cs.RO / 1 / 2610.12549
Fly-by-Code: Embodied Coding Agents for Aerial Manipulation with Active Visual and Physical Feedback
飞行编码：具身编码代理在主动视觉和物理反馈下的空中操作
Abstract
Embodied coding agents construct robot policies from natural-language instructions by generating executable programs and refining them from execution feedback. This refinement assumes that execution feedback reveals whether the task succeeded and, after a failure, provides sufficient evidence to guide program revision. Aerial manipulation violates this assumption in two ways. Payload constraints restrict onboard camera coverage and leave task-critical outcomes outside the robot's view (partial observability), while strong dynamic coupling and limited control margins make physical interactions fragile (limited interaction robustness) and leave failure causes unresolved. Re-examining the same feedback cannot recover observations that were never collected or separate failure causes that the available evidence does not distinguish. To acquire this missing evidence, we introduce Active Feedback (AF), an active "View and Probe" framework. When execution feedback leaves unresolved questions, the agent synthesizes a separate diagnostic program that runs in simulation to inspect unseen outcomes from informative viewpoints (View) or test hypotheses about failure causes through targeted physical interactions (Probe). Across six simulated aerial manipulation tasks with 60 trials per condition, AF reduces false completion declarations from 25 to 5 and raises final task success from 31.7% to 75% compared with terminal feedback under the same coding-agent configuration and task-execution limit. Using shared robot APIs, task programs refined with AF in simulation run on a physical aerial manipulator without task-program source-code modification and complete multi-stage manipulation tasks in the real world. These results show that executable diagnosis fills gaps in execution feedback for embodied coding agents, improving both completion assessment and program refinement in aerial manipulation.
Chinese Translation
具身编码代理通过生成可执行程序并根据执行反馈进行精炼，从自然语言指令中构建机器人策略。这种精炼假设执行反馈能够揭示任务是否成功，并在失败后提供足够的证据以指导程序修订。然而，空中操作在两个方面违反了这一假设。有效载荷限制限制了机载摄像头的覆盖范围，使得任务关键结果超出机器人的视野（部分可观察性），而强动态耦合和有限控制余量使得物理交互变得脆弱（有限交互鲁棒性），并使得失败原因无法解决。重新审视相同的反馈无法恢复从未收集的观察结果或区分可用证据无法区分的失败原因。为了获取这些缺失的证据，我们引入了主动反馈（Active Feedback, AF），一种主动的“查看与探测”（View and Probe）框架。当执行反馈留下未解的问题时，代理合成一个单独的诊断程序，该程序在模拟中运行，以从信息丰富的视角（查看）检查未见结果或通过有针对性的物理交互（探测）测试关于失败原因的假设。在六个模拟空中操作任务中，每个条件下有60次试验，AF将错误完成声明从25减少到5，并将最终任务成功率从31.7%提高到75%，与相同编码代理配置和任务执行限制下的终端反馈相比。通过共享机器人API，使用AF在模拟中精炼的任务程序可以在物理空中操纵器上运行，而无需修改任务程序源代码，并在现实世界中完成多阶段操作任务。这些结果表明，可执行诊断填补了具身编码代理执行反馈中的空白，改善了空中操作中的完成评估和程序精炼。
cs.RO / 2 / 2610.12558
PanTact: Teaching Whole-Body Dexterity with a Wearable Contact Interface
PanTact：通过可穿戴接触接口教授全身灵巧性
Abstract
Humanoid robots have many degrees of freedom and contact surfaces, yet most loco-manipulation research restricts manipulation to the hands and avoids other body contacts with the environment. To extend manipulation beyond the hands, we present PanTact, a system that learns humanoid whole-body dexterity, the capability to manipulate with any suitable body surface, directly from robot-free human demonstrations. Our key idea is to use contact as manipulation-centric information that transfers across embodiments. Contact locations and timing specify the interactions needed to accomplish a task, while the robot chooses how to coordinate its body to realize them within its own physical constraints. Specifically, PanTact captures, learns, and executes whole-body contact via: 1) a wearable, robot-free data collection interface that augments egocentric vision and bimanual UMI grippers with whole-body modular contact patches to record body contact locations and states; 2) a visuomotor policy that mitigates the egocentric observation gap and predicts whole-body contact goals; and 3) contact-aware whole-body control formulations that realize active contact goals using suitable robot surfaces. We deploy PanTact on a Unitree G1 humanoid without additional body contact sensors, demonstrating whole-body dexterity involving body support, pushing, sweeping, and load-sharing. Website: https://pantact.github.io
Chinese Translation
类人机器人具有多个自由度和接触面，但大多数运动操作研究将操作限制在手部，并避免与环境的其他身体接触。为了扩展超越手部的操作能力，我们提出了PanTact，一个学习类人全身灵巧性的系统，该系统能够直接从无机器人的人类示范中学习使用任何适合的身体表面进行操作的能力。我们的关键思想是将接触作为以操作为中心的信息，这些信息可以跨不同的体现进行传递。接触位置和时机指定了完成任务所需的交互，而机器人选择如何协调其身体以在自身的物理限制内实现这些交互。具体而言，PanTact通过以下方式捕捉、学习和执行全身接触：1）一个可穿戴的、无机器人数据收集接口，增强了自我中心视觉和双手UMI夹具，并配备全身模块化接触贴片，以记录身体接触位置和状态；2）一个视觉运动策略，减轻自我中心观察差距并预测全身接触目标；3）接触感知的全身控制公式，利用适当的机器人表面实现主动接触目标。我们在没有额外身体接触传感器的情况下，将PanTact部署在Unitree G1类人机器人上，展示了涉及身体支撑、推、扫和负载共享的全身灵巧性。网站：https://pantact.github.io
cs.RO / 3 / 2610.12561
What Should Embodied Agents Remember? Benchmarking and Evaluating Memory Modality for Embodied Agents
具身智能体应记住什么？具身智能体记忆模式的基准测试与评估
Abstract
Memory allows large language model (LLM) agents to use past experience, often stored as textual records, when responding to requests. For embodied agents that control robots, visual and spatial observations introduce a broader choice of memory modalities, affecting both the information available for subsequent tasks and the model inference costs of constructing and using memory. We introduce MeMoREval, a benchmark for studying this problem through a controlled experience-request protocol. Agents first experience a sequence of events without knowing the subsequent tasks, and subsequently they use memory to respond to information and action requests without receiving new observations. The benchmark comprises 53 tasks across four simulated indoor environments and measures task success alongside the inference costs of memory construction and use. Our reference evaluations reveal that storing frames outperforms storing generated descriptions at lower write-time inference cost for the primary model, while storing frames with scene graphs achieves the highest aggregate task success for every tested model. Extensions to accumulated experience, selective frame retention, and visual input to memory writers demonstrate how the benchmark accommodates further memory-design investigations. Two physical-robot demonstrations establish the feasibility of applying its protocol beyond simulation.
Chinese Translation
记忆使大型语言模型（LLM）智能体能够在响应请求时利用过去的经验，这些经验通常以文本记录的形式存储。对于控制机器人的具身智能体而言，视觉和空间观察引入了更广泛的记忆模式选择，这影响了后续任务可用的信息以及构建和使用记忆的模型推理成本。我们提出了MeMoREval，一个通过受控体验-请求协议研究该问题的基准。智能体首先经历一系列事件，而不知道后续任务，随后它们利用记忆响应信息和行动请求，而不接收新的观察。该基准包括53个任务，涵盖四个模拟室内环境，并测量任务成功率以及记忆构建和使用的推理成本。我们的参考评估表明，对于主要模型，存储帧在较低的写入时间推理成本下优于存储生成的描述，而存储带有场景图的帧在每个测试模型中实现了最高的整体任务成功率。对累积经验、选择性帧保留和视觉输入到记忆写入器的扩展展示了该基准如何适应进一步的记忆设计研究。两个物理机器人演示验证了其协议在模拟之外应用的可行性。
cs.RO / 4 / 2610.12567
CASTER: Formalizing Zero-Shot Manipulation from Synthetic Videos as Constrained Optimization
CASTER：将合成视频中的零-shot 操作形式化为约束优化
Abstract
Prior research on learning from humans often requires real human demonstrations. But advances in video generation models offer a new paradigm: we can now prompt these generators to make synthetic videos of the task. On the plus side, these synthetic videos take place in the robot's current environment, and do not require actual human teaching. However, while the generated videos appear to show plausible motions, they lack both physical consistency and robot grounding, and thus the robot cannot directly roll out the trajectory shown by the video. In this paper we seek to formalize the process of going from synthetic video generation to real-world robot trajectories for zero-shot manipulation tasks. We recognize that this problem is an instance of constrained optimization. The object trajectories extracted from the synthetic video provide an initial guess at the correct motion, and the robot must refine these trajectories (for instance, within a digital twin) while enforcing physical feasibility and semantic consistency. Our optimization-based formalism provides insights into the video generation properties necessary for success. Specifically, we show that the video generation must probabilistically seed the search within a semantic basin of attraction where the trajectory can be locally modified without losing its task alignment. Our analysis further reveals that state-of-the-art approaches can be viewed as instances of our unified formalism with key simplifications. In our experiments we remove these simplifications to more closely match the underlying optimization, leading to more robust zero-shot trajectories from synthetic videos and improving average task success by 34.1% compared to the best-performing baseline. See videos at our project website: https://collab.me.vt.edu/CASTER
Chinese Translation
以往关于从人类学习的研究通常需要真实的人类示范。然而，视频生成模型的进展提供了一种新范式：我们现在可以提示这些生成器制作任务的合成视频。好处在于，这些合成视频发生在机器人的当前环境中，并且不需要实际的人类教学。然而，尽管生成的视频似乎展示了合理的运动，但它们缺乏物理一致性和机器人基础，因此机器人无法直接执行视频中展示的轨迹。在本文中，我们试图将从合成视频生成到现实世界机器人轨迹的过程形式化，以应对零-shot 操作任务。我们认识到这个问题是约束优化的一个实例。从合成视频中提取的物体轨迹提供了对正确运动的初步猜测，机器人必须在强制物理可行性和语义一致性的情况下（例如，在数字双胞胎中）细化这些轨迹。我们的基于优化的形式化方法提供了成功所需的视频生成特性的见解。具体而言，我们表明，视频生成必须以概率方式在一个语义吸引盆地内播种搜索，在这个盆地中，轨迹可以在不失去任务对齐的情况下局部修改。我们的分析进一步揭示，最先进的方法可以被视为我们统一形式化的实例，并进行了关键简化。在我们的实验中，我们去除了这些简化，以更接近底层优化，从合成视频中获得更稳健的零-shot 轨迹，并将平均任务成功率提高了 34.1%，与表现最佳的基线相比。请访问我们项目网站查看视频：https://collab.me.vt.edu/CASTER
cs.RO / 5 / 2610.12609
Graph-Based Heuristics for Minimum Constraint Removal in Cluttered Robot Motion Planning
基于图的启发式方法用于杂乱环境中的最小约束移除机器人运动规划
Abstract
Humanoid and service robots operating in clut- tered human environments may encounter motion planning problems that are infeasible when all surrounding objects are treated as fixed obstacles. The Minimum Constraint Removal problem, which we refer to as MCR, addresses this situation by identifying the minimum number of constraints, such as movable objects, whose removal enables a collision-free path. However, computing an optimal MCR solution is NP-hard. In this work, we propose a graph-based heuristic pipeline that combines bidirectional feasible-set generation with fixed- length searches over reduced MCR graphs. We further derive an extended feasible-set generator for instances in which the candidates produced by the first generator do not contain a complete optimal removal set. Across six benchmarks, the evaluated pipeline matches the exact solution in three cases and, in the remaining cases, selects at most 31.25% more obstacles than the optimum, while being between 3.4 and 33.7 times faster than exact search. In one of the benchmarks, involving a manipulation task amidst clutter, selecting two rather than seven obstructing objects reduces the combined MCR, inverse- kinematics, and motion planning time from 84.72 s to 30.59 s. These results indicate that the proposed approach provides a practical compromise between computational efficiency and solution quality for motion planning among movable obstacles.
Chinese Translation
在杂乱的人类环境中操作的人形机器人和服务机器人可能会遇到运动规划问题，当所有周围物体被视为固定障碍物时，这些问题是不可行的。我们称之为最小约束移除问题（Minimum Constraint Removal，MCR），该问题通过识别最少数量的约束（如可移动物体），使得移除这些约束能够实现无碰撞路径。然而，计算最优的 MCR 解决方案是 NP-困难的。在本研究中，我们提出了一种基于图的启发式管道，该管道结合了双向可行集生成与在简化的 MCR 图上进行固定长度搜索。我们进一步推导出一种扩展的可行集生成器，用于处理第一生成器产生的候选项不包含完整最优移除集的情况。在六个基准测试中，评估的管道在三种情况下与精确解匹配，而在其余情况下，选择的障碍物最多比最优解多 31.25%，同时速度比精确搜索快 3.4 到 33.7 倍。在其中一个基准测试中，涉及在杂乱环境中的操作任务，选择两个而不是七个障碍物将综合 MCR、逆运动学和运动规划时间从 84.72 秒减少到 30.59 秒。这些结果表明，所提出的方法在可移动障碍物的运动规划中提供了计算效率与解决方案质量之间的实用折衷。
cs.RO / 6 / 2610.12631
Image-Based Path-Planning for Navigation of Soft-Growing Robots in the Spinal Sub-Arachnoid Space
基于图像的软生长机器人在脊髓蛛网膜下腔导航的路径规划
Abstract
Recently, soft robotic procedures are being developed aiming at achieving targeted minimally invasive operations or drug administration in specific locations in the human body. Accordingly, specialized path-planning algorithms need to be investigated for the estimation of efficient paths that would protect the delicate tissue structures and simultaneously satisfy a set of requirements for the efficient navigation of the robot. In this context, we propose an image-based 3D pathplanning algorithm, which is an extension of the A* algorithm, for the navigation of a soft-growing robot inside the spinal subarachnoid space (SSS). The proposed algorithm is capable of estimating a safe pathway towards the goal location, while ensuring the establishment of anchor points that facilitate the efficient steering of the soft robot. The algorithm is evaluated using a highly detailed model of the SSS with respect to its capacity to satisfy the robot movement requirements and path efficiency.
Chinese Translation
近年来，软机器人技术正在发展，旨在实现针对人体特定部位的靶向微创手术或药物给药。因此，需要研究专门的路径规划算法，以估计有效路径，保护脆弱的组织结构，并同时满足机器人高效导航的一系列要求。在此背景下，我们提出了一种基于图像的三维路径规划算法，该算法是 A* 算法的扩展，用于在脊髓蛛网膜下腔（SSS）内导航软生长机器人。所提算法能够估计通往目标位置的安全路径，同时确保建立锚点，以促进软机器人高效转向。该算法在一个高度详细的 SSS 模型上进行评估，以验证其满足机器人运动要求和路径效率的能力。
cs.RO / 7 / 2610.12636
Selectively Actuated Fiberbots for High-precision Minimally Invasive Surgery
用于高精度微创手术的选择性驱动光纤机器人
Abstract
Long, slender instruments are essential for minimally invasive surgery (MIS), but achieving precise distal manipulation becomes increasingly difficult when actuation extends along the entire instrument. Here, we present an electrothermally actuated fiber-based robot designed for MIS procedures, in which Joule heating is concentrated in the distal active segment. The fiber is fabricated using a thermal drawing process and uses embedded metallic wires for thermal actuation. This robotic fiber, 1.9 mm in diameter and 1.2 m in length, can precisely manipulate surgical laser fibers with positional precision below 100 $\mu$m. The compact size of the fiber robot allows it to pass through the working channel of a flexible endoscope, while keeping both devices mechanically uncoupled to allow independent motion. A phantom demonstration combining the robotic fiber with a da Vinci instrument conceptually demonstrated its compatibility with existing endoscopic systems for clinical applications. Our robotic fiber enables precise distal motion and holds the potential to improve clinical outcomes by enhancing precision in MIS procedures.
Chinese Translation
细长的仪器对于微创手术（MIS）至关重要，但当驱动沿整个仪器延伸时，实现精确的远端操作变得越来越困难。在此，我们提出了一种电热驱动的基于光纤的机器人，旨在用于微创手术，其中焦耳加热集中在远端的主动段。该光纤采用热拉伸工艺制造，并使用嵌入的金属线进行热驱动。这种直径为1.9毫米、长度为1.2米的机器人光纤能够以低于100微米的定位精度精确操作外科激光光纤。光纤机器人的紧凑尺寸使其能够通过柔性内窥镜的工作通道，同时保持两个设备在机械上解耦，以实现独立运动。结合机器人光纤与达芬奇（da Vinci）仪器的虚拟演示概念上展示了其与现有内窥镜系统在临床应用中的兼容性。我们的机器人光纤能够实现精确的远端运动，并有潜力通过提高微创手术的精确性来改善临床结果。
cs.RO / 8 / 2610.12637
Distributed Harmonic-Covariance Feedback for Hole Closure in Planar Robot Swarms
用于平面机器人群体孔闭合的分布式谐波协方差反馈
Abstract
We consider hole closure in a planar multi-agent system whose sensing disks form a time-varying \v{C}ech complex. We propose a harmonic-covariance feedback that uses harmonic edge signals to locate holes and converts those signals into agent forces. For distributed evaluation, we use a fixed-degree polynomial of the regularized Hodge Laplacian. This polynomial is evaluated through local operator applications. The closed loop has a unique global trajectory from every safe initial state. The model data and initial state are supplied with procedures for approximation to any requested accuracy with explicit error bounds. We assume that hole area is measurable and bounded, holes evolve continuously without splitting or merging, and only finitely many temporary new holes appear. Under these conditions, offline finite calculations produce rational lower bounds for the cumulative hole area on the time horizon without querying trajectory data. These bounds converge to that cumulative area, whether finite or infinite. If it is finite, holes close in the time-averaged sense: the average normalized hole area over every fixed-length time window tends to zero as time tends to infinity. In a 15-second numerical demo, all holes close by 9.9s and remain closed through the end.
Chinese Translation
我们考虑一个平面多智能体系统中的孔闭合问题，该系统的感知圆盘形成一个时变的 Cech 复形。我们提出了一种谐波协方差反馈，利用谐波边缘信号来定位孔，并将这些信号转换为智能体的作用力。为了进行分布式评估，我们使用了正则化的 Hodge 拉普拉斯算子的固定度多项式。该多项式通过局部算子应用进行评估。闭环从每个安全的初始状态出发具有唯一的全局轨迹。模型数据和初始状态提供了近似到任何所需精度的程序，并具有明确的误差界限。我们假设孔的面积是可测量且有界的，孔的演变是连续的，不会发生分裂或合并，并且只会出现有限数量的临时新孔。在这些条件下，离线有限计算可以在时间范围内生成累积孔面积的有理下界，而无需查询轨迹数据。这些下界收敛于该累积面积，无论其是有限还是无限。如果是有限的，孔在时间平均意义上闭合：在每个固定长度时间窗口内的平均归一化孔面积随着时间趋向无穷大而趋向于零。在一个15秒的数值演示中，所有孔在9.9秒时闭合，并在结束时保持闭合状态。
cs.RO / 9 / 2610.12642
PriMoRetarget: Humanoid Motion Transfer Through Prioritized Motion Retargeting
PriMoRetarget：通过优先级运动重定向实现类人运动转移
Abstract
Motion retargeting is a critical preprocessing step for transferring human motion data to humanoid robots. While existing methods are effective for single-character motions, interactions involving multiple characters and objects introduce additional contacts and spatial relationships that must be preserved. Retargeting is commonly formulated as a weighted multi-objective optimization problem, where conflicting objectives can require careful tuning of relative weights and can still compromise important motion characteristics. We introduce PriMoRetarget, a prioritized motion-retargeting framework that replaces relative objective weighting with an explicit priority hierarchy. Lower-priority updates are constrained to preserve higher-priority objectives, allowing critical characteristics such as foot placement and interaction contacts to take precedence over lower-priority pose objectives. The method is evaluated on four datasets spanning single-character motion, character-object manipulation, and character-character interactions. The resulting motions better preserve task-relevant interactions and provide reference trajectories that can be tracked more reliably than those produced by widely used retargeting methods.
Chinese Translation
运动重定向是将人类运动数据转移到类人机器人上的一个关键预处理步骤。尽管现有方法在单角色运动方面有效，但涉及多个角色和物体的交互引入了额外的接触和空间关系，这些必须得到保留。重定向通常被表述为一个加权的多目标优化问题，其中相互冲突的目标可能需要仔细调整相对权重，并且仍可能妨碍重要的运动特征。我们提出了PriMoRetarget，一种优先级运动重定向框架，它用显式的优先级层次结构替代了相对目标加权。低优先级的更新受到限制，以保留高优先级目标，从而使得关键特征（如脚的位置和交互接触）优先于低优先级的姿态目标。该方法在四个数据集上进行了评估，这些数据集涵盖了单角色运动、角色-物体操作和角色-角色交互。结果表明，该方法更好地保留了与任务相关的交互，并提供了比广泛使用的重定向方法生成的轨迹更可靠的参考轨迹。
cs.RO / 10 / 2610.12670
Robust and Scalable UAV-UGV Planning for Aerial Monitoring under Energy and Precedence Constraints
考虑能量和优先顺序约束的无人机-无人地面车辆联合规划用于空中监测
Abstract
We consider planning for unmanned aerial vehicle (UAV) and unmanned ground vehicle (UGV) teams in aerial monitoring missions with ordering requirements on the monitoring points to be visited by the UAV. The system consists of multiple UAV-UGV teams, each comprising one UAV and one UGV. UAV energy limitations are modeled through endurance, defined as the maximum flight time on a full battery. The UGVs serve as mobile charging stations and transport the UAVs between consecutive aerial tours. Given a set of monitoring points subject to ordering constraints and the teams' initial and terminal positions, the objective is to minimize mission completion time while satisfying UAV energy limits and the required visit order under bounded travel-time disturbances that may arise during execution. Unlike prior work, the plan jointly determines vehicle routes and their timing, allowing teams to slow down as needed to satisfy the ordering constraints. We develop a polynomial-time algorithm that generates efficient solutions to the underlying NP-hard planning problem. We establish feasibility guarantees, characterize the algorithm's computational complexity, and evaluate its performance via numerical comparisons.
Chinese Translation
我们考虑在空中监测任务中对无人机（UAV）和无人地面车辆（UGV）团队进行规划，监测点的访问顺序有要求。该系统由多个UAV-UGV团队组成，每个团队包括一架UAV和一辆UGV。UAV的能量限制通过耐力模型进行描述，耐力被定义为在满电状态下的最大飞行时间。UGV作为移动充电站，负责在连续的空中巡航之间运输UAV。给定一组受顺序约束的监测点以及团队的初始和终止位置，目标是最小化任务完成时间，同时满足UAV的能量限制和所需的访问顺序，并应对执行过程中可能出现的有限旅行时间干扰。与之前的研究不同，该计划联合确定车辆的路线和时间安排，允许团队根据需要减速以满足顺序约束。我们开发了一种多项式时间算法，能够为基础的NP难度规划问题生成高效解决方案。我们建立了可行性保证，描述了算法的计算复杂性，并通过数值比较评估其性能。
cs.RO / 11 / 2610.12684
Non-Prehensile Multi-Object Placement with Mobile Robots via Diffusion-Based Trajectory Planning
基于扩散的移动机器人非抓取多物体放置轨迹规划
Abstract
Non-prehensile manipulation with mobile robots has primarily targeted placement in coarse goal regions. Precision placement has been studied with manipulators on benchmarks such as Push-T, where an omnidirectional end effector can readily re-approach an object from any side. For a mobile robot, correcting the same placement errors may require wide repositioning maneuvers around the object. We formalize multi-object pushing with a mobile robot as a multi-target assignment problem at tight tolerances. We propose a diffusion planner with a causal factorization: each robot denoising step is followed by an object step conditioned on the current clean robot estimate. This factorization allows us to sample multiple object outcomes for a single robot plan, capturing the multi-modality of pushing. We also design a set of pushing-specific costs for mobile robots, used for both guidance and best-of-$N$ sampling at inference. On Push-T, our planner attains the highest success rate of the methods we compare, while having the lowest rate of implausible object motion. Across three multi-box pushing environments resembling warehouse scenarios, it raises the fraction of boxes placed by $21.5$ percentage points over an unguided baseline, and replans $1.8\times$ faster than a planner that denoises the two trajectories together. Lastly, we deploy the guided planner on a TurtleBot 3 Burger with object shapes absent from the training data, where it places $85.0\%$ of objects within tolerance against Diffusion Policy's $37.5\%$.
Chinese Translation
移动机器人进行非抓取操作主要针对粗略目标区域的放置。精确放置已在如 Push-T 等基准上通过操纵器进行研究，其中全向末端执行器可以从任何侧面轻松重新接近物体。对于移动机器人，纠正相同的放置错误可能需要围绕物体进行大范围的重新定位操作。我们将移动机器人进行的多物体推送形式化为一个在严格容差下的多目标分配问题。我们提出了一种具有因果分解的扩散规划器：每个机器人去噪步骤后跟随一个基于当前干净机器人估计的物体步骤。这种分解使我们能够为单个机器人计划采样多个物体结果，从而捕捉推送的多模态性。我们还为移动机器人设计了一组特定于推送的成本，用于推理时的指导和最佳 $N$ 采样。在 Push-T 上，我们的规划器在比较的方法中达到了最高的成功率，同时具有最低的不合理物体运动率。在三个类似仓库场景的多箱推送环境中，它将放置的箱子比例提高了 $21.5$ 个百分点，相较于无指导的基线，并且比一个同时去噪两个轨迹的规划器快 $1.8 imes$ 重新规划。最后，我们在缺乏训练数据的物体形状上，将指导规划器部署在 TurtleBot 3 Burger 上，成功将 $85.0 ext{%}$ 的物体放置在容差范围内，而扩散策略的成功率为 $37.5 ext{%}$。
cs.RO / 12 / 2610.12696
Spa-Bench: A Real-Robot Benchmark for Spatial Grounding in Vision-Language-Action Models
Spa-Bench：用于视觉-语言-动作模型空间定位的真实机器人基准测试
Abstract
Vision-language-action models (VLAs) offer a promising route toward general-purpose robotics by transferring semantic knowledge from large-scale vision-language pretraining into language-conditioned control. However, most VLA evaluations emphasize aggregate manipulation success under changes in objects, instructions, or environments, leaving unclear how reliably these policies ground spatial descriptions in the physical scenes where they act. We therefore ask a sharper question: how reliably can current VLAs translate spatial descriptions into appropriate real-robot behavior? To investigate this, we introduce Spa-Bench, a real-robot benchmark that evaluates spatial grounding across six tabletop tasks covering Object State, Relative Size, Referential Description, Relational Placement, Ordinal Position, and Counting. Our primary evaluation compares in-distribution (ID) and out-of-distribution (OOD) spatial configurations while retaining training support for the relevant objects, concepts, and actions, reducing object and motor unfamiliarity as alternative explanations for failure. Four diagnostic conditions further probe physical execution, sensitivity to wording, object-task associations, and recognition that no action is required. We release 4,092 episodes collected on the SO-101 robot arm: 1,200 human-teleoperated demonstrations and 2,892 real-robot evaluation rollouts from five state-of-the-art VLA policies. Across the three fully evaluated policies, success falls from 67.5-73.3% on ID spatial configurations to 40.7-52.0% on OOD spatial configurations, while matched direct instructions improve success by 28.3-40.0 percentage points. These results show that the evaluated VLAs can often execute the required manipulations but lack robust spatial grounding. Project materials, including videos, code, and benchmark data, are available at https://justintiensmith.github.io/Spa-Bench/.
