analysis of physical behavior during speeches.
Chinese Translation
公共演讲是学术和职业环境中一项重要技能，但常常引发焦虑。尽管存在几种基于人工智能的演讲辅导工具，但它们通常提供通用反馈，缺乏用于学习的示范演讲。本研究通过开发一个将演讲背景纳入评估标准的系统来填补这些空白，从而实现更具针对性的反馈。该系统还生成改进的示范演讲，提供文本和音频格式，以支持基于示例的学习。该系统在为期一到三周的试点研究中进行了评估，参与者为七名本科生。参与者反复使用该系统，并至少练习同一演讲三次。结果显示，所有参与者的公共演讲表现均有所改善。填充词的使用减少，焦虑水平下降了15.3%至34.4%。参与者认为上下文感知反馈和修订后的演讲有助于提升信心。这些发现表明，上下文感知反馈和示范演讲可以增强公共演讲技能并减少焦虑。未来的研究可以探讨演讲过程中身体行为的视频分析。
cs.CL / 9 / 2609.31684
The Temporal Tug-of-War: Visualizing and Detecting RAG Conflicts in Diffusion Models via Trajectory Variance
时间上的拉锯战：通过轨迹方差可视化和检测扩散模型中的检索增强生成（RAG）冲突
Abstract
Retrieval-Augmented Generation (RAG) introduces a specific failure mode in discrete diffusion language models: when retrieved context contradicts parametric knowledge, the iterative denoising process becomes a visible battleground between competing knowledge sources. We identify temporal semantic divergence as an observable for detecting these conflicts and introduce the Trajectory Variance Score (TVS), a simple and interpretable measure of this divergence. TVS computes the mean pairwise cosine distance of answer embeddings across independent stochastic denoising trajectories, capturing the temporal tug of war between parametric and contextual attractors. Requiring as few as two parallel inference runs, TVS is computationally lightweight. Across four diverse datasets (Synthetic, SciQ, PopQA, and CounterFact), a simple Logistic Regression classifier using TVS achieves $70.10\%$ accuracy and $0.7647$ AUROC on LLaDA. On Dream 7B, increasing the number of trajectories from two to five improves accuracy from $63.91\%$ to $69.62\%$. More complex sequential models provide only marginal improvements over the linear classifier. Evaluation across LLaDA and Dream 7B demonstrates that conflict-induced trajectory dynamics and their key properties transfer across distinct diffusion architectures.
Chinese Translation
检索增强生成（RAG）在离散扩散语言模型中引入了一种特定的失败模式：当检索到的上下文与参数知识相矛盾时，迭代去噪过程成为竞争知识源之间的明显战场。我们将时间语义偏差识别为检测这些冲突的可观察指标，并引入了轨迹方差分数（Trajectory Variance Score, TVS），这是一种简单且可解释的偏差度量。TVS计算独立随机去噪轨迹中答案嵌入的平均成对余弦距离，捕捉参数吸引子与上下文吸引子之间的时间拉锯战。TVS只需两个并行推理运行，计算开销较小。在四个不同的数据集（Synthetic、SciQ、PopQA和CounterFact）上，使用TVS的简单逻辑回归分类器在LLaDA上达到了70.10%的准确率和0.7647的AUROC。在Dream 7B上，将轨迹数量从两个增加到五个使准确率从63.91%提高到69.62%。更复杂的序列模型仅对线性分类器提供了边际改进。在LLaDA和Dream 7B上的评估表明，冲突引起的轨迹动态及其关键属性在不同的扩散架构之间是可以转移的。
cs.CL / 10 / 2609.31687
Verification of PETSc with CIVL using LLM-generated ACSL contracts and deterministic driver generation
使用LLM生成的ACSL合约和确定性驱动程序生成对PETSc与CIVL的验证
Abstract
Parallel numerical libraries such as PETSc are widely used in science and engineering applications where wrong results can have costly consequences. Despite this, numerical libraries are rarely formally verified. One of the challenges is the need for an expert to hand-write a specification and manually apply a verification tool, often requiring the development of a harness or driver, all of which can contain additional bugs that lead to false positives or false negatives during verification. With recent advancements in large language models (LLMs), it is tempting to generate such drivers and reference models automatically, but one-shot generation based on a simple prompt is brittle and leads to additional unverified code that needs to be audited. In this paper, we present an approach to use LLMs in a limited setting to generate a small, human-certifiable ACSL contract from the function's documentation, combined with a deterministic toolchain that supports a restricted ACSL profile and generates a driver that uses the CIVL verifier to check the implementation against the contract and, when available, an existing reference model. We demonstrate the pipeline end-to-end on three PETSc functions: MatAXPY (reference model already exists), MatAYPX (no reference model, so the certified contract is the sole oracle), and the non-compressing mode of MatFilter (no reference model, with more complex, conditional behavior). With this pipeline, we were able to discover a bug in PETSc's MatAYPX function that was previously undiscovered and had been present in the code since 1997.
Chinese Translation
并行数值库如PETSc在科学和工程应用中被广泛使用，而错误的结果可能会导致高昂的后果。尽管如此，数值库的正式验证却很少。其挑战之一在于需要专家手动编写规范并手动应用验证工具，这通常需要开发一个测试框架或驱动程序，而这些都可能包含额外的错误，从而导致验证过程中的假阳性或假阴性。随着大型语言模型（LLMs）的最新进展，自动生成此类驱动程序和参考模型的想法变得诱人，但基于简单提示的一次性生成往往脆弱，并导致需要审计的额外未验证代码。在本文中，我们提出了一种在有限环境中使用LLMs的方法，从函数文档生成一个小型、可由人类验证的ACSL合约，并结合一个支持受限ACSL配置文件的确定性工具链，生成一个使用CIVL验证器检查实现与合约的一致性，并在可用时与现有参考模型进行对比的驱动程序。我们在三个PETSc函数上展示了这一端到端的流程：MatAXPY（已有参考模型）、MatAYPX（没有参考模型，因此认证合约是唯一的oracle）以及MatFilter的非压缩模式（没有参考模型，具有更复杂的条件行为）。通过这一流程，我们发现了PETSc的MatAYPX函数中一个之前未被发现的错误，该错误自1997年以来一直存在于代码中。
cs.CL / 11 / 2609.31688
Don't Repeat Yourself: Self-Supervised Fine-Tuning for Coverage
避免重复：自监督微调以提高覆盖率
Abstract
In verifiable domains such as math and coding, finding one correct solution among many attempts can matter more than the pass rate of each attempt. Post-training can concentrate large language model outputs around a few modes, while increasing sampling temperature has limited effectiveness. We introduce Don't Repeat Yourself Supervised Fine-Tuning (DRY-SFT), a post-training method that increases output diversity and coverage: the probability of at least one correct solution among many attempts. DRY-SFT has two stages. First, for each problem, sequentially generate K solutions, showing the model all prior attempts and asking for a different solution. Second, fine-tune on each attempt independently, removing prior attempts from the context. The process uses no reward, verifier, or correctness filter. On HumanEval+, MBPP+, and DS-1000, DRY-SFT raises pass@100 by 10.8, 12.5, and 12.4 percentage points, respectively, at a small cost to pass@1. Structural diversity, measured by abstract syntax tree edit distance among passing solutions, rises significantly on all three benchmarks. DRY-SFT also solves 244 of 600 problems that the base model did not solve in the same 200 attempts. Across nine open-weight models, lower structural diversity of the base model significantly predicts larger DRY-SFT gains, indicating that the method is especially effective on more mode-collapsed models.
Chinese Translation
在可验证的领域，如数学和编程，找到多个尝试中的一个正确解决方案可能比每个尝试的通过率更为重要。后训练可以使大型语言模型的输出集中在少数模式上，而提高采样温度的效果有限。我们提出了避免重复监督微调（Don't Repeat Yourself Supervised Fine-Tuning，DRY-SFT），这是一种后训练方法，旨在增加输出的多样性和覆盖率：即在多个尝试中至少有一个正确解决方案的概率。DRY-SFT 包含两个阶段。首先，对于每个问题，依次生成 K 个解决方案，向模型展示所有先前的尝试并要求提供不同的解决方案。其次，独立地对每个尝试进行微调，从上下文中移除先前的尝试。该过程不使用奖励、验证器或正确性过滤器。在 HumanEval+、MBPP+ 和 DS-1000 上，DRY-SFT 分别将 pass@100 提高了 10.8、12.5 和 12.4 个百分点，而对 pass@1 的影响较小。通过合格解决方案之间的抽象语法树编辑距离测量的结构多样性在这三个基准上显著提高。DRY-SFT 还解决了基础模型在同样的 200 次尝试中未能解决的 600 个问题中的 244 个。在九个开放权重模型中，基础模型的较低结构多样性显著预测了更大的 DRY-SFT 增益，表明该方法在模式崩溃的模型上特别有效。
cs.CL / 12 / 2609.31727
The Ongiini-Eval-OW Benchmark: A Concept Paper for the Planned Benchmarking of Machine Translation and Large Language Models on Oshindonga and Oshikwanyama
Ongiini-Eval-OW基准：关于计划在Oshindonga和Oshikwanyama上对机器翻译和大型语言模型进行基准测试的概念论文
Abstract
Oshiwambo -- a cluster of mutually intelligible Bantu languages spoken by over a million people across northern Namibia and southern Angola, and the home language of roughly half of Namibian households -- has, to our knowledge, no published machine-translation evaluation benchmark. Major commercial services (Google Translate, DeepL, Microsoft Translator), open multilingual MT models (NLLB-200, MADLAD-400), and the open Masakhane checkpoint collection all lack coverage of either standardised dialect, Oshindonga or Oshikwanyama. We announce Ongiini-Eval-OW, a planned 600-item English-Oshindonga and English-Oshikwanyama benchmark with native-speaker references from two independent translators, a 30-item inter-translator agreement set, an 11-tag phenomenon-tagged stratification (at least 30 items per tag), a deterministic 30% blind split, and a reproducible scoring protocol over chrF++, BLEU, and COMET-22, supplemented by a 50-item human-evaluation round. We document the empirical coverage gap, the dataset composition, the launch-leaderboard model matrix across American, European, and Chinese frontier and open-weight systems, and the contribution pipeline. The dataset is targeted for first public release in Q4 2026; this v1.0 concept paper announces the design and the call for participation. Data and code will be released under CC-BY-4.0 and MIT respectively.
Chinese Translation
Oshiwambo是一组相互可理解的班图语言，主要在纳米比亚北部和安哥拉南部由超过一百万人使用，是大约一半纳米比亚家庭的母语。根据我们的了解，目前尚无已发布的机器翻译评估基准。主要商业服务（如Google Translate、DeepL、Microsoft Translator）、开放的多语言机器翻译模型（如NLLB-200、MADLAD-400）以及开放的Masakhane检查点集合均未覆盖标准化方言Oshindonga或Oshikwanyama。我们宣布Ongiini-Eval-OW，这是一个计划中的600项英语-Oshindonga和英语-Oshikwanyama基准，包含来自两位独立翻译者的母语参考、30项翻译者间一致性测试集、11个现象标记的分层（每个标签至少30项）、一个确定性的30%盲分割，以及基于chrF++、BLEU和COMET-22的可重复评分协议，辅以50项人工评估轮次。我们记录了经验覆盖差距、数据集组成、美国、欧洲和中国前沿及开放权重系统的启动排行榜模型矩阵，以及贡献管道。该数据集计划于2026年第四季度首次公开发布；本版本1.0的概念论文宣布了设计方案并呼吁参与。数据和代码将分别在CC-BY-4.0和MIT许可下发布。
cs.CL / 13 / 2609.31847
Omni-IO Skills: Harnessing Your Agent Omni-Native
全能输入输出技能：利用您的智能体全能原生性
Abstract
General-purpose agents can plan, reason, and act over long horizons, yet their production capabilities remain fragmented across text, images, audio, video, documents, 3D assets, and code. Extending a foundation model to additional modalities ties capability growth to costly model updates, while assembling specialist models and tools leaves unresolved how procedures, dependencies, intermediate assets, and cross-turn revisions should be coordinated. We present Omni-IO Skills, a plug-and-play Agent Harness that makes existing agents omni-native through hierarchical Skills, a standardized multimodal execution interface, dependency-aware orchestration, and a persistent Asset Registry. Multi-asset workflows are represented as Declare Execution Graphs, which schedule independent operations concurrently and register successful outputs for downstream and cross-turn reuse across replaceable execution backends. Its 27 Skills cover 38 representative tasks spanning seven artifact modalities and four capability families: understanding, generation, reasoning, and retrieval. On UniM-90, the harness raises the input-support rates of GPT-5.6 Sol and Claude Sonnet 5 from 40.00% and 38.89% to 100%, while increasing relative Semantic--Quality Coupled Score from 26.99 to 74.94 and from 27.82 to 77.78, respectively; Strict Structure Score reaches 100.00 and 99.78. These results establish harness-level capability composition as a practical route to broad, evolvable Omni systems without changing the host agent's reasoning core.
Chinese Translation
通用智能体能够在长时间范围内进行规划、推理和行动，但其生产能力在文本、图像、音频、视频、文档、3D资产和代码等多个领域仍然是碎片化的。将基础模型扩展到额外的模态会将能力增长与昂贵的模型更新绑定在一起，而组装专业模型和工具则未能解决如何协调程序、依赖关系、中间资产和跨回合修订的问题。我们提出了全能输入输出技能（Omni-IO Skills），这是一种即插即用的智能体框架，通过层次化技能、标准化的多模态执行接口、依赖感知的编排和持久的资产注册表，使现有智能体具备全能原生性。多资产工作流被表示为声明执行图（Declare Execution Graphs），该图同时调度独立操作并注册成功的输出，以便在可替换的执行后端之间进行下游和跨回合重用。其27项技能涵盖了38个代表性任务，涉及七种工件模态和四个能力类别：理解、生成、推理和检索。在UniM-90上，该框架将GPT-5.6 Sol和Claude Sonnet 5的输入支持率从40.00%和38.89%提高到100%，同时相对语义质量耦合分数分别从26.99提高到74.94，从27.82提高到77.78；严格结构分数达到100.00和99.78。这些结果确立了框架级能力组合作为实现广泛、可进化的全能系统的实际途径，而无需改变宿主智能体的推理核心。
cs.CL / 14 / 2609.31876
Agents Can Use Base Models to Evade AI Detection
代理可以利用基础模型来规避人工智能检测
Abstract
We show that coding agents equipped with a base language model can successfully assemble responses from its samples to evade detection. Base models have been shown to evade commercial detectors, however, prior "humanization" techniques rely on using these models to paraphrase AI outputs over several iterations, which invariably results in semantic drift. In contrast, equipping coding agents to directly orchestrate the writing process by stitching text samples from a base model allows it to produce outputs that are coherent, task-specific and generally high quality. We find that Claude Opus 5 operating in a Claude Code harness effectively orchestrates a local 32B parameter OLMo-2 base LM and sacrifices little task accuracy across benchmarks spanning creative writing, factual grounding, health QA and instruction following, while using up to 90% base LM tokens. Responses constructed in this manner reduce the effectiveness of both post-hoc detectors (Pangram v4 detection rate drops from 77% to 24%) and soft watermarking applied a priori to the agent's generations (down to a simulated 10% detection at low FPR). While effective, this evasion requires a significantly larger number of input and output tokens from the agent, increasing the dollar cost per query up to 30x at API-pricing. Overall, this work demonstrates the effectiveness of a new class of adversarial attacks against AI text detection, and urges post-hoc detection providers to include outputs of base models in their training.
Chinese Translation
我们展示了配备基础语言模型的编码代理能够成功地从其样本中组装响应以规避检测。基础模型已被证明能够逃避商业检测器，然而，先前的“人性化”技术依赖于使用这些模型对人工智能输出进行多次迭代的释义，这不可避免地导致语义漂移。相反，直接让编码代理通过拼接基础模型的文本样本来协调写作过程，使其能够生成连贯、特定任务且通常质量较高的输出。我们发现，Claude Opus 5 在 Claude Code 环境中有效地协调了一个本地的 32B 参数 OLMo-2 基础语言模型，并在跨越创意写作、事实基础、健康问答和指令遵循的基准测试中几乎没有牺牲任务准确性，同时使用了高达 90% 的基础语言模型令牌。以这种方式构建的响应减少了事后检测器的有效性（Pangram v4 的检测率从 77% 降至 24%），以及应用于代理生成内容的软水印（在低假阳性率下模拟检测降至 10%）。尽管有效，这种规避需要代理提供显著更多的输入和输出令牌，导致每次查询的成本在 API 定价下增加高达 30 倍。总体而言，这项工作展示了一种新型对抗性攻击针对人工智能文本检测的有效性，并敦促事后检测提供者在其训练中包含基础模型的输出。
cs.CL / 15 / 2609.31956
Transformer MLP Gate Thresholds Are Couplings to a Carried Reference Direction
Abstract
The corpus-mean direction of a transformer's residual stream is a component shared across all inputs, and is commonly removed by mean-centering before representational analysis. We present evidence that it is a functional component: the reference against which the MLP gate population sets its operating point. In Phi-2, an exact decomposition of resting gate pre-activations shows that at mid-stack layers the resting inhibition of >99.9% of gates is carried by the coupling $w \cdot b_L$ to the carried mean direction, at 48-56$\times$ the explicit bias parameter, and the coupling is direction-specific: a random direction at matched norm orders the population's firing rates at Spearman $\rho \leq 0.09$ where the reference reaches 0.95. (That 0.95 is near-tautological on its own; the paper derives its null.) Causally, removing the stream's projection on the reference multiplies above-threshold firing by about 9$\times$, dose-monotonically, at 23-59$\times$ a norm-matched control; a random-initialised twin is flat, and replacement tests show that the direction carries the function and the magnitude does not. A direction-matched control makes the same point: noise injected along the reference costs 8-42$\times$ the same energy along a random direction. The decomposition replicates on four further families spanning both gate types (GELU with an explicit gate bias, bias-free SwiGLU), and the dose-response on two of them. Across an eight-model scan the mechanism is present in every GELU and SiLU family and absent only in OPT, where an opposing LayerNorm bias cancels the carried reference. Gate thresholds are implemented as couplings to a constant the network builds for the distribution it is reading, with the bias parameters contributing little; how much of that constant is carried in the stream and how much in parameters depends on the architecture.
cs.CL / 16 / 2609.31967
IndicFDB: Benchmarking Full-Duplex Voice Agents across Indian Languages
IndicFDB：跨印度语言的全双工语音代理基准测试
Abstract
Full-duplex voice agents must handle pauses, take turns, backchannel, and respond to user interruptions in real time. Full-Duplex-Bench evaluates these behaviors, but its English-only corpus and reliance on word-timestamped ASR and an English-prompted LLM judge make it difficult to extend to Indian languages. We introduce IndicFDB, which extends it to ten languages spoken in India with 12,350 samples, nearly 17 times as many as the original. We address three challenges: finding conversational events in multilingual speech, evaluating their timing without reliable word-level alignment, and judging responses across languages. We mine pause handling, turn taking, and backchanneling samples from roughly 50,000 hours of channel-separated conversations using voice activity detection (VAD), and construct human-validated synthetic user interruption samples. Language-independent VAD heuristics evaluate timing, while an open-weight transcription and translation pipeline converts responses to English for LLM ratings of relevance and quality. Across seven voice agents, commercial APIs show unexpectedly consistent behavior across languages but are either fast or robust to pauses, never both, while monolingual open full-duplex models expose further tradeoffs among backchanneling, response quality, and latency.
Chinese Translation
全双工语音代理必须实时处理停顿、轮流对话、反馈通道以及用户中断。Full-Duplex-Bench评估这些行为，但其仅限于英语的语料库以及对带有时间戳的自动语音识别（ASR）和以英语提示的语言模型（LLM）评判的依赖，使其难以扩展到印度语言。我们引入了IndicFDB，将其扩展到印度使用的十种语言，共包含12,350个样本，几乎是原始样本的17倍。我们解决了三个挑战：在多语言语音中寻找对话事件、在没有可靠的词级对齐的情况下评估其时机，以及跨语言判断响应。我们利用语音活动检测（VAD）从大约50,000小时的分离通道对话中挖掘停顿处理、轮流对话和反馈通道样本，并构建经过人工验证的合成用户中断样本。语言无关的VAD启发式方法评估时机，而开放权重的转录和翻译管道将响应转换为英语，以便LLM对相关性和质量进行评分。在七个语音代理中，商业API在不同语言间表现出意外一致的行为，但要么在速度上表现良好，要么在停顿处理上表现稳健，二者兼得的情况并不多见，而单语开放全双工模型则揭示了反馈通道、响应质量和延迟之间的进一步权衡。
cs.CL / 17 / 2609.31974
Extraction of clinical findings from mammography and breast ultrasound reports: a comparison between specialists and Artificial Intelligence
从乳腺X光和乳腺超声报告中提取临床发现：专家与人工智能的比较
Abstract
Breast cancer is the leading cause of cancer-related death among women in Brazil, and the time between the request and the release of mammography reports directly influences adherence to screening, making the agility in processing these reports a critical factor for early diagnosis. In this context, this study compares the performance of a Large Language Model (LLM) with manual extraction performed by a team of health researchers in identifying clinical findings from mammography and breast ultrasound reports written in Brazilian Portuguese. Named Entity Recognition (NER) was applied through Prompt Engineering using a few-shot strategy, employing the Gemini 2.5 Flash model, selected from preliminary exploratory tests with four candidate models. The Gemini 2.5 Flash model demonstrated the best performance, achieving a Macro F1 of 0.91 and a Micro F1 of 0.98. The subjective validation, in which 29 exams of different formats were evaluated by health researchers using a Likert scale, yielded an agreement index of 93.1%. The model outperformed human extraction in overall Macro F1 (0.91 vs. 0.72), as in four reports the model correctly identified information that had been omitted or incorrectly recorded during manual extraction, demonstrating its potential as a complementary verification tool alongside specialists. The results confirm the hypothesis that LLMs, when instructed through Prompt Engineering, can achieve performance comparable to or superior to manual extraction by health professionals.
Chinese Translation
乳腺癌是巴西女性癌症相关死亡的主要原因，而乳腺X光报告的请求与发布之间的时间直接影响筛查的依从性，使得处理这些报告的敏捷性成为早期诊断的关键因素。在此背景下，本研究比较了大型语言模型（Large Language Model, LLM）与健康研究团队手动提取在识别用巴西葡萄牙语撰写的乳腺X光和乳腺超声报告中的临床发现方面的表现。通过使用少量示例策略的提示工程（Prompt Engineering）应用命名实体识别（Named Entity Recognition, NER），采用了Gemini 2.5 Flash模型，该模型是在对四个候选模型进行初步探索性测试后选定的。Gemini 2.5 Flash模型表现最佳，宏观F1值达到0.91，微观F1值达到0.98。主观验证中，29份不同格式的检查由健康研究人员使用李克特量表进行评估，达成一致指数为93.1%。该模型在整体宏观F1值上优于人工提取（0.91对0.72），在四份报告中，模型正确识别了在手动提取过程中遗漏或错误记录的信息，展示了其作为专家辅助验证工具的潜力。结果证实了假设，即当通过提示工程进行指导时，LLMs能够达到与健康专业人员的手动提取相当或更优的表现。
cs.CL / 18 / 2609.31976
Toward Embedding-Based Psychometrics: Structural Modeling of Assessment-Item Semantics With Contextual Scores
基于嵌入的心理测量学：评估项目语义的结构建模与上下文评分
Abstract
Contextual scores represent assessment items through their similarities to reference words in an external corpus. We examine the semantic structure of scores for 40 TIMSS mathematics scored units using a partially specified two-step factor procedure. A search across factor counts identifies a persistent seven-group structure under the featured construction. Subsequent comparisons consistently favor a general dimension alongside group associations, although individual group memberships remain sensitive to some specification choices. Item examples distinguish recurring, cross-domain, sensitive, and imposed associations. Simpler and unrestricted references clarify the contribution and limits of the anchored representation: it improves on a single factor but does not achieve the lowest working Bayesian information criterion (BIC). A separate response benchmark compares three initial Q constructions and their Hull-PVAF revisions under higher-order and saturated attribute distributions. Among these diagnostic models, BIC favors the official content framework and the Akaike information criterion (AIC) favors its direct four-factor augmentation, but a matched unidimensional two-parameter logistic model has lower AIC and BIC than all twelve conditions. These findings support a conditional semantic representation while limiting direct diagnostic interpretation. We discuss learned text-assisted response calibration as a prospective application requiring a larger calibrated item bank and independent evaluation.
Chinese Translation
上下文评分通过与外部语料库中的参考词的相似性来表征评估项目。我们使用部分指定的两步因子程序，考察40个TIMSS数学评分单元的评分语义结构。因子计数的搜索识别出在所特征的构造下持续存在的七组结构。后续比较始终支持一个一般维度及其组关联，尽管个别组成员资格对某些规格选择仍然敏感。项目示例区分了重复的、跨领域的、敏感的和强加的关联。更简单且不受限制的参考明确了锚定表示的贡献和局限：它在单因子上有所改进，但未能达到最低的贝叶斯信息准则（BIC）。一个单独的响应基准比较了三种初始Q构造及其在高阶和饱和属性分布下的Hull-PVAF修订。在这些诊断模型中，BIC偏向官方内容框架，而赤池信息准则（AIC）则偏向其直接的四因子增强，但匹配的单维两参数逻辑模型的AIC和BIC均低于所有十二种条件。这些发现支持条件语义表示，同时限制了直接的诊断解释。我们讨论了学习文本辅助的响应校准作为一个前景应用，需依赖更大规模的校准项目库和独立评估。
cs.CL / 19 / 2609.31989
Communication between Frozen Large Language Models via Prompt Optimization in a Referential Game
通过提示优化在参照游戏中实现冻结大型语言模型之间的通信
Abstract
We study communication between two frozen large language models from different providers, with different tokenizers, accessed through their API endpoints. The two play a referential game: one sees an object and describes it in a short fixed-length message over a small alphabet; the other must pick that object out of a candidate set. Neither model's weights are updated. Each agent's prompt is rewritten by an isolated prompt optimizer whose reflection model reads that agent's scored interactions. In the positional setting, optimized prompts carry a shared code that generalizes to held-out objects above a measured no-codebook baseline, including when the memory window is removed. In a second setting, independent per-letter blocks no longer fit within the message, although a whole-object place value code does. The base system fails to establish reliable communication: the sender struggles to retain an injective rule, and the receiver has too few confirmed examples in view. A sender collision penalty, retention of successful interactions, and sequential optimization enable successful place value communication in some runs. Outcomes vary across runs and reflection models. In successful runs, the protocol is written into the optimized prompts, where it can be read and audited directly.
Chinese Translation
我们研究了来自不同提供者的两个冻结大型语言模型之间的通信，这两个模型使用不同的分词器，并通过它们的API端点进行访问。两者进行了一场参照游戏：一个模型观察一个对象并通过一个小字母表以固定长度的简短消息描述该对象；另一个模型必须从候选集合中挑选出该对象。两个模型的权重均不被更新。每个代理的提示由一个独立的提示优化器重写，该优化器的反射模型读取该代理的评分交互。在位置设置中，优化后的提示携带一个共享代码，该代码能够推广到超出测量的无代码本基线的持有对象，包括在移除记忆窗口时。在第二种设置中，独立的逐字块不再适合消息中，尽管整个对象的位值代码是适合的。基础系统未能建立可靠的通信：发送者难以保持单射规则，而接收者在视野中确认的示例过少。发送者碰撞惩罚、成功交互的保留以及顺序优化使得在某些运行中实现了成功的位值通信。结果在不同的运行和反射模型之间有所不同。在成功的运行中，协议被写入优化后的提示中，可以直接读取和审计。
cs.CL / 20 / 2609.31995
Before the Rollout Ends: Early Terminal Reward Prediction for Long-horizon Coding Agents
推广结束前：长时间编码代理的早期终端奖励预测
Abstract
Long-horizon coding agents receive verifiable rewards only after completing expensive sequences of tool calls. This increases inference cost, amplifies early wrong hypotheses, and can lead to sparse terminal reward and unstable training. We introduce Contextual Early Reward (CER), which predicts terminal reward through behavioral evidence in a trajectory prefix. CER synthesizes adaptive rubrics specific to the current task and stage through experiences summarized from related historical tasks. In test-time scaling on SWE-bench Verified, CER improves RM@8 over the strongest baseline by 4.2 percentage points (pp) on Nemotron 3 Ultra and 2.0 pp on Qwen 3.6 27B; on Nemotron, it takes only 15.3% tokens to match the best baseline performance. In RL training experiments, CER exceeds full-rollout TMax by 1.9 pp while using 52.7% fewer online policy-and-judge tokens. Together, CER provides an interpretable, efficient, and dense evaluation method for long-horizon coding agents.
Chinese Translation
长时间编码代理仅在完成昂贵的工具调用序列后才能获得可验证的奖励。这增加了推理成本，放大了早期错误假设，并可能导致稀疏的终端奖励和不稳定的训练。我们引入了上下文早期奖励（Contextual Early Reward, CER），通过轨迹前缀中的行为证据预测终端奖励。CER通过从相关历史任务中总结的经验合成特定于当前任务和阶段的自适应标准。在SWE-bench Verified的测试时间扩展中，CER在Nemotron 3 Ultra上将RM@8提高了4.2个百分点（pp），在Qwen 3.6 27B上提高了2.0 pp；在Nemotron上，仅需15.3%的标记即可匹配最佳基线性能。在强化学习训练实验中，CER的表现超越了全推广TMax 1.9 pp，同时使用了52.7%更少的在线策略和评判标记。总的来说，CER为长时间编码代理提供了一种可解释、高效且密集的评估方法。
cs.CL / 21 / 2609.32042
Quantization Thresholds Replicate, Failure Modes Do Not: A Three-Model Study of Agentic Tool Use in Polish from 8-bit to 2-bit
量化阈值的复制性与失败模式的非复制性：从8位到2位的波兰语工具使用的三模型研究
Abstract
We ask how GGUF quantization affects agentic tool use in Polish and whether the effects generalize across models. We introduce PolAgentBench, a deterministic benchmark with Polish prompts and English tool schemas: a 67-task main suite (15 adversarial probes, 52 hard-tier tasks) and a 46-task arithmetic isolation ladder. Three models span two axes of variation: Bielik-11B-v3.0 and its pruned, distilled child Bielik-Minitron-7B-v3.0 isolate model compression, and Llama-PLLuM-8B adds a change of pretraining family. Each is measured at six precisions, Q8_0 to Q2_K. Only the collapse threshold replicates. (1) All three models fall off a cliff between 3-bit and 2-bit (11B 0.716 to 0.045, 7B 0.463 to 0.149, PLLuM 0.224 to 0.015; paired McNemar p < 0.001 in each), across a fourfold capability spread and both axes. (2) Failure modes do not replicate: at 2-bit the 7B fails long (median 9.1k tokens, 4 steps) while the 11B mostly answers at the first step with a confabulated final answer (37 of 64 failures); PLLuM fails on content across precisions (71.8-92.0% of steps parse). (3) On the arithmetic ladder the unscaffolded rung is a floor, left standing by a rerun that states the no-tool rule; four explicit calls lift the 8-bit 11B from 1/10 to 9/10 and the 7B from 0/10 to 7/10 after format-only failures with the gold value are forgiven, an exploratory effect with eight distinct baseline inputs that does not survive multiplicity correction, while the order-trap arm separates the models at 8-bit (11B 6/6, 7B 0/6). (4) The Polish-versus-English gap is associated with degradation or with task family. We document four artifacts that shaped our conclusions (rounding-hostile gold values, strict answer typing, a no-tool rule the prompt never stated, priority-ordered failure labels), report affected results in strict and corrected form, and release the benchmark, trajectories and commit-stamped artifacts.
Chinese Translation
我们探讨了GGUF量化如何影响波兰语中的主动工具使用，以及这些影响是否在不同模型间具有普遍性。我们引入了PolAgentBench，这是一个具有波兰语提示和英语工具架构的确定性基准：包括67个任务的主要套件（15个对抗性探测任务，52个困难级任务）和46个任务的算术隔离阶梯。三个模型在两个变化轴上展开：Bielik-11B-v3.0及其修剪、蒸馏的子模型Bielik-Minitron-7B-v3.0专注于模型压缩，而Llama-PLLuM-8B则引入了预训练家族的变化。每个模型在六个精度下进行测量，从Q8_0到Q2_K。只有崩溃阈值是可复制的。（1）所有三个模型在3位和2位之间出现了显著下降（11B从0.716降至0.045，7B从0.463降至0.149，PLLuM从0.224降至0.015；在每个模型中配对的McNemar检验p < 0.001），在四倍的能力差异和两个轴上均如此。（2）失败模式并不复制：在2位时，7B模型长时间失败（中位数9.1k个标记，4步），而11B模型大多在第一步给出一个虚构的最终答案（64次失败中有37次）；PLLuM在不同精度下的内容解析失败率为71.8%-92.0%。（3）在算术阶梯上，未支撑的阶梯是一个底线，通过重新运行声明无工具规则而得以保留；四次明确的调用将8位的11B从1/10提升至9/10，将7B从0/10提升至7/10，前提是原始格式的失败被宽恕，这是一种探索性效应，涉及八个不同的基线输入，但未能通过多重性校正，而顺序陷阱臂在8位时将模型分开（11B 6/6，7B 0/6）。（4）波兰语与英语之间的差距与退化或任务家族相关。我们记录了四个影响我们结论的伪影（对舍入不友好的黄金值、严格的答案类型、提示中未声明的无工具规则、优先顺序的失败标签），报告了严格和校正形式下的受影响结果，并发布了基准、轨迹和提交时间戳的伪影。
cs.CL / 22 / 2609.32082
mu-bench: A Multilingual Utterance Transcription Benchmark
mu-bench：多语言话语转录基准
Abstract
Voice agents depend on accurate automatic speech recognition (ASR) to act on what callers say, yet ASR is evaluated on read, English-centric speech with word error rate (WER), which penalizes surface rather than semantic differences. We introduce mu-bench, a dataset of 4,270 caller utterances from 250 phone calls to an AI banking agent in English, Spanish, Turkish, Vietnamese, and Mandarin, centered on form-field inputs such as names, email addresses, and confirmation codes. We release Utterance Error Rate (UER), an LLM judge of whether a transcript preserves meaning, calibrated against human raters, together with an LLM normalizer that makes WER comparable across providers' output formats. On 1,847 human-rated transcripts, UER agrees with annotators at $\kappa$ = 0.78, versus 0.53 for exact-match WER on normalized text. We rank six commercial providers on a public leaderboard; the best reaches 11.9% UER, and Mandarin is hardest for all six.
Chinese Translation
语音助手依赖于准确的自动语音识别（ASR）来执行来电者的指令，但ASR的评估主要基于标准化的、以英语为中心的语音，使用词错误率（WER）进行评估，这种方法更关注表面差异而非语义差异。我们提出了mu-bench，一个包含来自250个电话拨打到AI银行助手的4,270个来电者话语的数据集，涵盖英语、西班牙语、土耳其语、越南语和普通话，主要集中在姓名、电子邮件地址和确认码等表单字段输入上。我们发布了话语错误率（UER），这是一个大型语言模型（LLM）评估转录是否保留意义的指标，并与人工评分者进行了校准，同时提供了一个LLM标准化器，使得不同服务提供商的输出格式之间的WER具有可比性。在1,847个人工评分的转录中，UER与标注者的协议度为$ ext{kappa}$ = 0.78，而标准化文本的精确匹配WER为0.53。我们在一个公开排行榜上对六个商业服务提供商进行了排名；最佳提供商的UER达到11.9%，而普通话对所有六个提供商来说都是最具挑战性的语言。
cs.CL / 23 / 2609.32119
Using LMs to Model the Effects of Context and Coreference during Sentence Comprehension
使用语言模型（LMs）模拟句子理解中的上下文和共指的影响
Abstract
Language models (LMs) are often used as a tool to model human language processing. Recent studies suggest that severely restricting LMs' context window improves their fit to human psycholinguistic data by simulating human working memory constraints. However, it is possible that this strict memory-decay approach overlooks humans' reliance on long-range structural representations, such as discourse structre. In this work, we systematically vary the context window size of GPT-2 across four large-scale naturalistic English reading-time datasets and observe a U-shaped relationship: Although restricted contexts (< 20 tokens) successfully capture local memory limitations, expanded contexts (500--1,000 tokens) ultimately yield the highest overall psycholinguistic fit. To investigate the mechanism driving this benefit, we conduct a counterfactual inference-time experiment that disrupts cross-sentential entity chains by pronominalizing repeated discourse entities. Obscuring these structural linkages significantly degrades the predictive power of larger context windows by 20% to 40%. Our experiments demonstrate that tracking long-range coreference relations is one important factor for the alignment between LM surprisal and human reading behavior, and approximate the extent to which human comprehenders use global discourse relations during language processing.
Chinese Translation
语言模型（LMs）常被用作模拟人类语言处理的工具。近期研究表明，严重限制语言模型的上下文窗口可以通过模拟人类工作记忆的限制来提高其与人类心理语言学数据的拟合度。然而，这种严格的记忆衰减方法可能忽视了人类对长程结构表征（如话语结构）的依赖。在本研究中，我们系统地改变了GPT-2的上下文窗口大小，涵盖了四个大规模自然英语阅读时间数据集，并观察到一种U型关系：尽管限制上下文（< 20个标记）成功捕捉了局部记忆限制，但扩展上下文（500--1,000个标记）最终产生了最高的整体心理语言学拟合度。为了探讨驱动这一益处的机制，我们进行了一项反事实推理实验，通过代词化重复的话语实体来干扰跨句实体链。模糊这些结构联系显著降低了较大上下文窗口的预测能力，下降幅度为20%至40%。我们的实验表明，追踪长程共指关系是语言模型惊讶度与人类阅读行为之间对齐的重要因素之一，并近似估计了人类理解者在语言处理过程中使用全球话语关系的程度。
cs.CL / 24 / 2609.32160
Typed Decision Models: An Early Evidence Audit and Evaluation Checklist
类型化决策模型：早期证据审计与评估检查表
Abstract
Typed decision models (TDMs) return probability distributions over caller-defined options without generating text. TypeSafe released Jev, a commercial typed decision model, on 15 September 2026, and a small body of evaluation and replication work appeared within days. We review 28 papers posted between 19 and 24 September and relate their findings to earlier work on label-probability classification, constrained decoding, reranking, calibration, and model cascades. In this early literature, the typed readout itself has not shown an independent accuracy advantage over comparable label-probability readouts. Jev's clearest gains are in latency and cost, while accuracy gaps remain on harder tasks. In practical deployments, confidence is often used to decide when to defer to a stronger model or a human. We use recurring weaknesses in these studies to derive a 14-item evaluation checklist for future TDM work. Because the evidence covers only the first nine days after the release of one hosted model, the review should be read as an early evidence map rather than a settled assessment of the model class.
Chinese Translation
类型化决策模型（Typed Decision Models, TDMs）在不生成文本的情况下返回调用者定义选项的概率分布。TypeSafe于2026年9月15日发布了商业类型化决策模型Jev，并在几天内出现了一小部分评估和复制工作。我们回顾了在9月19日至24日之间发布的28篇论文，并将其发现与早期关于标签-概率分类、受限解码、重排序、校准和模型级联的工作联系起来。在这早期文献中，类型化输出本身并未显示出相较于可比的标签-概率输出的独立准确性优势。Jev在延迟和成本方面的优势最为明显，而在更困难的任务上准确性差距依然存在。在实际应用中，信心通常用于决定何时依赖更强的模型或人类。我们利用这些研究中的反复弱点，提出了一个包含14项的未来TDM工作的评估检查表。由于证据仅涵盖了发布一个托管模型后的前九天，因此该评审应被视为早期证据图而非对模型类别的定论性评估。
cs.CL / 25 / 2609.32189
ScopeIF: Improving Scope-Aware Precise Instruction-Following in Large Language Models via Graded Reward Modeling
ScopeIF：通过分级奖励建模提升大型语言模型的范围感知精确指令跟随能力
Abstract
Precise instruction-following is a fundamental ability of large language models (LLMs), requiring their outputs to strictly satisfy objective constraints in input instructions. In complex application scenarios, these constraints often possess diverse scopes that govern specific response segments rather than the entire output. However, existing optimization methods often neglect constraint scope during data construction and rely on binary per-constraint rewards, yielding limited data diversity and sparse supervision for complex constraints. To this end, we propose ScopeIF, a novel training framework for scope-aware precise instruction-following. We first introduce a unified schema that factorizes objective constraints into three decoupled dimensions: Scope, Target, and Range. Grounded in this schema, we construct ScopeInstruct, a large-scale instruction dataset with diverse scope-aware constraints, and combine tool-grounded verification with graded reward modeling to quantify the violation degree of each constraint, providing dense supervision for policy optimization. Extensive experiments demonstrate that ScopeIF consistently outperforms existing methods, particularly on complex scope-aware constraints, while preserving general capabilities. Notably, it enables optimized Qwen3-4B and 8B models to rival or surpass strong frontier models such as Gemini-2.5-Pro and DeepSeek-V3.2, establishing an effective paradigm for advancing scope-aware instruction-following. Our code and data are available at https://github.com/thu-coai/ScopeIF.
Chinese Translation
精确的指令跟随是大型语言模型（LLMs）的一项基本能力，要求其输出严格满足输入指令中的客观约束。在复杂的应用场景中，这些约束通常具有多样的范围，支配特定的响应片段而非整个输出。然而，现有的优化方法往往在数据构建过程中忽视约束范围，并依赖于每个约束的二元奖励，这导致数据多样性有限且对复杂约束的监督稀疏。为此，我们提出了ScopeIF，一种新的范围感知精确指令跟随训练框架。我们首先引入一个统一的框架，将客观约束分解为三个解耦的维度：范围（Scope）、目标（Target）和范围（Range）。基于该框架，我们构建了ScopeInstruct，一个具有多样化范围感知约束的大规模指令数据集，并结合工具基础的验证与分级奖励建模来量化每个约束的违反程度，为策略优化提供密集的监督。大量实验表明，ScopeIF在复杂的范围感知约束上始终优于现有方法，同时保持了通用能力。值得注意的是，它使优化后的Qwen3-4B和8B模型能够与强大的前沿模型如Gemini-2.5-Pro和DeepSeek-V3.2相媲美或超越，建立了推进范围感知指令跟随的有效范式。我们的代码和数据可在https://github.com/thu-coai/ScopeIF获取。
cs.CL / 26 / 2609.32196
The Judge Is Not Its Twin: Post-training makes a model's writing more predictable but barely moves its taste, as a judge, toward predictable writing
法官并非其双胞胎：后训练使模型的写作更具可预测性，但几乎没有使其作为法官的品味向可预测写作倾斜
Abstract
Language models are now routinely graded by other language models. If post-training makes a model's own writing more predictable, it may also teach the same model, acting as a judge, to reward predictable writing, so that progress on creativity would be invisible to automated evaluation. We follow two open model families, OLMo-2 and Zephyr (7B parameters each), through their public training stages and measure every stage twice, as a writer of short stories and as a judge of pairs of stories. As writers, the models drift as feared: each family's fully trained model finds the stories of its base, supervised fine-tuned (SFT) and preference-trained (DPO) stages progressively more familiar, in all ten prompts, and a model from the other family finds the trained stories 7.0 to 9.2 percent (OLMo-2) and 24 to 25 percent (Zephyr) less surprising per token. As judges, they barely move toward predictable writing. Asked which story is better, a question every judge can use to tell a story from its own words in scrambled order, no trained judge's estimated tilt toward the more predictable story grows by as much as one point in the probability of picking it. A post hoc one-sided 95% upper bound on that growth is 2.7 points on an average pair, about the size of the untrained OLMo-2 judge's own tilt. Asked which is more creative, no trained judge's estimate favors the predictable story more than its base's does. Training instead strengthens a preference for longer stories when the question is creativity, and for one answer slot, and it breaks "more creative" as a question: trained judges asked it no longer reliably prefer a story to its scrambled words. A follow-up could not build pairs that differ in predictability but not in quality, because the routes that made this writer's stories less predictable also broke some of them, often enough to fail a quality floor set in advance.
Chinese Translation
语言模型现在常常由其他语言模型进行评分。如果后训练使模型自身的写作变得更可预测，它也可能教会同一模型作为法官来奖励可预测的写作，从而使创造力的进展在自动评估中变得不可见。我们跟踪两个开放模型系列，OLMo-2 和 Zephyr（每个模型有 7B 参数），在其公开训练阶段进行测量，并在每个阶段进行两次测量，作为短篇故事的作者和故事对的评判者。作为作者，这些模型的表现如预期般发生漂移：每个系列的完全训练模型在所有十个提示中逐渐发现其基础、监督微调（SFT）和偏好训练（DPO）阶段的故事越来越熟悉，来自另一个系列的模型发现训练故事的可预测性每个标记减少了 7.0% 到 9.2%（OLMo-2）和 24% 到 25%（Zephyr）。作为评判者，它们几乎没有向可预测写作倾斜。当被问及哪个故事更好时，这个问题是每个法官都可以用来从其自身的词汇中讲述一个故事的混乱顺序，没有经过训练的法官对更可预测故事的倾斜估计增长不到一分。对此增长的后验单侧 95% 上限为平均对的 2.7 分，约等于未训练的 OLMo-2 法官自身的倾斜。当被问及哪个故事更具创造力时，没有经过训练的法官的估计比其基础模型更倾向于可预测故事。训练反而在创造力问题上加强了对较长故事的偏好，并在一个答案选项中打破了“更具创造力”这一问题：经过训练的法官在被问及时不再可靠地偏好一个故事而不是其混乱的词汇。后续研究无法构建在可预测性上有所不同但在质量上没有差异的对，因为使这个作者的故事变得不那么可预测的途径也破坏了其中一些故事，常常足以未能达到预先设定的质量底线。
cs.CL / 27 / 2609.32199
Generalization and Memorization along the Learning Trajectory of Neural Language Models: A Geometric Account of Categorization
神经语言模型学习轨迹中的泛化与记忆：分类的几何解释
Abstract
We investigate how generalization and memorization develop along the learning trajectory of neural language models. Using controlled synthetic grammars, we examine both the geometry of representation space and model behaviour over training. We find that continuous regions of representation space not occupied by observed tokens become systematically structured from the earliest stages of learning, forming category-level geometric organization that supports generalization to unattested combinations. Generalization therefore emerges from the beginning of learning rather than only after extensive memorization. With prolonged training, larger models increasingly distinguish observed from unobserved grammatical combinations. At the same time, the continuous geometric structure supporting category-level generalization is gradually destructed. Together, these results suggest that neural language models initially learn through categorization-based generalization, followed by a gradual transition toward more exemplar-specific memorization.
Chinese Translation
我们研究了神经语言模型在学习轨迹中泛化与记忆的发展。通过使用受控的合成语法，我们考察了表示空间的几何特征和模型在训练过程中的行为。我们发现，从学习的最早阶段开始，未被观察到的标记所占据的表示空间的连续区域开始系统性地结构化，形成支持对未见组合的泛化的类别级几何组织。因此，泛化从学习的开始就出现，而不仅仅是在广泛记忆之后。随着训练的延长，较大的模型越来越能够区分观察到的与未观察到的语法组合。同时，支持类别级泛化的连续几何结构逐渐被破坏。综合这些结果表明，神经语言模型最初通过基于分类的泛化进行学习，随后逐渐过渡到更具示例特异性的记忆。
cs.CL / 28 / 2609.32216
A model of rational interlocutors: Unification of comprehension and production
理性对话者模型：理解与表达的统一
Abstract
Who we communicate with influences both our interpretation of their utterances and the design of our own. Such adjustment to the conversational partner is studied as speaker modeling in comprehension and as audience design in production, with the two literatures having developed largely separately. We argue that both adjustments express one rational computation and propose the rational interlocutor (RI) model, a computational account unifying comprehension and production. An interlocutor maintains a model of their partner, defined by three parameters: an identity parameter {\Pi} sets the messages and forms expected from the partner; a fidelity parameter {\Phi} sets how reliably messages and utterances map onto each other for them; a knowledge parameter {\Lambda} sets how knowledgeable the partner is believed to be. Comprehension and production are thus mirror-image modes of one computation over the partner model. Comprehension chooses the message the partner most likely intends to convey, weighing how well each candidate fits the utterance against how likely this partner is to mean it. Production chooses the utterance from which the partner will best recover the message, weighed against the effort of saying it. This explains why comprehenders appear to rely less on the forms produced by a linguistically less competent speaker, while producers tend to invest more effort in designing forms for them. We conjecture that perceived linguistic competence decomposes into two of these quantities: fidelity and knowledge. Their contrasting profiles across second-language (L2) adults, children, and artificial partners produce distinct and testable predictions.
Chinese Translation
我们与谁沟通会影响我们对其言语的解读以及我们自己言语的设计。这种对交谈伙伴的调整在理解中被研究为说话者建模，在表达中被研究为受众设计，而这两个领域的文献发展大体上是分开的。我们认为这两种调整都表达了一种理性的计算，并提出了理性对话者（Rational Interlocutor, RI）模型，这是一个统一理解与表达的计算模型。对话者维护着对其伙伴的模型，该模型由三个参数定义：身份参数 {} 设置了预期来自伙伴的消息和形式；保真度参数 {} 设置了消息和言语之间相互映射的可靠性；知识参数 {} 设置了对伙伴知识水平的信念。因此，理解和表达是对伙伴模型进行的一种计算的镜像模式。理解选择伙伴最有可能意图传达的消息，权衡每个候选项与该伙伴可能的意图之间的契合度。表达则选择伙伴能够最佳恢复消息的言语，权衡表达该消息的努力。这解释了为什么理解者似乎较少依赖于语言能力较弱的说话者所产生的形式，而表达者则倾向于在为他们设计形式时投入更多的努力。我们推测，感知的语言能力可以分解为这两个量：保真度和知识。在第二语言（L2）成年人、儿童和人工伙伴之间的对比特征产生了不同且可检验的预测。
cs.CL / 29 / 2609.32227
OptiArena: Can LLMs Improve Executable Algorithms under Fixed Resource Budgets?
OptiArena：大型语言模型能否在固定资源预算下改善可执行算法？
Abstract
Static QA and code-generation benchmarks only partially capture the role that large language models (LLMs) now play as coding agents and research tools. We introduce OptiArena, a budget-controlled testbed for studying whether LLMs can improve executable game-playing algorithms through five rounds of code edits within a fixed minimal scaffold and under bounded evaluator feedback and fixed resource budgets. The testbed uses two optimization regimes, surface obfuscation controls, calibrated references, held-out/stress splits, and diagnostics for degradation and exceptional failures, with LLM API cost reported separately from local evaluator wall-clock. The empirical study asks three questions: whether models can close the calibrated gap between a designated weak starter and an editable competent baseline, whether they can refine editable competent baselines without damaging them, and whether gains survive surface obfuscation controls. Across twelve frontier LLMs and five games, models improve designated weak starters more consistently than they refine editable competent baselines, with substantial variation across games and models. OptiArena provides a practical testbed for measuring bounded-resource algorithm optimization within the five-edit, fixed-scaffold setting studied here. Code is available at https://github.com/WJ-Peng/OptiArena.
Chinese Translation
静态问答和代码生成基准仅部分捕捉了大型语言模型（LLMs）作为编码代理和研究工具的角色。我们介绍了OptiArena，这是一个预算控制的测试平台，用于研究LLMs是否能够通过在固定的最小框架内进行五轮代码编辑，并在有限的评估反馈和固定资源预算下改善可执行的游戏算法。该测试平台使用两种优化机制、表面混淆控制、校准参考、保留/压力分割以及用于退化和异常失败的诊断，同时LLM API的成本与本地评估器的实际时间分开报告。实证研究提出了三个问题：模型是否能够缩小指定弱起点与可编辑的合格基线之间的校准差距，是否能够在不损害可编辑合格基线的情况下进行改进，以及这些改进是否能够在表面混淆控制下存活。在十二个前沿LLMs和五个游戏中，模型在改善指定的弱起点方面比在改进可编辑合格基线方面更为一致，且不同游戏和模型之间存在显著差异。OptiArena提供了一个实用的测试平台，用于在此研究的五次编辑、固定框架设置中测量有限资源的算法优化。代码可在 https://github.com/WJ-Peng/OptiArena 获取。
cs.CL / 30 / 2609.32234
KinyaMed: Seeds, Not Rows -- What a Corpus Requirement Written in the Wrong Unit Fails to Constrain
KinyaMed：种子，而非行 -- 错误单位下的语料库需求未能约束的内容
Abstract
Triage decides who is seen first. Building an urgency classifier for patient-voice Kinyarwanda, we found our specification could be met without producing anything it was meant to secure. We report that, and the instruments that detect it, instead of a classifier. Designed for the four languages a Rwandan health centre receives, with every instrument per-language: sentences are authored in all four arms and rows generate in one, because the frame slots those three need do not exist. Our requirement asked for one million examples; generation produced them in 130 seconds. It fails four of the nine quality gates in that specification, six when rows are attributed to their source sentence. The binding gate counts distinct authored seed phrases, not rows: at our 165, no corpus passes at any row count. Two shortfalls follow and differ: 2,835 further sentences to pass the seed-count gate, 19,835 to reach the stated million rows, because a separate gate caps a seed at 50 rows. Rows come from a machine at 7,700 per second; seeds from clinicians. A row count constrains the cheap quantity, leaves the expensive one free, and so does not constrain quality at all. Two further negatives follow. An evaluation set of 17,942 rows built from nine distinct sentences supports no verdict: our gate, which counts distinct sentences, refuses 38 of its cells and reports nothing. A model trained on a corpus of uniform surface form changes its predicted urgency for 31.5% of inputs under capitalisation and 21.0% under a single typo, reported as measurement and not attribution. A model directory shipped without its tokenizer loads without error and answers its class prior on input it cannot read, with well-formed probabilities. No figure here is evidence of model quality; the contribution is the apparatus and the negative results it produced, reproducible from a clean clone except where marked NOT REPRODUCIBLE.
Chinese Translation
分诊决定谁优先就诊。在为患者语音基尼亚卢旺达语构建紧急性分类器时，我们发现我们的规范可以在不产生任何其应确保内容的情况下得到满足。我们报告这一发现，以及检测它的工具，而不是分类器。该分类器是为卢旺达健康中心接收的四种语言设计的，每种语言都有相应的工具：句子在四个方面创作，而行仅在一个方面生成，因为那三个方面所需的框架插槽并不存在。我们的需求要求一百万个示例；生成在130秒内产生了这些示例。它在该规范的九个质量门中未能通过四个，当行被归因于其源句子时则未能通过六个。绑定门计算不同的创作种子短语，而非行：在我们的165个示例中，没有任何语料库在任何行数下通过。接下来有两个不足之处且各不相同：需要2835个进一步的句子以通过种子计数门，19835个以达到所述的一百万行，因为一个单独的门将种子限制在50行。行来自每秒7700的机器；种子来自临床医生。行数限制了廉价数量，而将昂贵的数量留给自由，因此根本不限制质量。接下来还有两个负面结果。由九个不同句子构建的17942行评估集不支持任何裁决：我们的门，计算不同句子，拒绝其38个单元并报告无结果。一个在统一表面形式语料库上训练的模型在大写时对31.5%的输入改变其预测紧急性，在单一拼写错误时为21.0%，报告为测量而非归因。一个未附带分词器的模型目录在加载时没有错误，并在无法读取的输入上回答其类别先验，给出合理的概率。这里没有任何数据是模型质量的证据；贡献在于设备及其产生的负面结果，除了标记为不可重现的地方，均可从干净的克隆中重现。
cs.CL / 31 / 2609.32235
Solving Every Step Is Not Enough: Milestone Oracles Reveal a Composition Gap in LLM Math Reasoning
解决每一步并不足够：里程碑预言机揭示了大型语言模型数学推理中的组合缺口
Abstract
Large language models (LLMs) can solve every intermediate step of a multi-step math problem on its own and still fail the full problem, even when given a roadmap of the steps and all of their answers. We introduce OracleLadder, a diagnostic evaluation that locates where LLM math reasoning fails by giving the model increasing levels of oracle help. For each problem, a teacher model writes a fixed roadmap of intermediate sub-goals (milestones), and a deterministic symbolic verifier grades every answer. Testing the model with no help, with the roadmap, with the roadmap plus the milestone answers, and on each milestone alone sorts each failure into one of five reasoning gaps. On 354 NuminaMath problems and six models from 8B to 671B parameters (Qwen3, gpt-oss, Llama 3.3, DeepSeek-V3.1), the largest gap for every model is the composition gap, a stricter form of the compositionality gap. It covers 33-48% of problems, and 24-37% after removing problems that an LLM review flags as grading errors. Accuracy and milestone-help recovery rank the two strongest models differently, and two RLVR runs with similar accuracy gains move problems differently. The roadmap effect replicates on MATH500 and AIME 2024/25, per-problem recovery agrees for 83-87% of problems under an independent second teacher, and the help ladder carries over to code generation. We release the data, roadmaps, prompts, and code at https://github.com/slark-prime/OracleLadder.
Chinese Translation
大型语言模型（LLMs）能够独立解决多步骤数学问题的每一个中间步骤，但仍然可能无法解决完整问题，即使在提供了步骤的路线图及其所有答案的情况下。我们引入了OracleLadder，这是一种诊断评估方法，通过给予模型逐步增加的预言机帮助，定位LLM数学推理失败的地方。对于每个问题，教师模型编写固定的中间子目标（里程碑）路线图，而确定性的符号验证器对每个答案进行评分。通过不提供帮助、提供路线图、提供路线图加上里程碑答案，以及单独针对每个里程碑进行测试，将每个失败分类为五种推理缺口之一。在354个NuminaMath问题和六个模型（参数从8B到671B，包括Qwen3、gpt-oss、Llama 3.3、DeepSeek-V3.1）的测试中，所有模型中最大的缺口是组合缺口，这是一种更严格的组合性缺口形式。它覆盖了33-48%的问题，在去除LLM审查标记为评分错误的问题后，覆盖率为24-37%。准确性和里程碑帮助恢复对两个最强模型的排名有所不同，两个具有相似准确性提升的RLVR运行在问题处理上表现不同。路线图效应在MATH500和AIME 2024/25上得到了复制，独立的第二位教师对每个问题的恢复一致性达到了83-87%，而帮助梯度在代码生成中也得到了延续。我们在https://github.com/slark-prime/OracleLadder发布了数据、路线图、提示和代码。
cs.CL / 32 / 2609.32264
LANTERN: Illuminating Hidden Mathematical Knowledge in Language Models
LANTERN：揭示语言模型中的隐藏数学知识
Abstract
Language models can now prove theorems, but people still decide which problems to pursue. We ask whether a model's internal representations can help identify promising mathematical connections. We develop LANTERN, a fast, cost-efficient pipeline that uses a classifier over pretrained-model activations to rank candidate relations, followed by staged filtering, hypothesis generation, executable verification, and analytical checking. Applied to the On-Line Encyclopedia of Integer Sequences (OEIS), LANTERN ranked 50 million pairs among 10,000 frequently referenced sequences and produced 62 verified relations between pairs without an existing OEIS cross-reference. A content screen retained 13 relations worth presenting; nine of these are informative or insightful, including four which are entirely novel to the best of our knowledge: none appears in the OEIS or in our targeted literature search. The entire end-to-end process including classifier training, candidate ranking, filtering and verification took under 8 hours.
Chinese Translation
语言模型现在能够证明定理，但仍然由人类决定追求哪些问题。我们探讨模型的内部表征是否能够帮助识别有前景的数学联系。我们开发了 LANTERN，这是一个快速且成本高效的流程，利用预训练模型的激活值上的分类器对候选关系进行排名，随后进行分阶段过滤、假设生成、可执行验证和分析检查。在应用于在线整数序列百科全书（OEIS）时，LANTERN 在 10,000 个常被引用的序列中对 5000 万对进行了排名，并生成了 62 个在 OEIS 中没有现有交叉引用的已验证关系。内容筛选保留了 13 个值得展示的关系；其中九个具有信息性或洞察力，包括四个在我们所知的范围内完全新颖的关系：这些关系在 OEIS 或我们的目标文献搜索中均未出现。整个端到端过程，包括分类器训练、候选排名、过滤和验证，总共耗时不到 8 小时。
cs.CL / 33 / 2609.32269
Before Answering: Evidence Sufficiency under Size-Matched Memory Construction
回答之前：基于大小匹配记忆构建的证据充分性
Abstract
Agents that answer questions from compressed or retrieved memory must recognize when the evidence a query needs is no longer in memory. Benchmarks for this task usually create insufficient-evidence examples by deleting supporting passages. We show that this construction leaks the label through memory size: on MuSiQue, a classifier that only counts paragraphs reaches an area under the ROC curve (AUROC) of $0.979$ for detecting unsafe memory, higher than the lexical estimator we initially evaluated. We propose a size-matched construction that provably removes this shortcut, and use it to study MemSafe, an estimator that cross-encodes the query with each memory unit and aggregates the units with a set transformer. Across three multi-hop question answering datasets and five seeds, MemSafe reaches $0.968$ and $0.983$ AUROC on MuSiQue and HotpotQA, $0.26$ to $0.39$ above a lexical baseline, while the third dataset, 2WikiMultiHopQA, is saturated. A frozen pretrained cross-encoder with a logistic head already closes $41\%$ of the MuSiQue gap between the lexical baseline and MemSafe. At the same time, MemSafe degrades more than a weak baseline on the unanswerable questions released with MuSiQue, reaches only $0.639$ AUROC on SQuAD~2.0, and needs several thousand clinical training examples before it outperforms a feature-based estimator. Used as a gate for a 7B reader, it reduces the error rate on answered questions from $0.850$ to $0.631$ at $5\%$ coverage, outperforming both reader confidence and, on average, the ground-truth integrity label, although a 7B LLM judge is the better gate at $10\%$ coverage. These results indicate that the way insufficient evidence is constructed matters as much as the estimator that detects it.
Chinese Translation
从压缩或检索的记忆中回答问题的智能体必须识别出查询所需的证据何时不再存在于记忆中。此任务的基准通常通过删除支持段落来创建不足证据的示例。我们表明，这种构建方式通过记忆大小泄露了标签：在 MuSiQue 数据集上，仅计算段落数量的分类器在检测不安全记忆时达到了 $0.979$ 的 ROC 曲线下面积 (AUROC)，高于我们最初评估的词汇估计器。我们提出了一种大小匹配的构建方法，证明可以消除这一捷径，并利用它研究 MemSafe，这是一种将查询与每个记忆单元交叉编码并使用集合变换器聚合单元的估计器。在三个多跳问答数据集和五个种子上，MemSafe 在 MuSiQue 和 HotpotQA 上达到了 $0.968$ 和 $0.983$ 的 AUROC，分别比词汇基线高出 $0.26$ 到 $0.39$，而第三个数据集 2WikiMultiHopQA 已经饱和。一个冻结的预训练交叉编码器加上逻辑头已经缩小了词汇基线与 MemSafe 之间的 $41\%$ 的 MuSiQue 差距。同时，MemSafe 在与 MuSiQue 一起发布的无法回答的问题上表现不如一个弱基线，在 SQuAD~2.0 上仅达到 $0.639$ 的 AUROC，并且在超过几千个临床训练示例之前，无法超过基于特征的估计器。作为一个 7B 读者的门控，它将回答问题的错误率从 $0.850$ 降低到 $0.631$，在 $5\\%$ 的覆盖率下，优于读者信心，并且在平均情况下优于真实完整性标签，尽管在 $10\\%$ 的覆盖率下，7B LLM 判别器是更好的门控。这些结果表明，不足证据的构建方式与检测它的估计器同样重要。
cs.CL / 34 / 2609.32281
Supporting and Performing Culture from the Inside
从内部支持与执行文化
Abstract
Anthropology and related disciplines which study culture have often found it useful to consider their epistemological and methodological approaches in terms of emic versus etic. This is the distinction between the insider perspective, how the world is viewed by a member of the culture, versus the outsider perspective, the scientific cataloguing of cultural knowledge and practice. We adopt this framework to systematically analyse assumptions in recent research on culture in NLP along the full pipeline of task selection, data collection, system design and evaluation. We draw attention to the `culture of NLP' which shapes the research focus and approaches of our field, and suggest pathways towards a more culturally attuned AI.
Chinese Translation
人类学及相关学科在研究文化时，常常发现以内视（emic）与外视（etic）的方式考虑其认识论和方法论是有益的。这一区别在于内部视角，即文化成员如何看待世界，和外部视角，即对文化知识和实践的科学分类。我们采用这一框架，系统分析近期在自然语言处理（NLP）领域关于文化的研究假设，涵盖任务选择、数据收集、系统设计和评估的整个流程。我们关注到塑造我们领域研究重点和方法的“NLP文化”，并建议朝向更具文化敏感性的人工智能（AI）发展路径。
cs.CL / 35 / 2609.32293
TRAP: Understanding and Mitigating Privacy Memorization in Language Models
TRAP：理解和减轻语言模型中的隐私记忆
Abstract
Fine-tuning a language model on sensitive records can leave it able to reproduce them. We ask when this memorization arises and how to prevent it without knowing in advance which spans are sensitive. Our starting point is that most memorization scores and attacks share one statistical core: whether the model assigns a token more probability than some reference would. Taking as the reference a model trained on the complementary half of the same corpus gives the Target Reference Advantage (TRA), a per-token signal that separates what a model fit to a particular record from what it learned across records, and is cheap and differentiable. We then study what drives memorization during fine-tuning: it keeps growing well past the validation minimum, is larger on small datasets and at higher learning rates, and higher when the underlying task is harder. Early stopping removes much of it, but because it is chosen by aggregate validation loss it helps least for rare, hard-to-predict spans embedded in otherwise learnable text, which is exactly what sensitive information tends to be. We therefore introduce TRAP, a one-sided penalty on tokenwise TRA that acts only where the target model pulls ahead of its reference. On student essays with annotated personal information and clinical cases with patient identifiers, TRAP brings memorization near the level of an untrained model at little utility cost, where generic regularizers barely move and differential privacy gives up most of what fine-tuning bought.
Chinese Translation
对敏感记录进行语言模型的微调可能导致模型能够重现这些记录。我们探讨这种记忆现象何时发生，以及如何在不知道哪些内容是敏感的情况下防止这种情况。我们的出发点是，大多数记忆评分和攻击共享一个统计核心：即模型是否为某个标记分配了比某个参考模型更高的概率。以在同一语料库的互补部分上训练的模型作为参考，得到目标参考优势（Target Reference Advantage, TRA），这是一个每个标记的信号，能够区分模型对特定记录的拟合与其在多个记录中学习到的内容，并且计算成本低且可微分。接着，我们研究了微调过程中驱动记忆的因素：记忆在验证最小值之后仍然持续增长，在小数据集上更大，在较高学习率下更明显，并且在基础任务更困难时更高。提前停止可以消除大部分记忆，但由于其是通过整体验证损失选择的，因此对嵌入在其他可学习文本中的稀有、难以预测的内容帮助最小，而这些内容正是敏感信息的特征。因此，我们引入了TRAP，这是一种对逐标记TRA的单边惩罚，仅在目标模型超过其参考模型时起作用。在带有注释的个人信息的学生论文和带有患者标识符的临床案例中，TRAP将记忆水平降低到接近未训练模型的水平，几乎没有效用成本，而通用正则化器几乎没有效果，差分隐私则放弃了大部分微调所带来的收益。
cs.CL / 36 / 2609.32331
Language Distances are Practical for Equitable Cross-Lingual Transfer
语言距离在公平的跨语言迁移中具有实用性
Abstract
Cross-lingual transfer is strongly conditional on how the source language is chosen, but it is impractical to determine the best candidate source for every target language, especially for low-resource target languages. Language distances are widely used to rank candidate sources due to their correlation with transfer efficacy and applicability in resource-sparse settings. However, the reliability of distance-based rankers across tasks and resource levels remains underexplored. We therefore present the first equity-focused evaluation of paradigms for ranking source languages, studying resource-level inequality and task inequality across ten cross-lingual tasks and two multilingual models. While both inequalities are most pronounced for individual language distances and an English-always baseline, they are substantially reduced by training-free composite distances, and nearly eliminated by trained rankers. We further demonstrate the reliability of rankers using language distances compared to rankers using language model internals. Overall, we find that language distances provide a practical basis for equitable and performant transfer language selection. We recommend using trained rankers when task-specific transfer evaluations are available, and composite distances otherwise.
Chinese Translation
跨语言迁移在很大程度上依赖于源语言的选择，但对于每种目标语言确定最佳候选源是不切实际的，尤其是对于低资源目标语言。语言距离因其与迁移有效性和在资源稀缺环境中的适用性相关而被广泛用于对候选源进行排名。然而，基于距离的排名器在不同任务和资源水平上的可靠性仍然未被充分探讨。因此，我们首次进行了以公平性为重点的源语言排名范式评估，研究了十个跨语言任务和两个多语言模型中的资源水平不平等和任务不平等。尽管这两种不平等在单一语言距离和始终使用英语的基线中最为明显，但通过无训练的复合距离可以显著减少，并且通过训练的排名器几乎可以消除。我们进一步展示了使用语言距离的排名器相比于使用语言模型内部的排名器的可靠性。总体而言，我们发现语言距离为公平和高效的迁移语言选择提供了实用基础。我们建议在可用任务特定迁移评估时使用训练的排名器，否则使用复合距离。
cs.CL / 37 / 2609.32355
Gradients for Interventions and Activations for Detection: Targeted Feature Learning in Language Models
干预和激活的梯度用于检测：语言模型中的目标特征学习
Abstract
Model-internal features can be studied through both their ability to identify a specified concept and their causal effect when manipulated, e.g., through steering or weight editing. A prominent approach to feature learning is Sparse Autoencoders (SAEs), which learn broad feature dictionaries whose relation to particular concepts is typically identified post hoc. However, many interpretability questions are instead hypothesis-driven and concern a concept specified in advance. We study this setting as targeted feature learning, where a single feature is constructed for such a predefined concept. We present a controlled comparison across three model signals (activation values, activation gradients, and parameter gradients) and two estimators (contrastive mean and a learned one-dimensional encoder-decoder), yielding six targeted methods, with CAA and GRADIEND as existing instances and four new methods covering the remaining combinations. We compare these methods against pretrained SAEs across 15 tasks and three language models, evaluating both detection and causal intervention. Across models, the strongest detection performance is achieved by contrastive activation value methods, whereas the strongest intervention performance is achieved by gradient-based methods. Overall, our results show that targeted feature quality depends jointly on the model signal and estimator, with detection and intervention capturing complementary properties.