Chinese Translation
视觉-语言-动作模型（VLA）通过将大规模视觉-语言预训练中的语义知识转移到语言条件控制中，为通用机器人提供了一条有前景的路径。然而，大多数VLA评估强调在对象、指令或环境变化下的整体操作成功，这使得这些策略在其所作用的物理场景中如何可靠地实现空间描述变得不清晰。因此，我们提出一个更为尖锐的问题：当前的VLA能够多可靠地将空间描述转化为适当的真实机器人行为？为此，我们引入了Spa-Bench，这是一个真实机器人基准测试，评估六个桌面任务中的空间定位，涵盖对象状态、相对大小、指称描述、关系放置、序数位置和计数。我们的主要评估比较了在保留相关对象、概念和动作的训练支持下的分布内（ID）和分布外（OOD）空间配置，从而减少了对象和运动的不熟悉性作为失败的替代解释。四个诊断条件进一步探讨了物理执行、对措辞的敏感性、对象-任务关联以及识别不需要采取行动的情况。我们发布了在SO-101机器人手臂上收集的4,092个实验：1,200个由人类远程操作的演示和2,892个来自五个最先进VLA策略的真实机器人评估回合。在三个完全评估的策略中，ID空间配置的成功率从67.5%-73.3%下降到OOD空间配置的40.7%-52.0%，而匹配的直接指令则将成功率提高了28.3-40.0个百分点。这些结果表明，被评估的VLA通常能够执行所需的操作，但缺乏稳健的空间定位。项目材料，包括视频、代码和基准数据，可在https://justintiensmith.github.io/Spa-Bench/获得。
cs.RO / 13 / 2610.12706
FallGown: A Soft Pneumatic Wearable Device for Fall Prevention
FallGown：一种用于防止摔倒的软型气动可穿戴设备
Abstract
Concerns about falling and potential resulting injuries prevent many older adults from participating in forms of health-improving exercise such as walking, dance, and tai chi. Existing passive devices like walkers require use of the arms for support, and active devices like exoskeletons and robotic frames are often rigid, bulky, and stigmatized, which limits their use. The FallGown is a soft wearable support structure consisting of pressurized pneumatic pouches within a fabric shell; it can support the full weight of a user's body and aims to prevent falls. It is portable (6.9 kg and 0.051 m3 when stored), minimally burdensome (7.24-20.8 N drag force depending on floor type and whether an optional rolling base frame is used), and convenient (less than 2.5 minutes on average to don and less than 0.5 minutes on average to doff). The optional rolling base lowers friction with the floor, but reduces portability and skirt deformation (which can be useful to fit through doors and other tight spaces). Fall prevention testing with a single user revealed a tradeoff between device size and stability against tipping; a skirt with 12 pneumatic pouches and a floor diameter of 1.4 meters with a base frame prevented all falls except in one case with a mechanical failure. On-board pressure sensors can sense the user's weight distribution, providing an opportunity for monitored exercise and adaptive support.
Chinese Translation
对摔倒及其可能导致的伤害的担忧使许多老年人无法参与改善健康的锻炼形式，如步行、舞蹈和太极。现有的被动设备如助行器需要手臂提供支撑，而主动设备如外骨骼和机器人框架通常刚性、笨重且存在污名化，限制了它们的使用。FallGown是一种软型可穿戴支撑结构，由布料外壳内的加压气动袋组成；它可以支撑用户身体的全部重量，旨在防止摔倒。该设备便携（存储时重6.9千克，体积0.051立方米），负担最小（根据地面类型和是否使用可选的滚动底座框架，拖曳力在7.24-20.8牛顿之间），且方便（平均穿戴时间少于2.5分钟，脱下时间少于0.5分钟）。可选的滚动底座降低了与地面的摩擦，但减少了便携性和裙摆变形（这在通过门和其他狭小空间时可能有用）。对单一用户进行的防摔测试显示设备大小与防倾覆稳定性之间存在权衡；配备12个气动袋和1.4米直径底座框架的裙摆在一次机械故障的情况下防止了所有摔倒。设备上的压力传感器可以感知用户的体重分布，为监测锻炼和自适应支持提供了机会。
cs.RO / 14 / 2610.12710
REMM: Revisable Embodied Memory for Manipulation
REMM：可修订的具身记忆用于操控
Abstract
Long-horizon manipulation requires robots to retain task-relevant history while revising conclusions no longer supported by physical evidence. Retaining observations alone does not specify which completed conditions should be withdrawn after a disturbance or which obligations require repair. We introduce REMM, Revisable Embodied Memory for Manipulation, a supervisory memory layer for a frozen G0.5 vision-language-action executor. REMM separates an append-only evidence history from revisable task state and explicitly links derived relations and completed conditions to their supporting facts. When accepted evidence changes a supporting fact, it withdraws dependent conclusions and schedules targeted recovery at verified skill boundaries without updating the executor. Across 51 physical trials, including operator-assisted runs, REMM improves mean recorded task progress over the strongest baseline from 57% to 91% in intervention recovery and from 43% to 70% in sequence recall. Complementary video diagnostics show improved grounding of retained history into the next instruction, including comparisons with $\pi_{0.5}$-compatible high-level planner adaptations. These results support revisable memory as a practical interface for continued manipulation without policy retraining. Project: https://revanthsenthil.github.io/remm/.
Chinese Translation
长时间的操控任务要求机器人在修正不再由物理证据支持的结论的同时，保留与任务相关的历史记录。仅仅保留观察结果并不能明确在干扰后应撤回哪些已完成的条件或哪些义务需要修复。我们提出了REMM（可修订的具身记忆用于操控），这是一个用于冻结的G0.5视觉-语言-动作执行器的监督记忆层。REMM将仅附加的证据历史与可修订的任务状态分开，并明确将派生关系和已完成条件与其支持事实联系起来。当接受的证据改变支持事实时，它会撤回依赖的结论，并在经过验证的技能边界处安排有针对性的恢复，而无需更新执行器。在51个物理实验中，包括操作员辅助的运行，REMM在干预恢复中的平均记录任务进展从57%提高到91%，在序列回忆中从43%提高到70%。补充的视频诊断显示，保留的历史记录在下一个指令中的基础得到了改善，包括与$ ext{π}_{0.5}$兼容的高级规划器适配的比较。这些结果支持可修订的记忆作为一种实用接口，以便在不重新训练策略的情况下继续进行操控。项目链接：https://revanthsenthil.github.io/remm/
cs.RO / 15 / 2610.12733
Visual-Wi-Fi Navigation to Signal-Emitting Devices
基于视觉的Wi-Fi导航至信号发射设备
Abstract
Indoor navigation to task-relevant targets remains difficult when the target is small, occluded, or visually indistinct. Meanwhile, smart appliances, mobile devices, and IoT tags emit Wi-Fi signals during everyday communication, providing a pervasive but underused spatial cue. We study visual-Wi-Fi navigation, where a robot reaches a signal-emitting device in an unmapped environment using onboard visual observations and Wi-Fi packets overheard from the device, without connecting to the device or modifying infrastructure. Realizing this requires pose-dependent Wi-Fi observations during closed-loop policy learning and reliable spatial cues from bearing measurements corrupted by indoor multipath propagation on a moving robot. We present WiFiSpaces, which extends the Habitat embodied simulator with online ray-traced Wi-Fi sensing, rendering Wi-Fi observations for a robot-mounted antenna array at continuous poses during policy rollouts. On top of it, we build ViFiNav, a localization-aware navigation framework: an online source-localization algorithm exploits robot motion to disambiguate and robustly aggregate bearings from multiple viewpoints into a confidence-aware source estimate, and a learned policy fuses this evolving estimate with depth observations and learns to act on it as it is refined. Experiments show that exploiting robot motion is key to accurate source estimates, and that fine-tuning on ray-traced Wi-Fi observations in WiFiSpaces improves the success rate of ViFiNav. The policy transfers zero-shot to unseen scene datasets, and we deploy it on a real robot with a Wi-Fi sensor built from a commodity router, without any real-world training.
Chinese Translation
在目标较小、被遮挡或视觉上不明显的情况下，室内导航至与任务相关的目标仍然困难。同时，智能家电、移动设备和物联网标签在日常通信中发射Wi-Fi信号，提供了一种普遍但未被充分利用的空间线索。我们研究了基于视觉的Wi-Fi导航，其中机器人在未映射环境中通过机载视觉观测和从设备接收到的Wi-Fi数据包到达信号发射设备，而无需连接设备或修改基础设施。实现这一目标需要在闭环策略学习过程中进行姿态依赖的Wi-Fi观测，并从受室内多径传播影响的移动机器人上获取可靠的空间线索。我们提出了WiFiSpaces，它扩展了Habitat具身模拟器，提供在线光线追踪的Wi-Fi感知，为在策略展开期间处于连续姿态的机器人安装的天线阵列渲染Wi-Fi观测。在此基础上，我们构建了ViFiNav，一个具有定位意识的导航框架：一个在线源定位算法利用机器人运动来消歧并稳健地将来自多个视角的方位角聚合成一个置信度感知的源估计，而一个学习的策略则将这一不断演变的估计与深度观测融合，并学习在其被细化时进行相应的操作。实验表明，利用机器人运动对于准确的源估计至关重要，并且在WiFiSpaces中对光线追踪的Wi-Fi观测进行微调可以提高ViFiNav的成功率。该策略能够零-shot转移到未见过的场景数据集，并且我们在一台由普通路由器构建的Wi-Fi传感器的真实机器人上部署了该策略，而无需任何现实世界的训练。
cs.RO / 16 / 2610.12734
StereoEngine: Scaling Stereo Bimanual Robot Data without Stereo Cameras
StereoEngine：在没有立体摄像头的情况下扩展立体双手机器人数据
Abstract
Stereo vision is increasingly used for robotic manipulation, with a growing body of work incorporating it into Vision-Language-Action (VLA) models. However, the scarcity of large-scale, real-world stereo datasets remains a fundamental bottleneck for scaling such policies. Recent generative monocular-to-stereo methods target settings such as VR content and autonomous driving. No prior work produces realistic synthetic stereo data for robotic manipulation, where stereo supervision has come predominantly from simulation. We propose StereoEngine, which introduces two complementary methods to synthesize stereo data for robotic manipulation. In the first, we warp a monocular (left) video into the right view using estimated depth and inpaint the disoccluded regions with a video diffusion model, lifting monocular demonstrations into stereo pairs without training on real stereo. In the second, we render a stereo pair in simulation and jointly generate both photorealistic views, producing consistent left-right data without any real video as input. StereoEngine outperforms existing baselines in generated stereo quality, and using its synthetic data for policy mid-training or task-specific fine-tuning substantially improves downstream bimanual manipulation. Project website: https://stereoengine.github.io
Chinese Translation
立体视觉在机器人操作中越来越多地被应用，越来越多的研究将其纳入视觉-语言-动作（Vision-Language-Action, VLA）模型。然而，大规模真实世界立体数据集的稀缺仍然是扩展此类策略的根本瓶颈。近期的生成性单目到立体的方法针对虚拟现实内容和自动驾驶等场景。之前的工作没有为机器人操作生成真实的合成立体数据，而立体监督主要来自于仿真。我们提出了StereoEngine，引入了两种互补的方法来合成用于机器人操作的立体数据。在第一种方法中，我们使用估计的深度将单目（左）视频扭曲为右视图，并用视频扩散模型填补被遮挡的区域，将单目演示提升为立体对，而无需在真实立体数据上进行训练。在第二种方法中，我们在仿真中渲染一对立体图像，并共同生成两个照片级真实的视图，生成一致的左右数据，而不需要任何真实视频作为输入。StereoEngine在生成的立体质量上优于现有基线，并且使用其合成数据进行策略中期训练或任务特定微调显著改善了下游双手操作。项目网站：https://stereoengine.github.io
cs.RO / 17 / 2610.12747
Trajectory Prediction for Autonomous Driving in Road Construction Zones Using Human-in-the-Loop Driving Data
基于人机协同驾驶数据的道路施工区域自主驾驶轨迹预测
Abstract
Ego-trajectory prediction in urban roadway work zones remains an underexplored topic in autonomous-driving research, with limited dedicated data and systematic evaluation. Roadway work zones involve temporary changes in the lane layouts and connectivity of the roads, complicating interactions with surrounding vehicles and workers. These changes make future vehicle trajectories difficult to predict. To address these challenges, we collect WZDrive-Data, a multimodal dataset comprising 1,183 work-zone driving segments from 33 drivers, using a human-in-the-loop driving simulator with prerecorded worker animations and VR-controlled workers. Driver control and gaze signals are recorded from human participants, while vehicle trajectories and scene information are obtained from the simulator. Our proposed multimodal ego-trajectory predictor, WZDrive, achieves a mean minADE (K=6) of 1.09 m and minFDE (K=6) of 2.12 m. Compared with the best-performing state-of-the-art baseline for each metric in our benchmark, WZDrive reduces minADE (K=6) by 7.18% and minFDE (K=6) by 13.18%. This research addresses a gap in autonomous-driving research on urban work zones by combining human-in-the-loop data collection with multimodal learning from human driving signals and in the context of work zones.
Chinese Translation
在城市道路施工区域中，自我轨迹预测仍然是自主驾驶研究中一个未被充分探索的主题，相关专门数据和系统评估有限。道路施工区域涉及车道布局和道路连接的临时变化，复杂化了与周围车辆和工人的互动。这些变化使得未来车辆轨迹的预测变得困难。为了解决这些挑战，我们收集了WZDrive-Data，这是一个多模态数据集，包含来自33名驾驶员的1,183段施工区域驾驶数据，使用带有预录工人动画和虚拟现实控制工人的人机协同驾驶模拟器。我们记录了人类参与者的驾驶控制和注视信号，同时从模拟器获取车辆轨迹和场景信息。我们提出的多模态自我轨迹预测模型WZDrive在平均最小ADE (K=6)上达到1.09米，最小FDE (K=6)为2.12米。与我们基准测试中每个指标的最佳表现的最先进基线相比，WZDrive在最小ADE (K=6)上减少了7.18%，在最小FDE (K=6)上减少了13.18%。本研究通过结合人机协同数据收集与来自人类驾驶信号的多模态学习，填补了城市施工区域自主驾驶研究中的空白。
cs.RO / 18 / 2610.12761
Verifier-Mediated Multi-LLM Coordination for Multi-Robot Path Planning
验证者介导的多LLM协调用于多机器人路径规划
Abstract
Foundation models are increasingly used for high-level multi-robot task planning, yet independently generated symbolic plans can become jointly infeasible after geometric grounding because of shared spatial and temporal constraints. We study execution-aware coordination of large-language-model (LLM) planners for multi-robot navigation. We compare a centralized architecture, in which one LLM jointly reasons over the robot team, with a decentralized multi-LLM architecture in which each robot independently proposes a task plan. A geometric verifier grounds the candidate plans into robot trajectories, detects predicted inter-robot conflicts, and converts conflict information into structured feedback for targeted replanning. This verifier-mediated architecture separates semantic task reasoning from execution-level feasibility while enabling iterative coordination among LLM planners. Experiments in CARLA and on Crazyflie aerial robots evaluate coordination performance as task complexity increases and characterize the trade-offs between centralized reasoning, decentralized verifier-mediated coordination, and physical executability.
Chinese Translation
基础模型越来越多地用于高层次的多机器人任务规划，但独立生成的符号计划在几何基础化后可能由于共享的空间和时间约束而变得共同不可行。我们研究了大型语言模型（LLM）规划器在多机器人导航中的执行感知协调。我们比较了一种集中式架构，其中一个LLM对机器人团队进行联合推理，以及一种去中心化的多LLM架构，其中每个机器人独立提出任务计划。几何验证器将候选计划基础化为机器人轨迹，检测预测的机器人间冲突，并将冲突信息转换为结构化反馈以便于有针对性的重新规划。这种验证者介导的架构将语义任务推理与执行层面的可行性分离，同时实现了LLM规划器之间的迭代协调。在CARLA和Crazyflie空中机器人上的实验评估了随着任务复杂性增加的协调性能，并描述了集中推理、去中心化验证者介导协调与物理可执行性之间的权衡。
cs.RO / 19 / 2610.12772
Reflex: Visually Grounded Reactive Humanoid Control
Reflex：视觉基础的反应式类人控制
Abstract
Humans effortlessly react to the world; they adeptly return a tennis shot, catch a falling cup, step aside from a passing cart, or pass a box tossed during truck unloading. These everyday actions require anticipating a moving target and reflexively responding with the whole body within about a second. Such visual reactive whole-body control remains challenging for humanoids, which must (1) coordinate whole-body control, (2) react to object dynamics in real time, and (3) ground perception in egocentric vision. We study catching thrown boxes as a representative task that demands all three capabilities; existing systems cover only part of it, relying on end-effectors, motion capture, or marked objects. We present Reflex, a three-stage framework that learns these capabilities stage by stage while connecting them through a compact representation of object dynamics. First, a state-based policy learns whole-body catching with reinforcement learning. Second, the policy learns to infer the dynamics of a moving box from delayed and incomplete observations. Third, an egocentric RGB-D encoder learns to recover this dynamics representation directly from visual observations while keeping the whole-body controller fixed. This decomposition provides each capability with a direct training signal and makes visual learning substantially more scalable. In simulation, Reflex achieves 85.7% catch success from egocentric RGB-D, compared with 87.9% when given privileged box observations. Temporal visual history is critical: removing it reduces success by 35.6 and 39.8 points on centered and laterally offset throws. On a real Unitree G1, Reflex catches human-thrown boxes with 65% and 40% success at 2.5 and 5 m, reaching 76% and 59% of human catchers' success at the same distances.
Chinese Translation
人类能够轻松地对世界做出反应；他们熟练地回击网球、接住掉落的杯子、避让经过的推车，或在卡车卸货时接住被扔出的箱子。这些日常动作需要在大约一秒内预测移动目标并全身反应。然而，这种视觉反应的全身控制对于类人机器人仍然具有挑战性，因为它们必须（1）协调全身控制，（2）实时响应物体动态，以及（3）在自我中心视觉中进行感知。我们研究接住被扔出的箱子作为一个代表性任务，这一任务要求具备上述三种能力；现有系统仅覆盖其中一部分，依赖于末端执行器、运动捕捉或标记物体。我们提出了Reflex，一个三阶段框架，逐步学习这些能力，并通过对物体动态的紧凑表示将它们连接起来。首先，基于状态的策略通过强化学习学习全身接球。其次，该策略学习从延迟和不完整的观察中推断移动箱子的动态。第三，自我中心的RGB-D编码器学习直接从视觉观察中恢复这一动态表示，同时保持全身控制器不变。这种分解为每种能力提供了直接的训练信号，并使视觉学习的可扩展性大大增强。在仿真中，Reflex在自我中心RGB-D下实现了85.7%的接球成功率，而在提供特权箱子观察时成功率为87.9%。时间视觉历史至关重要：去除它会使中心和侧偏投掷的成功率分别降低35.6和39.8个百分点。在真实的Unitree G1上，Reflex以65%和40%的成功率接住人类投掷的箱子，距离为2.5米和5米，分别达到人类接球者在相同距离下成功率的76%和59%。
cs.RO / 20 / 2610.12780
Multimodal Image-based Robotic Assistance and Guidance Environment (MIRAGE): A Novel Unified Framework for Surgical Navigation and Visualization
基于多模态图像的机器人辅助与引导环境（MIRAGE）：一种用于外科导航与可视化的新统一框架
Abstract
This paper presents the computational architecture and quantitative preclinical evaluation of MIRAGE, a multimodal image-guided robotic surgery framework for real-time visualization and navigation. It integrates stereovision scene reconstruction, augmented-reality (AR) video overlay, quantified heads-up display (HUD) navigation, and intraoperative ultrasound within the da Vinci console's TilePro display (Intuitive Surgical). Additional modules include optical/electromagnetic tracker registration, stereo calibration, fiducial rigid registration with RMSE feedback, and 3D point-cloud reconstruction. Video I/O and rendering run in a C++ backend, while a Python GUI coordinates calibration, registration, reconstruction, visualization, and user interaction through an in-process zero-copy binding layer. The framework formalizes spatial transforms, module states, and synchronization of heterogeneous video and tracking streams. Preoperative calibration uses synchronized checkerboard image pairs and optimization, with estimated intrinsics, extrinsics, and rectification reused intraoperatively. MIRAGE was validated in phantom experiments across da Vinci S, Si, Xi, and SP platforms. Across clinically relevant working distances, mean 3D reprojection error was 0.6mm, with best 0.1mm RMSE at 100mm on Xi. Results support MIRAGE as a preclinical technical-validation platform for accurate, responsive, and extensible multimodal surgical guidance.
Chinese Translation
本文介绍了MIRAGE的计算架构和定量的临床前评估，MIRAGE是一个用于实时可视化和导航的多模态图像引导机器人手术框架。它集成了立体视觉场景重建、增强现实（AR）视频叠加、量化的抬头显示（HUD）导航以及在达芬奇控制台的TilePro显示器内的术中超声。其他模块包括光学/电磁追踪器注册、立体标定、带有均方根误差（RMSE）反馈的标志物刚性注册，以及三维点云重建。视频输入/输出和渲染在C++后端运行，而Python图形用户界面（GUI）通过进程内零拷贝绑定层协调标定、注册、重建、可视化和用户交互。该框架形式化了空间变换、模块状态以及异构视频和追踪流的同步。术前标定使用同步的棋盘图像对和优化，估计的内参、外参和校正在术中重复使用。MIRAGE在达芬奇S、Si、Xi和SP平台的幻影实验中得到了验证。在临床相关的工作距离下，平均三维重投影误差为0.6毫米，在Xi平台上最佳为100毫米时的均方根误差为0.1毫米。结果支持MIRAGE作为一个用于准确、响应迅速和可扩展的多模态外科引导的临床前技术验证平台。
cs.RO / 21 / 2610.12790
Tactile Sensing-based Secure and Gentle Grasping of Unknown Objects
基于触觉感知的未知物体安全且温和的抓取
Abstract
Reliable robotic manipulation requires maintaining a firm grasp while strictly avoiding both contact slip and structural damage to the object. Satisfying these inherent \textit{secure-and-gentle} grasping constraints is challenging for unknown objects. We present a novel online data-driven predictive control framework for safe dynamic robotic grasping of unknown objects. The proposed system features a pair of compact, single-channel tactile-sensing modules that characterize the gripper-object dynamic interaction in real time, entirely bypassing the need for prior knowledge or offline training. A finite-state-machine-based grasping force probing mechanism is developed to establish the initial grasping force and explicitly identify grasping force constraints. Through online dynamics learning, a Gaussian Process (GP) model captures gripper-object interaction dynamics. The model uncertainties are further used to form robust grasping constraints, which are embedded into a model predictive control scheme. We demonstrate error convergence in a probabilistic sense. The proposed control framework runs at 100~Hz in real time to dynamically manipulate and handle objects rather than static holding. Extensive experimental evaluations across diverse contact surface topologies and various types of objects validate superior performance and efficiency compared to baseline strategies. The proposed method is able to manipulate objects ranging from delicate items, such as raw tofu and marshmallows, to rough objects, such as rocks.
Chinese Translation
可靠的机器人操作需要在保持稳固抓握的同时，严格避免接触滑移和对物体的结构损害。满足这些固有的“安全且温和”抓取约束对于未知物体来说是具有挑战性的。我们提出了一种新颖的在线数据驱动预测控制框架，用于安全动态地抓取未知物体。该系统具有一对紧凑的单通道触觉感知模块，能够实时表征夹持器与物体之间的动态交互，完全绕过了对先前知识或离线训练的需求。我们开发了一种基于有限状态机的抓取力探测机制，以建立初始抓取力并明确识别抓取力约束。通过在线动态学习，采用高斯过程（Gaussian Process, GP）模型捕捉夹持器与物体之间的交互动态。模型的不确定性进一步用于形成稳健的抓取约束，这些约束嵌入到模型预测控制方案中。我们展示了在概率意义上的误差收敛。所提出的控制框架以100 Hz的实时频率运行，以动态操控和处理物体，而非静态保持。针对多种接触表面拓扑和不同类型物体的广泛实验评估验证了其相较于基线策略的优越性能和效率。该方法能够操控从精致物品（如生豆腐和棉花糖）到粗糙物体（如石头）等多种物体。
cs.RO / 22 / 2610.12793
Gamma: Grounded Multi-Agent Memory with Mechanical Audit for Memory-Intensive Manipulation
Gamma：具有机械审计的基础多智能体记忆用于内存密集型操作
Abstract
Pretrained vision-language-action (VLA) policies fail on memory tasks, behaviour that depends on events observed long before the current frame, because their training data is overwhelmingly Markov and mapping episodic memory directly to action chunks is hard. Yet the same policy, finetuned once to follow a privileged oracle's grounded-subgoal text, nearly solves these tasks: the difficulty lies entirely in mapping memory to a grounded symbolic subgoal. Existing work entrusts this stage to a single VLM that selects keyframes and reasons over raw frames, and falls far short of the oracle: key identities and dynamics are buried in noisy pixels, the reasoning spans hundreds of frames in an unbounded context, and hallucination corrupts the memory itself. We instead present GAMMA (Grounded multi-Agent Memory with Mechanical Audit), which builds the stage as a collaborating two-agent memory system: a detector tool (SAM-3) for spatial grounding, a VLM writer that distils each replan window into one grounded text line, an append-only text memory bank, and a VLM reasoner that reads the bank and emits the next grounded subgoal at the oracle's cadence, wrapped in a propose-verify harness that admits, defers, corrects, or rejects every agent claim against mechanical visual evidence, at training-corpus construction and at deployment alike. On a sixteen-task memory-manipulation suite GAMMA recovers a large fraction of the oracle ceiling with the policy held fixed, reaching the ceiling on several tasks, and transfers with unchanged contracts to a second benchmark.
Chinese Translation
预训练的视觉-语言-动作（VLA）策略在内存任务上表现不佳，这些任务依赖于在当前帧之前观察到的事件，因为它们的训练数据主要是马尔可夫的，直接将情节记忆映射到动作块是困难的。然而，同样的策略在经过一次微调以遵循特权神谕的基础子目标文本后，几乎能够解决这些任务：困难完全在于将记忆映射到一个基础的符号子目标。现有的工作将这一阶段委托给一个单一的视觉-语言模型（VLM），该模型选择关键帧并对原始帧进行推理，但远未达到神谕的水平：关键身份和动态被埋没在嘈杂的像素中，推理跨越数百帧在一个无限制的上下文中进行，而幻觉则会破坏记忆本身。我们提出了GAMMA（具有机械审计的基础多智能体记忆），将该阶段构建为一个协作的双智能体记忆系统：一个用于空间基础的检测工具（SAM-3），一个将每个重新规划窗口提炼为一个基础文本行的VLM写入器，一个仅附加的文本记忆库，以及一个读取该库并以神谕的节奏发出下一个基础子目标的VLM推理器，包裹在一个提议-验证的框架中，该框架在训练语料库构建和部署时，允许、延迟、纠正或拒绝每个智能体对机械视觉证据的主张。在一个包含十六个任务的内存操作套件中，GAMMA在保持策略不变的情况下恢复了神谕上限的很大一部分，在多个任务上达到了上限，并以不变的合同转移到第二个基准测试上。
cs.RO / 23 / 2610.12818
LeWAM: End-to-End World and Action Modeling with JEPAs
LeWAM：基于 JEPAs 的端到端世界与动作建模
Abstract
Joint-embedding predictive architectures (JEPAs) learn compact latent world models without pixel reconstruction. Existing JEPA pipelines typically train visual representations and action-conditioned dynamics before adding a control policy or solver. We introduce LeWAM, an end-to-end JEPA for control with: (1) a joint training objective over next latent state prediction and action flow matching, and (2) an isotropic Gaussian constraint on the latent state, enabling a single model to support direct policy execution and efficient planning by evaluating sampled actions with its learned latent dynamics. Compared with prior JEPAs in visual planning, LeWAM improves success rate from 28.6% to 89.7% on contact-rich tasks and from 10.9% to 46.2% on long-horizon tasks, and plans 32.3 times faster. On behavior cloning tasks in simulation and physical robots, LeWAM is also competitive with established methods. Interpretability analyses show that LeWAM learns control-relevant representations and action distributions effectively guiding test-time planning and control.
Chinese Translation
联合嵌入预测架构（JEPAs）在不进行像素重构的情况下学习紧凑的潜在世界模型。现有的 JEPA 流程通常在添加控制策略或求解器之前，先训练视觉表示和基于动作的动态。我们提出了 LeWAM，一种用于控制的端到端 JEPA，具有：（1）对下一个潜在状态预测和动作流匹配的联合训练目标，以及（2）对潜在状态的各向同性高斯约束，使得单一模型能够通过评估其学习到的潜在动态的采样动作来支持直接策略执行和高效规划。与之前的视觉规划中的 JEPA 相比，LeWAM 在接触丰富任务上的成功率从 28.6% 提高到 89.7%，在长时间任务上的成功率从 10.9% 提高到 46.2%，并且规划速度提高了 32.3 倍。在模拟和物理机器人中的行为克隆任务中，LeWAM 也与现有方法具有竞争力。可解释性分析表明，LeWAM 有效地学习了与控制相关的表示和动作分布，从而有效指导测试时的规划和控制。
cs.RO / 24 / 2610.12826
Cubic Spline Interpolation for Smooth Trajectory Generation in Autonomous Motion Systems
用于自主运动系统平滑轨迹生成的三次样条插值
Abstract
We presented a program that automates the construction and solution of the cubic spline interpolation problem for a sequence of position-time waypoints, producing trajectories that are continuous in position, velocity, and acceleration and therefore avoid the infinite-acceleration problem inherent to piecewise-linear waypoint following. By assembling the interpolation and continuity conditions into a single linear system and solving directly for the spline coefficients, the program removes the need to manually derive per-segment equations for each new waypoint set. We illustrated the approach on free-form curve drawing and signature digitization, two tasks that use the same underlying machinery required for smooth robot and drone trajectory generation. Future work includes extending the implementation to multi-axis trajectories with explicit velocity and acceleration bounds, adopting the more computationally efficient tridiagonal solution method for larger waypoint sets, and evaluating the approach quantitatively against alternative trajectory generation methods.