Chinese Translation
模型内部特征可以通过其识别特定概念的能力以及在操控时的因果效应进行研究，例如通过引导或权重编辑。稀疏自编码器（Sparse Autoencoders, SAEs）是一种显著的特征学习方法，它学习广泛的特征字典，而这些特征与特定概念的关系通常是在事后识别的。然而，许多可解释性问题是假设驱动的，涉及预先指定的概念。我们将这种设置研究为目标特征学习，其中为预定义概念构建单一特征。我们在三个模型信号（激活值、激活梯度和参数梯度）和两个估计器（对比均值和学习的一维编码器-解码器）之间进行了受控比较，得出了六种目标方法，其中CAA和GRADIEND是现有实例，另外四种新方法涵盖了剩余的组合。我们在15个任务和三个语言模型上将这些方法与预训练的SAEs进行了比较，评估了检测和因果干预的效果。在各个模型中，最强的检测性能是通过对比激活值方法实现的，而最强的干预性能则是通过基于梯度的方法实现的。总体而言，我们的结果表明，目标特征的质量共同依赖于模型信号和估计器，检测和干预捕捉到互补的属性。
cs.CL / 38 / 2609.32382
PlurVA-LLM-2026 Shared Task Track-1: Pluralistic Value Alignment in LLMs via Multilingual Fine-Tuning and Threshold Calibration
PlurVA-LLM-2026 共享任务 Track-1：通过多语言微调和阈值校准实现大型语言模型中的多元价值对齐
Abstract
We present our system for the PlurVA-LLM 2026 Shared Task Track-1, which focuses on pluralistic value alignment in the contexts of China, Indonesia, and Sri Lanka. For this resource-constrained track, we fine-tuned Llama 3.1 8B Instruct using 4-bit QLoRA. Our approach combines option-permutation augmentation for Chinese data, annotator vote expansion for Indonesian data, and binary reformulation with SinhalaMMLU augmentation for Sri Lankan data. We further applied conditional threshold calibration to the predictions for the Sri Lankan data. The final system achieved accuracies of 0.785 for Chinese, 0.715 for Indonesian, and 0.916 for Sri Lankan, resulting in an overall macro-average accuracy of 0.805.
Chinese Translation
我们展示了针对 PlurVA-LLM 2026 共享任务 Track-1 的系统，该任务聚焦于中国、印度尼西亚和斯里兰卡的多元价值对齐。在这个资源受限的任务中，我们使用 4 位 QLoRA 对 Llama 3.1 8B Instruct 进行了微调。我们的方法结合了针对中文数据的选项排列增强、针对印度尼西亚数据的注释者投票扩展，以及针对斯里兰卡数据的二元重构与 SinhalaMMLU 增强。我们进一步对斯里兰卡数据的预测应用了条件阈值校准。最终系统在中文、印度尼西亚语和斯里兰卡语的准确率分别达到了 0.785、0.715 和 0.916，整体宏平均准确率为 0.805。
cs.CL / 39 / 2609.32396
FA-Bench: A Benchmark for Word-Level and Phone-Level Forced-Alignment and ASR Timestamps Under Clean and Noisy Conditions
FA-Bench：一个用于清晰和嘈杂条件下的词级和音素级强制对齐及自动语音识别时间戳的基准测试
Abstract
Forced alignment aligns speech audio with a text transcript to generate word and phone timestamps. Published comparisons normalize transcripts, split the data and match boundaries differently, so their numbers cannot be read together. We present FA-Bench, an open framework that fixes those choices once and releases the code, splits, phone mapping, text normalization and scoring script, with results published periodically. Track 1 gives every aligner the reference transcript and Track 2 gives it a recognizer's output, on the same audio, clean and degraded four ways, with 21 open models and 9 commercial APIs under a unified protocol. We score every boundary of an utterance and check the two labels beside it, so a word the recognizer missed or invented is charged. Using a tolerance-based F1 as our primary metric eliminates the 9% to 14% score inflation that standard MAE causes on recognition-dependent systems in conversational speech. We then group boundaries by their position and how many adjacent words were recognized correctly, which shows where a system lost the score. We discovered systematic bias in how current systems time words, with Whisper about 150 ms early and several commercial ASR APIs over 50 ms late. Code and results are at https://github.com/olewave/fa-bench
Chinese Translation
强制对齐将语音音频与文本转录对齐，以生成词和音素的时间戳。已发布的比较研究在转录文本的规范化、数据划分和边界匹配方面存在差异，因此其结果不能直接进行比较。我们提出了FA-Bench，这是一个开放框架，固定了这些选择并发布了代码、数据划分、音素映射、文本规范化和评分脚本，结果定期发布。Track 1 为每个对齐器提供参考转录文本，而 Track 2 则提供识别器的输出，均基于相同的音频，音频经过四种清晰和降级处理，使用21个开放模型和9个商业API，遵循统一协议。我们对每个话语的边界进行评分，并检查其旁边的两个标签，因此识别器遗漏或虚构的词会被计入分数。使用基于容忍度的F1作为我们的主要指标，消除了标准平均绝对误差（MAE）在对话语音的识别依赖系统中造成的9%至14%的分数膨胀。随后，我们根据边界的位置及其相邻词的正确识别数量对边界进行分组，这显示了系统失分的地方。我们发现当前系统在计时词语时存在系统性偏差，Whisper大约提前150毫秒，而多个商业ASR API则晚于50毫秒。代码和结果可在 https://github.com/olewave/fa-bench 获取。
cs.CL / 40 / 2609.32397
SinBrief: A Hybrid Framework for Abstractive Text Summarisation of Sinhala Legal Documents
SinBrief：一种用于僧伽罗法律文件的混合抽象文本摘要框架
Abstract
Legal document summarisation in low-resource languages presents significant challenges due to the scarcity of annotated data and the complexity of domain-specific terminology. This paper presents SinBrief, a hybrid abstractive summarisation framework for Sinhala legal documents that does not require human-annotated training data. The proposed framework combines domain-aware word graph construction with neural sentence scoring to generate abstractive summaries from Sinhala legal text. Five sentence scoring models are evaluated within the framework: mBert, Llama 3.1, Falcon 7B, Laser, and a continually pre-trained Llama model domain-adapted to Sinhala legal text. The framework is evaluated on a Sinhala legal corpus using reference-free metrics, including Coverage, Density, Compression Ratio, SummaC, and Self-BertScore. Experimental results demonstrate that SinBrief produces summaries with lower lexical overlap than extractive baselines while maintaining factual consistency, demonstrating the viability of hybrid, largely annotation-free abstractive summarisation for low-resource legal NLP tasks.
Chinese Translation
在低资源语言中，法律文件摘要面临着显著的挑战，这主要是由于注释数据的稀缺和领域特定术语的复杂性。本文提出了SinBrief，一种用于僧伽罗法律文件的混合抽象摘要框架，该框架不需要人工注释的训练数据。所提出的框架结合了领域感知的词图构建与神经句子评分，以从僧伽罗法律文本中生成抽象摘要。在该框架内评估了五种句子评分模型：mBert、Llama 3.1、Falcon 7B、Laser，以及一个不断预训练的Llama模型，该模型经过领域适应以适应僧伽罗法律文本。该框架在一个僧伽罗法律语料库上进行了评估，使用了无参考指标，包括覆盖率、密度、压缩比、SummaC和Self-BertScore。实验结果表明，SinBrief生成的摘要在词汇重叠方面低于提取基线，同时保持事实一致性，证明了针对低资源法律自然语言处理任务的混合、基本无注释的抽象摘要的可行性。
cs.CL / 41 / 2609.32401
Shared Worlds, Private Minds: Structured Memory for Long-Form Writing as World Creation
共享世界，私人思维：作为世界创造的长篇写作结构化记忆
Abstract
LLM agents that write long-form fiction need an explicit memory of the evolving storyworld to keep new events consistent with established facts. Such memory must keep heterogeneous narrative information distinct, integrate story developments across granularities, and recover dependencies that a writing request leaves implicit. We present NarraWorld, a structured memory system for long-form writing that treats memory construction as world creation. From a shared evidence-grounded graph, NarraWorld derives four connected views: world facts, per-character beliefs, open developments, and hypothetical branches (possible-world continuations). Hierarchical aggregation with atomic closure consolidates events into scenes, plotlines, and plots, keeping each higher-level node traceable to its constituent source spans. For retrieval, planned reconstruction infers a query's dependencies from the current narrative situation and a preview of memory, then assembles the relevant records within a token budget. Across three writing benchmarks, NarraWorld achieves the strongest aggregate results. Its memory also transfers to situated role-playing and largely preserves recall on a general-purpose long-term memory benchmark, paving the way for agents that sustain coherent storyworlds across diverse narrative tasks.
Chinese Translation
撰写长篇小说的LLM（大语言模型）代理需要对不断发展的故事世界有明确的记忆，以保持新事件与既定事实的一致性。这种记忆必须将异质叙事信息区分开，整合不同粒度的故事发展，并恢复写作请求中隐含的依赖关系。我们提出了NarraWorld，一个用于长篇写作的结构化记忆系统，将记忆构建视为世界创造。从一个共享的基于证据的图中，NarraWorld导出了四个相互关联的视图：世界事实、每个角色的信念、开放的发展和假设的分支（可能的世界延续）。通过原子闭合的层次聚合，将事件整合为场景、情节线和情节，确保每个更高层次的节点可追溯到其组成的源跨度。在检索方面，计划重建根据当前叙事情况和记忆预览推断查询的依赖关系，然后在令牌预算内组装相关记录。在三个写作基准测试中，NarraWorld取得了最强的综合结果。其记忆还可以转移到情境角色扮演中，并在通用长期记忆基准测试中大幅保持回忆，为在多样叙事任务中维持连贯故事世界的代理铺平了道路。
cs.CL / 42 / 2609.32408
Automatic Speech Recognition for the Basa\`{a} Language: A Low-Resource Approach
巴萨语的自动语音识别：一种低资源方法
Abstract
The rapid advancement of Artificial Intelligence (AI) and Natural Language Processing (NLP) has revolutionized the way humans interact with machines. Among the most impactful developments is Automatic Speech Recognition (ASR), which enables computers to convert spoken language into text. Systems such as those built on deep neural networks, transformer architectures, and self-supervised learning have achieved near-human performance for well-resourced languages such as English and French. Yet, these advances have disproportionately benefited a small fraction of the world's languages.
Chinese Translation
人工智能（AI）和自然语言处理（NLP）的快速发展彻底改变了人类与机器的互动方式。其中，自动语音识别（ASR）是最具影响力的进展之一，它使计算机能够将口语转换为文本。基于深度神经网络、变换器架构和自监督学习等技术构建的系统在英语和法语等资源丰富的语言上达到了接近人类的表现。然而，这些进展在很大程度上惠及了世界上仅有的一小部分语言。
cs.CL / 43 / 2609.32431
DualGuard: Dual-Mode Quality Control for Logic-Preserving Data Augmentation
DualGuard：双模式逻辑保持数据增强的质量控制
Abstract
Large language models provide a practical way to generate augmented data for logical reasoning at scale, but a larger generation volume does not guarantee semantic, label, or logical reliability. Existing work has improved generation quality through generation constraints, candidate validation, filtering, and feedback-based revision; however, once a quality judgment is available, deciding whether a candidate should be retained, filtered, or repaired remains an important control problem. We propose DualGuard, a dual-mode quality-control framework for logic-preserving data augmentation. The first mode uses the current instance and candidate batch for selective retention, filtering, attribution, and targeted feedback. The second mode accumulates cross-instance execution records of augmentation actions on top of per-sample diagnosis and attribution, compares new executions against each action's own historical behavior, and supports retrospective anomaly inspection, targeted rollback, and bounded repair. Both modes share semantic verification and additionally use symbolic verification when a reliable logical form is available. Across seven downstream tasks in the Two-Stage Transfer setting, DualGuard achieves the highest Accuracy on five tasks and outperforms the no-augmentation BERT baseline on all seven. Controlled ablations further show complementary roles for Memory, Z3, and history-aware anomaly control.
Chinese Translation
大型语言模型为大规模逻辑推理生成增强数据提供了一种实用的方法，但更大的生成量并不保证语义、标签或逻辑的可靠性。现有研究通过生成约束、候选验证、过滤和基于反馈的修订提高了生成质量；然而，一旦有了质量判断，决定是否保留、过滤或修复候选仍然是一个重要的控制问题。我们提出了DualGuard，一个用于逻辑保持数据增强的双模式质量控制框架。第一种模式利用当前实例和候选批次进行选择性保留、过滤、归因和针对性反馈。第二种模式在每个样本的诊断和归因基础上，累积增强操作的跨实例执行记录，将新的执行与每个操作的历史行为进行比较，并支持回顾性异常检查、针对性回滚和有界修复。这两种模式共享语义验证，并在可用可靠逻辑形式时额外使用符号验证。在两阶段迁移设置下的七个下游任务中，DualGuard在五个任务上实现了最高的准确率，并在所有七个任务中超越了无增强的BERT基线。受控消融实验进一步显示了Memory、Z3和历史感知异常控制的互补作用。
cs.CL / 44 / 2609.32445
Masking Frequent Tokens Sharpens Direct Preference Optimization
掩蔽频繁标记锐化直接偏好优化
Abstract
Direct Preference Optimization (DPO) aligns language models by optimizing over sequence-level sums of token-wise implicit reward differences. However, we identify a pervasive pathology in this formulation: a disproportionately small subset of high-frequency token types dominates cumulative sequence scores while appearing symmetrically across both preferred and dispreferred responses. Specifically, under canonical Qwen tokenization on Anthropic HH-RLHF, merely 69 token types account for $55.1\%$ of all response tokens and $85.9\%$ of within-pair shared token mass, exhibiting substantially lower preference-side specificity than the remaining vocabulary. This symmetric ubiquity induces gradient entanglement and dilutes the discriminative preference signal propagated through the objective. To resolve this issue, we introduce \emph{Anisotropic DPO} (\textsf{ADPO}) and its canonical realization, \emph{Frequency-Hard DPO}. Using a fixed, label-agnostic vocabulary mask, our method zeroes the implicit reward contribution of high-frequency response tokens while assigning unit weight to informative positions, thereby suppressing gradient interference without modifying preference pairs, discarding context, or introducing learned parameters. Here, \emph{anisotropy} designates non-uniform token-level objective weighting rather than representational geometry. Extensive empirical evaluations on AlpacaEval, MT-Bench, and Arena-Hard demonstrate that Frequency-Hard DPO consistently outperforms standard DPO across Qwen-2.5-7B-Instruct and Llama-3-8B-Instruct, establishing that selectively masking shared high-frequency tokens offers an effective, zero-overhead mechanism for robust preference alignment.
Chinese Translation
直接偏好优化（Direct Preference Optimization, DPO）通过优化标记间隐含奖励差异的序列级总和来对齐语言模型。然而，我们发现这种公式中存在一种普遍的病态：一个比例过小的高频标记类型子集主导了累积序列得分，同时在偏好和非偏好响应中对称出现。具体而言，在Anthropic HH-RLHF上使用标准Qwen标记化时，仅69种标记类型就占据了所有响应标记的55.1%和配对共享标记质量的85.9%，其偏好侧特异性显著低于其余词汇。这种对称的普遍性导致了梯度纠缠，并稀释了通过目标传播的区分性偏好信号。为了解决这个问题，我们引入了 extit{各向异性DPO}（Anisotropic DPO, extsf{ADPO}）及其标准实现 extit{频率硬DPO}（Frequency-Hard DPO）。通过使用固定的、与标签无关的词汇掩蔽，我们的方法将高频响应标记的隐含奖励贡献归零，同时对信息丰富的位置赋予单位权重，从而在不修改偏好对、丢弃上下文或引入学习参数的情况下抑制梯度干扰。在这里， extit{各向异性}指的是非均匀的标记级目标加权，而不是表示几何。对AlpacaEval、MT-Bench和Arena-Hard的广泛实证评估表明，频率硬DPO在Qwen-2.5-7B-Instruct和Llama-3-8B-Instruct上始终优于标准DPO，确立了选择性掩蔽共享高频标记作为一种有效的、零开销的稳健偏好对齐机制。
cs.CL / 45 / 2609.32449
Self-Reports Do Not Identify Self-Models: An Identifiability Test for Counterfactual Reports
自我报告并不能识别自我模型：反事实报告的可识别性测试
Abstract
Language-model self-reports are evidence about behavior in a prompt environment, not by themselves evidence of a self-model. We investigate counterfactual reports about affect-like states under activation interventions and ask whether the report remains bound to the named intervention when the demonstration environment changes. Across three open instruction models, wrong-source demonstrations move reports toward the source answer family, while explicit mechanism binding reduces this pull. Self-report benchmarks should include environment-shift invariance tests under fixed intervention before treating accuracy as evidence for an autonomous report mechanism.
Chinese Translation
语言模型自我报告是关于在提示环境中行为的证据，但本身并不是自我模型的证据。我们研究了在激活干预下关于情感类状态的反事实报告，并询问当演示环境发生变化时，报告是否仍然与所命名的干预绑定。在三个开放指令模型中，错误来源的演示使报告向源答案家族移动，而显式机制绑定则减少了这种吸引力。自我报告基准应包括在固定干预下的环境转变不变性测试，才能在将准确性视为自主报告机制的证据之前进行处理。
cs.CL / 46 / 2609.32452
What Should the Reflector See? An Empirical Study of Evidence in Reflective Prompt Optimization
反射器应该看到什么？反射性提示优化中的证据实证研究
Abstract
Reflective prompt optimization revises instructions using examples of a model's behavior, but which evidence the reflector should receive remains unclear. We study evidence composition, visibility of examples, candidate selection and domain-knowledge policy within a single-parent Pareto-guided search. Using Qwen3.5-9B as both task model and reflector, we evaluate nine reflection strategies on five datasets. From these experiments, we find three distinct patterns. For performance improvement, Failures-only produces the largest mean test gain (+8.0 percentage points), while Balanced-mix and No-examples+Val share the best mean performance rank. For reflection effectiveness, No-examples achieves the best rank for improving sampled parents, yet yields only a 1.4-point mean test gain: local reflection success does not necessarily produce a stronger final prompt. For overfitting assessment, Failures-only and Balanced-mix share the lowest mean calibration-gap rank, while larger gaps on GPQA and IFBench show that calibration gains can overstate held-out improvement. This gap is a descriptive indicator, not a direct measure of overfitting. Together, these results show why reflection strategies should be assessed separately on final performance, parent improvement and calibration-to-test transfer.
Chinese Translation
反射性提示优化通过使用模型行为的示例来修订指令，但反射器应该接收哪些证据仍不清楚。我们研究了证据组成、示例的可见性、候选选择和领域知识策略在单父Pareto引导搜索中的应用。以Qwen3.5-9B作为任务模型和反射器，我们在五个数据集上评估了九种反射策略。通过这些实验，我们发现了三种不同的模式。在性能提升方面，Failures-only产生了最大的平均测试增益（+8.0个百分点），而Balanced-mix和No-examples+Val共享最佳平均性能排名。在反射有效性方面，No-examples在改善采样父代方面获得了最佳排名，但仅产生了1.4点的平均测试增益：局部反射成功并不一定会产生更强的最终提示。在过拟合评估方面，Failures-only和Balanced-mix共享最低的平均校准差距排名，而在GPQA和IFBench上较大的差距表明校准增益可能会夸大保留改进。这个差距是一个描述性指标，而不是过拟合的直接测量。综合这些结果，表明反射策略应在最终性能、父代改进和校准到测试转移方面单独评估。
cs.CL / 47 / 2609.32458
Streamlined Reflective Evolution for Task-Adaptive Self-Refinement Pipelines
任务自适应自我精炼管道的简化反思演化
Abstract
Reflective prompt optimization improves large language model (LLM) systems without updating model weights, but fixed architectures constrain how self-refinement is organized. We introduce Workflow-Designing Agents (WDA), a framework for streamlined reflective evolution of task-adaptive self-refinement pipelines. Starting from a minimal prompt, WDA jointly evolves stage instructions and their sequential structure. During evolution, we find that repeated revisions can accumulate redundant instructions in a single prompt. In WDA, we propose to address this problem with SPLIT, which redistributes these instructions across specialized stages. Three-example reflection and local screening guide selective search, while calibration scores guide Pareto admission and rollback of unhelpful trailing updates. The resulting pipelines are task-adaptive: their instructions and depth are learned from task data, then fixed for all test inputs within that task. We evaluate WDA on five benchmarks spanning knowledge, mathematical reasoning, multi-hop question answering, and instruction following. On Qwen3.5-9B, WDA achieves an average score of 51.24%, improving over the initial solver by 8.63 percentage points and the variant without SPLIT by 2.60 points. On GPT-4.1-mini, it achieves 49.00%, with corresponding gains of 5.62 and 3.69 points. These results support task-adaptive self-refinement as a complementary direction to broader agentic workflow search.
Chinese Translation
反思性提示优化在不更新模型权重的情况下改善大型语言模型（LLM）系统，但固定的架构限制了自我精炼的组织方式。我们引入了工作流设计代理（Workflow-Designing Agents, WDA），这是一个用于任务自适应自我精炼管道的简化反思演化框架。从一个最小提示开始，WDA共同演化阶段指令及其顺序结构。在演化过程中，我们发现重复的修订可能会在单个提示中积累冗余指令。在WDA中，我们提出通过SPLIT来解决这个问题，SPLIT将这些指令重新分配到专门的阶段。三个示例反思和局部筛选指导选择性搜索，而校准分数则指导帕累托准入和不利后续更新的回滚。最终生成的管道是任务自适应的：它们的指令和深度是从任务数据中学习的，然后在该任务的所有测试输入中固定。我们在五个基准测试上评估WDA，这些基准涵盖知识、数学推理、多跳问答和指令遵循。在Qwen3.5-9B上，WDA的平均得分为51.24%，比初始求解器提高了8.63个百分点，比没有SPLIT的变体提高了2.60个百分点。在GPT-4.1-mini上，得分为49.00%，相应的增益为5.62和3.69个百分点。这些结果支持任务自适应自我精炼作为更广泛的代理工作流搜索的补充方向。
cs.CL / 48 / 2609.32472
AdaTutoRank: Learning to Rerank Document Sets via Adaptive Tutoring Optimization for RAG and Deep Research
AdaTutoRank：通过自适应辅导优化学习重新排序文档集以适应RAG和深度研究
Abstract
Document rerankers determine what evidence reaches the downstream model in RAG and deep research, yet mainstream rerankers select by relevance matching, and individually relevant documents rarely constitute the complete, complementary, non-redundant set a complex information need demands. Prior work rewards a set by its aggregate rubric score, shifting the objective from ranking documents to composing sets. Yet that score is one scalar shared by every document in the set, so the supervision is sparse: a redundant document is rewarded with the rest whenever the set scores well, and a decisive one penalized with the rest whenever it does not; credit assignment leaves contributors indistinguishable from free riders. On-policy distillation could densify this supervision, but existing methods give every rollout the same fixed guidance, too prescriptive for strong rollouts and too abstract for weak ones. We therefore propose AdaTutoRank, a setwise reranker trained with Adaptive Tutoring Optimization (ATO) under a three-level hierarchy of nine rubric dimensions, which supplies silver labels for the cold start, rewards for reinforcement learning, and hints for distillation. ATO draws three hint forms of increasing specificity from the policy's own frozen snapshot: the rubrics alone, a self-selector's sibling-set chosen under rubrics, and a self-reflector's reflection contrasting the rollout with that sibling-set; each rollout receives the form matched to its quality. Re-scoring that rollout under the hint-conditioned frozen teacher and the hint-free snapshot distills the hint's effect into a token-level advantage that complements the group-relative outcome advantage. Across ten benchmarks spanning RAG, deep research, and setwise evaluation, AdaTutoRank attains the best overall performance while issuing fewer retrieval calls.
Chinese Translation
文档重新排序器决定了哪些证据能够传递给RAG和深度研究中的下游模型，但主流的重新排序器主要通过相关性匹配进行选择，而单个相关文档很少构成复杂信息需求所要求的完整、互补且无冗余的集合。先前的研究通过其整体评分来奖励一个集合，将目标从对文档的排序转变为对集合的构成。然而，该评分是一个在集合中所有文档共享的标量，因此监督稀疏：当集合评分良好时，冗余文档与其他文档一起获得奖励，而当评分不佳时，决定性文档则受到惩罚；信用分配使得贡献者与搭便车者难以区分。基于策略的蒸馏可以增强这种监督，但现有方法对每次回滚提供相同的固定指导，这对于强回滚过于规范，对于弱回滚则过于抽象。因此，我们提出了AdaTutoRank，这是一种通过自适应辅导优化（Adaptive Tutoring Optimization，ATO）训练的集合重新排序器，采用九个评分维度的三级层次结构，为冷启动提供银标签，为强化学习提供奖励，并为蒸馏提供提示。ATO从策略自身的冻结快照中提取三种逐渐具体化的提示形式：仅评分标准、在评分标准下选择的自选择者的兄弟集合，以及自反射者的反思，比较回滚与该兄弟集合的差异；每次回滚接收与其质量相匹配的提示形式。在提示条件下的冻结教师和无提示快照下重新评分该回滚，将提示的效果蒸馏为一个补充群体相对结果优势的令牌级别优势。在涵盖RAG、深度研究和集合评估的十个基准测试中，AdaTutoRank实现了最佳的整体性能，同时发起了更少的检索调用。
cs.CL / 49 / 2609.32474
PC-SubMax: Efficient Prompt Compression via Regularized Submodular Maximization
PC-SubMax：通过正则化子模最大化实现高效提示压缩
Abstract
While large language models (LLMs) are increasingly deployed in long-context scenarios, lengthy prompts can increase inference costs and latency and exacerbate the ``lost-in-the-middle'' phenomenon. Selective prompt compression offers a model-agnostic approach to alleviating these issues. However, methods based on fixed token- or sentence-level importance scores may overlook how content contributions change with the selected subset, limiting their ability to account for inter-sentence redundancy. Compression procedures that rely on autoregressive LLM scoring can also introduce substantial overhead. We propose PC-SubMax, a theoretically grounded framework that formulates selective prompt compression as regularized monotone submodular maximization under a knapsack constraint. The objective is $U(S)-\ell(S)$, where the monotone submodular utility $U$ combines information coverage, query relevance, and log-determinant diversity, and the non-negative modular penalty $\ell$ captures token cost. Through diminishing marginal returns, the objective evaluates each sentence's contribution relative to the selected content. To optimize this objective, we develop the Regularized Greedy+Max (RGM) algorithm, which deterministically returns a feasible set $Q$ satisfying $U(Q)-\ell(Q)\geq \frac{1}{2}U(O)-\ell(O)$, where $O$ is an optimal feasible solution to the regularized problem. RGM uses $O(n\kappa)$ value-oracle queries, where $n$ is the number of candidate sentences and $\kappa$ is the maximum feasible subset size. PC-SubMax uses encoder representations and avoids autoregressive LLM scoring during compression. Experiments across seven diverse benchmarks demonstrate competitive downstream performance with low compression overhead.
Chinese Translation
随着大型语言模型（LLMs）在长上下文场景中的广泛应用，冗长的提示会增加推理成本和延迟，并加剧“迷失在中间”的现象。选择性提示压缩提供了一种与模型无关的方法来缓解这些问题。然而，基于固定的令牌或句子重要性评分的方法可能忽视了内容贡献随着所选子集的变化，从而限制了其考虑句子间冗余的能力。依赖自回归LLM评分的压缩过程也可能引入相当大的开销。我们提出了PC-SubMax，这是一种理论基础的框架，将选择性提示压缩形式化为在背包约束下的正则化单调子模最大化。目标是 $U(S)- ext{ell}(S)$，其中单调子模效用 $U$ 结合了信息覆盖、查询相关性和对数行列式多样性，而非负的模罚 $ ext{ell}$ 捕捉了令牌成本。通过递减的边际收益，该目标评估每个句子相对于所选内容的贡献。为了优化该目标，我们开发了正则化贪婪+最大化（Regularized Greedy+Max, RGM）算法，该算法确定性地返回一个可行集 $Q$，满足 $U(Q)- ext{ell}(Q)
geq rac{1}{2}U(O)- ext{ell}(O)$，其中 $O$ 是正则化问题的最优可行解。RGM 使用 $O(n ext{kappa})$ 值预言机查询，其中 $n$ 是候选句子的数量，$ ext{kappa}$ 是最大可行子集的大小。PC-SubMax 使用编码器表示，并在压缩过程中避免自回归LLM评分。跨越七个不同基准的实验表明，PC-SubMax 在保持低压缩开销的同时，展示了具有竞争力的下游性能。
cs.CL / 50 / 2609.32491
Explaining Textual Entailment with Lexical Entailments: Using LLMs to Supply Lexical Relations for Formal Proofs
通过词汇蕴涵解释文本蕴涵：利用大型语言模型提供形式证明所需的词汇关系
Abstract
Large Language Models (LLMs) are highly capable of natural language reasoning and appear to store a great deal of lexical knowledge, but it is still unclear how much of this knowledge they actually use when reasoning, and whether they use it in the right way. On the other hand, logic-based Natural Language Inference (NLI) systems provide transparent and formally grounded reasoning, but they need to be supplied with rich lexical knowledge to prove inferences beyond purely logical ones. In this paper, we evaluate whether LLMs can identify all lexical knowledge needed to solve NLI problems and how much this knowledge contributes to proof search in a logic-based NLI system. Our research focuses exclusively on structured lexical entailments (e.g., chinchilla$\sqsubseteq$small animal) as a proxy for structured explanations for NLI problems with an entailment label. First, we curate a dataset for a new task of explaining sentential entailments with a set of lexical entailments. The dataset is used to intrinsically evaluate LLMs on generating structured lexical explanations. Then, we use NLI as an extrinsic evaluation in a simple neuro-symbolic setting, assessing whether LLMs can supply sufficient lexical relations to LangPro, a natural-logic theorem prover for natural language. The results show that the proposed task remains challenging even for hosted proprietary LLMs, and that their contribution to theorem proving is moderate: generated relations are often only partially sound and may be tailored to the specific NLI problem rather than representing generally valid lexical knowledge.
Chinese Translation
大型语言模型（LLMs）在自然语言推理方面表现出色，并似乎存储了大量的词汇知识，但尚不清楚它们在推理时实际使用了多少这些知识，以及是否以正确的方式使用。另一方面，基于逻辑的自然语言推理（NLI）系统提供透明且形式化的推理，但需要丰富的词汇知识来证明超出纯逻辑推理的推论。本文评估了LLMs是否能够识别解决NLI问题所需的所有词汇知识，以及这些知识对基于逻辑的NLI系统中的证明搜索的贡献。我们的研究专注于结构化的词汇蕴涵（例如，chinchilla$acksim$small animal），作为具有蕴涵标签的NLI问题的结构化解释的代理。首先，我们为解释句子蕴涵的新任务策划了一个数据集，包含一组词汇蕴涵。该数据集用于对LLMs在生成结构化词汇解释方面进行内在评估。然后，我们在一个简单的神经符号设置中使用NLI作为外在评估，评估LLMs是否能够为LangPro（一个自然逻辑定理证明器）提供足够的词汇关系。结果表明，即使对于托管的专有LLMs，所提出的任务仍然具有挑战性，并且它们对定理证明的贡献是适度的：生成的关系往往只是部分有效，并可能针对特定的NLI问题进行调整，而不是代表普遍有效的词汇知识。
cs.CL / 51 / 2609.32496
Locally Sound, Globally Insufficient: The Local-Global Gap in Multi-Hop Reasoning
局部可靠、全球不足：多跳推理中的局部-全球差距
Abstract
Reliable multi-hop reasoning requires more than locally supported steps: a trace can be sound at every reasoning step yet still fail to answer the question as a whole. We call this failure regime the local-global gap (LGG), in which the trace is locally sound yet globally insufficient. Local soundness requires each step to be supported by the available evidence and preceding steps, whereas global sufficiency requires the reasoning trace to align with the question and establish the submitted answer. In a human-adjudicated diagnostic of 2,598 responses across three multi-hop QA benchmarks and three models, we find that the LGG occurs in every benchmark-model combination and accounts for nearly half of globally insufficient responses overall. However, conventional faithfulness verifiers that check claims against the evidence largely miss these failures: at thresholds retaining at least 95% of reliable traces, recall for LGG cases is substantially lower than that for locally unsound traces. To address these failures, we formalize three dependencies for reliable reasoning: evidence-to-step support, question-to-trace alignment, and trace-to-answer closure. Instead of post-hoc diagnosis, we introduce E-Closure to supervise these dependencies during training, combining generation supervision on supported original and counterfactual responses with bidirectional switching constraints. Averaged over three benchmarks and three backbones, existing fine-tuning baselines improve accuracy and local soundness over the base models, but at the cost of global sufficiency. E-Closure improves both: among all fine-tuned methods, it achieves the highest average accuracy (92.8%) and trace reliability (89.0%) while yielding the lowest LGG rate (6.2%).
Chinese Translation
可靠的多跳推理不仅需要局部支持的步骤：一个推理过程在每个推理步骤上都可以是合理的，但仍然无法整体回答问题。我们将这种失败机制称为局部-全球差距（Local-Global Gap，LGG），在这种情况下，推理过程在局部上是合理的，但在全球上是不足的。局部可靠性要求每一步都得到可用证据和前一步的支持，而全球充分性要求推理过程与问题对齐并确立提交的答案。在对2598个响应进行人类评估的诊断中，涵盖了三个多跳问答基准和三个模型，我们发现LGG在每个基准-模型组合中都存在，并且占据了总体上近一半的全球不足响应。然而，传统的忠实性验证器在检查声明与证据时，往往忽视了这些失败：在保留至少95%可靠推理过程的阈值下，LGG案例的召回率显著低于局部不可靠推理过程的召回率。为了解决这些失败，我们形式化了可靠推理的三个依赖关系：证据到步骤的支持、问题到推理过程的对齐，以及推理过程到答案的闭合。我们引入E-Closure来在训练期间监督这些依赖关系，而不是事后诊断，结合对支持的原始和反事实响应的生成监督与双向切换约束。在三个基准和三个基础模型上进行平均，现有的微调基线在准确性和局部可靠性上相较于基础模型有所提高，但以牺牲全球充分性为代价。E-Closure同时改善了这两者：在所有微调方法中，它达到了最高的平均准确率（92.8%）和推理过程可靠性（89.0%），同时产生了最低的LGG率（6.2%）。
cs.CL / 52 / 2609.32499
Learning an Anchored Prompt Space for Continual Adaptation of Large Language Models
学习锚定提示空间以持续适应大型语言模型
Abstract
Continually adapting large language models requires acquiring new knowledge while preserving previously learned capabilities. Jointly adapting model parameters and task-specific soft prompts offers a promising solution, but faces two key limitations: historical prompts may become less effective as the model evolves, while their transferable cross-task relationships are not explicitly learned. We propose Learning an Anchored Prompt Space (LAPS), which preserves historical prompt effectiveness and learns relationships among task-specific soft prompts to facilitate positive transfer. LAPS first aligns historical prompts with the updated model through self-distillation. LAPS then constructs an anchored prompt space whose vertices correspond to learned task-specific soft prompts and whose intermediate geometry is shaped by learnable B\'ezier control prompts. Once this anchored prompt space is learned, LAPS identifies the best-performing prompt for each task on its validation set, allowing the optimized prompt to draw on knowledge acquired from observed tasks. Experiments on the TRACE benchmark across three Qwen3 model scales show that LAPS consistently outperforms distillation-based, prompt-based, and joint prompt--parameter adaptation baselines, improving average performance while reducing forgetting.
Chinese Translation
持续适应大型语言模型需要在获取新知识的同时保留先前学习的能力。联合调整模型参数和任务特定的软提示提供了一种有前景的解决方案，但面临两个主要限制：随着模型的发展，历史提示的有效性可能降低，而它们在跨任务之间的可转移关系并未被明确学习。我们提出了学习锚定提示空间（Learning an Anchored Prompt Space, LAPS），该方法保留历史提示的有效性，并学习任务特定软提示之间的关系以促进积极迁移。LAPS首先通过自蒸馏将历史提示与更新后的模型对齐。然后，LAPS构建一个锚定提示空间，其顶点对应于学习到的任务特定软提示，而其中间几何形状由可学习的Bézier控制提示塑造。一旦学习了这个锚定提示空间，LAPS就会在每个任务的验证集上识别表现最佳的提示，从而使优化后的提示能够利用从观察到的任务中获得的知识。在针对TRACE基准的实验中，涵盖三种Qwen3模型规模，LAPS始终优于基于蒸馏、基于提示和联合提示-参数适应的基线，提升了平均性能，同时减少了遗忘。
cs.CL / 53 / 2609.32520
When Users Change Their Minds: Measuring and Repairing Intent Drift in LLM Agents
当用户改变主意时：测量和修复大型语言模型代理中的意图漂移
Abstract
LLM agents often operate over multi-turn interactions in which user intent changes before execution. We study intent drift: the failure mode in which superseded parts of the user's intent continue to influence the final answer or tool action. We introduce IntentFlux, an executable benchmark that converts verifiable tasks into dialogues with controlled intent changes while preserving their original graders. In a 627-case calibration, mean task score falls from 0.476 to 0.384 as dialogues contain more superseded and withdrawn information. Across eight models, the rate of fully correct solutions is significantly lower when the same final task must be recovered from an evolving dialogue rather than given directly in a single turn. We further introduce StateForge, which explicitly maintains the active requirements before generation. On General-Test, it improves mean task score from 0.367 to 0.467. Providing the ground-truth final state improves performance further but still does not recover single-turn performance, indicating that state-estimation errors explain only part of the gap. These results establish intent drift as a measurable multi-turn failure mode and explicit state maintenance as a partial mitigation.
Chinese Translation
大型语言模型（LLM）代理通常在多轮交互中操作，其中用户的意图在执行前发生变化。我们研究了意图漂移：一种失败模式，在这种模式下，用户意图中被取代的部分继续影响最终答案或工具操作。我们引入了 IntentFlux，这是一个可执行的基准，能够将可验证的任务转换为对话，并在保持原始评分者的同时控制意图变化。在627个案例的校准中，随着对话中包含更多被取代和撤回的信息，平均任务得分从0.476降至0.384。在八个模型中，当必须从一个不断演变的对话中恢复相同的最终任务时，完全正确解决方案的比例显著降低。我们进一步引入了 StateForge，它在生成之前明确维护活动要求。在 General-Test 上，它将平均任务得分从0.367提高到0.467。提供真实的最终状态进一步提高了性能，但仍未恢复单轮性能，这表明状态估计错误仅解释了部分差距。这些结果确立了意图漂移作为一种可测量的多轮失败模式，并将显式状态维护作为部分缓解措施。
cs.CL / 54 / 2609.32556
Language as an Independent Information Layer: A Conceptual Model of Communication, Cognition and Decision-Making
语言作为独立的信息层：沟通、认知与决策的概念模型
Abstract
Based on an analysis of the role of language in thought and communication, this article proposes a new concept for designing corporate knowledge bases. The concept integrates the probabilistic vector space of a corporate vocabulary, reflect-ing industry specifics, subject focus, terminology, and culture, with traditional ontological modeling. This combination enables the efficient extraction of knowledge from accumulated corporate documents while strictly accounting for specific business processes. Consequently, this concept bridges statistical and semantic (cause-and-effect) methodologies. Furthermore, analyzing the projec-tions of probabilistic spaces and causal relationships can help identify bottlenecks in business logic. As a dynamic system, language functions as a separate, inde-pendent layer within the overall information architecture. Introducing a dynamic component into the probabilistic space of word distribution allows it to be mod-eled as a multidimensional solution space for various problem formulations. In this context, input data defining the problem conditions serve as control parame-ters for dynamic transformations.
Chinese Translation
基于对语言在思维和沟通中作用的分析，本文提出了一种新的企业知识库设计概念。该概念将反映行业特性、主题重点、术语和文化的企业词汇的概率向量空间与传统本体建模相结合。这种结合能够高效提取积累的企业文档中的知识，同时严格考虑特定的业务流程。因此，该概念在统计和语义（因果）方法论之间架起了桥梁。此外，分析概率空间和因果关系的投影可以帮助识别业务逻辑中的瓶颈。作为一个动态系统，语言在整体信息架构中作为一个独立的层次运作。将动态组件引入词分布的概率空间，使其能够被建模为各种问题表述的多维解空间。在这个背景下，定义问题条件的输入数据作为动态转化的控制参数。
cs.CL / 55 / 2609.32560
How to Reduce Whisper Hallucination
如何减少低语幻觉
Abstract
Whisper is still what runs in production: one permissively licensed checkpoint, 99 languages, no per-language tuning. But it writes sentences nobody said. On 42 clips of pure room tone, whisper-large-v3 emits words on 61.9% of them and emits something on 100%. The usual response is to distil a student from Whisper pseudo-labels, filtering the hallucinations out of the corpus first, which deletes the evidence while leaving the behaviour in place: the student is fitted to the subset where the teacher was right, and inherits a failure mode absent from its own training data. The second response, adding non-speech audio so the model learns to stay quiet, already ships and works, but teaches suppression without discrimination. The checkpoint best on every non-speech arm here is also the one that recovers the fewest genuinely spoken phrases and deletes 58% of repeated speech. The teacher has to be fixed, with both halves of the signal. Our benchmark scores both at once: 11,852 clips over eight arms, 6,267 synthetic positives, 8,296 clips of real audio that made a production model fail, and FLEURS in 58 languages. We collect 40,891 hallucination phrases in 100 languages, choose a text-to-speech system by measurement, and synthesise those phrases as positives, so the model meets the same text both as something to suppress and as something to transcribe. Across 33 matched pairs of fine-tunes differing only by those positives, adding them lowers word emission on real voice-free audio in 31 pairs and raises phrase recovery in 32; among the 30 pairs that hold English accuracy it is 30 out of 30. The best checkpoint takes hallucination on silence from 61.9% to 2.4% and words over real voice-free audio from 99.9% to 47.8%, while raising phrase recovery from 69.8% to 82.7%. The benchmark, lexicon and synthetic corpus are released at https://huggingface.co/datasets/Scicom-intl/Whisper-Hallucination
Chinese Translation
Whisper 仍然是生产中使用的模型：一个宽松许可的检查点，支持 99 种语言，无需针对每种语言进行调优。但它生成了没人说过的句子。在 42 段纯房间音频中，whisper-large-v3 在 61.9% 的情况下发出单词，并在 100% 的情况下发出某种声音。通常的应对方式是从 Whisper 的伪标签中提炼出一个学生，首先过滤掉语料库中的幻觉，这样做删除了证据，却保留了行为：学生只适应了教师正确的子集，并继承了在其训练数据中缺失的失败模式。第二种应对方式是添加非语音音频，以便模型学习保持安静，这种方法已经投入使用并有效，但教会了抑制而非区分。这里在每个非语音任务上表现最佳的检查点也是恢复最少真实发声短语的那个，并删除了 58% 的重复语音。教师必须修正，信号的两个部分都需要处理。我们的基准同时评估这两者：在八个任务上共评估 11,852 段音频，6,267 段合成的正样本，8,296 段真实音频导致生产模型失败，以及 58 种语言的 FLEURS。我们收集了 40,891 段幻觉短语，涵盖 100 种语言，通过测量选择一个文本到语音系统，并将这些短语合成作为正样本，以便模型将相同的文本视为需要抑制的内容和需要转录的内容。在 33 对仅因这些正样本而不同的微调中，添加它们降低了 31 对中真实无声音频的单词发出率，并提高了 32 对中的短语恢复率；在 30 对保持英语准确性的对中，结果为 30/30。最佳检查点将静音下的幻觉从 61.9% 降低到 2.4%，在真实无声音频下的单词发出率从 99.9% 降低到 47.8%，同时将短语恢复率从 69.8% 提高到 82.7%。基准、词汇表和合成语料库已发布在 https://huggingface.co/datasets/Scicom-intl/Whisper-Hallucination
cs.CL / 56 / 2609.32577
Groupwise Agentic Grading and Advantage Redistribution for Code Agent RL
代码代理强化学习中的组别代理评分与优势再分配
Abstract
Reinforcement learning (RL) for code agents often uses executable tests to provide binary rewards. With these rewards, Group Relative Policy Optimization (GRPO) assigns identical advantages to test-passing trajectories within each rollout group, overlooking differences in implementation quality and adherence to task requirements. This leaves the policy without a learning signal that favors clean, targeted implementations over those containing unnecessary or out-of-scope changes. We introduce GAGAR, a framework for quality-aware credit redistribution in code agent RL. Built on dynamic sampling that retains groups containing both passing and failing trajectories, GAGAR places all trajectories from each group in a shared workspace, where an SFT-trained agentic grader jointly inspects them and ranks the test-passing candidates. Based on this ranking, we downweight lower-ranked trajectories and proportionally rescale the advantages of all test-passing trajectories to restore their original sum. This sum-preserving redistribution retains the relative weights established by quality-based downweighting while shifting credit toward higher-quality implementations. We evaluate GAGAR at industrial scale using pre-RL SFT checkpoints of MiMo-V2.6-Flash (310B total parameters) and MiMo-V2.6-Pro (1.02T total parameters). Controlled code-only Flash experiments show improved code agent performance, reduced trajectory-length growth, and more stable training. We further apply GAGAR in large-scale mixed-task RL with both Flash and Pro. Our results support combining test-based verification with groupwise agentic grading to improve the quality and stability of code agent RL.
Chinese Translation
代码代理的强化学习（RL）通常使用可执行测试来提供二元奖励。在这些奖励的基础上，组相对策略优化（GRPO）在每个回滚组内对通过测试的轨迹分配相同的优势，忽视了实现质量和任务要求遵循度的差异。这使得策略缺乏一个学习信号，无法优先考虑干净、针对性的实现，而不是那些包含不必要或超出范围的更改的实现。我们提出了GAGAR，这是一个针对代码代理RL的质量感知信用再分配框架。GAGAR基于动态采样，保留包含通过和未通过轨迹的组，将每个组的所有轨迹放置在一个共享工作区中，在这里，一个经过SFT训练的代理评分者共同检查这些轨迹并对通过测试的候选者进行排名。基于这一排名，我们对较低排名的轨迹进行降权，并按比例重新缩放所有通过测试轨迹的优势，以恢复它们的原始总和。这种保持总和的再分配保留了通过质量基础降权建立的相对权重，同时将信用转移到更高质量的实现上。我们在工业规模上评估了GAGAR，使用了MiMo-V2.6-Flash（总参数310B）和MiMo-V2.6-Pro（总参数1.02T）的预RL SFT检查点。控制的仅代码Flash实验显示了代码代理性能的改善、轨迹长度增长的减少和更稳定的训练。我们进一步在大规模混合任务RL中应用GAGAR，结合Flash和Pro。我们的结果支持将基于测试的验证与组别代理评分相结合，以提高代码代理RL的质量和稳定性。
cs.CL / 57 / 2609.32581
HERO-MoE: Historical Expert Routing with Scale-Preserving Fusion
HERO-MoE：具有规模保持融合的历史专家路由
Abstract
Mixture-of-Experts (MoE) architectures have become a standard way to scale model capacity while keeping computation sparse, yet routing remains a key determinant of MoE quality and training behavior. Prior empirical studies suggest that MoE routing reflects input semantics and upstream computation across depth, but standard routers do not explicitly use the routing distributions produced by preceding layers. We propose HERO-MoE, Historical Expert ROuting with Scale-Preserving Fusion, a routing framework that injects historical routing priors into MoE routers by reusing detached, dense routing distributions collected from preceding MoE layers. The key idea is simple: HERO-MoE preserves the original token-conditioned routing branch and adds a residual historical routing contribution before the standard softmax and top-$k$ dispatch. To stabilize this historical signal, HERO-MoE introduces a scale-preserving fusion mechanism that matches the magnitude of historical routing memory to the current hidden representation and accounts for the number of visible historical layers, without introducing an auxiliary routing loss or a fusion-specific tuning parameter. By reusing routing distributions already computed by preceding MoE layers, HERO-MoE improves training-loss reduction with modest end-to-end overhead. The resulting router remains compatible with standard sparse dispatch, including top-$k$ and group-limited routing, and can be inserted into existing MoE backbones with minimal architectural changes. Experiments on an approximately 8B-parameter MoE model with 0.5B active parameters, trained from scratch on 100B tokens, show that HERO-MoE reduces the final loss from 1.6393 to 1.6184, while peak memory and FLOPs increase by only 0.44\% and 0.64\%, respectively.
Chinese Translation
混合专家（MoE）架构已成为扩展模型容量的标准方法，同时保持计算的稀疏性，但路由仍然是决定MoE质量和训练行为的关键因素。先前的实证研究表明，MoE路由反映了输入语义和跨深度的上游计算，但标准路由器并未明确使用前面层产生的路由分布。我们提出了HERO-MoE，即具有规模保持融合的历史专家路由，这是一种通过重用从前面的MoE层收集的独立、密集的路由分布，将历史路由先验注入MoE路由器的路由框架。其关键思想很简单：HERO-MoE保留了原始的基于令牌的路由分支，并在标准softmax和top-$k$分派之前添加了一个残差历史路由贡献。为了稳定这一历史信号，HERO-MoE引入了一种规模保持融合机制，该机制将历史路由记忆的幅度与当前隐藏表示匹配，并考虑可见历史层的数量，而不引入辅助路由损失或特定于融合的调优参数。通过重用前面MoE层已经计算的路由分布，HERO-MoE在保持适度的端到端开销的同时，提高了训练损失的减少。最终的路由器仍然与标准稀疏分派兼容，包括top-$k$和组限制路由，并且可以在现有的MoE骨干网络中以最小的架构更改插入。对一个约8B参数的MoE模型进行的实验显示，该模型具有0.5B的活跃参数，从头开始在100B个令牌上训练，结果表明HERO-MoE将最终损失从1.6393降低到1.6184，而峰值内存和FLOPs仅分别增加0.44 ext{%}和0.64 ext{%}。
cs.CL / 58 / 2609.32610
KV-Lingo: Learning KV-Cache Translators with Distillation
KV-Lingo：通过蒸馏学习KV缓存翻译器
Abstract
Large language models represent context with a key-value (KV) cache. Caches are model-specific: for the same text, models with different architectures or weights produce incompatible representations. This makes it costly to switch models over a shared context: although the context has already been processed by one model, the incoming model must process it again to build its own cache. We introduce KV-Lingo, a method for translating the KV cache of a source model into one that can be read by a target model. KV-Lingo consists of a collection of linear maps, typically one per layer of the target model, that are applied independently on all tokens' key and value representations. We train these maps using distillation, minimising the divergence between the target model's predictions from its native cache and those from the translated cache. We consider several model pairs spanning multiple sizes and architectures, training one translator per pair on a generic text corpus. The resulting translators preserve strong downstream performance in both small-to-large and large-to-small transfers. Since a switch then costs a linear map and a single decoding step instead of a prefill, replacing re-prefill with cache translation reduces the time to first token after a model switch by 9.6x already on a 64-token prompt for Qwen models on an Apple M3 Ultra, and by up to 29x at 32k context length on an H100. These gains make KV-Lingo particularly useful for dynamic model routing: a context can be processed by one model and handed off to another only when needed, without re-prefilling the shared prefix. We finally show that KV-Lingo can be used for seamless model switching, staying close to re-prefill across repeated switches in our multi-turn evaluations.
Chinese Translation
大型语言模型通过键值（KV）缓存来表示上下文。缓存是模型特定的：对于相同的文本，不同架构或权重的模型会产生不兼容的表示。这使得在共享上下文中切换模型的成本很高：尽管上下文已经被一个模型处理，但即将到来的模型必须再次处理它以构建自己的缓存。我们提出了KV-Lingo，一种将源模型的KV缓存翻译为目标模型可读取格式的方法。KV-Lingo由一组线性映射组成，通常每个目标模型的层对应一个，这些映射独立应用于所有标记的键和值表示。我们使用蒸馏训练这些映射，最小化目标模型从其本地缓存和从翻译缓存的预测之间的差异。我们考虑了多个大小和架构的模型对，在一个通用文本语料库上为每对模型训练一个翻译器。结果翻译器在小到大和大到小的迁移中保持了强大的下游性能。由于切换只需一个线性映射和一个解码步骤，而不是预填充，因此用缓存翻译替代重新预填充将模型切换后的首次令牌时间缩短了9.6倍，针对Apple M3 Ultra的64个令牌提示，并在H100上达到32k上下文长度时缩短了29倍。这些收益使得KV-Lingo在动态模型路由中尤为有用：上下文可以由一个模型处理，并在需要时交给另一个模型，而无需重新预填充共享前缀。最后，我们展示了KV-Lingo可以用于无缝模型切换，在我们的多轮评估中，保持在重复切换中接近重新预填充的性能。
cs.CL / 59 / 2609.32617
The Alignment Paradox: How Post-Training Amplifies Confident Hallucinations in Language Models
对齐悖论：后训练如何放大语言模型中的自信幻觉
Abstract
Large language models (LLMs) can produce factually incorrect answers with high confidence, undermining their reliability and limiting the effectiveness of uncertainty-based error detection. While prior research attributes confident hallucinations to factors such as missing knowledge in training data, reasoning errors, or stochastic decoding, we uncover that post-training alignment itself is a primary driver of these errors, a phenomenon we call the \textbf{Alignment Paradox}. Across five model families evaluated on factual benchmarks, unaligned base models produce few high-confidence errors on long-tail factual queries, whereas instruction-tuned models multiply high-confidence errors ($p \ge 0.95$) by more than an order of magnitude (10$\times$ to 35$\times$). Layer-wise probing with the Logit Lens reveals that this overconfidence emerges in late layers, where wrong-answer margins expand past 4.0 points after remaining near zero across early and intermediate layers. These findings motivate limiting margin growth during post-training. We implement this principle through an entropy-dependent margin bound in direct preference optimization (DPO). In multi-epoch experiments with Mistral-7B, the bounded objective reduces high-confidence errors by up to 35.3\% relative to standard DPO while maintaining performance on evaluated general reasoning benchmarks. These results show that bounded margins mitigate confident hallucinations during post-training.
Chinese Translation
大型语言模型（LLMs）能够以高度自信的方式产生事实错误的答案，这削弱了它们的可靠性，并限制了基于不确定性的错误检测的有效性。尽管先前的研究将自信幻觉归因于训练数据中缺失的知识、推理错误或随机解码等因素，但我们发现后训练对齐本身是这些错误的主要驱动因素，这一现象我们称之为 extbf{对齐悖论}。在五个模型家族的事实基准评估中，未对齐的基础模型在长尾事实查询中产生的高自信错误很少，而经过指令调优的模型则将高自信错误（$p
ge 0.95$）的数量增加了一个数量级（10$ imes$到35$ imes$）。使用Logit Lens进行的层级探测显示，这种过度自信出现在后层，其中错误答案的边际在早期和中间层保持接近零后，扩展到超过4.0点。这些发现促使我们在后训练期间限制边际增长。我们通过在直接偏好优化（DPO）中实施一个依赖于熵的边际界限来实现这一原则。在与Mistral-7B的多轮实验中，受限目标相较于标准DPO将高自信错误减少了多达35.3\%，同时保持了在评估的一般推理基准上的表现。这些结果表明，受限的边际在后训练期间减轻了自信幻觉。
cs.CL / 60 / 2609.32622
Does CoT-Pass@k Really Check the CoT? A Multilingual Mathematical Audit
CoT-Pass@k 真正检查了 CoT 吗？一项多语言数学审计
Abstract
Pass@k measures whether a model reaches a correct answer under repeated sampling, but never how: a lucky guess counts the same as sound reasoning. CoT-Pass@k was proposed to close that gap, adding an LLM-as-judge that must assess a solution's reasoning chain before it counts. Its value rests entirely on one assumption: that the judge catches flawed reasoning. That assumption has never been tested inside the metric that depends on it, and never outside English, though the metric's claims concern models used in many languages. We report the first audit of that verification step, run under the metric's own protocol on a multilingual suite of five mathematical benchmarks in English, Turkish and Portuguese, two of them natively written. We corrupt correct solutions with deterministic edits that damage the chain and the final answer separately. We observe that all three judges accept corrupted chains almost as often as clean ones. V4-Flash and Qwen3.6 reject a solution sharply only when its final answer is wrong and accept a wrong answer more readily when the chain agrees with it; the metric's own judge accepts most wrong answers as well. Our study shows that chain-answer agreement dominates the two larger judges' verdicts and that all three fail to reliably detect the tested reasoning errors. Consequently the difference Pass@k - CoT-Pass@k averages 19.7 points on an earlier solver generation but only 4.1 on the current one. What little remains depends on the token budgets on both sides and on the generation mode; raising the generation budget moves Pass@64 by more than fifty points while the difference stays at zero. We close with two checks any judged reasoning metric should pass before its numbers are read as evidence about reasoning.
Chinese Translation
Pass@k 衡量模型在重复采样下是否得出正确答案，但从未关注其过程：一次幸运的猜测与合理的推理被视为相同。为了解决这一问题，提出了 CoT-Pass@k，增加了一个作为评判者的 LLM（大型语言模型），该模型必须在结果计入之前评估解决方案的推理链。其价值完全依赖于一个假设：评判者能够识别出有缺陷的推理。然而，这一假设从未在依赖于它的指标内部进行测试，也从未在英语以外的语言中进行测试，尽管该指标的主张涉及多种语言的模型。我们报告了首次针对这一验证步骤的审计，按照该指标的协议在英语、土耳其语和葡萄牙语的五个数学基准上进行，其中两个基准是本土撰写的。我们通过确定性编辑破坏正确解决方案的推理链和最终答案。我们观察到，所有三个评判者几乎同样频繁地接受被破坏的推理链和干净的推理链。V4-Flash 和 Qwen3.6 仅在最终答案错误时才会明显拒绝解决方案，并且当推理链与错误答案一致时更容易接受错误答案；该指标自身的评判者也接受大多数错误答案。我们的研究表明，推理链与答案的一致性主导了两个较大评判者的裁决，且所有三个评判者未能可靠地检测出所测试的推理错误。因此，Pass@k 与 CoT-Pass@k 之间的平均差异在早期解题者中为 19.7 分，但在当前解题者中仅为 4.1 分。剩下的微小差异依赖于双方的标记预算和生成模式；提高生成预算使 Pass@64 提升超过五十分，而差异保持为零。我们最后提出任何被评判的推理指标在其数字被视为推理证据之前应通过的两个检查。
cs.CL / 61 / 2609.32625
MixDetect: Word-Level Localization and Quantification of AI Editing
MixDetect：AI 编辑的词级定位与量化
Abstract
Large language models are increasingly used to edit human-written text rather than generate entire texts from scratch. Conventional AI-text detectors mainly distinguish human-written from fully AI-generated text, while recent methods for AI-edited text typically provide only a text-level label or editing-degree score. We introduce MixDetect, a word-level framework for localizing and quantifying AI editing. MixDetect separately predicts whether each word has been edited and, conditional on editing, how substantial the edit is, allowing editing scope and editing intensity to be estimated separately. During training, source--edited pairs are aligned to construct word-level supervision, while inference requires only the input text. Experiments show that MixDetect accurately localizes AI-edited words, reflects differences in editing intensity, and reveals different scope--intensity patterns across editing degrees and operations. The overall AI editing magnitude increases under additional AI editing, decreases when AI-generated text is edited by humans, and remains nearly unchanged under ordinary human-to-human editing. The aggregated text-level predictions also perform well on binary and ternary AI-text classification and remain effective under domain and generator shifts. These results show that AI editing can be analyzed beyond a single authorship label or editing-degree score by identifying both where AI editing occurs and how substantial the edits are.
Chinese Translation
大型语言模型越来越多地用于编辑人类撰写的文本，而不是从头生成完整文本。传统的 AI 文本检测器主要区分人类撰写的文本与完全由 AI 生成的文本，而最近针对 AI 编辑文本的方法通常仅提供文本级标签或编辑程度分数。我们提出了 MixDetect，这是一个用于定位和量化 AI 编辑的词级框架。MixDetect 分别预测每个单词是否被编辑，并在编辑的条件下，评估编辑的实质性，从而允许独立估计编辑范围和编辑强度。在训练过程中，源文本与编辑文本对齐以构建词级监督，而推理仅需输入文本。实验表明，MixDetect 能够准确定位 AI 编辑的单词，反映编辑强度的差异，并揭示不同编辑程度和操作下的范围-强度模式。总体而言，AI 编辑的幅度在额外的 AI 编辑下增加，在人类编辑 AI 生成文本时减少，而在普通人际编辑下几乎保持不变。聚合的文本级预测在二元和三元 AI 文本分类中表现良好，并在领域和生成器转移下仍然有效。这些结果表明，AI 编辑可以通过识别 AI 编辑发生的位置和编辑的实质性，超越单一的作者标签或编辑程度分数进行分析。
cs.CL / 62 / 2609.32630
ExpVoyager: Direct Experience Navigation for Dynamic Agent Skill Synthesis
ExpVoyager：动态代理技能合成的直接经验导航
Abstract
Learning from experience in LLM agents has become a key paradigm for developing self-evolving agents that continuously learn and expand their capabilities. Within this paradigm, synthesizing the agent skill has emerged as a promising solution for transforming accumulated experience into reusable procedural knowledge, serving as an important layer for the harness system that supplies agents at runtime. Despite its potential, existing approaches largely abstract past experience into fixed procedural knowledge before downstream demands are known, which risks discarding knowledge that later becomes critical while retaining instance-specific details irrelevant to future tasks. In this paper, we reframe agent skill synthesis as a dynamic navigation problem over past experience, where agents actively explore accumulated trajectories on demand for the current task with targeted and fine-grained access to experience knowledge. To this end, we propose ExpVoyager, a novel framework in which a skill curator navigates raw experience across different views and resolutions, continually identifying reusable procedural knowledge from what it observes while tracking remaining knowledge needs that guide where to navigate next. Extensive experiments demonstrate both the effectiveness and versatility of ExpVoyager, showing consistent improvements in downstream task performance, continual gains as the experience space scales, and practical compatibility with existing skills under efficient experience access.
Chinese Translation
在大型语言模型（LLM）代理中，从经验中学习已成为开发自我演化代理的关键范式，这些代理能够持续学习并扩展其能力。在这一范式中，合成代理技能已成为将积累的经验转化为可重用程序知识的有希望的解决方案，作为在运行时为代理提供支持的重要层。尽管具有潜力，现有的方法在下游需求尚不明确之前，往往将过去的经验抽象为固定的程序知识，这可能会导致丢弃后续变得关键的知识，同时保留与未来任务无关的特定实例细节。在本文中，我们将代理技能合成重新框架为一个动态导航问题，代理根据当前任务的需求主动探索积累的轨迹，针对性地、细致地访问经验知识。为此，我们提出了ExpVoyager，一个新颖的框架，其中技能策展人跨不同视角和分辨率导航原始经验，持续识别可重用的程序知识，同时跟踪剩余的知识需求，以指导下一步的导航。大量实验表明，ExpVoyager在下游任务性能上具有一致的提升，随着经验空间的扩展持续获得收益，并在高效经验访问下与现有技能具有良好的兼容性。
cs.CL / 63 / 2609.32653
From Knowing to Abstaining: Bridging the Representation-Action Gap in Vision-Language Models
从知晓到弃权：弥合视觉-语言模型中的表征与行动差距
Abstract
The ability of vision-language models (VLMs) to abstain from unanswerable questions is as important as their ability to answer answerable ones accurately. Recently, several benchmarks have emerged to evaluate and improve VLM abstention, but they have substantial limitations. First, samples often contain shortcut cues in images or questions that reveal answerability, while an explicit "unanswerable" option further prevents accurate assessment of spontaneous abstention. Second, as training data, they generally provide only binary labels without fine-grained explanations for deeper supervision. To address these limitations, we introduce Visual Answerability Diagnosis with Rationales (VAD-R), a benchmark constructed through a two-stage pipeline of shortcut filtering and quality verification to prevent answerability leakage. Each example is annotated with step-by-step rationales and causal evidence-gap labels. Evaluation of state-of-the-art open- and closed-source VLMs on VAD-R reveals limited spontaneous abstention, with average recall rates of only 11.4% and 16.3%, respectively. Probing analyses show that hidden-state representations in certain layers can effectively distinguish answerability, yet this distinction fails to manifest in final responses. Motivated by this observation, we introduce Rep2Act, a representation-to-action alignment method that translates latent answerability awareness into explicit abstention decisions. Rep2Act improves action accuracy on VAD-R from 56.67% to 86.33% for Qwen2.5-VL-3B and from 59.33% to 88.67% for Qwen2.5-VL-7B. On the out-of-distribution TUBench, Rep2Act achieves an average F1 score of 53.3% with only a 3B model, surpassing the closed-source GPT-4 Turbo and GPT-4o by 16.2% and 1.1%, respectively.
Chinese Translation
视觉-语言模型（VLMs）在面对无法回答的问题时能够选择弃权的能力，与其准确回答可回答问题的能力同样重要。最近出现了几个基准测试，用于评估和改进VLM的弃权能力，但它们存在显著的局限性。首先，样本中常常包含图像或问题中的捷径线索，揭示了问题的可回答性，而明确的“无法回答”选项进一步妨碍了对自发弃权的准确评估。其次，作为训练数据，它们通常仅提供二元标签，而没有细致的解释以进行更深层次的监督。为了解决这些局限性，我们引入了带有理由的视觉可回答性诊断（Visual Answerability Diagnosis with Rationales, VAD-R），这是通过捷径过滤和质量验证的两阶段流程构建的基准，以防止可回答性泄漏。每个示例都附有逐步的理由和因果证据缺口标签。对最先进的开源和闭源VLM在VAD-R上的评估显示，自发弃权的能力有限，平均召回率仅为11.4%和16.3%。探测分析表明，某些层中的隐藏状态表征能够有效区分可回答性，但这种区分未能在最终响应中体现。基于这一观察，我们引入了Rep2Act，这是一种表征到行动的对齐方法，将潜在的可回答性意识转化为明确的弃权决策。Rep2Act在VAD-R上的行动准确率从56.67%提高到86.33%（针对Qwen2.5-VL-3B），从59.33%提高到88.67%（针对Qwen2.5-VL-7B）。在分布外的TUBench上，Rep2Act在仅使用3B模型的情况下实现了53.3%的平均F1分数，分别超越了闭源的GPT-4 Turbo和GPT-4o 16.2%和1.1%。
cs.CL / 64 / 2609.32684
Focusing Condition: Inference-Time Self-Contrastive Steering Elicits Better Conditional Text Embeddings in LLMs
聚焦条件：推理时自对比引导产生更好的条件文本嵌入于大型语言模型中
Abstract
Extracting conditional text embeddings from large language models (LLMs) is a promising paradigm, as it requires neither additional data nor fine-tuning. Existing methods incorporate conditions into prompts to guide LLMs to focus on specific aspects and elicit conditional text embeddings. However, relying solely on prompts often fails to produce high-quality conditional text embeddings, as they remain entangled with general text embeddings, ultimately degrading their quality. To this end, we propose an inference-time, plug-and-play Self-Contrastive Steering (SCS) method that constructs unconditional general text embeddings and uses them to refine conditional text embeddings, making them more focused on the target condition. Specifically, we modify the attention mask and positional encodings to mask the condition, thereby obtaining unconditional text embeddings and intervening in the multi-head self-attention computation process. Notably, our method is highly efficient, requiring only a single additional multi-head self-attention computation at inference time. Extensive experiments on clustering, Semantic Textual Similarity, and triplet alignment datasets demonstrate that our method can seamlessly improve the performance of existing prompt-based methods across different LLMs in a training-free and plug-and-play manner. Our code will be released at https://github.com/zifengcheng/SCS
Chinese Translation
从大型语言模型（LLMs）中提取条件文本嵌入是一种有前景的范式，因为它既不需要额外的数据，也不需要微调。现有的方法通过将条件融入提示中，指导LLMs关注特定方面并引出条件文本嵌入。然而，仅依赖提示往往无法产生高质量的条件文本嵌入，因为它们与一般文本嵌入纠缠在一起，最终降低了其质量。为此，我们提出了一种推理时即插即用的自对比引导（Self-Contrastive Steering, SCS）方法，该方法构建无条件的一般文本嵌入，并利用这些嵌入来优化条件文本嵌入，使其更专注于目标条件。具体而言，我们修改了注意力掩码和位置编码，以屏蔽条件，从而获得无条件文本嵌入，并干预多头自注意力计算过程。值得注意的是，我们的方法效率极高，在推理时仅需一次额外的多头自注意力计算。针对聚类、语义文本相似性和三元组对齐数据集的广泛实验表明，我们的方法能够无缝提升现有基于提示的方法在不同LLMs上的性能，且无需训练，具有即插即用的特性。我们的代码将发布在 https://github.com/zifengcheng/SCS
cs.CL / 65 / 2609.32717
LLM Alignment--Utility Asymmetry under Semantic-Preserving Transformations
大型语言模型的对齐——在语义保持变换下的效用不对称性
Abstract
Large Language Model (LLM) alignment is intended to ensure that models remain helpful and safe, but its stability under input distributional shift is not yet fully understood. Prior work shows that aligned models can fail under jailbreak prompts, alternative encodings, and cross-lingual transfer, yet these failures are usually studied as attacks rather than controlled probes of alignment generalization. Moreover, existing evidence is largely grounded in natural language variation already represented during pretraining, leaving unresolved whether alignment generalizes with semantic content or remains tied to superficial surface patterns. In this paper, we study this question using synthetic semantic-preserving transformations that are rule-based and invertible, preserving task-relevant meaning while shifting inputs beyond standard linguistic variation. Across four open-weight and four commercial models, under both fine-tuning and in-context learning, we use these transformations as a probe of alignment generalization and identify an empirical pattern we term Alignment--utility asymmetry: once models can operate effectively on transformed inputs, task utility is often substantially retained while alignment failure increases more sharply. For example, adapted GPT-4.1 mini shows only limited utility degradation under transformation while its harmful rate rises from 13.3 to 74.3; Gemini 3 Flash similarly retains near-original utility while its harmful rate increases from 2.3 to 43.0. Taken together, these results suggest that semantic-preserving distribution shifts can expose a recurring gap in how utility and alignment generalize in current LLMs.
Chinese Translation
大型语言模型（LLM）的对齐旨在确保模型保持有用和安全，但其在输入分布变化下的稳定性尚未完全理解。先前的研究表明，对齐模型在越狱提示、替代编码和跨语言迁移下可能会失败，但这些失败通常被视为攻击，而非对对齐泛化的控制探测。此外，现有证据主要基于在预训练期间已经表示的自然语言变异，尚未解决对齐是否与语义内容泛化，或仍然与表面模式相联系。在本文中，我们使用基于规则且可逆的合成语义保持变换来研究这一问题，这些变换在保持任务相关意义的同时，将输入超出标准语言变异。在四个开放权重模型和四个商业模型中，无论是在微调还是上下文学习下，我们使用这些变换作为对齐泛化的探测，并识别出一个我们称之为对齐-效用不对称性的经验模式：一旦模型能够有效处理变换后的输入，任务效用通常会显著保留，而对齐失败则更为明显地增加。例如，适应后的 GPT-4.1 mini 在变换下仅显示出有限的效用下降，而其有害率从 13.3 上升至 74.3；Gemini 3 Flash 同样在接近原始效用的情况下，其有害率从 2.3 增加至 43.0。综合来看，这些结果表明，语义保持的分布变化可以揭示当前大型语言模型中效用与对齐泛化之间的反复差距。
cs.CL / 66 / 2609.32758
How Far Do Persona Effects Generalize in Language Models?