Chinese Translation
我们提出了一种程序，该程序自动构建和求解三次样条插值问题，针对一系列位置-时间的航点，生成在位置、速度和加速度上连续的轨迹，从而避免了分段线性航点跟踪中固有的无限加速度问题。通过将插值和连续性条件组合成一个线性系统，并直接求解样条系数，该程序消除了为每组新的航点手动推导每个段方程的需要。我们在自由曲线绘制和签名数字化这两个任务上展示了该方法，这两个任务使用了平滑机器人和无人机轨迹生成所需的相同基础机制。未来的工作包括将实现扩展到具有显式速度和加速度限制的多轴轨迹，采用更具计算效率的三对角解法处理更大的航点集，并对该方法与其他轨迹生成方法进行定量评估。
cs.RO / 25 / 2610.12832
VLMs are Native Action Models: Unleashing the Manipulation Capabilities of VLMs with Explicit Action-Space Grounding
VLMs是原生动作模型：通过显式动作空间基础释放VLM的操控能力
Abstract
Pretrained vision-language models (VLMs) already exhibit strong visual, spatial and reasoning capabilities relevant to robotic manipulation. The key challenge in turning pretrained VLMs into generalist manipulation policies is to expose these capabilities through an appropriate action interface. Vision-language-action (VLA) models provide a unified trainable policy, but typically introduce learned action tokens, action heads, or action experts whose semantics must be learned from robot data pretraining. In contrast, modular foundation-model-based manipulation methods can exploit pretrained VLM capabilities for zero-shot manipulation, but rely on external planning and control scaffolds that are not end-to-end trainable as a unified continuous-action policy. In this work, we introduce EASG, an Explicit Action-Space Grounding framework that bridges these two paradigms. By projecting a tool center point (TCP)-centered action frame into the wrist camera observation and expressing actions in the form of VLM's native textual generation interface, EASG makes the semantics of a continuous robot action space directly observable in the VLM's perceptual context. This allows pretrained VLMs to directly generate executable robot actions using their existing visual-spatial reasoning capabilities, without robot-action pretraining or additional action modules. Besides, the same action interface supports direct fine-tuning with robot trajectories, allowing the models to continuously improve beyond their manipulation capabilities unleashed by EASG. Overall, EASG enables zero-shot manipulation and generalization across object positions, categories, and robot embodiments, while supporting effective fine-tuning in the same continuous action space. These results demonstrate that, given an appropriate action interface, pretrained VLMs can themselves serve as native action models for robotic manipulation.
Chinese Translation
预训练的视觉-语言模型（VLMs）已经展现出与机器人操控相关的强大视觉、空间和推理能力。将预训练的VLM转化为通用操控策略的关键挑战在于通过适当的动作接口来揭示这些能力。视觉-语言-动作（VLA）模型提供了一个统一的可训练策略，但通常引入了学习的动作标记、动作头或动作专家，其语义必须通过机器人数据预训练来学习。相比之下，基于模块化基础模型的操控方法可以利用预训练的VLM能力进行零-shot操控，但依赖于外部规划和控制框架，这些框架不能作为统一的连续动作策略进行端到端训练。在本研究中，我们引入了EASG（显式动作空间基础）框架，连接这两种范式。通过将以工具中心点（TCP）为中心的动作框架投影到手腕摄像头观察中，并以VLM的原生文本生成接口的形式表达动作，EASG使连续机器人动作空间的语义在VLM的感知上下文中直接可观察。这使得预训练的VLM能够直接生成可执行的机器人动作，利用其现有的视觉-空间推理能力，而无需机器人动作的预训练或额外的动作模块。此外，相同的动作接口支持与机器人轨迹的直接微调，使模型能够在EASG释放的操控能力之外持续改进。总体而言，EASG实现了零-shot操控，并在物体位置、类别和机器人体现之间实现了泛化，同时支持在相同的连续动作空间中进行有效的微调。这些结果表明，给定适当的动作接口，预训练的VLM可以自身作为机器人操控的原生动作模型。
cs.RO / 26 / 2610.12861
Evidence-Guided Counterfactual Learning for Long-Tailed Robotic Policies
基于证据的反事实学习用于长尾机器人策略
Abstract
In the practical deployment of vision--language--action policies, handling long-tailed task distributions is a common challenge due to the highly imbalanced composition of robot demonstration datasets. To investigate this issue, we study the head-task action bias in shared robot policies, particularly focusing on its relationship with task-specific visual evidence. Our experimental analysis shows that, when the visual evidence required by a tail task is insufficient, the policy increasingly falls back on action patterns reinforced by data-rich head tasks, resulting in degraded performance on under-represented tasks. Motivated by this observation, we propose Evidence-Guided Counterfactual Learning, a training framework for long-tailed robot policies. Our method identifies decision-relevant visual evidence using instruction-related and action-related signals, and constructs counterfactual observations by removing the selected evidence. The resulting counterfactual predictions are used to strengthen the association among task instructions, visual evidence, and actions, with the supervision adaptively adjusted according to the number of demonstrations available for each task. The proposed framework requires neither additional training data nor region-level annotations and incurs no additional computation during deployment. Extensive experiments across simulated benchmarks, different VLA architectures, and real-world robot platforms demonstrate that our method robustly and consistently improves long-tailed robot policy learning, particularly for tail tasks, while maintaining reliable performance on head tasks.
Chinese Translation
在视觉-语言-动作策略的实际部署中，处理长尾任务分布是一个常见挑战，因为机器人演示数据集的组成高度不平衡。为了研究这个问题，我们分析了共享机器人策略中的主任务动作偏差，特别关注其与任务特定视觉证据之间的关系。我们的实验分析表明，当尾任务所需的视觉证据不足时，策略越来越依赖于由数据丰富的主任务强化的动作模式，导致在代表性不足的任务上性能下降。基于这一观察，我们提出了基于证据的反事实学习（Evidence-Guided Counterfactual Learning），这是一个用于长尾机器人策略的训练框架。我们的方法通过与指令相关和动作相关的信号识别决策相关的视觉证据，并通过移除所选证据构建反事实观察。生成的反事实预测用于增强任务指令、视觉证据和动作之间的关联，同时根据每个任务可用的演示数量自适应调整监督。该框架不需要额外的训练数据或区域级注释，并且在部署过程中不会增加额外的计算负担。在模拟基准、不同的视觉-语言-动作（VLA）架构和真实世界机器人平台上的大量实验表明，我们的方法在长尾机器人策略学习中稳健且一致地提高了性能，特别是在尾任务上，同时在主任务上保持可靠的表现。
cs.RO / 27 / 2610.12877
TAME-Nav: Temporal Asymmetric Mixture-of-Experts Distillation for Agile and Versatile Robot Navigation
TAME-Nav：用于灵活多样机器人导航的时间非对称专家混合蒸馏
Abstract
Agile and reliable navigation is a core capability for embodied robots navigating complex spatial constraints and dynamic interactions. However, a single policy often struggles to accommodate the distinct decision-making demands of diverse scenarios. We present TAME-Nav, an asymmetric Teacher-Student navigation framework built around a temporal soft mixture of experts (Soft-MoE). The Teacher combines clean observations with causally maintained dynamic-obstacle tracks to construct privileged representations, while the Student learns complementary navigation features from corrupted observation histories through temporal encoding, soft expert routing, and latent distillation. Procedural scenario generation and parallel task-level simulation broaden training coverage, while deployment-informed asymmetric domain randomization supports sim-to-real transfer. Across in-distribution and unseen compositional scenarios, TAME-Nav achieves higher success rates, lower collision rates, and higher navigation scores than representative baselines. On a real quadruped robot, it completes all twenty trials across local-trap escape and narrow-passage navigation with sudden dynamic obstacles, with only one minor contact.
Chinese Translation
灵活且可靠的导航是具身机器人在复杂空间约束和动态交互中导航的核心能力。然而，单一策略往往难以满足多样场景的不同决策需求。我们提出了TAME-Nav，一种围绕时间软混合专家（Soft-MoE）构建的非对称教师-学生导航框架。教师结合干净的观测与因果维护的动态障碍轨迹来构建特权表示，而学生则通过时间编码、软专家路由和潜在蒸馏，从损坏的观测历史中学习补充的导航特征。程序化场景生成和并行任务级仿真扩展了训练覆盖范围，而基于部署的非对称领域随机化支持了从仿真到现实的转移。在分布内和未见的组合场景中，TAME-Nav实现了比代表性基线更高的成功率、更低的碰撞率和更高的导航评分。在一台真实的四足机器人上，它在局部陷阱逃脱和窄通道导航中完成了所有二十次试验，且仅发生了一次轻微接触。
cs.RO / 28 / 2610.12886
LeapBot-JAC: Joint Action Chains for World Action Models
LeapBot-JAC：用于世界行动模型的联合行动链
Abstract
Action-chunking policies predict future controls but execute only a prefix before replanning. How should a fixed action-token budget be organized to support execution? We propose LeapBot-JAC, which allocates sparse future-control references and dense execution targets within one shared action expert. Both chains are supervised by demonstrated controls and jointly denoised at aligned physical-time positions. Directed attention allows execution tokens to read reference tokens subject to the host's visibility constraints. JAC adds neither action tokens nor parameters and retains the host's world-modeling process; both chains participate in inference, while only the execution chain drives the robot. On LIBERO, JAC reaches 98.3% success with FastWAM and LeapBot-WA, up from 97.0% and 97.3%. On zero-shot LIBERO-Plus, their success rates improve from 53.5% to 70.4% and from 73.1% to 77.3%, respectively. RoboTwin and two real-world tasks provide additional deployment evaluations. Single-training-seed comparisons and FastWAM ablations support action-role organization as a practical design choice under a fixed action-token budget.
Chinese Translation
行动分块策略预测未来控制，但仅在重新规划之前执行前缀。如何组织固定的行动令牌预算以支持执行？我们提出了LeapBot-JAC，它在一个共享的行动专家内分配稀疏的未来控制参考和密集的执行目标。这两个链条都由示范控制进行监督，并在对齐的物理时间位置共同去噪。定向注意力允许执行令牌在主机的可见性约束下读取参考令牌。JAC既不增加行动令牌也不增加参数，并保留主机的世界建模过程；两个链条都参与推理，但只有执行链驱动机器人。在LIBERO上，JAC与FastWAM和LeapBot-WA的成功率达到98.3%，相比之下，之前为97.0%和97.3%。在零样本LIBERO-Plus上，它们的成功率分别从53.5%提高到70.4%，从73.1%提高到77.3%。RoboTwin和两个真实世界任务提供了额外的部署评估。单次训练种子比较和FastWAM消融实验支持在固定行动令牌预算下将行动角色组织作为一种实用的设计选择。
cs.RO / 29 / 2610.12910
Sampling-based Kinodynamic Planning with World Models
基于采样的动力学运动规划与世界模型
Abstract
Visual world models predict the consequences of actions from images, but planning with them is typically done by optimizers such as CEM and MPPI. However, these optimizers stall on long-horizon tasks where progress requires motion the objective does not reward, such as moving around a block before pushing it. This paper revisits kinodynamic sampling-based motion planning for world models, using the learned model as the propagator of a tree search in latent space. Our key insight is that the sampling strategies devised for high-dimensional search in configuration space apply directly to the latent space of a world model, where they rapidly expand through intermediate states that myopic optimizers never sample. We present two exploration strategies for the high-dimensional latent space, nearest-neighbor-guided RRT and projection-based KPIECE, with GPU-native batched expansion for efficient device utilization. We evaluate with an off-the-shelf DINO-WM on four simulated domains and a real-world block-pushing task. Our approaches match or exceed the success rate of CEM and MPPI at up to 13.8x lower planning time in simulation, and improve the success rate by 20% on the real robot.
Chinese Translation
视觉世界模型能够从图像中预测动作的后果，但通常使用诸如CEM（交叉熵法）和MPPI（模型预测路径积分）等优化器进行规划。然而，这些优化器在长时间尺度任务中往往陷入停滞，因为任务进展需要执行目标函数未奖励的动作，例如在推动物体前绕行障碍物。本文重新审视了基于动力学采样的运动规划方法，将学习到的世界模型作为潜在空间中树搜索的传播器。我们的核心观点是，为配置空间中的高维搜索设计的采样策略可以直接应用于世界模型的潜在空间，在该空间中，这些策略能够迅速通过短视优化器从未采样的中间状态进行扩展。我们提出了两种针对高维潜在空间的探索策略：基于最近邻引导的RRT（快速随机树）和基于投影的KPIECE（关键状态采样），并采用GPU原生的批量扩展以实现高效的设备利用率。我们在四个模拟环境和一个真实的推块任务中，使用现成的DINO-WM模型进行了评估。实验结果表明，我们的方法在模拟中以最高13.8倍更低的规划时间匹配或超过了CEM和MPPI的成功率，并在真实机器人上成功率提升了20%。
cs.RO / 30 / 2610.12936
Evo-World: A Lightweight Vision-Language-Action Model with Future Motion Prediction via Optical-Flow Supervision
Evo-World：一种通过光流监督实现未来运动预测的轻量级视觉-语言-动作模型
Abstract
Vision-language-action (VLA) models provide a unified framework for visual understanding, language grounding, and robot control, yet anticipating how the scene will evolve beyond the current observation can provide additional information for control. Action supervision alone does not explicitly encourage such anticipation, while future-image prediction entangles task-relevant motion with largely static appearance. To address these limitations, we introduce Evo-World, a lightweight VLA model that augments policy learning with future motion prediction through optical-flow supervision, thereby providing a task-relevant predictive signal focused on motion rather than appearance. We further design the training procedure to preserve these predictive representations during policy learning, regulating representation diversity while decoupling representation learning from action optimization. Across simulation benchmarks, Evo-World demonstrates strong performance, achieving a tier-averaged success rate of 91.1% on Meta-World MT50 , with a 13.0% relative improvement over the non-predictive baseline. On real-world AgileX robots, Evo-World outperforms the evaluated baselines across single-arm and bimanual manipulation tasks while using only 0.87B parameters and delivering lower GPU memory usage and higher inference efficiency. These results demonstrate that future motion prediction through optical-flow supervision provides an effective and lightweight approach to enhancing the predictive capabilities of VLA models.
Chinese Translation
视觉-语言-动作（VLA）模型为视觉理解、语言基础和机器人控制提供了一个统一的框架，但预测场景在当前观察之外的演变可以为控制提供额外信息。仅依赖动作监督并不能明确鼓励这种预测，而未来图像预测则将与任务相关的运动与基本静态的外观纠缠在一起。为了解决这些局限性，我们提出了Evo-World，这是一种轻量级的VLA模型，通过光流监督增强策略学习与未来运动预测，从而提供一个关注运动而非外观的任务相关预测信号。我们进一步设计了训练过程，以在策略学习过程中保留这些预测表示，调节表示的多样性，同时将表示学习与动作优化解耦。在模拟基准测试中，Evo-World表现出强大的性能，在Meta-World MT50上实现了91.1%的平均成功率，相较于非预测基线提高了13.0%。在真实世界的AgileX机器人上，Evo-World在单臂和双臂操作任务中超越了评估的基线，同时仅使用0.87亿参数，提供了更低的GPU内存使用和更高的推理效率。这些结果表明，通过光流监督实现的未来运动预测为增强VLA模型的预测能力提供了一种有效且轻量级的方法。
cs.RO / 31 / 2610.12957
JoyInside-VLM-Agent: Product-Conditioned Multimodal Policy Learning for Embodied Agents
JoyInside-VLM-Agent：基于产品条件的多模态策略学习用于具身智能体
Abstract
Embodied agents must translate multimodal observations and user requests into executable actions while adapting to product constraints, changing goals, and execution feedback. We present JoyInside-VLM-Agent, a ReAct-style multimodal agent developed through product-conditioned trajectory construction and staged policy learning for robotic platforms and smart hardware. A closed-loop sandbox generates multimodal interaction trajectories covering goal revisions and failure recovery, which are checked against execution outcomes and semantic criteria. Supervised fine-tuning and multi-teacher on-policy distillation precede reinforcement learning with DPER (Dual-Source Privileged On-Policy Self-Distillation with Token-Level Evidence Routing). DPER routes token-level evidence between a frozen privileged teacher and a privileged student sharing the rollout policy's parameters, modulating reward-derived token advantages while preserving their signs. Training rewards combine generative process evaluation, multi-judge rankings, and task-queue consistency. JoyInside-VLM-Agent achieves a BFCL v4 overall score of 77.44, exceeding the evaluated baselines; results on the other public benchmarks are mixed. On JoyInside-Devices and JoyInside-Robotics, the evaluated system, which includes domain-specific skill refinement, leads the baselines on all reported metrics. Two real-world quadruped case studies illustrate visually grounded search, replanning after execution failure, and explicit failure reporting.
Chinese Translation
具身智能体必须将多模态观察和用户请求转化为可执行的动作，同时适应产品约束、变化的目标和执行反馈。我们提出了JoyInside-VLM-Agent，这是一种通过基于产品条件的轨迹构建和分阶段策略学习开发的ReAct风格多模态智能体，适用于机器人平台和智能硬件。一个闭环沙箱生成涵盖目标修订和失败恢复的多模态交互轨迹，这些轨迹会根据执行结果和语义标准进行检查。监督微调和多教师在线蒸馏在使用DPER（双源特权在线自蒸馏与令牌级证据路由）的强化学习之前进行。DPER在一个冻结的特权教师和一个共享回放策略参数的特权学生之间路由令牌级证据，调节奖励派生的令牌优势，同时保持其符号。训练奖励结合了生成过程评估、多评审排名和任务队列一致性。JoyInside-VLM-Agent在BFCL v4整体得分中达到了77.44，超过了评估基准；在其他公共基准上的结果则表现不一。在JoyInside-Devices和JoyInside-Robotics上，评估系统（包括领域特定的技能精炼）在所有报告的指标上领先于基准。两个真实世界的四足案例研究展示了视觉基础的搜索、执行失败后的重新规划以及明确的失败报告。
cs.RO / 32 / 2610.12972
Adaptive Whole-Body MPC with Optimal Posture Planning for Humanoid Robots under Unknown Payloads
在未知负载下具有最佳姿态规划的自适应全身模型预测控制（MPC）用于类人机器人
Abstract
Humanoid robots performing lifting and material-handling tasks must maintain balance despite unknown and varying payloads. However, payload uncertainty significantly affects whole-body dynamics and may destabilize conventional controllers. We propose an adaptive whole-body control architecture that integrates three components: (i) a centroidal model predictive controller (MPC) that generates dynamically feasible reference motions, (ii) an adaptive fuzzy balance controller that compensates for payload-induced model mismatch by generating corrective torques distributed across the supporting-leg joints, and (iii) a waypoint-based optimal posture planner built on a triple-integrator model that produces smooth sit-to-stand and stand-to-sit trajectories satisfying center-of-mass (CoM) stability constraints, joint limits, and whole-body feasibility requirements. Together, these components enable stable lowering, grasping, lifting, carrying, and placing behaviors under significant payload uncertainty. Extensive simulation and hardware experiments demonstrate that the proposed framework improves balance robustness and reduces tracking degradation during posture transitions and lifting tasks compared to non-adaptive baselines.
Chinese Translation
执行举升和物料搬运任务的类人机器人必须在面对未知和变化的负载时保持平衡。然而，负载的不确定性显著影响全身动态，可能导致传统控制器的不稳定。我们提出了一种自适应全身控制架构，集成了三个组件：(i) 一个质心模型预测控制器（MPC），生成动态可行的参考运动；(ii) 一个自适应模糊平衡控制器，通过生成分布在支撑腿关节上的修正扭矩来补偿由负载引起的模型不匹配；(iii) 一个基于三重积分器模型的路径点优化姿态规划器，产生平滑的坐立和立坐轨迹，满足质心（CoM）稳定性约束、关节限制和全身可行性要求。这些组件共同使得在显著负载不确定性下实现稳定的降低、抓取、举升、搬运和放置行为。大量的仿真和硬件实验表明，与非自适应基线相比，所提出的框架提高了平衡的鲁棒性，并减少了在姿态过渡和举升任务中的跟踪退化。
cs.RO / 33 / 2610.12999
EgoReveal: Learning Active Vision and Object Manipulation under Partial Occlusions from Egocentric Demonstrations
EgoReveal：从自我中心示范中学习部分遮挡下的主动视觉和物体操作
Abstract
Learning from Demonstrations (LfD) enables robots to learn object manipulation directly from human demonstrations. In contrast to resource-intensive and labor-intensive teleoperation, egocentric demonstrations are easy to collect, enabling scalable data collection. Manipulated objects are often partially occluded in everyday environments, yet prior egocentric demonstration-based methods that require no robots or specialized teaching devices do not explicitly address occlusions. To fill this gap, we propose EgoReveal, a framework that learns viewpoint-repositioning together with manipulation actions from egocentric human demonstrations to reposition the viewpoint and obtain target visibility under partial occlusions. EgoReveal instantiates this as two imitation policies: one predicts the future camera-arm pose sequence from the current view and camera-arm pose, and the other predicts the future control commands from object keypoints and a history of control commands. EgoReveal switches between the two based on target visibility. In experiments where the manipulated object is partially occluded initially, EgoReveal repositions the camera to obtain visibility and completes the task more reliably than fixed-camera baselines.
Chinese Translation
从示范学习（LfD）使机器人能够直接从人类示范中学习物体操作。与资源密集和劳动密集的遥控操作相比，自我中心示范易于收集，从而实现可扩展的数据收集。在日常环境中，被操作的物体通常会部分遮挡，然而，先前基于自我中心示范的方法不需要机器人或专门的教学设备，但并未明确解决遮挡问题。为填补这一空白，我们提出了EgoReveal，一个框架，通过自我中心人类示范学习视角重定位和操作动作，以在部分遮挡下重新定位视角并获得目标可见性。EgoReveal将其具体化为两个模仿策略：一个从当前视图和相机-手臂姿态预测未来的相机-手臂姿态序列，另一个从物体关键点和控制命令历史预测未来的控制命令。EgoReveal根据目标可见性在这两者之间切换。在实验中，当被操作的物体最初部分遮挡时，EgoReveal重新定位相机以获得可见性，并比固定相机基线更可靠地完成任务。
cs.RO / 34 / 2610.13007
CeDR-Nav: Cross-Embodiment Mapless Navigation via Offline Demonstration Recasting
CeDR-Nav：通过离线演示重塑实现跨体现无地图导航
Abstract
Learning-based methods have demonstrated strong performance in mapless robot navigation. However, a learned policy remains restricted to the embodiment on which it was trained. Many cross-embodiment navigation methods require time-consuming online training and can still underperform target-specific policies. In fact, demonstrations collected on a source embodiment already contain navigation experience that is potentially reusable for deriving a policy for each target embodiment, but the embodiment mismatch prevents these demonstrations from directly serving as valid supervision for the target. In this article, we present CeDR-Nav, a cross-embodiment demonstration recasting framework that trains a dedicated navigation policy for each target embodiment from existing source demonstrations without online data collection. Geometry-consistent scaling with per-demonstration validity assessment converts the source demonstrations into target training data, and a target-specific Gaussian mixture policy generates feasible avoidance maneuvers for each target embodiment. Extensive quantitative simulations and real-world deployments across multiple robot platforms demonstrate the effectiveness of CeDR-Nav, with an average success-rate gain of 13.8 percentage points over DRL-DCLP in cross-embodiment simulations and target-specific policy training completed within seconds.
Chinese Translation
基于学习的方法在无地图机器人导航中表现出色。然而，学习到的策略仍然局限于其训练所用的体现。许多跨体现导航方法需要耗时的在线训练，并且仍然可能无法达到特定目标策略的性能。实际上，在源体现上收集的演示已经包含了潜在可重用的导航经验，用于为每个目标体现推导策略，但体现不匹配阻止了这些演示直接作为目标的有效监督。在本文中，我们提出了CeDR-Nav，一种跨体现演示重塑框架，它能够从现有的源演示中为每个目标体现训练专用的导航策略，而无需在线数据收集。通过几何一致的缩放和每个演示的有效性评估，将源演示转换为目标训练数据，并且目标特定的高斯混合策略为每个目标体现生成可行的避障动作。广泛的定量模拟和多个机器人平台的实际部署证明了CeDR-Nav的有效性，在跨体现模拟中，相较于DRL-DCLP，成功率平均提高了13.8个百分点，并且目标特定策略的训练在几秒内完成。
cs.RO / 35 / 2610.13049
Failure-Resilient Multi-Robot Coordination with $k$-Connectivity Constrained Submodular Maximization
具有 $k$-连通性约束的抗失效多机器人协调与子模最大化
Abstract
Multi-robot coordination is essential in applications such as coverage, target tracking, and information gathering, where robots must jointly select actions to maximize team performance. Submodular maximization provides an effective framework for such coordination, and resilient variants further account for robot failures or adversarial removals. However, existing methods typically overlook the impact of robot loss on the communication network, which may disconnect the surviving team and prevent continued coordination. To address this issue, we introduce Resilient $k$-Connected Submodular Maximization, which jointly considers task performance, resilience, and communication connectivity. The resulting feasible family is not a matroid, preventing the direct application of classical curvature-greedy guarantees. We overcome this challenge through a Menger-based connectivity reduction, a connectivity-aware greedy algorithm with a feasibility-preserving fallback, and a new curvature-based analysis that captures the cost of enforcing connectivity through a single factor $\rho$. Our method achieves a worst-case post-failure objective within a factor of $[\max(1-\nu_f(\mathcal{I}),h(N,\beta))]^2 \rho \frac{1}{\kappa_f}(1-e^{-\kappa_f})$ of the optimal connectivity-respecting resilient assignment, while recovering the classical unconstrained guarantee in the appropriate limits. Experiments on multi-robot target tracking validate the proposed method and characterize the practical impact of connectivity constraints.
Chinese Translation
多机器人协调在覆盖、目标跟踪和信息收集等应用中至关重要，其中机器人必须共同选择行动以最大化团队表现。子模最大化为这种协调提供了有效的框架，而抗失效变体进一步考虑了机器人故障或对抗性移除的情况。然而，现有方法通常忽视了机器人损失对通信网络的影响，这可能导致幸存团队的断开并阻碍持续协调。为了解决这个问题，我们引入了抗失效 $k$-连通子模最大化，它共同考虑了任务表现、抗失效能力和通信连通性。由此产生的可行家族不是一个矩阵，从而阻止了经典曲率贪婪保证的直接应用。我们通过基于Menger的连通性减少、具有可行性保护回退的连通性感知贪婪算法，以及一种新的基于曲率的分析来克服这一挑战，该分析通过单一因子 $
ho$ 捕捉强制连通性的成本。我们的方法在最坏情况下的故障后目标达到最优连通性尊重的抗失效分配的 $[ ext{max}(1-
u_f( ext{I}),h(N,eta))]^2
ho rac{1}{ au_f}(1-e^{- au_f})$ 的因子，同时在适当的极限中恢复经典的无约束保证。在多机器人目标跟踪实验中验证了所提方法，并表征了连通性约束的实际影响。
cs.RO / 36 / 2610.13077
Towards Safer Cities: A Design Process for Perception Datasets in Urban Driving Scenarios
迈向更安全的城市：城市驾驶场景中感知数据集的设计过程
Abstract
Urban environments present significant challenges for the perception systems of automated driving due to their high complexity, characterized by dense traffic, diverse obstacles, or a wide range of possible scenarios. While numerous automotive datasets exist, few focus on urban settings, and the majority are acquired exclusively on passenger cars and overlook other transport modalities such as heavy-duty trucks, busses and trams. In contrast to passenger vehicles, these platforms operate under different geographical and operational constraints; including traversing pedestrian zones or platform-specific tasks such as movement of trailer or winter service. In this work a method is proposed for a guided dataset creation based on an analysis of existing datasets using user-generated map data. Furthermore, representative truck-specific use cases are identified and exemplified to demonstrate the applicability of the proposed methodology and to support targeted acquisition planning.