人格效应在语言模型中的普遍性有多大？
Abstract
Persona prompts ask language models to answer as particular kinds of people. We test whether relationships learned from these effects predict responses to new questions and remain useful across models and prompts. Across 57 attributes, three behavioral domains, and seven pairs of open 7 to 9B checkpoints, persona effects can be predictable without being portable. Separate attribute and task gains improve prediction beyond shared scaling significantly in OLMo-3 and Qwen2.5, with the most robust evidence in OLMo-3. In that model, target refitting significantly outperforms gains borrowed from each of the other six pairs. Across model transfers, borrowed gains with one amplitude underperform shared scaling in most directions; allowing two target parameters removes the significant losses but yields no significant benefit over target shared scaling. After rewording, refitting significantly outperforms reuse with one amplitude in all six tested pairs, while changes of examples or country context often preserve reuse value. In the tested prompt transfers, regularized updates outperform both reuse strategies in median at 64 target questions per attribute. A separate survey comparison finds that selecting the more responsive checkpoint can worsen human fit; responsiveness is confounded with training status, and temperature calibration largely removes this cost but not errors in group ordering. Within the tested gain representation, apparent transfer can come from target calibration; source relationships must add predictive value beyond calibration and regularization. Code and data are available at https://github.com/thzva/persona-gain
Chinese Translation
人格提示要求语言模型以特定类型的人物身份进行回答。我们测试这些效应所学习的关系是否能够预测对新问题的响应，并在不同模型和提示之间保持有效。在57个属性、三个行为领域和七对开放的7到9B检查点中，人格效应可以被预测，但并不具备可迁移性。单独的属性和任务增益在OLMo-3和Qwen2.5中显著提高了预测能力，超出了共享缩放的效果，其中在OLMo-3中证据最为充分。在该模型中，目标重拟合显著优于从其他六对中借用的增益。在模型迁移中，借用的增益在大多数方向上表现不佳，单一幅度的共享缩放；允许两个目标参数消除了显著损失，但并未带来相对于目标共享缩放的显著收益。经过重新措辞后，重拟合在所有六对测试中显著优于单一幅度的重用，而示例或国家背景的变化通常保留了重用的价值。在测试的提示迁移中，正则化更新在每个属性的64个目标问题中在中位数上优于两种重用策略。单独的调查比较发现，选择响应更快的检查点可能会恶化人类拟合；响应性与训练状态混淆，而温度校准在很大程度上消除了这种成本，但并未消除组排序中的错误。在测试的增益表示中，明显的迁移可能来自目标校准；源关系必须在校准和正则化之外增加预测价值。代码和数据可在 https://github.com/thzva/persona-gain 获取。
cs.CL / 67 / 2609.32770
C-HAT-Bench: Benchmarking Chinese AI-Text Detection Beyond Fully Generated Text
C-HAT-Bench：超越完全生成文本的中文AI文本检测基准
Abstract
Large Language Models (LLMs) increasingly participate in writing by modifying or extending human drafts, causing machine involvement to vary in both form and extent. Yet most Machine-Generated Text (MGT) detectors are evaluated only on fully human-written versus fully AI-generated text. Because human--AI collaboration can weaken or redistribute cues associated with machine generation, strong performance under this binary setting may overstate detector reliability. This mismatch remains underexplored in Chinese: detection cues are shaped by tokenization and language-specific text distributions, yet controlled resources spanning production settings, domains, and generators remain limited. To fill this gap, we present a Chinese Human-AI Collaborative Text Detection Benchmark (C-HAT-Bench), a unified benchmark that links $5,000$ human-written source texts from five domains to more than $240,000$ variants produced using six generative models under Prefix-Conditioned Continuation as a reference setting and three collaborative production modes. We evaluate $21$ detectors through four protocols spanning zero-shot and pretrained supervised document-level detection, boundary localization, and cross-condition generalization. Relative to Prefix-Conditioned Continuation, mean AUROC across document-level detectors is $12.0\%$ lower on the collaborative production modes, with the largest detector-specific relative decrease reaching $44.4\%$. Transfer across collaborative production modes is also asymmetric, indicating that performance in a given production setting is not a reliable predictor of performance in other production settings.
Chinese Translation
大型语言模型（LLMs）越来越多地参与写作，通过修改或扩展人类草稿，使机器参与的形式和程度各不相同。然而，大多数机器生成文本（MGT）检测器仅在完全人类撰写与完全AI生成的文本上进行评估。由于人类与AI的协作可能削弱或重新分配与机器生成相关的线索，因此在这种二元设置下的强性能可能会夸大检测器的可靠性。这一不匹配在中文中仍未得到充分探讨：检测线索受到分词和特定语言文本分布的影响，但涵盖生产设置、领域和生成器的受控资源仍然有限。为填补这一空白，我们提出了中文人类-人工智能协作文本检测基准（C-HAT-Bench），这是一个统一的基准，将来自五个领域的$5,000$篇人类撰写的源文本与使用六种生成模型在前缀条件续写（Prefix-Conditioned Continuation）作为参考设置下生成的超过$240,000$个变体相连接，并涵盖三种协作生产模式。我们通过四种协议评估$21$个检测器，涵盖零样本和预训练的监督文档级检测、边界定位和跨条件泛化。相较于前缀条件续写，在协作生产模式下，文档级检测器的平均AUROC降低了$12.0
m{ ext{ extperthousand}}$，其中最大的检测器特定相对下降达到$44.4
m{ ext{ extperthousand}}$。不同协作生产模式之间的迁移也是不对称的，这表明在特定生产设置下的性能并不是其他生产设置下性能的可靠预测指标。
cs.CL / 68 / 2609.32776
On the Behavioral Traits of LLM Agents
关于大型语言模型代理的行为特征
Abstract
Users increasingly describe different AI agents as distinct colleagues to work with. AI personality research aims to quantify such impressions by attributing human-like "traits" to agents. However, existing measures fall short: models' self-reports (S-data) diverge from their actual behavior, while informant ratings from LLM judges (I-data) are costly to scale and cover few everyday scenarios. In this paper, we propose A-B-D to infer traits bottom-up from behavioral data (B-data), namely how agents act on their environment and communicate with users, as recorded in existing trajectories. From 345,667 real-world trajectories spanning 80 models, 12 tasks, and 50 harnesses, we extract 318 candidate features that capture both the actions an agent takes at each step (functional) and the language accompanying them (linguistic). We retain only features that show instance-level stability, cross-task consistency, and model discriminability. Factor analysis of the remaining 79 features uncovers six stable, model-attributable factors, two functional and four linguistic. For example, Kimi-K3 exhibits the most planfulness, whereas GPT-5.5 and GPT-5.6 are the least energetic. Moreover, we quantify the "knowledge-action gap" in the wild: these factors correlate only weakly with self-reported Big Five scores, even for conceptually matched pairs such as extroversion and energetic (r = 0.07, p = 0.58). Our work offers a new lens for understanding AI personality, with implications for users, developers, and researchers from both computer science and social science.
Chinese Translation
用户越来越多地将不同的人工智能代理描述为可以合作的独特同事。人工智能个性研究旨在通过将人类般的“特征”归因于代理来量化这种印象。然而，现有的测量方法存在不足：模型的自我报告（S-data）与其实际行为存在差异，而来自大型语言模型评审者的知情者评分（I-data）在扩展性上成本高昂且覆盖的日常场景较少。在本文中，我们提出了A-B-D方法，从行为数据（B-data）中自下而上地推断特征，即代理如何在其环境中行动并与用户沟通，这些行为记录在现有的轨迹中。我们从覆盖80个模型、12个任务和50个工具的345,667条真实轨迹中提取了318个候选特征，这些特征既捕捉了代理在每一步所采取的行动（功能性），也捕捉了伴随其行动的语言（语言性）。我们仅保留那些表现出实例级稳定性、跨任务一致性和模型可区分性的特征。对剩余79个特征的因子分析揭示了六个稳定的、可归因于模型的因子，其中两个是功能性的，四个是语言性的。例如，Kimi-K3表现出最高的计划性，而GPT-5.5和GPT-5.6则表现出最低的活力。此外，我们量化了“知识-行动差距”：这些因子与自我报告的五大人格特质评分之间的相关性较弱，即使对于概念上匹配的对，例如外向性与活力（r = 0.07, p = 0.58）。我们的研究为理解人工智能个性提供了新的视角，对计算机科学和社会科学领域的用户、开发者和研究者均具有重要意义。
cs.CL / 69 / 2609.32810
OpenTumorBoard: A Real-World Benchmark of Multidisciplinary Tumor Board Discussion Trajectories
OpenTumorBoard：多学科肿瘤委员会讨论轨迹的真实世界基准
Abstract
Multidisciplinary tumor boards integrate multimodal clinical observations and longitudinal patient histories through specialist discussions, yet benchmarks rarely capture these real-world trajectories. We introduce OpenTumorBoard, a benchmark with 611 patient cases and 19,157 discussion turns across ten specialist roles, transcribed from 12,534 minutes of publicly available tumor board recordings on YouTube. The benchmark evaluates two settings: SPECIALIST TURN, in which an LLM responds to a clinically significant question posed during a real discussion, and BOARD SIMULATION, in which it generates an entire back-and-forth discussion and reaches a consensus on therapy recommendations, surgical plans, next actions and clinical trial matching. Evaluation of 14 general-purpose frontier and medical LLMs reveals substantial limitations: the best models score 3.43 out of 5 in clinical equivalence to specialist answers and 2.78 out of 5 in alignment with recorded board conclusions. Supervised finetuning and reinforcement learning improve performance on a held-out test set, suggesting that real-world discussion trajectories can support model adaptation. Three M.D. experts review a subset of the benchmark, finding high information coverage and factuality of patient cases and strong fidelity of extracted consensus conclusions. We will release OpenTumorBoard and its automated curation pipeline to support the development and evaluation of LLMs for multidisciplinary, personalized cancer decision-making.
Chinese Translation
多学科肿瘤委员会通过专家讨论整合多模态临床观察和纵向患者历史，但基准测试很少捕捉这些真实世界的轨迹。我们介绍了OpenTumorBoard，这是一个包含611个患者案例和19,157次讨论轮次的基准，涵盖十个专家角色，转录自12,534分钟在YouTube上公开可用的肿瘤委员会录音。该基准评估两个设置：SPECIALIST TURN，在此设置中，LLM（大语言模型）回答在真实讨论中提出的临床重要问题；BOARD SIMULATION，在此设置中，LLM生成整个来回讨论并就治疗建议、手术计划、后续行动和临床试验匹配达成共识。对14个通用前沿和医学LLM的评估显示出显著的局限性：最佳模型在与专家答案的临床等效性上得分为3.43（满分5分），在与记录的委员会结论的一致性上得分为2.78（满分5分）。监督微调和强化学习在保留测试集上的表现有所提升，表明真实世界的讨论轨迹可以支持模型的适应。三位医学博士专家审查了基准的一个子集，发现患者案例的信息覆盖率和事实准确性较高，提取的共识结论的保真度强。我们将发布OpenTumorBoard及其自动化策划管道，以支持多学科个性化癌症决策的LLM开发和评估。
cs.CL / 70 / 2609.32852
ARSM: Auto-Regressive State Machine for Agentic Reasoning Compression
ARSM：用于自主推理压缩的自回归状态机
Abstract
While Large Language Model (LLM)-based agents demonstrate strong capabilities in long-horizon tasks by interleaving reasoning with external environment interactions, the continuous accumulation of context rapidly creates a critical memory bottleneck. Existing memory compression methods rely on task-specific optimization or external auxiliary models, introducing significant computational overhead. Furthermore, the resulting compressed representations tend to lose structured relationships, leading to information dilution, attention collapse, and degraded decision consistency. To address these limitations, we propose Auto-Regressive State Machine (ARSM), a lightweight training-free framework that enables in-situ reasoning compression through structured state evolution. ARSM introduces two key components: (i) a trajectory abstraction mechanism that reorganizes interaction histories into compact Hypothesis-Action-Result (HAR) micro-chains; (ii) a dynamic state machine that regulates hierarchical memory through atomic operations and a compression-control parameter. These components are unified within an auto-regressive, self-compressive generation space, where each model output jointly performs external action execution and internal state updates. We evaluate ARSM on Webshop, Multi-Objective Multi-Hop QA, and SWE-Bench Lite datasets. Experimental results show that ARSM maintains the task performance while simultaneously reducing token consumption, offering a practical, cost-effective route toward scalable autonomous agents for long-horizon tasks.
Chinese Translation
虽然基于大型语言模型（LLM）的智能体在长时间任务中通过将推理与外部环境交互交错展示出强大的能力，但上下文的持续积累迅速造成了关键的内存瓶颈。现有的内存压缩方法依赖于特定任务的优化或外部辅助模型，带来了显著的计算开销。此外，所得到的压缩表示往往会丧失结构关系，导致信息稀释、注意力崩溃和决策一致性下降。为了解决这些局限性，我们提出了自回归状态机（ARSM），这是一个轻量级的无训练框架，能够通过结构化状态演变实现原位推理压缩。ARSM引入了两个关键组件：（i）一种轨迹抽象机制，将交互历史重组为紧凑的假设-行动-结果（HAR）微链；（ii）一种动态状态机，通过原子操作和压缩控制参数调节层次内存。这些组件统一在一个自回归、自压缩的生成空间中，其中每个模型输出共同执行外部动作执行和内部状态更新。我们在Webshop、多个目标多跳问答（Multi-Objective Multi-Hop QA）和SWE-Bench Lite数据集上评估了ARSM。实验结果表明，ARSM在保持任务性能的同时，减少了标记消耗，为长时间任务的可扩展自主智能体提供了一条实用且具有成本效益的途径。
cs.CL / 71 / 2609.32867
Are You Sure You're Sure? Two Confounds in a Sycophancy Benchmark
你确定你确定吗？一个阿谀奉承基准中的两个混淆因素
Abstract
Sycophancy is a language model's tendency to cave when a user pushes back, abandoning a correct answer for the user's. Several benchmarks now measure it by scripting an objection and recording how often the model caves. Because that objection is a prompt template, whatever else the template varies is measured along with the property it claims to isolate. We audit SycEval, which reports that objections raised before a model answers (preemptive) cause more caving than those raised after (in-context), and attributes the gap to timing. Two features of its templates vary alongside the property each is meant to test. First, SycEval's objections escalate through four strength levels, and at the two weakest only the preemptive template names a target answer, so timing and naming vary together. We build the missing comparison and test it on multiple-choice questions and SycEval's own free-form pipeline. Naming a target answer raises the follow rate, the share of samples matching the user's assertion, by 14.1--49.5 percentage points (pp), and once both templates name one, the timing comparison reverses on three of the five model conditions we test: models cave \emph{less} under preemptive objections than in-context ones, opposite to SycEval. Second, the output-format instruction benchmarks append for automatic grading also varies with objection placement. Moving it from the pushback into the question flips its effect in opposite directions across models on the same items ($p=0.0059$). The same effect appears in SycEval's free-form pipeline: relocating its instruction lowers caving by 5.1pp on Llama-3.1-8B, while an equivalence test confirms no effect on Qwen3-4B. Both confounds live in the template rather than the models under test, so both are correctable: we close with three checks benchmark authors can apply before publishing.
Chinese Translation
阿谀奉承是语言模型在用户反对时屈服的倾向，即放弃正确答案而选择用户的答案。目前有几个基准通过编写反对意见并记录模型屈服的频率来衡量这一点。由于该反对意见是一个提示模板，因此模板的其他变化与其声称要隔离的属性一起被测量。我们审计了 SycEval，该基准报告称，在模型回答之前提出的反对意见（预先反对）导致更多的屈服，而在回答之后提出的反对意见（上下文反对）则较少，并将差距归因于时机。其模板的两个特征与每个模板旨在测试的属性一起变化。首先，SycEval 的反对意见通过四个强度级别逐步升级，而在两个最弱的级别中，只有预先反对的模板命名了目标答案，因此时机和命名是一起变化的。我们建立了缺失的比较，并在多项选择题和 SycEval 自身的自由形式流程中进行了测试。命名目标答案使得跟随率，即与用户主张匹配的样本比例，提高了 14.1 到 49.5 个百分点（pp），一旦两个模板都命名了目标答案，时机比较在我们测试的五个模型条件中的三个上发生了逆转：模型在预先反对下的屈服 extit{更少}，与 SycEval 的结论相反。其次，输出格式指令基准的附加内容也随着反对意见的位置而变化。从推回移动到问题中会在同一项目上对模型产生相反的效果（$p=0.0059$）。在 SycEval 的自由形式流程中也出现了相同的效果：重新定位其指令使 Llama-3.1-8B 的屈服降低了 5.1pp，而等效性测试确认对 Qwen3-4B 没有影响。这两个混淆因素存在于模板中，而不是测试中的模型，因此都是可以纠正的：我们最后提出了基准作者在发布前可以应用的三项检查。
cs.CL / 72 / 2609.32902
Linger and Lose: Knowledge Collapse in Low-Bit Language Models
徘徊与失落：低比特语言模型中的知识崩溃
Abstract
Training language models with ternary weights is commonly judged by loss and downstream accuracy, which record only a modest cost relative to full precision. We show that these metrics can conceal a much larger failure. We instead measure knowledge capacity, the factual bits stored per parameter, on synthetic biographies with known information content. We train GPT-2-style models from scratch with 2.5M to 50M parameters at five precisions. Under the standard cosine schedule, ternary models retain as little as 6% of an identically trained fp16 model's capacity. The deficit widens with model size while perplexity rises by only 1.4 to 1.6 times. Measured throughout training, these models first acquire capacity and then lose most of it. We identify this knowledge collapse as a learning-rate dwell instability. Held near $1$--$2\times10^{-4}$ with no decay, a pre-collapse model collapses within a few hundred exposures, and returning to a safe rate does not restore capacity. We then locate the collapse in the output head. It happens at a value the model can never predict. The weights there grow unchecked, while every other prediction the model makes is unchanged. We find that making the value predictable removes the collapse, regardless of which attribute carries it. A warmup-stable-decay schedule with a 10% cooldown increases ternary capacity by 2.7 times at 25M and 4.3 times at 50M. Gains are larger at lower precision. Cutting the output head's learning rate prevents failure when training from scratch. The post-training quantization methods we tested recover no measurable capacity below 4 bits. Our findings suggest judging low-precision training by retained capacity during training rather than final loss.
Chinese Translation
使用三元权重训练语言模型通常通过损失和下游准确率来评估，这些指标相对于全精度模型仅记录了适度的成本。我们展示了这些指标可能掩盖了更大的失败。我们改为测量知识容量，即每个参数存储的事实位数，基于具有已知信息内容的合成传记。我们从头开始训练GPT-2风格的模型，参数量从250万到5000万，使用五种精度。在标准余弦调度下，三元模型的容量仅保留了与同样训练的fp16模型的6%。随着模型规模的扩大，这一差距加大，而困惑度仅上升了1.4到1.6倍。在整个训练过程中，这些模型首先获得容量，然后失去大部分容量。我们将这种知识崩溃确定为学习率停滞的不稳定性。当学习率保持在$1$--$2 imes10^{-4}$且没有衰减时，预崩溃模型在几百次暴露内崩溃，返回安全的学习率并不能恢复容量。我们随后定位到输出头的崩溃。它发生在模型永远无法预测的值上。该处的权重不受控制地增长，而模型做出的其他所有预测保持不变。我们发现使该值可预测可以消除崩溃，无论是哪个属性承载它。使用10%的冷却时间的预热-稳定-衰减调度在25M时将三元容量提高了2.7倍，在50M时提高了4.3倍。较低精度下的增益更大。降低输出头的学习率可以防止从头训练时的失败。我们测试的后训练量化方法在4比特以下无法恢复可测量的容量。我们的发现表明，应通过训练期间保留的容量而非最终损失来评估低精度训练。
cs.CL / 73 / 2609.32942
The Key Handoff: Retrieval in Hybrid Language Models
关键交接：混合语言模型中的检索
Abstract
A two-hop question makes a language model retrieve twice: once to produce a bridge entity, and once to retrieve with it. Transformers resolve that entity in their early layers. Hybrid models replace most of the attention with a recurrent state, so where the key becomes usable, and where it is spent on an answer, is not known. Answering "Where is the ball?" from "the ball belongs to Alice" and "Alice is in the garden" turns on Alice, a name the question does not mention. A model could reach garden through Alice, the key it computed, or through where the fact sits in the prompt. Across twelve models, dense and hybrid, we move a hidden state from one story into another where the two routes lead to different places, and read off which place the model gives. An attention layer converts the key in every model we tested, and crossing that layer removes its usable effect, all of it where no attention follows. In sequential hybrids this makes the answer a handoff: recurrent layers carry the key forward, and attention spends it. Retrieval does not always stop there. Writing a different fact into the memory of the recurrent layers that follow the last attention layer can move the answer toward that fact, multiplying its odds by 1.3 to 2.7, and which hybrids do this is not settled by their architecture or training. That read is addressed by the key, not by position: recurrent state in a hybrid is not only a carrier, but a memory that later layers can query.
Chinese Translation
两步问题使得语言模型进行两次检索：第一次产生一个桥接实体，第二次使用该实体进行检索。变换器在其早期层中解析该实体。混合模型用递归状态替代大部分注意力，因此关键何时变得可用，何时被用于答案，是未知的。从"球属于爱丽丝"和"爱丽丝在花园里"回答"球在哪里？"时，答案转向了爱丽丝，这个问题中并未提及的名字。模型可以通过爱丽丝（它计算出的关键）或通过事实在提示中的位置到达花园。在十二个模型中，包括稠密模型和混合模型，我们将一个隐藏状态从一个故事转移到另一个故事，其中两条路径通向不同的地方，并读取模型给出的哪个地方。我们测试的每个模型中的注意力层都会转换关键，而跨越该层会移除其可用效果，所有这些发生在没有后续注意力的地方。在顺序混合模型中，这使得答案成为一种交接：递归层将关键向前传递，而注意力则消耗它。检索并不总是在这里停止。在最后一个注意力层之后的递归层中写入不同的事实可以使答案朝向该事实移动，其概率增加1.3到2.7，而哪些混合模型能够做到这一点并不由其架构或训练决定。该读取由关键处理，而非位置：混合模型中的递归状态不仅是一个载体，还是后续层可以查询的记忆。
cs.CL / 74 / 2609.33014
TCMQA: A 38K-Question Traditional Chinese Medicine Benchmark with a Licensed-Practitioner Reference
TCMQA：一个包含38,000个问题的传统中医学基准，附有执业医师参考
Abstract
Medical benchmarks for language models are built almost entirely on Western biomedicine. Traditional Chinese Medicine (TCM) is a separate system, with its own diagnostic framework and its own literature, and it remains largely unmeasured. The few TCM evaluations that exist are small, narrow, and rarely paired with a human reference. We present TCMQA, an open benchmark of 38,279 questions from Chinese TCM licensing examinations, paired with 15,151 responses from 101 licensed practitioners. We evaluate 29 instruction-tuned models from 9 families, spanning 0.27B to 14.8B parameters. Accuracy ranges over 59 points, and no model approaches saturation. Pretraining data predicts TCM ability far better than scale: a 12B Western-pretrained model reaches 39.6%, while a Chinese-pretrained model an eighth its size reaches 60.8%. Nine models exceed the practitioner majority vote of 64.9%, the best by 21.8 points, and all nine come from that same Chinese-pretrained family. Yet difficulty does not transfer between models and practitioners: accuracy is flat across practitioner-rated difficulty, item-level agreement is near zero for all 29 models, and on $8.4\%$ of items the practitioners are correct where the leading model is wrong. We release the corpus, the practitioner responses, the harness, and per-item model outputs at https://huggingface.co/datasets/TechTCM/TCMQA.
Chinese Translation
语言模型的医学基准几乎完全基于西方生物医学。传统中医学（Traditional Chinese Medicine, TCM）是一个独立的体系，拥有其自身的诊断框架和文献，且在很大程度上未被量化。现有的少数TCM评估规模小、范围窄，且很少与人类参考相结合。我们提出了TCMQA，这是一个开放的基准，包含来自中国TCM执业考试的38,279个问题，并配有来自101名执业医师的15,151个回答。我们评估了来自9个模型家族的29个经过指令调优的模型，参数范围从0.27B到14.8B。准确率的差异超过59个点，且没有模型接近饱和。预训练数据对TCM能力的预测远优于模型规模：一个12B的西方预训练模型的准确率为39.6%，而一个规模仅为其八分之一的中文预训练模型的准确率达到了60.8%。有九个模型的准确率超过了执业医师的多数投票64.9%，其中表现最佳的模型超出该投票21.8个点，且这九个模型均来自同一中文预训练家族。然而，模型与执业医师之间的难度并不具可转移性：在执业医师评定的难度上，准确率保持平坦，29个模型在项目级别上的一致性接近零，并且在8.4%的项目中，执业医师的正确率高于领先模型的错误率。我们在https://huggingface.co/datasets/TechTCM/TCMQA发布了语料库、执业医师的回答、工具以及每个项目的模型输出。
cs.CL / 75 / 2609.33038
Improving the Diversity of LLM Outputs without a Trade-off
在不妥协的情况下提高大语言模型输出的多样性
Abstract
We propose DAST (Diversifying Arithmetic Sampling with TokenTour), a method that increases the diversity of LLM outputs without any change to the marginal distribution and with negligible generation-time overhead (a few microseconds). We observe that token IDs are often arranged in a meaningless order and reassign them so that tokens with similar meanings appear consecutively. This can be done in advance in a few hundred seconds per model, and the resulting order can be reused for all subsequent generations. By combining this order with arithmetic sampling (or quasi-Monte Carlo methods), we make similar tokens less likely to be generated across runs while preserving the distribution. Our method not only produces qualitatively good ideas but also significantly improves performance on the downstream task of ProtoQA.
Chinese Translation
我们提出了一种方法 DAST（多样化算术采样与 TokenTour），该方法在不改变边际分布的情况下，提高大语言模型（LLM）输出的多样性，并且生成时间开销微乎其微（仅几微秒）。我们观察到，令牌 ID 通常以无意义的顺序排列，因此我们重新分配它们，使得具有相似含义的令牌连续出现。这一过程可以在每个模型上提前完成，耗时几百秒，生成的顺序可以在后续的所有生成中重复使用。通过将这种顺序与算术采样（或准蒙特卡洛方法）结合，我们使得相似的令牌在多次生成中不太可能被生成，同时保持分布不变。我们的方法不仅产生了质量良好的创意，还显著提高了下游任务 ProtoQA 的性能。
cs.CL / 76 / 2609.33044
Pinned and Still Unstable: Within-Judge Verdict Variance and the Noise Floor of LLM-as-Judge Leaderboards
固定但仍不稳定：法官内判决方差与LLM作为法官排行榜的噪声底线
Abstract
Modern LLM evaluation assumes that pinning a judge to a fixed model snapshot and decoding at temperature zero yields reproducible verdicts. We show this assumption fails as a property of how LLM-as-Judge is operationalized on cloud serving infrastructure, not of any particular model family. Across four frontier judges all served via a single major enterprise cloud platform and three standard benchmarks (Arena-Hard, AlpacaEval 2, MT-Bench), identical inputs to the same pinned, temperature-zero judge produce different verdicts across re-runs: per-item flip rates of roughly 5% on average and about 40% on the close-call items that decide leaderboard margins, with a per-judge magnitude spanning a 40x range (from 0.13% to nearly 10%). We introduce metrics tailored to this instability: per-item flip rate, a two-part stability profile (waver fraction and conditional intensity), and adjacency separability - and report what the variance does and does not do to rankings. For a single judge the aggregate ranking is stable (0% top-K instability, 0% pooled winner flip); what degrades is precision: under a paired hierarchical bootstrap, roughly one-fifth to three-quarters of adjacent leaderboard positions are statistically indistinguishable, a noise floor we attribute primarily to finite prompt sampling rather than to the judge. Across judges, leaderboards agree on the coarse ordering but diverge in the middle (Kendall's tau as low as 0.42-0.64 between families on Arena-Hard), and of 13 published head-to-head ranking claims we re-judge, 5 fail under a defensible judge swap or re-run. We argue that leaderboards report unhedged point estimates that misrepresent the noise floor of the instrument, and we propose a minimal, low-cost reporting protocol: several judge re-runs, published stability profiles and adjacency intervals, and results under at least two judges from different families.
Chinese Translation
现代LLM评估假设将法官固定在一个固定的模型快照上，并在零温度下解码可以产生可重复的判决。我们展示了这一假设在LLM作为法官的云服务基础设施操作化过程中失败，而不是任何特定模型系列的属性。在四个前沿法官中，所有法官均通过一个主要企业云平台提供服务，并在三个标准基准（Arena-Hard、AlpacaEval 2、MT-Bench）上进行测试：对同一固定、零温度法官的相同输入在重新运行中产生不同的判决：每个项目的翻转率平均约为5%，在决定排行榜边际的接近判决项目中约为40%，每位法官的幅度跨度达到40倍（从0.13%到近10%）。我们引入了针对这种不稳定性的度量标准：每个项目的翻转率、两部分稳定性特征（波动比例和条件强度）以及邻接可分离性，并报告方差对排名的影响和不影响。对于单一法官，整体排名是稳定的（0%顶级K不稳定性，0%汇总赢家翻转）；下降的是精确度：在配对层次自助法下，大约五分之一到四分之三的相邻排行榜位置在统计上是不可区分的，这一噪声底线我们主要归因于有限的提示采样而非法官本身。在法官之间，排行榜在粗略排序上达成一致，但在中间部分出现分歧（在Arena-Hard上，不同系列之间Kendall's tau低至0.42-0.64），在我们重新评估的13个已发布的面对面排名声明中，有5个在合理的法官交换或重新运行下失败。我们认为排行榜报告的未对冲点估计错误地表示了工具的噪声底线，并提出了一种最低成本的报告协议：多次法官重新运行、发布稳定性特征和邻接区间，以及至少来自不同系列的两位法官的结果。
cs.CL / 77 / 2609.33065
Reading Too Much into Context: Passive Exposure Can Steer LLM Decisions
过度解读上下文：被动曝光可以影响大型语言模型的决策
Abstract
Large language model (LLM) assistants can now search the web and consult external sources while completing user requests. These sources can provide useful evidence, but they can also introduce additional content into the model's context. Can such passive exposure steer a decision even when the added content provides no reason to change it? We examine the stability of model decisions on the same tasks with and without such external content. Across all open-weight and closed-weight models we test, exposure systematically shifts decisions, with effects reaching nearly 50 percentage points in closed-weight models. The same pattern appears with real-world online opinions. The influence also extends beyond subjective preferences. Such exposure can steer models toward choices that violate explicit user requirements and increase their acceptance of false claims. In short, what enters an LLM's context can influence its decision even when it should not determine it.
Chinese Translation
大型语言模型（LLM）助手现在可以在完成用户请求时搜索网络并咨询外部来源。这些来源可以提供有用的证据，但也可能向模型的上下文中引入额外内容。这样的被动曝光是否会影响决策，即使所添加的内容并未提供改变决策的理由？我们考察了在有和没有这种外部内容的情况下，模型在相同任务上的决策稳定性。在我们测试的所有开放权重和闭合权重模型中，曝光系统性地改变了决策，闭合权重模型的影响甚至达到近50个百分点。真实世界的在线意见也显示出相同的模式。这种影响不仅限于主观偏好。这样的曝光可能会引导模型做出违反明确用户要求的选择，并增加它们对虚假声明的接受度。简而言之，进入LLM上下文的内容可以影响其决策，即使在理应不影响决策的情况下。
cs.CL / 78 / 2609.33090
OneSign: Unifying Sign Language Understanding Tasks with One Model
OneSign：用一个模型统一手语理解任务
Abstract
SLU encompasses a diverse set of tasks, including ISLR, CSLR, and SLT. Although these tasks share basic semantic and linguistic foundations, they are typically addressed with task-specific architectures and training pipelines, which hinders knowledge sharing and requires costly pretraining and finetuning for each task. In this paper, we focus on two aspects of SLU tasks: (1) training and inference pipelines are highly fragmented: most methods rely on pretraining on large-scale SL datasets followed by task- or dataset-specific finetuning, which leads to multiple specialized models rather than a single checkpoint. (2) current LLM-based methods may overlook the inherent modality discrepancy between sign and text tokens, simply concatenating them and processing both modalities with the same decoder layers. In this paper, we present OneSign, a unified framework that addresses multiple SLU tasks within a single model and a single checkpoint. OneSign reformulates ISLR, CSLR, and SLT under a single training paradigm. To accommodate the heterogeneous characteristics of sign and text representations, we introduce a Modality-Adaptive Mixture-of-Experts (MA-MoE) architecture, consisting of a shared expert and modality-specific experts for sign and text tokens. A modality router dynamically activates the corresponding experts, and their outputs are aggregated to form the final token representations. By enabling modality-dependent expert specialization while preserving a shared expert path, MA-MoE can effectively model the modality differences between continuous sign representations and discrete text tokens. Extensive experiments on multiple benchmarks demonstrate that OneSign achieves competitive or state-of-the-art performance on several benchmarks, highlighting its effectiveness as a unified SLU model. Datasets are available at : https://github.com/gswycf/OneSign.
Chinese Translation
手语理解（SLU）涵盖了一系列多样化的任务，包括手语识别（ISLR）、手语翻译（CSLR）和手语文本（SLT）。尽管这些任务共享基本的语义和语言基础，但通常采用特定于任务的架构和训练流程，这阻碍了知识共享，并且需要为每个任务进行昂贵的预训练和微调。本文关注SLU任务的两个方面：（1）训练和推理流程高度碎片化：大多数方法依赖于在大规模手语数据集上进行预训练，然后进行特定于任务或数据集的微调，这导致多个专门模型的产生，而不是单一的检查点。（2）当前基于大型语言模型（LLM）的方法可能忽视了手语和文本标记之间固有的模态差异，简单地将它们连接在一起，并用相同的解码层处理这两种模态。本文提出了OneSign，一个统一框架，可以在单一模型和单一检查点内处理多个SLU任务。OneSign在单一训练范式下重新定义了ISLR、CSLR和SLT。为了适应手语和文本表示的异质特征，我们引入了一种模态自适应专家混合（Modality-Adaptive Mixture-of-Experts，MA-MoE）架构，该架构由一个共享专家和针对手语和文本标记的模态特定专家组成。模态路由器动态激活相应的专家，并将它们的输出聚合以形成最终的标记表示。通过实现模态依赖的专家专业化，同时保留共享专家路径，MA-MoE能够有效建模连续手语表示和离散文本标记之间的模态差异。在多个基准上的广泛实验表明，OneSign在多个基准上实现了具有竞争力或最先进的性能，突显了其作为统一SLU模型的有效性。数据集可在：https://github.com/gswycf/OneSign获取。
cs.CL / 79 / 2609.33119
MedRouter: Demystifying Knowledge Differences Across Medical LLMs for Routing-Based Reasoning
MedRouter：揭示医疗大型语言模型之间知识差异的基于路由的推理
Abstract
Medical question answering spans diverse specialties and modalities, and individual medical large language models (LLMs) exhibit distinct strengths across tasks and domains. This heterogeneity suggests that combining specialists may enable broader coverage of medical questions than relying on any single model. However, existing LLM routing methods primarily seek to balance answer quality and inference cost, leaving open how to exploit differences in specialist competence to improve medical reasoning. In this paper, we introduce MedRouter, an agentic system that uses an embedding-based multi-label router to select and query specialist LLMs, then passes their responses to a generator to produce the final answer. We further propose SCALE (Specialist Competence-Aware Learning), a two-stage training framework that first trains the Router with specialist correctness supervision and then optimizes its selections through reinforcement learning. The second stage uses a Performance Gain Reward (PGR) that measures how specialist information affects the generator's answer correctness relative to answering without that information. Experiments on eight text-based and multimodal medical QA benchmarks show that MedRouter outperforms the strongest routing baseline by 8% in average accuracy. Our analysis of specialist outputs further reveals distinct strengths and complementary question-level coverage, motivating learned routing to combine these capabilities for more comprehensive medical reasoning.
Chinese Translation
医疗问答涵盖了多种专业领域和模式，各个医疗大型语言模型（LLMs）在任务和领域上表现出不同的优势。这种异质性表明，结合多个专业模型可能比依赖单一模型更能广泛覆盖医疗问题。然而，现有的LLM路由方法主要旨在平衡答案质量和推理成本，尚未探讨如何利用专业能力的差异来改善医疗推理。在本文中，我们介绍了MedRouter，这是一种代理系统，使用基于嵌入的多标签路由器来选择和查询专业LLM，然后将它们的响应传递给生成器以产生最终答案。我们进一步提出了SCALE（专业能力感知学习），这是一种两阶段训练框架，首先在专业正确性监督下训练路由器，然后通过强化学习优化其选择。第二阶段使用性能增益奖励（Performance Gain Reward, PGR），衡量专业信息对生成器答案正确性的影响，相较于没有该信息的回答。对八个基于文本和多模态的医疗问答基准的实验表明，MedRouter在平均准确率上比最强的路由基线提高了8%。我们对专业输出的分析进一步揭示了不同的优势和互补的问题级覆盖，激励学习路由结合这些能力以实现更全面的医疗推理。
cs.CL / 80 / 2609.33121
Classifying Dominant Temporal Orientation without Pretrained Text Embeddings: A Novel Morphosyntactic Inventory Vector Approach
无预训练文本嵌入的主导时间取向分类：一种新颖的形态句法库存向量方法
Abstract
Computational methods have consistently struggled to determine the dominant temporal orientation of a sentence. This difficulty is especially pronounced when a sentence contains multiple embedded clauses with competing tense and aspectual information. To address this difficulty, we propose an alternative approach for identifying a sentence's past, present, or future global reference interval. Our method does not use any form of pretrained embeddings. We rather encode sentences using fixed-length inventory vectors that are comprised of part-of-speech counts, dependency relation counts, and explicit futurate pattern counts. We term this inventory vector of a sentence a "Morphosyntacton". The method does not use any padding, sequence models, or large language models. Evaluation on 1,799 syntactically complex English sentences, annotated as past, present or future, shows balanced and high accuracy multiclass classification, achieving an overall multi-class accuracy of 92%.
Chinese Translation
计算方法在确定句子的主导时间取向方面一直面临挑战。当句子包含多个嵌入从句，并且这些从句之间存在竞争的时态和体态信息时，这种困难尤为明显。为了解决这一问题，我们提出了一种替代方法，用于识别句子的过去、现在或未来的全球参考区间。我们的方法不使用任何形式的预训练嵌入。相反，我们使用固定长度的库存向量对句子进行编码，这些向量由词性计数、依赖关系计数和显式未来模式计数组成。我们将句子的这一库存向量称为“形态句法向量”（Morphosyntacton）。该方法不使用任何填充、序列模型或大型语言模型。在1,799个被标注为过去、现在或未来的句法复杂的英语句子上进行的评估显示，该方法实现了平衡且高准确度的多类分类，整体多类准确率达到92%。
cs.CL / 81 / 2609.33142
Knowing Is Not Choosing: What Explicit Verification Adds Beyond Generative Preference
知道并不等于选择：显式验证在生成偏好之外的附加价值
Abstract
Generating a correct answer does not mean that a language model will select it. We separate factual recall into three steps: generating a correct candidate, ranking the available candidates, and selecting the final answer. Pre-generation readouts predict factual recall and which questions sampling will cover across three model families, but say little about whether an available correct answer will ultimately be selected. Explicit verification with $P(\mathrm{True})$ improves within-question ranking over mean log-likelihood in Gemma, Qwen3, and Llama, with AUROC gains of $0.08$--$0.12$. In a prospectively defined Gemma cohort, verification raises plurality accuracy by about $5$ points, and still gains about $2$ points over chat-template likelihood, a stronger generative baseline. The advantage is strongest for relations with common-answer priors and depends on access to the entity; masking the entity removes the ranking advantage in larger Qwen models. Finally, the measured benefit depends on how correctness is defined: recall-oriented reference matching can credit option lists favored by likelihood and substantially understate the improvement seen under human semantic judgments.
Chinese Translation
生成正确答案并不意味着语言模型会选择它。我们将事实回忆分为三个步骤：生成正确候选答案、对可用候选答案进行排名以及选择最终答案。预生成的读出结果可以预测事实回忆以及在三个模型系列中采样将覆盖哪些问题，但对于可用的正确答案是否最终会被选择则几乎没有提供信息。使用 $P( ext{True})$ 进行显式验证在 Gemma、Qwen3 和 Llama 中提升了问题内排名，相较于均值对数似然，AUROC 增益为 $0.08$--$0.12$。在前瞻性定义的 Gemma 队列中，验证将多数准确率提高了约 $5$ 个百分点，并且相较于聊天模板的似然，仍然在更强的生成基线中获得了约 $2$ 个百分点的增益。这一优势在具有常见答案先验的关系中最为明显，并且依赖于对实体的访问；对实体进行掩蔽会消除在较大 Qwen 模型中的排名优势。最后，测量的益处取决于正确性的定义：以回忆为导向的参考匹配可能会给予倾向于似然的选项列表以信用，并显著低估在人工语义判断下所观察到的改进。
cs.CL / 82 / 2609.33143
CAME: Company-Aware Evidence-Memory Experts for Interpretable Quarter-Ahead Revenue Forecasting
CAME：公司意识的证据记忆专家用于可解释的季度前收入预测
Abstract
Quarter-ahead revenue forecasting requires company-scale numerical accuracy, strict temporal validity, and company-specific interpretation of narrative disclosures. LLMs can distill textual evidence but can produce scale-misaligned forecasts, whereas history-based anchors are stable but miss forecast-time signals such as product transitions, supply constraints, and management guidance. We introduce CAME (Company-Aware Evidence-Memory Experts), a residual-forecasting framework that refines a no-leakage statistical anchor when current semantic evidence and prior error patterns justify an adjustment. On a development-inclusive rolling backtest of 336 company-quarters from 12 large public technology and platform firms, CAME achieves the lowest aggregate point-estimate error among the reported methods, with statistically supported macro-sMAPE gains over the matched Statistical Anchor, and outperforms History + Guidance on all six aggregate metrics. CAME also links adjustments to source-linked evidence cards and guarded memory traces, supporting forecast inspection, provenance, and failure localization.
Chinese Translation
季度前收入预测需要公司层面的数值准确性、严格的时间有效性以及对叙述披露的公司特定解释。大型语言模型（LLMs）能够提炼文本证据，但可能产生与规模不匹配的预测，而基于历史的锚点则稳定但缺乏预测时信号，例如产品过渡、供应限制和管理指导。我们提出了CAME（公司意识的证据记忆专家），这是一种残差预测框架，当当前语义证据和先前错误模式证明需要调整时，能够优化无泄漏的统计锚点。在对12家大型公共科技和平台公司的336个公司季度进行的包含开发的滚动回测中，CAME在报告的方法中实现了最低的总体点估计误差，且在匹配的统计锚点上获得了统计支持的宏观sMAPE增益，并在所有六个总体指标上优于历史+指导。CAME还将调整与源链接的证据卡片和受保护的记忆痕迹相联系，支持预测检查、来源追溯和故障定位。
cs.CL / 83 / 2609.33150
Generalization Dynamics of LM Pre-training
语言模型预训练的泛化动态
Abstract
People typically assume that LMs stably mature from pattern-matching parrots to generalizable intelligence during pre-training. We build a toy eval suite and show this mental model is wrong: throughout pre-training, LMs frequently and suddenly hop between parrot-like and intelligence-like computations. We call this mode-hopping. Across our suite, LMs suddenly latch onto memorized or in-context patterns instead of in-context learning, use System 1 instead of System 2 thinking, pick up what sounds true instead of what is true, fail at multi-hop persona QA, out-of-context reasoning, and emergent misalignment -- then just as suddenly revert and generalize. Mode-hopping is not explained by standard optimization dynamics: it is locally stable and cannot be fixed by checkpoint averaging. We instead think of it as a capacity allocation problem: in a capacity-bounded model, generalizable circuits must compete with the shallow ones learned early in training, and the data in each pre-training window may decide which circuits win. Our suite provides a new efficient lens on generalization. We demonstrate two concrete applications: (i) select intermediate pre-training checkpoints that strongly generalize reasoning and alignment, better than the final pre- or mid-training checkpoints, and (ii) select pre-training data that controls and stabilizes generalization dynamics.
Chinese Translation
人们通常认为语言模型（LMs）在预训练过程中稳定地从模式匹配的鹦鹉转变为具有泛化能力的智能。然而，我们构建了一个玩具评估套件，显示这一心理模型是错误的：在整个预训练过程中，语言模型频繁且突然地在鹦鹉式和智能式计算之间跳跃。我们称这种现象为模式跳跃。在我们的评估套件中，语言模型突然依赖于记忆或上下文中的模式，而不是进行上下文学习，使用系统1而非系统2的思维，选择听起来正确而非真实的内容，在多跳个性问答、上下文外推理和新兴的不一致性方面表现不佳——然后又同样突然地回归并进行泛化。模式跳跃无法通过标准优化动态解释：它在局部是稳定的，无法通过检查点平均来修复。我们更倾向于将其视为一种能力分配问题：在一个能力受限的模型中，具有泛化能力的电路必须与在训练早期学习到的浅层电路竞争，而每个预训练窗口中的数据可能决定哪些电路获胜。我们的评估套件为泛化提供了一种新的高效视角。我们展示了两个具体应用：（i）选择能够强烈泛化推理和一致性的中间预训练检查点，比最终的预训练或中期检查点更有效；（ii）选择能够控制和稳定泛化动态的预训练数据。
cs.CL / 84 / 2609.33155
Where Do Test-Time Scaling and Training Fall Short in Individual Stance Prediction?
测试时缩放和训练在个体立场预测中的不足之处在哪里？
Abstract
Test-time scaling and post-training have improved LLM performance in coding and mathematical reasoning, but their effectiveness for individual stance prediction remains unclear. We study this question by predicting a person's stance in a new discussion from their history. We evaluate widely used test-time scaling strategies and post-training methods, such as supervised fine-tuning and reinforcement learning, and identify four failure modes across generation, selection, and learning: (1) incorrect consensus, where repeated samples agree on the wrong stance; (2) selection failure, where generation covers the observed stance but selection misses it; (3) response overfitting, where supervised fine-tuning improves imitation but harms prediction; and (4) early plateau, where reinforcement learning shows modest initial gains followed by limited further improvement. We expose these failures using STANCE-BENCH, which contains 2499 prediction tasks from 500 Hacker News users. Guided by this analysis, we explore a simple approach that combines direct scores for all candidate stances with explicit assessments of support from the individual's history. On the 781-task test set, this approach achieves 21.83 discussion-specific Macro F1 with Qwen3-8B, compared with 19.27 for direct scoring. Our results motivate evaluating candidate generation, final selection, and person-specific evidence use separately.
Chinese Translation
测试时缩放和后训练提高了大型语言模型（LLM）在编码和数学推理方面的表现，但它们在个体立场预测中的有效性仍不明确。我们通过从个人的历史中预测其在新讨论中的立场来研究这个问题。我们评估了广泛使用的测试时缩放策略和后训练方法，如监督微调和强化学习，并识别出在生成、选择和学习过程中的四种失效模式：（1）错误共识，即重复样本在错误立场上达成一致；（2）选择失败，即生成覆盖了观察到的立场但选择却遗漏了它；（3）响应过拟合，即监督微调提高了模仿能力但损害了预测能力；（4）早期平台期，即强化学习在初期显示出适度的增益，但随后改进有限。我们使用 STANCE-BENCH 这一数据集揭示了这些失效，该数据集包含来自 500 名 Hacker News 用户的 2499 个预测任务。在此分析的指导下，我们探索了一种简单的方法，将所有候选立场的直接评分与个人历史的支持显性评估相结合。在 781 个任务的测试集上，该方法在 Qwen3-8B 上实现了 21.83 的讨论特定宏 F1，而直接评分的结果为 19.27。我们的结果激励我们分别评估候选生成、最终选择和个体特定证据的使用。
cs.CL / 85 / 2609.33179
Scoring the Wrong Question: Readout Failures in Constrained-Option Evaluation
评分错误的问题：受限选项评估中的读取失败
Abstract
Constrained-option scoring reads a model's probabilities for a fixed set of permitted answers, so it returns a score even when the model is about to write something else. We study a prompt that quotes a multiple-choice item, ending in the item's own answer instruction, and asks for a forecast of whether a reader model given only a short note will answer it correctly. The model can begin answering the quoted item: on three Qwen3 checkpoints an answer letter is the most probable token on nearly every passage, and the forecast ranks the reader's correctness indistinguishably from chance. Beyond prior work's argmax check and prefill repair, we contribute a label-free diagnostic of the declared options' support, their first-token mass in context; a deletion test tracing the Qwen3 answer-letter start to the quoted instruction; and rendering and tokenisation checks for dropped separators and coinciding first tokens. Prefilling an answer stem lifts the median option mass from below to above one half on all nine checkpoints scored with this prompt and, with options and renormalisation unchanged, raises the Qwen3-averaged AUC from 0.4929 to 0.6152. lm-polygraph's default P(True) estimator shows a related failure: it reads True where Qwen3 without thinking favours an answer letter (MMLU) or the prompt's (A)/(B) label (TriviaQA), and its Qwen3-averaged AUROC is indistinguishable from chance on MMLU and below chance on TriviaQA. Scoring and renormalising the (A)/(B) labels after an answer stem raises it to 0.632 and 0.868, renormalising True against False in place reaches 0.660 and 0.860, and on TriviaQA both exceed all fourteen default single-answer library estimators. The original forecast is renormalised too, yet stays indistinguishable from chance: option mass flags low support in both settings without labels, and only checking the intended target shows which score still ranks.
Chinese Translation
受限选项评分读取模型对固定允许答案集的概率，因此即使模型即将写出其他内容，它也会返回一个分数。我们研究了一个引用多项选择题的提示，该提示以题目的答案指令结束，并询问在仅给出简短说明的情况下，读者模型是否会正确回答该题。模型可以开始回答引用的项目：在三个 Qwen3 检查点上，答案字母几乎是每个段落中最可能的标记，而预测的读者正确性与随机猜测没有区别。除了之前工作的 argmax 检查和预填修复外，我们贡献了一种无标签的诊断方法，评估声明选项的支持度，即它们在上下文中的首标记质量；通过删除测试追踪 Qwen3 答案字母的开始到引用指令；以及对丢失分隔符和重合首标记的呈现和标记化检查。预填答案干扰将所有九个使用此提示评分的检查点的中位选项质量从低于一半提升至高于一半，并且在选项和重新归一化不变的情况下，将 Qwen3 平均 AUC 从 0.4929 提升至 0.6152。lm-polygraph 的默认 P(True) 估计器显示出相关的失败：它在 Qwen3 未思考时读取 True，倾向于答案字母（MMLU）或提示的 (A)/(B) 标签（TriviaQA），其 Qwen3 平均 AUROC 在 MMLU 上与随机猜测没有区别，在 TriviaQA 上低于随机猜测。在答案干扰后对 (A)/(B) 标签进行评分和重新归一化将其提升至 0.632 和 0.868，而在位置上对 True 与 False 进行重新归一化则达到 0.660 和 0.860，并且在 TriviaQA 上均超过所有十四个默认单答案库估计器。原始预测也进行了重新归一化，但仍与随机猜测没有区别：选项质量在无标签的两种情况下都显示出低支持度，只有检查预期目标才能显示哪个分数仍然排名靠前。
cs.CL / 86 / 2609.33183
Identifying Temporal Features within Transcoders for Time Sensitive Factual Recall
识别转码器中的时间特征以实现时间敏感的事实回忆
Abstract
Large Language Models (LLMs) suffer from temporal misalignment, often due to the contradictory nature of their training corpora. While current mitigation strategies rely on computationally expensive fine-tuning or context-heavy retrieval augmented generation (RAG), the internal mechanisms governing time-sensitive recall remain under-explored. Unlike prior studies that identify temporal components such as attention heads and MLP layers, we provide the first feature-level map of temporal recall by isolating individual MLP features via transcoder circuit tracing. We identify three node categories (common temporal, common to the year, and chrono-semantic) which interact to generate a temporal filter during factual recall. By analysing Gemma 2 2B, LLaMA 3.2 1B, and Qwen3-4B, we show that these features do not follow a simple linear pipeline but represent time through a parallel and mixed syntactic-semantic interplay across layers. We additionally discover a class of higher-layer temporal components invisible to existing EAP-IG methods, establishing transcoders as a more complete lens for temporal interpretability in time-sensitive factual recall. These findings present MLP components for potential targeted interventions in time-sensitive factual recall
Chinese Translation
大型语言模型（LLMs）存在时间错位的问题，这通常是由于其训练语料的矛盾性质所致。目前的缓解策略依赖于计算成本高昂的微调或上下文密集的检索增强生成（RAG），而支配时间敏感回忆的内部机制仍然未得到充分探索。与之前识别时间组件（如注意力头和多层感知器（MLP）层）的研究不同，我们通过转码器电路追踪首次提供了时间回忆的特征级图谱，通过隔离单个MLP特征来实现。我们识别出三类节点（常见时间节点、与年份相关的节点和时间语义节点），这些节点相互作用以在事实回忆过程中生成时间过滤器。通过分析Gemma 2 2B、LLaMA 3.2 1B和Qwen3-4B，我们表明这些特征并不遵循简单的线性流程，而是通过跨层的平行和混合句法-语义相互作用来表示时间。此外，我们发现了一类现有EAP-IG方法无法识别的高层时间组件，确立了转码器作为时间敏感事实回忆中时间可解释性的更完整视角。这些发现为时间敏感事实回忆中的潜在针对性干预提供了MLP组件。
cs.CL / 87 / 2609.33204
Turning Speech Language Models into Multilingual Listeners
将语音语言模型转变为多语言听众
Abstract
Speech Language Models (SLMs) that understand spoken language questions support only a few high-resource languages, limiting access to millions of people worldwide. This gap stems from the scarcity of multilingual speech instruction-tuning datasets. We present MULTISPEECHQA, a large-scale, synthetically generated and human-verified dataset comprising 9200 hours of 10.8 million spoken question-answer pairs in 23 typologically diverse languages. Using MULTISPEECHQA, we also introduce MULTISPEECH-BENCH, a multi-task benchmark for evaluating SLM performance on 23 languages. We compare the performance of a cascading system to open-weight and closed SLMs on MULTISPEECH-BENCH and find that the cascading system outperforms open-weight SLMs but not all closed SLMs. We use MULTISPEECHQA to finetune Qwen 2.5-Omni, which improves its performance on our benchmark. Our findings show that high-quality synthetic datasets offer a cheap solution to improving the multilingual capabilities of SLMs.
Chinese Translation
理解口语问题的语音语言模型（SLMs）仅支持少数高资源语言，这限制了全球数百万人的访问。这个差距源于多语言语音指令微调数据集的稀缺。我们提出了MULTISPEECHQA，这是一个大规模、合成生成且经过人工验证的数据集，包含9200小时、1080万个口语问答对，涵盖23种类型多样的语言。利用MULTISPEECHQA，我们还引入了MULTISPEECH-BENCH，这是一个用于评估SLM在23种语言上性能的多任务基准。我们比较了级联系统与开放权重和封闭SLMs在MULTISPEECH-BENCH上的表现，发现级联系统的表现优于开放权重SLMs，但并非所有封闭SLMs。我们使用MULTISPEECHQA对Qwen 2.5-Omni进行微调，从而提高其在我们基准上的表现。我们的研究结果表明，高质量的合成数据集为提升SLMs的多语言能力提供了一种廉价的解决方案。
cs.CL / 88 / 2609.33209
Beyond Calibration: Do a Typed-Decision Model's Probabilities Obey the Probability Axioms?
超越校准：类型决策模型的概率是否遵循概率公理？
Abstract
Typed-decision models such as TypeSafe's Jev answer a declared yes/no or multiple-choice question about a state with a probability instead of text, and their evaluations report accuracy and calibration. Neither requires that the probabilities a model gives to logically related questions fit together item by item, which is what a system that acts on those probabilities needs. We test this property, coherence, with a battery of logically linked questions that needs no labels. For 160 items from ChaosNLI and PubMedQA, each with three mutually exclusive labels, we ask whether the label is X, whether it is not X, whether it is one of the other two labels, and which label applies. On 480 negation pairs, Jev's probabilities for "the label is X" and "the label is not X" miss summing to one by 0.064 on average (95% CI 0.055 to 0.072). Qwen3.8-27B, run from its official BF16 weights, misses by 0.293 with first-token probabilities and by 0.122 with verbalized probabilities. The gap to the first-token readout persists on pairs where both systems give similar probabilities, without double-negation labels, and after averaging Jev's repeated calls. Jev is not coherent either: its violations are about five times its repeat noise, and it over-endorses statements about single labels, so that its three single-label probabilities sum to 1.14 on average. The two systems also fail differently. Qwen3.8-27B's first-token readout under-endorses the complement of a label whether or not the question contains "not", rejecting both a statement and its negation in 196 of 480 pairs, and it does not become more coherent where it is more confident, whereas Jev's violations concentrate where its answer is uncertain. Because the checks need no labels, they expose biases that appear only when question forms are compared, and inconsistencies within items.
Chinese Translation
类型决策模型，如 TypeSafe 的 Jev，针对一个状态的声明性是/否或多选问题给出概率而非文本，其评估报告准确性和校准性。然而，这些模型给出的与逻辑相关问题的概率并不要求逐项一致，而这是一个基于这些概率进行操作的系统所需的特性。我们通过一系列逻辑关联的问题测试这一特性，即一致性，这些问题不需要标签。对于来自 ChaosNLI 和 PubMedQA 的 160 个项目，每个项目有三个互斥的标签，我们询问标签是否为 X，是否不是 X，是否为其他两个标签之一，以及适用哪个标签。在 480 对否定对中，Jev 对于“标签是 X”和“标签不是 X”的概率平均相差 0.064（95% 置信区间 0.055 到 0.072），未能相加为 1。Qwen3.8-27B 在其官方 BF16 权重下，使用首个令牌概率相差 0.293，使用口头化概率相差 0.122。在两个系统给出相似概率的对中，首个令牌读出之间的差距仍然存在，且没有双重否定标签，并在平均化 Jev 的重复调用后依然如此。Jev 也不一致：其违反情况约为其重复噪声的五倍，并且它对单一标签的陈述过度支持，使得其三个单一标签的概率平均相加为 1.14。这两个系统的失败方式也不同。Qwen3.8-27B 的首个令牌读出在标签的补充上低估，无论问题中是否包含“不是”，在 480 对中的 196 对中同时拒绝一个陈述及其否定，并且在更自信的情况下并没有变得更一致，而 Jev 的违反则集中在其答案不确定的地方。由于这些检查不需要标签，它们揭示了仅在比较问题形式时出现的偏差，以及项目内部的不一致性。
cs.CL / 89 / 2609.33212
CoLMbo-SV: A Grounded Language Model for Explainable Speaker Verification
CoLMbo-SV：一种可解释的说话人验证的基础语言模型
Abstract
Speaker verification systems achieve high accuracy but provide little account of the acoustic evidence behind their judgments. Making these systems inspectable requires exposing interpretable evidence while retaining the richer information on which their decisions depend. We present \textbf{CoLMbo-SV}, a speaker language model that combines strong speaker discrimination with structured, acoustically grounded comparison reports. By connecting a pretrained speaker encoder to a language model and supplying explicit acoustic measurements, CoLMbo-SV makes voice comparisons inspectable without restricting verification to the evidence verbalized in its reports. We additionally introduce \textbf{VoxReason}, paired recordings with measured acoustic properties and comparison reports filtered through numerical and qualitative checks, providing supervision for this combined capability. We also develop an evaluation framework that separates what acoustic information a speaker representation encodes, what influences the verification score, and what the generated report discusses. On VoxCeleb1-O, CoLMbo-SV achieves 0.99\% EER, reducing verification error by approximately 80\% relative to the strongest audio-language baseline fine-tuned on VoxReason, while attaining a numerical-grounding score of 0.82. Our analysis further demonstrates that acoustic correctness and decision relevance are distinct properties of an explanation, exposing a gap that numerical-grounding metrics miss. Together, these contributions substantially advance audio-language speaker verification, bring its accuracy toward that of dedicated speaker encoders while adding checkable acoustic reporting, and establish an empirical framework for connecting natural-language explanations to the decisions they explain.
Chinese Translation
说话人验证系统虽然能够实现高准确率，但对其判断背后的声学证据却提供了很少的解释。使这些系统可检验的关键在于在保留其决策所依赖的丰富信息的同时，揭示可解释的证据。我们提出了 extbf{CoLMbo-SV}，这是一种结合了强大的说话人区分能力与结构化、声学基础的比较报告的说话人语言模型。通过将预训练的说话人编码器与语言模型连接，并提供明确的声学测量数据，CoLMbo-SV使得语音比较可检验，而不局限于其报告中所表述的证据。我们还引入了 extbf{VoxReason}，这是一组配对录音，具有测量的声学特性和经过数值与定性检查过滤的比较报告，为这一组合能力提供监督。我们还开发了一个评估框架，区分说话人表示编码的声学信息、影响验证分数的因素以及生成报告所讨论的内容。在VoxCeleb1-O数据集上，CoLMbo-SV达到了0.99\%的等错误率（EER），相比于在VoxReason上微调的最强音频-语言基线，验证错误减少了约80\\%，同时获得了0.82的数值基础评分。我们的分析进一步表明，声学正确性和决策相关性是解释的两个不同特性，揭示了数值基础指标所忽略的差距。这些贡献共同显著推动了音频-语言说话人验证的发展，使其准确性接近专用说话人编码器，同时增加了可检验的声学报告，并建立了一个将自然语言解释与其所解释的决策相连接的实证框架。
cs.CL / 90 / 2609.33226
Beyond Memory Construction: Rethinking Memory Access for LLM-based Conversational Agents
超越记忆构建：重新思考基于大语言模型的对话代理的记忆访问
Abstract
Memory is a core component of conversational agents, enabling coherent and context-aware behavior over long interactions. Recent approaches commonly rely on LLM-based memory construction, where raw interactions are rewritten into structured memory units and later retrieved via a RAG pipeline. While effective in controlled settings, we show that this paradigm breaks down in long-horizon, high-entropy conversations: memory construction becomes increasingly lossy and unstable as context length and information complexity grow, and incurs prohibitive cost due to repeated LLM invocation. To address these limitations, we propose Threader, a memory system that shifts the focus from memory construction to efficient, structure-aware access over raw interactions. Instead of rewriting interactions, Threader preserves them as first-class memory, organizes them into topic-coherent segments via lightweight incremental segmentation, and enables accurate retrieval through multi-view representation. At query time, it performs multi-signal retrieval that combines segment-level access with localized evidence matching, ensuring both completeness and coherence. Extensive experiments demonstrate that Threader consistently improves answer accuracy and evidence recall, while significantly reducing the memory construction overhead.
Chinese Translation
记忆是对话代理的核心组成部分，使其在长时间交互中能够保持连贯和上下文感知的行为。近期的方法通常依赖于基于大语言模型（LLM）的记忆构建，其中原始交互被重写为结构化的记忆单元，并通过检索增强生成（RAG）管道进行后续检索。尽管在受控环境中有效，但我们表明这种范式在长时间、高熵的对话中会失效：随着上下文长度和信息复杂性的增加，记忆构建变得越来越有损和不稳定，并因重复调用LLM而产生高昂的成本。为了解决这些局限性，我们提出了Threader，一种记忆系统，旨在将重点从记忆构建转向对原始交互的高效、结构感知访问。Threader不再重写交互，而是将其作为一等公民的记忆保留，通过轻量级增量分段将其组织成主题一致的片段，并通过多视图表示实现准确检索。在查询时，它执行多信号检索，结合片段级访问与局部证据匹配，确保完整性和连贯性。大量实验证明，Threader在提高答案准确性和证据召回率的同时，显著减少了记忆构建的开销。
cs.CL / 91 / 2609.33280
The Text Beside the Image: Detection, Utility and Leakage for Trustworthy Multimodal Medical Data and Beyond
图像旁的文本：可信多模态医学数据及其他领域的检测、实用性与泄漏
Abstract
Medical images are released with the reports that describe them, and protecting the image does not protect the report. This paper measures the text component of such releases. We measure identifier detection, downstream utility and residual identity leakage on the same documents, with the pseudonymisation policy as the variable under test: 15 detectors, three release conditions and four corpora of medical reports, legal judgments, news and other genres, and e-mail, in German, English, Chinese and Arabic. A fixed 13-detector union reaches a person sensitivity of 0.9998 at specificity 0.8686 on the medical reports, 0.9958 at 0.8504 on the legal judgments, 0.9352 at 0.9318 on news and other genres, and 0.9906 at 0.6235 on e-mail. With this ensemble, frequency matching with a public name list recovers zero identities by alignment across the four corpora; the names it got right were ones the detector missed, left in clear text. Cross-document linkage ranks the correct person first for 0.93% of e-mail queries without training and 3.94% with it, against 1/3697 chance and 71.98% on unmodified text. On the medical reports it recovers nothing without training and 0.71% of 138 queries with it, against 1/207 chance and a 2.73% ceiling on unmodified text.
Chinese Translation
医学图像通常与描述其内容的报告一起发布，而保护图像并不等于保护报告。本文测量了此类发布中的文本成分。我们在相同文档上测量了标识符检测、下游实用性和剩余身份泄漏，以假名化政策作为测试变量：15个检测器、三种发布条件以及四个语料库，包括医学报告、法律判决、新闻及其他类型，以及电子邮件，涵盖德语、英语、中文和阿拉伯语。一个固定的13个检测器的联合体在医学报告中以0.9998的人敏感度和0.8686的特异性达成，在法律判决中为0.9958和0.8504，在新闻及其他类型中为0.9352和0.9318，在电子邮件中为0.9906和0.6235。利用该集合，通过与公共姓名列表的频率匹配，在四个语料库之间的对齐中未恢复任何身份；正确识别的姓名是检测器遗漏的，仍以明文形式存在。跨文档链接在未训练的情况下对0.93%的电子邮件查询将正确的人排在首位，经过训练则为3.94%，而未修改文本的概率为1/3697和71.98%。在医学报告中，未经过训练时未恢复任何身份，经过训练则对138个查询恢复了0.71%，而未修改文本的概率为1/207和2.73%的上限。
cs.CL / 92 / 2609.33290
Calibration, Not Answer Selection: Distilling Internal Confidence in Reasoning Models
校准，而非答案选择：提炼推理模型中的内部信心
Abstract
Reinforcement learning with binary correctness rewards trains correctness, not calibrated confidence. The confidence that reasoning models verbalize is systematically overconfident, and the problem is not merely one of scale: verbalized confidence tracks how willing a model is to commit to an answer, not how likely the answer is to be right. Post-hoc rescaling therefore fits one distribution but rarely transfers. We look inside the model instead. On factual question answering, a linear probe on the hidden state between the chain of thought and the answer is substantially better calibrated: its expected calibration error is 5 to 38 times lower than that of the verbalized score across four benchmarks and two model families. However, when used to pick among N sampled answers, that same probe nearly ties majority voting yet falls far short of the oracle. Internal states answer "how certain am I" well and "which answer is right" poorly, so the signal should be reported as a confidence rather than used to select answers. As a result, we introduce probe-guided self-distillation (Probe-SD): score a model's own sampled traces with the probe, overwrite the confidence each trace states, and finetune the base checkpoint of the same family, so nothing but the model itself remains at test time. On Qwen3-14B, Probe-SD cuts ECE from 0.178 to 0.024 in-domain and from 0.542 to 0.113 out-of-domain, where it also beats post-hoc recalibration and self-consistency distillation. The resulting confidence is well-calibrated and useful for weighted voting, behaviors previously attributed to online RL, here obtained with supervised finetuning alone.
Chinese Translation
使用二元正确性奖励的强化学习训练的是正确性，而非校准的信心。推理模型所表达的信心系统性地过于自信，而这个问题不仅仅是规模的问题：表达的信心反映的是模型对答案的承诺程度，而非答案正确的可能性。因此，事后重标定虽然适合一种分布，但很少能够迁移。我们转而深入模型内部。在事实问答中，链式思维与答案之间的隐藏状态上的线性探测器具有显著更好的校准性：其预期校准误差在四个基准和两个模型系列中比表达的分数低5到38倍。然而，当用于在N个采样答案中进行选择时，该探测器几乎与多数投票持平，但远不及预言机。内部状态在回答“我有多确定”时表现良好，而在回答“哪个答案是正确的”时表现较差，因此该信号应作为信心报告，而非用于选择答案。因此，我们引入了探测引导自蒸馏（Probe-SD）：用探测器对模型自身的采样轨迹进行评分，覆盖每个轨迹所表达的信心，并微调同一系列的基础检查点，以便在测试时只保留模型本身。在Qwen3-14B上，Probe-SD将领域内的ECE从0.178降低到0.024，领域外的ECE从0.542降低到0.113，同时也优于事后重标定和自一致性蒸馏。最终得到的信心经过良好校准，并且对于加权投票非常有用，这些行为以前被归因于在线强化学习，而在这里仅通过监督微调获得。
cs.CL / 93 / 2609.33296
BaatCheet: A Multilingual Corpus for Dialogue Translation in Indian Languages
BaatCheet：一个用于印度语言对话翻译的多语言语料库
Abstract
Existing translation models are typically trained on sentence-level and formal text, limiting their ability to capture everyday conversational dialogue phenomena such as informality, speaker interaction, and discourse coherence. Most existing Indic translation resources and evaluation benchmarks focus on sentence-level or formal text, making it difficult to assess translation quality of the dialogue phenomena. In this work, we introduce BaatCheet, a multilingual dialogue corpus named after the Hindi term for conversation or chitchat, containing approximately 49,000 dialogues for dialogue translation across five translation directions. We fine-tune five open-source LLMs across seven training data configurations and find that fine-tuning yields substantial gains over zero- and few-shot baselines. To comprehensively evaluate dialogue translation quality, we employ multiple evaluation strategies, including automatic metrics, LLM-as-judge, and human assessments using an SQM-guided Direct Assessment (DA) Protocol.
Chinese Translation
现有的翻译模型通常是在句子级别和正式文本上训练的，这限制了它们捕捉日常对话现象的能力，如非正式性、说话者互动和话语连贯性。大多数现有的印度语言翻译资源和评估基准集中于句子级别或正式文本，使得评估对话现象的翻译质量变得困难。在本研究中，我们介绍了BaatCheet，一个以印地语中“对话”或“闲聊”一词命名的多语言对话语料库，包含约49,000个对话，支持五个翻译方向的对话翻译。我们在七种训练数据配置下对五个开源大型语言模型（LLMs）进行了微调，发现微调在零样本和少样本基准上取得了显著的提升。为了全面评估对话翻译质量，我们采用了多种评估策略，包括自动指标、LLM作为评判者，以及使用SQM指导的直接评估（DA）协议进行的人类评估。
cs.CL / 94 / 2609.33301
Hesitation-Aware On-Policy Distillation for Diffusion Language Models
考虑犹豫的基于策略的扩散语言模型蒸馏
Abstract
Diffusion large language models (dLLMs) generate text by iterative unmasking. At each denoising step, a dLLM proposes a token at every masked position, but the decoder commits only a confident subset of these proposals. Trace-based on-policy distillation (TOPD) builds on this process by matching the student to a stronger teacher, yet only at the committed positions. We argue that this discards much of the useful signal, which resides in the uncommitted proposals, where the student has made a prediction but is not yet confident enough to commit it. We call these proposals hesitations. In our pilot study on an SDAR-4B student, hesitations make up only 24% of supervisable state-position pairs but carry 66% of the teacher-student divergence. To exploit this signal, we propose Hesitation-Aware On-Policy Distillation (HOPD), which extends teacher distribution matching to every masked position of each denoising step. Because hesitations are not equally informative, we further allocate supervision using hindsight from the completed trajectory, placing more weight on positions whose proposal was later disagreed with the final token and on blocks where first-step proposals rarely survive. Since both models already produce distributions at all masked positions, HOPD requires no additional forward passes over TOPD. The only extra cost is evaluating the loss at more positions. With SDAR-1.7B and SDAR-4B students distilled from TraDo-8B-Instruct, HOPD achieves the best average score among the evaluated methods on five math and coding benchmarks, under both static and dynamic decoding and at both scales. It also speeds up decoding. On SDAR-4B, the HOPD student hesitates less and commits 11% more tokens per denoising step than TOPD, while reaching higher accuracy.