Chinese Translation
城市环境对自动驾驶的感知系统提出了重大挑战，其复杂性体现在密集的交通、多样的障碍物以及广泛的可能场景上。虽然存在众多汽车数据集，但很少有专注于城市环境的，大多数数据集仅在乘用车上获取，忽视了重型卡车、公交车和电车等其他交通方式。与乘用车不同，这些平台在不同的地理和操作约束下运行，包括穿越行人区或执行特定平台的任务，如拖车移动或冬季服务。在本研究中，提出了一种基于对现有数据集分析的引导数据集创建方法，利用用户生成的地图数据。此外，识别并示例了一些具有代表性的卡车特定用例，以展示所提方法的适用性，并支持有针对性的获取规划。
cs.RO / 37 / 2610.13079
VICAR: Video Imitation with Contact-Aware Retargeting for Humanoid Interactive Skills
VICAR：具有接触感知重定向的人形互动技能视频模仿
Abstract
Third-person human videos offer a scalable alternative to task-specific reward engineering, teleoperation, and manually scripted demonstrations for humanoid loco-manipulation. Recent video-to-humanoid systems have shown that noisy human motion, object trajectories, and contacts can be extracted from unstructured videos and refined into deployable skills. However, a key gap remains for contact-rich imitation: the task is often defined by world-frame contacts whose poses, timing, and speed must survive embodiment mismatch and runtime scene changes. Towards this, we present VICAR (Video Imitation with Contact-Aware Retargeting), a pipeline for learning humanoid interactive skills from third-person human videos. The system first recovers human motion, reconstructs task-relevant obstacles from monocular video, and then retargets the motion to a humanoid while preserving global contact poses, motion speed, and collision clearance. Unlike retargeting methods that only remove penetration or preserve local interaction geometry, our formulation represents user-specified contact poses and collision clearance as high-priority Lagrangian penalty terms and optimizes the full motion to remain smooth through contact-rich transitions. The resulting motions are used to train a contact-conditioned motion generator whose inputs include movable contact points enabling runtime adaptation to new contact positions. We demonstrate the framework on ten diverse third-person video skills, including six table-tennis serves, three object-pickup tasks, and ladder climbing, and deploy the learned skills on hardware. These results suggest that contact-preserving retargeting is a practical step toward humanoids that learn useful interactive behavior by watching humans.
Chinese Translation
第三人称人类视频为人形运动操作提供了一种可扩展的替代方案，替代了任务特定的奖励工程、远程操作和手动脚本演示。最近的视频到人形系统表明，可以从非结构化视频中提取嘈杂的人类运动、物体轨迹和接触，并将其精炼为可部署的技能。然而，对于接触丰富的模仿，仍然存在一个关键的缺口：任务通常由世界坐标系中的接触定义，其姿态、时机和速度必须在体现不匹配和运行时场景变化中保持稳定。为此，我们提出了VICAR（具有接触感知重定向的视频模仿），这是一个从第三人称人类视频中学习人形互动技能的管道。该系统首先恢复人类运动，从单目视频中重建与任务相关的障碍物，然后在保留全局接触姿态、运动速度和碰撞间隙的情况下，将运动重定向到人形。与仅去除穿透或保留局部交互几何的重定向方法不同，我们的公式将用户指定的接触姿态和碰撞间隙表示为高优先级的拉格朗日惩罚项，并优化整个运动以在接触丰富的过渡中保持平滑。生成的运动用于训练一个接触条件的运动生成器，其输入包括可移动的接触点，从而实现对新接触位置的运行时适应。我们在十个不同的第三人称视频技能上演示了该框架，包括六个乒乓球发球、三个物体拾取任务和梯子攀爬，并在硬件上部署了学习到的技能。这些结果表明，保留接触的重定向是朝着人形机器人通过观察人类学习有用互动行为的实际一步。
cs.RO / 38 / 2610.13087
Belief-Overlap Partial Inversion: Object-Centric Flow Replanning for Humanoid Loco-Manipulation
信念重叠部分反演：面向人形机器人运动操控的物体中心流动重新规划
Abstract
People often learn how an unfamiliar object moves through a small exploratory push, using its response to guide their next motion. Inspired by this behavior, we study humanoids pushing carts whose mass and caster stiffness are initially unknown. Our object-centric framework closes online estimation, flow-based planning, and whole-body control into a loop, creating a tension: replacing the plan whenever the belief shifts disrupts the maneuver under way, while keeping it strands the robot on a path its evidence has ruled out. Belief-Overlap Partial Inversion (BOPI) scales revision to agreement, inverting the flow under the old belief only as far as the beliefs disagree before reconstructing under the new one. Across 18 simulated goals it lowers pooled final position and yaw error by roughly 33% and 47% against fixed-depth replanning, and on hardware it translates and turns a non-holonomic cart.
Chinese Translation
人们通常通过小幅探索性推力来学习一个不熟悉物体的运动，并利用其反应来指导下一步的动作。受到这一行为的启发，我们研究了推动车辆的人形机器人，其质量和万向轮刚度最初是未知的。我们的物体中心框架将在线估计、基于流动的规划和全身控制结合成一个循环，形成了一种张力：每当信念变化导致当前操作受到干扰时，替换计划会中断正在进行的操作，而保持计划则会使机器人停留在其证据已排除的路径上。信念重叠部分反演（Belief-Overlap Partial Inversion, BOPI）将修正扩展到一致性，仅在信念不一致的情况下反转旧信念下的流动，然后在新信念下重构。在18个模拟目标中，它将最终位置和偏航误差的汇总降低了大约33%和47%，相比于固定深度的重新规划，并且在硬件上能够使非全向车辆进行平移和转向。
cs.RO / 39 / 2610.13093
ReAT: Residual Action Tokenization for Autoregressive Robot Policies
ReAT：自回归机器人策略的残差动作标记化
Abstract
Autoregressive robot policies rely on action tokenizers to convert continuous action chunks into compact sequences of discrete prediction targets. Existing tokenizers can reconstruct action chunks from discrete codes, but often leave implicit what each individual token contributes to the reconstructed action. We introduce Residual Action Tokenization (ReAT), which gives each token an explicit role in refining the current action reconstruction. It uses the decoded reconstruction as feedback to construct the next token from the remaining action error. This closes the loop between encoding and decoding, making each new token responsive to the errors left by the actual decoded prefix. Across 19 manipulation tasks spanning three simulation benchmarks and real-world settings, 8-token ReAT policies outperform the evaluated baselines in task success; on LIBERO, even 4-token ReAT surpasses the evaluated 8-token action-tokenization baselines at half the token budget. These results suggest a path toward action tokenization that supports self-corrective policy learning, helping autoregressive policies learn to progressively refine their own action plans.
Chinese Translation
自回归机器人策略依赖于动作标记器将连续的动作块转换为紧凑的离散预测目标序列。现有的标记器能够从离散代码重构动作块，但通常未明确每个单独标记对重构动作的贡献。我们提出了残差动作标记化（ReAT），为每个标记在细化当前动作重构中赋予明确的角色。它利用解码后的重构作为反馈，从剩余的动作误差中构建下一个标记。这在编码和解码之间闭合了循环，使每个新标记对实际解码前缀留下的误差做出响应。在涵盖三个仿真基准和现实世界环境的19个操作任务中，8标记的ReAT策略在任务成功率上超越了评估的基准；在LIBERO上，甚至4标记的ReAT在一半的标记预算下也超越了评估的8标记动作标记化基准。这些结果表明了一条支持自我修正策略学习的动作标记化路径，帮助自回归策略逐步细化其自身的行动计划。
cs.RO / 40 / 2610.13117
Forward Dynamics Models
前向动力学模型
Abstract
Vision-language-action (VLA) models and world action models (WAMs) have advanced robotic policy learning by building on pretrained vision-language models (VLMs) and video generative models (VGMs), respectively. Yet VLAs learn without explicit supervision from future world states, while WAMs that condition actions on predicted futures incur extra inference cost and let visual prediction errors propagate into actions. We present Forward Dynamics Models (FDM), a family of robotic action models that couples action and future world-state prediction through an action-to-future paradigm. Future prediction is conditioned on action features, never the reverse: end-to-end training lets ground-truth future states supervise the policy through these features, while prediction errors cannot propagate into actions. At inference, FDM skips future generation entirely, avoiding its computational cost. The paradigm is backbone-agnostic, and we instantiate it with both VLM and VGM backbones. Experiments in simulation and on real robots demonstrate the effectiveness of FDM across both backbone families, achieving a state-of-the-art 99.4% average success rate on LIBERO and absolute gains of up to 13% over VLA and WAM methods on real-world tasks.
Chinese Translation
视觉-语言-动作（VLA）模型和世界动作模型（WAMs）通过基于预训练的视觉-语言模型（VLMs）和视频生成模型（VGMs）推动了机器人策略学习的发展。然而，VLA在没有来自未来世界状态的明确监督的情况下进行学习，而WAM在基于预测的未来条件下进行动作时则会产生额外的推理成本，并使视觉预测误差传播到动作中。我们提出了前向动力学模型（FDM），这是一类通过动作到未来的范式将动作与未来世界状态预测相结合的机器人动作模型。未来预测以动作特征为条件，而不是反向：端到端训练使得真实的未来状态通过这些特征对策略进行监督，而预测误差无法传播到动作中。在推理时，FDM完全跳过未来生成，避免了其计算成本。该范式与骨干网络无关，我们用VLM和VGM骨干网络进行了实例化。在仿真和真实机器人上的实验表明，FDM在这两类骨干网络中均表现出有效性，在LIBERO上达到了99.4%的先进平均成功率，并在真实世界任务中相较于VLA和WAM方法实现了高达13%的绝对增益。
cs.RO / 41 / 2610.13118
The value of information and the emergence of communication in evolved e-puck robots
信息的价值与进化型 e-puck 机器人中通信的出现
Abstract
We study when communication emerges between autonomous agents and what role an innate action selection architecture plays. Two e-puck robots, reconstructed in PyBullet from their Webots model and validated against it, must find a feeding zone that can only be detected when a robot passes over it. Each robot can emit a one-bit signal whose direction is detected by the receiver. Control is provided by a GRU recurrent network with a residual path, either operating alone or as a channel within an Embedded Action Selection Architecture (EASA). Fitness measures only the time the robots spend over the food. When the robots receive a compass pointing to the food, use of the signal disappears completely. Without a compass, in a small arena, three independent runs achieved similar performance using different strategies: a referential signal, a permanent beacon, or no use of the signal. As the arena expands, the strategy based on a referential signal generalizes best. In contrast, evolving directly in the large arena does not increase the frequency of communication, but it does separate the strategies, and the referential signal clearly outperforms solitary searching. Within EASA, innate reflexes constrain the network and reduce performance. As its priorities evolve, evolution consistently weakens those reflexes. The results suggest that communication emerges in proportion to the value of the information it transmits, and that an innate scaffolding designed for isolated agents can be an impediment in a social task.
Chinese Translation
我们研究自主体之间何时会出现通信以及内在动作选择架构所起的作用。两个 e-puck 机器人在 PyBullet 中重建自其 Webots 模型并经过验证，必须找到一个只能在机器人经过时被检测到的喂食区。每个机器人可以发出一个方向被接收者检测到的一位信号。控制由一个具有残差路径的 GRU 循环网络提供，该网络可以单独操作或作为嵌入式动作选择架构（EASA）中的一个通道。适应度仅衡量机器人在食物上花费的时间。当机器人接收到指向食物的指南针时，信号的使用完全消失。在一个小的竞技场中，三次独立的实验使用不同的策略达到了相似的表现：参考信号、永久信标或不使用信号。随着竞技场的扩大，基于参考信号的策略表现出最佳的推广性。相比之下，直接在大竞技场中进化并未增加通信的频率，但确实使策略分化，参考信号明显优于单独搜索。在 EASA 中，内在反射限制了网络并降低了性能。随着其优先级的进化，进化过程持续削弱这些反射。结果表明，通信的出现与其传递的信息价值成正比，而为孤立代理设计的内在支架在社会任务中可能成为一种障碍。
cs.RO / 42 / 2610.13119
Human2Dex: Dexterous Robot Learning from Wrist-Centric Human Demonstrations
Human2Dex：基于手腕中心的人类演示学习灵巧机器人
Abstract
Collecting dexterous manipulation demonstrations for each robot hand requires repeated task-specific data acquisition. We ask whether one shared corpus of natural human demonstrations can instead serve as a reusable data interface for different dexterous hands, without collecting task demonstrations on each target robot. We present Human2Dex, a wrist-centric human demonstration interface that separates shared human observation and motion semantics from embodiment-specific execution. It aligns observations through local hand-appearance adaptation and a shared 21-point hand representation with consistent structural semantics across human and robot observations. It constructs executable supervision by retargeting fused human hand motion to each target hand, and introduces the Grasp-Pocket Interaction Representation (GPIR) to express tracked objects relative to a hand-centered Functional Grasp Pocket. The corpus is converted into separate training sets for Linker O6 and Wuji, while action labels, normalization, training, and diffusion-policy checkpoints remain hand-specific. Across six real-world tasks, Human2Dex achieves 109/120 (90.8%) successes on Linker O6 and 97/120 (80.8%) on Wuji. Across both hands, Human-Raw achieves 63/240 (26.3%) successes, whereas the complete Human2Dex configuration achieves 206/240 (85.8%). These results support the use of wrist-centric human demonstrations as a reusable interface across calibrated dexterous hands while retaining embodiment-specific policy learning. The project website is publicly available at https://human2dex.pages.dev/.
Chinese Translation
为每个机器人手收集灵巧操作演示需要重复进行特定任务的数据采集。我们探讨是否可以利用一个共享的自然人类演示语料库作为不同灵巧手的可重用数据接口，而无需在每个目标机器人上收集任务演示。我们提出了Human2Dex，这是一个以手腕为中心的人类演示接口，它将共享的人类观察和运动语义与特定体现的执行分离。该接口通过局部手部外观适应和共享的21点手部表示对观察进行对齐，确保人类和机器人观察之间具有一致的结构语义。它通过将融合的人类手部运动重新定向到每个目标手来构建可执行的监督，并引入了抓取-口袋交互表示（Grasp-Pocket Interaction Representation, GPIR）来表达相对于手中心功能抓取口的跟踪对象。该语料库被转换为Linker O6和Wuji的单独训练集，而动作标签、归一化、训练和扩散策略检查点仍然是手部特定的。在六个现实世界任务中，Human2Dex在Linker O6上实现了109/120（90.8%）的成功率，在Wuji上实现了97/120（80.8%）的成功率。在两个手上，Human-Raw实现了63/240（26.3%）的成功率，而完整的Human2Dex配置实现了206/240（85.8%）的成功率。这些结果支持将以手腕为中心的人类演示作为跨校准灵巧手的可重用接口，同时保留特定体现的策略学习。项目网站可公开访问，网址为https://human2dex.pages.dev/
cs.RO / 43 / 2610.13158
Low-Latency Mapless ESDF Queries for MAV Reactive Collision Avoidance
低延迟无地图的欧几里得有符号距离场查询用于无人机反应式避碰
Abstract
Reactive collision avoidance in unknown, cluttered environments must perceive obstacles and correct motion within milliseconds. Existing methods rely on maps for high-rate collision queries, but their acceleration structures require pre- processing. That cost is amortized in long-horizon planning, where a map serves millions of queries; reactive planning, however, is a near-field, short-horizon problem with potentially only hundreds of queries per measurement, where building a map may not pay off. We propose a mapless, Graphics Processing Unit (GPU)-accelerated Euclidean Signed Distance Field (ESDF) query with negligible pre-processing that operates directly on posed depth images, with resolution set by pooling the image rather than by a voxel size. On an embedded GPU it is simultaneously more accurate and lower in total latency than every voxel-grid configuration we measured between 2 cm and 20 cm, at one to two orders of magnitude less memory at matched resolution. With our query integrated into a Model Predictive Controller (MPC), simulated runs in laser-scanned forests at up to 9 m/s collide in 0.4 % through dense branches, against 5.9 % with the best voxel grid and 5.5 % with the best mapless baseline at our resolution, whose failures we attribute to voxel resolution and total latency respectively. We deploy the system onboard at two forest sites and fly at commanded speeds of up to 4 m/s.
Chinese Translation
在未知且杂乱的环境中，反应式避碰必须在毫秒内感知障碍物并修正运动。现有方法依赖于地图进行高频率的碰撞查询，但它们的加速结构需要预处理。这种成本在长时间规划中得以摊销，因为地图可以服务于数百万次查询；然而，反应式规划是一个近场短期问题，每次测量可能只有数百次查询，此时构建地图可能得不偿失。我们提出了一种无地图的、基于图形处理单元（GPU）加速的欧几里得有符号距离场（ESDF）查询，几乎不需要预处理，直接在姿态深度图上操作，其分辨率通过池化图像而非体素大小来设定。在嵌入式GPU上，它的准确性更高且总延迟更低，相较于我们测量的2厘米到20厘米之间的每种体素网格配置，其内存使用量减少了一个到两个数量级。将我们的查询集成到模型预测控制器（MPC）中，在激光扫描的森林中进行的模拟运行中，以高达9米/秒的速度飞行时，碰撞率为0.4%，而最佳体素网格的碰撞率为5.9%，最佳无地图基线的碰撞率为5.5%，我们将这些失败归因于体素分辨率和总延迟。我们在两个森林地点部署该系统，并以高达4米/秒的指令速度飞行。
cs.RO / 44 / 2610.13164
Grounded-Dreamer: Robot Learning from World Model Synthetic Data with Physical Grounding
Grounded-Dreamer：基于物理基础的世界模型合成数据进行机器人学习
Abstract
Scaling robot manipulation policies requires large amounts of diverse, action-labeled data, yet collecting such data in the real world remains expensive and difficult to scale. Existing methods either depend on human teleoperation, rely on simulation with limited visual realism, or use human videos that lack embodiment-specific annotations. In this paper, we introduce Grounded-Dreamer, a framework for synthesizing physically grounded robot manipulation data with video world models while avoiding human teleoperation demonstrations. Grounded-Dreamer combines the visual realism of generative video models with simulator physics engines. Starting from a small set of real images depicting robot task initial states, our method automatically reconstructs task-relevant objects and camera geometry, aligns them in simulation, and collects physically valid trajectories. These trajectories are used to post-train both a video world model and an inverse dynamics model, enabling the generation of visually realistic robot manipulation videos and corresponding robot actions from real images. To filter physically implausible rollouts, Grounded-Dreamer replays generated trajectories in the reconstructed simulation and retains only those that successfully complete the tasks. The real-world robot policies are trained solely on the grounded synthetic trajectories without any human demonstrations. Empirical results in real-world tasks demonstrate that Grounded-Dreamer can synthesize physically grounded data and substantially facilitate policy learning without requiring human teleoperation effort.
Chinese Translation
扩展机器人操作策略需要大量多样化的、带有动作标签的数据，但在现实世界中收集此类数据仍然昂贵且难以扩展。现有方法要么依赖于人工遥控，要么依赖于视觉真实感有限的模拟，或者使用缺乏特定体现注释的人类视频。本文介绍了Grounded-Dreamer，一个通过视频世界模型合成物理基础的机器人操作数据的框架，同时避免了人工遥控演示。Grounded-Dreamer将生成视频模型的视觉真实感与模拟物理引擎相结合。从一小组描绘机器人任务初始状态的真实图像开始，我们的方法自动重建与任务相关的物体和相机几何，在线性模拟中对齐它们，并收集物理有效的轨迹。这些轨迹用于对视频世界模型和逆动力学模型进行后训练，使得能够从真实图像生成视觉真实的机器人操作视频和相应的机器人动作。为了过滤物理上不合理的轨迹，Grounded-Dreamer在重建的模拟中重放生成的轨迹，仅保留成功完成任务的轨迹。现实世界中的机器人策略完全基于基础合成轨迹进行训练，而无需任何人工演示。实证结果表明，Grounded-Dreamer能够合成物理基础的数据，并在不需要人工遥控努力的情况下显著促进策略学习。
cs.RO / 45 / 2610.13190
Prediction-Informed Landing Window Planning for UAV Landing on USV in High Sea States
基于预测的信息化着陆窗口规划：无人机在无人水面载体上的着陆
Abstract
Autonomous landing of unmanned aerial vehicle (UAV) on unmanned surface vehicle (USV) is difficult in high sea states. Existing hierarchical landing methods often infer future USV motion only implicitly from short observation histories, which limits their ability to adapt to severe and rapidly changing sea conditions. This paper presents a prediction-informed hierarchical control framework. A pretrained model is developed to explicitly predict the USV attitude during high-level landing decisions. The prediction helps identify suitable landing windows and choose the right touchdown time, while a conventional low-level flight controller maintains UAV stability and tracks motion commands. To evaluate the framework under realistic disturbances, a real-sim validation setup is constructed in which USV-motion data collected in real sea conditions are transferred to a MuJoCo USV. The experiments demonstrate the feasibility of using explicit attitude prediction to improve landing-window selection and support timing-aware UAV landing planning in rapidly varying marine environments.
Chinese Translation
在恶劣海况下，无人机（UAV）在无人水面载体（USV）上的自主着陆是一个挑战。现有的分层着陆方法通常仅通过短期观察历史隐式推断未来的USV运动，这限制了它们在严重且快速变化的海洋条件下的适应能力。本文提出了一种基于预测的信息化分层控制框架。我们开发了一个预训练模型，以显式预测USV在高层次着陆决策过程中的姿态。该预测有助于识别合适的着陆窗口并选择正确的着陆时间，同时传统的低层次飞行控制器保持UAV的稳定性并跟踪运动指令。为了在现实干扰下评估该框架，我们构建了一个真实-仿真验证设置，将在真实海况下收集的USV运动数据转移到MuJoCo USV中。实验表明，使用显式姿态预测来改善着陆窗口选择并支持在快速变化的海洋环境中进行时序感知的UAV着陆规划是可行的。
cs.RO / 46 / 2610.13195
PATACON: GPU-Parallel Tangent-Bundle RRT-Connect for Constrained Motion Planning
PATACON：用于约束运动规划的GPU并行切线束RRT-Connect
Abstract
Fast whole-body motion planning enables continuous replanning for humanoid robots in dynamic environments while satisfying collision avoidance and kinematic constraints. However, constrained motion planning incurs substantial computational cost due to repeated constraint projection in addition to collision checking and nearest-neighbor search. Recently, GPU-based massively parallel planners have been proposed to reduce computational cost. Nevertheless, these planners still steer in the ambient configuration space, so projecting candidates that deviate from the manifold remains a significant source of overhead, particularly in highly constrained, high-dimensional problems. In this work, we introduce PATACON, a GPU-parallel bidirectional planner that explores tangent space to exploit the local manifold geometry induced by known constraint functions. Experiments on dual-arm closed-chain and humanoid whole-body constrained planning tasks show that PATACON outperforms the prior work as configuration-space dimension and constraint codimension grow, achieving up to an 11.25 times speedup. These results suggest that GPU-parallel tangent-space exploration brings constrained whole-body motion planning closer to real-time performance.
Chinese Translation
快速的全身运动规划使得类人机器人能够在动态环境中进行连续重新规划，同时满足避碰和运动学约束。然而，约束运动规划由于需要重复进行约束投影，除了碰撞检测和最近邻搜索外，带来了巨大的计算成本。最近，基于GPU的大规模并行规划器被提出以降低计算成本。然而，这些规划器仍然在环境配置空间中进行引导，因此投影偏离流形的候选点仍然是一个显著的开销来源，尤其是在高度约束的高维问题中。在本研究中，我们提出了PATACON，这是一种GPU并行的双向规划器，探索切线空间以利用已知约束函数所诱导的局部流形几何。对双臂闭链和类人全身约束规划任务的实验表明，随着配置空间维度和约束余维的增加，PATACON的性能优于先前的工作，实现了最高11.25倍的加速。这些结果表明，GPU并行切线空间探索使得约束全身运动规划更接近实时性能。
cs.RO / 47 / 2610.13212
Beyond Actions: Learning Future Operation Targets for Vision-Language-Action Models
超越动作：为视觉-语言-动作模型学习未来操作目标
Abstract
Actions specify what a robot should execute, but do not describe what those commands will achieve. Vision-language-action (VLA) policies nevertheless learn primarily by predicting action chunks from images, language, and state. We introduce Action--Future Coupled Effect Learning (AFCE), which makes a future-grounded physical Effect the policy's primary prediction target. An offline tokenizer constructs temporally ordered Effect tokens from demonstrated actions, realized robot responses, and sparse visual transitions. Reconstruction objectives encourage the representation to retain executable detail and observed physical change. A flow-matching VLA predicts Effects from current images, instruction, and robot state, while a state-conditioned decoder converts them into action chunks. Embodiment-specific interfaces allow single- and dual-arm tasks to share the Effect model. On DexJoCo, AFCE achieves $50.97\%$ overall success, improving over $\pi_{0.5}$ and UVT by 11.82 and 15.21 percentage points, respectively. It leads on all five bimanual tasks, highlighting the value of Effect supervision for coordinated dexterous manipulation. The project page is available at https://afceproject.github.io/AFCE-Web/.
Chinese Translation
动作指定了机器人应执行的内容，但并未描述这些指令将实现的目标。尽管如此，视觉-语言-动作（VLA）策略主要通过从图像、语言和状态中预测动作片段来学习。我们提出了动作-未来耦合效应学习（AFCE），使得以未来为基础的物理效应成为策略的主要预测目标。离线分词器从示范动作、实现的机器人响应和稀疏视觉过渡中构建时间顺序的效应标记。重建目标鼓励表示保留可执行细节和观察到的物理变化。流匹配VLA从当前图像、指令和机器人状态中预测效应，而状态条件解码器将其转换为动作片段。特定于体现的接口允许单臂和双臂任务共享效应模型。在DexJoCo上，AFCE实现了$50.97\%$的整体成功率，分别比$ ext{π}_{0.5}$和UVT提高了11.82和15.21个百分点。它在所有五个双手任务中领先，突显了效应监督在协调灵巧操作中的价值。项目页面可访问 https://afceproject.github.io/AFCE-Web/。
cs.RO / 48 / 2610.13262
Dual-Stream Memory for Vision-Language-Action Policies
视觉-语言-动作策略的双流记忆
Abstract
Memory-dependent manipulation requires a policy to track execution state and recover visual evidence that is no longer observable. We propose dual-stream memory for vision-language-action policies, assigning these requirements to complementary representations. A task-agnostic Gated DeltaNet compresses history into one temporal token, supervised by progress prediction and action-conditioned feature reconstruction. A task-conditioned retriever retains frames above a fixed relevance threshold; linear multi-scale fusion encodes each frame into sixteen spatial tokens. We then freeze the temporal compressor and jointly optimize retrieval and policy learning. Across sixteen RoboMME tasks, the complete model achieves 49.22% mean success versus 44.51% for the published FrameSamp+Modul reference, improving fourteen tasks and all four suite means. On four real-robot tasks, mean success reaches 62.5% versus 37.5%, with the largest gains on repeated actions requiring execution-stage discrimination. These results support combining execution context with selective visual access, while failures under cup swaps expose persistent identity tracking as an unresolved requirement.
Chinese Translation
依赖记忆的操作需要一种策略来跟踪执行状态并恢复不再可观察的视觉证据。我们提出了用于视觉-语言-动作策略的双流记忆，将这些需求分配给互补的表征。一种与任务无关的门控DeltaNet（Gated DeltaNet）将历史压缩为一个时间令牌，受进度预测和基于动作的特征重建的监督。一个与任务相关的检索器保留高于固定相关性阈值的帧；线性多尺度融合将每帧编码为十六个空间令牌。然后，我们冻结时间压缩器，并联合优化检索和策略学习。在十六个RoboMME任务中，完整模型的平均成功率达到49.22%，而已发布的FrameSamp+Modul参考模型为44.51%，在十四个任务和所有四个套件均值上有所改善。在四个真实机器人任务中，平均成功率达到62.5%，而37.5%，在需要执行阶段区分的重复动作中获得了最大的提升。这些结果支持将执行上下文与选择性视觉访问相结合，而在杯子交换下的失败则暴露出持续的身份跟踪作为一个未解决的需求。
cs.RO / 49 / 2610.13345
Evo-RL: Iterative Robot Reinforcement Learning from Selective Experience
Evo-RL：基于选择性经验的迭代机器人强化学习
Abstract
Existing real-world reinforcement learning methods improve robot policies by combining human corrections with autonomous rollouts. However, accumulated experience varies in its value for policy improvement, and even successful trajectories can contain inefficient or redundant behavior, motivating more selective experience reuse. To this end, we propose Evo-RL, an iterative reinforcement learning method that combines human interventions with advantage-guided experience selection. Evo-RL uses human corrections and experience with positive estimated advantages from successful trajectories to iteratively refine the policy. To support learning from these different experience sources, we introduce RW-RL-Dataset, which spans over 1,000 hours across four robot embodiments. On a real-world high-precision manipulation task, Evo-RL improves attempt success rate over RECAP by 30.46\%. Our code, model checkpoints, and dataset are publicly available.
Chinese Translation
现有的现实世界强化学习方法通过结合人类纠正与自主执行来改善机器人策略。然而，积累的经验在政策改进中的价值各不相同，即使是成功的轨迹也可能包含低效或冗余的行为，这促使我们更有选择性地重用经验。为此，我们提出了Evo-RL，这是一种迭代强化学习方法，结合了人类干预与优势引导的经验选择。Evo-RL利用来自成功轨迹的人类纠正和具有正估计优势的经验，迭代地优化策略。为了支持从这些不同经验来源的学习，我们引入了RW-RL-Dataset，该数据集涵盖了四种机器人形态下超过1000小时的经验。在一个现实世界的高精度操作任务中，Evo-RL的尝试成功率比RECAP提高了30.46%。我们的代码、模型检查点和数据集均已公开。
cs.RO / 50 / 2610.13373
PartiGen: Part-Aware Representations for Language-Conditioned Humanoid Motion Generation
PartiGen：面向部件的语言条件下类人运动生成表示
Abstract
Translating language into humanoid motion requires coordinating body parts, determining motion duration, and producing references suitable for robot execution. We introduce PartiGen, a part-aware latent diffusion framework that incorporates anatomical structure into a shared whole-body motion representation. Part-specific attention processes anatomically masked frame features, and its outputs are fused to combine local limb information with whole-body context. Conditioned on text and motion history, the model generates motion primitives autoregressively. Forward-kinematics supervision encourages kinematic consistency, while a learned end-of-sequence predictor selects a frame-level endpoint after generation. On HumanML3D retargeted to the Unitree G1, PartiGen achieves a segment-level FID of 3.165, a 16.8% reduction over TextOp, and improves R@1 by 1.4 percentage points. In the duration study with 16-frame primitives, endpoint prediction reduces mean absolute duration error from 12.471 to 3.080 frames compared with a fixed-length cutoff. We further evaluate robot execution using a pretrained, frozen GRIT tracker: 409 of 500 generated references (81.8%) pass the posture-based stability screen in MuJoCo. Following simulation screening and manual selection, 26 of 27 attempted motions are judged successful on the real G1.
Chinese Translation
将语言转化为类人运动需要协调身体部位、确定运动持续时间，并生成适合机器人执行的参考。我们提出了PartiGen，这是一种关注部件的潜在扩散框架，将解剖结构纳入共享的全身运动表示。特定部件的注意力过程对解剖遮罩的帧特征进行处理，其输出被融合以结合局部肢体信息与全身上下文。在文本和运动历史的条件下，该模型自回归地生成运动原语。前向运动学监督促进运动学一致性，而学习的序列结束预测器在生成后选择帧级的结束点。在针对Unitree G1的HumanML3D重定向上，PartiGen实现了3.165的段级FID，相较于TextOp减少了16.8%，并提高了R@1的准确率1.4个百分点。在使用16帧原语的持续时间研究中，结束点预测将平均绝对持续时间误差从12.471帧降低到3.080帧，相较于固定长度的截止。我们进一步使用预训练的冻结GRIT跟踪器评估机器人执行：在500个生成的参考中，有409个（81.8%）通过了MuJoCo中的基于姿态的稳定性筛选。经过模拟筛选和手动选择，27个尝试的动作中有26个被判断为成功在真实的G1上执行。
cs.RO / 51 / 2610.13376
EndoRSI: Physical-Evidence-Driven Recursive Improvement of Endoscopic Robot Policies
EndoRSI：基于物理证据的内窥镜机器人策略递归改进
Abstract
Artificial intelligence is beginning to automate scientific discovery, yet physical robot-policy research still relies on scientists and engineers to interpret experimental failures and decide how controllers should be revised. We present EndoRSI (Endoscopic Recursive Self-Improvement), a human-governed, self-improving robotic research system in which the embedded EndoForge coding agent iteratively develops and refines endoscopic robot policies using evidence from physical experiments. Physical execution remains isolated within a safety-constrained runtime under human authorisation. In motorised endoscopy, validation evidence first caused the agent to reject an initial action representation, while subsequent physical experiments falsified an offline-supported insertion strategy by revealing hazardous closed-loop behaviour absent from the source data. Failed physical rollouts were then incorporated into policy revision; insertion success subsequently increased from 2/21 in the first generation to 17/21 in the fourth. The frozen final policy retained competence across visual and hardware shifts and was the best performing autonomous policy across six independent ex-vivo porcine specimens. These results show that physical robot interaction can serve as experimental evidence for agent-driven robot policy research.
Chinese Translation
人工智能开始自动化科学发现，但物理机器人策略研究仍依赖科学家和工程师解读实验失败，并决定如何修订控制器。我们提出了EndoRSI（内窥镜递归自我改进），这是一个由人类主导的自我改进机器人研究系统，其中嵌入的EndoForge编码代理利用物理实验的证据迭代开发和完善内窥镜机器人策略。物理执行仍在经过人类授权的安全约束运行时内隔离。在电动内窥镜中，验证证据首先导致代理拒绝初始动作表示，而随后的物理实验通过揭示源数据中缺失的危险闭环行为，驳斥了离线支持的插入策略。失败的物理执行结果随后被纳入策略修订；插入成功率从第一代的2/21提高到第四代的17/21。最终冻结的策略在视觉和硬件变化中保持了能力，并且在六个独立的体外猪标本中表现出最佳的自主策略。这些结果表明，物理机器人交互可以作为代理驱动的机器人策略研究的实验证据。
cs.RO / 52 / 2610.13403
Robot-WATCHDOG : failure detection through object-centric graph representation
Robot-WATCHDOG：通过面向对象的图表示进行故障检测
Abstract
Reliable real-time failure detection is critical for deploying learned robotic policies, yet defining failures explicitly is infeasible in open-world settings. This motivates failure detection from nominal demonstrations alone. Existing methods, however, struggle to balance robustness to benign variations with the ability to capture semantic and interaction-level errors. We introduce Robot-WATCHDOG, a self-supervised framework that models manipulation scenes as object-centric spatio-temporal graphs. Building on this representation, we investigate two complementary failure scores: Trajectory Projection and Correlation (TPC), a non-parametric measure of geometric deviation and relational consistency, and GnnT, a Graph Attention Network with a Transformer that predicts future object states. We evaluate both on a public benchmark and on a new dataset of semantically complex manipulation tasks, across six tasks and ten random seeds. Both scores outperform image-based failure detectors, particularly on semantic and interaction-level failures. TPC provides the strongest performance in AUPR and MCC on both dataset averages, with low variance across seeds. Qualitative results are made available at https://cea-list.github.io/Robot-WATCHDOG/.
Chinese Translation
可靠的实时故障检测对于部署学习的机器人策略至关重要，但在开放世界环境中明确定义故障是不可行的。这促使我们仅从正常演示中进行故障检测。然而，现有方法在应对良性变化的鲁棒性与捕捉语义和交互级别错误的能力之间难以取得平衡。我们提出了Robot-WATCHDOG，一个自监督框架，将操作场景建模为面向对象的时空图。在此表示的基础上，我们研究了两种互补的故障评分：轨迹投影与相关性（Trajectory Projection and Correlation，TPC），这是一种几何偏差和关系一致性的非参数度量，以及GnnT，一个结合了变换器（Transformer）的图注意力网络，用于预测未来的物体状态。我们在一个公共基准和一个新的语义复杂操作任务数据集上对这两种评分进行了评估，涵盖六个任务和十个随机种子。这两种评分在语义和交互级别的故障检测上均优于基于图像的故障检测器。TPC在两个数据集平均值的AUPR和MCC上表现最佳，且在不同种子间的方差较低。定性结果可在 https://cea-list.github.io/Robot-WATCHDOG/ 获取。
cs.RO / 53 / 2610.13409
ManeuverFlow: Maneuver Hypothesis Generation with Future Evidence for Trajectory Evaluation and Refinement in End-to-End Autonomous Driving
ManeuverFlow：基于未来证据的机动假设生成用于端到端自主驾驶中的轨迹评估与优化
Abstract
End-to-end planning in complex scenes must model diverse driving intentions and select the most appropriate trajectory using scene context and future risks. However, multi-query regression may collapse to one mode, while anchor-based and generative planners share a scaling bottleneck: small candidate sets limit maneuver coverage or discovery of high-quality trajectories, whereas large sets increase generation and scoring costs. We propose ManeuverFlow to generate trajectories for distinct intentions in one step and evaluate and refine them using current and future evidence. The model selects relevant maneuver modes, draws corresponding noise from the Gaussian mixture prior, and maps it to intention-specific trajectories through a maneuver-routed mixture-of-experts MeanFlow field. Current features and future features predicted by the world model are aggregated along each candidate path into trajectory-specific evidence for PDM-factor prediction. Using the same evidence, the refiner predicts local Frenet residuals for selected candidates; refined and retained candidates are then jointly re-scored for selection. With a lightweight TransFuser-style ResNet-34 backbone, ManeuverFlow achieves 90.6 PDMS and 90.1 EPDMS on NAVSIM v1/v2 navtest, respectively. On WorldEngine's recent 3DGS closed-loop benchmark, it attains the highest listed PDMS$^\ast$ scores of 75.19 and 74.58 under logged and IDM-controlled traffic, respectively. Code: https://github.com/wyh-autonomy/maneuverflow.
Chinese Translation
在复杂场景中的端到端规划必须建模多样的驾驶意图，并利用场景上下文和未来风险选择最合适的轨迹。然而，多查询回归可能会崩溃为单一模式，而基于锚点和生成式规划器则面临一个扩展瓶颈：小候选集限制了机动覆盖或高质量轨迹的发现，而大候选集则增加了生成和评分的成本。我们提出了ManeuverFlow，以一步生成针对不同意图的轨迹，并利用当前和未来证据对其进行评估和优化。该模型选择相关的机动模式，从高斯混合先验中提取相应的噪声，并通过机动路由的专家混合均值场映射到特定意图的轨迹。当前特征和世界模型预测的未来特征沿每个候选路径聚合成轨迹特定的证据，用于PDM因子的预测。利用相同的证据，优化器为选定的候选轨迹预测局部Frenet残差；经过优化并保留的候选轨迹随后被联合重新评分以进行选择。凭借轻量级的TransFuser风格ResNet-34骨干网络，ManeuverFlow在NAVSIM v1/v2 navtest上分别达到了90.6的PDMS和90.1的EPDMS。在WorldEngine最近的3DGS闭环基准测试中，它在记录和IDM控制的交通下分别获得了75.19和74.58的最高PDMS$^ullet$得分。代码： https://github.com/wyh-autonomy/maneuverflow.
cs.RO / 54 / 2610.13453
Reflect-VLA: Internalizing Reflective Guidance via Dual-Path On-Policy Self-Distillation for Autonomous Driving
Reflect-VLA：通过双路径在线自蒸馏内化反思指导以实现自主驾驶
Abstract
Reinforcement Learning (RL) substantially improves planning performance in autonomous-driving Vision-Language-Action (VLA) models. However, existing RL methods rely on scalar trajectory rewards that measure overall quality but fail to identify specific planning errors or provide actionable correction signals. In long-tail scenarios, errors in scene understanding, behavior selection, and trajectory execution are compressed into a single score. We observe that Reflective Guidance turns trajectory evaluation into explicit diagnoses and corrective instructions, substantially improving planning accuracy. Yet such guidance requires an initial prediction and its evaluation results, making it unavailable for direct single-pass inference. We propose Reflect-VLA, which treats Reflective Guidance as training-time privileged information and internalizes its planning benefits through Dual-Path On-Policy Self-Distillation (Dual-OPSD). Unlike scalar-reward optimization, Dual-OPSD uses the driving score only to route student-generated trajectories. It distills dense token-level corrective distributions from the teacher for risky trajectories while applying self-imitation to preserve reliable ones. To provide reliable corrective distributions on student rollouts, the teacher is progressively constructed through Driving-Knowledge Pretraining, Planning-and-Refinement SFT, and Self-Refinement Rejection Sampling. At inference time, Reflect-VLA generates trajectories in a single pass without explicit guidance or iterative refinement. Experiments on NAVSIM v1/v2 and Bench2Drive demonstrate competitive planning performance while reducing training time by over 90% compared with RL.
Chinese Translation
强化学习（RL）显著提高了自主驾驶视觉-语言-动作（VLA）模型的规划性能。然而，现有的RL方法依赖于标量轨迹奖励来衡量整体质量，但未能识别特定的规划错误或提供可操作的纠正信号。在长尾场景中，场景理解、行为选择和轨迹执行中的错误被压缩为一个单一的评分。我们观察到，反思指导将轨迹评估转化为明确的诊断和纠正指令，从而显著提高了规划的准确性。然而，这种指导需要初始预测及其评估结果，使其无法直接用于单次推理。我们提出了Reflect-VLA，将反思指导视为训练时的特权信息，并通过双路径在线自蒸馏（Dual-OPSD）内化其规划优势。与标量奖励优化不同，Dual-OPSD仅使用驾驶评分来引导学生生成的轨迹。它从教师那里提炼出密集的标记级纠正分布用于风险轨迹，同时应用自我模仿以保留可靠的轨迹。为了在学生的回滚中提供可靠的纠正分布，教师通过驾驶知识预训练、规划与精炼的监督微调（SFT）和自我精炼拒绝采样逐步构建。在推理时，Reflect-VLA能够在没有明确指导或迭代精炼的情况下生成轨迹。在NAVSIM v1/v2和Bench2Drive上的实验表明，与RL相比，规划性能具有竞争力，同时训练时间减少超过90%。
cs.RO / 55 / 2610.13516
RESCUE: Is This Alarm Worth Acting On? Counterfactual Evaluation of VLA Runtime Monitors
RESCUE：这个警报值得采取行动吗？VLA运行时监控器的反事实评估
Abstract
Runtime failure monitors for robot policies are typically evaluated by how accurately they distinguish failed from successful executions. Yet an accurate alarm may arrive too late for recovery, while an unnecessary intervention may disrupt an execution that would otherwise succeed. We introduce RESCUE, a counterfactual framework for evaluating whether a monitor's alarms enable useful interventions. By restoring simulator and policy states at execution checkpoints, RESCUE compares alternative actions, identifies the latest checkpoint satisfying a specified recovery criterion, and evaluates whether acting on alarms reduces expected task-completion cost. We evaluate RESCUE with OpenVLA-OFT and $\pi_{0.5}$ over LIBERO-Spatial and LIBERO-Plus using 128,340 counterfactual branches. Recoverability depends on the state--intervention pair rather than on robot state alone: at $\tau=0.5$, 14 of 66 OpenVLA-OFT/LIBERO-Plus failures contain a recoverable checkpoint under continuation, compared with 34 of 66 under continuation or scripted recovery, while recovery reduces recoverability in other conditions. On the primary strong-policy conditions, favourable detection scores do not establish lower intervention cost. Under the evaluated stop-and-reset model, even an idealised detector can afford only a 2.0--3.3% false-alarm rate before unnecessary resets outweigh the savings from stopping failures early. These results show that monitor value depends jointly on the intervention available and the workload in which it is deployed.
Chinese Translation
机器人策略的运行时故障监控器通常通过其区分失败与成功执行的准确性来评估。然而，准确的警报可能在恢复时已为时已晚，而不必要的干预可能会干扰本可以成功的执行。我们引入了RESCUE，一个反事实框架，用于评估监控器的警报是否能够实现有用的干预。通过在执行检查点恢复模拟器和策略状态，RESCUE比较替代行动，识别满足特定恢复标准的最新检查点，并评估对警报采取行动是否降低了预期的任务完成成本。我们使用128,340个反事实分支，结合OpenVLA-OFT和$ ext{π}_{0.5}$在LIBERO-Spatial和LIBERO-Plus上评估RESCUE。可恢复性依赖于状态-干预对，而不仅仅是机器人状态：在$ au=0.5$时，66个OpenVLA-OFT/LIBERO-Plus故障中有14个在继续执行下包含可恢复的检查点，而在继续执行或脚本恢复下则有34个，而在其他条件下恢复会降低可恢复性。在主要的强策略条件下，良好的检测得分并未确立较低的干预成本。在评估的停止-重置模型下，即使是理想化的检测器也只能承受2.0%至3.3%的误报率，否则不必要的重置将超过提前停止故障所带来的节省。这些结果表明，监控器的价值共同依赖于可用的干预和其部署的工作负载。
cs.RO / 56 / 2610.13522
Beyond Successful Robot Demonstrations: Learning from Human Interactions and Modeling Failure Dynamics
超越成功的机器人演示：从人类交互中学习与建模失败动态
Abstract
Action-conditioned world models offer a promising foundation for generalist robotic intelligence by predicting the consequences of actions. However, existing robotic datasets provide limited coverage of the diverse interaction dynamics required for generalizable world modeling. Large-scale egocentric human videos offer complementary interaction experience, yet the embodiment gap between human hands and robotic manipulators hinders effective cross-embodiment learning. Meanwhile, the predominance of successful demonstrations leaves failure dynamics substantially underrepresented, limiting the ability of world models to anticipate unsuccessful executions. To address these limitations, we introduce HF-WM, an action-conditioned world model that combines Human interaction learning with Failure-aware post-training. To bridge the embodiment gap, we develop a dual-branch framework that couples video generation with auxiliary pointflow prediction, using stabilized pointflow as a unified motion representation across human and robot interactions while suppressing egocentric camera motion. To improve failure modeling, we further perform Group Relative Policy Optimization (GRPO) post-training on perturbed actions derived from successful demonstrations, guided by action alignment and physical plausibility rewards, enabling the model to learn plausible failure dynamics without paired ground-truth videos. Extensive experiments demonstrate that HF-WM achieves state-of-the-art performance in action-conditioned video generation and exhibits strong utility for policy evaluation in downstream robotic tasks.
Chinese Translation
基于动作条件的世界模型通过预测动作的后果，为通用机器人智能提供了一个有前景的基础。然而，现有的机器人数据集在覆盖多样化交互动态方面有限，这对于可推广的世界建模至关重要。大规模的自我中心人类视频提供了互补的交互经验，但人类手部与机器人操控器之间的体现差距阻碍了有效的跨体现学习。同时，成功演示的主导地位使得失败动态在很大程度上被低估，限制了世界模型预测不成功执行的能力。为了解决这些局限性，我们提出了HF-WM，一个结合了人类交互学习与失败意识后训练的基于动作条件的世界模型。为了弥补体现差距，我们开发了一个双分支框架，将视频生成与辅助点流预测相结合，使用稳定的点流作为人类与机器人交互的统一运动表示，同时抑制自我中心相机运动。为了改善失败建模，我们进一步在来自成功演示的扰动动作上执行了群体相对策略优化（Group Relative Policy Optimization, GRPO）后训练，受动作对齐和物理合理性奖励的指导，使模型能够在没有配对真实视频的情况下学习合理的失败动态。大量实验表明，HF-WM在基于动作条件的视频生成方面达到了最先进的性能，并在下游机器人任务的策略评估中展现出强大的实用性。
cs.RO / 57 / 2610.13572
ATHLETE: Learning Reactive Humanoid Tennis via Trajectory-Guided Motion Matching
ATHLETE：通过轨迹引导的运动匹配学习反应性人形网球
Abstract
Humanoid tennis requires reactive whole-body coordination under rapidly changing ball dynamics, yet collecting task-specific motion-capture demonstrations that cover diverse approach states, contact locations, and timing conditions is costly. Inspired by how humans acquire new skills by combining reusable motor experience with task-specific demonstrations, we present ATHLETE, a framework for learning reactive humanoid tennis from generic locomotion data and ordinary tennis videos. ATHLETE combines LAFAN1 locomotion with video-derived striking motions and uses Trajectory-Guided Motion Matching (TGMM) to construct diverse task-conditioned whole-body references. These references are transformed into a reactive policy through physics-based teacher--student learning with online DAgger distillation, followed by PPO fine-tuning for the tennis task. The resulting policy responds directly to robot and ball observations without requiring reference-motion inputs at deployment. On a real humanoid robot, ATHLETE achieves forehand and backhand hit rates of $90.00\%$ and $86.67\%$, respectively, and enables multi-shot rallies with human players. These results demonstrate that ATHLETE can learn dynamic humanoid tennis skills from reusable locomotion data and video demonstrations. Result videos and code can be found at https://anonymous.4open.science/w/ATHLETE2027/.
Chinese Translation
人形网球要求在快速变化的球动态下进行反应性的全身协调，但收集涵盖多样化接近状态、接触位置和时机条件的任务特定动作捕捉示范是成本高昂的。受到人类通过结合可重用运动经验与任务特定示范来获取新技能的启发，我们提出了ATHLETE，一个从通用运动数据和普通网球视频中学习反应性人形网球的框架。ATHLETE结合了LAFAN1运动与视频衍生的击球动作，并使用轨迹引导的运动匹配（Trajectory-Guided Motion Matching, TGMM）构建多样化的任务条件全身参考。这些参考通过基于物理的师生学习与在线DAgger蒸馏转化为反应性策略，随后通过PPO微调以适应网球任务。在真实的人形机器人上，ATHLETE实现了正手和反手击球率分别为90.00%和86.67%，并使得与人类玩家进行多次击球的对抗成为可能。这些结果表明，ATHLETE能够从可重用的运动数据和视频示范中学习动态的人形网球技能。结果视频和代码可以在https://anonymous.4open.science/w/ATHLETE2027/找到。
cs.RO / 58 / 2610.13578
RARA: Robust Policy Fine-Tuning through Retrieval-Anchored Representation Alignment for Robotic Manipulation
RARA：通过检索锚定的表示对齐实现稳健的策略微调用于机器人操作
Abstract
Pre-trained robot policies, such as Vision-Language-Action (VLA) models, provide a strong foundation for generalist robotic manipulation, but adapting them to new tasks with only a few demonstrations remains challenging. Directly fine-tuning a pre-trained policy often achieves strong in-distribution performance, yet it can distort the representations learned during pre-training and weaken the model's generalization ability under out-of-distribution shifts. We propose Retrieval-Anchored Representation Alignment (RARA), a two-stage adaptation method, to address this issue. First, RARA retrieves representation features from pretraining with similar task intent and motion, and aligns the policy's representations of the target demonstrations with the pre-trained representations of the retrieved demonstrations. Second, RARA fine-tunes the policy on the target demonstrations while limiting how much its representations of the retrieved demonstrations change during fine-tuning. By reducing the representation mismatch before action learning and preserving pre-trained knowledge during adaptation, RARA limits unnecessary representation drift during fine-tuning. Experiments with Diffusion Policy and GR00T on 13 tasks in RoboTwin, RoboCasa, and on a real robot show that RARA improves the average success of fine-tuning by 9.5 points in-distribution and 13.6 points out-of-distribution, and outperforms all compared baselines. Project page: https://shua-kang.github.io/RARA_robot/
Chinese Translation
预训练的机器人策略，如视觉-语言-动作（Vision-Language-Action, VLA）模型，为通用机器人操作提供了坚实的基础，但仅通过少量示例将其适应于新任务仍然具有挑战性。直接微调预训练策略通常能够在同类分布上取得良好的性能，但可能会扭曲预训练期间学习到的表示，从而削弱模型在分布外变化下的泛化能力。我们提出了检索锚定的表示对齐（Retrieval-Anchored Representation Alignment, RARA），这是一种两阶段的适应方法，以解决这一问题。首先，RARA从具有相似任务意图和动作的预训练中检索表示特征，并将目标示例的策略表示与检索到的示例的预训练表示进行对齐。其次，RARA在目标示例上微调策略，同时限制在微调过程中检索到的示例的表示变化。通过在动作学习之前减少表示不匹配，并在适应过程中保留预训练知识，RARA限制了微调过程中不必要的表示漂移。在RoboTwin、RoboCasa上的13个任务以及在真实机器人上的实验表明，RARA在同类分布上提高了微调的平均成功率9.5个百分点，在分布外提高了13.6个百分点，并且优于所有对比基线。项目页面：https://shua-kang.github.io/RARA_robot/
cs.RO / 59 / 2610.13628
JitterVLA: A Replay Benchmark for Motion Quality in VLA-Driven Human-Robot Manipulation
JitterVLA：一种用于VLA驱动的人机操控运动质量的重放基准
Abstract
Vision-language-action (VLA) policies issue sparse targets whose arrival at the robot controller is delayed and irregular, so a reasonable predicted action can still produce abrupt motion near people. Remedies are evaluated on different streams, delays and metrics. JitterVLA replays identical commands and arrival traces through alternative controllers, isolating the controller's effect. We apply it to synthetic reaches, hand trajectories from 20 people, three open VLA pipelines (838 episodes) and a physical UR3. Interpolation lowered modeled p99 jerk for every source, by 83% (615 to 106 m/s^3) on human motion with 200 ms delay, at a 2.0 cm tracking cost, and the effect replicated on 598 new episodes; jerk-limited generation lowered jerk further at a larger cost. Entropy-discontinuity associations had opposite, replicated signs in two pipelines, and entropy gating helped in one but not the other. On the UR3, interpolation lowered accelerometer-derived jerk by 3.0-14.0% in all six exploratory pairs, while a position-derived metric did not resolve the contrast. JitterVLA supports screening before human-centred studies; code and data will be released.
Chinese Translation
视觉-语言-动作（VLA）策略发出稀疏目标，这些目标到达机器人控制器的时间延迟且不规律，因此合理预测的动作仍可能在靠近人类时产生突发运动。我们在不同的流、延迟和指标上评估了补救措施。JitterVLA通过替代控制器重放相同的命令和到达轨迹，从而隔离控制器的影响。我们将其应用于合成的到达、20个人的手部轨迹、三个开放的VLA管道（838个实验）和一个物理UR3。插值降低了每个来源的建模p99颤动，在200毫秒延迟下的人类运动中，颤动降低了83%（从615降至106 m/s^3），在2.0厘米的跟踪成本下，该效果在598个新实验中得到了复制；受颤动限制的生成在更大的成本下进一步降低了颤动。熵不连续性关联在两个管道中表现出相反的、可复制的符号，熵门控在一个管道中有所帮助，但在另一个管道中没有。在UR3上，插值在所有六对探索性实验中将加速度计导出的颤动降低了3.0-14.0%，而基于位置的指标未能解决这种对比。JitterVLA支持在以人为中心的研究之前进行筛选；代码和数据将被发布。
cs.RO / 60 / 2610.13643
AutoBio+: Benchmarking Dexterous Manipulation with Dynamic Experimental Phenomena in Realistic Laboratories
AutoBio+: 在真实实验室中基于动态实验现象的灵巧操作基准测试
Abstract
Automating biological laboratories requires robots to perform dexterous tool interactions under realistic visual conditions while handling dynamic experimental phenomena, presenting significant challenges for robotic manipulation. To enable systematic evaluation under these conditions, we introduce AutoBio+, a simulation benchmark and framework for dexterous manipulation in realistic biological laboratory environments. AutoBio+ comprises 14 tasks across 4 robot embodiments, spanning functional tool use, precision manipulation, language-conditioned visual grounding, and instrument operation. We introduce controllable dynamic experimental phenomena, including flames, vapor, and bubbles, to capture the evolving visual conditions of laboratory procedures. To further enhance environmental realism, we develop a hybrid rendering pipeline that integrates interactive simulated foreground objects with Gaussian-splatting backgrounds reconstructed from 6 real laboratory scenes. For efficient large-scale data generation and evaluation, we further build a parallel simulation framework on top of mjlab, proposing Parallel Finite-State Machine (PFSM) and integrating batched rendering with decoupled trajectory replay for demonstration generation and model-independent closed-loop policy evaluation. Our evaluation reveals substantial variation in policy performance across tasks, exposing distinct strengths and weaknesses of current manipulation policies. Real-world experiments further demonstrate a reduced sim-to-real gap, supporting the validity of AutoBio+ for realistic manipulation evaluation. Supplementary materials are available at https://autobio-plus.github.io/AutoBioPlus.
Chinese Translation
自动化生物实验室要求机器人在真实的视觉条件下执行灵巧的工具交互，同时处理动态实验现象，这对机器人操作提出了重大挑战。为在这些条件下进行系统评估，我们引入了AutoBio+，这是一个用于在真实生物实验室环境中进行灵巧操作的仿真基准和框架。AutoBio+包括4种机器人形态下的14个任务，涵盖功能性工具使用、精确操作、语言条件下的视觉定位和仪器操作。我们引入了可控的动态实验现象，包括火焰、蒸汽和气泡，以捕捉实验程序中不断变化的视觉条件。为了进一步增强环境的真实感，我们开发了一种混合渲染管道，将交互式仿真前景对象与从6个真实实验室场景重建的高斯散点背景相结合。为了高效地进行大规模数据生成和评估，我们在mjlab的基础上构建了一个并行仿真框架，提出了并行有限状态机（Parallel Finite-State Machine, PFSM），并将批量渲染与解耦轨迹重放相结合，以生成演示和模型无关的闭环策略评估。我们的评估揭示了不同任务之间政策性能的显著差异，暴露了当前操作政策的独特优势和劣势。现实世界实验进一步证明了减少了仿真与现实之间的差距，支持了AutoBio+在真实操作评估中的有效性。补充材料可在 https://autobio-plus.github.io/AutoBioPlus 获取。
cs.RO / 61 / 2610.13665
Cooperative Aerial Transportation of a Payload with a Sliding Cable Attachment
带滑动电缆连接的负载协作空中运输
Abstract
This paper presents a dual-quadrotor hook- suspended transport system for cooperative aerial manipulation of a payload that is not rigidly attached to a fixed point along the cable. Two quadrotors are connected by a single inextensible cable that passes through a frictionless hook mounted on the payload, allowing the cable to slide between its two segments. This introduces a cable-distribution variable that couples the quadrotor motion, cable geometry, and payload dynamics. A feasible reference-generation framework is formulated using the desired payload pose, cable distribution, and system shape while satisfying the cable-length constraint and elliptical workspace. The quadrotors track the geometrically generated trajectories while maintaining positive cable tension and respecting actuator limits. The payload pose is estimated from the system geometry, and its measured position is used only to validate the desired trajectory. The approach is evaluated through experimental demonstrations.