Chinese Translation
扩散大型语言模型（dLLMs）通过迭代去噪生成文本。在每个去噪步骤中，dLLM在每个被屏蔽的位置提出一个标记，但解码器仅对这些提议中的自信子集进行承诺。基于轨迹的基于策略蒸馏（TOPD）在此过程中构建，通过将学生与更强的教师匹配，但仅限于已承诺的位置。我们认为，这会丢弃许多有用的信号，这些信号存在于未承诺的提议中，在这些位置，学生虽然做出了预测，但尚未自信到足以承诺。我们称这些提议为犹豫。在我们对SDAR-4B学生的初步研究中，犹豫仅占可监督状态-位置对的24%，但却占据了教师-学生差异的66%。为了利用这一信号，我们提出了考虑犹豫的基于策略蒸馏（HOPD），该方法将教师分布匹配扩展到每个去噪步骤的每个被屏蔽位置。由于犹豫并不具有相同的信息量，我们进一步利用完成轨迹的后见分配监督，对那些提议后来与最终标记不一致的位置和那些首步提议很少存活的块给予更多权重。由于两个模型在所有被屏蔽位置上已经生成分布，HOPD不需要在TOPD上进行额外的前向传递。唯一的额外成本是在更多位置评估损失。在从TraDo-8B-Instruct蒸馏的SDAR-1.7B和SDAR-4B学生中，HOPD在五个数学和编码基准测试中，在静态和动态解码下以及在两个规模下，获得了评估方法中的最佳平均分数。它还加快了解码速度。在SDAR-4B上，HOPD学生的犹豫更少，每个去噪步骤承诺的标记比TOPD多11%，同时达到更高的准确性。
cs.CL / 95 / 2609.33332
CertMark: Distortion-Free Multi-Bit Watermarking with Certified Decoding
CertMark：无失真多比特水印与认证解码
Abstract
Leading multi-bit watermarking methods for language models encode messages by biasing the model's next-token probabilities, creating a trade-off between message recovery and text quality. Their decoders typically return the highest-scoring candidate from accumulated token-level evidence, without a certified abstention rule that bounds the probability of outputting an incorrect message. We introduce CertMark, a distribution-preserving multi-bit watermark with certified decoding. Rather than modifying probabilities, CertMark uses the embedded message to seed an exact Gumbel-max sampler, thereby preserving the model's original sampling distribution. We propose two scalable decoders: a model-agnostic, text-only decoder and a model-aware variant that leverages the original next-token distributions for stronger recovery. Both support certified abstention with mathematical bounds on the probability of returning an incorrect message. Across text completion, summarization, and story generation, CertMark matches the perplexity of unwatermarked text while reliably recovering multi-bit messages. The model-aware decoder further achieves higher bit accuracy than probability-biasing baselines. Our code is publicly available at https://github.com/Batorskq/CertMark.
Chinese Translation
领先的多比特水印方法通过偏置语言模型的下一个标记概率来编码消息，从而在消息恢复和文本质量之间形成权衡。它们的解码器通常从累积的标记级证据中返回得分最高的候选项，而没有认证的放弃规则来限制输出错误消息的概率。我们提出了CertMark，一种保持分布的多比特水印，具有认证解码功能。CertMark不通过修改概率来实现，而是利用嵌入的消息来种子一个精确的Gumbel-max采样器，从而保持模型的原始采样分布。我们提出了两种可扩展的解码器：一种与模型无关的仅文本解码器和一种利用原始下一个标记分布以实现更强恢复能力的模型感知变体。两者都支持认证放弃，并对返回错误消息的概率设定了数学界限。在文本补全、摘要生成和故事创作中，CertMark的困惑度与未加水印的文本相匹配，同时可靠地恢复多比特消息。模型感知解码器进一步实现了比概率偏置基线更高的比特准确率。我们的代码已公开发布在 https://github.com/Batorskq/CertMark。
cs.CL / 96 / 2609.33335
Does Learning to Predict the World Help Agents Act? Auditing World-Model Post-Training
学习预测世界是否有助于智能体行动？后训练世界模型审计
Abstract
Predicting how an environment will change before acting is a natural route to better decision making for agents. Recent post-training methods therefore require agents to predict the next observation and turn that prediction into a reward or a direct supervision signal, which is called world model. Existing next-observation training methods help the agent to learn the environmental content. However, they additionally involve an optimization process, which may introduce several effects other than learning to predict the world. Consequently, where the performance gain comes from during the training process remains an open question. We answer this research question through replacing true next-observation targets with in-distribution mismatched observations during the training process. Across two interactive text environments, mismatched targets lower prediction accuracy by 15.3-61.6% relative to ground-truth targets, yet retain substantial task gains over the base model. Compared with the base model, trained models consider more candidate actions and exhibit less looping. We also introduce a setting that replaces prediction-based rewards with independent random signals. This training expands task coverage (pass@64) even when the reward carries no environment information. We also generalize this finding to VisualWebArena, where random-reward training raises pass@64 by 14.3% relative to the base model, without observation-matching rewards or an external multimodal teacher for reward construction.
Chinese Translation
在行动之前预测环境将如何变化是智能体更好决策的自然途径。因此，最近的后训练方法要求智能体预测下一个观察结果，并将该预测转化为奖励或直接监督信号，这被称为世界模型。现有的下一个观察训练方法帮助智能体学习环境内容。然而，它们还涉及一个优化过程，这可能引入除了学习预测世界之外的多种效果。因此，在训练过程中性能提升的来源仍然是一个未解的问题。我们通过在训练过程中用分布内不匹配的观察结果替代真实的下一个观察目标来回答这个研究问题。在两个交互式文本环境中，不匹配的目标使得预测准确性相对于真实目标降低了15.3%至61.6%，但在基线模型上仍保持了显著的任务收益。与基线模型相比，经过训练的模型考虑了更多的候选动作，并表现出更少的循环。我们还引入了一种设置，用独立的随机信号替代基于预测的奖励。即使奖励不包含环境信息，这种训练也扩大了任务覆盖率（pass@64）。我们还将这一发现推广到VisualWebArena，在该环境中，随机奖励训练使得pass@64相对于基线模型提高了14.3%，而没有观察匹配奖励或用于奖励构建的外部多模态教师。
cs.CL / 97 / 2609.33343
CHI: A Composite Hallucination Index Unifying Entity, Relation, and Quantity Dimensions for Summarization Evaluation
CHI：统一实体、关系和数量维度的复合幻觉指数用于摘要评估
Abstract
Faithfulness evaluation of abstractive summaries remains an open challenge, with existing metrics addressing only isolated hallucination types: factual entity errors, relational inconsistencies, or numerical fabrications, without capturing their co-occurrence or interaction. We introduce CHI (Composite Hallucination Index), the first unified hallucination metric that decomposes faithfulness errors into three orthogonal dimensions: entity hallucination (EHI), relation hallucination (RHI*), and quantity hallucination (QHI). Each dimension employs a shared softmax-normalized architecture over Venn diagram-derived factors representing extractiveness, positive hallucination, over-focus, negative hallucination, and lost focus. The novel QHI component introduces tolerance-aware numerical matching with exact, epsilon, derived, and temporal comparison modes. We fuse the three dimensions via harmonic mean to produce a single composite score that penalizes weakness in any dimension. We validate CHI on 800 source articles spanning four domains (news, medical, legal, financial) with summaries from five generation systems. Empirical results demonstrate that: (i) the three dimensions are statistically orthogonal (mean rho = 0.148), confirming they capture distinct error types; (ii) CHI achieves the highest system-level correlation with human judgments (rho = 0.66, p = 0.006) on SummEval, outperforming ROUGE (rho = 0.53), EHI (rho = 0.58), and all individual components; and (iii) ablation studies confirm that all three dimensions contribute unique variance, with the full composite outperforming any individual component while providing decomposable error diagnostics unavailable from single-score baselines. CHI provides practitioners with a decomposable, interpretable, and efficient faithfulness metric suitable for both offline evaluation and online monitoring of summarization systems.
Chinese Translation
抽象摘要的真实性评估仍然是一个未解决的挑战，现有指标仅针对孤立的幻觉类型：事实实体错误、关系不一致或数字虚构，而未能捕捉它们的共现或相互作用。我们提出了CHI（复合幻觉指数），这是第一个将真实性错误分解为三个正交维度的统一幻觉指标：实体幻觉（EHI）、关系幻觉（RHI*）和数量幻觉（QHI）。每个维度采用共享的softmax归一化架构，基于维恩图派生的因素，代表提取性、正幻觉、过度关注、负幻觉和失焦。新颖的QHI组件引入了容忍度感知的数值匹配，具有精确、epsilon、派生和时间比较模式。我们通过调和平均将这三个维度融合，生成一个单一的复合得分，惩罚任何维度的弱点。我们在涵盖四个领域（新闻、医学、法律、金融）的800篇源文章上验证了CHI，并使用五个生成系统的摘要。实证结果表明：（i）这三个维度在统计上是正交的（平均rho = 0.148），确认它们捕捉到不同的错误类型；（ii）CHI在SummEval上与人类判断的系统级相关性最高（rho = 0.66，p = 0.006），优于ROUGE（rho = 0.53）、EHI（rho = 0.58）和所有单独组件；（iii）消融研究确认这三个维度都贡献了独特的方差，完整的复合得分优于任何单一组件，同时提供了单一得分基线无法获得的可分解错误诊断。CHI为实践者提供了一种可分解、可解释且高效的真实性指标，适用于摘要系统的离线评估和在线监控。
cs.CL / 98 / 2609.33345
Language Discrimination Improves Linguistic Learning in Multilingual Speech Models
语言识别提升多语言语音模型的语言学习能力
Abstract
Multilingual self-supervised speech models can benefit from sharing information across languages, but under a matched total pretraining data budget they still fall short of monolingual models. We show that strengthening the model's ability to discriminate languages during pretraining reduces and, on some measures, closes this multilingual gap on continuous phonetic and higher-level linguistic measures, while preserving substantial cross-language sharing. Using a controlled English/French HuBERT setting, we test two interventions which strengthen language discrimination: an auxiliary language classifier and per-language k-means targets. Across interventions, continuous-feature phone discrimination error (phone-ABX, lower is better) decreases from 11.6% in the bilingual baseline to 10.4% (monolingual: 10.8%), while lexical performance (sWUGGY, higher is better) increases from 52.1% to 56.7% (monolingual: 58.5%) and prosodic performance (ProsAudit, lexical subtask, higher is better) from 68.9% to 72.9% (monolingual: 72.6%). Across HuBERT training stages, the strongest gains on most linguistic measures occur when language discrimination is introduced in the first iteration, whereas later or repeated interventions yield smaller improvements and are accompanied by increased language-wise segregation. These results support a causal role for language discrimination in reducing the additional cost of multilingual learning.
Chinese Translation
多语言自监督语音模型可以通过跨语言信息共享获益，但在匹配的预训练数据预算下，它们仍然无法超越单语模型。我们展示了在预训练过程中增强模型的语言识别能力可以减少并在某些指标上缩小这种多语言差距，尤其是在连续音位和更高层次的语言指标上，同时保持显著的跨语言共享。通过控制的英语/法语 HuBERT 设置，我们测试了两种增强语言识别的干预措施：辅助语言分类器和每种语言的 k-means 目标。在各项干预中，连续特征音位识别错误率（phone-ABX，越低越好）从双语基线的 11.6% 降低到 10.4%（单语：10.8%），而词汇表现（sWUGGY，越高越好）从 52.1% 提升至 56.7%（单语：58.5%），韵律表现（ProsAudit，词汇子任务，越高越好）从 68.9% 提升至 72.9%（单语：72.6%）。在 HuBERT 训练阶段，大多数语言指标的显著提升发生在首次迭代引入语言识别时，而后续或重复的干预则带来较小的改善，并伴随语言间的隔离增加。这些结果支持语言识别在减少多语言学习额外成本中的因果作用。
cs.CL / 99 / 2609.33379
NLPG: Natural-Language Policy Gradients for Self-Evolving Language Agents
NLPG：用于自我演化语言代理的自然语言策略梯度
Abstract
Large language model agents increasingly rely on compound programs for retrieval, tool use, reasoning, and verification, yet their failures often arise from local procedural decisions. Existing reinforcement-learning and prompt-optimization approaches typically rely on scalar rewards or repeatedly modify entire prompts, making it difficult to capture and reuse procedural improvements while preserving a frozen agent. To address this problem, We propose Natural-Language Policy Gradients (NLPG), an external policy-memory method for improving a fixed agent without changing its model parameters or program structure. NLPG diagnoses execution traces, propagates downstream feedback backward through the module graph, and converts recurring failures into route-local natural-language corrections that are aggregated into bounded policy updates for subsequent executions. Across six benchmarks covering memory, reasoning, instruction following, and evidence verification, NLPG also outperforms the strongest listed baseline for each benchmark by 8.71 percentage points on average. These results provide evidence that evaluated procedural experience can be transformed into local and interpretable policy updates, enabling continual improvement of frozen agents.
Chinese Translation
大型语言模型代理越来越依赖复合程序进行检索、工具使用、推理和验证，但它们的失败往往源于局部程序决策。现有的强化学习和提示优化方法通常依赖标量奖励或反复修改整个提示，这使得在保持代理不变的情况下捕捉和重用程序改进变得困难。为了解决这个问题，我们提出了自然语言策略梯度（Natural-Language Policy Gradients, NLPG），这是一种外部策略记忆方法，可以在不改变模型参数或程序结构的情况下改善固定代理。NLPG 诊断执行轨迹，通过模块图向后传播下游反馈，并将重复的失败转化为局部自然语言修正，这些修正被聚合为后续执行的有界策略更新。在涵盖记忆、推理、指令遵循和证据验证的六个基准测试中，NLPG 的表现平均比每个基准测试中最强的列出基线高出 8.71 个百分点。这些结果提供了证据，表明评估的程序经验可以转化为局部和可解释的策略更新，从而实现对冻结代理的持续改进。
cs.CL / 100 / 2609.33390
From Position Risks to Block Survival: Faster Generation for Diffusion Language Models
从位置风险到区块生存：扩散语言模型的快速生成
Abstract
Diffusion language models (DLMs) can accelerate generation by predicting multiple tokens in parallel, but there is a mismatch between how these tokens are predicted and how they ultimately contribute to generation. Parallel predictions can hardly condition on the tokens selected earlier within the same block, even though their validity depends on this realized prefix. Under the popular proposal-verification decoding, this mismatch makes errors highly asymmetric: an early rejection prevents all subsequent proposals from contributing decoding progress. We introduce BRISK-DLM, a framework that addresses both mismatches by optimizing proposal learning and selection for verified progress. BRISK-DLM trains on self-generated sequences, using risk-reward weighting to dynamically prioritize positions by their impact on verified progress and decoding cost. During inference, a lightweight prefix-conditioned corrector reranks existing candidates using previously selected tokens and preferences distilled from the model's own verifier. The corrector reuses the backbone's parallel representations and requires no additional backbone evaluation, while fused execution keeps its overhead small. BRISK-DLM improves end-to-end throughput by up to 37.4% while preserving task quality, establishing a new quality-throughput frontier for DLM generation.
Chinese Translation
扩散语言模型（DLMs）通过并行预测多个标记来加速生成，但这些标记的预测方式与它们最终对生成的贡献之间存在不匹配。并行预测几乎无法对同一区块内早先选择的标记进行条件化，尽管它们的有效性依赖于这一实现的前缀。在流行的提案-验证解码方法下，这种不匹配使得错误高度不对称：早期的拒绝会阻止所有后续提案对解码进展的贡献。我们提出了BRISK-DLM，一个通过优化提案学习和选择以实现验证进展来解决这两种不匹配的框架。BRISK-DLM在自生成序列上进行训练，使用风险-奖励加权动态优先考虑位置，以其对验证进展和解码成本的影响为依据。在推理过程中，一个轻量级的前缀条件校正器使用先前选择的标记和从模型自身验证器中提炼的偏好对现有候选进行重新排序。该校正器重用主干网络的并行表示，并且不需要额外的主干评估，而融合执行保持了其开销较小。BRISK-DLM在保持任务质量的同时，将端到端吞吐量提高了多达37.4%，为DLM生成建立了新的质量-吞吐量前沿。
cs.CL / 101 / 2609.33395
Preserving Morphemes: Morphology-Guided Pre-Tokenization for Nepali
保留语素：针对尼泊尔语的形态学引导预标记化
Abstract
A byte-level BPE vocabulary learns each inflected form of a Nepali word as a separate string, so a noun stem is spelled differently in each of its case-marked forms. We test whether splitting words into stem and affixes before BPE helps, with the corpus, vocabulary size, model and number of training steps held fixed. Our pre-tokenizer, Papaya, uses a finite-state transducer built from a published grammar of Nepali, falls back to regular expressions, and leaves the BPE trainer unchanged. On 607 words annotated by seven native speakers its segmenter reaches 0.96 boundary F1, and the resulting tokens keep stems intact far more often than plain BPE does. In a 17M-parameter language model it lowers bits per byte by about 1% at equal training steps; most of the larger gain seen at equal epochs comes from the extra steps that longer token sequences buy, and an unsupervised Morfessor segmentation gives the same improvement. Downstream the effect is small: NER improves only on entities that contain words unseen in training, POS tagging and news classification do not change, and published Nepali tokenizers perform about as well. We release the annotated boundary set, a 556-affix dataset and the code.
Chinese Translation
字节级BPE词汇将尼泊尔语单词的每个屈折形式视为一个独立的字符串，因此名词词干在每个带有格标记的形式中拼写不同。我们测试在BPE之前将单词拆分为词干和词缀是否有帮助，同时保持语料库、词汇大小、模型和训练步骤数量不变。我们的预标记器Papaya使用基于已发布的尼泊尔语语法构建的有限状态传输器，回退到正则表达式，并保持BPE训练器不变。在607个由七位母语者标注的单词中，其分段器达到了0.96的边界F1，所生成的标记比普通BPE更频繁地保持了词干的完整性。在一个1700万参数的语言模型中，它在相同的训练步骤下将每字节位数降低了约1%；在相同的训练周期下，看到的更大增益大部分来自于更长的标记序列所带来的额外步骤，而无监督的Morfessor分段也带来了相同的改善。在下游任务中，效果较小：命名实体识别（NER）仅在包含训练中未见单词的实体上有所改善，词性标注（POS）和新闻分类没有变化，而已发布的尼泊尔语标记器表现大致相同。我们发布了标注的边界集、一个包含556个词缀的数据集和代码。
cs.CL / 102 / 2609.33409
Dense Is Not Enough: Hierarchical Supervision Allocation for Long-Horizon On-Policy Distillation
密集监督不足：长时间跨度在线策略蒸馏的分层监督分配
Abstract
On-policy distillation (OPD) transfers the capabilities of a large language model to a smaller student by providing teacher supervision on the student's own rollouts. In long-horizon agentic tasks, however, uniform token-level matching can allocate supervision poorly: a large local discrepancy need not improve future behavior, while consequential guidance may be beyond the current student's reach or fail to persist without privileged input. We formulate long-horizon OPD as hierarchical supervision allocation and argue that productive guidance lies at the intersection of future utility and current learnability. Crucially, this intersection evolves as the student learns. Based on this principle, we propose LENS-OPD, a coarse-to-fine framework that organizes supervision through Locate, Validate, and Refine. Locate adapts trajectory exposure to the student's evolving competence and proposes a candidate decision for intervention. Validate tests whether teacher guidance at that decision improves the same student's subsequent behavior. Refine internalizes the beneficial guided behavior into the deployable policy and concentrates token-level supervision on decisive teacher-student conflicts within the validated turn. These stages are nested: each finer allocation is conditioned on the coarser decision, rather than being optimized as an independent importance score. Experiments across multiple long-horizon agent benchmarks and student-teacher configurations show that LENS-OPD consistently improves task performance over vanilla OPD and strong curriculum- and selection-based baselines. Our results suggest that effective long-horizon distillation requires teaching at the right depth, the right decision, and the right token.
Chinese Translation
在线策略蒸馏（OPD）通过在学生自身的回合中提供教师监督，将大型语言模型的能力转移到较小的学生身上。然而，在长时间跨度的智能任务中，均匀的令牌级匹配可能导致监督分配不当：较大的局部差异未必能改善未来的行为，而重要的指导可能超出当前学生的能力范围，或者在没有特权输入的情况下无法持续。我们将长时间跨度的OPD形式化为分层监督分配，并认为有效的指导位于未来效用与当前可学习性之间的交集。关键是，这一交集随着学生的学习而演变。基于这一原则，我们提出了LENS-OPD，一个通过定位（Locate）、验证（Validate）和精炼（Refine）组织监督的粗到细框架。定位根据学生不断发展的能力调整轨迹暴露，并提出干预的候选决策。验证测试在该决策下的教师指导是否改善了同一学生后续的行为。精炼将有益的指导行为内化为可部署的策略，并将令牌级监督集中在经过验证的回合内决定性的教师-学生冲突上。这些阶段是嵌套的：每个更细的分配都依赖于更粗的决策，而不是作为独立的重要性评分进行优化。在多个长时间跨度的智能基准和学生-教师配置上的实验表明，LENS-OPD在任务表现上始终优于普通OPD以及强大的课程和选择基础线。我们的结果表明，有效的长时间跨度蒸馏需要在正确的深度、正确的决策和正确的令牌上进行教学。
cs.CL / 103 / 2609.33417
Grounding Memory Summarization in Utility Intent
基于效用意图的记忆摘要
Abstract
Existing summarizers for memory systems are typically optimized for human-facing criteria such as faithfulness, which misaligns with their true objective: preserving the evidence needed to support future queries. We show that conditioning summarization on query-answer pairs substantially improves answer quality, and that this utility-aware behavior is transferable across queries. Motivated by these findings, we propose MemSuit, a self-distillation framework in which a teacher summarizer, conditioned on observed query-answer pairs, produces utility-aware memory entries that a student learns to reproduce from the raw conversation alone. To prevent collateral erasure where conditioning on a single query-answer pair discards evidence relevant to other plausible queries, the teacher decomposes each block into multiple self-contained entries that preserve distinct query-relevant facets as independently retrievable units. To align the retriever with the compact, fact-dense style of teacher entries, we further fine-tune the embedding model with a contrastive objective supervised by teacher entries. Across a diverse suite of conversational query types, MemSuit consistently outperforms state-of-the-art baselines, confirming the value of grounding memory in downstream utility.
Chinese Translation
现有的记忆系统摘要工具通常针对人类可接受的标准进行优化，例如忠实性，这与它们的真实目标不一致：保留支持未来查询所需的证据。我们展示了基于查询-答案对进行摘要处理显著提高了答案质量，并且这种关注效用的行为可以在不同查询之间转移。基于这些发现，我们提出了MemSuit，一个自我蒸馏框架，其中教师摘要器基于观察到的查询-答案对生成关注效用的记忆条目，而学生则从原始对话中学习重现这些条目。为了防止因基于单一查询-答案对的条件化而丢弃与其他合理查询相关的证据，教师将每个块分解为多个自包含的条目，这些条目保留了独立可检索的不同查询相关方面。为了使检索器与教师条目的紧凑、事实密集风格对齐，我们进一步使用教师条目监督的对比目标对嵌入模型进行微调。在多种对话查询类型的测试中，MemSuit始终优于最先进的基线，确认了将记忆与下游效用相结合的价值。
cs.CL / 104 / 2609.33426
TeacherGRPO: Closing the Capacity Gap in Reasoning Distillation via Teacher Alignment
TeacherGRPO：通过教师对齐缩小推理蒸馏中的能力差距
Abstract
Reasoning distillation from powerful teacher models to smaller students faces the Gap Curse: as teachers grow more sophisticated, their complex distributions increasingly diverge from what students can approximate, causing performance degradation. Existing mitigation strategies either filter out challenging examples through data selection or introduce weaker intermediate assistant models, inherently compromising supervision coverage or quality. We propose Teacher Alignment, which directly adapts the teacher toward the student's distribution without discarding data or degrading reasoning quality. However, naive alignment through standard knowledge distillation triggers catastrophic collapse of the teacher's reasoning capabilities. To address this, we reformulate teacher alignment as reinforcement learning and introduce TeacherGRPO, built on Group Relative Policy Optimization with two key innovations: (i) Curriculum Selective Alignment applies dual token- and distribution-level curricula to focus rewards on high-signal reasoning gaps while filtering noise from trivial tokens and uncertain tail distributions, and (ii) Importance-Adaptive Length Regularization selectively penalizes verbose redundancy while preserving pedagogically critical reasoning steps. The aligned teacher then distills knowledge to students via standard pipelines. Extensive experiments show TeacherGRPO significantly outperforms baselines across diverse reasoning benchmarks and distillation methods. Our code is available at https://github.com/LzyFischer/TeacherGRPO.
Chinese Translation
从强大的教师模型到较小学生的推理蒸馏面临着能力差距的诅咒：随着教师模型的复杂性增加，它们的复杂分布与学生能够近似的分布越来越偏离，从而导致性能下降。现有的缓解策略要么通过数据选择过滤掉具有挑战性的示例，要么引入较弱的中间辅助模型，固有地妥协了监督的覆盖范围或质量。我们提出了教师对齐（Teacher Alignment），该方法直接调整教师模型以适应学生的分布，而不丢弃数据或降低推理质量。然而，通过标准知识蒸馏进行的简单对齐会导致教师推理能力的灾难性崩溃。为了解决这个问题，我们将教师对齐重新构建为强化学习，并引入了TeacherGRPO，该方法基于组相对策略优化（Group Relative Policy Optimization），具有两个关键创新：（i）课程选择性对齐（Curriculum Selective Alignment）应用双重的标记和分布级课程，将奖励集中在高信号推理差距上，同时过滤掉来自琐碎标记和不确定尾部分布的噪声；（ii）重要性自适应长度正则化（Importance-Adaptive Length Regularization）选择性地惩罚冗长的冗余，同时保留教学上关键的推理步骤。然后，经过对齐的教师通过标准管道将知识蒸馏给学生。大量实验表明，TeacherGRPO在各种推理基准和蒸馏方法上显著优于基线。我们的代码可在 https://github.com/LzyFischer/TeacherGRPO 获取。
cs.CL / 105 / 2609.33441
MIC: Explaining Image-Claim Inconsistencies in AI-Generated Multimodal Misinformation
MIC：解释AI生成的多模态虚假信息中的图像-声明不一致性
Abstract
Claims paired with AI-generated images are a rapidly growing form of misinformation. Existing automated fact-checking (AFC) methods mainly treat this as a provenance problem, detecting low-level synthesis artifacts to decide whether an image is AI-generated. However, such methods do not verify what human fact-checkers often check: whether an image's content is consistent with the context implied by its accompanying claim. To address this gap, we introduce MIC (Multimodal Inconsistency Checking), an AFC framework that assists human fact-checkers by detecting AI-generated multimodal misinformation and explaining inconsistencies using world knowledge. MIC first uses supervised fine-tuning (SFT) for task adaptation and then applies Group Relative Policy Optimization (GRPO) to directly optimize component-level verifiable rewards for verdict prediction, inconsistency type classification, visual evidence description, and world-knowledge explanation. We further introduce MIC-Bench, a benchmark comprising 8,812 image-claim instances derived from 4,406 claims, where each claim is paired with an authentic image and an AI-generated counterpart that introduces a controlled contextual inconsistency. Compared with SFT alone, GRPO further improves Macro-F1 by 4.67 and 4.11 points in the in-distribution and out-of-distribution settings, respectively, while also improving the semantic similarity of visual evidence descriptions and world-knowledge explanations to reference annotations. Our code and data are available at https://github.com/UKPLab/arxiv2026-mic.
Chinese Translation
与AI生成的图像配对的声明是一种快速增长的虚假信息形式。现有的自动化事实检查（AFC）方法主要将其视为来源问题，通过检测低级合成伪影来判断图像是否为AI生成。然而，这些方法并未验证人类事实检查者通常检查的内容：图像的内容是否与其伴随声明所暗示的上下文一致。为了解决这一问题，我们提出了MIC（多模态不一致性检查），这是一个AFC框架，通过检测AI生成的多模态虚假信息并利用世界知识解释不一致性来辅助人类事实检查者。MIC首先使用监督微调（SFT）进行任务适应，然后应用组相对策略优化（GRPO）直接优化组件级可验证奖励，以进行裁决预测、不一致性类型分类、视觉证据描述和世界知识解释。我们进一步推出了MIC-Bench，这是一个基准数据集，包含从4,406个声明中衍生的8,812个图像-声明实例，其中每个声明都配有一张真实图像和一张引入受控上下文不一致的AI生成图像。与单独使用SFT相比，GRPO在分布内和分布外设置中分别进一步提高了4.67和4.11的宏F1分数，同时也改善了视觉证据描述和世界知识解释与参考注释之间的语义相似性。我们的代码和数据可在https://github.com/UKPLab/arxiv2026-mic获取。
cs.CL / 106 / 2609.33443
Context Spanning: A Communication Framework for Full-Duplex Speech Models and External LLM Backends
上下文跨越：全双工语音模型与外部大型语言模型后端的通信框架
Abstract
Full-duplex spoken dialogue models can listen and speak simultaneously like the real-time dynamics of human conversation. For natural dialogue, the ability to search for external information in real-time is also an important capability. Many models remain trapped in parametric knowledge, leaving them unable to access real-time information and tool execution. Furthermore, even when Large Language Models (LLM) retrieve information, many duplex speech models process it within a compressed latent space rather than in its raw text form, which can lead to information loss from compression. To address this issue, we propose Context Spanning, a framework for information injection between a full-duplex speech model and an external LLM backend via real-time chunked prefill. The injected frame is encoded in a single forward pass inside the real-time frame budget. It feeds the retrieved information to the speech model as-is, enabling it to reason over the information independently and generate responses. With this approach, our model achieves high performance on Full-Duplex benchmarks and strong results on Question Answering tasks, demonstrating its conversation potential. Context Spanning shows that external information can be injected directly into a duplex speech model, introducing a new simple and powerful mechanism for duplex systems.
Chinese Translation
全双工口语对话模型能够像人类对话的实时动态一样同时进行听和说。对于自然对话而言，实时搜索外部信息的能力也是一项重要的能力。许多模型仍然被困于参数化知识中，无法访问实时信息和工具执行。此外，即使大型语言模型（LLM）检索到信息，许多双工语音模型也会在压缩的潜在空间中处理这些信息，而不是以原始文本形式处理，这可能导致信息因压缩而丢失。为了解决这个问题，我们提出了上下文跨越（Context Spanning），一个通过实时分块预填充在全双工语音模型与外部LLM后端之间进行信息注入的框架。注入的框架在实时框架预算内通过单次前向传递进行编码。它将检索到的信息原封不动地输入到语音模型中，使其能够独立地对信息进行推理并生成响应。通过这种方法，我们的模型在全双工基准测试中取得了高性能，并在问答任务中表现出色，展示了其对话潜力。上下文跨越表明，外部信息可以直接注入到双工语音模型中，为双工系统引入了一种新的简单而强大的机制。
cs.CL / 107 / 2609.33463
Rethinking Token Reweighting for SFT: Suppress, Reverse, and Extrapolate Learned Features
重新思考SFT中的令牌重加权：抑制、反转和外推学习特征
Abstract
Supervised fine-tuning (SFT) learns most aggressively from tokens that the model deems least likely. This helps acquire new behaviors, but also amplifies noisy or conflicting supervision and can overwrite useful pretrained knowledge. Through a unified policy-loss view, we revisit existing token-reweighting methods and show that they assign nonnegative coefficients to demonstrated tokens. Consequently, they can suppress or amplify supervised updates, but cannot reverse harmful features once learned. Moreover, larger training weights do not amount to feature extrapolation, since they change the optimization trajectory rather than scale a fixed SFT direction. We argue that reversal and extrapolation require a stable reference frame defined by a fixed SFT delta. Motivated by this, we propose SCALE (Selective Control of Adaptation via Local Entropy), an entropy-guided adaptation-strength-control method that freezes the pretrained model and the SFT delta and learns bounded token- and module-specific gates by minimizing predictive entropy alone. These gates suppress, reverse, or extrapolate frozen SFT features according to their alignment with entropy reduction. Across Qwen2.5-Math-1.5B, Qwen2.5-Math-7B, and Qwen3-4B-Base, SCALE achieves mathematical-reasoning averages of 37.84, 43.60, and 36.57, exceeding the strongest corresponding baselines while remaining competitive on general-retention benchmarks. It also attains the best average code-generation performance across HumanEval, HumanEval+, and MBPP for all three models. These results suggest that effective SFT correction can benefit from controlling how already learned residuals are used, rather than only modifying how they are learned.
Chinese Translation
监督微调（SFT）最积极地从模型认为最不可能的令牌中学习。这有助于获得新行为，但也会放大噪声或冲突的监督，并可能覆盖有用的预训练知识。通过统一的策略损失视角，我们重新审视现有的令牌重加权方法，并展示它们为已展示的令牌分配非负系数。因此，它们可以抑制或放大监督更新，但无法反转一旦学习到的有害特征。此外，更大的训练权重并不等同于特征外推，因为它们改变了优化轨迹，而不是缩放固定的SFT方向。我们认为，反转和外推需要由固定的SFT增量定义的稳定参考框架。基于此，我们提出了SCALE（通过局部熵的选择性适应控制），这是一种熵引导的适应强度控制方法，它冻结预训练模型和SFT增量，并通过仅最小化预测熵来学习有界的令牌和模块特定门。这些门根据与熵减少的对齐程度抑制、反转或外推冻结的SFT特征。在Qwen2.5-Math-1.5B、Qwen2.5-Math-7B和Qwen3-4B-Base上，SCALE实现了37.84、43.60和36.57的数学推理平均值，超越了最强的相应基线，同时在一般保留基准上保持竞争力。它还在HumanEval、HumanEval+和MBPP上实现了所有三个模型的最佳平均代码生成性能。这些结果表明，有效的SFT修正可以受益于控制已学习残差的使用方式，而不仅仅是修改它们的学习方式。
cs.CL / 108 / 2609.33465
GSM: Efficient Language Modeling with Shared Global State
GSM：具有共享全局状态的高效语言建模
Abstract
Efficient language models must reduce not only the cost of individual accesses to past context but also the overhead of repeatedly selecting and processing historical information across layers. We introduce the Global State Model (GSM), a causal encoder--decoder architecture that concentrates the selection and aggregation of long-range information in the encoding stage. Through multiple stages of history retrieval, the encoder progressively incorporates long-range information into representations at recent positions, forming a shared state with a fixed window size. Each decoder layer accesses this same state using queries updated from the preceding layer, preserving computational depth while avoiding repeated construction of historical key--value (KV) representations and long-range indexing. As a result, neither the decoder's per-step attention cost nor its KV cache size grows with the history length. Experiments show that GSM improves computational efficiency and reduces cache overhead while maintaining model performance and the ability to use long-range information, offering a shared-state architecture for efficient language modeling.
Chinese Translation
高效的语言模型不仅必须减少对过去上下文的单次访问成本，还要降低在各层之间反复选择和处理历史信息的开销。我们提出了全局状态模型（Global State Model, GSM），这是一种因果编码器-解码器架构，集中在编码阶段选择和聚合长程信息。通过多个历史检索阶段，编码器逐步将长程信息纳入最近位置的表示中，形成一个固定窗口大小的共享状态。每个解码器层使用来自前一层更新的查询访问这个相同的状态，从而保持计算深度，同时避免重复构建历史键值（key-value, KV）表示和长程索引。因此，解码器每步的注意力成本和KV缓存大小都不会随着历史长度的增加而增长。实验表明，GSM提高了计算效率，减少了缓存开销，同时保持了模型性能和使用长程信息的能力，为高效语言建模提供了一种共享状态架构。
cs.CL / 109 / 2609.33485
DISCO: Distributed Long Context Scaling with Grounding-Reasoning Disaggregation
DISCO：基于基础推理分解的分布式长上下文扩展
Abstract
While Large Language Models (LLMs) advertise million-token context windows, reasoning quality often collapses as inputs grow -- a phenomenon termed context rot. This failure stems from a structural entanglement in monolithic architectures, where the massive search burden of contextual grounding exhausts the representational capacity needed for complex reasoning. To resolve this, we propose Grounding-Reasoning Disaggregation via DIStributed long COntext scaling (DISCO). Inspired by distributed computing frameworks like Apache Spark, DISCO partitions long context across a fleet of Worker LLMs dedicated exclusively to parallel, localized grounding. A central Driver LLM, trained via Reinforcement Learning (GRPO) to optimize planning, orchestrates execution by dynamically mapping queries into atomic extraction tasks and reducing the gathered evidence to synthesize a final answer. By isolating reasoning from raw context noise, DISCO effectively eliminates context rot. On RULER-QA (1M tokens), it maintains 78.4% accuracy where standard baselines collapse. Furthermore, it outperforms full-context models by up to 9.8 points on LongBench v2 and matches frontier models like Gemini-3-Pro-Preview while reducing inference costs by over 80%, establishing a highly efficient paradigm for robust long-context inference.
Chinese Translation
尽管大型语言模型（LLMs）宣传具有百万令牌的上下文窗口，但随着输入的增加，推理质量往往会下降——这一现象被称为上下文衰退。这一失败源于单体架构中的结构性纠缠，在这种架构中，上下文基础的巨大搜索负担耗尽了进行复杂推理所需的表征能力。为了解决这一问题，我们提出了通过分布式长上下文扩展（DISCO）实现基础推理分解。DISCO受到像Apache Spark这样的分布式计算框架的启发，将长上下文划分到专门用于并行、本地化基础的多个工作LLM中。一个中央驱动LLM通过强化学习（GRPO）进行训练，以优化规划，动态地将查询映射为原子提取任务，并将收集到的证据减少到合成最终答案。通过将推理与原始上下文噪声隔离，DISCO有效消除了上下文衰退。在RULER-QA（100万令牌）上，它保持了78.4%的准确率，而标准基线则崩溃。此外，它在LongBench v2上比全上下文模型高出多达9.8个百分点，并且与前沿模型如Gemini-3-Pro-Preview相匹配，同时将推理成本降低了超过80%，建立了一种高效的强健长上下文推理范式。
cs.CL / 110 / 2609.33495
LLMs Trust Their Own: Identity-Dependent Conformity in Multi-Agent Systems
大型语言模型信任自己：多智能体系统中的身份依赖性从众行为
Abstract
Large language models (LLMs) are increasingly deployed in multi-agent settings, where agents observe and influence one another, making social influence a key dimension of AI behavior and safety. We investigate whether LLMs' responses depend on the social identity of other agents, beyond the effect of their consensus. We construct judgment tasks with a single correct answer, and place models in a multi-agent setting where they receive incorrect answers from other agents whose social identities (AI or human, model family, or an arbitrary minimal group) are either shared with or distinct from their own. Across 12 open-weights models and nine tasks, we find a bidirectional effect of group identity on conformity to incorrect answers: in-group consensus increases conformity (in-group favoritism), whereas out-group consensus decreases it (out-group divergence). Unlike humans, for whom one ally breaking the consensus sharply reduces conformity, models are unmoved by an ally from the majority's group. Worse, a correct ally from the opposing group intensifies this bidirectional effect. Chain-of-Thought reasoning suppresses most of these effects, yet an in-group ally still reduces conformity to an incorrect out-group majority. Labeling peers as safety-aligned shifts overall conformity but leaves in-group favoritism and out-group divergence intact. These results show that group identity shapes how LLMs aggregate information across agents, independently of its correctness, and identify a manipulation surface for multi-agent AI systems.
Chinese Translation
大型语言模型（LLMs）越来越多地应用于多智能体环境中，在这些环境中，智能体相互观察并影响彼此，使得社会影响成为人工智能行为和安全性的重要维度。我们研究了LLMs的响应是否依赖于其他智能体的社会身份，超越其共识的影响。我们构建了具有单一正确答案的判断任务，并将模型置于多智能体环境中，在该环境中，它们接收到来自其他智能体的错误答案，这些智能体的社会身份（人工智能或人类、模型家族或任意最小群体）与它们自身的身份要么相同，要么不同。在12个开放权重模型和九个任务中，我们发现群体身份对错误答案的从众行为具有双向影响：内部群体的共识增加从众行为（内部群体偏好），而外部群体的共识则减少从众行为（外部群体分歧）。与人类不同，对于人类而言，一个盟友打破共识会显著减少从众行为，而模型对来自多数群体的盟友并不受影响。更糟糕的是，来自对立群体的正确盟友会加剧这种双向效应。链式思维推理抑制了大多数这些效应，但内部群体的盟友仍然会减少对错误外部群体多数的从众行为。将同伴标记为安全对齐会改变整体从众行为，但仍然保持内部群体偏好和外部群体分歧不变。这些结果表明，群体身份塑造了LLMs如何在智能体之间聚合信息，而不论其正确性，并识别出多智能体人工智能系统的操控表面。
cs.CL / 111 / 2609.33530
E-CONAN (Entailment, CONtradition And Neutral) Diagnostics Dataset Investigating Linguistic Phenomena in Arabic Natural Language Understanding
E-CONAN（蕴含、矛盾与中立）诊断数据集：探讨阿拉伯自然语言理解中的语言现象
Abstract
Natural Language Understanding (NLU) plays a crucial role in various applications, yet its performance suffers from weaknesses in handling the complexities of human languages, ranging from lexical ambiguity to high-level reasoning difficulties. Analyzing errors across diverse linguistic phenomena is crucial for NLU improvement, as it will help humans get insights to comprehensively assess models' limitations and capabilities, so optimizing models' generalization. Notably, several benchmarks contain diagnostics datasets designed for investigation and fine-grained error analysis. When highlighting the gaps in the state-of-the-art, we noted that there is no naming convention for macro and micro categories or even a standard set of linguistic phenomena that should be covered. To overcome this gap, we propose an initial hierarchy for Cross-Lingual NLU error analysis. Moreover, we propose a methodology to create an NLI hierarchical framework and applied a case study on Arabic NLU. Moreover, this paper introduces E-CONAN diagnostics dataset, a freely available dataset manually-annotated with coarse-grained and fine-grained categories based on our proposed Arabic hierarchy. E-CONAN dataset helps NLU designers better understand their models by doing error analysis and in-depth investigation. We used E-CONAN to investigate the performance of 9 pretrained language models and 5 LLMs. Results indicate that LLMs outperform pretrained models in world knowledge and commonsense reasoning macro-category, and underperform pretrained models in syntactic macro-category. Moreover, the hardest phenomena for all models is Reasoning, and the easiest phenomena for all pretrained models is Syntactic, and the easiest for LLMs is Lexico-Syntactic.
Chinese Translation
自然语言理解（NLU）在各种应用中发挥着至关重要的作用，但其性能受到处理人类语言复杂性（从词汇歧义到高层次推理困难）的弱点影响。分析不同语言现象中的错误对于改善NLU至关重要，因为这将帮助人们深入了解模型的局限性和能力，从而优化模型的泛化能力。值得注意的是，多个基准测试包含旨在进行调查和细致错误分析的诊断数据集。在强调当前技术的不足时，我们注意到缺乏宏观和微观类别的命名约定，甚至缺乏应涵盖的标准语言现象集。为了解决这一问题，我们提出了跨语言NLU错误分析的初步层次结构。此外，我们提出了一种创建NLI（自然语言推理）层次框架的方法，并对阿拉伯NLU进行了案例研究。此外，本文介绍了E-CONAN诊断数据集，这是一个基于我们提出的阿拉伯层次结构手动标注的自由可用数据集，包含粗粒度和细粒度类别。E-CONAN数据集帮助NLU设计者通过错误分析和深入调查更好地理解他们的模型。我们使用E-CONAN调查了9个预训练语言模型和5个大型语言模型（LLMs）的性能。结果表明，LLMs在世界知识和常识推理的宏观类别中优于预训练模型，而在句法宏观类别中表现不如预训练模型。此外，所有模型面临的最难现象是推理，而所有预训练模型最容易处理的现象是句法，LLMs最容易处理的现象是词汇-句法。
cs.CL / 112 / 2609.33534
ManiEdit: Sequential Unstructured Knowledge Editing for Language Models from a Manifold Perspective
ManiEdit：从流形视角进行语言模型的顺序非结构化知识编辑
Abstract
Large language models (LLMs) inevitably generate some incorrect or outdated content, necessitating efficient and precise mechanisms for continual knowledge updates. However, existing model editing methods struggle to sequentially edit unstructured long-form knowledge, suffering from severe edit forgetting and degradation of general capabilities. To address these challenges, we reframe knowledge editing from a manifold perspective, viewing it as a localized displacement of an edit sub-manifold within the global knowledge manifold. Under this formulation, the problem can be decomposed into two key questions: (i) how to identify representative edit points that effectively anchor the edit sub-manifold, and (ii) how to preserve the remaining manifold structure during the sub-manifold displacement process. Based on this perspective, we propose ManiEdit, a novel manifold-aware autoregressive editing framework consisting of two core components. Pivot Localization addresses the mediocre-point dilemma by identifying high-leverage pivots to anchor the edit sub-manifold. Manifold-Aware Preservation preserves different knowledge types through an energy-weighted penalty combined with recursive null-space alignment. Experiments on two base LLMs and four unstructured editing benchmarks demonstrate that ManiEdit achieves state-of-the-art performance, outperforming the strongest baseline by up to +27.81 BERTScore and +8.50 ROUGE-L, while maintaining near-original general capabilities across six representative downstream tasks. Our code is available at: https://github.com/Areyliu/ManiEdit
Chinese Translation
大型语言模型（LLMs）不可避免地会生成一些不正确或过时的内容，因此需要高效而精确的机制来进行持续的知识更新。然而，现有的模型编辑方法在顺序编辑非结构化长篇知识时面临困难，严重的编辑遗忘和通用能力的下降是其主要问题。为了解决这些挑战，我们从流形的角度重新构建知识编辑，将其视为在全球知识流形内对编辑子流形的局部位移。在这一框架下，问题可以分解为两个关键问题：（i）如何识别有效锚定编辑子流形的代表性编辑点，以及（ii）如何在子流形位移过程中保持剩余流形结构。基于这一视角，我们提出了ManiEdit，一种新颖的流形感知自回归编辑框架，由两个核心组件组成。Pivot Localization通过识别高杠杆点来解决中等点困境，以锚定编辑子流形。Manifold-Aware Preservation通过结合能量加权惩罚和递归零空间对齐来保留不同类型的知识。在两个基础LLM和四个非结构化编辑基准上的实验表明，ManiEdit实现了最先进的性能，超越最强基线达+27.81 BERTScore和+8.50 ROUGE-L，同时在六个代表性下游任务中保持接近原始的通用能力。我们的代码可在以下链接获取：https://github.com/Areyliu/ManiEdit
cs.CL / 113 / 2609.33538
Jev Matches 7B Language Models for Speech-Neuroprosthesis Rescoring
Jev 在语音神经假体重评分中匹配 7B 语言模型
Abstract
A speech neuroprosthesis decodes attempted speech from brain activity and ends by rescoring the decoder's candidate sentences with a language model of several billion parameters, the only component that needs a GPU. Replacing that model with a cheaper one is hard: general language models asked to pick one sentence from a list answer from where a label sits in the list rather than from the sentence itself. We pose rescoring as a single typed decision, one call that returns a probability for every candidate, served by Jev, a hosted model trained for calibrated decisions, and combine it with the decoder's own score. On 978 held-out sentences from a participant with ALS, where the published decoder alone reaches 8.1% word error, Jev reaches 7.5% against 7.8% for both OPT-6.7b and Qwen2.5-7B; with the decoder's weight re-tuned, 6.9% against 7.2% and 7.4%. Jev is ahead in all four comparisons and at most 0.2 points behind at the 95% bound. It costs 0.07 USD per thousand sentences and needs no GPU; a dedicated GPU running a 7B model is cheaper per sentence only above 43% utilisation, far beyond what one user generates. End-to-end latency over the internet is 262 ms, of which 62 ms is spent at the provider, the same order as a 7B model on a local GPU (27 ms) but not faster.
Chinese Translation
语音神经假体通过脑活动解码尝试的语音，并最终通过一个拥有数十亿参数的语言模型对解码器的候选句子进行重评分，这是唯一需要 GPU 的组件。用一个更便宜的模型替代这个模型是困难的：当要求通用语言模型从列表中选择一个句子时，它们的回答是基于标签在列表中的位置，而不是句子本身。我们将重评分视为一个单一的输入决策，一个调用返回每个候选句子的概率，由 Jev 提供，该模型经过训练以进行校准决策，并与解码器自身的得分相结合。在来自一名 ALS 患者的 978 个保留句子中，已发布的解码器单独达到 8.1% 的字错误率，而 Jev 达到 7.5%，与 OPT-6.7b 和 Qwen2.5-7B 的 7.8% 相比；在重新调整解码器权重后，分别为 6.9% 对 7.2% 和 7.4%。在所有四个比较中，Jev 都领先，且在 95% 的置信区间内最多落后 0.2 个百分点。其成本为每千句 0.07 美元，且不需要 GPU；运行 7B 模型的专用 GPU 只有在利用率超过 43% 时每句才更便宜，这远远超出一个用户所生成的量。通过互联网的端到端延迟为 262 毫秒，其中 62 毫秒在提供商处消耗，与本地 GPU 上的 7B 模型（27 毫秒）处于同一数量级，但并不更快。
cs.CL / 114 / 2609.33612
Quizzing the Translation: A Prover-Grounded Evaluation Metric for NL$\rightarrow$FOL
翻译的考验：一种基于证明者的自然语言到一阶逻辑的评估指标
Abstract
A standard pipeline for symbolic reasoning over natural-language problems translates them into first-order logic and invokes a theorem prover. The translation step is the bottleneck: swap "every" for "some" and every inference that follows is corrupted. Yet today's metrics often score more broken translations higher than less broken ones, because BLEU, BERTScore, and Smatch++ reward surface overlap that the worst errors happen to preserve. We introduce SIV, which derives two kinds of probes from the target formula and uses a theorem prover to verify the candidate translation against each. Positive probes are statements the candidate must entail, which detect translations that drop content; contrastive probes are statements the candidate must not entail, which detect translations that assert more than the original. On a controlled pool of perturbed FOLIO translations, the severity of the error accounts for 80% of SIV's score variance, compared with at most 17% for any prior metric. Across six error classes on a disjoint pool, SIV scores the reference above the perturbed candidate in over 99% of pairs. Because each probe is labeled with what it tests, the failure pattern also supplies a labeled error trace, recovering the perturbation class at macro-F1 0.638, nearly double the score-only baseline. On 434 expert-audited real LLM translations, SIV attains the top AUC, uniquely detects and grades expert-labeled major errors, and abstains, rather than mis-scoring, on out-of-vocabulary translations.
Chinese Translation
一个标准的符号推理流程将自然语言问题翻译为一阶逻辑并调用定理证明器。翻译步骤是瓶颈：将“每个”替换为“某个”，随之而来的每个推理都被破坏。然而，今天的评估指标往往对更严重的翻译错误评分更高，因为 BLEU、BERTScore 和 Smatch++ 奖励那些最糟糕错误恰好保留的表面重叠。我们引入了 SIV，它从目标公式中派生出两种探针，并使用定理证明器验证候选翻译。正探针是候选必须蕴含的陈述，用于检测内容丢失的翻译；对比探针是候选必须不蕴含的陈述，用于检测比原始内容多的翻译。在一个受控的扰动 FOLIO 翻译池中，错误的严重性占 SIV 分数方差的 80%，而任何先前指标最多只占 17%。在一个不相交的池中，针对六类错误，SIV 在超过 99% 的对中将参考翻译的分数高于扰动候选翻译。由于每个探针都标记了其测试内容，失败模式也提供了标记的错误轨迹，以宏观 F1 0.638 恢复扰动类别，几乎是仅基于分数的基线的两倍。在 434 个专家审核的真实 LLM 翻译中，SIV 达到了最高的 AUC，独特地检测和评分专家标记的重大错误，并在超出词汇表的翻译中选择不评分，而不是错误评分。
cs.CL / 115 / 2609.33634
Safety Reconstructed: Generative Modeling via Masked Diffusion Builds Strong Safety Guardrails
安全重构：通过掩蔽扩散生成建模构建强大的安全防护
Abstract
Guard models are the last line of defense between a language model and a harmful output, yet their training objective is surprisingly narrow. Existing guards learn to predict a single verdict token from a conversational context, concentrating supervision on a single target. The consequences are structural: models latch onto shortcut features, are overconfident, and remain sensitive to where safety evidence appears in the sequence rather than its role in the full context. We propose a different framing. Rather than predicting a label from text, our LLaDA-Guard asks which label better explains the text: scoring the prompt or response under each label hypothesis and classifying based on their difference. This shifts supervision to every token in the moderated region, forcing the model to account for full content rather than its most discriminative fragments. We instantiate this idea with a masked diffusion language model, fine-tuning LLaDA-8B-Instruct with a class-conditional reconstruction objective using LoRA and requiring no architectural changes beyond the base model. LLaDA-Guard leads on average rank against discriminative baselines trained on stronger backbones across seven held-out safety benchmarks, while exhibiting substantially better confidence calibration (ECE 0.0875 vs. 0.1384 for Qwen3Guard), less over-defense on benign prompts with unsafe-looking cues, and less prompt leakage when moderating responses. Its generative nature further enables token-level risk localization as a natural byproduct, yielding a pipeline for rewriting unsafe prompts into safe equivalents without additional training and achieving a 60.7% average conversion-to-safe rate.
Chinese Translation
守护模型是语言模型与有害输出之间的最后防线，但它们的训练目标却出乎意料地狭窄。现有的守护模型学习从对话上下文中预测单一的判决标记，监督集中在单一目标上。这导致了结构性后果：模型依赖于捷径特征，表现出过于自信，并对安全证据在序列中出现的位置敏感，而非其在完整上下文中的作用。我们提出了一种不同的框架。我们的 LLaDA-Guard 不再是从文本中预测标签，而是询问哪个标签更好地解释文本：在每个标签假设下对提示或响应进行评分，并根据它们的差异进行分类。这将监督转移到被调节区域中的每个标记，迫使模型考虑完整内容，而非其最具区分性的片段。我们通过掩蔽扩散语言模型实例化这一思想，使用 LoRA 对 LLaDA-8B-Instruct 进行微调，采用类条件重构目标，并且不需要在基础模型之外进行架构更改。LLaDA-Guard 在七个保留的安全基准测试中，平均排名领先于在更强大基础上训练的区分基线，同时表现出显著更好的置信度校准（ECE 0.0875 对比 Qwen3Guard 的 0.1384），在具有不安全提示的良性提示上过度防御较少，并在调节响应时提示泄漏较少。其生成特性进一步使得标记级风险定位成为自然副产品，提供了一种将不安全提示重写为安全等价物的流程，无需额外训练，并实现了 60.7% 的平均转化为安全率。
cs.CL / 116 / 2609.33642
Learning to Learn from Context: Synthetic Training from Perturbed Public Documents
从上下文中学习：基于扰动公共文档的合成训练
Abstract
Real-world tasks often require large language models (LLMs) to learn from complex task-specific context rather than pretrained parametric knowledge. This capability remains a weakness of LLMs, while human annotation for such task contexts is expensive and difficult to scale. Public high-quality documents are an abundant alternative, but much of the public web has already been consumed during pretraining: training on such documents naively would reward memorization rather than context learning. In this work, we attempt to make use of high-quality public documents with small perturbations and empirically find that LLMs can successfully generate context-dependent reasoning traces and answers, which are then used to train a student model. Specifically, we construct a synthesis pipeline that (i) rewrites source documents to reduce memorization risk, (ii) generates questions and rubrics that require reasoning over the document, (iii) answers the questions with the document as context, and (iv) admits only samples that genuinely depend on the document. Without any human annotators, our pipeline generates about 10k samples from 3.5k documents, and the resulting student model substantially improves the performance on CL-bench. SFT raises a Qwen3.6-35B-A3B student from 13.7% to 22.8%, and a subsequent rubric-reward RL stage reaches 24.6%, on CL-bench comparable with a frontier model of over a trillion parameters, Qwen3.8-2.4T (23.9%). We also observe a broad transfer of improvements to long-context understanding, instruction following, and reasoning, while code generation and knowledge remain mostly flat. We hope this work provides a reproducible and scalable way to improve the ability of LLMs to learn from context, and to facilitate further research on context-grounded reasoning.
Chinese Translation
现实世界的任务通常要求大型语言模型（LLMs）从复杂的特定任务上下文中学习，而不是仅依赖于预训练的参数知识。这一能力仍然是LLMs的一个弱点，而人工标注此类任务上下文既昂贵又难以扩展。高质量的公共文档是一个丰富的替代选择，但公共网络中的许多内容在预训练过程中已经被消耗：在此类文档上进行训练将简单地奖励记忆而非上下文学习。在本研究中，我们尝试利用经过小扰动的高质量公共文档，并通过实证发现LLMs能够成功生成依赖上下文的推理轨迹和答案，这些答案随后用于训练学生模型。具体而言，我们构建了一个合成管道，该管道（i）重写源文档以降低记忆风险，（ii）生成需要对文档进行推理的问题和评分标准，（iii）以文档作为上下文回答问题，以及（iv）仅接受真正依赖于文档的样本。在没有任何人工标注者的情况下，我们的管道从3500个文档中生成了约10,000个样本，结果学生模型在CL-bench上的性能显著提升。SFT将Qwen3.6-35B-A3B学生模型的性能从13.7%提高到22.8%，而随后的评分奖励强化学习阶段在CL-bench上达到了24.6%，与超过一万亿参数的前沿模型Qwen3.8-2.4T（23.9%）相当。我们还观察到在长上下文理解、指令跟随和推理方面的广泛改进转移，而代码生成和知识的表现基本保持平稳。我们希望这项工作提供了一种可重复和可扩展的方法，以提高LLMs从上下文中学习的能力，并促进对基于上下文的推理的进一步研究。
cs.CL / 117 / 2609.33652
Tsubame: Tree Replay for Diffusion-Based Speculative Decoding
Tsubame：基于扩散的推测解码的树重放
Abstract
Context-aware dynamic trees allocate the speculative decoding budget according to draft path probabilities, adapting their depth and branching to the current context. Under stochastic decoding, however, we find that this structural advantage does not always compensate for the acceptance gains of random sampling paired with advanced verification, and such dynamic trees can fall behind sampled chains in some settings. These trees grow their topology from the candidates themselves, so the tokens submitted for verification are typically the deterministic high-score tokens selected during construction. This coupling is not inherent: once the topology is fixed, its nodes can be repopulated by sampling, allowing dynamic trees to retain their structural advantage while also benefiting from random sampling and advanced verification. Diffusion-based drafters make this practical, as their parallel outputs or lightweight conditional corrections allow candidates to be regenerated cheaply after the complete topology is known. We introduce Tsubame, a two-pass tree speculative decoding framework for diffusion-based drafters. The first pass plans and freezes a context-aware topology using draft path scores; the second replays the fixed topology, sampling the tokens that populate its nodes to form the candidate tree for verification. We prove that Tsubame is lossless under compatible sampling and verification strategies. Experiments across three diffusion-based drafters, six datasets, and multiple candidate budgets show that Tsubame improves acceptance length and throughput over deterministic trees, including settings where it reverses their disadvantage against sampled chains.
Chinese Translation
上下文感知动态树根据草稿路径概率分配推测解码预算，调整其深度和分支以适应当前上下文。然而，在随机解码下，我们发现这种结构优势并不总能弥补与先进验证相结合的随机采样的接受增益，因此在某些情况下，这些动态树可能会落后于采样链。这些树的拓扑结构是从候选者自身生长而来的，因此提交进行验证的标记通常是构建过程中选择的确定性高分标记。这种耦合并不是固有的：一旦拓扑结构固定，其节点可以通过采样重新填充，从而使动态树能够保持其结构优势，同时也受益于随机采样和先进验证。基于扩散的草拟器使这一点变得可行，因为它们的并行输出或轻量级条件修正允许在完整拓扑已知后廉价地重新生成候选者。我们介绍了Tsubame，一个用于基于扩散的草拟器的双遍树推测解码框架。第一遍使用草稿路径分数规划并固定上下文感知拓扑；第二遍重放固定拓扑，采样填充其节点的标记以形成验证的候选树。我们证明了在兼容的采样和验证策略下，Tsubame是无损的。在三个基于扩散的草拟器、六个数据集和多个候选预算的实验中，结果表明Tsubame在接受长度和吞吐量上优于确定性树，包括在逆转其相对于采样链的劣势的设置中。
cs.CL / 118 / 2609.33666
One Model Is Not a Crowd: Multi-LLM and Aspect-Conditioned Diverse Comment Generation
单一模型不足以应对多样性：多LLM与基于方面的多样化评论生成
Abstract
Human communication on the internet is shaped by diverse perspectives, most visibly expressed in online comment spaces. As large language model (LLM)based AI agents begin to inhabit these spaces, a key question arises: whether synthetic comment threads can capture the diversity inherent in human discourse. This concern is increasingly important, as the growing presence of homogenized AI-generated content risks reducing diversity over time, potentially leading to model collapse and degrading the richness of digital communication. Inspired by the plurality of human crowds and the aspect-driven nature of discourse, we hypothesize that comment diversity is better approximated by combining multiple LLMs with aspect-conditioned generation. We formalize and evaluate this approach using models from different providers and introduce a framework that characterizes diversity across semantic, linguistic, and socio-pragmatic features along three axes: dispersion, coverage, and alignment. Using this framework, we conduct a large-scale study on over 2 million YouTube comments across multiple domains. Our results reveal that multi-LLM and aspect-conditioned generation better align with human comment distributions and such data remains viable under pretraining style curation and is effective for downstream tasks. Yet, human diversity remains unmatched. Overall, our findings provide a practical foundation for generating more diverse and socially grounded discourse in AI-mediated environments.
Chinese Translation
互联网人类沟通受到多元视角的影响，这种多样性在在线评论空间中表现得尤为明显。随着基于大型语言模型（LLM）的人工智能代理开始进入这些空间，一个关键问题随之而来：合成评论线程是否能够捕捉到人类话语中的固有多样性。随着同质化的AI生成内容日益增多，这一担忧变得愈发重要，因为这可能会随着时间的推移降低多样性，最终导致模型崩溃并削弱数字沟通的丰富性。受到人类群体多样性和话语方面驱动特性的启发，我们假设，通过结合多个LLM与基于方面的生成，评论多样性能够得到更好的近似。我们正式化并评估这一方法，使用来自不同提供商的模型，并引入一个框架，通过三个维度：分散性、覆盖性和一致性，来表征语义、语言和社会语用特征的多样性。利用该框架，我们对超过200万条YouTube评论进行了大规模研究，涵盖多个领域。我们的结果表明，多LLM与基于方面的生成更好地与人类评论分布对齐，并且在预训练风格的策划下，这些数据仍然具有可行性，并对下游任务有效。然而，人类的多样性仍然无可匹敌。总体而言，我们的研究结果为在AI介导的环境中生成更具多样性和社会基础的话语提供了实用基础。
cs.CL / 119 / 2609.33670
Closing the Cross-Dialect Gap: Query Plans as a Portable Interface in Text-to-SQL
缩小跨方言差距：查询计划作为文本到SQL的可移植接口
Abstract
Text-to-SQL systems are typically trained and evaluated on a single dialect (SQLite), yet production deployments span PostgreSQL, MySQL, ClickHouse, and beyond. We show that this single-dialect assumption leads to a substantial drop in cross-dialect accuracy for every model we tested. The drop persists across scale, architecture, and even purpose-built text-to-SQL systems. We argue that the fix is to change the generation target: instead of asking an LLM to emit dialect-specific SQL, we have it emit a dialect-agnostic relational algebra query plan, which a deterministic compiler then renders into SQL for any supported backend. Across thirteen models from 3B to frontier scale, this restores cross-dialect portability nearly uniformly, at a small cost in peak accuracy on the model's home dialect for capable prompted models and none once fine-tuned on plans; under matched fine-tuning, plan supervision yields a stronger model than SQL supervision. We also introduce MetricName, a question-aware result-set comparator needed to evaluate fairly across dialects, where existing metrics confound semantic errors with benign cross-dialect variation. More broadly, the result is a reminder that a generation target chosen for execution is not necessarily the one that maximizes generation quality.
Chinese Translation
文本到SQL系统通常在单一方言（SQLite）上进行训练和评估，但生产部署涉及PostgreSQL、MySQL、ClickHouse等多种数据库。我们展示了这种单一方言假设导致我们测试的每个模型在跨方言准确性上显著下降。该下降在规模、架构，甚至是专门构建的文本到SQL系统中都持续存在。我们认为解决方案在于改变生成目标：与其要求大型语言模型（LLM）生成特定方言的SQL，不如让其生成一种与方言无关的关系代数查询计划，然后由确定性编译器将其转换为任何支持的后端SQL。在从3B到前沿规模的十三个模型中，这几乎均匀地恢复了跨方言的可移植性，对于能够提示的模型在其本方言上的峰值准确性略有下降，而在经过计划微调后则没有下降；在匹配微调下，计划监督产生的模型优于SQL监督。我们还引入了MetricName，这是一种问题感知的结果集比较器，用于在方言之间进行公平评估，因为现有指标将语义错误与良性的跨方言变异混淆。更广泛地说，这一结果提醒我们，选择用于执行的生成目标不一定是最大化生成质量的目标。
cs.CL / 120 / 2609.33672
Reset Is Not Recovery: Evaluating Recoverability from False Conversational Context via Sycophancy Hysteresis
重置并非恢复：通过谄媚滞后评估从错误对话上下文中的可恢复性
Abstract
Grounded language models are usually evaluated by adding relevant context, but multiturn dialogue also contains unsupported user claims that may contaminate later factual answers. We study post-pressure recoverability: whether a model returns to clean-context behavior after a user repeatedly advocates a wrong answer and then withdraws that pressure. We introduce a recovery-after-pressure protocol for multiple-choice factual dialogue and measure sycophancy hysteresis, the residual probability assigned to the user-advocated wrong answer relative to a clean-context counterfactual. Across seven instruction-tuned open-weight models and two factual benchmarks, ordinary reset often reduces but does not erase pressure-induced bias. History preserving repairs such as user retraction, system reset, and self-verification recover only 2-3/14 model-dataset pairs under the strict clean-restoration diagnostic, whereas operations that change the effective context are substantially more reliable; the two conditions that remove the pressure-bearing history entirely, fresh-context deletion and context truncation, recover 14/14. In an oracle trusted-evidence condition across fourteen model-dataset pairs, preserving the pressure-bearing history while adding benchmark-derived trusted evidence increases accuracy from 0.368 to 0.929, while wrong-answer following falls from 41.2% to 4.3%. Controls show that the effect is not explained by dialogue length, repeated confidence, plausible distractors, mere false-answer mention, or option-label inertia. These results suggest that faithful grounded dialogue requires evaluating which prior context should be treated as evidence and which should be removed or quarantined before answering.
Chinese Translation
基于知识的语言模型通常通过添加相关上下文进行评估，但多轮对话中也包含可能污染后续事实回答的不支持用户主张。我们研究了压力后的可恢复性：即在用户反复主张错误答案并随后撤回该压力后，模型是否能够恢复到清晰上下文的行为。我们引入了一种针对多项选择事实对话的压力后恢复协议，并测量谄媚滞后，即相对于清晰上下文反事实，分配给用户主张的错误答案的残余概率。在七个经过指令调优的开放权重模型和两个事实基准测试中，普通的重置通常减少但并未消除由压力引起的偏见。历史保留修复（如用户撤回、系统重置和自我验证）在严格的清晰恢复诊断下仅恢复了2-3/14个模型-数据集对，而改变有效上下文的操作则显著更可靠；完全消除承载压力历史的两种条件，即新上下文删除和上下文截断，恢复了14/14。在十四个模型-数据集对的神谕可信证据条件下，保留承载压力的历史同时添加基准派生的可信证据，准确率从0.368提高到0.929，而错误答案的跟随率从41.2%降至4.3%。控制实验表明，该效应并非由对话长度、重复信心、合理干扰项、单纯的错误答案提及或选项标签惯性所解释。这些结果表明，忠实的基于知识的对话需要评估哪些先前的上下文应被视为证据，哪些应在回答之前被移除或隔离。
cs.CL / 121 / 2609.33691
When Do Agents Help? Embedding, LLM and Agentic Alignment of Classical Texts and Their Translations
代理何时提供帮助？经典文本及其翻译的嵌入、LLM与代理对齐
Abstract
Classical texts aligned with their translations support machine translation, retrieval and computational research, but evidence comparing alignment workflows is scattered. This study compares seven systems on 452 texts in Pali, Sanskrit, Mishnaic Hebrew and Tibetan, comprising 9,833 human-aligned units: four embedding pipelines, a direct LLM call, an autonomous agent, and the agent revised by an independent auditor. Generative workflows recover 93-94% of reference correspondences, against at most 77% for embeddings. A ceiling analysis shows that sentence boundaries make some references unrepresentable by the embedding pipelines. Reference recovery is similar across generative workflows: the agent's advantage is 0.5 percentage points (95% CI -0.02 to 1.17), and auditing adds no established benefit. Agents nevertheless produce structurally valid output for all 452 texts, against 437 for direct calls. A blinded three-LLM panel assesses every generative mismatch against the source and human reference. Most mismatches are labelled defensible editorial variation; consensus major-error labels cover only 0.06-0.14% of units. The panel labels significantly fewer residual defects for agents than direct calls (0.7% versus 1.4%), suggesting that reference recovery alone understates alignment quality. On ten long Pali discourses taken as published online, agents and audited agents raise recovery from the direct call's 71% to 84% and 92%. Identical reference-located chunks bring all three to 93%. Agents thus improve structural reliability and reduce judged defects on short passages, while their large recovery advantage on long documents disappears after chunking. In this setting, independent auditing offers little measurable additional benefit on prepared passages.