Chinese Translation
本文提出了一种双四旋翼钩悬挂运输系统，用于协作空中操作一个未刚性固定在电缆某一点的负载。两个四旋翼通过一根单一的不可伸缩电缆连接，该电缆穿过安装在负载上的无摩擦钩，从而允许电缆在其两个段之间滑动。这引入了一个电缆分布变量，该变量将四旋翼运动、电缆几何形状和负载动力学耦合在一起。基于所需的负载姿态、电缆分布和系统形状，制定了一个可行的参考生成框架，同时满足电缆长度约束和椭圆工作空间。四旋翼在跟踪几何生成的轨迹时，保持正向电缆张力并遵循执行器限制。负载姿态通过系统几何形状进行估计，其测量位置仅用于验证期望轨迹。该方法通过实验演示进行了评估。
cs.RO / 62 / 2610.13688
Towards a Unified Taxonomy of Agricultural Harvesting Technologies: Linking Physical Interaction, Automation, and Selectivity
迈向农业收获技术的统一分类法：连接物理交互、自动化与选择性
Abstract
Harvesting technologies have evolved from conventional mechanized systems based on bulk mechanical interaction toward increasingly automated and robotic systems capable of perception, decision-making, and localized actuation. This evolution has also created a classification problem, as harvesting systems are commonly described according to crop type, mechanical removal principle, sensing technology, robotic architecture, or degree of automation, making direct comparison difficult. This paper proposes a unified taxonomy for agricultural harvesting technologies based on two primary classification axes: the physical principle governing plant-machine interaction and the level of automation incorporated into the harvesting process. Selectivity is treated as an emergent functional attribute that expresses the biological and spatial scale at which perception and decision-making influence physical actuation. Conventional mechanized methods, including cutting, pulling, vibration, impact, brushing, pneumatic action, and combined systems, are analyzed together with semi-automated and robotic harvesting approaches. The analysis shows that increasing automation does not necessarily result in greater selectivity. A qualitative transition occurs when perception and decision-making directly influence product-level actuation, shifting harvesting from distributed or bulk interaction toward localized and selective manipulation. The proposed framework provides a common conceptual basis for positioning and comparing mechanized, automated, and robotic harvesting technologies. By separating the physical mechanism of product removal from the automation level and resulting selectivity, the taxonomy clarifies technological boundaries, identifies intermediate configurations, and supports the analysis and development of future harvesting systems.
Chinese Translation
收获技术已从基于大宗机械交互的传统机械化系统演变为越来越自动化和机器人化的系统，这些系统能够进行感知、决策和局部驱动。这一演变也带来了分类问题，因为收获系统通常根据作物类型、机械去除原理、传感技术、机器人架构或自动化程度进行描述，使得直接比较变得困难。本文提出了一种基于两个主要分类轴的农业收获技术统一分类法：支配植物与机器交互的物理原理和收获过程中所包含的自动化水平。选择性被视为一种新兴的功能属性，表达了感知和决策影响物理驱动的生物和空间尺度。传统机械化方法，包括切割、拉拔、振动、冲击、刷洗、气动作用和组合系统，与半自动化和机器人收获方法一起进行了分析。分析表明，增加自动化并不一定导致更高的选择性。当感知和决策直接影响产品级别的驱动时，会发生一种定性转变，使收获从分散或大宗交互转向局部和选择性操作。所提出的框架为定位和比较机械化、自动化和机器人收获技术提供了共同的概念基础。通过将产品去除的物理机制与自动化水平和由此产生的选择性分开，分类法澄清了技术边界，识别了中间配置，并支持未来收获系统的分析和开发。
cs.RO / 63 / 2610.13740
SmolWAM: A Latent World-Action Model for Robotic Manipulation on Edge Devices
SmolWAM：一种用于边缘设备的潜在世界-动作模型在机器人操控中的应用
Abstract
World-action models (WAMs) predict a robot's actions together with the observations those actions will produce. However, the strongest models rely on video diffusion transformers that are far too computationally expensive for resource-constrained edge devices. We present SmolWAM, a latent WAM that we derive from the 2B-parameter Cosmos Policy by keeping its pretrained trunk and rebuilding its inference interface: single-frame visual latents, cached text features that preserve the trunk's attention sinks, 12 of 28 transformer blocks, and 1,008 tokens from which actions and future visual latents are read out in one rectified-flow step. Trained in 3.6 GPU-hours, SmolWAM succeeds in 95.5% of 400 LIBERO episodes at 57 ms and 2.3 TFLOP per 16-action chunk with 2.4 GiB of GPU memory, 11x faster than its teacher with 18.5x fewer FLOPs and 3.2x less memory, and its world head still forecasts scene motion.
Chinese Translation
世界-动作模型（WAMs）预测机器人的动作以及这些动作将产生的观察结果。然而，最强大的模型依赖于视频扩散变换器，这对于资源受限的边缘设备来说计算开销过大。我们提出了SmolWAM，一种潜在的WAM，我们通过保留其预训练主干并重建其推理接口来从2B参数的Cosmos Policy中推导出该模型：单帧视觉潜变量、缓存文本特征以保留主干的注意力汇聚、28个变换器块中的12个，以及从中读取动作和未来视觉潜变量的1,008个标记，所有这些都在一个修正流步骤中完成。经过3.6 GPU小时的训练，SmolWAM在400个LIBERO实验中以57毫秒和每16个动作块2.3 TFLOP的速度成功率达到95.5%，其GPU内存使用为2.4 GiB，速度比其教师模型快11倍，FLOP数少18.5倍，内存使用量少3.2倍，并且其世界头仍然能够预测场景运动。
cs.RO / 64 / 2610.13788
When LiDAR Fades, Radar Takes Over: LiDAR Reliability-Aware Adaptive Localization Under Airborne Obscurants
当激光雷达失效时，雷达接管：基于激光雷达可靠性的自适应定位在空气遮挡物下的应用
Abstract
LiDAR-inertial odometry (LIO) is widely used for real-time localization and mapping, but its reliability can degrade substantially in subterranean environments affected by smoke, dust, and other airborne obscurants, whereas radar remains comparatively robust. This article presents a one-class temporal learning framework for LiDAR degradation detection and reliability-aware localization under such conditions. A broad set of degradation-sensitive cues, including spatial, cross-modal, geometric/motion, intensity, and measurement-recency features, is evaluated to characterize changes in LiDAR sensing quality. Their temporal evolution is modelled over sliding windows using an LSTM autoencoder trained exclusively on nominal operation, enabling deviations from healthy sensor behaviour to be detected without requiring labelled degradation data. The resulting LiDAR reliability estimate governs adaptive selection between LIO and radar inertial odometry (RIO), retaining LIO during reliable sensing and transferring localization authority to RIO when degradation is detected. The framework is evaluated using real-world ground robot data collected in GNSS-denied subterranean tunnels under nominal and dense-smoke conditions. Experimental results demonstrate degradation detection, adaptive modality switching, and continued localization output under severe LiDAR degradation. A feature family ablation study further shows that a reduced feature representation can improve degradation-aware localization performance in the evaluated dense-smoke scenarios, highlighting the importance of feature parsimony in one-class sensor degradation detection. The code will be released as open source to support reproducibility and further research.
Chinese Translation
激光雷达惯性里程计（LIO）广泛应用于实时定位和地图构建，但在受到烟雾、灰尘和其他空气遮挡物影响的地下环境中，其可靠性可能显著下降，而雷达则相对稳健。本文提出了一种一类时间学习框架，用于在此类条件下进行激光雷达退化检测和可靠性感知定位。评估了一系列对退化敏感的线索，包括空间、跨模态、几何/运动、强度和测量新近性特征，以表征激光雷达传感质量的变化。通过使用仅在正常操作下训练的LSTM自编码器，在滑动窗口上对其时间演变进行建模，从而能够在不需要标记退化数据的情况下检测到健康传感器行为的偏离。生成的激光雷达可靠性估计控制LIO和雷达惯性里程计（RIO）之间的自适应选择，在可靠传感时保留LIO，并在检测到退化时将定位权转移给RIO。该框架使用在GNSS失效的地下隧道中收集的真实地面机器人数据进行了评估，涵盖正常和浓烟条件。实验结果展示了退化检测、自适应模态切换以及在严重激光雷达退化下持续的定位输出。一项特征家族消融研究进一步表明，减少特征表示可以提高在评估的浓烟场景中的退化感知定位性能，突显了特征简约性在一类传感器退化检测中的重要性。代码将作为开源发布，以支持可重复性和进一步研究。
cs.RO / 65 / 2610.13816
Dream to Remember: Continual Steering of Generalist Robot Policies with World Models
梦境记忆：利用世界模型持续引导通用机器人策略
Abstract
Generalist robot policies are trained with imitation learning and are bounded by the quality and coverage of their demonstrations. Reinforcement learning can improve and extend them, either by fine-tuning the policy weights or by training a small module to steer the input noise of a frozen generative policy. When tasks arrive sequentially, both approaches suffer from catastrophic forgetting. Fine-tuning on a task degrades prior tasks, and steering only shifts forgetting to the shared steering module. We propose Dream to Remember (D2R), which continually steers a frozen generalist policy toward new tasks, while retaining its earlier skills. As the steering module learns a new task, a frozen pretrained world model rehearses prior tasks through imagined rollouts, anchoring the module to its earlier behavior without a replay buffer or new data. When the world model is faithful enough, D2R learns the new task entirely in imagination. Otherwise, it learns the task in the environment while using the world model only to rehearse prior tasks. On CALVIN, D2R improves eight tasks in sequence entirely in imagination, raising average success from 63.7% to 74.8%, while LoRA fine-tuning on the same sequence drops it to 11.7%. With a frozen off-the-shelf generalist policy and world model, $\pi_0$ and A2World, D2R learns unseen LIBERO-90 tasks in the environment while keeping $\pi_0$'s performance on the LIBERO suites it was fine-tuned on, even though the world model is not accurate enough for policy optimization. Finally, on a real Franka robot, D2R improves three tasks in sequence entirely in imagination while retaining the earlier improvements.
Chinese Translation
通用机器人策略通过模仿学习进行训练，其性能受到演示质量和覆盖范围的限制。强化学习可以通过微调策略权重或训练一个小模块来引导冻结生成策略的输入噪声，从而改善和扩展这些策略。当任务顺序到达时，这两种方法都会遭遇灾难性遗忘。对某一任务的微调会降低先前任务的性能，而引导仅仅是将遗忘转移到共享的引导模块上。我们提出了梦境记忆（Dream to Remember, D2R），该方法持续引导冻结的通用策略朝向新任务，同时保留其早期技能。当引导模块学习新任务时，一个冻结的预训练世界模型通过想象的回放演练先前任务，将模块锚定到其早期行为，而无需重放缓冲区或新数据。当世界模型足够忠实时，D2R完全在想象中学习新任务。否则，它在环境中学习任务，同时仅使用世界模型来演练先前任务。在CALVIN上，D2R在想象中连续改善八个任务，将平均成功率从63.7%提高到74.8%，而在同一序列上进行LoRA微调则将成功率降至11.7%。借助一个冻结的现成通用策略和世界模型，$ ext{π}_0$和A2World，D2R在环境中学习未见过的LIBERO-90任务，同时保持$ ext{π}_0$在其微调过的LIBERO套件上的表现，尽管世界模型在策略优化上并不够准确。最后，在真实的Franka机器人上，D2R在想象中连续改善三个任务，同时保留早期的改进。
cs.RO / 66 / 2610.13829
MC-EvoBench: Auditing Experience-Driven Self-Improvement in Minecraft Agents
MC-EvoBench：审计以经验驱动的Minecraft代理自我改进
Abstract
A central challenge in embodied recursive self-improvement is enabling agents to convert interaction history into capabilities that persist beyond one world and transfer to later tasks. A saved memory or successful final action does not by itself establish that a reusable capability was acquired; evaluation must separate capability change from carried-over resources, unequal source exposure, and early execution failures. We construct MC-EvoBench as a controlled Minecraft benchmark with a fixed catalog of 19 registered evolution chains, including 12 primary evaluation chains and seven calibration or probe chains, together with three graph-aligned paired probe families for transfer. The benchmark represents each task as a task-state graph, defines stage-level environment verifiers, assigns matched source and target roles, resets the target world, and permits only a declared agent artifact to cross the source-target boundary. For each probe, no-prior, related, unrelated, and shuffled source conditions are evaluated under matched environment-step budgets and five seeds; we report TSR, WGP, and VMP and retain provenance records. The benchmark reveals a consistent direct-capability gap between controller and planner-backed variants, while related experience yields no stable transfer: it does not improve either tool probe, its small wood-species gain hinges on one seed, and all intervals include zero. Stage audits further show that early wood-collection and crafting bottlenecks often leave later capabilities unobserved. MC-EvoBench thus offers a controlled testbed for durable, transferable embodied capabilities. Code and data are available at https://github.com/xzw-ustc/mc-bench-open-source
Chinese Translation
在具身递归自我改进中，一个核心挑战是使代理能够将交互历史转化为超越单一世界并可转移到后续任务的能力。仅仅保存的记忆或成功的最终行动并不能证明可重用能力的获得；评估必须将能力变化与延续资源、不平等的源暴露和早期执行失败区分开来。我们构建了MC-EvoBench，作为一个控制的Minecraft基准，包含19个注册的进化链的固定目录，其中包括12个主要评估链和7个校准或探测链，以及三个与图形对齐的配对探测家族以便于转移。该基准将每个任务表示为任务状态图，定义阶段级环境验证器，分配匹配的源和目标角色，重置目标世界，并仅允许声明的代理工件跨越源-目标边界。对于每个探测，评估在匹配的环境步骤预算和五个种子下进行无先前、相关、不相关和打乱的源条件；我们报告TSR、WGP和VMP，并保留来源记录。基准揭示了控制器和规划器支持变体之间的一致直接能力差距，而相关经验并未产生稳定的转移：它没有改善任何工具探测，其小的木材种类增益依赖于一个种子，所有区间均包含零。阶段审计进一步表明，早期的木材收集和制作瓶颈往往使得后续能力未被观察到。因此，MC-EvoBench为持久、可转移的具身能力提供了一个控制测试平台。代码和数据可在 https://github.com/xzw-ustc/mc-bench-open-source 获取。
cs.RO / 67 / 2610.13838
Mapless, Stabilizing, and Anti-Stuck Flipper Control for Articulated Tracked Robots
无地图、稳定且防卡滞的关节履带机器人翻转控制
Abstract
Nearly every autonomous or semi-autonomous flipper controller assumes access to an elevation map of the terrain under and in front of the robot. The terrain beneath the chassis is occluded in the sensors, so the map is remembered through odometry integration, whose drift corrupts it. We present Stabilizing Anti-Stuck (SAS), a mapless controller that combines a stabilization term for body roll and pitch with an anti-stuck term that frees the robot when it gets stuck. As input, SAS uses the IMU, the flipper joint angles, and a motion cue that only tells whether the robot is moving. The cue is computed over a short window and never accumulated, so SAS does not inherit the odometry-integration drift that corrupts the map. We deploy SAS on a tracked robot with four flippers and evaluate it against manual teleoperation and prior flipper controllers in a large comparison, both in simulation and on a real obstacle. In simulation and on the real robot, SAS achieves traversal quality comparable to state-of-the-art map-based controllers. On a log hidden in tall grass, invisible to the LiDAR, the best map-based controller gets stuck while SAS traverses the hidden obstacle without a problem.
Chinese Translation
几乎所有自主或半自主的翻转控制器都假设可以访问机器人下方和前方地形的高程地图。然而，底盘下方的地形在传感器中是被遮挡的，因此地图是通过里程计积分来记忆的，而这种积分的漂移会导致地图的失真。我们提出了稳定防卡滞控制器（Stabilizing Anti-Stuck, SAS），这是一种无地图控制器，它结合了用于机身滚转和俯仰的稳定项与在机器人被卡住时解放机器人的防卡滞项。SAS的输入包括惯性测量单元（IMU）、翻转关节角度以及一个仅能判断机器人是否在移动的运动提示。该提示在一个短时间窗口内计算，且不进行累积，因此SAS不会继承导致地图失真的里程计积分漂移。我们在一款配备四个翻转器的履带机器人上部署了SAS，并在大规模比较中将其与手动遥控和先前的翻转控制器进行了评估，比较内容包括仿真和真实障碍物。在仿真和真实机器人上，SAS实现了与最先进的基于地图的控制器相当的行驶质量。在一根隐藏在高草中的木头上，该木头对激光雷达（LiDAR）不可见，最佳的基于地图的控制器被卡住，而SAS则顺利通过了这个隐藏的障碍。
cs.RO / 68 / 2610.13849
HorizonNav: Geometry-Guided Video Diffusion for Anticipatory 3D Robot Navigation
HorizonNav：基于几何引导的视频扩散用于预期的3D机器人导航
Abstract
We present HorizonNav, a video diffusion-based framework for anticipatory robot navigation that transforms a single RGB image and a text prompt into a geometrically aligned representation of plausible surrounding 3D space. Given the initial observation and a textual description of the reconstruction objective, a vision-language model, guided by designed instructions, interprets the scene and constructs structured prompts for a video diffusion model. The resulting video sequence provides plausible future observations of the environment, which are subsequently processed using monocular ORB-SLAM3 for camera pose estimation and Video Depth Anything for dense depth prediction. Our central contribution is a geometry-guided depth alignment stage that combines the estimated camera poses with triangulated visual landmarks to constrain dense depth predictions across multiple views. This alignment improves the geometric consistency of the reconstructed 3D environment, addressing inconsistencies that temporal coherence of the generated video alone cannot resolve. The experimental evaluation of navigation in imagined 3D environment shows that across a 19-scene horizon study, planning-feasible scenes increase from 13/19 at P20 to 15/19 at P81, while successful RRT trials increase from 62/90 to 75/90. In a separate corridor study, 6 of 25 evaluable generated-camera trajectories violate the reconstructed collision constraints; RRT planner recovers feasible alternatives in 5 of those 6 cases. A larger archive contains 58 usable reconstructions from 67 attempted cases, of which 36 support a downstream plan and 22 do not, exposing reconstruction and free-space failure modes rather than hiding them without lidar. Across 15 corridor reconstructions, the mean reconstruction error is approximately 0.30 m at matched structural landmarks.
Chinese Translation
我们提出了HorizonNav，一个基于视频扩散的框架，用于预期的机器人导航，该框架将单个RGB图像和文本提示转化为几何对齐的可行周围3D空间的表示。给定初始观察和重建目标的文本描述，一个视觉-语言模型在设计的指令指导下，解释场景并为视频扩散模型构建结构化提示。生成的视频序列提供了环境的可行未来观察，这些观察随后通过单目ORB-SLAM3进行相机位姿估计，并使用Video Depth Anything进行密集深度预测。我们的核心贡献是一个几何引导的深度对齐阶段，该阶段将估计的相机位姿与三角测量的视觉地标结合，以限制多个视角下的密集深度预测。这种对齐改善了重建3D环境的几何一致性，解决了仅靠生成视频的时间一致性无法解决的不一致性。对想象的3D环境中导航的实验评估表明，在19场景的研究中，规划可行场景从P20的13/19增加到P81的15/19，而成功的RRT试验从62/90增加到75/90。在一个独立的走廊研究中，25条可评估的生成相机轨迹中有6条违反了重建的碰撞约束；RRT规划器在这6种情况下恢复了5种可行的替代方案。一个更大的档案包含67个尝试案例中的58个可用重建，其中36个支持下游计划，22个不支持，暴露了重建和自由空间的失败模式，而不是在没有激光雷达的情况下隐藏它们。在15个走廊重建中，匹配结构地标的平均重建误差约为0.30米。
cs.RO / 69 / 2610.13862
BT-Agent: Deploying Onboard Robot Agents with Behavior Trees and Multi-Fidelity Evaluation
BT-Agent：基于行为树和多保真度评估的机载机器人代理部署
Abstract
Deploying language-model agents onboard mobile robots requires task reasoning and responsive execution within limited compute and memory. We propose BT-Agent, an onboard architecture combining LLM-driven task decisions with predefined behavior-tree (BT) skills. These skills enforce execution conditions and return structured failure feedback. We demonstrate fully onboard operation with a 9B model on a quadruped across five task categories. Intervention tests show that local checks stop ongoing navigation without waiting for another model decision, while the agent remains active to retry or report constraints. To compare agent designs efficiently, we introduce a 161-task multi-fidelity benchmark, evaluated in text on all tasks and in Habitat on 39. Text supports 10.9$\times$ higher evaluation throughput, while simulation reveals how perception and motion failures change task outcomes. BT-Agent's largest success gains over the strongest Act/ReAct baseline occur at 4B and 9B: approximately 10 and 5 percentage points. Ablations identify structured tool calls and informative failure feedback as the most useful components.
Chinese Translation
在移动机器人上部署语言模型代理需要在有限的计算和内存条件下进行任务推理和响应执行。我们提出了BT-Agent，这是一种机载架构，结合了基于大语言模型（LLM）的任务决策与预定义的行为树（BT）技能。这些技能强制执行条件并返回结构化的失败反馈。我们在一款四足机器人上展示了使用9B模型的完全机载操作，涵盖五个任务类别。干预测试表明，本地检查可以在不等待其他模型决策的情况下停止正在进行的导航，同时代理仍然保持活跃状态以重试或报告约束。为了高效比较代理设计，我们引入了一个包含161个任务的多保真度基准，在所有任务上进行文本评估，并在39个任务上使用Habitat进行评估。文本评估支持10.9倍更高的评估吞吐量，而仿真则揭示了感知和运动失败如何改变任务结果。BT-Agent在4B和9B模型上相较于最强的Act/ReAct基线获得了最大的成功增益，分别约为10和5个百分点。消融实验表明，结构化工具调用和信息丰富的失败反馈是最有用的组成部分。
cs.RO / 70 / 2610.13929
Chronos Policy: Learning Robot Policies on Physical Time
Chronos策略：在物理时间上学习机器人策略
Abstract
Robot policies often convert sensing and control into fixed-rate observation-action sequences, leaving physical time implicit in sequence position. This assumption breaks when observation rates or delays vary, or when inference latency changes. We introduce Chronos Policy, which makes physical time explicit by conditioning on observation capture times and desired action execution times. We study two ways to incorporate time into a diffusion transformer: adding timestamp embeddings to token features, or encoding physical time in positional encoding through time-aware rotary positional encodings (time-aware RoPE). We find that time-aware RoPE is more robust, allowing attention to depend directly on physical time while keeping token content unchanged. During training, we temporally randomize which observations are sampled and when the action chunk begins, exposing the policy to a wider range of sensing and execution schedules that may occur during deployments. This also enables learning from asynchronous, multi-rate data without forcing all measurements onto a fixed temporal grid. Across simulation and real-world experiments, Chronos Policy improves performance on dynamic tasks, where observation-to-action timing is critical, and on multisensory tasks, where the relative timing between observations matters. On standard quasi-static tasks, it preserves nominal performance while improving robustness to changes in sensing and inference timing.
Chinese Translation
机器人策略通常将感知和控制转换为固定速率的观察-动作序列，使物理时间在序列位置中隐含。当观察速率或延迟变化，或推理延迟发生变化时，这一假设会失效。我们提出了Chronos策略，通过对观察捕获时间和期望动作执行时间进行条件化，使物理时间显式化。我们研究了将时间纳入扩散变换器的两种方法：将时间戳嵌入添加到标记特征中，或通过时间感知旋转位置编码（time-aware RoPE）在位置编码中编码物理时间。我们发现，时间感知RoPE更具鲁棒性，允许注意力直接依赖于物理时间，同时保持标记内容不变。在训练过程中，我们随机化哪些观察被采样以及动作块何时开始，从而使策略接触到在部署过程中可能出现的更广泛的感知和执行时间表。这也使得从异步、多速率数据中学习成为可能，而无需将所有测量强制到固定的时间网格上。在模拟和现实世界实验中，Chronos策略在动态任务上提高了性能，其中观察到动作的时机至关重要；在多感官任务中，观察之间的相对时机也很重要。在标准的准静态任务中，它保持了名义性能，同时提高了对感知和推理时机变化的鲁棒性。
cs.RO / 71 / 2610.13943
Noise Space Learning for VLA Adaptation
噪声空间学习用于视觉-语言-动作（VLA）适应
Abstract
Pretrained vision-language-action (VLA) models encode broad manipulation priors, yet adapting them to specific tasks requires balancing limited robot interaction, computational resources, and safety constraints. Noise space learning (NSL) adapts flow-matching-based VLAs by learning a context-conditioned distribution over noise while keeping the pretrained VLA fixed. Building on prior noise-steering methods, we formulate NSL as an interface for VLA adaptation that accommodates different learning algorithms and instantiate it with PPO, SAC, and AWR. Theoretically, we provide an information-theoretic analysis of how NSL enables efficient adaptation. Empirically, we study these NSL algorithms on LIBERO, RoboTwin, and four real-robot manipulation tasks under interaction, computation, and execution constraints. Across these settings, we develop a suite of effective NSL variants and derive practical recommendations for choosing among them under different resource and deployment regimes.
Chinese Translation
预训练的视觉-语言-动作（VLA）模型编码了广泛的操作先验，但将其适应于特定任务需要平衡有限的机器人交互、计算资源和安全约束。噪声空间学习（NSL）通过学习一个基于上下文的噪声分布来适应基于流匹配的VLA，同时保持预训练的VLA不变。在先前的噪声引导方法基础上，我们将NSL构建为一个VLA适应的接口，能够容纳不同的学习算法，并使用PPO、SAC和AWR进行实例化。从理论上讲，我们提供了NSL如何实现高效适应的信息论分析。从经验上看，我们在LIBERO、RoboTwin以及四个实际机器人操作任务中研究这些NSL算法，考虑到交互、计算和执行的约束。在这些设置中，我们开发了一系列有效的NSL变体，并提出了在不同资源和部署模式下选择它们的实用建议。
cs.RO / 72 / 2610.13945
Splat2Act: From Captured Scenes to Personalized VLA Manipulation
Splat2Act：从捕获场景到个性化的视觉-语言-动作（VLA）操控
Abstract
A user who brings a generalist manipulation policy into their home or office wants it to pick up their objects from their shelves, counters, and tables. Yet even a strong generalist like $\pi_{0.5}$ achieves 0% pickup success in our captured deployment scenes without adaptation, and collecting real demonstrations in every user's environment is prohibitively expensive. We present Splat2Act, an automated real-to-sim-to-real pipeline that personalizes a pretrained vision-language-action (VLA) policy to a specific scene and object set using only a handheld phone scan and images of the user's objects. Unlike prior real-to-sim work that reconstructs individual tabletops, Splat2Act operates on building-scale indoor environments: it builds a photorealistic digital twin with 3D Gaussian Splatting and collision geometry, automatically discovers tabletop and non-tabletop receptacles, constructs feasible pickup tasks, and generates successful demonstrations with a privileged simulation oracle, with only scene and object capture required from a human. Applied to 101 scanned campus environments with a median navigable area of 104.87 m$^2$ and 7 receptacles, the pipeline yields 852 receptacles and over 68,000 pickup demonstrations. We fine-tune $\pi_{0.5}$ in two stages: Stage 1 adapts to hundreds of public object assets, drawn either from the deployment scene alone or from all 101 scenes; Stage 2 personalizes to the user's own objects on the deployment scene's receptacles. Our strongest personalized policy reaches 62.5% aggregate pickup success in the target scene. Deployed on a real Stretch 3, the same policy achieves 63.9% success on a 12-task subset, closely tracking its 66.7% simulated success on those tasks. These results suggest that automated real-to-sim-to-real training can personalize VLA manipulation policies using only simulation-oracle data.