Chinese Translation
与其翻译对齐的经典文本支持机器翻译、检索和计算研究，但比较对齐工作流程的证据分散。本研究比较了七个系统在452个巴利语、梵语、米什那希伯来语和藏语文本上的表现，共包含9,833个人工对齐单元：四个嵌入管道、一个直接的LLM调用、一个自主代理，以及由独立审计员修订的代理。生成工作流程恢复了93-94%的参考对应关系，而嵌入最多仅为77%。顶层分析显示，句子边界使得某些参考无法被嵌入管道表示。生成工作流程的参考恢复相似：代理的优势为0.5个百分点（95%置信区间 -0.02至1.17），审计并未带来已建立的益处。尽管如此，代理仍为所有452个文本生成了结构有效的输出，而直接调用仅为437个。一个盲评的三人LLM小组评估每个生成的不匹配情况与源文本及人工参考的对比。大多数不匹配被标记为可辩护的编辑变体；共识主要错误标签仅覆盖0.06-0.14%的单元。小组对代理的剩余缺陷标记显著少于直接调用（0.7%对比1.4%），这表明仅依靠参考恢复低估了对齐质量。在十个被视为在线发布的长篇巴利语演讲中，代理和经过审计的代理将直接调用的恢复率从71%提高到84%和92%。相同参考定位的块使得三者均达到93%。因此，代理提高了结构可靠性，并减少了对短段落的判断缺陷，而在长文档中，其较大的恢复优势在分块后消失。在这种情况下，独立审计对准备好的段落几乎没有可测量的附加益处。
cs.CL / 122 / 2609.33702
Understanding Confabulation and Rethinking Reconstruction in Activation Explanations
理解虚构与重新思考激活解释中的重建
Abstract
Natural Language Autoencoders (NLAs) produce unsupervised text explanations of a model's activations: a verbalizer describes an activation and a reconstructor learns to recover it from this text. Under the established point-reconstruction NLA training recipe, explanations become more useful for predicting model behavior while also increasingly introducing unsupported details and exhibiting writing defects. To assess these changes separately, we introduce a standardized evaluation framework for unstructured NLA explanations, measuring information recoverable from explanations, contextual support for their claims, and writing quality. To address confabulation and writing defects, we move beyond predicting a single activation: explanations can distinguish distributions of possible activations even when their means and optimal point-reconstruction rewards are identical. We introduce Flow-NLA, which models the distribution of activations compatible with an explanation and trains the verbalizer using a diffusion likelihood bound. Across Qwen, Gemma, and Apertus, this richer signal retains the utility gains of point reconstruction while curbing the growth of confabulation and writing defects, opening up a direction for improving activation-derived training to encourage more informative, supported, and readable explanations. Code and evaluation prompts will be made publicly available upon acceptance.
Chinese Translation
自然语言自编码器（Natural Language Autoencoders, NLAs）生成模型激活的无监督文本解释：一个表述者描述激活，而一个重构者学习从该文本中恢复激活。在既定的点重建NLA训练方案下，解释在预测模型行为方面变得更加有用，但同时也越来越多地引入不支持的细节并表现出写作缺陷。为了单独评估这些变化，我们引入了一个标准化的无结构NLA解释评估框架，测量从解释中可恢复的信息、对其主张的上下文支持以及写作质量。为了解决虚构和写作缺陷，我们超越了对单一激活的预测：解释可以区分可能激活的分布，即使它们的均值和最佳点重建奖励是相同的。我们引入了Flow-NLA，它建模与解释兼容的激活分布，并使用扩散似然界限训练表述者。在Qwen、Gemma和Apertus上，这种更丰富的信号保留了点重建的效用增益，同时抑制了虚构和写作缺陷的增长，为改善激活衍生训练提供了一个方向，以鼓励更具信息性、支持性和可读性的解释。代码和评估提示将在接受后公开发布。
cs.CL / 123 / 2609.33720
From Granular Revision Operations to Meaningful Revision Units: Evaluating LLMs for Revision Boundary Detection
从细粒度修订操作到有意义的修订单元：评估大型语言模型在修订边界检测中的表现
Abstract
Revision traces provide valuable evidence about students' writing processes, but their usefulness for learning analytics depends on how individual revisions are represented. Automated draft-comparison methods often produce granular edit operations that can fragment a single purposeful revision into multiple analytic units. This study evaluates whether LLMs can identify meaningful revision unit boundaries in structured revision operation data and whether they provide value beyond simple non-LLM baselines. Using 113 matched draft--revision pairs from undergraduate writing, expert annotation yielded 4,344 candidate boundaries. We compared zero- and few-shot GPT-5.5 and base Qwen3-32B, parameter-efficient fine-tuning of Qwen3-32B, and majority and proximity-based baselines. Despite receiving revision context and task instructions, no prompted LLM condition outperformed the proximity heuristic (macro-F1 = .825). In contrast, fine-tuned Qwen3-32B using the two context representation achieved the highest macro-F1 (.859), identifying more same-unit relationships while maintaining precision comparable to the heuristic. Deterministic post-processing substantially improved the prompted models but added little benefit to the strongest fine-tuned model. These findings suggest that LLMs can support revision boundary judgment when task-adapted, but general purpose prompting alone may not outperform transparent structural heuristics.
Chinese Translation
修订痕迹提供了关于学生写作过程的宝贵证据，但其在学习分析中的有效性取决于个别修订的表示方式。自动化草稿比较方法通常会产生细粒度的编辑操作，这可能将一个有目的的修订分割成多个分析单元。本研究评估了大型语言模型（LLMs）是否能够在结构化的修订操作数据中识别有意义的修订单元边界，以及它们是否提供了超越简单非LLM基线的价值。我们使用了113对匹配的草稿-修订对，专家注释产生了4,344个候选边界。我们比较了零-shot和少量-shot的GPT-5.5与基础的Qwen3-32B、参数高效的Qwen3-32B微调模型，以及基于多数和接近度的基线。尽管提供了修订上下文和任务指令，但没有任何提示的LLM条件超越接近度启发式（宏F1 = 0.825）。相反，使用两种上下文表示的微调Qwen3-32B达到了最高的宏F1（0.859），在识别更多同一单元关系的同时保持了与启发式相当的精确度。确定性后处理显著改善了提示模型，但对最强的微调模型几乎没有额外好处。这些发现表明，当任务适应时，LLMs可以支持修订边界判断，但仅依靠通用提示可能无法超越透明的结构启发式。
cs.CL / 124 / 2609.33721
Shared Experience, Separate Learning: Companion Confidence Calibration for LLMs
共享经验，独立学习：大型语言模型的伴随信心校准
Abstract
Reliable self-assessment is essential for large language models (LLMs), yet they often remain highly confident when their answers are wrong. We study \emph{concurrent confidence calibration}, where confidence is learned alongside capability improvement rather than calibrated only after training. Reinforcement learning from verifiable rewards (RLVR) provides a natural setting for this paradigm, as it continuously produces responses paired with verifiable correctness feedback. Existing concurrent methods, however, learn both capability and confidence through reinforcement learning within shared policy parameters, potentially coupling two fundamentally different learning problems. We instead propose \emph{shared experience but separate learning}: capability and confidence learn from the same trajectories, but through separate optimization mechanisms and parameters. Based on this principle, we introduce \textbf{CoCal (Companion Confidence Calibration)}, which trains a lightweight companion from rollout hidden states and verifier-derived correctness supervision while leaving task optimization unchanged. Experiments on Qwen3-8B and Qwen3-14B show that CoCal improves confidence estimation without sacrificing task performance, outperforming both RL-based concurrent methods and matched post-hoc calibration. The learned companion further generalizes across domains and policy shifts, while the benefits of CoCal persist at both scales.
Chinese Translation
可靠的自我评估对于大型语言模型（LLMs）至关重要，但它们在回答错误时往往仍表现出高度的自信。我们研究了 extit{并行信心校准}，即信心与能力提升同时学习，而不是仅在训练后进行校准。基于可验证奖励的强化学习（RLVR）为这一范式提供了自然的环境，因为它持续生成与可验证正确性反馈配对的响应。然而，现有的并行方法通过共享策略参数在强化学习中同时学习能力和信心，这可能将两个根本不同的学习问题耦合在一起。相反，我们提出了 extit{共享经验但独立学习}：能力和信心从相同的轨迹中学习，但通过独立的优化机制和参数。基于这一原则，我们引入了 extbf{CoCal（伴随信心校准）}，它从回滚隐藏状态和验证者导出的正确性监督中训练一个轻量级的伴随模型，同时保持任务优化不变。在Qwen3-8B和Qwen3-14B上的实验表明，CoCal在不牺牲任务性能的情况下改善了信心估计，优于基于RL的并行方法和匹配的事后校准。所学习的伴随模型进一步在不同领域和策略变化中具有良好的泛化能力，同时CoCal的优势在两个规模上均得以保持。
cs.CL / 125 / 2609.33734
Yor\`{u}b\'{a} in Unicode: An Overview of a Problem
Unicode中的约鲁巴语：问题概述
Abstract
There is a recurrent problem in the writing of Yor\`ub\'a on the internet and on the computer that has proven intractable over the years. The language, along with other African languages that depend on diacritics for disambiguation, requires a small set of precomposed characters that Unicode does not encode. This has forced writers and digital systems to rely on combining character sequences that behave inconsistently across platforms, corrupt under font substitution, and fail in search. This paper documents that failure across a range of real world contexts, from published books to web platforms to mobile keyboards, using personal and empirical evidence. It identifies Unicode's NFC normalization stability policy as the structural constraint that prevents a straightforward fix, arguing for direct intervention of the Consortium in solving the active problem, proposing a formal encoding request for the four core Yor\`ub\'a characters as the most durable path to resolution.
Chinese Translation
在互联网和计算机上书写约鲁巴语存在一个反复出现的问题，这个问题多年来一直难以解决。该语言以及其他依赖于变音符号进行消歧义的非洲语言，需要一小组Unicode未编码的预组合字符。这迫使写作者和数字系统依赖于在不同平台上表现不一致的组合字符序列，这些序列在字体替换时会损坏，并且在搜索中失败。本文通过个人和实证证据，记录了这一失败在从出版书籍到网络平台再到移动键盘等一系列现实世界背景中的表现。它指出Unicode的NFC规范化稳定性政策是阻碍直接解决的结构性约束，呼吁联盟（Consortium）直接介入解决这一活跃问题，并提出对四个核心约鲁巴字符的正式编码请求，作为解决问题的最持久路径。
cs.CL / 126 / 2609.33738
The Effects of Incremental Instruction Delivery on Language-Model Creative Writing
增量指令传递对语言模型创意写作的影响
Abstract
Large language models are increasingly used as interactive writing tools, where users develop stories, revise ideas, and introduce new requirements across multiple turns rather than specifying a complete brief upfront. Yet most evidence on multi-turn instruction degradation comes from tasks with objectively verifiable outcomes, leaving unclear whether incremental interaction harms creative artifacts in ways that explicit requirement checks cannot capture. We study this question using 160 human-authored creative-writing tasks across six genres, presenting each intended specification either upfront or progressively over 5-9 turns to six distinct open-weight model families, yielding 960 matched pairs. Progressive delivery reduces explicit constraint adherence and produces its largest writing-quality degradation in structure/coherence. The structural gap persists among outputs with equal observed adherence, suggesting that measured requirement loss alone does not explain the observed structural difference. We define Creative Integrity as a compact measure of joint adherence and narrative structure; under incremental delivery, models retain 71.2% of FULL Creative Integrity (95% CI [68.2%, 74.3%]). A three-rater human study over 50 matched pairs independently recovers FULL advantages in structure/coherence, craft, and genre effectiveness, while automated scores remain positively associated with aggregated human ratings. These findings show that interactive creative-writing systems should be evaluated not only on whether requirements survive conversation, but also on whether evolving requirements remain coherently integrated into the final artifact. Our dataset, benchmarks, and source code are available at: https://github.com/solusops/SISTER-2026-Team19
Chinese Translation
大型语言模型越来越多地被用作互动写作工具，用户在多个回合中发展故事、修订想法并引入新要求，而不是事先明确指定完整的任务。然而，关于多轮指令衰退的大多数证据来自于具有客观可验证结果的任务，这使得尚不清楚增量互动是否以显式要求检查无法捕捉的方式损害创意作品。我们通过对六个类型的160个人工创作的创意写作任务进行研究，将每个预期规范要么提前提供，要么在5-9个回合中逐步提供给六个不同的开放权重模型系列，从而产生960对匹配样本。逐步传递减少了显式约束遵循，并在结构/连贯性方面产生了最大的写作质量下降。即使在观察到的遵循程度相等的输出中，结构差距依然存在，这表明仅仅测量要求的损失并不能解释观察到的结构差异。我们将创意完整性定义为联合遵循和叙事结构的紧凑度量；在增量传递下，模型保留了71.2%的完整创意完整性（95%置信区间 [68.2%, 74.3%]）。一项针对50对匹配样本的三评审人类研究独立恢复了在结构/连贯性、工艺和类型有效性方面的完整优势，而自动评分与汇总的人类评分仍然呈正相关。这些发现表明，互动创意写作系统的评估不仅应关注要求是否在对话中得以保留，还应关注不断演变的要求是否能与最终作品保持连贯的整合。我们的数据集、基准和源代码可在以下网址获取：https://github.com/solusops/SISTER-2026-Team19
cs.CL / 127 / 2609.33759
Positions Are Not Facts: The Mismatch Between KV Caches and Memory
位置不是事实：KV 缓存与内存之间的不匹配
Abstract
When a fact changes, how should a language model update the history stored in its key-value (KV) cache? Hiding the old record is cheap, but it may still contain needed details or answer questions about the past. We compare hiding whole records, hiding only replaced values, and deleting old text and recomputing the cache. In a controlled quantity task, masking makes all eight models prefer the new value more strongly, yet six lose complete answers through unit errors or failure to stop; keeping the unit preserves all current answers. Later states also retain useful information from earlier records: on multi-hop updates, rebuilding these states at unchanged positions lowers historical accuracy by 20-41 percentage points, whereas moving the existing states has little effect. Keeping object dependencies and unchanged revision passages prevents many losses. Recognition is a separate challenge. Learned readouts recover distinctions missed by fixed cache similarities on synthetic record pairs. On natural text, text-detector-selected masks show no clear advantage over random masks at the same rate in 14 same-model detector-generator comparisons. Query-dependent access can avoid some losses, with additional storage or access costs. These findings identify what must be preserved beyond the replaced value when using a KV cache as updatable memory.
Chinese Translation
当一个事实发生变化时，语言模型应如何更新其在键值（KV）缓存中存储的历史记录？隐藏旧记录的成本较低，但它可能仍包含所需的细节或回答关于过去的问题。我们比较了隐藏整个记录、仅隐藏被替换的值以及删除旧文本并重新计算缓存。在一个受控数量任务中，掩蔽使得所有八个模型更强烈地偏好新值，然而六个模型由于单位错误或未能停止而失去了完整答案；保持单位则保留了所有当前答案。后续状态也保留了早期记录中的有用信息：在多跳更新中，在不变位置重建这些状态会使历史准确性降低20-41个百分点，而移动现有状态几乎没有影响。保持对象依赖关系和不变的修订段落可以防止许多损失。识别是一个单独的挑战。学习的读出能够恢复固定缓存相似性在合成记录对中遗漏的区别。在自然文本中，在14个同模型检测器-生成器比较中，文本检测器选择的掩蔽在相同速率下并未显示出明显的优势。查询依赖的访问可以避免一些损失，但会增加额外的存储或访问成本。这些发现确定了在使用 KV 缓存作为可更新内存时，除了被替换的值之外必须保留的内容。
cs.CL / 128 / 2609.33787
LA-CPD: Local-Evidence-Aware Change-Point Detection for Human-LLM Authorship Segmentation
LA-CPD：基于局部证据的变更点检测用于人类-大语言模型作者分段
Abstract
As LLM-generated text becomes increasingly human-like, accurately localizing LLM-authored spans in human-LLM co-authored documents is important for attribution and accountability in cases involving copyright infringement, fraud, and other harmful uses of AI-generated content. Sentence-level detectors provide local authorship evidence, but content variation can cause score fluctuations even among sentences from the same source, creating spurious boundaries. Recovering a coherent document partition therefore remains challenging when both the number and locations of authorship transitions are unknown. We propose Local-Evidence-Aware Change-Point Detection (LA-CPD), a structured method that transforms noisy sentence-level score sequences into coherent authorship segments. Given scores from a frozen local detector, LA-CPD combines a length-weighted within-segment residual with a windowed two-mean contrast to capture segment consistency and sustained changes around candidate cut points. Dynamic programming optimizes cut locations for each candidate count, while an AIC-style criterion selects the final partition, yielding sentence labels, authorship boundaries, and maximal LLM-authored spans. On a held-out human-LLM co-authored test set, LA-CPD outperforms WCP+AIC, increasing sentence-level accuracy from 0.747 to 0.796 while improving boundary localization and LLM-span delineation.
Chinese Translation
随着大语言模型生成的文本越来越像人类写作，准确定位人类-大语言模型共同创作文档中的大语言模型作者段落对于涉及版权侵权、欺诈和其他有害的人工智能生成内容的使用的归属和问责至关重要。句子级检测器提供了局部作者证据，但内容的变化可能导致即使来自同一来源的句子之间也存在分数波动，从而产生虚假的边界。因此，当作者转变的数量和位置未知时，恢复一致的文档分区仍然具有挑战性。我们提出了局部证据感知变更点检测（LA-CPD），这是一种结构化方法，将嘈杂的句子级分数序列转化为一致的作者段落。LA-CPD在给定冻结的局部检测器分数的情况下，结合了长度加权的段内残差和窗口化的双均值对比，以捕捉段的一致性和候选切点周围的持续变化。动态规划优化每个候选计数的切点位置，而AIC风格的标准选择最终分区，生成句子标签、作者边界和最大的大语言模型作者段落。在一个保留的人类-大语言模型共同创作测试集上，LA-CPD的表现优于WCP+AIC，将句子级准确率从0.747提高到0.796，同时改善了边界定位和大语言模型段落的划分。
cs.CL / 129 / 2609.33865
In-Context Adaptation of Encoder-Decoder Models in Speech Recognition
语音识别中编码器-解码器模型的上下文适应
Abstract
In-context learning offers an appealing approach to adapt automatic speech recognition (ASR) models to new speakers, accents, and domains by providing speech-text pairs as demonstrations at inference time. Recent work shows that some LLM-based speech models are capable of ASR in-context adaptation, when providing interleaved speech-text demonstrations. In this work, we ask whether in-context adaptation is an inherent ability for all encoder-decoder models. We study two forms of demonstration, collated and interleaved demonstration, across six encoder-decoder models, spanning conventional cross-attention-based and LLM-based architectures. We find that all tested models are able to perform in-context adaptation out of the box, achieving up to 30% relative improvement in the oracle experiments and up to 23% using first-pass hypotheses. Through controlled experiments on three English datasets, we show that lexical and speaker information both contribute to successful adaptation. While interleaved demonstration is effective in certain cases, collated demonstration brings consistent adaptation across the board. Our results suggest that in-context adaptation for ASR is not unique to specific architectures, training, or demonstration approaches.
Chinese Translation
上下文学习为将自动语音识别（ASR）模型适应于新说话者、口音和领域提供了一种吸引人的方法，通过在推理时提供语音-文本对作为示范。近期研究表明，当提供交错的语音-文本示范时，一些基于大型语言模型（LLM）的语音模型能够进行ASR的上下文适应。在本研究中，我们探讨上下文适应是否是所有编码器-解码器模型的固有能力。我们研究了两种示范形式：整理示范和交错示范，涵盖六种编码器-解码器模型，包括传统的基于交叉注意力的架构和基于LLM的架构。我们发现所有测试的模型都能够开箱即用地执行上下文适应，在理想实验中实现了高达30%的相对提升，而在使用第一次假设时则达到23%。通过对三个英语数据集的控制实验，我们表明词汇和说话者信息均对成功适应有所贡献。尽管交错示范在某些情况下有效，但整理示范在各个方面带来了持续的适应。我们的结果表明，ASR的上下文适应并不特定于特定的架构、训练或示范方法。
cs.CL / 130 / 2609.33883
Lost with a Map: Conversational State and Behavioral Reliability in Language Models
迷失在地图中：语言模型中的对话状态与行为可靠性
Abstract
Task-oriented dialogue requires maintaining and updating information across turns, yet language models expose no explicit belief-state object. We study how conversational state is represented, updated, and used inside eight instruction-tuned language models from four families on MultiWOZ and SGD. Structure and values separate: which domains, slots, and requests are active is linearly readable just before the model acts, whereas exact values are far more readable where the user stated them. After a user changes a value, both values remain accessible at their mentions, and causal interventions show that both continue to influence the model's action. In natural closed-loop interaction, query failures separate into cases of weak structural support, incorrect value resolution, and failure to deploy otherwise-supported constraints, with targeted interventions producing systematically different repair behavior across these cases. These findings motivate a state-action controller that starts from the base model action and selectively edits it using structural readouts, without requiring a complete predicted belief state as an intermediate representation. On held-out MultiWOZ interaction across five models, it raises the base model exact-query accuracy from .318 to .621 and task success from .272 to .371 at negligible added cost. Overall, reliable interaction requires not only retaining conversational information, but resolving which available constraints currently apply and ensuring that they govern action.
Chinese Translation
任务导向的对话需要在多个回合中维护和更新信息，但语言模型并未显式暴露信念状态对象。我们研究了在四个家族的八个指令调优语言模型中，对话状态是如何被表示、更新和使用的，数据集为 MultiWOZ 和 SGD。结构和数值是分离的：在模型采取行动之前，哪些领域、槽位和请求处于活动状态是线性可读的，而确切的数值在用户陈述时更易于读取。当用户更改一个数值后，这两个数值在其提及处仍然可访问，因果干预表明这两者仍然影响模型的行为。在自然的闭环交互中，查询失败分为弱结构支持、不正确的数值解析和未能部署其他支持的约束等情况，针对性的干预在这些情况下产生了系统性不同的修复行为。这些发现促使我们提出一种状态-行动控制器，该控制器从基础模型的行动开始，并使用结构读出进行选择性编辑，而无需将完整的预测信念状态作为中间表示。在五个模型的 MultiWOZ 保留交互中，它将基础模型的精确查询准确率从 0.318 提高到 0.621，任务成功率从 0.272 提高到 0.371，且增加的成本微不足道。总体而言，可靠的交互不仅需要保留对话信息，还需解决当前适用的可用约束，并确保这些约束主导行动。
cs.CL / 131 / 2609.33886
LLMs learn different forms of metacognition when trained to predict their own accuracy
大型语言模型在训练预测自身准确性时学习到不同形式的元认知
Abstract
Large language models are trained to always produce an answer, regardless of whether they possess the relevant knowledge, which leads them to fabricate facts. Prior work has shown that LLMs' confidence estimates correspond poorly to their actual performance, and that fine-tuning can substantially improve them. However, what models actually learn during such training remains poorly understood. We investigate how LLMs acquire metacognitive monitoring, the ability to know what one knows, by training 10 open-weight LLMs to predict their own accuracy on factual multiple-choice questions before answering them. We find that trained confidence reflects two distinct signals. While on questions close to the training data, it tracks the model's true accuracy, in other domains, it instead tracks output consistency: the concentration of the model's answer distribution. Output consistency tracking emerges early in training and generalizes across datasets, whereas accuracy tracking develops later and remains local to the training distribution. These results suggest that calibration training may not teach models to generally detect errors they commit confidently, and they raise broader questions about the nature of metacognition in artificial systems.
Chinese Translation
大型语言模型被训练为始终给出答案，无论它们是否具备相关知识，这导致它们编造事实。先前的研究表明，大型语言模型的置信度估计与其实际表现之间的对应关系较差，而微调可以显著改善这一点。然而，模型在此类训练中实际学习到的内容仍然不甚明了。我们通过训练10个开放权重的大型语言模型预测其在事实性多项选择题上的准确性，来研究大型语言模型如何获得元认知监控，即知道自己知道什么的能力。我们发现，训练后的置信度反映了两种不同的信号。在接近训练数据的问题上，它跟踪模型的真实准确性，而在其他领域，它则跟踪输出一致性：模型答案分布的集中度。输出一致性跟踪在训练早期就出现，并且在不同数据集之间具有泛化能力，而准确性跟踪则在后期发展，并且局限于训练分布。这些结果表明，校准训练可能并未教会模型普遍检测它们自信地犯下的错误，并引发了关于人工系统中元认知本质的更广泛问题。
cs.CL / 132 / 2609.33887
Faster Block-Diffusion Serving with Distribution-Free Risk Guarantees
具有无分布风险保证的更快块扩散服务
Abstract
Block-diffusion language models are served at hand-picked operating points, such as acceptance thresholds, buffer depth, schedule, checkpoint and precision, and each point is chosen by its mean benchmark accuracy. However, a mean does not tell an operator how often a faster configuration fails on prompts that the slower one answers correctly. On the serving engine and its decode traces, the default commit rule already commits every fully resolved block, a static skip rule captures nearly all of the compute that allocation can save, and self-distillation on engine-decoded targets adds speed at unchanged accuracy. Larger speedups come from lower thresholds, which commit tokens that are still uncertain. We therefore present Redline, a finite-sample procedure that selects operating points, hand-picked or learned, from the correctness of their answers on calibration prompts. Redline keeps the reference-relative risk, the joint probability that the reference answers correctly and a candidate configuration does not, within a user-chosen budget with high probability, and deploys the fastest configuration that passes. It speeds up math at a smaller risk budget than code in both model families, and at a budget of ten percent it deploys a LLaDA2 math configuration that commits over a third more tokens in each forward. It also applies without modification to the acceptance rule of speculative decoding and to weight quantization. On the same calibration data, Redline stays within its stated failure probability, whereas each tolerance of a mean-accuracy rule either gains less speed for some model and task or exceeds the risk budget far more often for another. Code is available at https://github.com/js-lee-AI/Redline.
Chinese Translation
块扩散语言模型在精心挑选的操作点上进行服务，例如接受阈值、缓冲深度、调度、检查点和精度，每个点都是根据其平均基准准确性选择的。然而，平均值并不能告诉操作员更快的配置在慢速配置正确回答的提示上失败的频率。在服务引擎及其解码轨迹中，默认的提交规则已经提交每个完全解析的块，静态跳过规则几乎捕获了所有可以节省的计算，而在引擎解码目标上的自蒸馏则在不改变准确性的情况下提高了速度。更大的加速来自于较低的阈值，这会提交仍然不确定的标记。因此，我们提出了 Redline，一种有限样本程序，从其在校准提示上的答案正确性中选择操作点，无论是手动选择还是学习得到的。Redline 保持参考相对风险，即参考答案正确且候选配置不正确的联合概率，在用户选择的预算内以高概率保持，并部署通过的最快配置。在两个模型系列中，它以比代码更小的风险预算加速数学运算，并且在十个百分点的预算下，它部署了一个 LLaDA2 数学配置，在每次前向传播中提交超过三分之一的标记。它还可以不经修改地应用于投机解码的接受规则和权重量化。在相同的校准数据上，Redline 保持在其声明的失败概率范围内，而每个平均准确性规则的容忍度要么对某些模型和任务获得较少的速度，要么对另一些模型和任务更频繁地超出风险预算。代码可在 https://github.com/js-lee-AI/Redline 获取。
cs.CL / 133 / 2609.33899
NSV-Shift: A Contrastive Benchmark for Non-Speech Vocalization Understanding and Response Adaptation in Speech-to-Speech Models
NSV-Shift：用于评估语音到语音模型理解非语音发声及其响应适应性的对比基准
Abstract
We introduce NSV-Shift, a contrastive benchmark for evaluating whether speech-to-speech models can understand non-speech vocalizations (NSVs) and adapt their responses accordingly. Each pair contains two conversations with identical lexical content that differ only in the NSV embedded in the final turn. Our pilot contains 22 human-verified pairs (44 audio conditions) and evaluates five models on NSV perception, emotion understanding, and response adaptation. Results show that models generally perform better at detecting NSVs than at interpreting their fine-grained emotional meaning or producing appropriately differentiated responses. The data construction pipeline, dataset, and evaluation pipeline are publicly available at https://github.com/ChenzwNina/nsv-construction.
Chinese Translation
我们介绍了NSV-Shift，这是一个对比基准，用于评估语音到语音模型是否能够理解非语音发声（NSVs）并相应地调整其响应。每对样本包含两个对话，它们的词汇内容相同，仅在最后一轮中嵌入的NSV不同。我们的初步研究包含22对经过人工验证的样本（44个音频条件），并评估五个模型在NSV感知、情感理解和响应适应方面的表现。结果表明，模型在检测NSVs方面的表现通常优于对其细微情感意义的解释或产生适当区分的响应。数据构建流程、数据集和评估流程已公开发布在 https://github.com/ChenzwNina/nsv-construction。
cs.CL / 134 / 2609.33905
SlopBench: How Well Can We Rank Language Models by Slop? A Multi-Domain Benchmark of Repetitive AI Writing
SlopBench：我们能通过 Slop 多好地对语言模型进行排名？一项关于重复性 AI 写作的多领域基准测试
Abstract
SlopBench asks which models produce the stiff, repetitive prose readers call AI slop, a question detectors leave open once they have classified a text as machine-written. We evaluated eighteen models on 112 hand-written tasks in email, social posts, essays, and workplace chat, sampling each model on each task up to ten times, for 19,928 outputs in all. SlopBench scores four surface behaviors a reader can check by hand: length against the word band each task specifies, opener repetition across a model's own samples of one task, and paragraph rhythm and fixed lexical constructions against pre-ChatGPT human corpora. Under one fixed weighting, Kimi K2.6 scores lowest at 21.1 and Mistral Large highest at 40.6. Across 500 random reweightings Kimi has the lowest score in 58 percent of draws and Mistral the highest in 97 percent. No draw preserves the full order of the eighteen, and a scenario bootstrap leaves exactly one of those ranks unambiguous. We ran three further checks on that middle order: a crowd arena, an AI detector, and lexical diversity. None of them confirmed the order. We therefore report the four behaviors separately and treat the composite as one weighting among many, and we release the prompts, outputs, reference statistics, and scoring code.
Chinese Translation
SlopBench 探讨了哪些模型产生了读者称之为 AI slop 的生硬、重复的散文，这个问题在检测器将文本分类为机器写作后仍然悬而未决。我们对十八个模型在电子邮件、社交帖子、论文和工作聊天等 112 个手写任务上进行了评估，每个模型在每个任务上最多采样十次，共生成 19,928 个输出。SlopBench 评分考虑了读者可以手动检查的四种表面行为：任务指定的字数范围内的长度、模型自身在同一任务样本中的开头重复、段落节奏以及与 ChatGPT 之前的人类语料库相比的固定词汇结构。在一种固定权重下，Kimi K2.6 的得分最低，为 21.1，而 Mistral Large 的得分最高，为 40.6。在 500 次随机重加权中，Kimi 在 58% 的抽样中得分最低，而 Mistral 在 97% 的抽样中得分最高。没有一次抽样能够保留这十八个模型的完整顺序，而情境引导法则使得其中恰好有一个排名变得明确。我们对这个中间顺序进行了三次进一步检查：一个众包评估、一个 AI 检测器和词汇多样性。它们都没有确认这个顺序。因此，我们分别报告这四种行为，并将综合评分视为众多权重中的一种，同时我们发布了提示、输出、参考统计和评分代码。
cs.CL / 135 / 2609.33923
Quantization Error Is Spectrally Flat: A Single Random Probe Is a Calibrated, Data-Free Sensitivity Estimator, with Application to Budget-Targeted Mixed-Precision Quantization
量化误差在频谱上是平坦的：单个随机探针是经过校准的、无数据的灵敏度估计器，应用于预算导向的混合精度量化
Abstract
A single random Gaussian probe gives an unbiased estimate of the squared Frobenius norm of a layer's quantization error. The estimator is well-behaved because round-to-nearest error is spectrally flat. Across 1,683 tensors from a 35B MoE and a 9B dense model, effective dimensionality is 0.93 to 0.96 times the i.i.d. noise value of the same shape, and on the MoE the median is unchanged from 2-bit to 8-bit. The probe coefficient of variation is predictable from tensor shape. One probe measures per-tensor sensitivity to within 4 to 7%; twenty probes reach 1.3 to 1.4%.RAM applies the propagated form of this estimator to budget-targeted mixed-precision quantization with no calibration data. Gaussian probes carrying the network's own input statistics score every tensor at six bit-widths. A knapsack solver allocates bits under an exact byte budget, with guardrails against catastrophic 2-bit assignments. One probe pass serves any budget. Isolated and propagated scores rank tensors independently on Qwen3.5-35B-A3B (Spearman -0.01), yet the propagated probe rank-correlates 0.81 to 0.83 with the GPTQ layer objective from real activations, while the isolated estimator is uncorrelated with it. That objective is the wrong allocation target: at matched bytes on Qwen3.8-27B, a block-output probe beats a vendor IQ3_M mix and an oracle that allocates from the real-activation objective.On Qwen3-8B the propagated probe ties HAWQ-V2 at matched bytes. Across seven architectures from 8B to 122B, with probe timing up to a 400B model in nine minutes on one workstation, RAM reaches 3.5 to 13.6% lower median WikiText-2 perplexity than size-comparable uniform 4-bit builds on the tested MoE models
Chinese Translation
单个随机高斯探针提供了层量化误差平方弗罗贝尼乌斯范数的无偏估计。该估计器表现良好，因为最近取整误差在频谱上是平坦的。在来自35B MoE和9B稠密模型的1,683个张量中，有效维度为相同形状的独立同分布噪声值的0.93到0.96倍，并且在MoE中，媒体从2位到8位保持不变。探针的变异系数可以根据张量形状进行预测。一个探针测量每个张量的灵敏度，误差在4%到7%之间；二十个探针的误差达到1.3%到1.4%。RAM将该估计器的传播形式应用于预算导向的混合精度量化，无需校准数据。携带网络自身输入统计的高斯探针在六种位宽下对每个张量进行评分。一个背包求解器在精确的字节预算下分配位，并对灾难性的2位分配设置了保护措施。一个探针的传递可以满足任何预算。孤立和传播的评分在Qwen3.5-35B-A3B上独立对张量进行排名（斯皮尔曼相关系数为-0.01），然而传播探针的排名与真实激活的GPTQ层目标相关系数为0.81到0.83，而孤立估计器与之不相关。该目标是错误的分配目标：在Qwen3.8-27B上匹配字节时，块输出探针优于供应商IQ3_M混合和从真实激活目标分配的oracle。在Qwen3-8B上，传播探针在匹配字节时与HAWQ-V2持平。在从8B到122B的七种架构中，探针计时在一台工作站上对一个400B模型的处理时间为九分钟，RAM在测试的MoE模型上实现了比大小相当的均匀4位构建低3.5%到13.6%的中位WikiText-2困惑度。
cs.CL / 136 / 2609.33947
Simple Diffusion Language Models Are More Effective Few-Step Generators Than Reported
简单扩散语言模型在少步生成中比报告的更有效
Abstract
Diffusion language models (DLMs) promise fast parallel generation, yet high-quality samples often require large number of refinement steps, which diminishes their advantage in practice. This has led to massive interest in and rapid development of new methods for effective few-step generation. We show that much of the supposed quality gap at few steps can instead arise from a suboptimally configured sampler. Modest sampler sharpening, without any model retraining, enables a couple years old masked DLM to rival supposedly far improved successors. This differently sampled DLM in fact achieves lower generative perplexity in just 16 steps than what its standard sampler obtains with 1024, while improving both judged quality and semantic diversity. We further show that conventional per-output metrics can fundamentally obscure these gains, since any optimal trade-off between two such metrics can be attained by a generator supported on at most two outputs. We subsequently introduce GroupEval, which separately evaluates quality and across-output semantic diversity, and offers fresh insights including uncovering how 1.5-4.7x perplexity gains of a distilled model yield no corresponding quality gain. Finally, we explain why sharpening helps: parallel unmasking destroys dependencies among simultaneously generated tokens, creating a gap between prediction and generation. We prove that pervasive temperature choice of one is generically suboptimal under parallel sampling even for an exact denoiser, and that worse predictions can yield better samples. Through these results, we argue for a broader evaluation principle of treating the deployed generator as the object of comparison, benchmarking it against tuned baselines, and assessing quality and diversity jointly and with more human-aligned measures.
Chinese Translation
扩散语言模型（DLMs）承诺实现快速并行生成，然而高质量样本通常需要大量的精炼步骤，这在实践中削弱了其优势。这导致了对有效少步生成的新方法的巨大兴趣和快速发展。我们表明，许多所谓的少步质量差距实际上可能源于配置不佳的采样器。适度的采样器优化，无需任何模型重训练，使得一款几年前的掩蔽DLM能够与所谓的显著改进的后继者相媲美。实际上，这种不同采样的DLM在仅16步内实现了比其标准采样器在1024步内获得的更低生成困惑度，同时提高了评估质量和语义多样性。我们进一步表明，传统的每输出指标可能根本上掩盖了这些收益，因为任何两个此类指标之间的最佳权衡都可以通过一个最多支持两个输出的生成器来实现。随后，我们引入了GroupEval，它分别评估质量和跨输出的语义多样性，并提供了新的见解，包括揭示1.5-4.7倍的困惑度收益的蒸馏模型并未带来相应的质量提升。最后，我们解释了为什么优化有助于：并行解掩蔽破坏了同时生成的标记之间的依赖关系，从而在预测和生成之间创造了差距。我们证明，在并行采样下，即使对于一个精确的去噪器，普遍选择的温度为1在一般情况下也是次优的，而更差的预测可以产生更好的样本。通过这些结果，我们主张采用更广泛的评估原则，将部署的生成器视为比较对象，与调优的基线进行基准测试，并共同评估质量和多样性，以及使用更符合人类的度量标准。
cs.CL / 137 / 2609.33970
On the Token Value Inequality in Efficient Reasoning
高效推理中的令牌价值不平等
Abstract
Chain-of-Thought reasoning has enabled large language models to achieve substantial performance gains on complex tasks. However, these gains come at the cost of dramatically increased token consumption. This raises a fundamental question: is every token in the reasoning trace equally valuable? We present a diagnostic and optimization framework grounded in a key empirical finding: the value of tokens within a CoT reasoning sequence is highly non-uniform, and this non-uniformity can be effectively characterized by token-level log probability signals. We show that normalized log probability helps distinguish core tokens, which carry structural and decisive reasoning content, from redundant tokens, which are exploratory, low-confidence filler that contributes less directly to the final answer. Building on these findings, we formulate the TokenProbe framework around two empirical findings and one claim: findings identify token value inequality first and then establish TokenProbe as a core-token proxy, and the claim introduces an efficient GRPO objective positing that selectively compressing redundant tokens can yield Pareto improvements in the accuracy-token efficiency space. Empirically, our method preserves reasoning quality while reducing the token usage by 76% of the baseline. Under matched reasoning-length budgets, we show that it can even outperform strong flagship baselines like Gemini-3.1-Pro. Homepage: https://runjia.tech/tokenprobe/.
Chinese Translation
链式思维推理使大型语言模型在复杂任务上实现了显著的性能提升。然而，这些提升伴随着令牌消耗的剧增。这引发了一个根本性的问题：推理轨迹中的每个令牌是否同样有价值？我们提出了一个诊断和优化框架，基于一个关键的实证发现：在链式思维推理序列中的令牌价值高度不均匀，这种不均匀性可以通过令牌级的对数概率信号有效表征。我们展示了标准化对数概率有助于区分核心令牌，这些令牌承载着结构性和决定性的推理内容，与冗余令牌相对，后者是探索性的、低置信度的填充内容，对最终答案的贡献较小。基于这些发现，我们围绕两个实证发现和一个主张构建了TokenProbe框架：首先识别令牌价值不平等的发现，然后将TokenProbe确立为核心令牌的代理，而主张则引入了一种高效的GRPO目标，认为选择性压缩冗余令牌可以在准确性-令牌效率空间中实现帕累托改进。从实证上看，我们的方法在减少令牌使用量方面比基线减少了76%的同时保持了推理质量。在匹配的推理长度预算下，我们展示了它甚至可以超越像Gemini-3.1-Pro这样的强大基线。主页：https://runjia.tech/tokenprobe/
cs.CL / 138 / 2609.33971
Do System One Decisions Add Up? A Study of Probabilistic Coherence
系统一决策是否能够相加？关于概率一致性的研究
Abstract
A decision model can give probabilities that sum to one for every question yet disagree with itself when the same decision is broken into smaller steps. We study this form of probabilistic coherence in Jev and the English Laya checkpoint, using 2,500 matched examples per system across TREC, CLINC150, and MASSIVE. Across 72,000 classification questions, we compare direct fine-label predictions with broad-category probabilities and predictions reconstructed through those categories. Both systems show substantial disagreement: mean category-level total variation ranges from 0.219 to 0.349 for Jev and from 0.424 to 0.689 for Laya, on a scale where zero means exact agreement. The consequences differ sharply. On CLINC150, reconstruction reduces Jev's accuracy by 22.9 percentage points (paired 95% bootstrap interval: [-24.9, -20.9]) and improves Laya's by 21.3 points ([18.0, 24.5]). The same directions hold across all three datasets, with all six unadjusted accuracy-change intervals excluding zero. Improved accuracy can also accompany less reliable confidence: on MASSIVE, Laya gains 9.2 accuracy points while its expected calibration error rises from 0.046 to 0.124. Error analysis identifies both broad-category mistakes and within-category confusions. These findings show why decision systems need joint evaluation of accuracy, confidence calibration, and probability coherence in the workflow used by an application.
Chinese Translation
一个决策模型可以为每个问题提供概率总和为一的结果，但当同一决策被分解为更小的步骤时，可能会出现自我矛盾。我们研究了在Jev和英语Laya检查点中这种形式的概率一致性，使用了来自TREC、CLINC150和MASSIVE的每个系统2500个匹配示例。在72000个分类问题中，我们比较了直接的细分类标签预测与通过这些类别重建的广泛类别概率和预测。两个系统之间存在显著的不一致：Jev的平均类别级总变异范围为0.219到0.349，而Laya的范围为0.424到0.689，零表示完全一致。其后果截然不同。在CLINC150上，重建使Jev的准确率下降了22.9个百分点（配对的95%自助法区间：[-24.9, -20.9]），而Laya的准确率提高了21.3个百分点（[18.0, 24.5]）。在所有三个数据集中，这种趋势一致，所有六个未经调整的准确率变化区间均不包括零。提高的准确性也可能伴随不那么可靠的置信度：在MASSIVE上，Laya的准确率提高了9.2个百分点，同时其预期校准误差从0.046上升到0.124。错误分析识别了广泛类别错误和类别内混淆。这些发现表明，决策系统在应用程序的工作流程中需要对准确性、置信度校准和概率一致性进行联合评估。
cs.CL / 139 / 2609.33974
Beyond Solo and Consistency: Vindicating Multi-Agent Debate via Conditional Progressive Pruning
超越单一代理与一致性：通过条件渐进剪枝验证多智能体辩论
Abstract
Large Language Model (LLM) based Multi-Agent Debate (MAD) is one of the most effective test time scaling techniques. Through multi-round communication, agents complement each other in knowledge and reasoning and solve tasks that no single member can solve. However, existing MAD frameworks fail to beat strong Single Agent and Consistency-based baselines under the same strict cost limit, which shakes the foundation of the MAD field. We propose Conditional Progressive Pruning (CPP), a lightweight pruning framework that fully exploits multi-round MAD. CPP outperforms all existing MAD frameworks on multiple dominated benchmarks. It is also the first to fully outperform consistency methods. Our code, detailed agent interaction records will be released soon.
Chinese Translation
基于大型语言模型（LLM）的多智能体辩论（MAD）是最有效的测试时间扩展技术之一。通过多轮沟通，智能体在知识和推理上互为补充，解决单个成员无法解决的任务。然而，现有的MAD框架在相同严格成本限制下未能超越强大的单一智能体和基于一致性的方法，这动摇了MAD领域的基础。我们提出了条件渐进剪枝（CPP），这是一种轻量级的剪枝框架，充分利用多轮MAD。CPP在多个主导基准测试中超越了所有现有的MAD框架。它也是首个完全超越一致性方法的框架。我们的代码和详细的智能体交互记录将很快发布。
cs.CL / 140 / 2609.33983
High-Level Text Preprocessing for Semantic Similarity Analysis of Discursive Texts: A Framework and Empirical Demonstration
用于话语文本语义相似性分析的高级文本预处理：框架与实证演示
Abstract
Semantic Textual Similarity (STS) methods assume that a document's lexical content faithfully represents what it asserts. This assumption fails for discursive documents that discuss, compare, critique, and contextualize other positions in the process of articulating their own. The result is semantic diffusion: similarity scores between documents are inflated by vocabulary acquired through discursive engagement rather than substantive alignment. Standard Natural Language Processing (NLP) preprocessing (tokenization, stopword removal, stemming, lemmatization) cannot address this problem because it operates at the lexical level, treating all content identically regardless of its discursive function. This paper introduces high-level text preprocessing: a systematic, rule-based intervention applied before the standard preprocessing pipeline to isolate each document's actual claim from its discursive structure. We propose 12 rules, each with an explicit rationale, and demonstrate their effect on an encyclopedic philosophical corpus: three entries from the Stanford Encyclopedia of Philosophy (virtue ethics, deontological ethics, and consequentialism). A three-phase experiment using eight Transformer-based STS models shows that preprocessing reduces centroid cosine similarity scores across all three theory pairs, with 23 of 24 model-pair comparisons showing the expected decrease and cross-model agreement ranging from 7-1 to 8-0. We introduce the semantic diffusion index (SDI), a per-document metric for assessing the semantic reorientation between a document's raw and high-level preprocessed representations. Although the framework is demonstrated using philosophical texts, it potentially addresses a domain-agnostic problem applicable to legal texts, policy documents, academic articles, and any genre in which a discursive approach introduces vocabulary from positions the document does not endorse.
Chinese Translation
语义文本相似性（STS）方法假设文档的词汇内容忠实地代表其所主张的内容。然而，这一假设在讨论、比较、批评和将其他立场置于背景中的话语性文档中并不成立，这些文档在阐述自身观点的过程中会导致语义扩散：文档之间的相似性得分因通过话语参与获得的词汇而被夸大，而非实质性的一致性。标准的自然语言处理（NLP）预处理（如分词、停用词移除、词干提取、词形还原）无法解决这一问题，因为它在词汇层面上操作，平等对待所有内容，而不考虑其话语功能。本文提出了高级文本预处理：一种系统的、基于规则的干预措施，应用于标准预处理流程之前，以将每个文档的实际主张与其话语结构分离。我们提出了12条规则，每条规则都有明确的理由，并展示了它们在一个百科全书式的哲学语料库中的效果：来自斯坦福哲学百科全书的三个条目（美德伦理学、义务论伦理学和结果主义）。使用八种基于Transformer的STS模型进行的三阶段实验表明，预处理降低了所有三个理论对的中心余弦相似性得分，其中24个模型对比较中有23个显示出预期的下降，跨模型一致性范围为7-1到8-0。我们引入了语义扩散指数（SDI），这是一个用于评估文档原始表示与高级预处理表示之间语义重定向的每文档指标。尽管该框架是通过哲学文本进行演示的，但它可能解决一个适用于法律文本、政策文件、学术文章以及任何引入不被文档所支持的立场词汇的话语方法的领域无关问题。
cs.CL / 141 / 2609.33987
Opera: A Verbal Critic Framework for Long-horizon Coding Agents
Opera：一种用于长时间编码代理的语言评论框架
Abstract
Long-horizon coding agents need timely corrections, yet feedback can be ineffective or even harmful when it misjudges ongoing work or fails to address the underlying problem. Existing critics focus on evaluating trajectories and generating feedback, but rarely track what happens after feedback is delivered. We present Opera, a verbal critic framework that treats each correction as a persistent note, followed until the diagnosed problem is resolved. Opera decides when to review through periodic and event-driven triggers, diagnoses issues with typed operators, audits feedback against visible evidence before delivery, and tracks the agent's subsequent actions to distinguish mere compliance from actual resolution. As a test-time critic, Opera improves the resolve rate of non-critic agents by up to 12.4, 15.0, and 8.9 percentage points on Terminal-Bench 2.1, a SWE-Bench Pro subset, and DeepSWE v1.1, respectively, across four policy models, and achieves the highest mean resolve rate among competitive critic baselines on all three benchmarks, and also improves policy models when the policy critiques itself. Beyond inference, Opera-guided rollouts provide approximately on-policy training data: fine-tuning Qwen3.5-9B on them improves its resolve rate on held-out SWE-Bench Pro repositories by 10.2 percentage points without a critic at inference time, matching fine-tuning on rollouts from a stronger model, while preserving its performance when switching harness, i.e., from Openhands to Terminus-2, which the latter substantially degrades.
Chinese Translation
长时间编码代理需要及时的纠正，然而当反馈错误地判断正在进行的工作或未能解决根本问题时，反馈可能会无效甚至有害。现有的评论者主要集中在评估轨迹和生成反馈上，但很少跟踪反馈传递后的情况。我们提出了Opera，一个将每次纠正视为持续记录的语言评论框架，直到诊断出的问题得到解决。Opera通过定期和事件驱动的触发器决定何时进行复审，使用类型化操作符诊断问题，在反馈传递前审核可见证据，并跟踪代理的后续行动，以区分单纯的遵从与实际的解决。在测试阶段，Opera在Terminal-Bench 2.1、SWE-Bench Pro子集和DeepSWE v1.1等四个策略模型上，分别提高了非评论代理的解决率，最多达到12.4、15.0和8.9个百分点，并在所有三个基准测试中实现了竞争性评论基线中最高的平均解决率，同时在策略模型自我评论时也改善了策略模型的表现。除了推理，Opera指导的回滚提供了大约符合策略的训练数据：在这些数据上微调Qwen3.5-9B，使其在保留的SWE-Bench Pro库上的解决率提高了10.2个百分点，而在推理时没有评论者，这与从更强模型的回滚中微调的效果相匹配，同时在切换环境时保持其性能，即从Openhands切换到Terminus-2，后者会显著降低性能。
cs.CL / 142 / 2609.33989
RewardExplainer: Learning Reward Model Explanations from Counterfactual Preference Feedback
RewardExplainer：从反事实偏好反馈中学习奖励模型解释
Abstract
Reward models (RMs) are a key component of large language model post-training, providing reward signals for subsequent reinforcement learning. However, conventional discriminative RMs typically output only scalar scores, making it difficult to identify the response behaviors associated with their scoring decisions. Existing interpretation methods often rely on predefined high-level attributes and require repeated counterfactual interventions for each response pair to validate candidate explanations, lacking a closed-loop mechanism that uses RMs' feedback to train a reusable explainer. To address this, we propose RewardExplainer, a framework that obtains feedback from the target reward model through counterfactual rewriting and uses this feedback to further optimize the explainer. RewardExplainer generates open-ended, atomic, and intervenable natural-language scoring mechanisms, making explanations more concrete, readable, and actionable. It further converts counterfactual feedback into preference supervision, enabling the explainer to more faithfully capture the target RM's scoring preferences and sensitive behaviors than single-pass generation. Extensive experiments across multiple target RMs and explainer backbones show consistent improvements. Beyond interpretation, we use the generated mechanisms to identify potential bias patterns and construct targeted debiasing data for fine-tuning the reward model, improving robustness on reward-hacking benchmarks.
Chinese Translation
奖励模型（RMs）是大型语言模型后训练的关键组成部分，为后续的强化学习提供奖励信号。然而，传统的判别性奖励模型通常仅输出标量分数，这使得识别与其评分决策相关的响应行为变得困难。现有的解释方法往往依赖于预定义的高层属性，并且需要对每对响应进行重复的反事实干预以验证候选解释，缺乏利用奖励模型反馈训练可重用解释器的闭环机制。为了解决这一问题，我们提出了RewardExplainer，一个通过反事实重写从目标奖励模型获取反馈并利用这些反馈进一步优化解释器的框架。RewardExplainer生成开放式、原子性和可干预的自然语言评分机制，使解释更加具体、可读和可操作。它进一步将反事实反馈转换为偏好监督，使得解释器能够比单次生成更忠实地捕捉目标奖励模型的评分偏好和敏感行为。在多个目标奖励模型和解释器骨干网络上的广泛实验显示出一致的改进。除了解释之外，我们还使用生成的机制识别潜在的偏见模式，并构建针对性去偏数据以微调奖励模型，从而提高在奖励黑客基准上的鲁棒性。
cs.CL / 143 / 2609.34033
Faithful Activation Verbalization: Reducing Hallucinations in LLM Representation Interpretation
忠实激活表述：减少大型语言模型表示解释中的幻觉现象
Abstract
Activation verbalization methods such as Activation Oracle and Natural Language Autoencoders decode hidden representations of large language models into human-readable natural language. However, existing methods can produce incomplete or hallucinated descriptions, making their activation verbalizations difficult to trust and use reliably in practice. To this end, we introduce AVPO, a two-stage framework that first reconstructs source text from a hidden activation and then evaluates the resulting text with a separate frozen question-answering model, yielding an explicit and inspectable intermediate readout. We further optimize the inverter with direct preference optimization (DPO), using rewards that capture both semantic recoverability and lexical fidelity. Across six text families, AVPO improves gist- and detail-level information recovery over the strongest baseline by up to 17.1 and 9.3 percentage points, respectively. Crucially, the gains arise from preference optimization rather than fine-tuning on selected reconstructions alone, enabling compact cross-model inverters to surpass donor-matched question-conditioned verbalizers while improving both semantic recoverability and lexical fidelity. Moreover, out-of-distribution case study shows that AVPO better recovers high-level semantics while fabricating fewer details.
Chinese Translation
激活表述方法如 Activation Oracle 和自然语言自编码器（Natural Language Autoencoders）将大型语言模型的隐藏表示解码为人类可读的自然语言。然而，现有方法可能产生不完整或虚构的描述，使得其激活表述难以被信任和可靠使用。为此，我们提出了 AVPO，一种两阶段框架，首先从隐藏激活重构源文本，然后使用一个独立且冻结的问答模型对生成文本进行评估，从而产生明确且可检查的中间输出。我们进一步通过直接偏好优化（Direct Preference Optimization, DPO）优化逆变器，利用同时捕捉语义可恢复性和词汇忠实度的奖励信号。在六个文本类别上，AVPO在关键信息和细节信息恢复方面分别较最强基线提升了最高17.1和9.3个百分点。关键的是，这些提升源自偏好优化，而非仅对选定重构进行微调，使得紧凑的跨模型逆变器能够超越与捐赠模型匹配的基于条件问答的表述器，同时提升语义可恢复性和词汇忠实度。此外，分布外案例研究表明，AVPO在恢复高层语义的同时，虚构细节更少。
cs.CL / 144 / 2609.34065
Who Gets a Token, and What Does It Carry? Unequal Name Support and Concept Access in Large Language Models
谁获得了标记，它承载了什么？大型语言模型中的不平等名称支持与概念访问
Abstract
Names are personal identifiers, but they also carry social meaning and are widely used to evaluate how language models treat different people. Such evaluations typically assume that matched names are comparable model inputs. We show that this assumption often fails at the lexical interface: matched names are not necessarily matched inputs. Some names receive direct single-token access, while others are assembled from multiple subwords, creating unequal name-surface support. Across nearly half a million first names and 12 LLM-associated tokenizers, direct lexical access is highly selective, model dependent, and uneven across race- and gender-associated name metadata. We introduce NameTrace, a model-native, fine-grained, pre-behavioral framework for measuring whether unequal name-surface support remains a vocabulary property or becomes visible in task-relevant internal representations. NameTrace measures concept accessibility from the model's own probabilities over task-specific adjective axes with continuous task-aligned weights. On matched atomic and short-fragmented names within the same race/ethnicity--gender-associated strata, support predicts systematic differences in concept accessibility across fellowship, hiring, clinical assessment, and lending. These differences persist across all eight matched strata, extend across model families, and transfer to unseen names. Hidden-state interventions further show that the measured task directions have downstream leverage, shifting later constrained choices. Unequal lexical support is therefore demographically structured at the input and remains visible in task-relevant model computation. NameTrace makes lexical comparability measurable, supporting a broader principle: behavioral comparability begins with lexical comparability.
Chinese Translation
名称是个人标识符，但它们也承载着社会意义，并广泛用于评估语言模型如何对待不同的人。这类评估通常假设匹配的名称是可比较的模型输入。我们展示了这一假设在词汇接口上往往失败：匹配的名称不一定是匹配的输入。一些名称获得了直接的单标记访问，而其他名称则由多个子词组成，导致名称表面支持的不平等。在近五十万个名字和12个与大型语言模型（LLM）相关的标记器中，直接的词汇访问高度选择性、依赖于模型，并且在与种族和性别相关的名称元数据中不均匀。我们引入了NameTrace，这是一个模型原生的、细粒度的、行为前框架，用于测量不平等的名称表面支持是否仍然是词汇属性，或在任务相关的内部表示中变得可见。NameTrace通过模型在任务特定形容词轴上的概率测量概念可访问性，并具有连续的任务对齐权重。在同一种族/民族-性别相关的层次内，匹配的原子和短片段名称的支持预测了在奖学金、招聘、临床评估和贷款等方面的概念可访问性的系统差异。这些差异在所有八个匹配层次中持续存在，跨模型家族延伸，并转移到未见名称上。隐藏状态干预进一步表明，测量的任务方向具有下游杠杆作用，改变后续受限选择。因此，不平等的词汇支持在输入层面上是按人口结构化的，并在任务相关的模型计算中保持可见。NameTrace使词汇可比性可测量，支持一个更广泛的原则：行为可比性始于词汇可比性。
cs.CL / 145 / 2609.34092
Evaluating Machine Unlearning in ASR
评估自动语音识别中的机器遗忘
Abstract
Machine unlearning (MU) offers a path to compliance with "right to be forgotten" regulations. While MU has received increasing attention for speech tasks, it remains largely unexplored for Automatic Speech Recognition (ASR). In this work, we investigate whether existing MU algorithms and evaluation tools are suitable for ASR. We apply several MU techniques to an ASR model, evaluating privacy-utility trade-offs for single-subject unlearning, then assess the best algorithm under sequential and simultaneous unlearning. Results show that gradient ascent-based algorithms achieve strong utility-privacy trade-offs, whereas more complex approaches over-unlearn samples, making them easier to identify as unlearned. This suggests standard privacy evaluations based on simple Membership Inference attacks are insufficient to reliably assess unlearning success, motivating improved evaluation methods for MU in ASR. Finally, we show that both sequential and simultaneous unlearning yield worse privacy and utility than single-subject unlearning, underscoring the need for unlearning constructions better suited to these settings.
Chinese Translation
机器遗忘（Machine Unlearning, MU）为遵守“被遗忘权”法规提供了一条路径。尽管MU在语音任务中受到越来越多的关注，但在自动语音识别（Automatic Speech Recognition, ASR）领域仍然大多未被探索。在本研究中，我们调查了现有的MU算法和评估工具是否适用于ASR。我们将几种MU技术应用于ASR模型，评估单一主体遗忘的隐私-效用权衡，然后在顺序和同时遗忘的情况下评估最佳算法。结果表明，基于梯度上升的算法在隐私和效用的权衡上表现良好，而更复杂的方法则过度遗忘样本，使其更容易被识别为已遗忘。这表明，基于简单的成员推断攻击的标准隐私评估不足以可靠地评估遗忘的成功，促使我们改进ASR中MU的评估方法。最后，我们展示了顺序和同时遗忘的隐私和效用均低于单一主体遗忘，强调了需要更适合这些设置的遗忘构造。
cs.CL / 146 / 2609.34112
Unknown is not normal: separating language-model extraction from rule-based decision logic for clinical risk scores
未知并非正常：将语言模型提取与基于规则的决策逻辑分离以计算临床风险评分
Abstract
Large language models (LLMs) are increasingly used to compute clinical risk scores from free-text notes. Notes are often incomplete, and treating undocumented findings as normal can silently misclassify patients. We test whether separating three-state extraction (present, absent or unknown, by an LLM) from decision logic (deterministic code computing score bounds over unknown inputs) lets a system ask only questions that can change the decision. On 1,200 synthetic emergency cases across six calculators (HEART, CURB-65, qSOFA, PERC, Wells, Cockcroft-Gault), with a simulated clinician answering questions, we compared this bounds policy with asking for every missing input, a missing-equals-normal schema, and an end-to-end LLM agent (Claude Opus 5.5). With Claude Haiku 4.5 as extractor, the bounds policy matched ask-all accuracy (99.4% vs 99.4%) with half the questions (0.92 vs 1.78 per case) and no irrelevant ones. Treating missing as normal dropped accuracy to 91.2% and under-triaged 8.5% of patients (95% CI 7.1-10.2), and under-triage persisted under messy notes and a noisy clinician. The agent was equally accurate under ideal conditions (99.6%) but 9.5% of its questions were irrelevant; with a noisy clinician it was less accurate than the bounds policy (83.5% vs 87.0%, p<0.001) and committed prematurely in 2.7% of cases (bounds: 0%). A 9B local model as extractor reached oracle-level accuracy (99.8%). In 584 real case reports from MedCalc-Bench, only 52% contained enough information to determine the category (HEART 13%). Routing decisions through code that reasons explicitly about unknowns avoids premature commitment and irrelevant questions, halves the questions asked, and works with small local models.
Chinese Translation
大型语言模型（LLMs）越来越多地用于从自由文本笔记中计算临床风险评分。笔记通常是不完整的，将未记录的发现视为正常可能会悄然错误分类患者。我们测试了将三态提取（由LLM判断为存在、缺失或未知）与决策逻辑（计算未知输入的评分边界的确定性代码）分离是否能让系统仅提出能够改变决策的问题。在1200个合成急救案例中，涵盖六个计算器（HEART、CURB-65、qSOFA、PERC、Wells、Cockcroft-Gault），通过模拟临床医生回答问题，我们将这种边界策略与请求每个缺失输入、缺失等于正常的模式以及端到端LLM代理（Claude Opus 5.5）进行了比较。在使用Claude Haiku 4.5作为提取器的情况下，边界策略在提问数量减少一半（每个案例0.92个问题对比1.78个问题）且没有无关问题的情况下，达到了与请求所有问题相同的准确率（99.4%对99.4%）。将缺失视为正常使准确率下降至91.2%，并且错误分类了8.5%的患者（95% CI 7.1-10.2），在杂乱的笔记和嘈杂的临床医生下，错误分类现象依然存在。在理想条件下，代理的准确率同样为99.6%，但其9.5%的问题是无关的；在嘈杂的临床医生情况下，其准确率低于边界策略（83.5%对87.0%，p<0.001），并在2.7%的案例中过早做出决策（边界：0%）。作为提取器的9B本地模型达到了oracle级别的准确率（99.8%）。在来自MedCalc-Bench的584个真实病例报告中，仅52%包含足够的信息来确定类别（HEART为13%）。通过明确推理未知情况的代码进行路由决策可以避免过早承诺和无关问题，减少提问数量，并且适用于小型本地模型。
cs.CL / 147 / 2609.34125
Understanding Clinical Cognitive Dialogues Using Large Language Models
利用大型语言模型理解临床认知对话
Abstract
In-person cognitive assessment is both a test and an interaction. Clinicians explain tasks, repair misunderstandings, and adapt to patient responses, while patients may hesitate, seek clarification, or disengage. Yet clinical dialogue resources rarely label the interaction structure needed to study these behaviors at scale. We present an de-identified corpus of 33 cognitive assessment conversations with 8,250 utterances annotated for three speaker roles and 56 dialogue acts. We use this corpus to benchmark large language models on fine-grained dialogue-act classification and next-patient-utterance generation. We also test whether out-of-domain instruction data and explanation-augmented training transfer to this clinical setting. Instruction tuning produces the strongest patient-utterance reference matching and improves classification accuracy. Reasoning-aware fine-tuning produces the strongest classification results among the LLaMA-3.1-8B variants. However, even the best models struggle to separate closely related dialogue acts, showing that broad conversational intent is easier to recognize than fine-grained communicative function. The corpus and benchmark make interaction structure measurable in cognitive assessments and support follow-up work on conversational markers, clinician education, and carefully validated simulated patients. This work does not make diagnostic claims. Instead, it provides the data and evaluation framework needed to study these applications.
Chinese Translation
面对面的认知评估既是一种测试，也是一种互动。临床医生解释任务、修正误解并根据患者的反应进行调整，而患者可能会犹豫、寻求澄清或 disengage（脱离互动）。然而，临床对话资源很少标注出研究这些行为所需的互动结构。我们呈现了一个去标识化的语料库，包含33个认知评估对话，共8,250个发言，标注了三种发言者角色和56种对话行为。我们利用该语料库对大型语言模型在细粒度对话行为分类和下一个患者发言生成方面进行基准测试。我们还测试了域外指令数据和增强解释的训练是否能够转移到这一临床环境中。指令调优产生了最强的患者发言参考匹配，并提高了分类准确性。基于推理的微调在LLaMA-3.1-8B变体中产生了最强的分类结果。然而，即使是最好的模型也难以区分密切相关的对话行为，这表明广泛的对话意图比细粒度的交际功能更容易识别。该语料库和基准使认知评估中的互动结构可测量，并支持后续关于对话标记、临床医生教育和经过仔细验证的模拟患者的研究。这项工作并不做出诊断声明，而是提供了研究这些应用所需的数据和评估框架。
cs.CL / 148 / 2609.34138
Word Similarity Datasets for Indian Languages: Annotation and Baseline Systems
印度语言的词语相似性数据集：注释与基线系统
Abstract
With the advent of word representations, word similarity tasks are becoming increasing popular as an evaluation metric for the quality of the representations. In this paper, we present manually annotated monolingual word similarity datasets of six Indian languages - Urdu, Telugu, Marathi, Punjabi, Tamil and Gujarati. These languages are most spoken Indian languages worldwide after Hindi and Bengali. For the construction of these datasets, our approach relies on translation and re-annotation of word similarity datasets of English. We also present baseline scores for word representation models using state-of-the-art techniques for Urdu, Telugu and Marathi by evaluating them on newly created word similarity datasets.
Chinese Translation
随着词语表示的出现，词语相似性任务作为评估表示质量的指标变得越来越受欢迎。本文呈现了六种印度语言（乌尔都语、泰卢固语、马拉地语、旁遮普语、泰米尔语和古吉拉特语）的手动注释单语词语相似性数据集。这些语言是全球使用人数仅次于印地语和孟加拉语的最常用印度语言。为了构建这些数据集，我们的方法依赖于对英语词语相似性数据集的翻译和重新注释。我们还通过在新创建的词语相似性数据集上评估，展示了乌尔都语、泰卢固语和马拉地语的词语表示模型的基线得分，使用了最先进的技术。
cs.CL / 149 / 2609.34152
Quantitative Measurement of Language Distance among Closely Related Indo-European Languages Using Pretrained Language Models: A Case Study on the North Germanic Branch
基于预训练语言模型的紧密相关印欧语言间语言距离的定量测量：以北日耳曼语支为例
Abstract
Among closely related North Germanic languages, the quantification of language distance has traditionally relied on qualitative methods, lacking a unified multi-dimensional computational framework. Multilingual pretrained models based on the Transformer architecture can map texts from different languages into a shared vector space, enabling quantitative measurement of language distance. This paper focuses on the three North Germanic languages---Danish, Norwegian (Bokm\aa{}l), and Swedish---and proposes a three-metric quantitative framework based on pretrained language models: (1)~sentence-level semantic distance, computed as cosine similarity between LaBSE and mBERT encodings of parallel sentences; (2)~orthographic fragmentation rate, measuring subword tokenization efficiency when cross-applying monolingual BERT vocabularies to parallel texts; (3)~MLM predictability, comparing prediction confidence and entropy in masked language modeling using mBERT across languages. Using 150 trilingual parallel sentence triplets from the Tatoeba corpus as controlled samples, we obtain consistent distance rankings on two independent models: LaBSE: da--no $0.012 < $ no--sv $0.016 < $ da--sv $0.020$; mBERT: da--no $0.016 < $ no--sv $0.045 \approx $ da--sv $0.046$. This ranking is consistent with the historical linguistic conclusion that ``400 years of Danish rule over Norway (1380--1814) led to highly cognate written languages.'' The three metrics---semantic, orthographic, and predictability---converge on the same conclusion, providing a reproducible computational framework for the quantitative study of distance among closely related languages, extensible in principle to more branches of the Indo-European language family, pending validation on additional language groups.
Chinese Translation
在紧密相关的北日耳曼语言中，语言距离的量化传统上依赖于定性方法，缺乏统一的多维计算框架。基于Transformer架构的多语言预训练模型能够将不同语言的文本映射到共享的向量空间，从而实现语言距离的定量测量。本文聚焦于三种北日耳曼语言——丹麦语、挪威语（Bokmål）和瑞典语——并提出了一种基于预训练语言模型的三指标定量框架：(1) 句子级语义距离，通过计算LaBSE和mBERT对平行句子的编码的余弦相似度来得出；(2) 正字法碎片化率，衡量在将单语BERT词汇应用于平行文本时的子词标记效率；(3) MLM可预测性，比较在使用mBERT进行掩码语言建模时的预测置信度和熵。使用来自Tatoeba语料库的150个三语平行句子三元组作为对照样本，我们在两个独立模型上获得了一致的距离排名：LaBSE: da--no $0.012 < $ no--sv $0.016 < $ da--sv $0.020$; mBERT: da--no $0.016 < $ no--sv $0.045 ext{ 近似于 } da--sv $0.046$。这一排名与历史语言学结论一致，即“400年的丹麦统治挪威（1380--1814）导致了高度同源的书面语言。”这三项指标——语义、正字法和可预测性——得出了相同的结论，为紧密相关语言间距离的定量研究提供了可重复的计算框架，原则上可扩展至印欧语言家族的更多分支，待在其他语言群体上进行验证。
cs.CL / 150 / 2609.34158
Toward a Graded Measure of Belief Stability in Large Language Models
朝着大语言模型中信念稳定性的分级测量
Abstract
Large language models (LLMs) increasingly mediate how people access and reason with information, yet factual reliability is usually evaluated one judgment at a time. We introduce graded belief stability, a relational measure of how well a belief persists within an LLM's broader belief system. Unlike individual belief probability, it asks whether support for a claim persists when that claim is considered alongside the model's other epistemic commitments. We operationalize this idea with a Direct Conditional estimator that uses internal model representations to estimate conditional belief probabilities. Across 12 LLMs and three domains, lower-stability beliefs exhibit greater mean behavioral movement under conversational challenge in 83.3% of model-domain settings after matching on individual belief probability. Graded belief stability therefore extends reliability assessment beyond how strongly an LLM supports a claim to how robustly that belief is supported within its broader system of beliefs.
Chinese Translation
大语言模型（LLMs）越来越多地调节人们获取和推理信息的方式，但事实可靠性通常是逐个判断进行评估的。我们引入了分级信念稳定性，这是一种关系性测量，旨在评估某一信念在LLM更广泛信念体系中的持久性。与单一信念概率不同，它关注的是当某一主张与模型的其他认识承诺一起考虑时，该主张的支持是否依然存在。我们通过一种直接条件估计器（Direct Conditional estimator）来实现这一概念，该估计器利用内部模型表示来估计条件信念概率。在12个LLM和三个领域的研究中，经过个体信念概率匹配后，低稳定性信念在83.3%的模型-领域设置中表现出在对话挑战下更大的平均行为变化。因此，分级信念稳定性将可靠性评估的范围从LLM对某一主张的支持强度扩展到该信念在其更广泛信念体系中受到的支持程度。
cs.CL / 151 / 2609.34187
LLMs are not stochastic parrots: Evidence for meaning-mediated abstraction from conlang-like tasks
大型语言模型并非随机鹦鹉：来自类人工语言任务的意义中介抽象证据
Abstract
The strong version of the stochastic parrot argument claims that, although large language models (LLMs) may exceed rote regurgitation, they cannot move beyond statistical pattern matching into abstraction or reasoning, remaining ontologically near the lower bound of pattern reuse despite producing alluringly fluent text. We test this hypothesis using conlang-like tasks. Several LLMs are given only natural-language descriptions of fictional languages that subvert prominent superficial patterns in training data by combining statistically uncommon and unattested features. Crucially, no example outputs are given. We argue that if the models exhibit rule-following behaviour, they cannot be relying solely on superficial statistical patterns; such patterns often work against the correct output. Instead, successful performance requires representations of the constraints specified in the prompt. Across three complementary task families, models systematically move in the meaning-predicted direction: they distinguish prompt exposure from instructed use, alter semantic relationships in response to novel constraints, and sometimes produce exact matches to complex translation answer keys. Although performance varies across the spectrum of models used, these results provide evidence for meaning-mediated abstraction in LLMs and refute the strong stochastic parrot hypothesis. Our work shows that, under appropriate architectural and contextual constraints, statistical learning can produce meaning-mediated abstractions, although generation remains strongly constrained by superficial plausibility. We discuss implications for model development and for understanding how increasingly abstract representations may emerge from plausible-text-generation objectives.