Chinese Translation
用户希望将通用操控策略引入他们的家或办公室，以便从架子、台面和桌子上拾取物品。然而，即使是像 $ ext{π}_{0.5}$ 这样的强大通用策略，在我们的捕获部署场景中没有适应时，其拾取成功率为0%。在每个用户的环境中收集真实演示是极其昂贵的。我们提出了 Splat2Act，这是一种自动化的真实-模拟-真实（real-to-sim-to-real）管道，它利用仅需手持手机扫描和用户物品图像的方式，将预训练的视觉-语言-动作（VLA）策略个性化到特定场景和物品集合。与之前重建单个桌面的真实-模拟工作不同，Splat2Act 在建筑规模的室内环境中运行：它使用3D高斯溅射和碰撞几何构建一个逼真的数字双胞胎，自动发现桌面和非桌面容器，构建可行的拾取任务，并利用特权模拟神谕生成成功的演示，仅需人类提供场景和物品捕获。该管道应用于101个扫描的校园环境，具有104.87 m$^2$的中位可导航区域和7个容器，产生了852个容器和超过68,000个拾取演示。我们在两个阶段对 $ ext{π}_{0.5}$ 进行微调：第一阶段适应于数百个公共物品资产，这些资产可以仅来自部署场景或所有101个场景；第二阶段则针对用户在部署场景容器上的自有物品进行个性化。我们最强的个性化策略在目标场景中达到了62.5%的整体拾取成功率。在真实的 Stretch 3 上部署时，同一策略在12个任务子集上达到了63.9%的成功率，紧密跟踪其在这些任务上的66.7%的模拟成功率。这些结果表明，自动化的真实-模拟-真实训练可以仅使用模拟神谕数据来个性化 VLA 操控策略。
cs.CV / 1 / 2610.12514
CORAL-KV: Compact Cross-Layer Prediction and Residual Coding for Autoregressive Video KV Caches
CORAL-KV：用于自回归视频KV缓存的紧凑跨层预测与残差编码
Abstract
Autoregressive video diffusion models maintain a growing attention key-value (KV) cache, making memory a key bottleneck for long video generation. Recently, clustering-based quantization has emerged as an effective method to compress video KV caches by grouping similar key and value vectors then quantizing residual errors. However, when later frames attend to the compressed history, small reconstruction errors can compound as the video grows. Beyond the within-layer similarity exploited by clustering, we find strong dependencies across cache layers in video models. This motivates Coral-KV - Calibrated Offline, Residual-coded, and Adapted Locally - a KV-cache compression framework that uses cache states in each layer to predict keys and values in subsequent layers, then quantizes the remaining prediction residuals. We construct compact block-diagonal-plus-low-rank (block-DPLR) predictors: block-diagonal components model channel relationships within each attention head, while a global low-rank term models dependencies between heads. Coral-KV exposes a flexible fidelity-memory trade-off: it can reuse its predictors across videos or refit lightweight block-diagonal predictors to the current video for higher fidelity. Across extensive evaluations on both HY-WorldPlay and MAGI-1, our reusable configuration improves PSNR over QVG-4 by 0.32 dB and 4.95 dB, respectively, while using 48% and 37% less cache memory. Against QVG-Pro, our video-adapted configuration improves MAGI-1 PSNR by 3.77 dB while using 33.7% less cache memory. Coral-KV also maintains higher reconstruction fidelity over extended generation and reduces temporal discontinuities while retaining comparable generation latency.
Chinese Translation
自回归视频扩散模型维护着一个不断增长的注意力键值（KV）缓存，使得内存成为长视频生成的关键瓶颈。最近，基于聚类的量化方法已成为压缩视频KV缓存的有效手段，通过将相似的键和值向量进行分组，然后量化残差误差。然而，当后续帧关注于压缩的历史时，随着视频的增长，小的重建误差可能会累积。除了聚类所利用的层内相似性外，我们发现视频模型中缓存层之间存在强依赖关系。这促使我们提出了Coral-KV——经过校准的离线、残差编码和局部适应的KV缓存压缩框架，该框架利用每层的缓存状态来预测后续层中的键和值，然后量化剩余的预测残差。我们构建了紧凑的块对角加低秩（block-DPLR）预测器：块对角成分建模每个注意力头内的通道关系，而全局低秩项建模头之间的依赖关系。Coral-KV展现了灵活的保真度-内存权衡：它可以在视频之间重用其预测器，或将轻量级块对角预测器重新调整到当前视频以获得更高的保真度。在对HY-WorldPlay和MAGI-1的广泛评估中，我们的可重用配置在使用48%和37%更少的缓存内存的同时，分别提高了PSNR值0.32 dB和4.95 dB。与QVG-Pro相比，我们的视频适应配置在使用33.7%更少的缓存内存的情况下，将MAGI-1的PSNR提高了3.77 dB。Coral-KV还在延长生成过程中保持了更高的重建保真度，并减少了时间不连续性，同时保持了可比的生成延迟。
cs.CV / 2 / 2610.12516
TraceICL: Thread-Resolved Demonstration Curation with Consistency Calibration for Robust Multimodal In-Context Learning
TraceICL：具有一致性校准的线程解析演示策划，用于稳健的多模态上下文学习
Abstract
Interleaved multimodal in-context learning (ICL) lets frozen large vision-language models (LVLMs) acquire new tasks from a handful of image-text-answer demonstrations, yet it remains fragile: demonstrations bind to the wrong images, redundant exemplars crowd out informative ones, and noisy answers are copied verbatim into predictions. We present AVA, a training-free framework that organizes the demonstration pipeline of multimodal ICL into three composable stages. A Thread-Resolved Decomposer (TRD) segments the interleaved context into per-sample threads and re-stitches them with explicit boundaries, restoring image-text binding. An Influence-scored Demonstration Selector (IDS) estimates each candidate's marginal influence with one extra forward pass, removes redundancy, and orders the survivors by descending value. A Weak-to-Strong Consistency Calibration (WSC) module scores each demonstration by a weak sibling model's reconstruction confidence and softly reweights its contribution during decoding. On five vision-language benchmarks with interleaved 4-shot prompts, AVA improves the strongest same-setting baseline by +1.7 points on average, with gains of up to +2.4 points on VizWiz, shrinks the require-visual performance gap on ScienceQA-IMG, and retains most of its gain under 30% noisy demonstrations, at a 26% latency overhead that an offline cache amortizes. No gradients are ever computed; the LVLM stays frozen.
Chinese Translation
交错的多模态上下文学习（ICL）使得冻结的大型视觉-语言模型（LVLMs）能够从少量的图像-文本-答案演示中获取新任务，但仍然脆弱：演示绑定到错误的图像，冗余示例挤占了信息丰富的示例，而嘈杂的答案被逐字复制到预测中。我们提出了AVA，一个无训练的框架，将多模态ICL的演示管道组织为三个可组合的阶段。线程解析分解器（TRD）将交错的上下文分割为每个样本的线程，并用明确的边界重新缝合它们，恢复图像-文本的绑定。影响评分演示选择器（IDS）通过一次额外的前向传播估计每个候选的边际影响，去除冗余，并按降序排列幸存者的价值。弱到强一致性校准（WSC）模块通过弱兄弟模型的重构置信度对每个演示进行评分，并在解码过程中柔性地重新加权其贡献。在五个具有交错4-shot提示的视觉-语言基准上，AVA在同一设置下的最强基线平均提高了1.7分，VizWiz的增益高达2.4分，缩小了ScienceQA-IMG上的视觉性能差距，并在30%的嘈杂演示下保留了大部分增益，且具有26%的延迟开销，离线缓存可以摊销这一开销。没有计算任何梯度；LVLM保持冻结状态。
cs.CV / 3 / 2610.12517
TerraAgent: Demonstration-Grounded Multi-Agent Reasoning for Long-Context Remote Sensing Visual Question Answering
TerraAgent：基于演示的多智能体推理用于长上下文遥感视觉问答
Abstract
Remote sensing visual question answering (RS-VQA) requires answers that truly depend on image evidence, yet current remote sensing multimodal large language models (RS MLLMs) are still evaluated mainly by accuracy, which cannot reveal whether a correct answer was produced without using the image. This setting introduces three challenges: demonstrations retrieved across different ground sample distances (GSD), sensors, and city textures may be poorly matched; real queries often involve long visual contexts from multi-scale tiles, multi-temporal frames, and retrieved knowledge, where evidence can be diluted and errors can propagate through multi-step reasoning; and prior-only answering remains invisible to standard metrics. We present AVA, a training-free framework that wraps a frozen RS MLLM with three modules. GDR performs GSD-bucketed, scene-graph-based demonstration retrieval and filters demonstrations whose answers are unsupported by their images. MACE decomposes each query into an evidence-typed sub-question DAG, routes sub-questions to suitable perception tools, and records verified provenance in a shared evidence ledger with bounded re-perception. CVA audits the final answer through counterfactual perturbations of cited evidence, including ablation and swap, and computes a visual-dependency score for confidence calibration and evidence-forcing re-query. On RSVQA-LR/HR, VRSBench, and LHRS-Bench, AVA improves average accuracy by up to 2.40 points over the strongest same-setting baselines, raises the visual-dependency score from 0.28 to 0.49, reduces the prior-only answer rate from 46.2 percent to 21.7 percent, and lowers the hallucination rate from 17.1 percent to 9.8 percent, with 2.6x the cost of a single backbone forward pass and no fine-tuning.
Chinese Translation
遥感视觉问答（RS-VQA）要求答案真正依赖于图像证据，然而当前的遥感多模态大语言模型（RS MLLMs）仍主要通过准确性进行评估，这无法揭示在未使用图像的情况下是否产生了正确答案。这种设置引入了三个挑战：跨不同地面样本距离（GSD）、传感器和城市纹理检索的演示可能匹配不佳；真实查询通常涉及来自多尺度切片、多时间帧和检索知识的长视觉上下文，其中证据可能被稀释，错误可能通过多步推理传播；而仅依赖先验的回答对标准指标来说是不可见的。我们提出了AVA，这是一个无训练的框架，它将一个冻结的RS MLLM与三个模块结合在一起。GDR执行基于GSD分桶的场景图演示检索，并过滤那些其答案未被图像支持的演示。MACE将每个查询分解为证据类型的子问题有向无环图（DAG），将子问题路由到合适的感知工具，并在共享证据账本中记录经过验证的来源，确保有限的再感知。CVA通过对引用证据的反事实扰动进行审计，包括消融和交换，计算视觉依赖分数以进行信心校准和证据强制再查询。在RSVQA-LR/HR、VRSBench和LHRS-Bench上，AVA在最强同设置基线之上提高了平均准确性最多2.40分，将视觉依赖分数从0.28提高到0.49，将仅依赖先验的回答率从46.2%降低到21.7%，并将幻觉率从17.1%降低到9.8%，其成本为单个主干前向传递的2.6倍且无需微调。
cs.CV / 4 / 2610.12568
MIRAGE: Metric Inference and Reliability-Aware Gaussian Estimation from Feed-Forward Vision
MIRAGE：基于前馈视觉的度量推断与可靠性感知高斯估计
Abstract
Feed-forward 3D priors enable high-quality reconstruction from image streams but often lack reliable metric scale and suffer from long-term trajectory drift. In contrast, LiDAR--inertial estimation provides metrically accurate motion but lacks the image-space consistency needed for high-fidelity reconstruction. We propose $MIRAGE$, which anchors feed-forward visual geometry to LiDAR--IMU metric motion and regulates online pose corrections according to correspondence support and cross-modal agreement, jointly enabling accurate metric pose estimation and high-fidelity Gaussian reconstruction. Experiments on Oxford Spires, M2DGR and INS show that MIRAGE achieves accurate metric trajectory estimation, outperforming COLMAP, a widely used but computationally expensive reference for vision-only benchmarks. Motivated by the scarcity of long-horizon monocular benchmarks for large-scale indoor reconstruction, we further build a multimodal data-collection platform and use MIRAGE to establish accurate 6-DoF reference trajectories for a new vision-only benchmark. We benchmark recent monocular large-scale reconstruction methods on this dataset, providing a systematic evaluation of their performance in long-horizon indoor scenes.
Chinese Translation
前馈3D先验能够从图像流中实现高质量重建，但通常缺乏可靠的度量尺度，并且会遭受长期轨迹漂移。相比之下，LiDAR-惯性估计提供了度量上准确的运动，但缺乏高保真重建所需的图像空间一致性。我们提出了MIRAGE，它将前馈视觉几何锚定到LiDAR-IMU度量运动，并根据对应支持和跨模态一致性调节在线姿态修正，从而共同实现准确的度量姿态估计和高保真的高斯重建。在Oxford Spires、M2DGR和INS上的实验表明，MIRAGE实现了准确的度量轨迹估计，优于COLMAP，这是一种广泛使用但计算成本高的视觉基准参考。鉴于大规模室内重建的长时间单目基准稀缺，我们进一步建立了一个多模态数据收集平台，并使用MIRAGE建立准确的6自由度参考轨迹，以便为新的视觉基准提供支持。我们在该数据集上对最近的单目大规模重建方法进行了基准测试，系统评估了它们在长时间室内场景中的性能。
cs.CV / 5 / 2610.12574
Scaling Multi-Teacher Distillation for Digital Pathology
扩展数字病理学中的多教师蒸馏
Abstract
Multi-teacher distillation allows transferring knowledge from multiple "teacher" networks to a single "student" network. This method is promising in fields such as digital pathology, where many powerful foundation models were proposed recently. However, as we show in this paper, the standard multi-teacher distillation approach reacts poorly to an increase in the number of teachers used to train a student encoder. We demonstrate that learning teacher-specific representations is a key to scaling in MTD. Importantly, different design choices, such as learnable teacher tokens, a tailored attention scheme, additional mixture-of-experts layers and a contrastive loss, are proposed to better learn such teacher-specific representations. We show that our method better scales with respect to the number of considered teachers, allowing us to train compact student encoders, up to 10x more efficient than larger teacher foundation models, while matching their performance and even outperforming them on a large set of tile-level and slide-level tasks. We conduct a thorough empirical validation, evaluating more than 10 foundation models on 39 tasks spanning tile and slide levels. As a result, we release a new collection of strong and efficient foundation models, named OMNI, trained from the knowledge of 10 state-of-the-art foundation models. We finally conduct an analysis of the learned teacher-specific representations, highlighting their complementarity and explaining why they can be easily aggregated at downstream time.
Chinese Translation
多教师蒸馏允许将来自多个“教师”网络的知识转移到单个“学生”网络。这种方法在数字病理学等领域具有很大潜力，最近提出了许多强大的基础模型。然而，正如我们在本文中所展示的，标准的多教师蒸馏方法在使用更多教师训练学生编码器时反应不佳。我们证明，学习教师特定的表示是扩展多教师蒸馏（MTD）的关键。重要的是，提出了不同的设计选择，如可学习的教师标记、定制的注意力机制、额外的专家混合层和对比损失，以更好地学习这些教师特定的表示。我们展示了我们的方法在考虑教师数量方面更具扩展性，使我们能够训练紧凑的学生编码器，其效率比更大的教师基础模型高出10倍，同时在性能上与之匹配，甚至在大量的切片级和幻灯片级任务上超越它们。我们进行了全面的实证验证，评估了超过10个基础模型在39个任务上的表现，这些任务涵盖了切片级和幻灯片级。最终，我们发布了一组新的强大且高效的基础模型，命名为OMNI，这些模型是基于10个最先进的基础模型的知识训练而成。最后，我们对学习到的教师特定表示进行了分析，强调了它们的互补性，并解释了为什么它们可以在下游任务中轻松聚合。
cs.CV / 6 / 2610.12575
Mind the Temporal Gap: Evaluating Video Models for Temporal Understanding in Driving
注意时间间隔：评估视频模型在驾驶中的时间理解能力
Abstract
Modern video-language models are increasingly treated as general-purpose video reasoners, yet it remains unclear whether they robustly capture temporal dynamics in safety-critical settings such as driving. We study this question through controlled ego-motion tasks that require identifying actions from driving videos. Across video-language models, we find that standard prompting, visual prompting, few-shot prompting, and chain-of-thought prompting remain unreliable and highly task- and model-dependent, with persistent class biases, although newer models show substantial task-specific few-shot gains. In contrast, lightweight LoRA finetuning and simple probes trained on intermediate visual features recover strong motion-discriminative performance, indicating that motion-relevant cues are often present inside the models but are not reliably used by the final video-to-language pathway. These results suggest that, in driving, current video-language models remain unreliable temporal reasoners and often behave more like frame-level, bias-driven recognizers than dependable temporal world models.
Chinese Translation
现代视频语言模型越来越被视为通用的视频推理器，但在安全关键的环境中，如驾驶，它们是否能够稳健地捕捉时间动态仍不清楚。我们通过控制的自我运动任务来研究这个问题，这些任务要求从驾驶视频中识别动作。在不同的视频语言模型中，我们发现标准提示、视觉提示、少量示例提示和思维链提示都不可靠，并且高度依赖于任务和模型，存在持续的类别偏差，尽管较新的模型在特定任务上显示出显著的少量示例收益。相比之下，轻量级的LoRA微调和在中间视觉特征上训练的简单探测器恢复了强大的运动区分性能，表明与运动相关的线索通常存在于模型内部，但并未被最终的视频到语言路径可靠地利用。这些结果表明，在驾驶中，当前的视频语言模型仍然是不可靠的时间推理器，往往更像是基于帧级别的、受到偏见驱动的识别器，而不是可靠的时间世界模型。
cs.CV / 7 / 2610.12624
MemReg: Streaming Outdoor LiDAR Point Cloud Registration with Hybrid Memory Buffers
MemReg：具有混合内存缓冲区的流式户外LiDAR点云配准
Abstract
Prior outdoor LiDAR registration works are commonly limited by the pair-wise input paradigm, which neglects temporal correlations inherent within streaming sequences. In this paper, we propose a novel multi-frame outdoor point cloud registration network for streaming LiDAR scans enhanced with pose-related memory buffers. The key observation is that long-term temporal LiDAR sequences can provide rich global contextual information to complete sparse measurements, filter outliers, and address low-overlap problems, thereby boosting registration performance. Specifically, two hybrid memory buffers are designed, including an implicit memory feature buffer and an explicit memory pose buffer, to store and dynamically update pose-related temporal features. Moreover, a novel dynamic history weighting module is developed to adaptively fuse current and history pose-related features. Extensive experiments on three outdoor datasets, including KITTI, nuScenes, and Apollo-Southbay, demonstrate state-of-the-art performance of MemReg, surpassing all previous pair-wise methods and also multi-frame SLAM systems. Our method also generalizes surprisingly well to multiview indoor registration scenarios with rather competitive performance on 3DMatch, 3DLoMatch, and ScanNet. Code will be released at https://github.com/IRMVLab/MemReg.
Chinese Translation
以往的户外LiDAR配准工作通常受到成对输入范式的限制，忽视了流式序列中固有的时间相关性。本文提出了一种新颖的多帧户外点云配准网络，用于增强姿态相关内存缓冲区的流式LiDAR扫描。关键观察是，长期的时间LiDAR序列可以提供丰富的全局上下文信息，以补全稀疏测量、过滤异常值并解决低重叠问题，从而提升配准性能。具体而言，设计了两个混合内存缓冲区，包括隐式内存特征缓冲区和显式内存姿态缓冲区，用于存储和动态更新与姿态相关的时间特征。此外，开发了一种新颖的动态历史加权模块，以自适应融合当前和历史的姿态相关特征。在三个户外数据集（包括KITTI、nuScenes和Apollo-Southbay）上的大量实验表明，MemReg的性能达到了最先进水平，超越了所有以前的成对方法以及多帧SLAM系统。我们的方法在多视角室内配准场景中也表现出惊人的泛化能力，在3DMatch、3DLoMatch和ScanNet上具有相当有竞争力的表现。代码将发布在https://github.com/IRMVLab/MemReg。
cs.CV / 8 / 2610.12625
ForcingBooster: Adaptive Patch-wise Denoising for Fast Autoregressive Video Diffusion
ForcingBooster：快速自回归视频扩散的自适应块状去噪
Abstract
Autoregressive video diffusion models have become a leading approach to real-time interactive video generation and video world modeling. However, existing methods typically allocate computation uniformly across space when generating the next video chunk, updating all patches with the same denoising schedule and number of function evaluations. This ignores the spatial heterogeneity of video content: static, simple, or recurring regions are often easier to denoise, while other regions require further refinement or additional context. A uniform budget thus leads to redundant computation on easy patches, whereas reducing the fixed step count can leave difficult patches insufficiently refined. We introduce \methodname{}, a general framework that dynamically allocates denoising steps to individual patches: easy patches stop early to improve efficiency, while difficult patches receive more refinement steps. This design accelerates generation with negligible quality degradation. Our method is compatible with most existing few-step autoregressive diffusion models, requiring only a lightweight routing module with modest training. Experiments demonstrate that \methodname{} reduces the total number of denoising steps applied across video patches by about 50\% while maintaining visual quality comparable to the four-step baseline.
Chinese Translation
自回归视频扩散模型已成为实时交互视频生成和视频世界建模的主要方法。然而，现有方法在生成下一个视频片段时通常在空间上均匀分配计算，使用相同的去噪计划和函数评估次数更新所有块。这忽视了视频内容的空间异质性：静态、简单或重复的区域通常更容易去噪，而其他区域则需要进一步的精细化或额外的上下文。因此，均匀的预算导致在简单块上进行冗余计算，而减少固定步骤数可能使困难块的精细化不足。我们提出了 extit{ForcingBooster}，这是一个动态分配去噪步骤到各个块的一般框架：简单块提前停止以提高效率，而困难块则获得更多的精细化步骤。这一设计加速了生成过程，几乎没有质量下降。我们的方法与大多数现有的少步自回归扩散模型兼容，仅需一个轻量级的路由模块和适度的训练。实验表明， extit{ForcingBooster}在保持与四步基线相当的视觉质量的同时，将应用于视频块的去噪步骤总数减少了约50%。
cs.CV / 9 / 2610.12650
Feature Competition and Its Representational Origins in Vision Language Models: Evidence from the Stroop Paradigm
视觉语言模型中的特征竞争及其表征来源：来自斯特鲁普范式的证据
Abstract
When you ask a vision-language model (VLM) a question with an image, it answers your question based on the image. But to act means to choose: saying one thing means choosing to respond not say another. This becomes an issue when the image contains multiple features that equally relate to the question at hand: how would VLMs choose to respond? This problem is precisely studied by the classical Stroop task in psychology: among a set of deliberately designed images whose text and color conflicts (e.g., \textcolor{blue}{RED}), participants are instructed to report the color. We adapt this psychological paradigm to a benchmark and prompted a range of contrastive and generative VLMs on their behavior. Across models, we observe a persistent bias toward the feature of text in Stroop images: when text and color disagree, models preferentially generate responses centered to the text rather than color. We then opened the black box and examined the source of this bias and why certain features influences the model's behavior more than others. In the embedding space, we found that textual cues are represented more strongly and separably than color cues. This representational asymmetry predicts intervention success: steering models toward text-related directions is substantially easier than steering them toward color-related directions. Using a classical psychological paradigm, we demonstrate, at both behavioral and representational levels, that distinct features within the same image can exert unequal influence on VLM behavior. Our findings suggest that VLMs assign different degrees of saliency to different features within the same image, highlighting the need to understand how cue-specific saliency shapes their performance in widely used downstream applications.
Chinese Translation
当你向视觉语言模型（VLM）提出带有图像的问题时，它会根据图像回答你的问题。但行动意味着选择：说一个东西意味着选择不说另一个。当图像包含多个与当前问题同等相关的特征时，这就成为一个问题：VLM将如何选择回应？这一问题正是心理学经典的斯特鲁普任务所研究的：在一组故意设计的文本和颜色冲突的图像中（例如， extcolor{blue}{RED}），参与者被指示报告颜色。我们将这一心理学范式适配为基准，并对一系列对比性和生成性VLM的行为进行了提示。在各个模型中，我们观察到在斯特鲁普图像中对文本特征的持续偏向：当文本和颜色不一致时，模型倾向于生成以文本为中心的响应，而非颜色。随后，我们打开了黑箱，检查了这种偏向的来源，以及为何某些特征对模型行为的影响大于其他特征。在嵌入空间中，我们发现文本线索的表征强度和可分离性均优于颜色线索。这种表征不对称性预测了干预的成功：引导模型朝向与文本相关的方向要显著容易于引导它们朝向与颜色相关的方向。通过使用经典的心理学范式，我们在行为和表征层面上证明了同一图像中的不同特征可以对VLM行为施加不等的影响。我们的发现表明，VLM在同一图像中对不同特征赋予了不同程度的显著性，强调了理解线索特异性显著性如何影响其在广泛应用的下游任务中的表现的必要性。
cs.CV / 10 / 2610.12676
WoodBR-Cross: A public wood-image dataset and individual-tree-disjoint benchmark for species identification
WoodBR-Cross：一个用于物种识别的公共木材图像数据集和个体树木不重叠基准
Abstract
Wood-image benchmarks can yield optimistic estimates when images from the same tree occur in both training and test sets. WoodBR-Cross contains 4,875 transverse-surface images in 38 operational taxonomic classes, linked to 303 distinct trees identified by numeric catalogue witness codes (codigos-testemunho) in the Brazilian Forest Service collection. A species-level subset comprises 3,406 images, 20 classes and 214 trees. Three tasks compare duplicate-aware image partitions for all classes (A), image partitions for the species subset (B1), and individual-tree-disjoint partitions for that subset (B2). In B2, all images from each tree are kept together, so no tree is shared across training, validation or test within a fold. ResNet-50, ConvNeXt-Tiny and ViT-B/16 were evaluated with three outer folds, internal validation and five training seeds. Handcrafted and frozen-feature baselines complement border-only and randomized-label controls. Validation-selected ConvNeXt-Tiny achieved B2 tree-level macro-F1 of 0.8621 +/- 0.0317; frozen DINOv2 features yielded 0.9000 +/- 0.0722. Values are means and standard deviations across three seed-averaged fold means. Results estimate performance on previously unseen trees from this collection.
Chinese Translation
当训练集和测试集中出现来自同一棵树的图像时，木材图像基准可能会产生乐观的估计。WoodBR-Cross包含4,875幅横截面图像，涵盖38个操作分类学类，与巴西森林服务局收藏中通过数字目录见证代码（codigos-testemunho）识别的303棵不同树木相关联。物种级子集包括3,406幅图像、20个类和214棵树。三个任务比较了所有类的重复感知图像划分（A）、物种子集的图像划分（B1）以及该子集的个体树木不重叠划分（B2）。在B2中，每棵树的所有图像被保留在一起，因此在一个折叠内没有树木在训练、验证或测试中共享。ResNet-50、ConvNeXt-Tiny和ViT-B/16在三个外部折叠、内部验证和五个训练种子下进行了评估。手工制作和冻结特征的基线补充了仅边界和随机标签的控制。经过验证选择的ConvNeXt-Tiny在B2树级宏F1值上达到了0.8621 +/- 0.0317；冻结的DINOv2特征则达到了0.9000 +/- 0.0722。这些值是基于三个种子平均折叠均值的均值和标准差。结果估计了在该集合中之前未见过的树木的性能。
cs.CV / 11 / 2610.12703
Bridging Human and Model Attention: Explainability Analysis of CNN, Mamba, and ViT Architectures with Gaze-Based Validation
桥接人类与模型注意力：基于注视验证的CNN、Mamba和ViT架构的可解释性分析
Abstract
High-risk applications of deep learning require precise characterization of model decision-making. We conduct a systematic explainability evaluation across Convolutional Neural Networks (CNNs), Vision Transformers (ViTs), and Vision Mamba (ViM). Unlike predominantly qualitative studies, we quantify the alignment between model attention and human visual perception using eye-tracking ground truth. We formalize this comparison via a standardized normalization pipeline mapping post-hoc attribution maps and gaze heatmaps onto spatial probability distributions. Alignment is evaluated across fine-grained classification, medical imaging, and emotion recognition using architecture-specific XAI methods. We measure fidelity using Jensen-Shannon similarity, Chi-square similarity, and Pearson correlation, three complementary estimators capturing support overlap, mass discrepancy, and structural co-variation. Our results reveal distinct attention dynamics: CNNs tend to exhibit a more global focus, with spatially more diffuse saliency maps that correlate well with human perception in general contexts, while ViTs demonstrate fine-grained, patch-level attention, often focused on discriminative microfeatures. The Mamba architecture shows biased behavior: it achieves strong explainability in medical tasks but exhibits divergent intensity distributions in natural images; our analytical characterization of the metrics makes this dissociation explicit, since near-unit support overlap is attained simultaneously with a vanishing linear correlation. We further analyze the effect of backbone freezing and ensemble strategies on model interpretability. The results highlight how architectural and training choices influence the alignment between model saliency and human visual attention.