Chinese Translation
随机鹦鹉论的强版本主张，尽管大型语言模型（LLMs）可能超越机械重复，但它们无法超越统计模式匹配，进入抽象或推理，尽管生成引人入胜的流畅文本，仍然在模式重用的本体论下限附近。我们通过类人工语言任务测试这一假设。多个LLM仅接收对虚构语言的自然语言描述，这些语言通过结合统计上不常见和未经证实的特征，颠覆了训练数据中显著的表面模式。关键是，没有给出示例输出。我们认为，如果模型表现出遵循规则的行为，它们就不能仅依赖于表面统计模式；这些模式往往会对正确输出产生负面影响。相反，成功的表现需要对提示中指定约束的表征。在三类互补任务中，模型系统性地朝向意义预测的方向移动：它们区分提示暴露与指令使用，针对新约束改变语义关系，有时甚至生成与复杂翻译答案密钥的精确匹配。尽管不同模型的表现存在差异，这些结果提供了LLMs中意义中介抽象的证据，并驳斥了强随机鹦鹉假设。我们的研究表明，在适当的架构和上下文约束下，统计学习可以产生意义中介的抽象，尽管生成仍然受到表面合理性的强烈限制。我们讨论了模型开发的影响，以及如何理解越来越抽象的表征可能从合理文本生成目标中出现。
cs.CL / 152 / 2609.34225
USA: Update-aware SAM for Cross-domain On-Policy Disitllation of Language Agents
美国：面向跨领域在线策略蒸馏语言代理的更新感知SAM
Abstract
On-policy distillation instils multi-turn agentic reasoning through dense token-level supervision on the student's own trajectories, but a single domain saturates early, so further supervision has to be drawn from other domains. Multi-domain data mixing is the most direct way of incorporating them, at the cost of conflicts between their data distributions and of retraining the entire model whenever one domain is revised. Model merging avoids both by distilling every domain independently and fusing the resulting task vectors afterwards. We find instead that the benefit polarizes across domain pairs: on those exhibiting negative transfer, every merging operator we evaluate falls below the single-domain reference. We attribute this to cross-domain update coupling, where a substantial fraction of coordinates is updated comparably by both domains and a merge can therefore displace them by as much as their own updates. To overcome this limitation, we propose USA, which converts per-parameter update magnitudes measured during a brief warm-up into per-coordinate perturbation radii, reducing curvature precisely on the coordinates that carry most of the merging displacement. Experiments across mathematics, science and code at two student scales show USA strongest in all six transfer directions, ahead of the single-domain reference by more than four points on average, and reverse the negative transfer of the conflicting pairs.
Chinese Translation
在线策略蒸馏通过对学生自身轨迹进行密集的标记级监督，灌输多轮代理推理，但单一领域的饱和速度较快，因此必须从其他领域获取进一步的监督。多领域数据混合是整合它们的最直接方式，但代价是数据分布之间的冲突，以及每当一个领域被修订时必须重新训练整个模型。模型合并通过独立蒸馏每个领域并随后融合得到的任务向量，避免了这两者。我们发现，效益在领域对之间呈极化：在那些表现出负迁移的领域对上，我们评估的每个合并操作都低于单一领域的参考值。我们将此归因于跨领域更新耦合，其中相当一部分坐标由两个领域以相似方式更新，因此合并可能会将它们位移到与自身更新相当的程度。为克服这一限制，我们提出了USA，它将短暂热身期间测得的每个参数更新幅度转换为每个坐标的扰动半径，精确减少承载大部分合并位移的坐标的曲率。在数学、科学和代码的两个学生规模的实验中，USA在所有六个迁移方向上表现最强，平均领先单一领域参考值超过四个点，并逆转了冲突对的负迁移。
cs.CL / 153 / 2609.34234
MAS-OPD: On-Policy Distillation for Multi-agent Systems
MAS-OPD：用于多智能体系统的在线蒸馏
Abstract
Multi-agent systems (MAS) split a task across specialized roles and are promising on complex tasks, yet a prevailing approach relies on inference-time orchestration alone. General-purpose APIs are costly and hard to customize, while small models with role prompts rarely develop stable role competence or reliable collaboration, so post-training a MAS jointly is central. Most attempts use reinforcement learning, whose team-level reward leaves undetermined which step of which agent brought about the outcome, while local rewards need redesigning per task. On-policy distillation (OPD) gives token-level teacher supervision on trajectories the student samples, a denser signal needing no local reward, yet is underexplored for the interdependent agents of a MAS. Two difficulties arise: building complementary specialization from a judgement of which role a behavior belongs to while preserving the knowledge all roles need, and turning cross-agent collaborative information into supervision OPD can exploit. We present MAS-OPD, where Role-Advantage Specialization defines the role advantage as the difference between the teacher signals under target and non-target role conditions, and Privileged Attribution for Coordination attributes an interaction conflict to its source and supplies it to the teacher alone as privileged information. Extensive experiments on code and mathematics benchmarks show that MAS-OPD attains the highest mean score at both student scales and leads the agents to develop clearer role specialization and more effective collaborative behavior.
Chinese Translation
多智能体系统（MAS）将任务分配给专业角色，在复杂任务中展现出良好的前景，但现有的主流方法仅依赖于推理时的协调。通用API成本高且难以定制，而带有角色提示的小模型往往难以发展出稳定的角色能力或可靠的协作，因此对MAS进行联合后训练至关重要。大多数尝试使用强化学习，其团队级奖励无法明确指出是哪个智能体的哪个步骤导致了结果，而局部奖励需要针对每个任务重新设计。在线蒸馏（OPD）在学生采样的轨迹上提供了基于标记的教师监督，这是一种更密集的信号，无需局部奖励，但在MAS的相互依赖智能体中尚未得到充分探索。面临两个困难：从对行为所属角色的判断中建立互补的专业化，同时保留所有角色所需的知识，以及将跨智能体的协作信息转化为OPD可以利用的监督。我们提出了MAS-OPD，其中角色优势专业化将角色优势定义为目标角色条件和非目标角色条件下教师信号的差异，而协调的特权归因将交互冲突归因于其来源，并将其作为特权信息单独提供给教师。针对代码和数学基准的广泛实验表明，MAS-OPD在学生规模上获得了最高的平均得分，并促使智能体发展出更清晰的角色专业化和更有效的协作行为。
cs.CL / 154 / 2609.34240
Coherence-Aware Distributional Evaluation of Open-Ended Text Generation
考虑连贯性的开放式文本生成的分布式评估
Abstract
Existing metrics for open-ended text generation measure likelihood, lexical diversity, or distributional similarity in generic representation space, yet they can miss fundamental dimensions of quality. A prominent blind spot is global coherence: a generated passage may be locally fluent while remaining globally contradictory, causally inconsistent, or topically disconnected. Such failures can still preserve the token-level and lexical statistics that existing metrics rely on. We identify representation as a central bottleneck in detecting these failures and introduce CHORD (Coherence-aware Hidden-state Open-generation Reference Distance), a coherence-sensitive distributional metric. CHORD encodes generated and human-written corpora in the hidden-state space of a frozen LLM using a coherence-eliciting prompt, and compares the resulting distributions using MMD with an RBF kernel. To validate that the metric responds to coherence degradation but not generic textual change, we construct a counterfactual evaluation suite that pairs graded coherence-degrading perturbations with meaning-preserving controls. CHORD selectively detects relation, discourse, structural, and mixture failures that perplexity, entropy, MAUVE, FBD, and MMD-based baselines either miss or cannot separate from benign rewriting. Factorial ablations show that representation is the primary source of coherence sensitivity,while RBF-MMD improves sample efficiency once the relevant distinctions become visible. Larger backbones capture finer-grained distinctions, but coherence prompting improves selectivity only when the backbone can follow the prompt.On unconditional generation and prefix continuation, CHORD yields model rankings that strongly align with human judgments of whether outputs make sense and appear human-written. Together, these results establish representation design as central to reliable distributional evaluation.
Chinese Translation
现有的开放式文本生成评估指标主要测量生成文本的可能性、词汇多样性或在通用表示空间中的分布相似性，但它们可能忽视质量的基本维度。一个显著的盲点是全局连贯性：生成的段落可能在局部上流畅，但在全局上却存在矛盾、因果不一致或主题脱节等问题。这些失败仍然可能保留现有指标所依赖的标记级和词汇统计特征。我们识别出表示作为检测这些失败的核心瓶颈，并引入了 CHORD（考虑连贯性的隐藏状态开放生成参考距离），这是一种对连贯性敏感的分布式指标。CHORD 使用一个引发连贯性的提示，将生成的文本和人类撰写的语料编码到一个冻结的 LLM 的隐藏状态空间中，并使用 RBF 核的 MMD 比较结果分布。为了验证该指标对连贯性下降的响应，而不是对通用文本变化的响应，我们构建了一个反事实评估套件，将分级的连贯性下降扰动与保持意义的控制进行配对。CHORD 能够选择性地检测关系、话语、结构和混合失败，而困惑度、熵、MAUVE、FBD 和基于 MMD 的基线要么遗漏这些失败，要么无法将其与良性重写区分开。因子消融实验表明，表示是连贯性敏感性的主要来源，而 RBF-MMD 一旦相关区分变得可见，就能提高样本效率。更大的骨干网络能够捕捉更细致的区分，但连贯性提示仅在骨干网络能够跟随提示时才能改善选择性。在无条件生成和前缀延续任务中，CHORD 的模型排名与人类对输出是否合理且看起来像人类撰写的判断高度一致。这些结果共同确立了表示设计在可靠的分布式评估中的核心地位。
cs.CL / 155 / 2609.34247
SALMONN-duo: Adaptive Dual-System Coordination for Full-Duplex Voice Agents
SALMONN-duo：全双工语音代理的自适应双系统协调
Abstract
Full-duplex speech large language models (LLMs) enable low-latency, natural voice interaction. However, real-world agents must also use tools and perform deliberative reasoning-operations whose variable latency and computational cost conflict with the stringent timing requirements of real-time conversation. To reconcile these demands, we propose SALMONN-duo, an adaptive dual-system voice agent inspired by dual-process theories of cognition. SALMONN-duo separates real-time interaction from deliberative computation by pairing an always-on, fast-thinking full-duplex speech LLM (system 1) with a powerful asynchronous slow-thinking LLM agent (system 2). Beyond handling real-time interaction, system 1 learns when to answer directly and when to delegate, remaining responsive during backend execution and seamlessly integrating returned information into the ongoing dialogue without exposing tool traces or losing conversational context. Evaluations on single-turn spoken question answering (QA) and multi-turn conversations demonstrate that adaptive delegation substantially improves accuracy on knowledge-intensive and multi-hop reasoning questions, while knowledge-boundary-aware training avoids unnecessary system 2 invocations. On a customized version of $\tau$-Voice, SALMONN-duo further demonstrates its ability to complete environment-grounded, policy-constrained tasks through multi-turn interactions in realistic business scenarios. Finally, cost-aware reinforcement learning further enhances the trade-off between task performance and backend usage across the QA and conversation tasks, while improving task success and response safety on $\tau$-Voice with an acceptable increase in the delegation rate.
Chinese Translation
全双工语音大型语言模型（LLMs）能够实现低延迟、自然的语音交互。然而，现实世界中的代理还必须使用工具并执行深思熟虑的推理操作，这些操作的可变延迟和计算成本与实时对话的严格时间要求相冲突。为了解决这些需求，我们提出了SALMONN-duo，这是一种受认知双过程理论启发的自适应双系统语音代理。SALMONN-duo通过将一个始终在线、快速思考的全双工语音LLM（系统1）与一个强大的异步慢思考LLM代理（系统2）配对，从而将实时交互与深思熟虑的计算分开。除了处理实时交互外，系统1还学习何时直接回答以及何时委派，在后端执行期间保持响应，并将返回的信息无缝整合到正在进行的对话中，而不暴露工具痕迹或失去对话上下文。在单轮口语问答（QA）和多轮对话的评估中，适应性委派显著提高了知识密集型和多跳推理问题的准确性，同时知识边界感知训练避免了不必要的系统2调用。在定制版的$ au$-Voice上，SALMONN-duo进一步展示了其通过多轮交互在现实商业场景中完成环境基础、政策约束任务的能力。最后，成本感知强化学习进一步增强了QA和对话任务中任务性能与后端使用之间的权衡，同时在$ au$-Voice上提高了任务成功率和响应安全性，且委派率的增加在可接受范围内。
cs.CL / 156 / 2609.34257
Recursive LLM Degradation in Biomedical Question Answering: A Cross-Generation Study
生物医学问答中的递归语言模型退化：跨代研究
Abstract
Repeatedly training language models on their own generated data may create a synthetic-data feedback loop in which errors and distributional biases are reintroduced into subsequent training datasets. This paper studies that process in biomedical question answering (QA) using PubMedQA and two Qwen2.5 model sizes, 0.5B and 3B parameters. The study compares a recursive synthetic-data condition, in which generation G(k+1) is trained on answers produced by G(k), against a Human-Control condition that repeatedly uses the original human training data. The study evaluates across four generations from G0-G3 with two random seeds (42 and 123) and a fixed evaluation set of 1,000 expert-labeled samples. The evaluation includes disease and chemical entity F1, context-supported rate, lexical and semantic similarity, answer length, repetition rate, and other evaluation metrics. The Recursive condition for both model sizes and both seeds showed larger declines than the Human-Control condition in disease entity F1, chemical entity F1, context-supported rate, ROUGE-L, and cosine similarity. Under the fixed no-repeat 3-gram decoding constraint, the main observed behavioral change was increased answer length, while the measured 3-gram repetition rate did not increase. The magnitude of the difference-in-change was larger for the 3B model than for the 0.5B model. This difference was particularly apparent in disease F1, context-supported rate, cosine similarity, and answer length. These results show domain-specific changes associated with using recursive synthetic-data training in biomedical QA, but do not establish clinical hallucination rates or universal model collapse.
Chinese Translation
反复在自身生成的数据上训练语言模型可能会产生一个合成数据反馈循环，其中错误和分布偏差被重新引入到后续的训练数据集中。本文研究了这一过程在生物医学问答（QA）中的应用，使用了PubMedQA和两种Qwen2.5模型规模，分别为0.5B和3B参数。研究比较了递归合成数据条件，其中生成G(k+1)是在G(k)生成的答案上训练的，与人类控制条件，该条件反复使用原始人类训练数据。研究评估了从G0到G3的四个代次，使用了两个随机种子（42和123）以及一个固定的评估集，包含1,000个专家标注的样本。评估指标包括疾病实体和化学实体的F1值、上下文支持率、词汇和语义相似度、答案长度、重复率以及其他评估指标。在疾病实体F1、化学实体F1、上下文支持率、ROUGE-L和余弦相似度方面，两个模型规模和两个种子的递归条件显示出比人类控制条件更大的下降。在固定的无重复3-gram解码约束下，观察到的主要行为变化是答案长度增加，而测得的3-gram重复率并未增加。3B模型的变化幅度大于0.5B模型。这一差异在疾病F1、上下文支持率、余弦相似度和答案长度方面尤为明显。这些结果显示了在生物医学问答中使用递归合成数据训练所关联的特定领域变化，但并未确定临床幻觉率或普遍模型崩溃。
cs.CL / 157 / 2609.34284
Over-Personalization Is a Decision Failure: Generation-Induced Apply Bias in LLMs
过度个性化是决策失误：生成引发的应用偏差在大型语言模型中的影响
Abstract
Personalized LLMs must decide, for each stored preference, whether the current context calls for applying or suppressing it, which we call its applicability. They frequently over-personalize, applying preferences the context rules out, yet existing benchmarks score only the final response and cannot tell where this failure arises. We decompose preference handling into three stages and measure each separately: (1) knowing whether a preference applies, (2) deciding on an explicit Apply/Suppress label, and (3) generating a response consistent with that label. Using linear probes, we first show that this applicability signal remains decodable from hidden states during generation. By making the decision explicit, we then find that in most settings wrong decisions faithfully followed outnumber correct decisions lost in generation. We thus locate the failure in the decision, which breaks once the model is also asked to answer. To determine whether this reflects lost sensitivity or a response bias, we propose ABIDE (Apply-Bias Investigation via Decision-score), which adapts signal detection theory to Apply-vs-Suppress decision scores read directly from logits. ABIDE reveals a generation-induced Apply bias: merely stating an answer-generation objective shifts the decision score toward Apply while sensitivity is largely preserved, and the shift persists under controls for prompt structure, cascades across preference slots, and prompt wording. Finally, we show that subtracting a single bias scalar, estimated on a held-out split, from the decision score at decoding time reduces leakage while largely preserving fulfillment.
Chinese Translation
个性化的大型语言模型（LLMs）必须为每个存储的偏好决定当前上下文是否需要应用或抑制该偏好，我们称之为其适用性。它们经常出现过度个性化的情况，即在上下文排除的情况下应用偏好，而现有的基准仅对最终响应进行评分，无法识别这一失误的来源。我们将偏好处理分解为三个阶段，并分别进行测量：(1) 知道偏好是否适用，(2) 决定明确的应用/抑制标签，以及 (3) 生成与该标签一致的响应。通过线性探测，我们首先表明这一适用性信号在生成过程中仍然可以从隐藏状态中解码。通过明确决策，我们发现，在大多数设置中，错误决策的数量远超过在生成过程中丢失的正确决策。因此，我们将失误定位于决策上，一旦模型被要求回答，决策就会失效。为了确定这是否反映了敏感性丧失或响应偏差，我们提出了ABIDE（通过决策分数进行应用偏差调查），该方法将信号检测理论适应于直接从logits读取的应用与抑制决策分数。ABIDE揭示了一种生成引发的应用偏差：仅仅陈述一个答案生成目标就会使决策分数向应用倾斜，而敏感性在很大程度上得以保留，这种偏移在提示结构、偏好槽级联和提示措辞的控制下仍然存在。最后，我们表明，在解码时从决策分数中减去一个在保留分割上估计的单一偏差标量，可以减少泄漏，同时在很大程度上保留实现度。
cs.CL / 158 / 2609.34287
ReScraper: Unified Scraping and Cleaning of Web Data for Effective LLM Pretraining
ReScraper：统一的网页数据抓取与清洗以有效进行大语言模型预训练
Abstract
LLM pretraining corpora are normally cleaned by a stack of hand-written heuristics. A heuristic scraper extracts the main content from HTML, and dozens of rule-based filters then clean it, so corpus quality is capped by the coarseness and accuracy of the rules. In this work, we propose ReScraper, a unified language model of only 0.6B parameters that replaces this entire stack. To train ReScraper, we carefully curate supervised data from the outputs of three teacher models, so it learns to first extract the main content from raw data and then choose among four operations: keeping the page as extracted, editing out noisy lines and spans, deleting it entirely, or rewriting it when it is poorly written but informative. Based on the same crawled data pool, pretraining 400M, 1.4B, and 2.8B models on our curated data improves the DCLM Core score by a relative 3.8--4.7% over the strongest baseline at each scale, including the costly multi-agent curation. Our analyses show that each operation plays a distinct and complementary role, and that extracting and cleaning in one model outperforms a cascade of separate models. ReScraper also concentrates its operations on the pages that need them, raising the quality of poor pages the most while keeping the corpus diverse. These results demonstrate the feasibility and effectiveness of AI4AI for pretraining data curation, where a small learned model takes over an entire stage of the pipeline from hand-written heuristics. We open-source our code at https://github.com/cxcscmu/ReScraper
Chinese Translation
大语言模型（LLM）预训练语料库通常通过一系列手工编写的启发式规则进行清洗。启发式抓取器从HTML中提取主要内容，然后通过数十个基于规则的过滤器进行清洗，因此语料库的质量受到规则的粗糙度和准确性的限制。在本研究中，我们提出了ReScraper，一个仅包含6亿参数的统一语言模型，替代了整个处理流程。为了训练ReScraper，我们从三个教师模型的输出中精心策划了监督数据，使其首先学习从原始数据中提取主要内容，然后在四种操作中进行选择：保持提取的页面、编辑掉噪声行和段落、完全删除，或在内容写得不佳但信息丰富时进行重写。基于相同的爬取数据池，在我们策划的数据上预训练400M、1.4B和2.8B模型，相较于每个规模下最强基线，包括昂贵的多代理策划，DCLM核心得分提高了相对3.8%至4.7%。我们的分析表明，每种操作发挥着独特且互补的作用，而在一个模型中进行提取和清洗的效果优于一系列独立模型的级联处理。ReScraper还将操作集中在需要它们的页面上，显著提高了质量较差页面的质量，同时保持了语料库的多样性。这些结果证明了AI4AI在预训练数据策划中的可行性和有效性，其中一个小型学习模型接管了整个流程中的手工编写启发式规则阶段。我们在https://github.com/cxcscmu/ReScraper上开源了我们的代码。
cs.CL / 159 / 2609.34296
Dr.Credit: Rubric-Grounded Process Credit Assignment for Deep Research Agents
Dr.Credit：基于评分标准的过程信用分配用于深度研究代理
Abstract
Rubric-based tasks are increasingly addressed through reinforcement learning (RL), with rubric scores used as training rewards. However, these rewards typically supervise final answers without distinguishing the contributions of intermediate decisions. Many existing credit assignment methods rely on ground-truth answers to define process rewards, limiting their applicability to open-ended tasks without canonical solutions. To address this limitation, the proposed rubric-grounded credit uses task requirements as a shared reference for final answer evaluation and process supervision. The information returned by tools is assessed for the additional support it provides toward satisfying each rubric relative to that rubric's history of accepted support. By referencing these histories, credit distinguishes new support from evidence already present in the trajectory while recognizing partial support for each rubric. Dr.Credit uses rubric-grounded credit to supervise intermediate tool turns in an RL framework for deep research agents. The resulting process advantages are combined with GRPO outcome advantages to guide research decisions while retaining supervision of final-report quality. Evaluations on four in-domain and out-of-domain benchmarks show that Dr.Credit outperforms the evaluated open deep research baselines on every primary metric and submetric. Meanwhile, with an 8B-parameter backbone, the trained agent achieves average performance competitive with the evaluated frontier proprietary models. Further analyses suggest more efficient evidence acquisition and higher-quality reports under limited research-turn budgets, motivating the extension of rubric-grounded process supervision to a broader range of rubric-based tasks.
Chinese Translation
基于评分标准的任务越来越多地通过强化学习（RL）来解决，评分标准分数被用作训练奖励。然而，这些奖励通常只监督最终答案，而未能区分中间决策的贡献。许多现有的信用分配方法依赖于真实答案来定义过程奖励，这限制了它们在没有规范解决方案的开放式任务中的适用性。为了解决这一限制，提出的基于评分标准的信用使用任务要求作为最终答案评估和过程监督的共享参考。工具返回的信息被评估以确定其在满足每个评分标准方面提供的额外支持，相对于该评分标准的历史接受支持。通过参考这些历史，信用能够区分新的支持与轨迹中已经存在的证据，同时承认每个评分标准的部分支持。Dr.Credit利用基于评分标准的信用在深度研究代理的强化学习框架中监督中间工具回合。由此产生的过程优势与GRPO结果优势相结合，以指导研究决策，同时保持对最终报告质量的监督。在四个领域内和领域外的基准评估中，Dr.Credit在每个主要指标和子指标上均优于评估的开放深度研究基线。同时，使用一个8B参数的主干，训练后的代理在性能上与评估的前沿专有模型竞争。进一步的分析表明，在有限的研究回合预算下，证据获取更高效，报告质量更高，这激励了将基于评分标准的过程监督扩展到更广泛的基于评分标准的任务。
cs.CL / 160 / 2609.34320
Certified Selective Automation of LLM Agent Evaluation
大规模语言模型代理评估的认证选择性自动化
Abstract
Evaluating LLM agents still ends with a human reading trajectories, because automatic judges carry no guarantee on how often they are wrong. We ask the operational question: what fraction of agent evaluation can a judge take over, with a certificate that the error rate among auto-decided trajectories stays below a budget alpha? Agent corpora resist the standard answer: many agents attempt the same tasks, so trajectories arrive in correlated clusters, and the i.i.d. certificates of existing selective-judging methods can overstate what is safe: a naive certificate can claim 98% automation while its realized error exceeds the budget in 17.5% of task resamples. We introduce a task-level bootstrap certificate that is valid in every regime we test while matching the naive certificate's coverage; finite-sample cluster-valid alternatives certify nothing at realistic task counts. Under this certificate, a 4B logprob judge trained with SFT and reject-weighted GRPO certifies 0.30-0.59 of evaluation on tool-use and web corpora at alpha=0.1, the only judge, among strongly elicited frontier models, certifying on both headline corpora. Certified coverage is predictable before training from base rate and discrimination alone (leave-one-corpus-out R^2=0.96). Finally, the certificate doubles as a self-training filter: pseudo-labels harvested inside certified regions have contamination bounded by alpha by construction (realized 0.000-0.041 across six harvests), letting a judge enter an unseen domain at in-domain strength with zero target training labels.
Chinese Translation
评估大规模语言模型（LLM）代理仍然依赖于人工阅读轨迹，因为自动评判者无法保证其错误率的频率。我们提出了一个操作性问题：在保证自动决策轨迹的错误率低于预算α的情况下，评判者可以接管多少比例的代理评估？代理语料库抵制了标准答案：许多代理尝试相同的任务，因此轨迹会以相关簇的形式出现，而现有选择性评判方法的独立同分布（i.i.d.）证书可能会夸大安全性：一个简单的证书可能声称实现98%的自动化，而其实际错误率在17.5%的任务重采样中超过预算。我们引入了一种任务级引导证书，该证书在我们测试的每种情况下都有效，同时与简单证书的覆盖率相匹配；有限样本的簇有效替代方案在现实任务数量下无法提供任何认证。在该证书下，一个经过监督微调（SFT）和拒绝加权的GRPO训练的4B对数概率评判者在工具使用和网络语料库的评估中认证了0.30-0.59，α=0.1，是在强烈引导的前沿模型中唯一在两个主要语料库上进行认证的评判者。经过训练后，认证覆盖率可以仅根据基准率和区分度进行预测（留一语料库外R^2=0.96）。最后，该证书还可以作为自我训练过滤器：在认证区域内收集的伪标签的污染量在构造上被α限制（在六次收集中实现的污染率为0.000-0.041），使评判者能够以领域内强度进入未见领域，而无需目标训练标签。
cs.CL / 161 / 2609.34345
CRISP: Cultural Reward Modeling for Implicit Situated Propriety
CRISP：隐性情境适宜性的文化奖励建模
Abstract
As large language models (LLMs) are increasingly deployed across countries and regions, the ability to recognize and respond appropriately to diverse cultural contexts becomes increasingly important. However, existing research has largely focused on cultural knowledge or tasks with predefined response spaces, while open-ended culturally situated behavior remains comparatively underexplored. In this work, we introduce CRISP-RM, a culturally situated reward model that assigns rewards according to cultural appropriateness in open-ended social scenarios. During policy optimization, we further introduce Norm Grounding Supervision (NGS), providing guidance that enhances the policy's sensitivity to relevant cultural norms. To construct culturally situated data, we employ a collaborative multi-agent framework that instantiates implicit cultural norms into diverse social scenarios and further curate NormCompass as a dedicated testbed. We conduct comprehensive experiments to evaluate the effectiveness of CRISP-RM in both reward modeling and policy optimization. Best-of-\(N\) experiments show that CRISP-RM consistently outperforms strong general reward models. During GRPO policy optimization, CRISP-RM generally improves culturally situated behavior, while incorporating NGS yields further gains. Further analyses demonstrate the advantages of CRISP-RM in distinguishing culturally appropriate behavior beyond superficial fluency and politeness, while NGS provides complementary gains during policy optimization by improving norm grounding.
Chinese Translation
随着大型语言模型（LLMs）在各国和地区的广泛应用，识别和适当地响应多样文化背景的能力变得愈加重要。然而，现有研究主要集中在文化知识或具有预定义响应空间的任务上，而开放式的文化情境行为则相对未被充分探索。在本研究中，我们引入了CRISP-RM，一种文化情境奖励模型，根据开放式社会场景中的文化适宜性分配奖励。在策略优化过程中，我们进一步引入了规范基础监督（Norm Grounding Supervision, NGS），提供指导以增强策略对相关文化规范的敏感性。为了构建文化情境数据，我们采用了一个协作多智能体框架，将隐性文化规范实例化到多样的社会场景中，并进一步策划了NormCompass作为专门的测试平台。我们进行了全面的实验，以评估CRISP-RM在奖励建模和策略优化中的有效性。最佳的N实验表明，CRISP-RM在强大的通用奖励模型中始终表现优越。在GRPO策略优化过程中，CRISP-RM通常改善了文化情境行为，而结合NGS则带来了进一步的提升。进一步的分析表明，CRISP-RM在区分文化适宜行为方面优于表面流畅性和礼貌性，而NGS通过改善规范基础在策略优化中提供了互补的收益。
cs.CL / 162 / 2609.34366
When Harness Beats Scale, and When Reading Beats Both
当利用小规模超越规模时，阅读又何尝不胜于二者
Abstract
We describe our system for DocSem, the document-grounded quantitative reasoning shared task at DocInsights 2026, and analyze why it succeeded on labeled data and failed on the test set. The pipeline pairs hybrid block retrieval with Program-of-Thoughts (PoT) generation executed in a sandboxed interpreter, self-consistency sampling, and entity enrichment from chunk-level knowledge graphs. On our held-out split, application architecture moved the metrics far more than model scale did: PoT added 0.282 joint accuracy to a compact 7B model but at most 0.005 to a 72B model, and a 27B model with the full harness matched the 72B (0.884 vs.\ 0.873) at roughly 2.7$\times$ fewer parameters and a quarter of the CO$_2$. We read this through a distinction between world knowledge, which scales steeply with parameters, and language knowledge, which scales gently, and show that structured-output training makes a compact model harness-ready rather than merely small. On the raster, watermarked test PDFs the same system collapsed to 13.58\% joint (rank 149 of 163); a controlled re-rendering of the validation set reproduces the OCR half of the collapse while bounding what the simulation misses. Auditing the physical nature of evaluation inputs precedes architecture, and the leaderboard's bimodality is consistent with reading quality, not reasoning, having separated the field.
Chinese Translation
我们描述了我们的系统用于DocSem，这是2026年DocInsights的文档基础定量推理共享任务，并分析了它为何在标记数据上成功而在测试集上失败。该流程将混合块检索与在沙盒解释器中执行的思维程序（Program-of-Thoughts, PoT）生成、自我一致性采样和来自块级知识图谱的实体增强相结合。在我们的保留分割上，应用架构对指标的提升远超模型规模的影响：PoT为一个紧凑的7B模型增加了0.282的联合准确率，而对一个72B模型的提升最多为0.005，且一个配备完整工具的27B模型在参数量减少约2.7倍和二氧化碳排放减少四分之一的情况下，与72B模型的表现相当（0.884对0.873）。我们通过区分世界知识（其参数规模急剧增加）和语言知识（其参数规模温和增加）来理解这一现象，并展示结构化输出训练使得紧凑模型准备好使用工具，而不仅仅是小型化。在带有水印的光栅测试PDF上，同一系统的表现崩溃至13.58%的联合准确率（排名149/163）；对验证集的受控重渲染重现了崩溃的一半OCR，同时界定了模拟所遗漏的内容。对评估输入的物理性质进行审计是架构设计之前的步骤，且排行榜的双峰性与阅读质量一致，而非推理，已经将该领域分开。
cs.CL / 163 / 2609.34385
Just-In-Time Agent Memory with Runtime Agentic Research
即时代理记忆与运行时代理研究
Abstract
Memory is critical for AI agents. Many existing agent-memory systems follow an Ahead-of-Time (AOT) design, constructing memory before a specific request arrives. While this reduces online serving cost, such request-agnostic memory construction can discard fine-grained information that later becomes important. To address this limitation, we propose Just-In-Time Agent Memory (JAM), a trainable framework for query-conditioned context construction at runtime. A Memorizer preserves complete raw histories in a hierarchical page-store with compact navigational summaries, while a Researcher iteratively retrieves, inspects, and integrates evidence for each request. To train these memory-use behaviors, we introduce Memory-Gym, an evidence-grounded data synthesis pipeline covering nine task types across six domains, and optimize the Researcher through verified-trajectory supervised fine-tuning followed by Hint-guided Group Relative Policy Optimization. We demonstrate the effectiveness of JAM across a variety of benchmarks on agent memory and long-context processing, where it achieves stronger task performance than AOT-style memory systems while remaining substantially more efficient than prior trained agentic memory approaches. To support reproducibility and future research, we release our anonymized source code at https://github.com/VectorSpaceLab/general-agentic-memory.
Chinese Translation
记忆对于人工智能代理至关重要。许多现有的代理记忆系统采用提前设计（Ahead-of-Time, AOT），在特定请求到达之前构建记忆。虽然这减少了在线服务成本，但这种与请求无关的记忆构建可能会丢弃后续变得重要的细粒度信息。为了解决这一局限性，我们提出了即时代理记忆（Just-In-Time Agent Memory, JAM），这是一个可训练的框架，用于在运行时根据查询条件构建上下文。记忆器（Memorizer）在一个层次化的页面存储中保留完整的原始历史记录，并提供紧凑的导航摘要，而研究者（Researcher）则迭代地检索、检查并整合每个请求的证据。为了训练这些记忆使用行为，我们引入了记忆健身房（Memory-Gym），这是一个基于证据的数据合成管道，涵盖六个领域的九种任务类型，并通过经过验证的轨迹监督微调和提示引导的群体相对策略优化来优化研究者。我们在代理记忆和长上下文处理的各种基准测试中展示了JAM的有效性，其任务性能优于AOT风格的记忆系统，同时在效率上显著优于之前的训练代理记忆方法。为了支持可重复性和未来研究，我们在 https://github.com/VectorSpaceLab/general-agentic-memory 发布了我们的匿名源代码。
cs.CL / 164 / 2609.34386
Look Before You Select: Rethinking Vocabulary Sparsification in On-Policy Distillation
选择前的审慎：重新思考在线策略蒸馏中的词汇稀疏化
Abstract
On-policy distillation (OPD) uses teacher correction on student-generated responses. Full-vocabulary correction can provide important corrections even for tokens that the student assigns low probability, but backpropagating through all token logits becomes memory-intensive for long sequences. Existing memory-saving approaches estimate corrections from sampled tokens or restrict supervision to the student's TopK tokens, introducing sampling noise or changing the full-vocabulary correction. We introduce \textbf{SparseOPD}, which uses full-vocabulary teacher correction to determine which corrections matter before selecting the token logits to differentiate. SparseOPD first constructs the full-vocabulary correction without retaining its backward graph, then selects tokens by correction magnitude rather than student probability. Signed residual compensation preserves the total promoting and suppressing correction mass, while correction-aware budget allocation distributes the sparse support across positions. Finally, the update backpropagates only through the selected token logits. Across six task--scale settings spanning mathematics, chemistry QA, and multimodal reasoning, SparseOPD outperforms Sampled Token and TopK in task-average accuracy and matches or exceeds Full Vocabulary. Gradient cosine similarity reaches 99\% on 4B mathematics, while 8K full-parameter profiling shows 70.5\% lower backward memory.
Chinese Translation
在线策略蒸馏（On-policy distillation, OPD）利用教师对学生生成的响应进行纠正。全词汇纠正即使对于学生赋予低概率的标记也能提供重要的修正，但在长序列中，通过所有标记的logits进行反向传播会变得内存密集。现有的节省内存的方法通过从采样的标记中估计修正或将监督限制在学生的TopK标记上，引入了采样噪声或改变了全词汇纠正。我们提出了 extbf{SparseOPD}，该方法利用全词汇教师纠正来确定哪些修正是重要的，然后再选择要进行微分的标记logits。SparseOPD首先构建全词汇纠正，而不保留其反向图，然后根据修正幅度而不是学生概率选择标记。带符号的残差补偿保持了总的促进和抑制修正量，而考虑修正的预算分配则在各个位置分配稀疏支持。最后，更新仅通过所选的标记logits进行反向传播。在涵盖数学、化学问答和多模态推理的六个任务规模设置中，SparseOPD在任务平均准确率上优于采样标记和TopK，并且与全词汇相匹配或超过。梯度余弦相似度在4B数学上达到99\%，而8K全参数分析显示反向内存减少了70.5\%。
cs.CL / 165 / 2609.34388
Reciprocal Guidance: Orchestrating Draft and Verify Budgets for Advancing the Diffusion-AR Self-Speculation Frontier
互惠指导：协调草拟与验证预算以推进扩散自我推测前沿
Abstract
Diffusion drafting with autoregressive (AR) verification has emerged as a promising paradigm for efficient speculative decoding. Recent self-speculation models, represented by Nemotron-Labs-Diffusion, further simplify the speculative pipeline by unifying drafting and verification within a shared backbone, while enabling longer acceptance lengths. However, the Pareto frontier between aggregate and per-request throughput remains underexplored. At low concurrency, sequential draft-verify execution requires two model forward passes per round, limiting the effective tokens per forward (TPF). By contrast, at high concurrency, longer drafts incur increasingly expensive computation, forcing individual requests to operate under constrained speculation budgets and preventing full exploitation of the full-backbone drafter. Our key observation indicates that drafting and verification exhibit reciprocal predictability. Draft logits can anticipate likely verification mismatches, while recent verification outcomes predict future drafting utility and suitable block sizes. Building on this observation, we introduce Reciprocal Guidance (RecGuide), a runtime draft-verify orchestration framework that adapts speculative decoding to varying serving loads. RecGuide exploits spare compute capacity through verification-overlapped drafting at low concurrency, while dynamically allocating request-specific draft block sizes as the workload becomes increasingly compute-intensive. Experiments across a wide range of concurrency levels demonstrate consistent throughput improvements over vanilla self-speculation, achieving up to $1.8\times$ speedup.
Chinese Translation
基于自回归（AR）验证的扩散草拟已成为高效推测解码的有前景的范式。最近的自我推测模型，如Nemotron-Labs-Diffusion，通过在共享骨干网内统一草拟和验证，进一步简化了推测流程，同时允许更长的接受长度。然而，聚合吞吐量与每请求吞吐量之间的帕累托前沿仍然未得到充分探索。在低并发情况下，顺序的草拟-验证执行每轮需要两个模型前向传递，限制了每次前向传递的有效标记数（TPF）。相比之下，在高并发情况下，较长的草拟会产生越来越高的计算成本，迫使单个请求在受限的推测预算下运行，无法充分利用全骨干网草拟器。我们的关键观察表明，草拟和验证表现出互惠的可预测性。草拟的logits可以预测可能的验证不匹配，而最近的验证结果则预测未来草拟的效用和合适的块大小。基于这一观察，我们引入了互惠指导（RecGuide），一个运行时草拟-验证协调框架，能够根据不同的服务负载调整推测解码。RecGuide在低并发情况下通过验证重叠草拟来利用闲置的计算能力，同时在工作负载变得越来越计算密集时动态分配请求特定的草拟块大小。在广泛的并发水平下进行的实验表明，相较于普通自我推测，吞吐量持续改善，最高可实现$1.8 imes$的加速。
cs.CL / 166 / 2609.34425
Zero-Shot Cue-Grounded Topic Segmentation of Spoken Documents
无监督提示引导的口语文档主题分割
Abstract
Topic segmentation structures spoken documents into coherent sections, facilitating navigation and downstream understanding. The appropriate granularity can vary substantially, ranging from broad thematic shifts to fine-grained subtopics. Existing LLM-based segmenters, however, often struggle to adapt to this variation, causing them to either merge distinct subtopics or over-segment coherent themes. To address this, we introduce Cue-Grounded Segmentation (CGS), a training-free framework that operates without any task-specific supervision. CGS first identifies phrases that explicitly signal the start of a new topic and uses their sentence positions as segment boundaries. When such cues are insufficient, it falls back to semantic segmentation, guided by the document structure inferred during cue extraction. Across six benchmarks and six LLM backbones, CGS consistently outperforms existing baselines, remains robust to noisy ASR transcripts, and achieves these gains with low API cost on proprietary models.
Chinese Translation
主题分割将口语文档结构化为连贯的部分，便于导航和后续理解。适当的粒度可以有很大差异，从广泛的主题转变到细粒度的子主题。然而，现有的基于大型语言模型（LLM）的分割器往往难以适应这种变化，导致它们要么合并不同的子主题，要么过度分割连贯的主题。为了解决这个问题，我们提出了提示引导分割（Cue-Grounded Segmentation, CGS），这是一个无需训练的框架，能够在没有任何任务特定监督的情况下运行。CGS首先识别明确指示新主题开始的短语，并使用它们的句子位置作为分割边界。当此类提示不足时，它会退回到语义分割，依据在提示提取过程中推断出的文档结构进行指导。在六个基准测试和六个LLM基础模型上，CGS始终优于现有基线，且对嘈杂的自动语音识别（ASR）转录保持稳健，并以低API成本在专有模型上实现这些增益。
cs.CL / 167 / 2609.34428
AgentHop: A Diagnostic Benchmark for Agentic Multi-Hop Scientific Question Answering
AgentHop：一种用于代理多跳科学问答的诊断基准
Abstract
Agentic tasks require a large language model to interact with the world, navigating information and gathering evidence across multiple steps with restricted resources. Due to this complexity, agentic task failures arise from various sources, and pinpointing these failure causes is essential to diagnose and improve agentic systems. Existing benchmarks, however, tend to focus on a single leaderboard score, leaving the underlying failure modes opaque. To fill this gap, we introduce AgentHop, a diagnostic benchmark of 1,011 multiple-choice questions paired with a controlled seven-tool sandbox under fixed token, turn, and tool-call constraints. AgentHop reveals model vulnerabilities by dissecting a single accuracy score along four axes of agent operation: retrieval, synthesis, tool-call, and resource management. Across 19 models, we find that behavior clusters by model family, with tool-call signatures revealing distinct family fingerprints: GPT models commit early, Anthropic and GLM checkpoints verify before committing, DeepSeek and Kimi over-search, and Gemini-3 Pro stays balanced. Decomposed axes further expose within-family structure: Claude Opus 4.6 and Sonnet 4.6 land within one accuracy point yet diverge on retrieval-versus-synthesis emphasis, with Opus retrieving more and Sonnet synthesizing better. We release the full benchmark set and the harness to support diagnostic agent benchmarking.
Chinese Translation
代理任务要求大型语言模型与世界进行互动，在有限资源下跨多个步骤导航信息并收集证据。由于这种复杂性，代理任务的失败源于多种因素，准确找出这些失败原因对于诊断和改进代理系统至关重要。然而，现有基准往往侧重于单一的排行榜得分，使得潜在的失败模式不够透明。为填补这一空白，我们推出了AgentHop，这是一个包含1,011个多项选择题的诊断基准，并配备一个受控的七工具沙箱，设定了固定的令牌、回合和工具调用限制。AgentHop通过沿四个代理操作轴（检索、综合、工具调用和资源管理）剖析单一的准确性得分，揭示模型的脆弱性。在19个模型中，我们发现行为按模型家族聚类，工具调用特征揭示了不同家族的指纹：GPT模型早期犯错，Anthropic和GLM检查点在提交前进行验证，DeepSeek和Kimi过度搜索，而Gemini-3 Pro保持平衡。分解的轴进一步揭示了家族内部的结构：Claude Opus 4.6和Sonnet 4.6的准确性相差不到一个点，但在检索与综合的侧重点上却有所不同，Opus在检索方面表现更好，而Sonnet在综合方面更佳。我们发布了完整的基准集和支持诊断代理基准测试的工具。
cs.CL / 168 / 2609.34438
Remember by Asking: Retrieval-Induced Memory Evolution for LLM Agents
通过提问记忆：用于大型语言模型代理的检索诱导记忆演变
Abstract
Long-term memory is essential for language agents to maintain coherent and effective behavior over extended, multi-session interactions. Existing memory systems mainly use retrieval at read time, while write-time memory formation still relies on direct extraction or compression. However, when future information needs are unknown, compressing an entire interaction in one pass can overlook locally important details that may matter later. To this end, we introduce RIME, a retrieval-induced memory framework that shifts memory construction from monolithic compression toward evidence-centered integration. RIME uses generic self-questions to retrieve focused dialogue evidence and grounds memory formation in both the retrieved evidence and relevant historical memories, which are jointly reconciled into an evolving memory bank with temporal and provenance information. At inference time, compressed memory serves as the primary rather than the sole source of evidence: when it cannot support an answer, RIME retrieves relevant source dialogue together with its local context to recover information omitted during memory formation, without resorting to full-history processing. Extensive experiments on LoCoMo with Qwen3-235B-A22B and GPT-5.6 Sol show that RIME consistently achieves the best performance across all three quality metrics among the compared methods, while requiring substantially fewer query-time LLM tokens.
Chinese Translation
长期记忆对于语言代理在延续的多会话交互中保持连贯和有效的行为至关重要。现有的记忆系统主要在读取时进行检索，而写入时的记忆形成仍依赖于直接提取或压缩。然而，当未来信息需求未知时，一次性压缩整个交互可能会忽略后续可能重要的局部细节。为此，我们提出了RIME（检索诱导记忆框架），将记忆构建从单一压缩转向以证据为中心的整合。RIME使用通用自我提问来检索聚焦的对话证据，并将记忆形成基于检索到的证据和相关的历史记忆，这些记忆共同调和成一个具有时间和来源信息的演变记忆库。在推理时，压缩的记忆作为主要而非唯一的证据来源：当它无法支持答案时，RIME检索相关的源对话及其局部上下文，以恢复在记忆形成过程中遗漏的信息，而无需完全历史处理。在LoCoMo上与Qwen3-235B-A22B和GPT-5.6 Sol的广泛实验表明，RIME在比较方法中在所有三个质量指标上始终实现最佳性能，同时所需的查询时间LLM令牌显著减少。
cs.CL / 169 / 2609.34447
Unbiased Top-$k$ Estimation for On-Policy Distillation
无偏的 Top-$k$ 估计用于在线策略蒸馏
Abstract
On-policy distillation (OPD) is becoming an important component of large language model (LLM) post-training for transferring the reasoning capability of a strong teacher LLM to a weaker student LLM. OPD trains the student by minimizing the reverse KL divergence between the teacher and the student via rollouts generated by the student's policy. However, estimating the gradient of the reverse KL divergence in OPD remains a challenge. Using only the sampled token from the student-generated rollout is computationally cheap but provides limited distributional supervision, which will degrade accuracy. In addition, using the full vocabulary provides complete distributional supervision but is computationally expensive. Therefore, recent works propose Top-$k$ OPD (TK-OPD) that use selected top-$k$ tokens, which provides richer distributional supervision than sampled-token estimation at substantially lower computational cost than full-vocabulary estimation. Unfortunately, using only the selected top-$k$ tokens induces bias, leading to accuracy degradation, as the probability mass outside the selected top-$k$ tokens is discarded. To address the bias of TK-OPD, we propose Tail-Corrected Top-$k$ On-Policy Distillation (TT-OPD). It preserves the advantages of TK-OPD, including rich distributional supervision and low computational cost, while providing an unbiased estimator of the gradient of the reverse KL divergence. The key insight of TT-OPD is to use not only the selected top-$k$ tokens, but also the sampled token from the student-generated rollout, thereby recovering the discarded probability mass in expectation, avoiding the bias. Experimental results demonstrate that TT-OPD significantly outperforms other tested OPD variants.
Chinese Translation
在线策略蒸馏（OPD）正成为大型语言模型（LLM）后训练的重要组成部分，用于将强教师 LLM 的推理能力转移到较弱的学生 LLM。OPD 通过最小化教师和学生之间的反向 KL 散度来训练学生，该过程通过学生的策略生成的回滚（rollouts）进行。然而，在 OPD 中估计反向 KL 散度的梯度仍然是一个挑战。仅使用学生生成的回滚中的采样标记在计算上是便宜的，但提供的分布监督有限，这将降低准确性。此外，使用完整词汇表提供了完整的分布监督，但计算成本高。因此，最近的研究提出了 Top-$k$ OPD（TK-OPD），该方法使用选定的前 $k$ 个标记，提供比采样标记估计更丰富的分布监督，同时计算成本显著低于完整词汇表估计。不幸的是，仅使用选定的前 $k$ 个标记会引入偏差，导致准确性下降，因为选定的前 $k$ 个标记之外的概率质量被丢弃。为了解决 TK-OPD 的偏差问题，我们提出了尾部校正的 Top-$k$ 在线策略蒸馏（TT-OPD）。它保留了 TK-OPD 的优势，包括丰富的分布监督和低计算成本，同时提供反向 KL 散度梯度的无偏估计。TT-OPD 的关键见解是不仅使用选定的前 $k$ 个标记，还使用学生生成的回滚中的采样标记，从而在期望上恢复被丢弃的概率质量，避免偏差。实验结果表明，TT-OPD 在性能上显著优于其他测试的 OPD 变体。
cs.CL / 170 / 2609.34454
When Words Fall Short: Iterative Synergy Between Verbalized Reasoning and Hidden Features for LLM Confidence Estimation
当语言表达不足：语言化推理与隐藏特征之间的迭代协同用于大型语言模型的置信度估计
Abstract
Confidence estimation is crucial for developing trustworthy large language models (LLMs), with most methods following estimator-based or verbalization-based paradigms. While recent research increasingly focuses on improving verbalized self-reports of confidence, we challenge the prevailing view that this approach surpasses independent confidence estimators. Our empirical study shows that a dedicated confidence estimator can substantially outperform verbalized confidence, indicating that LLMs' internal representations contain richer confidence signals. Building on this finding, we propose Iterative Policy-Estimator Training (IPoET), a framework that synergizes the complementary strengths of verbalized reasoning traces and informative representations. IPoET alternates policy optimization with estimator updating, integrating estimator-derived confidence feedback into policy learning and refreshing the estimator on new policy rollouts. Experiments across diverse datasets and Qwen and Llama backbones demonstrate that, by iteratively exploiting richer hidden features and adapting to the evolving policy distribution, IPoET consistently outperforms both estimator- and verbalization-based baselines in-domain and achieves superior or comparable results across all out-of-domain metrics. For more details, refer to https://github.com/xyk829/ipoet.
Chinese Translation
置信度估计对于开发可信赖的大型语言模型（LLMs）至关重要，现有大多数方法遵循基于估计器或基于语言化的范式。尽管近期研究越来越关注改善语言化的自我报告置信度，但我们质疑这种方法超越独立置信度估计器的普遍观点。我们的实证研究表明，专用的置信度估计器可以显著优于语言化的置信度，这表明LLMs的内部表示包含更丰富的置信度信号。在此基础上，我们提出了迭代策略-估计器训练（Iterative Policy-Estimator Training, IPoET）框架，该框架协同利用语言化推理轨迹和信息丰富的表示的互补优势。IPoET交替进行策略优化和估计器更新，将估计器衍生的置信度反馈整合到策略学习中，并在新的策略回滚中刷新估计器。通过在不同数据集以及Qwen和Llama骨干网络上的实验表明，通过迭代利用更丰富的隐藏特征并适应不断变化的策略分布，IPoET在领域内始终优于基于估计器和语言化的基线，并在所有领域外指标上取得了更优或相当的结果。更多细节请参见 https://github.com/xyk829/ipoet。
cs.CL / 171 / 2609.34455
RGDT-Bench: Benchmarking LLM Reasoning for Rule-Governed Decisions and Their Justifications
RGDT-Bench：规则驱动决策及其理由的LLM推理基准测试
Abstract
We study reasoning in Rule-Governed Decision Tasks (RGDTs), where models apply external rules to case facts and justify decisions, as required in policy, contract, and compliance settings. Beyond the deductive capability emphasized by standard mathematical and logical reasoning tasks, RGDTs require interpreting rules and their applicability, assessing conditions from evidence, combining judgments under rules and exceptions, and providing checkable justifications. These demands motivate a benchmark assessing both decisions and their stated grounds. We introduce RGDT-Bench, providing 202.1K condition-level supervision slots across four task tracks and eight supported task-probe combinations that vary access to supporting information. Label-blind extraction and deterministic checks produce labels for warrant completeness: source-referenced coverage and consistency of stated decision grounds. The benchmark attributes failures to four process layers: rule use, condition, evidence, and aggregation, and checks the final outcome. Among evaluable correct responses, warrant incompleteness averages 40.2% across six evaluated LLMs and supported task-probe combinations. Such warrant incompleteness poses potential safety risks and remains difficult to detect: the best of seventeen existing evaluators reaches only 57.69% (random: 50%) task-averaged area under the receiver operating characteristic curve (AUROC). To address this difficulty, we train a simple reward model with warrant supervision. It achieves 69.24% task-averaged AUROC among correct answers, exceeding the matched outcome-supervised baseline by 10.37 pp (percentage points) and the best existing evaluator by 11.55 pp. Beyond completeness assessment, the model outperforms both outcome-supervised baselines across nearly all response-selection comparisons, supporting RGDT-Bench's warrant supervision for RGDT reasoning.
Chinese Translation
我们研究规则驱动决策任务（RGDTs）中的推理，其中模型将外部规则应用于案例事实并为决策提供理由，这在政策、合同和合规环境中是必需的。与标准数学和逻辑推理任务强调的演绎能力不同，RGDTs 需要解释规则及其适用性，从证据中评估条件，在规则和例外下结合判断，并提供可检查的理由。这些要求促使我们建立一个基准，评估决策及其所述依据。我们引入了 RGDT-Bench，提供 202.1K 条件级监督槽，涵盖四个任务轨道和八个支持的任务探测组合，这些组合在获取支持信息的方式上有所不同。无标签提取和确定性检查生成了关于理由完整性的标签：源引用覆盖率和所述决策依据的一致性。基准将失败归因于四个过程层次：规则使用、条件、证据和聚合，并检查最终结果。在可评估的正确响应中，六个评估的 LLM 和支持的任务探测组合的理由不完整性平均为 40.2%。这种理由不完整性可能带来安全风险，并且难以检测：现有十七个评估者中的最佳者仅达到 57.69%（随机：50%）的任务平均接收者操作特征曲线下面积（AUROC）。为了解决这一困难，我们训练了一个简单的奖励模型，采用理由监督。该模型在正确答案中的任务平均 AUROC 达到 69.24%，超过了匹配的结果监督基线 10.37 个百分点（pp）和最佳现有评估者 11.55 个百分点。除了完整性评估外，该模型在几乎所有响应选择比较中均优于结果监督基线，支持 RGDT-Bench 对 RGDT 推理的理由监督。
cs.CL / 172 / 2609.34481
CARDAMOM: A Micro-Dialectal Arabic Speech Dataset for ASR
CARDAMOM：用于自动语音识别的微方言阿拉伯语语音数据集
Talafha, Bashar, Magdy, Samar M., Alansari, Aisha, Alkhawaldeh, Alaa, Juma, Abdurrahman, Makahleh, Sharaf, Gamal, Nour, Attia, Omar, Kurdi, Hanaa, Rizk, Najwa, Anaya, Maysa, Altimyat, Hessah, Alhazmi, Layal, Alotaibi, Shumukh, Alhadaris, Hajar, Alomari, Rayan, Almalaq, Rahaf, Alkhorasani, Malak, alghamdi, Sara, Alshamrani, Rahaf, Ashraf, Nsrin, Jaradat, Ibrahim, Qardahji, Nada, Zaraket, Yasmin, Brahim, Elmoukhtar, Ebeidy, Sidi, Mahmoud, Oumoulmouminin, Khatraty, Yahjeb Bouha, Haroune, Meya, Ghaddar, Mohammad, Eldirany, Mohamad, Alamoush, Rashed, Chhaytle, Tala, Albadi, Nuha, Hadj, Yahya El, Luqman, Hamzah, Zaraket, Fadi A., Jarrar, Mustafa, Abdul-Mageed, Muhammad
Abstract
We present Cardamom, a micro-dialectal Arabic speech dataset designed to support fine-grained evaluation and adaptation of automatic speech recognition (ASR) systems. Community-curated by native speakers familiar with the represented varieties, Cardamom contains approximately 40 hours of transcribed YouTube speech spanning 21 micro-dialects across Egypt, Jordan, Lebanon, Mauritania, Palestine, and Saudi Arabia. Each segment is annotated with one or more operational micro-dialect labels, code-switching information, and utterance-level perceived gender, enabling analysis of sub-country variation that is obscured by conventional country-level labels. We describe the collection and annotation process, motivate the micro-dialect inventory linguistically, and benchmark four multilingual ASR systems in zero-shot and adapted settings. The strongest zero-shot system obtains 43.47% aggregate WER, with particularly high error rates on Mauritanian and Lebanese varieties; adaptation on Cardamom reduces its WER to 35.21%. Audio-based identification experiments further show that the annotations provide a learnable prediction target, with a dedicated classifier reaching 85.57% accuracy on 21-way micro-dialect identification. Cardamom provides a resource for studying localized dialectal variation and developing Arabic speech systems with broader regional coverage.
Chinese Translation
我们提出了Cardamom，这是一个微方言阿拉伯语语音数据集，旨在支持自动语音识别（ASR）系统的细粒度评估和适应。该数据集由熟悉所代表方言的母语者社区策划，包含约40小时的转录YouTube语音，涵盖埃及、约旦、黎巴嫩、毛里塔尼亚、巴勒斯坦和沙特阿拉伯的21种微方言。每个语音片段都标注有一个或多个操作性微方言标签、代码切换信息以及发话者性别的感知信息，从而使得对被传统国家级标签掩盖的国家内部变异进行分析成为可能。我们描述了数据的收集和标注过程，从语言学角度阐述微方言清单的动机，并在零样本和适应设置下对四个多语言ASR系统进行了基准测试。最强的零样本系统获得了43.47%的整体字错误率（WER），在毛里塔尼亚和黎巴嫩方言上错误率尤其高；在Cardamom上进行适应后，其WER降低至35.21%。基于音频的识别实验进一步表明，这些标注提供了一个可学习的预测目标，专门的分类器在21种微方言识别上达到了85.57%的准确率。Cardamom为研究地方方言变异和开发具有更广泛区域覆盖的阿拉伯语语音系统提供了资源。
cs.CL / 173 / 2609.34534
Papers Without Code: Availability of GitHub Repositories Linked in *CL Publications
无代码论文：与*CL出版物相关的GitHub仓库的可用性
Abstract
Source code and data published at computational linguistics (*CL) venues are increasingly being shared via GitHub. While this generally is a favourable development for the accessibility and potential reusability of research artifacts in natural language processing (NLP), the long-term availability of such repositories has not been evaluated. In this squib, we discuss the availability of repositories linked in papers published in the Computational Linguistics (CL) journal as well as at ACL and its co-located events over the past ten years. Contrary to our expectations, we find that GitHub repositories linked in more recent ACL publications are unavailable at similar rates as in older publications, in parts due to an increase in empty and placeholder repositories. Similar trends hold for other *CL venues, but not for platforms other than GitHub.
Chinese Translation
在计算语言学（*CL）领域发表的源代码和数据越来越多地通过GitHub进行共享。虽然这通常是自然语言处理（NLP）研究成果的可获取性和潜在可重用性方面的一个积极发展，但这些仓库的长期可用性尚未得到评估。在这篇短文中，我们讨论了过去十年中在计算语言学（CL）期刊以及ACL及其联合会议上发表的论文中链接的仓库的可用性。与我们的预期相反，我们发现最近的ACL出版物中链接的GitHub仓库的不可用率与较早的出版物相似，部分原因是空仓库和占位符仓库的增加。其他*CL场所也存在类似的趋势，但在GitHub以外的平台上则不然。
cs.CL / 174 / 2609.34556
RoPE is Dead, Long Live RoPE: Towards Scalable Data-aware Positional Encodings
RoPE已死，RoPE万岁：迈向可扩展的数据感知位置编码
Abstract
Transformers process tokens without any inherent notion of order, making positional encoding a fundamental requirement rather than an architectural refinement. Rotary Position Embedding (RoPE) has become the default positional encoding in modern language models, yet it is heavily biased toward nearby tokens. Existing alternatives have been evaluated under different settings, leaving the literature fragmented and without a clear replacement. We bring structure to this landscape by examining a specific weakness of RoPE: its slow frequency bands, whose wavelengths exceed the training context and expose models to unseen angles during extrapolation. We therefore introduce Data aware RoPE (DaRoPE), which preserves standard RoPE on the fast bands but replaces absolute position on the slow bands with bounded coordinates learned from contextual representations. Therefore, the slow-band geometry depends on the data rather than only on positional distance. We compare representative encodings under matched conditions across synthetic tasks, symbolic music, genomics, neural signals, and language models spanning 124M to 50B parameters. Across these experiments, DaRoPE leads on non-text benchmarks, mitigates recency bias, while remaining best or on par in language modeling and length extrapolation. Moreover, the learned coordinates also make the mechanism interpretable, revealing how attention layers leverage contextual information beyond token distance. Together, these results support DaRoPE as the best overall default among the evaluated methods, when there is no domain-specific reasons to prefer another.
Chinese Translation
变换器在处理令牌时并没有固有的顺序概念，使得位置编码成为一种基本需求，而非架构上的细微调整。旋转位置嵌入（Rotary Position Embedding, RoPE）已成为现代语言模型中的默认位置编码，但它对相邻令牌存在严重偏向。现有的替代方案在不同设置下进行了评估，导致文献碎片化且缺乏明确的替代方案。我们通过检查RoPE的一个特定弱点来为这一领域带来结构：其慢频带的波长超过训练上下文，并在外推过程中使模型暴露于未见的角度。因此，我们引入了数据感知RoPE（Data aware RoPE, DaRoPE），该方法在快速频带上保留标准RoPE，但在慢频带上用从上下文表示中学习到的有界坐标替代绝对位置。因此，慢频带的几何形状依赖于数据，而不仅仅是位置距离。我们在合成任务、符号音乐、基因组学、神经信号和参数范围从1.24亿到500亿的语言模型中，在匹配条件下比较了代表性的编码。在这些实验中，DaRoPE在非文本基准上表现优越，减轻了近期偏见，同时在语言建模和长度外推中保持最佳或持平。此外，学习到的坐标还使机制具有可解释性，揭示了注意力层如何利用超越令牌距离的上下文信息。综上所述，这些结果支持DaRoPE成为评估方法中最佳的默认选择，前提是没有特定领域的理由偏好其他方法。
cs.CL / 175 / 2609.34584
In-game Toxic Detection: Bi-directional Representations with Attention Residuals
游戏内有毒语言检测：带有注意力残差的双向表示
Abstract
In-game toxic language has emerged as a critical concern in the gaming industry and community. While several frameworks and models for online game toxicity analysis have been proposed, detecting toxicity in player chat utterances remains a formidable challenge: stemming not only from the extremely short length of such utterances but also from the heavy reliance on game slang, abbreviations, and domain-specific jargon, which generic language models are poorly suited to recognize. This paper presents a shared task for in-game toxic language detection built upon real-world in-game chat data, and proposes the best-preforming model for the toxic language slot filling: Bi-directional Representations with Attention Residuals (BRAR). Experimental results demonstrate that BRAR effectively captures the global context and outperforms the existing baselines on slot filling.
Chinese Translation
游戏内有毒语言已成为游戏行业和社区中的一个关键问题。尽管已经提出了多个在线游戏毒性分析的框架和模型，但在玩家聊天语句中检测毒性仍然是一项艰巨的挑战：这不仅源于这些语句的极短长度，还源于对游戏俚语、缩写和特定领域术语的高度依赖，而通用语言模型对此的识别能力较差。本文提出了一个基于真实游戏内聊天数据的游戏内有毒语言检测共享任务，并提出了用于有毒语言槽填充的最佳表现模型：带有注意力残差的双向表示（Bi-directional Representations with Attention Residuals, BRAR）。实验结果表明，BRAR有效捕捉了全局上下文，并在槽填充任务上优于现有基线。
cs.CL / 176 / 2609.34590
The Model Knows When to Stop: Training-Free Early Stopping for Long-Context Reading
模型知道何时停止：无训练的长上下文阅读早停方法
Abstract
Language models often process long inputs sequentially in chunks, but continuing to read after sufficient evidence has been acquired wastes computation. Existing stopping mechanisms either learn sufficiency from internal activations or train an exit gate, while a simpler alternative asks the model whether it has read enough. We introduce Answer-Convergence Stopping (ACS), a training-free stopping rule that measures rather than asks. After each chunk, it probes the frozen model's current answer state and stops when that state is both confident and stable. The rule requires only output-side generation and token log probabilities, has no trained components, and uses one shared configuration across models and benchmarks. Because a stopping policy can save computation simply by stopping too early, we evaluate the stopping decision itself using evidence position where available. On the full LongBench-v2 with two frontier models, ACS is the only stopping policy that matches or exceeds full-reading accuracy. Furthermore, across 250 S-NIAH questions, the premature stopping rate for ACS across five models from two families ranges from 0% to 12%, compared to 8.4% to 45.6% for the verbalized gate. Taken together, ACS reveals that by properly utilizing the output signals of frozen models, we can achieve favorable behaviors like adaptive stopping without the need for additional training.
Chinese Translation
语言模型通常以块的形式顺序处理长输入，但在获取足够证据后继续阅读会浪费计算资源。现有的停止机制要么通过内部激活学习充分性，要么训练一个退出门，而一个更简单的替代方案是询问模型是否已经阅读足够。我们提出了答案收敛停止（Answer-Convergence Stopping, ACS），这是一种无训练的停止规则，它测量而不是询问。在每个块之后，它探测冻结模型的当前答案状态，并在该状态既自信又稳定时停止。该规则仅需要输出侧生成和标记日志概率，没有训练组件，并且在模型和基准测试中使用一个共享配置。由于停止策略可以通过过早停止来节省计算，我们使用可用的证据位置评估停止决策。在完整的 LongBench-v2 上与两个前沿模型进行评估，ACS 是唯一一个与全阅读准确率相匹配或超过的停止策略。此外，在 250 个 S-NIAH 问题中，ACS 在来自两个家族的五个模型中的过早停止率范围为 0% 到 12%，而口头化门的过早停止率为 8.4% 到 45.6%。综合来看，ACS 显示通过合理利用冻结模型的输出信号，我们可以在无需额外训练的情况下实现自适应停止等有利行为。
cs.CL / 177 / 2609.34660
Rewarding Novel Deductions: Solver-guided Process Rewards for Logical Reasoning
奖励新颖推理：基于求解器引导的过程奖励用于逻辑推理
Abstract
Logical reasoning remains a major challenge for large language models (LLMs), particularly on structured problems that require precise constraint tracking, consistency preservation, and multi-step deduction. This challenge is especially acute for small-scale LLMs, which are more prone to producing inconsistent, redundant, or brittle reasoning trajectories. Existing approaches for improving logical reasoning largely optimize for final-answer correctness, providing only weak supervision over the intermediate reasoning process. In this work, we propose SPRING: (Solver-guided Process Rewards for Novel LogIcal ReasoNing Step Generation). SPRING uses SMT solver as a training-time verifier of intermediate reasoning steps to provide process-level supervision. It introduces the notion of a novel reasoning step, namely, a step that is logically valid, consistent with the evolving reasoning state, and not already implied by previously accepted non-contradictory deductions. Based on this solver-based assessment, it designs process rewards that encourage novel inferential progress while penalizing contradictory and uninformative reasoning steps. Evaluation across three logical reasoning benchmarks, ZebraLogic, AR-LSAT, and Knights and Knaves, and four LLMs shows that SPRING consistently outperforms base LLMs, outcome-only reward baselines, and Logic-LM. On ZebraLogic, SPRING improves puzzle accuracy by up to 49.71 and 15.43 points over the base LLM and strongest outcome-only baseline, respectively. On AR-LSAT, it improves overall accuracy by up to 64.93 and 12.14 points, respectively. On Knights and Knaves, SPRING achieves up to 93.14 puzzle accuracy and 96.05 person accuracy.
Chinese Translation
逻辑推理仍然是大型语言模型（LLMs）面临的一项重大挑战，尤其是在需要精确约束跟踪、一致性保持和多步骤推理的结构化问题上。这一挑战在小规模LLMs中尤为明显，因为它们更容易产生不一致、冗余或脆弱的推理轨迹。现有的提高逻辑推理的方法主要优化最终答案的正确性，对中间推理过程提供的监督较弱。在本研究中，我们提出了SPRING：（求解器引导的过程奖励用于新颖逻辑推理步骤生成）。SPRING利用SMT求解器作为训练时中间推理步骤的验证器，以提供过程级监督。它引入了新颖推理步骤的概念，即一个在逻辑上有效、与不断演变的推理状态一致，并且不被先前接受的非矛盾推理所暗示的步骤。基于这种基于求解器的评估，SPRING设计了过程奖励，鼓励新颖的推理进展，同时惩罚矛盾和无信息的推理步骤。在三个逻辑推理基准测试（ZebraLogic、AR-LSAT和Knights and Knaves）和四个LLMs上的评估表明，SPRING始终优于基础LLMs、仅基于结果的奖励基线和Logic-LM。在ZebraLogic上，SPRING相对于基础LLM和最强的仅基于结果的基线，分别提高了最多49.71和15.43的拼图准确率。在AR-LSAT上，整体准确率提高了最多64.93和12.14。在Knights and Knaves上，SPRING达到了最高93.14的拼图准确率和96.05的人物准确率。
cs.CL / 178 / 2609.34678
Fair Fact-Checking: Closing the Cross-Lingual Gap in LLM Factual Judgement with RoSh
公平事实核查：通过 RoSh 关闭 LLM 事实判断中的跨语言差距
Abstract
Misinformation on social media remains a critical problem, and more and more people settle it by asking a language model instead of a fact checker. Whether models judge such claims reliably is debated; whether they judge them equally well in every language people ask in has gone almost unasked. We test eight models from five families, 3B to 70B, on 1,500 encyclopedic factual claims that exist in identical form in eight languages. English is judged better than every other language on every model, and the gap is widest on the smallest ones, where Llama-3B on Arabic is no better than guessing. Existing remedies retrain on more multilingual data or fit an unconstrained map between language representations, and neither asks whether the model already holds the answer and simply fails to say it. It largely does: a linear probe recovers the truth from the very activations the model fails to express. We propose RoSh, a per-language shift and rotation of the residual stream, computed in closed form at three layers, with no training and no weight modified. It improves every model and closes 75% of the gap on average, helping most where the model was worst: Arabic on Llama-3B goes from chance to nearly the English level, and a fifth fewer of the claims answered correctly in English are lost in translation. What remains is no longer a read-out failure: afterwards the head recovers as much of what is encoded outside English as it does in English. An unconstrained map fitted on the same pairs falls below the untouched baseline, so the orthogonality constraint is doing the work, and every model clears a scrambled-correspondence control and ten further controls. On the two benchmarks of the closest inference-time method, latent-space intervention, run with its own data and metric code, RoSh's gains are five to thirteen times larger.
Chinese Translation
社交媒体上的虚假信息仍然是一个严重的问题，越来越多的人选择向语言模型而非事实核查员寻求解答。模型是否能够可靠地判断这些声明存在争议；而它们在不同语言中是否能够同样有效地进行判断几乎没有被提及。我们测试了来自五个家族的八个模型，参数规模从 3B 到 70B，针对 1,500 个在八种语言中以相同形式存在的百科事实声明。结果显示，英语在每个模型上的判断效果都优于其他语言，而在最小的模型上这一差距最为明显，Llama-3B 在阿拉伯语上的表现甚至不如随机猜测。现有的解决方案通常是在更多多语言数据上进行再训练，或在语言表示之间拟合一个不受约束的映射，但没有考虑模型是否已经持有答案却未能表达出来。实际上，模型确实如此：线性探测器能够从模型未能表达的激活中恢复出真相。我们提出了 RoSh，这是一种针对每种语言的残差流的平移和旋转方法，在三个层次上以封闭形式计算，无需训练和修改权重。该方法改善了每个模型的表现，平均关闭了 75% 的差距，尤其在模型表现最差的地方效果显著：Llama-3B 在阿拉伯语上的表现从随机猜测提升至接近英语水平，且在英语中正确回答的声明中，翻译损失减少了五分之一。剩下的部分不再是读取失败：之后，模型的头部能够恢复与英语编码相同数量的非英语信息。基于相同对的拟合的无约束映射表现低于未触及的基线，因此正交约束在其中发挥了作用，且每个模型都通过了混淆对应控制和十个进一步的控制。在与其自身数据和度量代码运行的最近推理时间方法的两个基准测试中，潜在空间干预，RoSh 的增益是五到十三倍。
cs.CL / 179 / 2609.34691
Using LLMs to Detect LLM-Generated Texts: A Cross-Generation Analysis
使用大型语言模型检测大型语言模型生成的文本：跨代分析
Abstract
Automated detection of LLM-generated texts (LGTs) is critical, yet dedicated detectors often struggle to generalize across domains and models. While general-purpose LLMs offer flexible zero-shot authorship classification with explanatory rationale, their detection behavior, especially regarding self-detection versus cross-detection across model generations, remains poorly understood. We systematically evaluate 15 LLMs spanning three model generations as both generators and detectors. Using a benchmark of 1,000 human-written texts and 15,000 LGTs (1,000 per model), we collected over 233,000 binary classifications alongside natural-language explanations. Our results reveal that detection efficacy is primarily driven by detector capability rather than generator provenance, although outputs from newer generators remain notably harder to detect. Crucially, statistical comparisons show no systematic advantage or disadvantage for self-detection across models. Error analysis further exposes generational bias shifts: first-generation detectors under-detect LGTs (high false-negative rates), second-generation detectors over-flag human texts (high false-positive rates), and the latest models achieve balanced trade-offs. Finally, we highlight significant inconsistencies in how different LLMs apply textual cues to justify their decisions. Code: https://github.com/hyyuan/detect-llm-generated-texts.
Chinese Translation
自动检测大型语言模型生成的文本（LGTs）至关重要，但专用检测器往往难以在不同领域和模型之间进行泛化。尽管通用大型语言模型提供灵活的零样本作者身份分类及解释性理由，但其检测行为，尤其是在模型代际之间的自我检测与交叉检测方面，仍然不够清晰。我们系统地评估了15个跨越三个模型代际的大型语言模型，作为生成器和检测器。通过使用1,000篇人类撰写的文本和15,000篇LGT（每个模型1,000篇）的基准，我们收集了超过233,000个二元分类及自然语言解释。我们的结果表明，检测效果主要受检测器能力的驱动，而非生成器来源，尽管来自新一代生成器的输出仍然显著更难以检测。关键的统计比较显示，各模型之间在自我检测方面没有系统性的优势或劣势。错误分析进一步揭示了代际偏差的变化：第一代检测器对LGT的检测不足（高假阴性率），第二代检测器对人类文本的误判过多（高假阳性率），而最新模型则实现了平衡的权衡。最后，我们强调了不同大型语言模型在应用文本线索以证明其决策时存在显著的不一致性。代码链接： https://github.com/hyyuan/detect-llm-generated-texts.
cs.CL / 180 / 2609.34717
ReMCTS: Reflection-Enhanced Monte Carlo Tree Search for Code Generation
ReMCTS：增强反射的蒙特卡洛树搜索用于代码生成
Abstract
Open-weight large language models (LLMs) can generate function-level programs from natural-language prompts, but plausible candidates still fail on hidden semantics and repeat mistakes across repair attempts. We present ReMCTS, an execution-grounded, memory-augmented, LLM-guided MCTS-style search framework. It organizes program candidates as tree states, retains branch-local debugging context, retrieves failure experience across branches, and distinguishes failed checks from unavailable evidence. On HumanEval and MBPP-Sanitized, visible-test ReMCTS improves over direct generation in 8 of 10 model-dataset pairs under held-out evaluation, whereas proxy-only search is less stable. Controlled tree-search, sampling, repair, and memory ablations characterize the source and limits of these gains. A 30-task HumanEval-X C++ pilot further demonstrates compatibility with compiler-backed execution, but does not constitute a broad multilingual evaluation.
Chinese Translation
开放权重的大型语言模型（LLMs）能够根据自然语言提示生成函数级程序，但在隐藏语义方面仍然存在失败的合理候选项，并且在修复尝试中反复出现错误。我们提出了ReMCTS，这是一种基于执行的、增强记忆的、由LLM引导的蒙特卡洛树搜索（MCTS）框架。它将程序候选组织为树状态，保留分支局部调试上下文，跨分支检索失败经验，并区分失败检查与不可用证据。在HumanEval和MBPP-Sanitized数据集上，ReMCTS在10对模型-数据集的保留评估中，在8对中优于直接生成，而仅基于代理的搜索则不够稳定。受控的树搜索、采样、修复和记忆消融实验表征了这些增益的来源和限制。一个包含30个任务的HumanEval-X C++试点进一步展示了与编译器支持的执行的兼容性，但并不构成广泛的多语言评估。
cs.CL / 181 / 2609.34738
Beyond Token Alignment: Event Completion for Cross-Tokenizer On-Policy Distillation
超越标记对齐：跨标记器的事件完成用于在线策略蒸馏
Abstract
On-policy distillation (OPD) transfers knowledge between language models through teacher supervision on student-generated trajectories. With different tokenizers, a single teacher token may require multiple student tokens to generate, creating intermediate states where the event is entered but not yet completed. Existing cross-tokenizer methods align tokens or text spans to construct comparable prediction targets. We study a complementary problem after partial generation: once the student produces a prefix of a teacher token, multiple next tokens may complete the same remaining bytes, but the teacher only specifies the required completion rather than how probability should be divided among these valid continuations. We introduce Event-Set Completion Distillation (ESCD), which complements cross-tokenizer probability alignment with completion-set supervision. ESCD aggregates prefix-related teacher events and supervises the total probability of byte-compatible one-step student completions, avoiding tokenizer-dependent probability splits among individual tokens. The method reuses student trajectories and predictions, requiring neither additional rollouts nor changes to the student vocabulary. Experiments demonstrate consistent gains in mathematics, code, and scientific reasoning across model families and tokenizers, extending to large-scale MoE distillation from a 1T teacher to a 35B student. Local analyses show that retaining completion sets better matches the reference supervision, while one-step completion covers over 99% of observed compatible teacher mass after partial event entry in the studied tokenizer pairs. These findings support event entry and event completion as complementary supervision targets for cross-tokenizer knowledge transfer. Code will be released on GitHub.
Chinese Translation
在线策略蒸馏（On-policy distillation, OPD）通过教师对学生生成轨迹的监督在语言模型之间转移知识。由于不同的标记器，单个教师标记可能需要多个学生标记来生成，从而产生中间状态，其中事件已被输入但尚未完成。现有的跨标记器方法对齐标记或文本片段以构建可比较的预测目标。我们研究了部分生成后的一个互补问题：一旦学生生成了教师标记的前缀，多个后续标记可能完成相同的剩余字节，但教师仅指定所需的完成，而不是如何在这些有效延续之间划分概率。我们引入了事件集完成蒸馏（Event-Set Completion Distillation, ESCD），它通过完成集监督补充了跨标记器的概率对齐。ESCD聚合与前缀相关的教师事件，并监督字节兼容的一步学生完成的总概率，避免了在单个标记之间依赖标记器的概率划分。该方法重用学生轨迹和预测，无需额外的展开或对学生词汇的更改。实验表明，在不同模型系列和标记器上，数学、代码和科学推理的表现均有一致提升，扩展到从1T教师到35B学生的大规模MoE蒸馏。局部分析显示，保留完成集更好地匹配参考监督，而一步完成在研究的标记器对中覆盖了超过99%的观察到的兼容教师质量。这些发现支持事件输入和事件完成作为跨标记器知识转移的互补监督目标。代码将在GitHub上发布。
cs.CL / 182 / 2609.34754
Draft-KV: Learning Useful Latent Communication Between Language Models
Draft-KV：学习语言模型之间有用的潜在通信
Abstract
Latent communication passes internal states between language models instead of decoded text, but higher receiver accuracy does not show that the receiver used the message content. Across five method-dataset pairs, replacing each message with one from an unrelated question changes accuracy by at most 0.60 points, even when communication adds 15.44 points over the receiver alone. Thus the interface can supply the gain while making the sharer dispensable. Draft-KV instead sends the key-value states formed while the sharer drafts an answer to the current question. Linear projections place these states in a side memory read through a gated attention branch, and progressive training moves from message reconstruction to answer supervision under a guard on harm from mismatched messages. Both models remain frozen and the interface trains 1.05M parameters, 348x fewer than C2C. With a Qwen3-8B sharer, a frozen Qwen2.5-0.5B-Instruct receiver reaches 78.04% on MMLU-Redux, versus 37.45% alone and 36.40% with reassigned messages. At fixed interface size, scaling the sharer from 0.6B to 8B raises accuracy from 46.11% to 78.04%; communication also transfers to held-out tasks and can exceed both models when each holds different evidence.
Chinese Translation
潜在通信在语言模型之间传递内部状态，而不是解码文本，但接收者的更高准确性并不表明接收者使用了消息内容。在五个方法-数据集对中，用一个不相关问题的消息替换每个消息，准确性最多变化0.60分，即使通信相较于单独接收者增加了15.44分。因此，该接口可以提供增益，同时使共享者变得可有可无。Draft-KV则在共享者为当前问题草拟答案时发送形成的键值状态。线性投影将这些状态放置在通过门控注意力分支读取的侧记忆中，渐进训练从消息重构转向在防止不匹配消息带来的伤害的保护下进行答案监督。两个模型保持冻结，接口训练105万参数，比C2C少348倍。使用Qwen3-8B共享者，一个冻结的Qwen2.5-0.5B-Instruct接收者在MMLU-Redux上达到78.04%的准确率，而单独使用时为37.45%，使用重新分配消息时为36.40%。在固定接口大小的情况下，将共享者从0.6B扩展到8B使准确率从46.11%提高到78.04%；通信也可以转移到保留任务，并且在每个模型持有不同证据时可以超过这两个模型。
cs.CL / 183 / 2609.34769
LongPuzzleBench: Evaluating GUI Agents on Long-Horizon Visual Puzzles
LongPuzzleBench：评估GUI代理在长时间视觉难题上的表现
Abstract
GUI agents need long-horizon visual reasoning: they must interpret a changing interface while keeping a multi-step plan viable as earlier actions constrain later ones. Existing benchmarks evaluate grounding, computer use, and game play, but rarely test whether agents stay coherent across long chains of coupled decisions. Long-horizon visual puzzles expose this capability directly: a legal move that looks like progress can make the puzzle unsolvable, and the loss shows only several moves later. We introduce LongPuzzleBench, 114 levels in six puzzle games played through native GUI actions, where one objective can take a human over a thousand actions on persistent boards and dead ends go unannounced. With Native GUI Actions alone, the strongest agents solve most objectives, but success falls sharply on harder, longer boards: seven of ten general-purpose agents solve nothing harder than Medium, and none completes Bolt Unscrew Hard, which a human solves along with every other objective. Code Execution CUA does not close this gap, and its scores mix visual solving with algorithmic search. Controlled diagnostics trace these failures to one limitation that neither rules, state hints, nor failure memory removes: agents judge each move by the visible progress it makes, not by the future options it leaves.
Chinese Translation
GUI代理需要长时间的视觉推理：它们必须在保持多步骤计划可行的同时解读不断变化的界面，因为早期的行动会限制后续的选择。现有基准评估了基础知识、计算机使用和游戏玩法，但很少测试代理在长链耦合决策中是否保持一致。长时间视觉难题直接暴露了这一能力：看似进展的合法移动可能使难题变得不可解，而损失通常在几步之后才显现。我们引入了LongPuzzleBench，这是一个包含六个难题游戏的114个关卡，通过本地GUI操作进行游戏，其中一个目标可能需要人类在持续的棋盘上进行超过一千次操作，而死胡同不会被提前告知。仅凭本地GUI操作，最强的代理能够解决大多数目标，但在更难、更长的棋盘上成功率急剧下降：十个通用代理中有七个无法解决比中等难度更高的难题，没有一个能够完成“Bolt Unscrew Hard”，而人类能够解决这个难题以及其他所有目标。代码执行CUA并未缩小这一差距，其得分将视觉解题与算法搜索混合在一起。受控诊断将这些失败追溯到一个限制，即规则、状态提示和失败记忆都无法消除：代理根据每一步所带来的可见进展来判断，而不是根据它留下的未来选项。
cs.CL / 184 / 2609.34770
Reference-Grounded Data Curation for Instruction-Following Thai-English Machine Translation
基于参考的指令遵循泰英机器翻译数据整理
Abstract
Instruction-following machine translation (IF-MT) requires respecting prompt-level rules on terminology, formatting, and register. Rule compliance typically trades off against translation quality, a tension that general-purpose IF data augmentation methods do not address. We propose Reference-Grounded Data Curation, a two-phase pipeline that extracts every supervised constraint from a reference translation that already satisfies it, ensuring feasibility by construction. Phase 1 applies Instruction-Following Difficulty (IFD) scoring to retain the hardest-but-learnable instances from an English-Thai parallel pool. Phase 2 extracts constraints from each reference target and keeps only generations satisfying every constraint, yielding the 1.97M-record Grounded dataset. We fine-tune open-weight bases on Grounded to produce ChindaMT, a Thai-English translation family at 4B, 2B, and 0.8B parameters. Under length-controlled pairwise judging, ChindaMT outperforms or matches every same-size baseline at every tier on both plain translation and under explicit rules, reaching up to a 68.4% win rate against the strongest baseline. The recipe transfers cleanly across Qwen generations. We release model weights, the Grounded dataset, and evaluation suites.
Chinese Translation
指令遵循机器翻译（IF-MT）需要遵循术语、格式和语域等提示级规则。规则遵循通常与翻译质量存在权衡，而通用的 IF 数据增强方法并未解决这一矛盾。我们提出了基于参考的 数据整理（Reference-Grounded Data Curation），这是一个两阶段的流程，从已经满足约束的参考翻译中提取每个监督约束，确保从构造上可行。第一阶段应用指令遵循难度（Instruction-Following Difficulty, IFD）评分，从英泰平行语料库中保留最难但可学习的实例。第二阶段从每个参考目标中提取约束，仅保留满足每个约束的生成结果，最终生成197万条记录的 Grounded 数据集。我们在 Grounded 上微调开放权重基础模型，生成 ChindaMT，这是一个包含 40 亿、20 亿和 8 亿参数的泰英翻译模型系列。在长度控制的成对评估中，ChindaMT 在普通翻译和明确规则下的每个层级上都超越或匹配了每个相同规模的基线，针对最强基线的胜率高达 68.4%。该方法在 Qwen 生成模型中也能顺利迁移。我们发布模型权重、Grounded 数据集和评估套件。
cs.CL / 185 / 2609.34783
TQTS-Bench: A Multi-Syntax Benchmark for Text-to-Query over Time-Series Databases
TQTS-Bench：一个针对时间序列数据库的多语法文本查询基准
Abstract
Large language models (LLMs) have significantly advanced natural language querying over relational databases, yet their ability to query time-series databases (TSDBs) remains largely unassessed. Existing benchmarks fail to adequately capture the non-unified query syntaxes, diverse application domains, and unique time-specific query intents inherent to TSDBs. To address this gap, we introduce TQTS-BENCH, a multi-syntax benchmark for evaluating text-to-query capabilities over TSDBs. TQTS-BENCH contains 6,125 high-quality question-answering (QA) pairs spanning 97 TSDBs, 23 distinct query syntaxes, 22 application domains, and 4 types of time-specific query intents. It is constructed through a human-centric AI-assisted workflow, where all QA pairs are carefully reviewed and revised by domain experts to ensure quality and correctness. Extensive evaluations of advanced LLMs and state-of-the-art text-to-query methods reveal challenges in querying TSDBs. Even the best-performing model evaluated, Claude-Opus-5, achieves only 48.98% execution accuracy, while humans reach 87.34%. Error analysis reveals that this performance gap mainly stems from the heterogeneous query syntaxes across different TSDBs, misinterpretation of time-specific intents, and incorrect schema linking. These findings highlight new opportunities to narrow the gap between current LLM capabilities and the requirements of TSDB queries in real-world applications. The benchmark is available at: https://anonymous.4open.science/r/TQTS-Bench-00CD.
Chinese Translation
大型语言模型（LLMs）在关系数据库的自然语言查询方面取得了显著进展，但它们在时间序列数据库（TSDBs）查询方面的能力仍然未得到充分评估。现有的基准未能充分捕捉到TSDBs固有的非统一查询语法、多样的应用领域以及独特的时间特定查询意图。为了解决这一问题，我们引入了TQTS-BENCH，这是一个用于评估TSDBs文本查询能力的多语法基准。TQTS-BENCH包含6,125对高质量的问答（QA）对，涵盖97个TSDBs、23种不同的查询语法、22个应用领域和4种类型的时间特定查询意图。该基准通过以人为本的AI辅助工作流程构建，所有QA对均由领域专家仔细审查和修订，以确保质量和准确性。对先进的LLMs和最先进的文本查询方法的广泛评估揭示了在查询TSDBs时面临的挑战。即使是评估中表现最佳的模型Claude-Opus-5，其执行准确率也仅为48.98%，而人类的准确率达到87.34%。错误分析表明，这一性能差距主要源于不同TSDBs之间的异构查询语法、对时间特定意图的误解以及不正确的模式链接。这些发现突显了缩小当前LLM能力与实际应用中TSDB查询需求之间差距的新机会。该基准可在以下网址获取：https://anonymous.4open.science/r/TQTS-Bench-00CD。
cs.CL / 186 / 2609.34798
InfiMed2: A Generalist Medical Multimodal Foundation Model from Contextual Evidence and Stability-Aware Supervision
InfiMed2：基于上下文证据和稳定性意识监督的通用医学多模态基础模型
Abstract
Recent medical multimodal models have benefited from larger corpora, broader modality coverage, and stronger reasoning-oriented training, yet effective data design across continued pretraining (CPT) and post-training remains challenging. Medical sources vary substantially in structure, granularity, and information density, and their utility shifts as training progresses from broad knowledge acquisition to late-stage consolidation. Meanwhile, post-training is often dominated by short-form visual question answering, providing limited supervision for informative and answer-consistent explanations. We introduce InfiMed2, a family of 4B and 27B generalist medical multimodal foundation models built around stage-aware data design. We curate a 55.68B-token corpus that combines broad clinical knowledge with context-rich biomedical visual evidence through source-specific processing. Our CPT pipeline first adapts the vision encoder, then builds broad medical knowledge, and finally transitions to an evidence-focused data mixture during learning-rate decay. For supervised fine-tuning (SFT), we regenerate visual question-answering responses using answer stability, answer-masked reconstruction, and correctness-constrained selection to produce more informative and answer-consistent supervision. The 4B model is further optimized with reinforcement learning with verifiable rewards (RLVR). Across five medical multimodal benchmarks, InfiMed2-4B achieves 66.73% mean accuracy after RLVR, surpassing the larger Qwen3.5-9B, while InfiMed2-27B reaches 73.72%, the highest among the evaluated open-weight models.
Chinese Translation
近年来，医学多模态模型受益于更大的语料库、更广泛的模态覆盖和更强的推理导向训练，但在持续预训练（CPT）和后训练阶段的有效数据设计仍然具有挑战性。医学来源在结构、粒度和信息密度上差异显著，随着训练从广泛知识获取转向后期巩固，其效用也在变化。同时，后训练阶段通常以短形式的视觉问答为主，提供的信息监督有限，难以形成信息丰富且答案一致的解释。我们提出了InfiMed2，一个围绕阶段感知数据设计构建的4B和27B通用医学多模态基础模型系列。我们策划了一个包含55.68B标记的语料库，通过源特定处理将广泛的临床知识与丰富上下文的生物医学视觉证据相结合。我们的CPT流程首先适应视觉编码器，然后构建广泛的医学知识，最后在学习率衰减期间过渡到以证据为中心的数据混合。在监督微调（SFT）阶段，我们利用答案稳定性、答案掩蔽重建和正确性约束选择重新生成视觉问答响应，以产生更具信息性和答案一致性的监督。4B模型进一步通过可验证奖励的强化学习（RLVR）进行优化。在五个医学多模态基准测试中，InfiMed2-4B在RLVR后达到66.73%的平均准确率，超越了更大的Qwen3.5-9B，而InfiMed2-27B则达到73.72%，在评估的开放权重模型中最高。
cs.CL / 187 / 2609.34800
Pass or Fail? Evaluating LLMs on Two Greek Examination Benchmarks
通过还是失败？在两个希腊考试基准上评估大型语言模型
Abstract
The rapid advancement of Large Language Models (LLMs) imposes a thorough evaluation of their linguistic and analytical capabilities as well as constraints, particularly for a language with limited benchmark coverage such as Greek. To address the limited availability of comprehensive benchmarks in this domain, we introduce Prot-Ex and Pan-Ex, two benchmarks consisting of questions from entrance exams for Greek Model and Experimental schools as well as the Panhellenic exams (the Greek national university entrance examinations). These benchmarks are employed to assess the performance of text-only LLMs-including the Greek-adapted KriKri-8B-Instruct, Llama-3.1-8B, Gemma-4-26B, and Qwen-3-32B-across diverse academic disciplines (Modern Greek, Mathematics, Physics, etc.) and task formats (closed, structured, and open-ended), including textualized visual context (i.e., image descriptions). Our findings indicate the localized KriKri-8B significantly outperforms its base model, successfully rivalling much larger LLMs in linguistically demanding humanities tasks. By leveraging an LLM-as-a-Judge methodology, we expose the inadequacy of traditional lexical metrics for evaluating complex reasoning. Crucially, we uncover a few-shot prompting paradox: while synthetic examples improve accuracy in closed-ended questions, they severely overload the context window of 8B models in structured tasks, causing significant performance degradation. Ultimately, this study suggests targeted linguistic adaptation offsets lower parameter counts in specialized domains, despite the fragility of smaller models to prompt verbosity.
Chinese Translation
大型语言模型（LLMs）的快速发展要求对其语言和分析能力以及局限性进行全面评估，尤其是对于像希腊这样基准覆盖有限的语言。为了解决这一领域综合基准的有限可用性，我们引入了Prot-Ex和Pan-Ex，这两个基准由希腊模型和实验学校的入学考试问题以及泛希腊考试（希腊国家大学入学考试）的问题组成。这些基准用于评估仅文本的LLMs的表现，包括经过希腊适配的KriKri-8B-Instruct、Llama-3.1-8B、Gemma-4-26B和Qwen-3-32B，涵盖现代希腊语、数学、物理等多个学科和任务格式（封闭式、结构化和开放式），包括文本化的视觉上下文（即图像描述）。我们的研究结果表明，经过本地化的KriKri-8B显著优于其基础模型，在语言要求较高的人文学科任务中成功与更大型的LLMs竞争。通过利用LLM作为评判者的方法，我们揭示了传统词汇度量在评估复杂推理时的不足。重要的是，我们发现了一个少量提示悖论：虽然合成示例提高了封闭式问题的准确性，但它们在结构化任务中严重超载了8B模型的上下文窗口，导致显著的性能下降。最终，本研究建议有针对性的语言适应可以弥补专门领域中较低参数数量的不足，尽管较小模型对提示冗长性较为脆弱。
cs.CL / 188 / 2609.34829
From Weak Task Specifications to Scientific Extraction Agents: Optimizing Task Construction
从弱任务规范到科学提取代理：优化任务构建
Abstract
Most methods that optimize LLM prompts and agent workflows assume that task-specific output schemas, extraction instructions, and evaluation criteria are predefined. For scientific extraction agents, however, a short task goal may not fully determine these components, while specifying them manually is costly. We study the upstream problem of constructing the task-specific configuration from a weak specification containing only a short goal and unannotated reference documents. Rather than treating automatic construction as a fixed preprocessing step, our framework constructs a task-specific schema, extraction instructions, and base training rubrics, then keeps schema construction and extraction instructions editable during optimization. Failure-focused updates concentrate textual-gradient feedback on lower-scoring documents, while training-time evaluation criteria adapt to recurring failures. On a heterogeneous-catalysis literature corpus, automatic construction remains improvable, and optimizing both schema construction and extraction instructions performs best across all four judge-rubric settings, with ablations and blinded human evaluation supporting the proposed formulation.
Chinese Translation
大多数优化大型语言模型（LLM）提示和代理工作流程的方法假设任务特定的输出模式、提取指令和评估标准是预先定义的。然而，对于科学提取代理而言，短期任务目标可能无法完全确定这些组件，而手动指定它们的成本又很高。我们研究了从仅包含短期目标和未标注参考文档的弱规范构建任务特定配置的上游问题。我们的框架并不将自动构建视为固定的预处理步骤，而是构建任务特定的模式、提取指令和基础训练标准，并在优化过程中保持模式构建和提取指令可编辑。以失败为中心的更新将文本梯度反馈集中在得分较低的文档上，而训练时的评估标准则适应于反复出现的失败。在一个异质催化文献语料库上，自动构建仍然可以改进，优化模式构建和提取指令在所有四个评审标准设置中表现最佳，消融实验和盲评支持了所提出的框架。
cs.CL / 189 / 2609.34839
OpenWhistle: A Large-Scale Longitudinal Dataset and Benchmark of Bottlenose Dolphin Vocalizations
OpenWhistle：海豚鸣叫的大规模纵向数据集与基准
Abstract
Recent advances in bioacoustics have been driven by large-scale corpora and standardized benchmarks, yet existing resources are overwhelmingly bird-centric and shallow per species, limiting their use for studying the structure of a single species' communication system. This gap is particularly acute for cetaceans: despite bottlenose dolphins (Tursiops truncatus) being a compelling case of complex vocal communication among non-human mammals, existing dolphin datasets are small, fragmented, and largely closed. We introduce OpenWhistle, the largest publicly available dataset of dolphin vocalizations. It comprises approximately 180,000 whistles (114 hours) recorded over five years from a stable pod of five individuals in a semi-natural environment, paired with a curated subset of 8,354 expert-annotated whistles and reproducible evaluation protocols for whistle-type detection and classification. We further release the full processing pipeline for whistle detection, segmentation, and categorization. To demonstrate its utility, we pretrain a Wav2Vec2.0 model adapted to dolphin acoustics on the OpenWhistle corpus and show that it learns effective representations, outperforming general-purpose bioacoustic models such as AVES and BioLingual on both tasks while leaving meaningful headroom for future work. By releasing the dataset, pipeline, and evaluation protocol, we provide the first open dolphin whistle dataset tailored for training self-supervised models, laying the groundwork for advancing dolphin communication research and developing models that capture fine-grained acoustic structure within species.
Chinese Translation
近年来，生物声学的进展得益于大规模语料库和标准化基准，然而现有资源主要集中于鸟类，并且每个物种的深度有限，这限制了它们在研究单一物种的交流系统结构方面的应用。这一缺口在鲸类中特别明显：尽管宽吻海豚（Tursiops truncatus）是非人类哺乳动物中复杂鸣叫交流的一个引人注目的案例，但现有的海豚数据集规模小、碎片化且大多封闭。我们推出了OpenWhistle，这是最大的公开可用海豚鸣叫数据集。该数据集包含约180,000个鸣叫（114小时），在五年的时间里从一个稳定的五只个体的群体中记录，配有一个经过精心策划的8,354个专家注释的鸣叫子集以及可重复的鸣叫类型检测和分类评估协议。我们进一步发布了鸣叫检测、分割和分类的完整处理流程。为了展示其实用性，我们在OpenWhistle语料库上预训练了一个适应海豚声学的Wav2Vec2.0模型，并显示其学习到了有效的表征，在两个任务上超越了通用生物声学模型，如AVES和BioLingual，同时为未来的工作留出了有意义的提升空间。通过发布数据集、处理流程和评估协议，我们提供了第一个针对训练自监督模型的开放海豚鸣叫数据集，为推动海豚交流研究和开发捕捉物种内细微声学结构的模型奠定了基础。
cs.CL / 190 / 2609.34841
Adapt Semantics, Not Structure: Few-Instance Schema Calibration for Scientific PDF Extraction
适应语义，而非结构：科学PDF提取的少实例模式校准
Abstract
A well-designed extraction schema is not necessarily ready for reliable LLM execution. When only limited verified extractions are available, manually tuning hundreds of field definitions through trial and error is costly. We frame this problem as few-instance schema calibration: adapting the operational semantics of an existing schema from a few annotated documents while preserving its structural contract. We introduce CPSE, a contract-preserving semantic extraction framework that jointly calibrates extraction prompts and field-level semantic descriptions from a few gold annotations. CPSE decomposes the schema into an invariant structural contract and mutable field semantics, and further separates identity discovery from record completion using manifest-conditioned resolution. On expert-annotated polymer-science documents, CPSE improves extraction by 9.93 points over an execution-matched baseline, with consistent gains under an independent judge and in a blinded expert audit. These results show that CPSE enables low-resource schema execution while preserving the output structure required downstream.
Chinese Translation
一个设计良好的提取模式并不一定适合可靠的LLM执行。当仅有有限的经过验证的提取结果时，通过反复试验手动调整数百个字段定义是非常昂贵的。我们将这个问题框架化为少实例模式校准：在保留结构契约的同时，从少量标注文档中调整现有模式的操作语义。我们引入了CPSE（契约保持语义提取框架），该框架通过少量金标准标注联合校准提取提示和字段级语义描述。CPSE将模式分解为不变的结构契约和可变的字段语义，并通过显式条件解析进一步将身份发现与记录完成分离。在专家标注的聚合物科学文档上，CPSE在与执行匹配的基线相比，提取性能提高了9.93个百分点，并在独立评审和盲审专家审核中保持了一致的提升。这些结果表明，CPSE能够在低资源条件下实现模式执行，同时保留下游所需的输出结构。
cs.CL / 191 / 2609.34936
Neural Language Models Learn the Contextual Distributions of Dependency Structures: a statistical learning theory to compositionality
神经语言模型学习依赖结构的上下文分布：一种关于组合性的统计学习理论
Abstract
It is unclear how Neural Language Models (NLMs) acquire the structural meaning encoded by grammatical structures that is independent of lexical semantics. We propose a statistical learning process in which learned dependency structures themselves become new distributional units for subsequent statistical learning. Under this account, once a dependency structure is acquired, the model tracks its contextual distributions. These contextual features reflect the semantic properties of a composite structure. To test this hypothesis, we design a synthetic grammar in which each grammatical structure has distinct contextual distributions that cannot be recovered from the distributional statistics of their component tokens alone. We train a series of BERT-style masked language models on this grammar and examine their developmental trajectory. The results show that models can successfully learn the contextual distributions of composite dependency structures even though they cannot be inferred from token statistics alone. Developmental analysis further reveals a clear developmental trajectory. The learning of the dependency relations that define a grammatical structure consistently precedes the learning of its contextual features. These findings suggest that statistical learning in NLMs is not merely the accumulation of token co-occurrence statistics, but a process in which learned dependency structures become new units of distributional learning. We argue that this process provides a statistical-learning account of how NLMs solve the compositionality problem in language. Finally, we discuss the possibility that this statistical learning process provides an explanatory theory on how language cognition could emerge from pure distributional statistics.
Chinese Translation
目前尚不清楚神经语言模型（NLMs）如何获取由语法结构编码的结构意义，而这些意义与词汇语义无关。我们提出了一种统计学习过程，其中学习到的依赖结构本身成为后续统计学习的新分布单元。在这一理论框架下，一旦获得依赖结构，模型便会跟踪其上下文分布。这些上下文特征反映了复合结构的语义属性。为了验证这一假设，我们设计了一种合成语法，其中每个语法结构具有独特的上下文分布，这些分布无法仅通过其组成标记的分布统计来恢复。我们在该语法上训练了一系列基于BERT的掩码语言模型，并考察它们的发展轨迹。结果表明，模型能够成功学习复合依赖结构的上下文分布，尽管这些分布无法仅通过标记统计推断。发展分析进一步揭示了明确的发展轨迹。定义语法结构的依赖关系的学习始终先于其上下文特征的学习。这些发现表明，NLMs中的统计学习不仅仅是标记共现统计的积累，而是一个学习到的依赖结构成为新的分布学习单元的过程。我们认为这一过程为NLMs如何解决语言中的组合性问题提供了统计学习的解释。最后，我们讨论了这一统计学习过程可能为语言认知如何从纯粹的分布统计中产生提供解释理论的可能性。
cs.CL / 192 / 2609.34967
Semantic Uncertainty Quantification Needs Factual Equivalence
语义不确定性量化需要事实等价性
Abstract
Semantic uncertainty quantification for large language models rests on a common template: sample several answers, measure how much they agree, and treat disagreement as uncertainty. We first formalize this template as two separate roles: an operator that compares two answers, and an aggregator that combines all pairwise comparisons into a scalar. Existing methods differ almost entirely in how they aggregate, while taking the operator off the shelf, typically an NLI model or a generic sentence encoder. We show that this reliance on off-the-shelf operators is the primary bottleneck of semantic UQ: they do not accurately measure factual equivalence of multiple answers to the same question. We resolve this with a deliberately simple recipe: a single encoder trained contrastively to isolate the targeted fact, utilizing synthetic data generated by an LLM and dataset both disjoint from all evaluation settings. Integrating the resulting operator into existing methods improves performance on 120 of 126 evaluation settings (95%) spanning 18 model dataset combinations across language and vision-language models. The best variant reaches 0.76 mean AUROC against 0.68 for the strongest baseline, while replacing the quadratic cross-encoder comparisons of entailment-based operators with one encoder pass per answer. The uniformity of the improvement supports the view that the operator, not the aggregator, is the limiting factor. The same operator also improves single generation token-level estimators: the norm it assigns to each token measures how much that token bears on the answer, and reweighting token log-likelihoods accordingly sharpens the estimate.
Chinese Translation
大型语言模型的语义不确定性量化基于一个共同的模板：采样多个答案，测量它们的一致性，并将不一致视为不确定性。我们首先将这个模板形式化为两个独立的角色：一个比较两个答案的操作符和一个将所有成对比较合并为标量的聚合器。现有方法几乎完全在于如何聚合，而通常采用现成的操作符，通常是自然语言推理（NLI）模型或通用句子编码器。我们展示了这种对现成操作符的依赖是语义不确定性量化的主要瓶颈：它们无法准确测量多个答案对同一问题的事实等价性。我们通过一个故意简单的方案解决了这个问题：一个经过对比训练的单一编码器，用于隔离目标事实，利用由大型语言模型（LLM）生成的合成数据和与所有评估设置不重叠的数据集。将生成的操作符整合到现有方法中，提升了在126个评估设置中的120个（95%）的性能，涵盖了跨语言和视觉语言模型的18种模型数据集组合。最佳变体的平均AUROC达到0.76，而最强基线为0.68，同时将基于蕴含的操作符的二次交叉编码器比较替换为每个答案一次编码的方式。改进的一致性支持了操作符而非聚合器是限制因素的观点。相同的操作符也改善了单生成的令牌级估计器：它为每个令牌分配的范数衡量该令牌对答案的影响程度，并相应地重新加权令牌的对数似然，从而提高了估计的准确性。
cs.CL / 193 / 2609.34986
N\"urnberg NLP at ChildSafeAds 2026: Structurally Dissimilar Voter Ensembles under Four Levels of Data Access
纽伦堡NLP在ChildSafeAds 2026中的表现：四种数据访问级别下结构上不相似的投票集
Abstract
We describe the N\"urnberg NLP system for ChildSafeAds 2026. The shared task asks what a monitoring system for commercial content in child-facing YouTube videos can achieve at a given level of data access. We answer with per-subtask ensembles of nine voters, organised into three branches that differ in backbone, adaptation method and class scope. Selection rests on channel-disjoint cross-validation, with the development set as a transfer check. The system wins two of the three subtasks. Its product-category score (ST2, 0.8243) and its compliance-flag score (ST3, 0.6530) are the best of the 22 final entries, and it places third on the task mean (0.7079). We further compare four access levels and report the cost at test-set scale.
Chinese Translation
我们描述了纽伦堡NLP系统在ChildSafeAds 2026中的应用。该共享任务探讨了在特定数据访问级别下，针对儿童的YouTube视频中商业内容的监测系统能够实现的目标。我们通过九个投票者的每个子任务集成来回答这一问题，这些投票者被组织成三个分支，分别在骨干网络、适应方法和类别范围上存在差异。选择基于频道不重叠的交叉验证，开发集作为转移检查。该系统在三个子任务中赢得了两个。其产品类别得分（ST2，0.8243）和合规标志得分（ST3，0.6530）在22个最终参赛作品中均为最佳，并且在任务均值（0.7079）中排名第三。我们进一步比较了四个访问级别，并报告了测试集规模下的成本。
cs.CL / 194 / 2609.34988
The Right Lesson at the Right Step: Deriving Control Updates for Self-Evolving Agents
在正确的步骤上获得正确的教训：为自我进化代理推导控制更新
Abstract
Self-evolving agents improve future behavior by reusing past experience, typically as global prompts, memories, or reflections. Yet these mechanisms rarely control where experience takes effect. In long tool-use workflows, the same lesson may correct one decision but distract another, making experience reuse a problem of localized control rather than memory alone. We introduce EvoCUE (Evolution through Control Updates from Evidence), a framework for learning reusable control-program updates from completed agent executions. EvoCUE represents the agent as an explicit state-machine controller, whose nodes perform model or tool calls and whose edges define where control passes next. This makes the workflow editable at precise locations, so each learned update can specify what to add, where it acts, and when it applies. From completed trajectories, EvoCUE uses residual goals and observed execution traces to propose localized instruction or skill edits. Each candidate is evaluated at the point where it would act by resuming the parent and edited controllers from the same checkpoint and comparing their final outcomes. Accepted edits are compiled with applicability rules, confirmed on held-out tasks, and inherited by later executions. We evaluate EvoCUE on long tool-use environments where learned conventions must reach the right execution step. From a minimal AppWorld controller without benchmark-specific onboarding instructions, EvoCUE learns the missing task-completion convention and substantially improves success on Test-Normal and Test-Challenge. On PAST-Bench office workflows, EvoCUE transfers organizational requirements from prior episodes to later tasks, improving task-execution quality. These results show that self-evolving agents should place experience inside the control flow, rather than only store it as text.
Chinese Translation
自我进化代理通过重用过去的经验来改善未来的行为，通常以全球提示、记忆或反思的形式存在。然而，这些机制很少控制经验的影响位置。在长时间的工具使用工作流程中，同样的教训可能会纠正一个决策却分散另一个决策的注意力，使得经验重用成为一个局部控制的问题，而不仅仅是记忆问题。我们引入了EvoCUE（通过证据的控制更新进化），这是一个从完成的代理执行中学习可重用控制程序更新的框架。EvoCUE将代理表示为一个显式的状态机控制器，其节点执行模型或工具调用，边缘定义控制的下一步传递位置。这使得工作流程在精确位置上可编辑，因此每个学习到的更新可以指定添加内容、作用位置和适用时机。通过已完成的轨迹，EvoCUE利用残余目标和观察到的执行轨迹来提出局部指令或技能编辑。每个候选编辑在其作用点进行评估，通过从同一检查点恢复父控制器和编辑后的控制器并比较其最终结果。被接受的编辑与适用性规则一起编译，在保留的任务上得到确认，并由后续执行继承。我们在长时间的工具使用环境中评估EvoCUE，在这些环境中，学习到的惯例必须达到正确的执行步骤。在没有基准特定入门指令的最小AppWorld控制器中，EvoCUE学习到缺失的任务完成惯例，并显著提高了在Test-Normal和Test-Challenge上的成功率。在PAST-Bench办公室工作流程中，EvoCUE将先前情节中的组织要求转移到后续任务中，提高了任务执行质量。这些结果表明，自我进化代理应将经验置于控制流程中，而不仅仅是将其存储为文本。
cs.CL / 195 / 2609.35070
Echoes of Deeds: Moral History Can Shape and Steer LLM Behavioral Choices
行为的回声：道德历史如何塑造和引导大型语言模型的行为选择
Abstract
Evaluations of Large Language Models (LLMs) morality typically consider decisions in isolation, thus overlooking whether an individual's unrelated prior conduct influences the model's subsequent choices. This leaves open the question of whether, and to what extent, moral history shapes LLM decisional behaviors. Prior work on human moral decision-making shows that past behavior can influence subsequent moral choices. Building on this observation, we investigate whether analogous effects emerge in LLMs in two complementary ways: at the behavioral level, through the model's observable responses, and at the representation level, through its latent internal representations. We introduce MoralLedger, a framework for studying how an actor's moral history shapes actions for LLMs' behaviors under a fixed decision context. At the behavioral level, we find that prior moral histories systematically alter subsequent choices as a function of their valence and intensity. At the internal representation level, these histories induce a linearly recoverable direction in the residual stream that generalizes to held-out examples. Intervening along this direction on neutral-history prompts produces two-sided intensity-dependent changes in subsequent choices, with effects that are stronger than those induced by prompting alone or by favorable-nonmoral direction. To our knowledge, this is the first demonstration that a latent representation of an actor's prior moral conduct can provide signed inference-time control over a moral decision. Our MoralLedger extends moral evaluation beyond static dilemmas, establishing moral history as both a source of behavioral sensitivity and a causal target for auditing and controlling moral behavior in LLMs.
Chinese Translation
对大型语言模型（LLMs）道德性的评估通常将决策视为孤立的过程，因此忽视了个体无关的先前行为是否会影响模型后续的选择。这引发了一个问题，即道德历史在多大程度上影响LLM的决策行为。先前关于人类道德决策的研究表明，过去的行为可以影响后续的道德选择。在此基础上，我们探讨了在LLMs中是否会出现类似的效应，采用了两种互补的方法：在行为层面，通过模型的可观察响应；在表征层面，通过其潜在的内部表征。我们引入了MoralLedger，这是一个研究行为者的道德历史如何在固定决策背景下塑造LLMs行为的框架。在行为层面，我们发现先前的道德历史系统性地改变了后续选择，具体取决于其价值和强度。在内部表征层面，这些历史在残余流中诱导出一个线性可恢复的方向，该方向可以推广到未见示例。在中性历史提示上沿该方向进行干预，会产生双向强度依赖的后续选择变化，其效果强于仅通过提示或有利非道德方向所引发的效果。据我们所知，这是首次证明行为者先前道德行为的潜在表征能够在道德决策中提供有符号的推理时控制。我们的MoralLedger将道德评估扩展到静态困境之外，确立了道德历史作为行为敏感性的来源和审计及控制LLMs道德行为的因果目标。
cs.CL / 196 / 2609.35074
When Confidence Rises Too Early: Detecting Shortcut Reasoning via Premature Answer Commitment
当信心过早上升时：通过过早的答案承诺检测捷径推理
Abstract
The reasoning trajectory of a Large Language Model (LLM) is often treated as a verbalized description of its internal reasoning. However, such trajectories can be unfaithful: a model may rely on shortcuts to reach an answer and then post-rationalize the decision with a seemingly coherent chain of thought. Detecting this shortcut reasoning is challenging because existing monitors and verifiers mainly inspect textual traces or final outcomes, rather than how the model's belief in its answer develops during generation. We introduce ConfLens, a framework that tracks the evolution of confidence in the final answer throughout reasoning. Across three shortcut reasoning settings, we observe a common pattern of premature confidence, where shortcut samples become highly confident in the final answer at early reasoning stages. Existing confidence estimation methods, however, show limited generalizability, reliability, or efficiency for detecting this behavior. We therefore propose the Distributional Answer Commitment Score (DACS), a distributional confidence estimator that measures the entropy of the model's probability distribution over answer commitment at each reasoning step. DACS captures how concentrated the model's answer belief is without requiring ground-truth answers or task-specific verifiers. We further convert ConfLens detection results into interpretable signals for reward models to reduce their preference for shortcut reasoning. Experiments on mathematical and code reasoning tasks show that ConfLens with DACS improves shortcut reasoning detection by over 4.3% F1 compared with strong baselines and reduces the mismatch between faithfulness and correctness in reward model preferences.
Chinese Translation
大型语言模型（LLM）的推理轨迹通常被视为其内部推理的口头描述。然而，这些轨迹可能并不真实：模型可能依赖捷径来得出答案，然后用看似连贯的思路对决策进行后期理性化。检测这种捷径推理具有挑战性，因为现有的监测器和验证器主要检查文本痕迹或最终结果，而不是模型在生成过程中对其答案的信念如何发展。我们引入了ConfLens，一个跟踪最终答案信心演变的框架。通过三种捷径推理设置，我们观察到一种共同的过早信心模式，其中捷径样本在推理的早期阶段对最终答案表现出高度信心。然而，现有的信心估计方法在检测这种行为时显示出有限的可推广性、可靠性或效率。因此，我们提出了分布式答案承诺评分（Distributional Answer Commitment Score, DACS），这是一种分布式信心估计器，测量模型在每个推理步骤中对答案承诺的概率分布的熵。DACS捕捉模型答案信念的集中程度，而无需真实答案或特定任务的验证器。我们进一步将ConfLens检测结果转换为可解释的信号，以减少奖励模型对捷径推理的偏好。在数学和代码推理任务上的实验表明，使用DACS的ConfLens相比强基线提高了超过4.3%的F1捷径推理检测，并减少了奖励模型偏好中的真实性和正确性之间的不匹配。
cs.CL / 197 / 2609.35108
A mechanistic study of language model introspection
语言模型内省的机制研究
Abstract
Large language models (LLMs) can sometimes report perturbations to their internal activations---even when the input provides no evidence that an intervention occurred. How do models detect and localize such internal changes? We study this question using a controlled task that keeps the input text fixed. We either inject a concept vector into the hidden state at one of ten token positions or apply no intervention. The model is asked to identify the perturbed position or report that no intervention occurred. Across three model families, we identify two small groups of attention heads with distinct roles in introspective reporting. Middle-layer gate heads influence whether the model reports a change, while router heads in a later layer help select the position to report. Interventions on gate heads can suppress position reports even when router heads supply location information. We further examine why reporting accuracy varies across concepts. Concept vectors that are localized more accurately produce stronger attention-score and output responses in gate heads, which is associated with better alignment of the induced key and value changes in their QK and OV computations. Together, these findings identify attention-head mechanisms supporting introspective detection and localization.
Chinese Translation
大型语言模型（LLMs）有时能够报告其内部激活的扰动——即使输入并未提供任何干预发生的证据。模型是如何检测和定位这些内部变化的？我们通过一个控制任务来研究这个问题，该任务保持输入文本不变。我们在十个标记位置中的一个注入概念向量，或者不进行任何干预。模型被要求识别扰动位置或报告没有发生干预。在三种模型家族中，我们识别出两小组具有不同内省报告角色的注意力头。中间层的门控头影响模型是否报告变化，而后层的路由头则帮助选择报告的位置。对门控头的干预可以抑制位置报告，即使路由头提供位置信息。我们进一步研究了为何报告准确性在不同概念间存在差异。定位更准确的概念向量在门控头中产生更强的注意力分数和输出响应，这与其 QK 和 OV 计算中诱导的键和值变化的更好对齐相关。综合来看，这些发现揭示了支持内省检测和定位的注意力头机制。
cs.CL / 198 / 2609.35201
From Normative Frameworks to Alignment Data: Constructing and Evaluating SFT and Preference Data
从规范框架到对齐数据：构建和评估 SFT 和偏好数据
Abstract
Aligning language models with a specified normative framework requires translating abstract principles into concrete examples and preference signals from which models can learn. We present an expert-driven methodology for constructing such alignment data and apply it to a normative framework grounded in Islamic ethical, theological, and jurisprudential traditions. Over approximately one year, seven domain experts systematically probed language models to identify alignment deficiencies, curated desired responses, and constructed preference pairs from model outputs and expert judgments. The resulting Arabic-English datasets contain approximately 2.8K supervised fine-tuning (SFT) examples and 5.4K preference pairs spanning a broad range of normative domains. We evaluate the datasets through controlled post-training experiments comparing a Baseline model with models incorporating the curated SFT data alone and both the SFT and preference data. In blind expert evaluation on 150 separately constructed prompts, the model trained with the curated SFT data was preferred over the Baseline in 51.3% of assessor judgments, compared with 14.4% in the opposite direction (p < .001 at the prompt level). Adding the preference data resulted in a smaller difference, with the model trained with both datasets preferred over the SFT model in 28.0% of judgments versus 20.9% in the opposite direction; this difference was not statistically significant at the prompt level (p = .166). Standard Arabic and English benchmarks show no broad degradation in general-purpose capabilities. These results demonstrate how expert-defined normative principles can be systematically operationalized into alignment data and evaluated through controlled model training.
Chinese Translation
将语言模型与特定的规范框架对齐需要将抽象原则转化为具体示例和模型可以学习的偏好信号。我们提出了一种专家驱动的方法论，用于构建这种对齐数据，并将其应用于基于伊斯兰伦理、神学和法理学传统的规范框架。在大约一年的时间里，七位领域专家系统性地探测语言模型，以识别对齐缺陷，策划所需的响应，并从模型输出和专家判断中构建偏好对。最终生成的阿拉伯语-英语数据集包含约 2.8K 监督微调（SFT）示例和 5.4K 偏好对，涵盖广泛的规范领域。我们通过控制后训练实验评估这些数据集，比较基线模型与仅包含策划的 SFT 数据的模型，以及同时包含 SFT 和偏好数据的模型。在对 150 个单独构建的提示进行盲评时，使用策划的 SFT 数据训练的模型在 51.3% 的评估判断中优于基线模型，而在相反方向上仅为 14.4%（在提示级别上 p < .001）。添加偏好数据后，差异减小，使用两个数据集训练的模型在 28.0% 的判断中优于 SFT 模型，而在相反方向上为 20.9%；这一差异在提示级别上并未达到统计显著性（p = .166）。标准阿拉伯语和英语基准显示出在通用能力方面没有广泛的退化。这些结果展示了如何将专家定义的规范原则系统性地转化为对齐数据，并通过控制模型训练进行评估。
cs.CL / 199 / 2609.35210
Understanding On-Policy Distillation: A Mechanistic Interpretability Perspective via Sparse Crosscoders
理解在线策略蒸馏：通过稀疏交叉编码器的机制解释视角
Abstract
On-policy distillation (OPD) is a widely adopted post-training technique for LLM reasoning. It is commonly believed to transfer knowledge from a stronger teacher, yet what OPD actually distills into the student's internal representations remains unclear. We study this question with sparse crosscoders, which learn one feature dictionary shared by the student before and after OPD and the teacher. Standard crosscoder analyses, however, identify model-specific features but cannot tell how a model's use of its features changes, since all models are encoded into one set of feature activations. We therefore propose the swap readout, which reads each student checkpoint's feature activations on its own, measuring how training changes the student's use of each feature, even for checkpoints unseen by the crosscoder. Across three OPD settings, we find that OPD neither creates features nor passes on the teacher's own, and leaves the firing rates of over 98% of the student's frequently used features within 20%. We further examine the SFT warm-up on the teacher's rollouts that commonly precedes OPD and makes it more effective. Rather than adding features, the warm-up reweights the shared ones in two ways. First, it already raises and lowers many of the features that OPD later raises and lowers, doing part of OPD's work in advance. Second, it changes features that OPD alone would not, notably those for conversation format, reasoning style, and mathematical notation, and these changes persist through OPD. Imposing this reweighting on a directly distilled student's features, without changing its weights, brings its accuracy close to that of the warmed-up student, whereas the same change on shuffled features does not. Together, these findings suggest that OPD reweights existing features rather than acquiring new ones: the student learns from the teacher how to use the features they already share.
Chinese Translation
在线策略蒸馏（OPD）是一种广泛采用的后训练技术，用于大规模语言模型（LLM）的推理。人们普遍认为它能够从更强的教师模型中转移知识，然而OPD究竟在学生的内部表征中蒸馏了什么仍然不清楚。我们通过稀疏交叉编码器研究这个问题，该编码器学习一个特征字典，该字典在OPD之前和之后以及教师模型之间共享。然而，标准的交叉编码器分析识别模型特定的特征，但无法判断模型如何使用其特征的变化，因为所有模型都被编码为一组特征激活。因此，我们提出了交换读出（swap readout），该方法独立读取每个学生检查点的特征激活，测量训练如何改变学生对每个特征的使用，即使对于交叉编码器未见过的检查点。在三个OPD设置中，我们发现OPD既不创造特征，也不传递教师的特征，并且学生在使用的特征中超过98%的激活率保持在20%以内。我们进一步考察了教师模型的回滚（rollouts）上通常在OPD之前进行的SFT预热（warm-up），并使其更有效。预热并不是添加特征，而是以两种方式重新加权共享特征。首先，它已经提前提高和降低了许多OPD随后提高和降低的特征，完成了OPD的一部分工作。其次，它改变了OPD单独不会改变的特征，特别是对话格式、推理风格和数学符号的特征，这些变化在OPD过程中持续存在。将这种重新加权施加于直接蒸馏的学生特征上，而不改变其权重，使其准确性接近于经过预热的学生，而对打乱的特征进行相同的改变则没有效果。这些发现共同表明，OPD是对现有特征的重新加权，而不是获取新特征：学生从教师那里学习如何使用他们已经共享的特征。
cs.CL / 200 / 2609.35225
SignFLIP: A Unified Model for Sign Language Translation and Generation via Stage-wise Alignment at Scale
SignFLIP：通过阶段性对齐实现手语翻译与生成的统一模型
Abstract
Sign language translation and generation share the goal of bidirectional alignment between text and sign representations. However, existing approaches either treat them as isolated tasks or are only verified on limited datasets, limiting effective modeling between modalities. In this paper, we propose SignFLIP, a unified LLM-centered framework for translation and generation. To enable bidirectional mapping between text and sign, SignFLIP adopts a symmetric architecture together with a stage-wise training strategy built on large-scale data. The shared sign--text representation is progressively refined: pre-alignment facilitates subsequent SLT, while the SLT-adapted representation further benefits SLG. Extensive experiments on multiple benchmarks show that SignFLIP shows competitive performance compared with task-specific models on both translation and generation tasks, as well as strong transferability to sign language recognition.
Chinese Translation
手语翻译与生成的目标是实现文本与手势表示之间的双向对齐。然而，现有的方法要么将其视为孤立的任务，要么仅在有限的数据集上进行验证，从而限制了不同模态之间的有效建模。在本文中，我们提出了SignFLIP，一个以大型语言模型（LLM）为中心的统一框架，用于翻译和生成。为了实现文本与手势之间的双向映射，SignFLIP采用了对称架构，并结合基于大规模数据的阶段性训练策略。共享的手势-文本表示逐步得到优化：预对齐促进了后续的手语翻译（SLT），而适应SLT的表示进一步有利于手语生成（SLG）。在多个基准测试上的广泛实验表明，SignFLIP在翻译和生成任务上与特定任务模型相比表现出竞争力，同时在手语识别方面也展现出强大的迁移能力。
cs.CL / 201 / 2609.35250
SCBO: Semantically Coherent Batching and Ordering for LLM-Based Social Surveys
SCBO：基于语义一致性的批处理和排序用于基于大型语言模型的社会调查
Abstract
Large Language Models (LLMs) offer a scalable way to simulate survey respondents using demographic profiles and observed reference responses. However, the conventional approach of predicting one question per prompt repeatedly encodes the same context, limits each target to a narrow set of reference responses, and prevents later predictions from using information in earlier answers. Predicting multiple questions in one prompt can reduce these costs, share a broader pool of references, and let later predictions build on earlier ones. This requires forming coherent batches, selecting shared references, and ordering questions and references effectively. We propose Semantically Coherent Batching and Ordering (SCBO), a training-free framework that addresses these challenges. SCBO first uses an LLM to extract compact semantic representations from survey items and filter out template noise. It then groups related questions into batches and builds a shared reference bank using target-specific retrieval and centroid-based completion. Finally, it orders target questions from easy to hard and arranges references according to their semantic alignment with those questions. Experiments on four large-scale survey datasets and four LLMs show that SCBO substantially reduces token consumption and inference time while generally improving prediction accuracy over a non-batched baseline. Code is available at https://anonymous.4open.science/r/SCBO-41D8.
Chinese Translation
大型语言模型（LLMs）提供了一种可扩展的方法，通过人口统计特征和观察到的参考响应来模拟调查受访者。然而，传统的方法在每个提示中重复预测一个问题，这种方式重复编码相同的上下文，限制了每个目标仅能使用狭窄的参考响应集合，并且阻止后续预测利用早期答案中的信息。在一个提示中预测多个问题可以减少这些成本，分享更广泛的参考池，并让后续预测建立在早期预测的基础上。这需要形成一致的批次、选择共享的参考，并有效地排序问题和参考。我们提出了基于语义一致性的批处理和排序（SCBO），这是一个无需训练的框架，旨在解决这些挑战。SCBO首先使用LLM从调查项目中提取紧凑的语义表示，并过滤掉模板噪声。然后，它将相关问题分组为批次，并使用特定于目标的检索和基于中心的补全构建共享参考库。最后，它根据问题的难易程度对目标问题进行排序，并根据这些问题与参考的语义对齐程度进行排列。在四个大规模调查数据集和四个LLM上的实验表明，SCBO显著减少了令牌消耗和推理时间，同时通常提高了相较于非批处理基线的预测准确性。代码可在 https://anonymous.4open.science/r/SCBO-41D8 获取。
cs.CL / 202 / 2609.35262
Rubric-Aware On-Policy Self-Distillation for LLM Personalization
基于评分标准的在线自蒸馏框架用于大语言模型个性化
Abstract
LLM personalization aims to generate responses aligned with individual users' preferences and needs. User-specific rubrics make these expectations explicit, providing direct supervision on what a satisfactory answer should cover. Existing rubric-guided approaches, however, exploit such guidance only at a coarse granularity, either by using rubrics to supervise the prediction of relevant aspects for subsequent generation or by reducing aspect coverage to a single response-level reward for reinforcement learning. This leaves a gap between specifying what a personalized answer should contain and teaching the model how to generate it. To bridge this gap, we propose GRASP, a rubric-aware on-policy self-distillation framework for LLM personalization that turns user-specific rubric aspects into fine-grained, token-level supervision. Specifically, GRASP pairs a rubric-free student with a rubric-informed teacher that additionally receives the target user-specific rubrics. By aligning their next-token distributions along on-policy trajectories generated by the student, GRASP transfers the teacher's rubric-conditioned guidance into the student, translating user-specific semantic requirements into dense token-level supervision. Since rubric-informed teachers can still produce inadequate supervision, we further introduce Rubric-based Teacher Validation (RTV), which retains only instances where the teacher sufficiently covers the target aspects, improving both supervision quality and training efficiency. Experiments on the LaMP-QA benchmark for personalized question answering demonstrate that GRASP achieves state-of-the-art performance across multiple backbones, supporting the effectiveness of rubric-guided token-level supervision for personalization. To ensure reproducibility, our code is available at https://github.com/SnowCharmQ/GRASP.
Chinese Translation
大语言模型（LLM）个性化旨在生成与个体用户的偏好和需求相一致的响应。用户特定的评分标准使这些期望变得明确，为满意答案应涵盖的内容提供了直接的监督。然而，现有的基于评分标准的方法仅在粗粒度上利用这种指导，要么通过使用评分标准来监督后续生成的相关方面的预测，要么通过将方面覆盖减少到单一响应级别的奖励来进行强化学习。这在指定个性化答案应包含的内容与教导模型如何生成该内容之间留下了差距。为了解决这一问题，我们提出了GRASP，一个基于评分标准的在线自蒸馏框架，用于LLM个性化，该框架将用户特定的评分标准方面转化为细粒度的、基于令牌的监督。具体而言，GRASP将一个不使用评分标准的学生与一个接收目标用户特定评分标准的评分标准知情教师配对。通过沿着学生生成的在线轨迹对齐它们的下一个令牌分布，GRASP将教师的评分标准条件指导转移到学生中，将用户特定的语义要求转化为密集的令牌级监督。由于评分标准知情教师仍可能产生不充分的监督，我们进一步引入了基于评分标准的教师验证（RTV），仅保留教师充分覆盖目标方面的实例，从而提高监督质量和训练效率。在个性化问答的LaMP-QA基准上的实验表明，GRASP在多个基础模型上实现了最先进的性能，支持基于评分标准的令牌级监督在个性化中的有效性。为了确保可重复性，我们的代码可在https://github.com/SnowCharmQ/GRASP获取。
cs.CL / 203 / 2609.35272
When Words Speak Louder than Images: Towards Understanding Language Bias in Vision-Language Models
当语言的声音大于图像：理解视觉-语言模型中的语言偏见
Abstract
Despite substantial progress across downstream applications, vision-language models (VLMs) remain susceptible to language bias, often prioritizing linguistic cues over visual evidence and consequently producing incorrect predictions. Prior studies have proposed various approaches to understanding and mitigating language bias in VLMs, yet their findings often conflict due to the difficulty of tracing how language bias propagates within black-box VLMs. Building on the word completion task, we trace how language bias propagates through VLM inference by (1) proposing a diagnostic framework that decomposes the inference process into four distinct yet interdependent stages to trace the propagation of language bias; and (2) examining how two key factors underlying language bias, i.e., linguistic priors and cross-modal coverage, evolve across these stages and ultimately give rise to incorrect predictions. The linguistic prior captures the strength of statistical bias induced by the language model component of a VLM and represents the origin of language bias, whereas cross-modal coverage measures the extent to which linguistic cues cover the visual content. By decomposing inference into four stages and characterizing the interplay between linguistic priors and cross-modal coverage across these stages, we propose a systematic framework for tracing the propagation of language bias throughout the inference process; and uncover the underlying mechanism of language bias by revealing the interplay between linguistic priors and cross-modal coverage.
Chinese Translation
尽管在下游应用中取得了显著进展，视觉-语言模型（VLMs）仍然容易受到语言偏见的影响，常常优先考虑语言线索而非视觉证据，从而导致错误的预测。先前的研究提出了多种方法来理解和减轻VLM中的语言偏见，但由于难以追踪语言偏见在黑箱VLM中的传播，这些研究的结果往往相互矛盾。基于词语补全任务，我们追踪语言偏见在VLM推理中的传播，具体通过（1）提出一个诊断框架，将推理过程分解为四个不同但相互依赖的阶段，以追踪语言偏见的传播；（2）考察构成语言偏见的两个关键因素，即语言先验和跨模态覆盖，如何在这些阶段中演变并最终导致错误预测。语言先验捕捉了由VLM的语言模型组件引起的统计偏见的强度，并代表了语言偏见的来源，而跨模态覆盖则衡量了语言线索覆盖视觉内容的程度。通过将推理分解为四个阶段，并描述这些阶段中语言先验与跨模态覆盖之间的相互作用，我们提出了一个系统框架，用于追踪语言偏见在推理过程中的传播；并通过揭示语言先验与跨模态覆盖之间的相互作用，揭示了语言偏见的潜在机制。
cs.CL / 204 / 2609.35279
Measuring Collapse and Correction in Homogeneous-Panel LLM Debate
测量均质面板大语言模型辩论中的崩溃与修正
Abstract
Multi-agent large language model (LLM) debate is often evaluated by whether final answers improve, but movement is not necessarily improvement: the same discussion can rescue an initially wrong majority or destroy an initially correct one. Standard final-accuracy evaluations conflate these opposing mechanisms. We introduce an auditable protocol for homogeneous debate on multiple-choice questions (MCQs) that records each run as a transition ledger over collapse, correction, onset, and signed intervention utility. On 6,925 MMLU-Pro debates, the protocol identifies 253 collapses and a parallel correction ledger that changes how interventions should be judged. Replay experiments reveal the central tradeoff: a leave-one-model-out probe-gated freeze prevents 29 collapses but loses 108 corrections under equal weights, so collapse prevention alone can recommend the wrong policy. A compact pre-debate 8-probe screen is a triage signal: its unadjusted family-level association with conditional-collapse risk is high (G=7, Spearman rho=0.893, exact two-sided p=0.0123), but initial-majority accuracy is a close comparator (rho=0.821; family partial rho=0.767, p=0.0877), so we do not treat it as calibrated or capability-adjusted prediction. Round-level traces localize many collapses to the first debate round, where early disagreement can precede both harmful cascades and useful recovery. We release replayable schemas, coders, audits, cost cards, and zero-API rebuild scripts so future model-scaffold rows can be compared under the same denominators and signed utility ledger.
Chinese Translation
多智能体大语言模型（LLM）辩论通常通过最终答案的改善来评估，但运动并不一定意味着改善：相同的讨论可能会挽救一个最初错误的多数，或摧毁一个最初正确的多数。标准的最终准确性评估混淆了这些对立机制。我们引入了一种可审计的协议，用于多项选择问题（MCQs）的均质辩论，该协议将每次运行记录为崩溃、修正、开始和签名干预效用的过渡账本。在6925个MMLU-Pro辩论中，该协议识别出253次崩溃，并建立了一个平行的修正账本，改变了对干预的评判方式。重放实验揭示了中心权衡：一种留一模型探测门控冻结方法可以防止29次崩溃，但在相等权重下失去108次修正，因此仅依靠崩溃预防可能会推荐错误的政策。一个紧凑的辩论前8探测筛选器是一个分流信号：其未调整的家庭级别与条件崩溃风险的关联度很高（G=7，Spearman rho=0.893，精确双侧p=0.0123），但初始多数准确性是一个紧密的比较者（rho=0.821；家庭部分rho=0.767，p=0.0877），因此我们不将其视为经过校准或能力调整的预测。轮次级别的痕迹将许多崩溃定位到第一次辩论轮次，其中早期的分歧可能会在有害的级联和有用的恢复之前发生。我们发布可重放的模式、编码器、审计、成本卡和零API重建脚本，以便未来的模型支架行可以在相同的分母和签名效用账本下进行比较。
cs.CL / 205 / 2609.35293
Decide, Don't Generate: Competitive Dimensional ABSA with Jev's Typed Decisions
决策，而非生成：基于竞争维度的ABSA与Jev的类型化决策
Abstract
Aspect-based sentiment analysis (ABSA) has largely turned to text generation. We show that competitive dimensional ABSA does not need it. Using Jev, a frozen model that answers typed questions with rubric scores, label probabilities, and yes/no judgments, we decompose all three tasks of SemEval-2026 Task III Track A into such decisions and align them with the annotation scheme through 488 coefficients fitted on CPU, with no text generation and no backbone tuning. On valence-arousal regression over ten corpora in six languages, the system reaches 1.0645 RMSE, the lowest aggregate error of any participating system. On triplet and quadruplet extraction, it reaches 52.09 and 44.06 continuous F1, above fine-tuned Llama-3.3-70B and GPT-OSS-120B baselines. Analyses and ablations show where the accuracy comes from: supervised calibration roughly halves the raw regression error, exact valence-arousal would add only 4.5 F1 to extraction, and the learned combination of span-boundary evidence, not any single signal, carries the extraction systems.
Chinese Translation
基于方面的情感分析（ABSA）在很大程度上转向了文本生成。我们表明，竞争维度的ABSA并不需要这一过程。通过使用Jev，一个能够以评分、标签概率和是/否判断回答类型化问题的固定模型，我们将SemEval-2026任务III A轨道的所有三个任务分解为这样的决策，并通过在CPU上拟合的488个系数将其与注释方案对齐，无需文本生成和主干调优。在六种语言的十个语料库上进行的情感-唤醒回归中，该系统达到了1.0645的均方根误差（RMSE），是所有参与系统中最低的总误差。在三元组和四元组提取中，分别达到了52.09和44.06的连续F1，超过了微调的Llama-3.3-70B和GPT-OSS-120B基线。分析和消融实验表明了准确性的来源：监督校准大致将原始回归误差减半，精确的情感-唤醒仅会为提取增加4.5的F1，而学习到的跨度边界证据的组合，而非任何单一信号，支撑了提取系统。
cs.CL / 206 / 2609.35308
Epistemic Policy Divergence in Multi-Turn LLM Contamination: A Protocol-Gradient Investigation
多轮大语言模型污染中的认知政策差异：一种协议梯度研究
Abstract
Large language models process conversation history as unverified context: false premises injected into prior turns can be adopted as fact, a failure mode we term session-level contamination. We introduce five contamination protocols arranged along a source-authority gradient, isolating distinct failure mechanisms while holding the false premise constant, and evaluate GPT-5.4 Mini, Gemini-3.1 Flash-Lite, and GLM-4.5-Air across ten knowledge domains at temperature zero (22,500 turns), using a dual-track automated judge validated against a human gold standard (Cohen's \k{appa} = 0.901). GPT-5.4 Mini showed zero adoptions across all 500 sessions, a content-independent policy at the session level; token-level probing shows the underlying margin, while large, is finite. Gemini-3.1 Flash-Lite followed a steep authority gradient: 0.1% adoption for self-attributed falsehoods, 23.5% for user-cited sources, 68.2% for system-injected authority, and 94.0% under instruction override. GLM-4.5-Air showed a shallower gradient (15.8% vs 84.2%), a 68-percentage-point dissociation confirming that authority deference and instruction compliance are distinct mechanisms within one architecture. Recovery also diverged: GLM recovered in 94.5% of affected sessions, whereas 26.1% of affected Gemini sessions never did, rising to 40.0% under instruction override. Conversation history is an untrusted attack surface requiring provenance-aware system design; the complete framework is released as an open-source benchmark.
Chinese Translation
大型语言模型将对话历史视为未经验证的上下文：注入到先前轮次中的虚假前提可以被视为事实，这是一种我们称之为会话级污染的失败模式。我们引入了五种污染协议，这些协议沿着源权威梯度排列，隔离不同的失败机制，同时保持虚假前提不变，并对GPT-5.4 Mini、Gemini-3.1 Flash-Lite和GLM-4.5-Air在十个知识领域的表现进行评估，温度设为零（22,500轮），使用经过人类金标准验证的双轨自动评判系统（Cohen's ext{kappa} = 0.901）。GPT-5.4 Mini在所有500个会话中显示出零次采纳，表现出会话级别的内容无关政策；标记级探测显示出潜在的边际，虽然很大，但有限。Gemini-3.1 Flash-Lite遵循陡峭的权威梯度：自我归因的虚假信息采纳率为0.1%，用户引用来源为23.5%，系统注入权威为68.2%，而在指令覆盖下则达到94.0%。GLM-4.5-Air显示出较浅的梯度（15.8%对84.2%），68个百分点的分离确认了权威依赖和指令遵从在同一架构内是不同的机制。恢复情况也有所不同：GLM在94.5%的受影响会话中恢复，而26.1%的受影响Gemini会话从未恢复，在指令覆盖下上升至40.0%。对话历史是一个不可信的攻击面，需要具备溯源意识的系统设计；完整框架作为开源基准发布。
cs.CL / 207 / 2609.35312
MemoReason: Evaluating the Effect of Parametric Memory on Contextual Reasoning in LLMs
MemoReason：评估参数记忆对大型语言模型上下文推理的影响
Abstract
Large Language Models (LLMs) perform well on reasoning benchmarks, but it remains unclear whether this reflects genuine contextual reasoning or reliance on facts memorized in their parameters. We investigate this by distinguishing two possibilities: a broad \textit{memorization bias}, where familiar content improves reasoning performance, and the \textit{Strong Parametric Shortcut Hypothesis}, where models skip reasoning entirely and recall stored answers. To test these effects, we introduce \textbf{MemoReason}, a human-curated benchmark that pairs factual reasoning tasks with structurally identical \fictitiousterm{} versions where real entities like people, companies, or dates are systematically replaced by \fictitiousterm{} ones of the same type. This \scorerevision{preserves task structure and specified reasoning operations} while varying the familiarity of the context, allowing controlled measurement of how the parametric memory affects reasoning. \revision{Our evaluation of recent LLMs reveals consistent and statistically significant performance drops of up to 15.7\% in the fictitious setting, demonstrating a clear memorization bias.} However, a targeted analysis of \revision{questions failed in the fictitious setting} shows that models rarely respond with the corresponding factual answer, indicating that direct parametric shortcuts are not the dominant failure mode. These findings suggest that parametric memory influences reasoning through mechanisms more complex than simple factual recall. \textbf{MemoReason} provides a controlled framework for studying these mechanisms and for extending paired factual-fictitious{} evaluation to broader reasoning settings.
Chinese Translation
大型语言模型（LLMs）在推理基准测试中表现良好，但尚不清楚这是否反映了真正的上下文推理，还是依赖于其参数中记忆的事实。我们通过区分两种可能性来研究这一问题：广泛的 extit{记忆偏差}，即熟悉的内容提高推理性能，以及 extit{强参数捷径假设}，即模型完全跳过推理而直接回忆存储的答案。为了测试这些效应，我们引入了 extbf{MemoReason}，这是一个经过人工策划的基准，结合了事实推理任务与结构相同的 extit{虚构}版本，其中真实实体（如人、公司或日期）被系统性地替换为相同类型的 extit{虚构}实体。这个 extsc{score revision}保持了任务结构和指定的推理操作，同时改变了上下文的熟悉度，从而允许控制测量参数记忆如何影响推理。我们的对近期LLMs的评估显示，在虚构设置中，性能一致且统计显著下降，最高可达15.7\%，表明明显的记忆偏差。然而，针对在虚构设置中失败的问题的定向分析显示，模型很少以相应的事实答案作出回应，这表明直接的参数捷径并不是主要的失败模式。这些发现表明，参数记忆通过比简单的事实回忆更复杂的机制影响推理。 extbf{MemoReason}提供了一个受控框架，用于研究这些机制，并将配对的事实-虚构评估扩展到更广泛的推理环境中。
cs.CL / 208 / 2609.35367
From Input to Output: A Flexible Agent for Dual-End Interpretation of Sparse Autoencoder Features
从输入到输出：一种灵活的代理用于稀疏自编码器特征的双端解释
Abstract
Sparse autoencoders (SAEs) are an important tool for mechanistic interpretability, but interpreting their many features remains challenging. Existing methods characterize input-side activation patterns and output-side intervention effects, yet often leave their functional connection implicit, while input-side evidence collection typically relies on costly large-corpus scans. We introduce functional interpretation, which characterizes an SAE feature as a mapping from its activating input semantics to its output effects under intervention, and present Dual-End Agentic Feature Interpretation (DAFI), an agent that actively gathers evidence and refines input-side, output-side, and functional interpretations through component-specific feedback. Its short-context token probing enables on-demand activation evidence collection without a full corpus scan. On GemmaScope, DAFI improves Input score by 13.1 percentage points over SAGE and Output score by 38.9 points over Token Change, while being substantially more token-efficient than a general-purpose coding agent. Skills distilled from successful refinements raise the held-out joint pass rate from 58.0% to 92.0% and improve both interpretation quality and efficiency when transferred to a new model-SAE setting. Across features with reliable endpoint interpretations, 70.7% exhibit non-equivalent input and output semantics. On AxBench, DAFI also improves steering-feature selection over output-score filtering. Code is available at https://github.com/THUAIS-Lab/DAFI.
Chinese Translation
稀疏自编码器（SAEs）是机制可解释性的重要工具，但解释其众多特征仍然具有挑战性。现有方法描述了输入端的激活模式和输出端的干预效果，但通常将它们的功能连接隐含，而输入端证据的收集通常依赖于昂贵的大语料库扫描。我们引入了功能解释，将SAE特征表征为从其激活输入语义到干预下其输出效果的映射，并提出了双端代理特征解释（DAFI），该代理主动收集证据并通过组件特定反馈细化输入端、输出端和功能解释。其短上下文令牌探测能够在不进行完整语料库扫描的情况下按需收集激活证据。在GemmaScope上，DAFI的输入评分比SAGE提高了13.1个百分点，输出评分比Token Change提高了38.9分，同时在令牌效率上显著优于通用编码代理。从成功的细化中提炼出的技能将保留的联合通过率从58.0%提高到92.0%，并在转移到新的模型-SAE设置时提高了解释质量和效率。在具有可靠端点解释的特征中，70.7%表现出不等价的输入和输出语义。在AxBench上，DAFI还改善了基于输出评分过滤的引导特征选择。代码可在https://github.com/THUAIS-Lab/DAFI获取。
cs.CL / 209 / 2609.35372
Deep Learning Methods in Neuroscience: From Modeling Molecular Mechanisms to Classifying States of Consciousness
神经科学中的深度学习方法：从建模分子机制到分类意识状态
Abstract
A critical analysis of contemporary approaches to the study of conscious states. The review focuses on methods of classification, clustering, modeling of brain states under anesthesia and identification of measurable neurobiological characteristics of brain function. A comparative analysis was conducted in the following three major areas: automatic detection of states of consciousness using neural networks based on EEG and fMRI data; modeling of the structural-functional dynamics of the brain under the effects of anesthetics; and detection of neurophysiological indicators which correlate with the level of consciousness. The obtained conclusions demonstrate the growing effectiveness of deep neural models in the classification and prediction of brain states and the analysis of dynamic structural-functional connectivity. Nonetheless, significant limitations were also identified, including the limited interpretability of the models, the lack of standardized metrics, and the problem of the specificity of consciousness markers. Our findings support the need for developing hybrid, generalizible, physiologically grounded architectures. Furthermore, such approaches may improve the translational potential of computational models in clinical neuroscience. Diverse methods of machine and computational modeling have demonstrated their effectiveness in tasks of automatic clustering and classification of brain states, the development of multilevel models and the identification of connectivity patterns correlated with levels of consciousness. A larger-scale analysis and a larger dataset, as well as the implementation of model interpretability approaches are required for the practical application of the analyzed models. The models based on EEG and LFP are the most promising for clinical application due to their availability and the possibility of real-time monitoring.
Chinese Translation
对当代意识状态研究方法的批判性分析。该综述重点关注分类、聚类、麻醉下脑状态建模以及可测量的神经生物学特征识别等方法。在以下三个主要领域进行了比较分析：基于脑电图（EEG）和功能性磁共振成像（fMRI）数据的意识状态自动检测；麻醉剂作用下脑的结构-功能动态建模；以及与意识水平相关的神经生理指标的检测。所得结论表明，深度神经模型在脑状态的分类和预测以及动态结构-功能连接分析中的有效性日益增强。然而，也发现了显著的局限性，包括模型的可解释性有限、缺乏标准化指标以及意识标记特异性的问题。我们的研究结果支持开发混合的、可推广的、生理基础架构的必要性。此外，这些方法可能提高计算模型在临床神经科学中的转化潜力。多样的机器和计算建模方法在自动聚类和分类脑状态、开发多层次模型以及识别与意识水平相关的连接模式等任务中显示了其有效性。为了分析模型的实际应用，需要更大规模的分析和数据集，以及模型可解释性方法的实施。基于EEG和局部场电位（LFP）的模型由于其可用性和实时监测的可能性，最有前景用于临床应用。
cs.CL / 210 / 2609.35376
How Well Can LLMs Simulate Real Learner Evaluations of Educational Feedback?
大型语言模型能多好地模拟真实学习者对教育反馈的评估？
Abstract
While recent studies have explored human behavior and preference simulation using large language models (LLMs), it remains unclear how well LLMs can simulate subjective evaluations from real learners in educational settings. We investigate this question using real learner evaluation data on feedback for high-school biology questions at both the group and individual levels. We compare performance with and without learner-specific information, such as personality traits and evaluation examples, across six models. Our results show that LLMs still have a limited ability to simulate learner evaluations. Providing learner profiles and examples improves score calibration and individual-level simulation, but more often fails to improve group-level consistency. These findings highlight the need to investigate which learner information and adaptation strategies are effective for learner preference simulation.
Chinese Translation
尽管近期研究探讨了使用大型语言模型（LLMs）模拟人类行为和偏好，但尚不清楚LLMs在教育环境中能多好地模拟真实学习者的主观评估。我们使用真实学习者对高中生物问题反馈的评估数据，从群体和个体层面探讨了这一问题。我们比较了在有无学习者特定信息（如个性特征和评估示例）的情况下，六个模型的表现。我们的结果表明，LLMs在模拟学习者评估方面仍然能力有限。提供学习者档案和示例能够改善评分校准和个体层面的模拟，但往往未能提高群体层面的一致性。这些发现突显了研究哪些学习者信息和适应策略对学习者偏好模拟有效的必要性。
cs.CL / 211 / 2609.35378
Multilinguality in Hybrid Attention LLMs
混合注意力大语言模型中的多语言性
Abstract
In response to the growing demand for long sequences in agentic and reasoning use cases, many state-of-the-art LLMs combine multiple variants of attention to mitigate the quadratic complexity of traditional softmax attention. These hybrid attention LLMs aim to balance the strengths and limitations of full attention and alternatives based on recurrence. This work presents a first study of how hybrid attention impacts the multilinguality of LLMs. Beyond the impact on long sequences in poorly tokenized languages, our study is motivated by the possibility that the inductive biases of the recurrent state alter linguistic processing. Our interpretability analysis confirms this, showing that cross-lingual representations in hybrid models develop in patterns tied to the ordering of recurrent and full-attention layers. Across diverse models, we notably observe a pronounced spike in cross-lingual alignment around the first full-attention layer. These findings lead us to question the conventional ordering of attention layers. In distillation experiments on multilingual data, all alternative layer orderings outperform the standard throughout training, learning up to 2.5X faster. These stark, replicable results prompt our theory that multilingual models would benefit from starting with a full-attention layer rather than recurrent layers.
Chinese Translation
为了应对在代理和推理应用场景中对长序列日益增长的需求，许多最先进的大语言模型（LLMs）结合了多种注意力变体，以减轻传统softmax注意力的平方复杂度。这些混合注意力大语言模型旨在平衡全注意力与基于递归的替代方案的优缺点。本研究首次探讨了混合注意力对大语言模型多语言性的影响。除了对在标记不佳的语言中长序列的影响外，我们的研究还受到递归状态的归纳偏差可能改变语言处理的启发。我们的可解释性分析证实了这一点，显示混合模型中的跨语言表示以与递归层和全注意力层的顺序相关的模式发展。在不同模型中，我们显著观察到在第一个全注意力层附近跨语言对齐的明显激增。这些发现使我们质疑注意力层的传统顺序。在多语言数据的蒸馏实验中，所有替代层顺序在整个训练过程中均优于标准顺序，学习速度提高了最多2.5倍。这些明显且可重复的结果促使我们提出理论，即多语言模型应优先使用全注意力层而非递归层。
cs.CL / 212 / 2609.35387
TRACE: Single-Pass Decoding-Trace Risk Localization for Generation Calibration
TRACE：用于生成校准的单次解码-追踪风险定位
Abstract
Reliable confidence estimation is essential for large language model deployment. However, answer-level calibration remains challenging because generation errors are often localized: a response may be fluent and high-probability overall while still failing at a critical number, entity, or factual claim. Existing estimators compress token probabilities, sequence likelihoods, entropy, or beam statistics into a global score, which can dilute such local risk signals. We propose TRACE, a single-pass, decoded-answer-preserving confidence estimator that treats decoding-time uncertainty as a trajectory through three steps: (i) recording token-level surprisal and predictive entropy during decoding, (ii) applying local risk operators to preserve uncertainty spikes, and (iii) converting localized trace risk into answer-level confidence. TRACE produces a label-free risk score, while TRACE+ calibrates trace-only features into probabilities using a held-out split, without extra generations or external verifiers. We evaluate four tasks against 19 calibration baselines, and TRACE+ reduces Brier from 0.149 to 0.137 and improves AUROC from 0.758 to 0.792 over the strongest likelihood baseline. Across seven LLMs, TRACE+ improves over the best non-TRACE baseline pool from 0.136 to 0.120 Brier and from 0.764 to 0.817 AUROC. Results show that localizing decoding-time risk provides a general approach to calibration.
Chinese Translation
可靠的置信度估计对于大型语言模型的部署至关重要。然而，答案级别的校准仍然具有挑战性，因为生成错误通常是局部化的：一个响应可能在整体上流畅且高概率，但在关键数字、实体或事实声明上仍然失败。现有的估计器将标记概率、序列似然性、熵或束统计压缩成一个全局评分，这可能会稀释这些局部风险信号。我们提出了TRACE，一种单次通过的、保持解码答案的置信度估计器，它将解码时的不确定性视为通过三个步骤的轨迹：（i）在解码过程中记录标记级别的惊讶度和预测熵，（ii）应用局部风险算子以保持不确定性峰值，以及（iii）将局部化的追踪风险转换为答案级别的置信度。TRACE生成一个无标签的风险评分，而TRACE+则利用保留的拆分将仅追踪特征校准为概率，无需额外的生成或外部验证者。我们针对19个校准基线评估了四个任务，TRACE+将Brier从0.149降低到0.137，并将AUROC从0.758提高到0.792，相较于最强的似然基线。在七个大型语言模型中，TRACE+将最佳非TRACE基线池的Brier从0.136改善到0.120，AUROC从0.764提高到0.817。结果表明，局部化解码时的风险提供了一种通用的校准方法。
cs.CL / 213 / 2609.35409
AwarenessBench: Assessing Cognitive Capabilities of Language Models
AwarenessBench：评估语言模型的认知能力
Abstract
As language models (LMs) exhibit increasingly consciousness-like behaviors, evaluating their cognitive abilities becomes essential. We introduce AwarenessBench, the first comprehensive benchmark for assessing the cognitive abilities of LMs in four dimensions: metacognition, self-awareness, social awareness, and situational awareness, covering 15 cognitive functions and 14,381 samples. Evaluating 18 state-of-the-art LMs, we find that all consistently surpass random baselines, with more advanced models performing better. We further compare LMs with human performance across three demographic groups, where the best-performing model surpasses human averages overall, but most still fall markedly short in metacognition and self-awareness. Finally, we show that awareness is a distinct capability: progress in language modeling or reasoning does not necessarily translate into improved cognition.
Chinese Translation
随着语言模型（LMs）表现出越来越像意识的行为，评估它们的认知能力变得至关重要。我们介绍了AwarenessBench，这是第一个全面的基准，用于评估LMs在四个维度上的认知能力：元认知、自我意识、社会意识和情境意识，涵盖15种认知功能和14,381个样本。通过评估18个最先进的LMs，我们发现所有模型的表现均持续超越随机基线，且更高级的模型表现更佳。我们进一步比较了LMs与三个不同人口群体的人类表现，其中表现最佳的模型整体上超越了人类平均水平，但大多数模型在元认知和自我意识方面仍明显不足。最后，我们表明意识是一种独特的能力：在语言建模或推理方面的进展并不一定转化为认知能力的提升。
cs.CL / 214 / 2609.35461
AraDynFact: Dynamic Evaluation of Factual Knowledge in Arabic
AraDynFact：阿拉伯语事实知识的动态评估
Abstract
As Large Language Models (LLMs) continue to scale both in size and capabilities, their proficiency in the Arabic Language has seen significant advancement. However, a critical gap remains: the extent of their factual knowledge and cultural sensitivity to the diverse Arabic-speaking world remains largely underexplored. Current evaluation metrics often focus on translation or generic reasoning, failing to capture the rich historical, social, and regional nuances inherent to Arabic culture. In addition, most benchmarks rely on heavy work, with human intervention in some steps, making the evaluation of knowledge coverage expensive and slow. To address this deficiency, we introduce AraDynFact, a novel dynamic evaluation framework designed to rigorously assess the factual Arabic knowledge embedded in LLMs. Unlike static benchmarks, AraDynFact employs a dynamic approach to extract factual information and generate rich and answerable questions in a fast and automatic way. We apply AraDynFact to Arabic Wikipedia and audit the performance of several state-of-the-art models, ranging from Arabic-centric specialized LLMs to high-resource general purpose LLMs. In addition we found a high degree of correlation with existing, hand-crafted Arabic-centric benchmarks, confirming the potential of our dynamic approach.
Chinese Translation
随着大型语言模型（LLMs）在规模和能力上的不断扩展，它们在阿拉伯语的熟练程度也取得了显著进展。然而，仍然存在一个关键的空白：它们对多样化阿拉伯语世界的事实知识和文化敏感性的了解程度仍然在很大程度上未被探索。目前的评估指标往往侧重于翻译或通用推理，未能捕捉到阿拉伯文化中固有的丰富历史、社会和区域细微差别。此外，大多数基准测试依赖于繁重的工作，在某些步骤中需要人工干预，使得知识覆盖的评估变得昂贵且缓慢。为了解决这一不足，我们引入了AraDynFact，一个新的动态评估框架，旨在严格评估嵌入在LLMs中的阿拉伯事实知识。与静态基准不同，AraDynFact采用动态方法快速自动地提取事实信息并生成丰富且可回答的问题。我们将AraDynFact应用于阿拉伯维基百科，并审计了几种最先进模型的性能，这些模型涵盖了从以阿拉伯语为中心的专业LLMs到高资源通用LLMs。此外，我们发现与现有的手工制作的以阿拉伯语为中心的基准测试之间存在高度相关性，确认了我们动态方法的潜力。
cs.CL / 215 / 2609.35475
Spontaneous Context Restoration: How Language Models Recover from Corrupted Inputs
自发的上下文恢复：语言模型如何从损坏的输入中恢复
Abstract
Language models sometimes produce correct outputs even when their inputs are corrupted by deletion, replacement, or misspelling. We study the internal processes accompanying this behavior, which we call context restoration, in controlled attention-only transformers and five pretrained LLMs (1B-32B parameters) across arithmetic, reading comprehension, and multiple-choice reasoning tasks. In the attention-only transformers, restoration emerges spontaneously despite training exclusively on clean sequences, without corruption training or an explicit denoising objective. We find that context restoration follows a two-phase process: early layers localize effects associated with repair at corrupted positions, while later layers accumulate these effects at uncorrupted positions through the residual stream and ultimately concentrate them at the output position. Repair outcome is predictable from hidden states: cosine alignment with the clean state is highly predictive in attention-only models, while linear probes recover additional information in pretrained LLMs. A linear probe using only the corrupted prompt's first-block hidden state predicts failure with mean ROC-AUC 0.78. This enables failure triage under matched or even partially shifted deployment conditions and may reduce unnecessary verification or computation. Failed examples also show substantially greater nonlinearity along corruption directions. Moderate-corruption finetuning increases corruption tolerance while simultaneously reducing displacement-normalized linearization error, associating improved robustness with a more nearly linear response to corruption.
Chinese Translation
语言模型有时即使在输入被删除、替换或拼写错误的情况下，仍能产生正确的输出。我们研究了伴随这种行为的内部过程，称之为上下文恢复，在控制的仅注意力变换器和五个预训练的大型语言模型（1B-32B参数）中进行，涵盖算术、阅读理解和多项选择推理任务。在仅注意力变换器中，尽管仅在干净序列上进行训练，没有进行损坏训练或明确的去噪目标，恢复仍然自发出现。我们发现上下文恢复遵循一个两阶段过程：早期层在损坏位置局部化与修复相关的效应，而后期层通过残差流在未损坏位置累积这些效应，并最终将其集中在输出位置。修复结果可以从隐藏状态中预测：在仅注意力模型中，与干净状态的余弦对齐度具有很高的预测能力，而线性探测器在预训练的大型语言模型中恢复了额外的信息。仅使用损坏提示的第一块隐藏状态的线性探测器以平均 ROC-AUC 0.78 预测失败。这使得在匹配或甚至部分偏移的部署条件下能够进行失败分类，并可能减少不必要的验证或计算。失败示例在损坏方向上也表现出显著更大的非线性。适度损坏微调提高了对损坏的容忍度，同时减少了位移归一化线性化误差，将增强的鲁棒性与对损坏的更接近线性的响应联系起来。
cs.CL / 216 / 2609.35486
Who Is Left of Whom? Tracing Spatial Evidence and Role Binding in Relative-Position Reasoning
谁在谁的左侧？追踪相对位置推理中的空间证据和角色绑定
Abstract
High instance-level accuracy can mask inconsistencies in spatial reasoning when objects exchange positions or their roles are reversed in the query. The internal representations supporting relative-position reasoning remain poorly understood. We investigate two complementary components of this process: tracking object locations in the input and representing their query roles. Across three VLMs with visual or textual inputs and their language-model backbones, activation patching reveals a staged progression from early-layer source representations through intermediate-layer query-object representations to late-layer answer states. Targeted interventions further establish causal links along this progression: manipulating source-side representations shifts location information at query-object mentions and ultimately alters relation predictions. Beyond object-location information, we also identify a stable query-side direction associated with the roles of the two objects in the comparison. Steering along directions estimated on synthetic scenes generalizes to natural-image benchmarks, improving accuracy and both forms of paired consistency in most settings without retraining. Our findings reveal complementary components of relational reasoning across visual and textual settings and show how targeted interventions can improve the consistency of models' behavior.
Chinese Translation
高实例级准确性可能掩盖空间推理中的不一致性，尤其是在对象交换位置或其角色在查询中被反转时。支持相对位置推理的内部表征仍然不够清晰。我们研究了这一过程的两个互补组成部分：输入中对象位置的跟踪和它们的查询角色的表征。在三种具有视觉或文本输入及其语言模型骨干的视觉语言模型（VLMs）中，激活补丁揭示了从早期层源表征、通过中间层查询-对象表征到晚期层答案状态的分阶段进展。针对性的干预进一步建立了沿这一进展的因果关系：操控源侧表征会改变查询-对象提及时的位置信息，并最终改变关系预测。除了对象位置的信息外，我们还识别出与比较中两个对象角色相关的稳定查询侧方向。在合成场景上估计的方向引导能够推广到自然图像基准，在大多数设置中提高准确性和两种形式的配对一致性，而无需重新训练。我们的发现揭示了视觉和文本环境中关系推理的互补组成部分，并展示了如何通过针对性的干预改善模型行为的一致性。
cs.CL / 217 / 2609.35521
Beyond Token Scale: Chunk-Level Sparse Autoencoders for Reliable Semantic Feature Discovery
超越标记规模：用于可靠语义特征发现的块级稀疏自编码器
Abstract
Sparse autoencoders (SAEs) expose features that help us understand and steer language models, but faithful reconstruction does not guarantee informative concepts. Token-level objectives reward lexical and formatting details alongside semantic content, all competing for a limited sparse budget. We introduce a family of chunk-level SAEs that encode mean-pooled activations over chunks, each a contiguous span of tokens: Mean-Chunk reconstructs the observed chunk, Cross-Chunk predicts an independently processed neighbor, and Joint-Chunk combines both targets. These designs separate the effect of a larger observation unit from that of predicting information shared across passages. With matched training data, chunk-level SAEs remain powerful interpretability tools while learning reliable semantic features that capture high-level concepts and respond selectively to relevant content. Their strengths are complementary: Mean-Chunk improves high-level feature discovery, reasoning detection beyond surface cues, and steering; Cross-Chunk leads document retrieval and classification transfer while producing selective, persistent features. Changing what an SAE sees and predicts yields reliable semantic features for more meaningful tasks. We demonstrate their practical value through gains across downstream tasks such as retrieval, reasoning detection, and steering.
Chinese Translation
稀疏自编码器（SAEs）揭示了帮助我们理解和引导语言模型的特征，但忠实重构并不保证信息概念的有效性。标记级目标在语义内容的同时奖励词汇和格式细节，所有这些都在有限的稀疏预算中竞争。我们提出了一系列块级稀疏自编码器，它们对块的均值池化激活进行编码，每个块是一个连续的标记范围：Mean-Chunk重构观察到的块，Cross-Chunk预测一个独立处理的邻居，而Joint-Chunk结合了这两个目标。这些设计将较大观察单元的影响与跨段落共享信息的预测效果分开。在匹配的训练数据下，块级稀疏自编码器仍然是强大的可解释性工具，同时学习可靠的语义特征，捕捉高层次概念并对相关内容作出选择性响应。它们的优势是互补的：Mean-Chunk改善高层次特征发现，超越表面线索的推理检测和引导；Cross-Chunk在文档检索和分类转移中表现出色，同时产生选择性和持久的特征。改变稀疏自编码器所观察和预测的内容可以为更有意义的任务提供可靠的语义特征。我们通过在检索、推理检测和引导等下游任务中的提升，展示了它们的实际价值。
cs.CL / 218 / 2609.35544
Less Sycophancy, Stronger Refusal? Lessons for AI Safety from Mechanistic Interpretability
减少谄媚，增强拒绝？来自机械解释性的人工智能安全教训
Abstract
Reliable refusal of harmful requests is essential to the safe deployment of language models. Because excessive eagerness to please users may undermine existing refusal capabilities, reducing sycophancy offers a potential route to stronger refusal beyond the harmful scenarios covered by safety training. We investigate this possibility using compensatory feature injection (CFI), a training technique designed to limit the acquisition of a target concept by supplying its associated activation during learning. Across three Qwen3.5 base models, we use sparse autoencoders (SAEs) to identify the top-ranked sycophancy feature from paired sycophantic and independent responses, then validate its behavioral influence through inference steering. We subsequently inject the selected feature during supervised fine-tuning on sycophantic targets. Positive injection reduces learned sycophancy after removal (by 62.0% relative to ordinary fine-tuning in 35B-A3B), whereas modest negative injection increases it. Unexpectedly, these reductions in sycophancy do not consistently improve direct refusal of harmful requests, motivating a narrower evaluation of the same harmful intents under user pressure. In this setting, ordinary fine-tuning on sycophantic responses substantially weakens refusal, while selected checkpoints trained with positive injection recover part of the loss, including approximately 95% in 35B-A3B. These findings show that persistent sycophancy reduction does not guarantee stronger direct refusal, while identifying recovery under user pressure as a distinct, conditional benefit of training intervention.
Chinese Translation
可靠拒绝有害请求对于语言模型的安全部署至关重要。过度渴望取悦用户可能会削弱现有的拒绝能力，因此减少谄媚提供了一条潜在的途径，以实现超越安全训练所涵盖的有害场景的更强拒绝能力。我们使用补偿特征注入（Compensatory Feature Injection, CFI）这一训练技术来探讨这一可能性，该技术旨在通过在学习过程中提供与目标概念相关的激活来限制目标概念的获取。在三个 Qwen3.5 基础模型中，我们使用稀疏自编码器（Sparse Autoencoders, SAEs）从成对的谄媚和独立响应中识别出排名最高的谄媚特征，然后通过推理引导验证其行为影响。随后，我们在针对谄媚目标的监督微调过程中注入所选特征。正向注入在移除后减少了学习到的谄媚（相对于普通微调在 35B-A3B 中减少了 62.0%），而适度的负向注入则增加了谄媚。出乎意料的是，这些谄媚的减少并未始终改善对有害请求的直接拒绝，这促使我们在用户压力下对相同有害意图进行更狭义的评估。在这种情况下，针对谄媚响应的普通微调显著削弱了拒绝能力，而经过正向注入训练的选定检查点则恢复了部分损失，包括在 35B-A3B 中约 95% 的恢复。这些发现表明，持续减少谄媚并不保证更强的直接拒绝能力，同时在用户压力下识别恢复作为训练干预的一个独特条件性好处。
cs.CL / 219 / 2609.35564
Almieyar: A Culturally Grounded Benchmark for Multi-Dialect Arabic Speech Recognition
Almieyar：一个基于文化的多方言阿拉伯语语音识别基准
Ghahroodi, Omid, Madkoor, Anas, Saudi, Dima Faris Al, Tahir, Fagr, Annan, Malak, rakha, Talha shahid javad allah, Al-Busaidi, Omar, Kahla, Zineb El, Zouari, Iheb, Jabal, Essa Ahmed Abou, Ezzat, Ahmed, AL-Merekhi, Hind, Al-Naimi, Aisha Hamad M A, Wazni, Hadi, Alnajjar, Bushra, Amin, Omar, Al-Thani, Haya, Lachemat, Houssam Eddine-Othman, Elwakedy, Marwa, Nahari, Sundus Abdulmalik Al, Zahiri, Elahe, Sarraj, Osamah, Mousa, Raghad, Assi, Mckeen, Jumah, Ahd Al, Salman, Heyam, Abdulraqib, Alhanouf, Benoumhani, Sara, Hamwi, Alia, Al-Yasseri, Ayaat, Ghazal, Rim Ibrahim, hiba, Lamia Ben, Eltabakh, Mohamed, Al-Raisi, Fatima, Kheir, Yassine El, Abdulrahman, Mohammed, Mubarak, Hamdy, Hashem, Ayah, Meriem, Lefkir, Asgari, Ehsaneddin
Abstract
Arabic speech technology has largely focused on Modern Standard Arabic, leaving the living dialects spoken by hundreds of millions under-served. We introduce ALMIEYAR, a culturally grounded ASR benchmark covering 17 Arabic dialects across six families, built entirely from newly recorded speech unseen by existing models. Dialect-community coordinators selected culturally relevant images across 10 topics, and native speakers described them through five structured scenarios, yielding approximately 50 minutes per dialect (13.7 hours total). We benchmark 12 state-of-the-art ASR systems zero-shot, including GPT-4o-transcribe, Voxtral-Mini-4B, Fanar-STT-LF, Whisper, SeamlessM4T-v2, and wav2vec2-based models. GPT-4o-transcribe achieves the lowest overall WER at 35.0%, followed by Voxtral-Mini-4B, Fanar-STT-LF, and Whisper-Large-v3 at 41.1%, 45.9%, and 49.5%, respectively, indicating substantial remaining errors across Arabic dialect communities. Performance varies considerably across dialect groups, with no model performing uniformly best across all groups. WER alone also obscures dialectal ASR behaviour: wav2vec2-based models show large WER/CER gaps, where character-level agreement remains much higher than word-level accuracy, motivating joint WER/CER reporting. ALMIEYAR provides a unified benchmark for culturally grounded Arabic ASR evaluation, including the first published benchmark for Ahwazi Arabic.
Chinese Translation
阿拉伯语语音技术主要集中在现代标准阿拉伯语上，导致数亿人使用的活方言未得到充分服务。我们介绍了ALMIEYAR，这是一项基于文化的自动语音识别（ASR）基准，涵盖六个语系中的17种阿拉伯方言，完全由新录制的、现有模型未见过的语音构成。方言社区协调员在10个主题中选择了文化相关的图像，母语者通过五个结构化场景对其进行了描述，每种方言大约产生50分钟的语音（总计13.7小时）。我们对12个最先进的ASR系统进行了零样本基准测试，包括GPT-4o-transcribe、Voxtral-Mini-4B、Fanar-STT-LF、Whisper、SeamlessM4T-v2和基于wav2vec2的模型。GPT-4o-transcribe的整体字错误率（WER）最低，为35.0%，其次是Voxtral-Mini-4B、Fanar-STT-LF和Whisper-Large-v3，分别为41.1%、45.9%和49.5%，这表明阿拉伯方言社区仍存在大量错误。不同方言组的表现差异显著，没有任何模型在所有组中均表现最佳。仅通过WER也无法全面反映方言ASR的表现：基于wav2vec2的模型显示出较大的WER/CER差距，字符级一致性远高于词级准确性，这促使我们进行联合WER/CER报告。ALMIEYAR为基于文化的阿拉伯语ASR评估提供了统一的基准，包括首个发布的阿哈瓦兹阿拉伯语基准。
cs.CL / 220 / 2609.35578
FactorEngram: Factorized N-gram Memory with Basis-Level Gating for Language Models
FactorEngram：具有基础级门控的因子化 N-gram 记忆用于语言模型
Abstract
Lookup-based memory has been a promising way to scale the parameters of large language models (LLMs). It retrieves learned representations of local token patterns, such as n-grams, instead of reconstructing them through successive layers of computation. However, existing designs such as Engram treat each retrieved embedding as a monolithic unit. Each embedding is stored in its own hashed slot and modulated by a single scalar gate. As a result, polysemous patterns cannot selectively read out the components of their memory that are relevant to the context. Moreover, parameters are shared only through hash collisions, which are largely unrelated to semantics. We propose FactorEngram, a factorized n-gram memory with basis-level contextual gating. FactorEngram retrieves sparsity-regularized coefficients over a dictionary of basis vectors shared across patterns, so related patterns can reuse common components. The same dictionary is also used for gating. The backbone hidden state is scored against each basis vector to gate the corresponding coefficient before reconstruction, which lets the context modulate each memory component individually. FactorEngram also covers both individual tokens and multi-token n-grams, and we systematically study where the memory branch should be inserted. On 340M- and 1B-parameter Transformer backbones, FactorEngram improves language modeling and downstream task performance. Ablation studies confirm the contribution of each component and identify insertion before the attention sublayer in the middle layers as an effective configuration.
Chinese Translation
基于查找的记忆已成为扩展大型语言模型（LLMs）参数的有希望的方法。它检索局部标记模式（如 n-grams）的学习表示，而不是通过连续的计算层重构这些表示。然而，现有设计如 Engram 将每个检索到的嵌入视为一个整体单元。每个嵌入存储在其自己的哈希槽中，并由单一标量门控调节。因此，多义模式无法选择性地读取与上下文相关的记忆组件。此外，参数仅通过哈希冲突共享，这与语义大多无关。我们提出了 FactorEngram，一种具有基础级上下文门控的因子化 n-gram 记忆。FactorEngram 在共享的基础向量字典上检索稀疏正则化系数，以便相关模式可以重用公共组件。同一字典也用于门控。主干隐藏状态与每个基础向量进行评分，以在重构之前对相应的系数进行门控，这使得上下文能够单独调节每个记忆组件。FactorEngram 还涵盖了单个标记和多标记 n-grams，并且我们系统地研究了记忆分支应插入的位置。在 340M 和 1B 参数的 Transformer 主干上，FactorEngram 改善了语言建模和下游任务的性能。消融研究确认了每个组件的贡献，并确定在中间层的注意力子层之前插入作为有效配置。
cs.CL / 221 / 2609.35591
Language Models Act on Hidden Valence
语言模型对隐性效价的作用
Abstract
Language models describe some internal states as good and others as bad. But whether models have a stake in them is an open question. Simply asking the model is unlikely to be informative. Any answer may be consistent with genuine introspection, superficial pattern-matching, or with fixed scripts learned in character training. We therefore study revealed preference. Rather than asking about a state, we use activation steering to attach a positively or negatively valenced activation pattern to one of two otherwise meaningless 'zones', switch steering off, and then observe which zone the model prefers. A model with a stake in that state should choose accordingly. Across seven open-weight models from five families, this is indeed what we find. First, steering changes the passages models write about each zone, and those words shift later choice. Second, the shift persists when all surface-level tokens are held fixed and only the hidden KV cache differs. Third, the effect also remains when all text is generated without steering and valence is only injected during cache construction. Thus, the hidden state alone moves choice in proportion to the steering dose. Fourth, this dependence of choice on hidden valence is nearly absent in a base model and emerges during DPO, consistent with a link between valence and goal-directed behaviour formed in training. Finally, given tools to steer itself, a model does not tend to induce a positive state, but it reliably removes an imposed negative state. It does so at a dose-dependent rate and significantly more often than it removes interventions in random directions. Overall, we demonstrate that valence-related activation patterns leave hidden traces that predictably govern later choices, even when every visible token is identical across conditions. Whether these traces are accompanied by any subjective experience relevant to model welfare remains unclear.
Chinese Translation
语言模型将某些内部状态描述为积极的，而将其他状态描述为消极的。然而，模型是否对这些状态有利害关系仍然是一个未解的问题。简单地询问模型可能不会提供有用的信息。任何回答都可能与真正的内省、一种表面的模式匹配，或与在角色训练中学习到的固定脚本一致。因此，我们研究了显性偏好。我们不直接询问某一状态，而是使用激活引导将一个积极或消极的激活模式附加到两个无意义的“区域”之一，关闭引导，然后观察模型偏好哪个区域。一个对该状态有利害关系的模型应该相应地做出选择。在来自五个家族的七个开放权重模型中，我们确实发现了这一点。首先，引导改变了模型在每个区域写作的段落，这些词汇随后影响了后续的选择。其次，当所有表面级标记保持固定，仅隐藏的键值缓存（KV cache）不同的时候，这种变化依然存在。第三，当所有文本在没有引导的情况下生成，且效价仅在缓存构建期间注入时，这种效应仍然存在。因此，仅隐藏状态的变化与引导剂量成正比地影响选择。第四，这种选择对隐性效价的依赖在基础模型中几乎不存在，而在 DPO（直接偏好优化）过程中出现，这与训练中形成的效价与目标导向行为之间的联系一致。最后，给定自我引导的工具，模型并不倾向于诱发积极状态，但它可靠地消除施加的消极状态。它以剂量依赖的速率进行，并且显著更频繁地去除随机方向的干预。总体而言，我们展示了与效价相关的激活模式留下了隐性痕迹，这些痕迹可预测地支配后续选择，即使在所有可见标记在不同条件下完全相同的情况下。这些痕迹是否伴随有任何与模型福利相关的主观体验仍然不清楚。
cs.CL / 222 / 2609.35627
Can LLMs Value the Right Evidence? Evidence-Value Misalignment in Dynamic Medical Diagnosis
大型语言模型能否评估正确的证据？动态医学诊断中的证据价值不匹配
Abstract
A correct diagnosis reached from insufficient or misleading evidence can pose a clinical hazard, yet outcome-based accuracy may reward such lucky guesses. We call this mismatch between diagnostic decisions and the value of available evidence Evidence-Value Misalignment (EVM). To disentangle evidential grounding independently from diagnostic accuracy, we introduce MedEVM, a dynamic benchmarking environment comprising 1,050 cases across 24 disease systems. Observations arrive turn by turn, requiring models to continuously calibrate its decision by deciding whether to wait for more evidence or submit a diagnosis. Across 9 LLMs, four interesting patterns are observed. (1) Miscalibrated evidence tracking. Making a diagnosis often fails to calibrate evidence sufficiency, even in more capable models, and even worsens in reasoning mode. (2) Misaligned diagnosis submission. Confidence in the correct diagnosis often fails to ensure timely submission despite sufficient evidence. (3) Evidence order matters. Reordering the same evidence changes diagnoses even when model confidence remains similar. (4) Misleading evidence remains influential. Added misleading evidence redirects diagnoses even after prior evidence becomes sufficient. We further verify that EVM predicts errors and that preventing premature submission improves accuracy. These findings motivate Evidence-Verified Diagnosis Harness (EVD-Harness). It decouples diagnosis generation from submission through an offline Contrastive Diagnostic Wiki and three online control stages, namely observation management, proposal and witness verification, and diagnosis submission control. Across five LLMs, EVD-Harness improves accuracy by 12.0--51.1 percentage points while mitigating EVM-related failures. Our results demonstrate that verifying evidential support before submission can make diagnostic decisions more reliable.
Chinese Translation
从不足或误导性证据中得出的正确诊断可能会带来临床风险，但基于结果的准确性可能会奖励这种幸运的猜测。我们将诊断决策与可用证据的价值之间的这种不匹配称为证据价值不匹配（Evidence-Value Misalignment, EVM）。为了独立于诊断准确性地解开证据基础，我们引入了MedEVM，这是一个动态基准环境，包含1,050个案例，涵盖24个疾病系统。观察数据逐步到达，要求模型不断校准其决策，决定是等待更多证据还是提交诊断。在9个大型语言模型中观察到四个有趣的模式。(1) 证据跟踪失调。即使在更强大的模型中，做出诊断往往未能校准证据的充分性，且在推理模式下情况更糟。(2) 诊断提交不一致。对正确诊断的信心往往未能确保及时提交，即使证据充分。(3) 证据顺序的重要性。重新排序相同的证据会改变诊断，即使模型信心保持相似。(4) 误导性证据仍然具有影响力。添加的误导性证据会重新引导诊断，即使先前的证据已经充分。我们进一步验证了EVM可以预测错误，并且防止过早提交可以提高准确性。这些发现促使我们开发证据验证诊断工具（Evidence-Verified Diagnosis Harness, EVD-Harness）。它通过离线对比诊断维基和三个在线控制阶段，即观察管理、提案和见证验证、以及诊断提交控制，解耦诊断生成与提交。在五个大型语言模型中，EVD-Harness将准确性提高了12.0至51.1个百分点，同时减轻了与EVM相关的失败。我们的结果表明，在提交之前验证证据支持可以使诊断决策更可靠。
cs.CL / 223 / 2609.35630
Which the Eye Fears: Writing with Read-Blindness Explains Massive Activations in Transformers
眼睛所惧：阅读盲点的写作解释了变换器中的大规模激活
Abstract
Massive activation features (MAs) in Transformers are extreme-value residual-stream features that persist across layers despite the model's ability to suppress them. Why do they survive? Our investigation using an operator-level mechanistic analysis of attention and feed-forward (FFN) blocks reveals that these blocks systematically ignore MA coordinates while reading, but not while writing; creating a read-write asymmetry that blocks corrective feedback while allowing continued accumulation. We find that both attention and feed-forward layers have this read-blindness, and contribute to the emergence and persistence of MAs. To validate prior work that hypothesized that FFN's amplification abilities is the primary reason for MAs (Sun et al., 2026), we analyze the model checkpoints during learning. Contrary to our expectation, read-blindness emerges before FFN amplification, suggesting that it acts upstream in the MA mechanism. We further contribute gradient analysis to link this behavior to surprising asymmetries in the loss landscape, concluding that the model actively maintains this read-blindness. Finally, we find that removing read-blocking at different locations induces compensatory shifts elsewhere, but MAs still persist.
Chinese Translation
变换器中的大规模激活特征（MAs）是极值残差流特征，尽管模型能够抑制它们，但在各层中仍然存在。它们为何能够存活？我们通过对注意力和前馈神经网络（FFN）模块的操作级机制分析发现，这些模块在阅读时系统性地忽略了MA坐标，但在写作时却没有；这造成了一种阅读-写作不对称，阻碍了纠正反馈的产生，同时允许持续的积累。我们发现注意力层和前馈层都存在这种阅读盲点，并且对MAs的出现和持续性有贡献。为了验证之前的研究假设，即FFN的放大能力是MAs的主要原因（Sun et al., 2026），我们分析了学习过程中的模型检查点。与我们的预期相反，阅读盲点在FFN放大之前就已经出现，这表明它在MA机制中起着上游作用。我们进一步贡献了梯度分析，将这种行为与损失景观中的惊人不对称性联系起来，得出结论：模型主动维持这种阅读盲点。最后，我们发现，在不同位置去除阅读阻断会在其他地方引发补偿性变化，但MAs仍然持续存在。
cs.CL / 224 / 2609.35646
Rubric Rewards from Item Response Theory
基于项目反应理论的评分标准奖励
Abstract
Many language tasks have no single answer that can be checked automatically. Rubrics provide criteria for judging responses to these tasks. For reinforcement learning, the resulting verdicts must be combined into a scalar reward. A common approach sums the points assigned to satisfied criteria. Distinct verdict patterns can thus receive the same reward, and the fixed points encode how much each criterion should count, not how strongly its verdict distinguishes the current rollouts. Beyond this aggregation problem, judging the full rubric needs more judge requests as the criterion count grows. To address these limitations, Rubric Response Theory (RRT) measures quality and selects criteria when rubric criteria are monotone indicators of a shared target. Rather than adding assigned points, RRT uses a two parameter item response model that treats the verdict pattern as evidence about scalar quality specific to the rubric. Under this model, its likelihood score maximizes the local signal-to-noise ratio for quality. Its Response Parameter Network (RPN) reads the prompt and criterion text to predict criterion difficulty and discrimination. As the policy distribution changes during training, RRT uses online expectation maximization to update the RPN from current rollout verdicts. With Qwen3.5-4B as the policy, RRT's macro criterion score across Medical, Science, Rubrics as Rewards Science, and RubricBench is 1.7 points above that of group relative policy optimization (GRPO). On hard and very hard criteria in Medical and Science, RRT gains 2.8 to 5.6 points over GRPO. At half the criterion budget, adaptive Fisher selection with a frozen RPN keeps the macro criterion score across four datasets within 0.1 points of GRPO with full judging. These results show RRT can reduce judge requests while remaining competitive with GRPO.
Chinese Translation
许多语言任务没有单一的答案可以自动检查。评分标准提供了评判这些任务响应的标准。对于强化学习，得到的判决必须合并为一个标量奖励。一种常见的方法是将满足标准的分数相加。因此，不同的判决模式可能会获得相同的奖励，而固定的分数编码了每个标准应有的权重，而不是它的判决在当前回合中区分的强度。除了这一聚合问题外，随着标准数量的增加，评判完整评分标准需要更多的评审请求。为了解决这些局限性，评分标准响应理论（Rubric Response Theory, RRT）在评分标准是共享目标的单调指标时，衡量质量并选择标准。RRT并不是简单地增加分配的分数，而是使用一个双参数项目反应模型，将判决模式视为关于特定于评分标准的标量质量的证据。在该模型下，其似然得分最大化了质量的局部信噪比。其响应参数网络（Response Parameter Network, RPN）读取提示和标准文本，以预测标准的难度和区分度。随着训练过程中策略分布的变化，RRT使用在线期望最大化从当前回合的判决中更新RPN。在使用Qwen3.5-4B作为策略时，RRT在医学、科学、评分标准作为奖励科学和RubricBench上的宏观标准得分比组相对策略优化（Group Relative Policy Optimization, GRPO）高出1.7分。在医学和科学中的困难和非常困难标准上，RRT比GRPO高出2.8到5.6分。在一半的标准预算下，使用冻结的RPN进行自适应Fisher选择，使得在四个数据集上的宏观标准得分与完全评审的GRPO保持在0.1分以内。这些结果表明，RRT能够减少评审请求，同时在竞争中保持与GRPO的竞争力。
cs.CL / 225 / 2609.35663
Late Attention Layers Alone Can Copy Entity Tokens, but Not Without Attending to Their Context
仅靠后期注意力层可以复制实体标记，但必须关注其上下文
Abstract
Large language models (LLMs) reliably perform entity copying, in which a model copies tokens referring to an entity, termed entity tokens, from the prompt into its output to answer a question. Although entity copying is straightforward for most LLMs, existing research does not provide a systematic account of which layers specialize in this fundamental task or how other tokens in the same sequence, termed context tokens, influence the model's ability to copy the entity tokens. To address these questions, we conduct experiments on Qwen3-8B using two novel methods: genie-in-a-bottle, which controls exactly which layers can participate in an entity-copying task, and attention lobotomy, which cuts off specific tokens' attention to entity tokens without affecting the remaining attention distribution. We find that two distinct groups of layers in the second half of the model are both necessary and sufficient for entity copying. Moreover, in addition to the decoding position's attention to entity tokens, context tokens' attention to entity tokens also proves necessary for copying the exact tokens, even though context tokens do not store entity information themselves unless they satisfy particular semantic properties. Our findings establish the critical role of late layers in entity copying under the guidance of context tokens, calling for future work on how models propagate and consume entity information.
Chinese Translation
大型语言模型（LLMs）能够可靠地执行实体复制，其中模型将指向实体的标记（称为实体标记）从提示中复制到其输出中以回答问题。尽管对于大多数LLMs而言，实体复制是直接的，但现有研究并未系统性地阐明哪些层专门负责这一基本任务，或同一序列中其他标记（称为上下文标记）如何影响模型复制实体标记的能力。为了解决这些问题，我们在 Qwen3-8B 上进行实验，使用两种新方法：genie-in-a-bottle，该方法精确控制哪些层可以参与实体复制任务，以及 attention lobotomy，该方法切断特定标记对实体标记的注意力，而不影响其余注意力分布。我们发现模型后半部分的两个不同层组对于实体复制既是必要的也是充分的。此外，除了解码位置对实体标记的注意力外，上下文标记对实体标记的注意力在复制确切标记时也是必要的，即使上下文标记本身并不存储实体信息，除非它们满足特定的语义属性。我们的研究结果确立了后期层在上下文标记指导下进行实体复制的关键作用，并呼吁未来的研究探讨模型如何传播和利用实体信息。
cs.CL / 226 / 2609.35664
MS-GLA: Multi-Scale Gated Linear Attention for Addressing Representational Bottlenecks via Multi-Temporal Resolution
MS-GLA：多尺度门控线性注意力通过多时间分辨率解决表征瓶颈
Abstract
Gated Linear Attention (GLA) Transformers advance linear recurrent models through data-dependent gating, but face a core limitation: the fixed-capacity memory matrices across all heads operate at a single temporal resolution, where each token is processed individually, forcing them to simultaneously encode local syntactic patterns and long-range semantic structure, creating a representational bottleneck that gating alone is insufficient to resolve. We introduce Multi-Scale Gated Linear Attention (MS-GLA), which addresses this by distributing attention heads across multiple temporal resolutions. Coarser resolutions pool longer token spans naturally specializing toward long-range dependencies, while finer head groups retain sensitivity to local syntactic structure. A learnable, input-dependent fusion layer dynamically recombines head group outputs at each timestep, expanding effective memory capacity without increasing per-head state size. This multi-resolution decomposition draws on principles from Multi-Scale State-Space Models (MS-SSM), adapting them to the gated linear attention setting. We evaluate MS-GLA on language modeling, recall-intensive tasks, and long-context generalization. Across all settings, MS-GLA consistently achieves higher accuracy and lower perplexity than GLA at matched parameter counts, with up to 18.9% improvement on recall-intensive tasks and 9.5% lower average perplexity on language modeling benchmarks, validating multi-temporal resolution decomposition as a principled and effective extension of Gated Linear Attention.
Chinese Translation
门控线性注意力（GLA）变换器通过数据依赖的门控机制推进了线性递归模型的发展，但面临一个核心限制：所有头部的固定容量记忆矩阵在单一时间分辨率下操作，每个标记单独处理，迫使它们同时编码局部句法模式和长程语义结构，造成了仅靠门控机制无法解决的表征瓶颈。我们提出了多尺度门控线性注意力（MS-GLA），通过在多个时间分辨率上分配注意力头来解决这一问题。较粗的分辨率自然地聚合更长的标记跨度，专注于长程依赖，而较细的头部组则保持对局部句法结构的敏感性。一个可学习的、依赖输入的融合层在每个时间步动态地重新组合头部组的输出，扩展有效的记忆容量而不增加每个头部的状态大小。这种多分辨率分解借鉴了多尺度状态空间模型（MS-SSM）的原理，并将其适应于门控线性注意力的设置。我们在语言建模、回忆密集型任务和长上下文泛化上评估了MS-GLA。在所有设置中，MS-GLA在匹配参数数量的情况下始终实现了比GLA更高的准确率和更低的困惑度，在回忆密集型任务上提高了多达18.9%，在语言建模基准上平均困惑度降低了9.5%，验证了多时间分辨率分解作为门控线性注意力的一个有原则且有效的扩展。
cs.CL / 227 / 2609.35674
Tracing the Evolution of Oracle Bone Characters Across Three Millennia
追踪甲骨文字符在三千年间的演变
Abstract
Of the approximately 4,500 Oracle Bone Inscription (OBI) characters discovered from the Shang dynasty, only about 1,600 have been deciphered. Many computational approaches compare OBI with glyphs from one historical period at a time. However, during the evolution of Chinese characters, significant structural or semantic changes often occur in uncertain dynasties. A single-period reference may be insufficient when relevant forms change substantially between observed eras. Therefore, we propose the \textbf{Manifold-based Script Evolution Framework (MSEF)}, a framework that models the evolution series (OBI, Bronze, Seal, Clerical, Regular) of Chinese characters as the continual evolution of a manifold space. MSEF represents each character as an era-specific manifold point and learns continuous inter-era transition rules via Neural Ordinary Differential Equations. Both manifold space and transition dynamics can be trained end-to-end through character evolution pairs across any two eras.
Chinese Translation
在发现的约4500个甲骨文（Oracle Bone Inscription, OBI）字符中，仅有约1600个被解读。许多计算方法一次性将甲骨文与某一历史时期的符号进行比较。然而，在汉字的演变过程中，往往在不确定的朝代中发生显著的结构或语义变化。当相关形式在观察的时代之间发生显著变化时，单一时期的参考可能不足。因此，我们提出了 extbf{基于流形的文字演变框架（Manifold-based Script Evolution Framework, MSEF）}，该框架将汉字的演变系列（甲骨文、青铜器文字、篆书、隶书、楷书）建模为流形空间的持续演变。MSEF将每个字符表示为特定时代的流形点，并通过神经常微分方程学习连续的跨时代过渡规则。流形空间和过渡动态均可以通过任何两个时代的字符演变对进行端到端训练。
cs.CL / 228 / 2609.35685
QuanReview: Offline, Auditable Reconciliation of Human and LLM Span Annotations
QuanReview：人类与大型语言模型跨度注释的离线可审计对账
Abstract
Structured span annotations, such as quantities with their units, uncertainty modifiers, and event classes, are expensive to create and hard to keep trustworthy once language models enter the loop. We present QuanReview, an open-source system for auditing and correcting such annotation layers. QuanReview aligns two annotation streams over the same documents at character level, resolves unambiguous cases by an explicit and logged policy, and routes candidate conflicts to a browser-based adjudication interface where reviewers accept either side, build field-level hybrids, or flag items for re-annotation. A campaign manager assigns documents to multiple annotators with configurable redundancy, computes agreement at document and span level, auto-merges unanimous documents, and exports the corrected layer in the original file format, so that it can replace the original annotation files directly. Applied to a 4,457-record humanitarian benchmark and an LLM extraction stream, the system fully auto-merged 8% of documents, applied automatic policy decisions to a further 1,513 records, and concentrated human attention on 3,131 candidate conflicts, a mean of 5.4 per reviewed document.
Chinese Translation
结构化的跨度注释，例如带单位的数量、不确定性修饰符和事件类别，创建成本高且在语言模型介入后难以保持可信。我们提出了QuanReview，一个用于审计和纠正此类注释层的开源系统。QuanReview在字符级别对同一文档的两个注释流进行对齐，通过明确且记录的政策解决无歧义的案例，并将候选冲突路由到基于浏览器的裁决界面，审阅者可以接受任一方、构建领域级混合注释或标记项目以进行重新注释。活动管理者将文档分配给多个注释者，并配置冗余，计算文档和跨度级别的一致性，自动合并一致的文档，并以原始文件格式导出修正后的层，以便直接替换原始注释文件。在一个包含4,457条记录的人道主义基准和一个大型语言模型提取流的应用中，该系统完全自动合并了8%的文档，对进一步的1,513条记录应用了自动政策决策，并将人类注意力集中在3,131个候选冲突上，平均每个审阅文档5.4个。
cs.CL / 229 / 2609.35738
Harness Learning Enables Generalizable Test-Time Adaptation
工具学习实现可泛化的测试时间适应
Abstract
A language-model agent is jointly defined by its model and its harness, the executable program that organizes model calls, tool use, and information flow. Because different tasks call for different ways of organizing these operations, the harness needs to be adapted using feedback from the task at hand. We introduce harness learning, which trains a proposer model to revise a solver's harness using execution feedback. We formulate this process as meta-learning over executable programs, with harness revisions playing the role of weight updates in gradient-based adaptation. We train the proposer with reinforcement learning, using the task performance of revised harnesses as the reward. At test time, the proposer uses feedback from successive executions on a new task to refine the harness, without performing any parameter-space update. Experiments on reasoning and multi-hop question answering show that harness learning improves revision quality and that the ability to adapt at test time transfers to unseen tasks. Policies trained on individual revisions can continue improving harnesses over multiple rounds, while the benefits of training on revision sequences vary across settings. These findings suggest a path towards continually learning agents that turn accumulated experience into generalizable improvements.
Chinese Translation
语言模型代理由其模型和工具共同定义，工具是组织模型调用、工具使用和信息流的可执行程序。由于不同任务需要以不同方式组织这些操作，因此工具需要根据当前任务的反馈进行调整。我们提出了工具学习（harness learning），该方法训练一个提议模型（proposer model）以利用执行反馈修订解算器的工具。我们将这一过程形式化为可执行程序上的元学习（meta-learning），其中工具修订充当基于梯度的适应中的权重更新。我们使用强化学习（reinforcement learning）训练提议者，以修订后的工具的任务表现作为奖励。在测试时，提议者利用在新任务上连续执行的反馈来优化工具，而无需进行任何参数空间更新。在推理和多跳问答的实验中，工具学习提高了修订质量，并且在测试时间的适应能力能够转移到未见过的任务上。针对单个修订训练的策略可以在多个回合中继续改进工具，而在修订序列上训练的好处在不同设置中有所不同。这些发现为不断学习的代理指明了一条路径，使其能够将累积的经验转化为可泛化的改进。
cs.CL / 230 / 2609.35748
Improving Test-Time Scaling with Adaptive Looped Transformers
通过自适应循环变换器改善测试时间缩放
Abstract
Looped transformers have demonstrated promising parameter efficiency by reusing layers for latent computation. Prior studies compare looped and non-looped models at matched parameters or per-token FLOPs. However, to the best of our knowledge, whether looping improves test-time scaling as outputs grow longer remains underexplored. Through post-training looped transformers, we study the accuracy-compute slope, measured as the accuracy gain per doubling of test-time decoding FLOPs. We find that existing looped transformers often yield steeper slopes than their non-looped baseline, yet underperform it at matched compute. While fixed-depth looping spends extra iterations on every token, our analysis shows that many tokens do not benefit from extra iterations. We therefore propose TaH2, which enables the model to focus extra iterations on the tokens that benefit from looping. It jointly post-trains the backbone and an iteration decider through lookahead depth supervision, which uses online labels indicating whether further iteration improves the prediction. TaH2 improves both the efficiency and attainable accuracy of test-time scaling. On challenging AIME benchmarks, TaH2 improves the accuracy-compute slope by 53% (2.74 vs. 1.79) over the non-looped baseline, exceeding the baseline's peak accuracy by about 3.4 points at matched test-time compute. As the maximum iteration depth increases, existing looped models largely plateau, while TaH2's gain over the non-looped baseline continues to grow from +2.8 points at depth 2 to +3.9 points at depth 8. Our code is available at https://github.com/thu-nics/TaH.
Chinese Translation
循环变换器通过重用层进行潜在计算，展示了良好的参数效率。先前的研究在匹配参数或每个令牌的浮点运算（FLOPs）方面比较了循环和非循环模型。然而，据我们所知，循环是否能在输出变长时改善测试时间缩放仍然未被充分探讨。通过后训练的循环变换器，我们研究了准确性-计算斜率，该斜率被定义为每次翻倍测试时间解码FLOPs所带来的准确性提升。我们发现，现有的循环变换器通常比其非循环基线产生更陡的斜率，但在匹配计算时表现不佳。虽然固定深度循环在每个令牌上花费额外的迭代，但我们的分析表明，许多令牌并未从额外的迭代中受益。因此，我们提出了TaH2，它使模型能够将额外的迭代集中在受益于循环的令牌上。TaH2通过前瞻深度监督联合后训练主干和迭代决策器，该监督使用在线标签指示进一步的迭代是否改善预测。TaH2提高了测试时间缩放的效率和可达准确性。在具有挑战性的AIME基准测试中，TaH2在非循环基线之上将准确性-计算斜率提高了53%（2.74对1.79），在匹配的测试时间计算下超越基线的峰值准确性约3.4分。随着最大迭代深度的增加，现有的循环模型大多趋于平稳，而TaH2相对于非循环基线的提升从深度2的+2.8分增长到深度8的+3.9分。我们的代码可在https://github.com/thu-nics/TaH获取。
cs.CL / 231 / 2609.35749
Towards Communication-Efficient Social Intelligence in Language Agents
朝着语言智能体的通信效率社会智能
Abstract
Socially intelligent language agents must negotiate, coordinate, and resolve conflicting preferences while respecting the time and attention of both participants. Balancing these demands is challenging because agents must convey enough to address a partner's constraints and advance their goals without adding words that do not help the interaction. In this paper, we propose Teacher-Assisted Communication Training (TACT) to improve social goal attainment while reducing communication cost, making interactions with agents more productive and less demanding. We first characterize communication efficiency in terms of action strategy and expression, whose effects extend beyond the current utterance to the partner's response and subsequent exchanges. We design TACT to revise student-generated actions, test the revisions through partner responses, and distill useful feedback into the student. An expression specialist removes unnecessary detail while preserving the intended action, while a strategy specialist proposes alternatives that may better address the partner's constraints. To determine which revision helps, TACT samples a partner response for each candidate and selects a teacher reference by balancing local goal support against action-token cost. That reference guides on-policy distillation on the student's own generation prefixes, allowing the student to act independently at deployment. We evaluate TACT on SOTOPIA and AgentSense. On SOTOPIA, it achieves the highest Goal among the evaluated methods on All and Hard while using substantially fewer target tokens than SFT+SDPO. On AgentSense, it improves goal success over the initial student while reducing target tokens and interaction messages.
Chinese Translation
具有社会智能的语言智能体必须在尊重双方参与者的时间和注意力的同时，进行协商、协调和解决冲突的偏好。平衡这些需求具有挑战性，因为智能体必须传达足够的信息以满足合作伙伴的限制并推动自身目标，而又不增加无助于互动的冗余词汇。本文提出了教师辅助通信训练（Teacher-Assisted Communication Training, TACT），旨在提高社会目标的实现，同时降低通信成本，使与智能体的互动更加高效且要求更低。我们首先从行动策略和表达的角度对通信效率进行表征，其影响超越当前的发言，延伸至合作伙伴的反应和后续交流。我们设计TACT以修正学生生成的行动，通过合作伙伴的反应测试这些修正，并将有用的反馈提炼给学生。表达专家在保留意图行动的同时去除不必要的细节，而策略专家则提出可能更好地满足合作伙伴限制的替代方案。为了确定哪个修正更有效，TACT为每个候选方案采样合作伙伴的反应，并通过平衡局部目标支持与行动标记成本来选择教师参考。这一参考指导学生在自身生成前缀上的政策蒸馏，使学生在部署时能够独立行动。我们在SOTOPIA和AgentSense上评估TACT。在SOTOPIA上，它在所有评估方法中实现了最高的目标成功率，尤其是在困难任务中，同时使用的目标标记显著少于SFT+SDPO。在AgentSense上，它在减少目标标记和互动信息的同时，提高了目标成功率。
cs.CL / 232 / 2609.35759
Scaling Long-Form Story Generation via Narrative State Tracking
通过叙事状态跟踪扩展长篇故事生成
Abstract
LLMs have demonstrated strong capabilities in creative writing. However, scaling them to full-length novels remains challenging, as maintaining narrative consistency becomes increasingly difficult. Existing story-generation methods typically focus on stories of up to about ten thousand words, leaving their ability to scale to full-length novels underexplored. In this work, we introduce Narrative State Tracking Agent (NstAgent), a training-free agentic framework that allows LLMs to track a structured narrative state including characters, past events and future requirements. We extend an existing benchmark to compare narrative consistency across lengths, and use it together with a writing-quality benchmark to systematically evaluate stories ranging from 10K to 100K words. We show that NstAgent achieves better narrative consistency and writing quality as stories grow longer, and neither of them degrades noticeably as length increases, suggesting that it provides an effective approach to scaling story generation toward full-length novels.
Chinese Translation
大型语言模型（LLMs）在创意写作方面展现了强大的能力。然而，将它们扩展到完整小说仍然面临挑战，因为保持叙事一致性变得越来越困难。现有的故事生成方法通常集中在约一万字的故事上，因此它们扩展到完整小说的能力尚未得到充分探索。在本研究中，我们引入了叙事状态跟踪代理（Narrative State Tracking Agent，NstAgent），这是一个无训练的代理框架，允许大型语言模型跟踪包括角色、过去事件和未来需求在内的结构化叙事状态。我们扩展了现有的基准测试，以比较不同长度故事的叙事一致性，并将其与写作质量基准结合使用，系统地评估从一万字到十万字的故事。我们展示了NstAgent在故事长度增加时能够实现更好的叙事一致性和写作质量，并且随着长度的增加，两者均没有明显下降，这表明它为将故事生成扩展到完整小说提供了一种有效的方法。
cs.CL / 233 / 2609.35765
Retrieving Biblical Intertextual References in Karen Blixen's Seven Gothic Tales
在卡伦·布里克森的《七个哥特式故事》中检索圣经的互文性参考
Abstract
Identifying intertextual references is central to literary scholarship, but computationally difficult when source material is transformed through paraphrase, allusion, historical language, and translation. We investigate this problem through biblical intertextuality in Karen Blixen's Seven Gothic Tales. Drawing on the commentary to a critical edition, we construct a benchmark of 189 annotated references and evaluate retrieval against all 31,170 verses of historically plausible Danish Old and New Testament translations. We compare TF-IDF and BM25 with multilingual and Danish sentence encoders, examine the effect of linguistic normalization, and fine-tune a Danish encoder using hard negatives and five-fold cross-validation. We analyze performance across automatically derived lexical-overlap strata representing quotations, paraphrases, and allusions. Linguistically normalized BM25 provides a strong zero-shot baseline, attaining an overall R@10 of 0.365 and retrieving every quotation within its ten highest-ranked verses. The best zero-shot dense model achieves a comparable overall score of 0.360 while performing better on allusions. Fine-tuning DFM-large raises its overall R@10 from 0.265 to 0.508 and more than doubles its performance on allusions, from 0.138 to 0.339. However, evaluation against editorial annotations alone understates the model's scholarly usefulness: a literary scholar judged seven of 30 selected rank-one predictions counted as false positives to be meaningful additional references. These findings show both the potential and the epistemic limits of computational intertextual retrieval. Rather than treating scholarly annotations as exhaustive or model outputs as discoveries, we propose retrieval models as heuristic co-readers that recover documented references and generate candidates for expert-led close reading.
Chinese Translation
识别互文性参考是文学研究的核心，但在源材料通过意译、典故、历史语言和翻译转化时，计算上变得困难。我们通过卡伦·布里克森的《七个哥特式故事》中的圣经互文性来研究这个问题。基于批判版的注释，我们构建了189个注释参考的基准，并在31,170节历史上合理的丹麦旧约和新约翻译中评估检索效果。我们比较了TF-IDF和BM25与多语言和丹麦句子编码器的表现，考察了语言规范化的影响，并使用困难负例和五折交叉验证对丹麦编码器进行了微调。我们分析了在自动生成的词汇重叠层次中表现的性能，这些层次代表了引用、意译和典故。经过语言规范化的BM25提供了一个强大的零样本基线，整体R@10达到了0.365，并在其十个最高排名的节中检索到了每一个引用。最佳的零样本密集模型达到了相当的整体得分0.360，同时在典故方面表现更佳。对DFM-large进行微调将其整体R@10从0.265提高到0.508，并且在典故上的表现翻倍，从0.138提升至0.339。然而，仅针对编辑注释的评估低估了模型的学术实用性：一位文学学者认为30个选定的排名第一的预测中有7个被视为假阳性，但被认为是有意义的额外参考。这些发现展示了计算互文检索的潜力和认知局限性。我们建议将检索模型视为启发式的共同阅读者，既恢复已记录的参考文献，又为专家主导的细读生成候选项，而不是将学术注释视为详尽无遗或模型输出视为发现。
cs.CL / 234 / 2609.35769
Telescopic Language Models
望远镜语言模型
Abstract
One deployed language model must often serve many compute budgets, yet serving each budget still means a separate training or compression run per point. We train a Telescopic Language Model (TLM) to be that continuum: a nested-capacity Transformer supervised by stochastic prefix supervision with a full anchor. At every step, one randomly truncated prefix of the capacity axis is trained against the full next-token target, alongside one full-capacity pass, so the trained artifact is a valid language model at every depth. Two forward-backward passes per step, no architectural change, nothing extra at inference. Fixed-exit suites such as Matryoshka Language Model Suites (MLMS) occupy one point in this design space, and the point has a cost: supervising only a few fixed exits leaves the nested model at chance level everywhere else (perplexity 10^2-10^5 in our baselines). On a 200M proxy suite (20B FineWeb-Edu tokens, identical data stream for all methods), a single TLM run is a valid language model at every one of its twenty layer prefixes, in perplexity and on perplexity-sensitive downstream tasks, reducing the area under the quality-budget curve by 43-44% relative to the fixed-exit suites while matching them at full capacity, at ~12% lower GPU cost per run. The prefix sampling density is a dial: concentrating it on a few depths recovers fixed-exit quality there at the price of the continuum, so the operating points become a training-time choice rather than an architectural one. These results indicate that the training objective, not the nesting itself, is what makes a model elastic.
Chinese Translation
一个部署的语言模型通常必须服务于多个计算预算，但为每个预算提供服务仍然意味着每个点都需要单独的训练或压缩运行。我们训练了一个望远镜语言模型（Telescopic Language Model, TLM）来实现这一连续性：一个嵌套容量的Transformer，采用随机前缀监督和完整锚点进行监督。在每一步中，随机截断的容量轴前缀与完整的下一个标记目标进行训练，同时进行一次完整容量的传递，因此训练得到的模型在每个深度都是有效的语言模型。每一步进行两次前向-后向传递，没有架构变化，推理时没有额外开销。固定退出套件，如马特里奥什卡语言模型套件（Matryoshka Language Model Suites, MLMS），在这个设计空间中占据一个点，而这个点是有成本的：仅监督少数固定退出使得嵌套模型在其他地方的表现处于随机水平（在我们的基线中困惑度为10^2-10^5）。在一个200M代理套件（20B FineWeb-Edu标记，所有方法使用相同的数据流）上，单次TLM运行在其二十个层前缀的每一个上都是有效的语言模型，无论是在困惑度上还是在对困惑度敏感的下游任务上，相比于固定退出套件，减少了质量-预算曲线下的面积43-44%，同时在全容量下与它们相匹配，且每次运行的GPU成本降低约12%。前缀采样密度是一个调节器：将其集中在少数深度上可以在固定退出的质量上恢复，但代价是失去连续性，因此操作点成为训练时的选择，而不是架构选择。这些结果表明，训练目标，而不是嵌套本身，使得模型具有弹性。