Chinese Translation
深度学习的高风险应用需要对模型决策进行精确表征。我们对卷积神经网络（CNN）、视觉变换器（ViT）和视觉Mamba（ViM）进行了系统的可解释性评估。与主要定性研究不同，我们量化了模型注意力与人类视觉感知之间的对齐程度，使用眼动追踪作为真实数据。我们通过标准化的归一化流程将后验归因图和注视热图映射到空间概率分布，从而形式化这一比较。我们在细粒度分类、医学成像和情感识别中使用特定架构的可解释人工智能（XAI）方法评估对齐情况。我们使用Jensen-Shannon相似性、卡方相似性和Pearson相关系数这三种互补估计量来测量保真度，捕捉支持重叠、质量差异和结构共变的情况。我们的结果揭示了不同的注意力动态：CNN倾向于表现出更全球的关注，具有空间上更分散的显著性图，与人类在一般上下文中的感知良好相关，而ViT则展示了细粒度的、基于补丁的注意力，通常集中于区分性微特征。Mamba架构表现出偏向行为：在医学任务中实现了强可解释性，但在自然图像中表现出不同的强度分布；我们对指标的分析特征使这种分离变得明确，因为几乎单位的支持重叠与消失的线性相关性同时出现。我们进一步分析了骨干网络冻结和集成策略对模型可解释性的影响。结果强调了架构和训练选择如何影响模型显著性与人类视觉注意力之间的对齐。
cs.CV / 12 / 2610.12727
Zebra3D: Efficient Hybrid Foundation Models For Scalable 3D Reconstruction
Zebra3D：可扩展3D重建的高效混合基础模型
Abstract
Feed-forward 3D foundation models reconstruct scenes from unposed images, but the quadratic cost of global attention limits their scalability to large view sets. We present Zebra3D, a layerwise sparse-linear hybrid attention framework that improves the efficiency of pretrained 3D foundation models without training from scratch. Across VGGT, HunyuanWorld-Mirror, and $\pi^3$, we observe a broadly consistent depth-wise pattern: early stages are relatively tolerant to linear replacement, while sensitive middle-to-late stages tend to exhibit more concentrated attention. Zebra3D exploits this structure by assigning linear-centered attention to tolerant stages and tile-sparse attention to sensitive stages, using landmark summaries to preserve unselected context. Closed-form affine fitting enables training-free conversion, while optional post-conversion training combines task supervision with output and relational feature distillation. Across all three backbones, Zebra3D achieves strong reconstruction and camera-pose accuracy at high attention sparsity, with training-free variants remaining close to their dense counterparts. End-to-end speedups increase with input view count, reaching up to $7.1\times$ at 1,024 views. After post-conversion training, Zebra3D-VGGT reaches 80.5\% RealEstate10K AUC@30 at 94\% mean attention sparsity, within 1.1 percentage points of dense VGGT fine-tuned on the same data and schedule.
Chinese Translation
前馈3D基础模型从未摆姿态的图像中重建场景，但全局注意力的二次成本限制了它们在大视图集上的可扩展性。我们提出了Zebra3D，一种分层稀疏线性混合注意力框架，能够在不从头训练的情况下提高预训练3D基础模型的效率。在VGGT、HunyuanWorld-Mirror和$ ext{π}^3$上，我们观察到一个普遍一致的深度模式：早期阶段对线性替换相对宽容，而中后期阶段则倾向于表现出更集中的注意力。Zebra3D利用这一结构，将线性中心注意力分配给宽容阶段，并将瓷砖稀疏注意力分配给敏感阶段，同时使用地标摘要来保留未选择的上下文。封闭形式的仿射拟合实现了无训练转换，而可选的后转换训练结合了任务监督与输出和关系特征的蒸馏。在所有三个基础模型中，Zebra3D在高注意力稀疏性下实现了强大的重建和相机姿态精度，且无训练变体与其密集对应物保持接近。端到端的加速随着输入视图数量的增加而增加，在1,024个视图时达到$7.1 imes$。在后转换训练后，Zebra3D-VGGT在94 ext{%}平均注意力稀疏性下达到80.5 ext{%} RealEstate10K AUC@30，距离在相同数据和计划上微调的密集VGGT仅相差1.1个百分点。
cs.CV / 13 / 2610.12735
Reference-Free Food Volume Estimation from Egocentric Stereo Video
基于无参考的自我中心立体视频食品体积估计
Abstract
Current stereo vision-based approaches for food portion estimation are hindered by the user burden of using a mobile phone to capture multiple images or placing a reference object in the scene for scale calibration. As a step towards passive dietary intake monitoring, we introduce a method that uses stereo video from a wearable egocentric device to reconstruct food items in true physical scale to accurately estimate volume. By utilizing FoundationStereo for zero-shot stereo matching, SAM 2 for food segmentation, and TSDF fusion over a set of keyframes, our method yields a visually and metrically accurate 3D mesh of any plated food item from a single stereo recording. We leverage the unique offerings of wearables, such as a factory-calibrated stereo baseline and camera poses estimated on-device via visual-inertial odometry (VIO), to perform fully automatic reconstruction without a scale reference object in the scene. We show strong initial volume estimation results across five recorded food items, achieving a mean absolute percentage error of 9.1%.
Chinese Translation
当前基于立体视觉的食品份量估计方法受到用户使用手机拍摄多张图像或在场景中放置参考物体进行尺度校准的负担限制。作为被动饮食摄入监测的一个步骤，我们提出了一种方法，利用可穿戴自我中心设备的立体视频重建食品项目的真实物理尺度，以准确估计体积。通过利用 FoundationStereo 进行零样本立体匹配，使用 SAM 2 进行食品分割，以及在一组关键帧上进行 TSDF 融合，我们的方法能够从单个立体录制中生成任何盘装食品项目的视觉和度量上准确的 3D 网格。我们利用可穿戴设备的独特优势，如工厂校准的立体基线和通过视觉惯性里程计（VIO）在设备上估计的相机姿态，实现了无需场景中尺度参考物体的全自动重建。我们在五种记录的食品项目上展示了强劲的初步体积估计结果，平均绝对百分比误差为 9.1%。
cs.CV / 14 / 2610.12783
Shift-Aware Identity-Guided Latent Refinement for Referring Audio--Visual Segmentation
基于身份引导的位移感知潜在细化用于指向音视频分割
Abstract
Referring Audio--Visual Segmentation (Ref-AVS) is an emergent multimodal task requiring the precise identification and segmentation of a specific object based on a natural language expression grounded in both visual and auditory cues. Unlike traditional AVS, which segments generic sound sources, or referring video object segmentation (Ref-VOS), which relies solely on visual--textual grounding, Ref-AVS necessitates fine-grained tri-modal interaction across vision, audio, and language. However, existing methods often fail to maintain consistency when subjected to target shift, a phenomenon where the model erroneously transfers its focus between objects due to transient occlusions, acoustic fading, or identity ambiguities. To address this, we propose a novel Shift-Aware Identity-guided Latent (SAIL) refinement framework. It introduces a dual-token strategy that decouples frame-level dynamics from a global semantic anchor to explicitly model target-shift cues. We design a shift-aware refinement module that rectifies drift by aligning tokens with domain-specific evidence. Finally, we leverage a decoder that integrates the refined embeddings with a continuous memory update to maintain spatio-temporal coherence and identity consistency throughout the video sequence. Extensive experiments conducted on two Ref-AVS benchmarks demonstrate the effectiveness of our proposed method, significantly outperforming existing AVS, Ref-VOS, and Ref-AVS approaches. The code will be released after publication.
Chinese Translation
指向音视频分割（Referring Audio–Visual Segmentation, Ref-AVS）是一项新兴的多模态任务，要求根据基于视觉和听觉线索的自然语言表达，精确识别和分割特定对象。与传统的音视频分割（AVS）仅对通用声源进行分割，或仅依赖视觉-文本基础的指向视频对象分割（Ref-VOS）不同，Ref-AVS需要在视觉、音频和语言之间进行细粒度的三模态交互。然而，现有方法在目标转移时常常无法保持一致性，目标转移是指模型由于瞬时遮挡、声学衰减或身份模糊而错误地在对象之间转移焦点的现象。为此，我们提出了一种新颖的位移感知身份引导潜在（Shift-Aware Identity-guided Latent, SAIL）细化框架。该框架引入了一种双标记策略，将帧级动态与全局语义锚点解耦，以明确建模目标转移线索。我们设计了一个位移感知细化模块，通过与领域特定证据对齐标记来纠正漂移。最后，我们利用一个解码器，将细化的嵌入与连续的记忆更新相结合，以在整个视频序列中保持时空一致性和身份一致性。在两个Ref-AVS基准上进行的大量实验表明，我们提出的方法有效性显著优于现有的AVS、Ref-VOS和Ref-AVS方法。代码将在发表后发布。
cs.CV / 15 / 2610.12786
Domain Adaptation of SAM 2.1 for Building-Component Segmentation in Image-Based 3D Reconstruction
基于图像的3D重建中SAM 2.1的建筑构件分割领域适应
Abstract
3D reconstruction of building components using image-based scanning requires appropriate segmentation of the objects in the 2D building images. In order to benefit from consistent grouping of the 2D masks in 3D environments, vision foundation models such as the Segment Anything Model have been utilized by Group Anything with Radiance Fields (GARField) to perform 3D segmentation. This study fine-tuned SAM 2.1 on the HBD dataset, consisting of 2,679 annotated images from 42 buildings, to enhance multi-class segmentation of 2D building images, covering 11 classes (Wall, Window, Ceiling, Door, Column, Beam, Floor, Opening, Curtain Wall, Roof, and Lift). The model was evaluated using per-class IoU, Dice, and mIoU metrics, and the results show an improvement in mIoU from 49.9% to 69.6%, indicating a meaningful improvement in class segmentation and providing a more accurate model for distilling the masks in 3D. SegFormer was then used as a baseline, generating per-class masks for comparison with SAM 2.1, which indicates improvement in some object classes with larger surface area. The masks in the building images were grouped in the neural radiance field, and the pretrained SAM 2.1 was compared with the fine-tuned SAM 2.1 in the newly rendered images. This comparison shows that fine-tuning improves the grouping of building components at a user-selected grouping scale. This advances the application of Group Anything with Radiance Fields (GARField) in 3D, enabling better segmentation of building structures for 3D auto-reconstruction.
Chinese Translation
使用基于图像的扫描进行建筑构件的3D重建需要对2D建筑图像中的对象进行适当的分割。为了在3D环境中实现2D掩膜的一致分组，视觉基础模型如Segment Anything Model（SAM）被Group Anything with Radiance Fields（GARField）用于执行3D分割。本研究在HBD数据集上对SAM 2.1进行了微调，该数据集包含来自42栋建筑的2,679张标注图像，以增强2D建筑图像的多类分割，涵盖11个类别（墙、窗、天花板、门、柱、梁、地板、开口、幕墙、屋顶和电梯）。模型使用每类IoU、Dice和mIoU指标进行评估，结果显示mIoU从49.9%提升至69.6%，表明类别分割有了显著改善，并为在3D中提取掩膜提供了更准确的模型。随后，SegFormer作为基线生成每类掩膜，与SAM 2.1进行比较，结果表明在某些表面积较大的物体类别上有所改善。建筑图像中的掩膜在神经辐射场中进行分组，预训练的SAM 2.1与微调后的SAM 2.1在新渲染的图像中进行比较。该比较显示微调改善了用户选择的分组尺度下建筑构件的分组。这推动了Group Anything with Radiance Fields（GARField）在3D中的应用，使建筑结构的分割在3D自动重建中更加精确。
cs.CV / 16 / 2610.12811
Spend Where It Matters: Particle Search with MH Refinement for Budget-Constrained Audio-Visual Alignment
在关键位置支出：基于粒子的搜索与预算受限的音视频对齐中的MH精炼
Abstract
Reward-guided inference-time alignment can improve joint audio-visual generation, but optimizing multiple alignment objectives requires many expensive model evaluations, making existing methods costly under limited compute. Particle-based search enables global exploration but incurs estimation error with a finite population, while Metropolis-Hastings (MH) can refine candidates locally but benefits from a strong initialization. We introduce \textbf{S}pend \textbf{W}here \textbf{I}t \textbf{M}atters (\modelname), a two-stage, training-free, reward-agnostic framework that combines particle search with MH refinement under a fixed compute budget. In the first stage, we adaptively allocate particles across denoising checkpoints using online variance estimates, minimizing estimation variance while concentrating compute where it is most useful. This search provides a promising candidate for MH refinement, which uses local noise-and-denoise proposals on completed outputs to further improve reward. By starting from a strong candidate, MH can recover additional reward without requiring a larger particle population or costly global exploration. We show that, at matched compute, combining particle search with MH refinement achieves higher rewards than either approach alone. On \textsc{JavisBench-Mini}, at closely matched compute, \modelname achieves mean relative gains across seven metrics of $30.7\%$ over BoN and $14.1\%$ over EvoSearch with JavisDiT++, and $9.2\%$ over BoN and $17.0\%$ over EvoSearch with LTX-2. Beyond joint audio-visual generation, we also demonstrate the benefits of adaptive particle allocation on SD1.5 image generation, improving AestheticScore by $2.2\%$ and ImageReward by $6.0\%$ over uniform allocation.
Chinese Translation
基于奖励的推理时间对齐可以改善音视频的联合生成，但优化多个对齐目标需要大量昂贵的模型评估，这使得在计算资源有限的情况下现有方法成本高昂。基于粒子的搜索能够进行全局探索，但在有限的种群中会产生估计误差，而Metropolis-Hastings (MH)可以在局部精炼候选者，但需要强大的初始化。我们提出了 extbf{S}pend extbf{W}here extbf{I}t extbf{M}atters ( extit{modelname})，这是一个两阶段的、无训练的、与奖励无关的框架，结合了粒子搜索与MH精炼，并在固定的计算预算下运行。在第一阶段，我们使用在线方差估计自适应地分配粒子到去噪检查点，最小化估计方差，同时将计算集中在最有用的地方。这种搜索为MH精炼提供了一个有前景的候选者，MH利用已完成输出的局部噪声和去噪提案进一步提高奖励。通过从强候选者开始，MH可以在不需要更大粒子种群或昂贵全局探索的情况下恢复额外的奖励。我们展示了在相同计算条件下，将粒子搜索与MH精炼结合可以获得比单独使用任何一种方法更高的奖励。在 extsc{JavisBench-Mini}上，在相近的计算条件下， extit{modelname}在七个指标上相较于BoN平均提高了$30.7\%$，相较于EvoSearch提高了$14.1\ ext{ }$（使用JavisDiT++），相较于BoN提高了$9.2\ ext{ }$，相较于EvoSearch提高了$17.0\ ext{ }$（使用LTX-2）。除了联合音视频生成外，我们还展示了自适应粒子分配在SD1.5图像生成中的优势，相较于均匀分配，AestheticScore提高了$2.2\ ext{ }$，ImageReward提高了$6.0\ ext{ }$。
cs.CV / 17 / 2610.12835
Structure-Aware Matching for Robust UAV Thermal-to-Satellite Geo-Localization
结构感知匹配用于鲁棒无人机热成像到卫星的地理定位
Abstract
UAV thermal-to-satellite geo-localization localizes UAVs by matching real-time thermal images against georeferenced satellite images, enabling autonomous navigation in GNSS-denied and low-visibility conditions. However, substantial cross-modal disparities in color, brightness and viewpoint hinder conventional appearance-based methods from establishing reliable correspondences. While existing approaches predominantly address global modality shifts, they overlook the invariant scene structures inherently shared between thermal and satellite views. To bridge this gap, we propose SAMF, a Structure-Aware Matching Framework that explicitly exploits stable cross-modal structural cues to improve UAV thermal geolocalization. Specifically, the Directional Structure Enhancement Module (DSEM) strengthens robust structural representations by capturing directional and low-frequency spatial patterns, and the Modality-Aware Affine Remapping Module (MARM) subsequently applies affine correction to these enhanced features, further promoting cross-modal alignment. Additionally, we introduce the Thermal-UAV-Paired benchmark derived from the Thermal-UAV dataset. Extensive experiments on both the Boson-nighttime and Thermal-UAV-Paired datasets show that SAMF achieves state-of-the-art performance. Ablation studies further validate the effectiveness and synergistic complementarity of DSEM and MARM. The code and dataset are released at: https://github.com/YuxinWang879/SAMF.git.
Chinese Translation
无人机热成像到卫星的地理定位通过将实时热成像与地理参考的卫星图像进行匹配，从而实现无人机在GNSS失效和低能见度条件下的自主导航。然而，颜色、亮度和视角等显著的跨模态差异阻碍了传统基于外观的方法建立可靠的对应关系。尽管现有方法主要解决全球模态转变，但它们忽视了热成像和卫星视图之间固有共享的不变场景结构。为了解决这一问题，我们提出了SAMF（结构感知匹配框架），该框架明确利用稳定的跨模态结构线索来改善无人机热成像地理定位。具体而言，方向结构增强模块（DSEM）通过捕捉方向性和低频空间模式来增强鲁棒的结构表示，而模态感知仿射重映射模块（MARM）随后对这些增强特征进行仿射校正，进一步促进跨模态对齐。此外，我们引入了基于热成像无人机数据集衍生的热成像无人机配对基准。对Boson夜间和热成像无人机配对数据集的广泛实验表明，SAMF实现了最先进的性能。消融研究进一步验证了DSEM和MARM的有效性及其协同互补性。代码和数据集已发布于：https://github.com/YuxinWang879/SAMF.git。
cs.CV / 18 / 2610.12843
APEX: Asynchronous Prefix Denoising for Extreme Compression
APEX：极端压缩的异步前缀去噪
Abstract
Fewer latent tokens promise cheaper diffusion. Yet aggressive latent compression creates a reconstruction-generation dilemma: adding channels improves reconstruction but makes generation harder. We introduce APEX, short for Asynchronous Prefix denoising for EXtreme compression, which pairs a hierarchical latent space with asynchronous denoising over channel groups. Our autoencoder orders channel prefixes from structure to detail, while one shared denoiser generates them at staggered local times without adding tokens. On ImageNet-512, APEX outperforms existing methods at every evaluated channel count under 64x and 128x compression. At matched training and guided sampling compute, it also outperforms a 256-token model at every evaluated budget, with larger gains at lower compute. Code is available at https://daviddinkevich.github.io/APEX.
Chinese Translation
更少的潜在标记承诺更便宜的扩散。然而，激进的潜在压缩造成了重建与生成之间的困境：增加通道可以改善重建，但却使生成变得更加困难。我们提出了APEX，即极端压缩的异步前缀去噪，结合了分层潜在空间和对通道组的异步去噪。我们的自编码器按照结构到细节的顺序排列通道前缀，而一个共享的去噪器在错开的局部时间生成它们，而不增加标记。在ImageNet-512上，APEX在每个评估的通道数量下，在64倍和128倍压缩下均优于现有方法。在匹配的训练和引导采样计算下，它在每个评估预算中也优于256个标记的模型，并在较低计算下获得更大的增益。代码可在 https://daviddinkevich.github.io/APEX 获取。
cs.CV / 19 / 2610.12866
CaptchaAgent: Verifiable Visual Reasoning for CAPTCHA Solving
CaptchaAgent：可验证的视觉推理用于 CAPTCHA 解题
Abstract
Modern reasoning CAPTCHAs require fine-grained perception, compositional relation reasoning, and grounded browser interaction. End-to-end vision-language model agents remain brittle in this setting because a missed object, incorrect count, or confused relation can propagate directly to the final action. To address this problem, we present CaptchaAgent, a structured visual reasoning framework for interactive CAPTCHA solving. Instead of predicting an action directly from a screenshot, CaptchaAgent first constructs a visual intermediate representation (V-IR) that explicitly records entities, attributes, and spatial or topological relations. A symbol-only ASCII protocol further requires the parser to commit to a concrete object set and count. Building on this global representation, localized evidence refinement revisits ambiguous regions at higher effective resolution, while counterfactual consistency verification tests whether the resulting action remains stable under task-preserving transformations. Across three proprietary backbones on Open CaptchaWorld, CaptchaAgent improves Pass@1 by 21.3-25.8 percentage points over a commercially available browser agent baseline evaluated with the same backbones. On the separate offline Oedipus-data benchmark, it further improves average Pass@1 by 9.2-12.3 percentage points over the Oedipus solver across three matched backbones. The intermediate traces show that exact counting, continuous geometry, and ordered execution remain major sources of automation difficulty for the evaluated agents, providing practical guidance for designing more resilient compositional CAPTCHAs.
Chinese Translation
现代推理 CAPTCHA 需要细致的感知、组合关系推理和基于浏览器的交互。端到端的视觉-语言模型代理在这种环境下仍然脆弱，因为遗漏的对象、错误的计数或混淆的关系会直接影响最终的行动。为了解决这个问题，我们提出了 CaptchaAgent，一个用于交互式 CAPTCHA 解题的结构化视觉推理框架。CaptchaAgent 首先构建一个视觉中间表示（V-IR），明确记录实体、属性以及空间或拓扑关系，而不是直接从屏幕截图预测行动。一个仅限符号的 ASCII 协议进一步要求解析器承诺具体的对象集和计数。在此全局表示的基础上，本地证据细化在更高有效分辨率下重新审视模糊区域，而反事实一致性验证则测试在任务保持变换下，结果行动是否保持稳定。在 Open CaptchaWorld 上的三个专有骨干网络中，CaptchaAgent 相较于使用相同骨干网络评估的商业浏览器代理基线，Pass@1 提高了 21.3-25.8 个百分点。在单独的离线 Oedipus 数据基准上，它在三个匹配的骨干网络上进一步提高了 Oedipus 解算器的平均 Pass@1 9.2-12.3 个百分点。中间跟踪显示，精确计数、连续几何和有序执行仍然是评估代理自动化困难的主要来源，为设计更具弹性的组合 CAPTCHA 提供了实际指导。
cs.CV / 20 / 2610.12868
Selective Spatio-Temporal Fruit Inspector (S$^2$FI): A Mamba-based Framework for Rotational Fruit Quality Assessment
选择性时空水果检测器 (S$^2$FI)：基于Mamba的旋转水果质量评估框架
Abstract
The global fruit and vegetable industry confronts severe post-harvest sorting challenges for small and medium-scale growers. Large intelligent sorting systems are too costly, while manual sorting suffers from low efficiency, inconsistent standards, and high labor demand. To bridge this gap between expensive equipment and inefficient manual work, this study develops a low-cost, high-efficiency, lightweight fruit sorting system driven by deep learning, suitable for small and medium growers. We propose the Selective Spatio-Temporal Fruit Inspector (S$^2$FI) architecture, which combines a 3D convolutional encoder with a Mamba-based backbone.It efficiently extracts spatiotemporal features, adaptively weighs informative frames during fruit rotation, and models long-range dependencies with linear complexity, reducing motion blur and occlusion. The hardware uses a dual-conveyor and curved rail design, supporting household power, fast deployment, and lightweight embedded deployment to improve affordability. We also built a conveyor-based fruit quality video dataset with 6,460 samples. Experiments on the held-out test set show that our method achieves the best trade-off among accuracy, efficiency, and practicality, outperforming the compared baselines on the test set, achieving an excellent balance of accuracy, speed, and practicality for real-world sorting.
Chinese Translation
全球水果和蔬菜行业面临着小型和中型种植者在后收获分拣方面的严重挑战。大型智能分拣系统成本过高，而人工分拣效率低下、标准不一致且劳动需求高。为了弥补昂贵设备与低效人工工作之间的差距，本研究开发了一种低成本、高效率、轻量化的水果分拣系统，该系统由深度学习驱动，适用于小型和中型种植者。我们提出了选择性时空水果检测器 (S$^2$FI) 架构，该架构结合了3D卷积编码器和基于Mamba的主干网络。它有效地提取时空特征，在水果旋转过程中自适应地加权信息帧，并以线性复杂度建模长程依赖关系，从而减少运动模糊和遮挡。硬件采用双输送带和弯曲轨道设计，支持家庭电源，快速部署，以及轻量嵌入式部署，以提高可负担性。我们还构建了一个基于输送带的水果质量视频数据集，包含6460个样本。在保留的测试集上的实验表明，我们的方法在准确性、效率和实用性之间实现了最佳平衡，优于比较基线，在测试集上取得了准确性、速度和实用性的优秀平衡，适用于现实世界的分拣。
cs.CV / 21 / 2610.12871
ROSETTA: Rotation Subspace Estimation for Training-free Transmodal Alignment
ROSETTA：无训练的跨模态对齐中的旋转子空间估计
Abstract
When a query must be matched against a database built with a different sensor or processed by a different encoder, the independently trained encoders impose distinct coordinate systems on the feature space. No canonical reference frame exists to which all encoders can be aligned, and no single training procedure has fully closed this gap. We show that the relative transformation between any two such systems is a spatially preserving rotation in whitened feature space, whose angular velocity exhibits white-noise characteristics: large in norm, temporally uncorrelated, and unpredictable from the features themselves. However, the cross-covariance between encoder outputs has low effective rank, constraining only a small fraction of the rotation parameters, therefore local alignment is tractable. ROSETTA is a training-free calibration layer that estimates the local rotation via an expectation-maximization procedure, requiring no coobservation at query time. We validate ROSETTA on three public driving datasets spanning LiDAR, camera, thermal, and event modalities in cross-modal, cross-device, cross-representation, and cross-backbone settings, with milliseconds overhead per query.
Chinese Translation
当查询必须与使用不同传感器构建或由不同编码器处理的数据库进行匹配时，独立训练的编码器在特征空间中施加了不同的坐标系统。不存在一个所有编码器都可以对齐的标准参考框架，也没有单一的训练过程能够完全弥补这一差距。我们表明，任何两个这样的系统之间的相对变换是在白化特征空间中保持空间不变的旋转，其角速度表现出白噪声特征：范数大、时间上不相关，并且无法从特征本身预测。然而，编码器输出之间的交叉协方差具有低有效秩，仅约束了一小部分旋转参数，因此局部对齐是可行的。ROSETTA是一个无训练的校准层，通过期望最大化程序估计局部旋转，在查询时无需共同观察。我们在三个公共驾驶数据集上验证了ROSETTA，这些数据集涵盖了激光雷达、相机、热成像和事件模态，涉及跨模态、跨设备、跨表示和跨骨干网的设置，每次查询的开销为毫秒级。
cs.CV / 22 / 2610.12887
Win-Win Fusion: Multi-Modal Image Fusion with Visual-Semantic Collaboration
双赢融合：视觉-语义协作的多模态图像融合
Abstract
Existing multi-modal image fusion (MMIF) methods generally face an optimization gap between the visual and semantic domains, limiting the enhancement of perception capabilities. To overcome this long-standing barrier, this work proposes W2F, a new visual-semantic mutual reinforcing paradigm for MMIF. First, we reinforce the foundation for visual-semantic collaboration by amplifying feature complementarity through two targeted designs. Specifically, the segmentation-guided fusion loss introduces object-aware structural guidance into visual fusion, while the embedded information restoration prior strengthens the ability of visual fusion to provide reliable perceptual evidence for the semantic branch under degraded conditions. Then, we introduce a strong multi-modal information interaction mechanism composed of interactive gated mixed attention (IGM-Att) and progressive coupling attention (PC-Att). These modules ensure comprehensive scene representation from both local and global perspectives, providing informational support for visual-semantic collaboration. Finally, we construct a collaborative mechanism to couple visual fusion and semantic fusion. It employs a feature search module to extract complementary key features from both tasks, enriching the representational potential of visual and semantic information. Meanwhile, a gradient correction strategy ensures collaboration from the optimization perspective, enhancing performance for both visual and semantic fusion through consistency learning. Extensive experiments demonstrate that W2F surpasses state-of-the-art methods in both visual appeal and decision accuracy. The code is publicly available at https://github.com/HaoZhang1018/W2F.
Chinese Translation
现有的多模态图像融合（MMIF）方法通常面临视觉和语义领域之间的优化差距，限制了感知能力的提升。为了解决这一长期存在的障碍，本文提出了W2F，一种新的视觉-语义相互强化的MMIF范式。首先，我们通过两种针对性的设计增强视觉-语义协作的基础，放大特征互补性。具体而言，基于分割的融合损失将对象感知的结构指导引入视觉融合，而嵌入的信息恢复先验则增强了视觉融合在退化条件下为语义分支提供可靠感知证据的能力。接着，我们引入了一种强大的多模态信息交互机制，由交互门控混合注意力（IGM-Att）和渐进耦合注意力（PC-Att）组成。这些模块确保从局部和全局视角全面表示场景，为视觉-语义协作提供信息支持。最后，我们构建了一种协作机制，将视觉融合和语义融合耦合在一起。该机制采用特征搜索模块，从两个任务中提取互补的关键特征，丰富视觉和语义信息的表征潜力。同时，梯度校正策略确保从优化的角度进行协作，通过一致性学习提升视觉和语义融合的性能。大量实验表明，W2F在视觉吸引力和决策准确性方面均超越了现有的最先进方法。代码已公开发布在 https://github.com/HaoZhang1018/W2F。
cs.CV / 23 / 2610.12895
MeshBind: Point-to-Mesh Binding for Non-Rigid Scene Flow Automatic Annotation
MeshBind：用于非刚性场景流自动标注的点到网格绑定
Abstract
Scene flow estimation is essential for many autonomous driving (AD) and robotic applications, but obtaining ground-truth scene flow for non-rigid objects such as pedestrians is highly difficult. We present MeshBind, an automatic annotation method that uses mesh models to generate per-point scene flow for non-rigid objects. The novelty of this work is threefold: 1) developing a multi-view module that reconstructs an articulated mesh model from six surround cameras and registers it to the LiDAR frame; 2) developing a local-frame binding technique that attaches each point to the mesh via a per-triangle tangent-normal frame, invariant under rigid motion yet adaptive to articulated deformation; 3) developing a three-layer flow registration pipeline that combines ego-vehicle motion compensation, box binding, and mesh binding to produce full-scene per-point flow. We implement MeshBind in SF-Studio, an interactive annotation tool that visualises flow and allows human mesh corrections with automatic recomputation. We also release SFB (SUScape-Flow Benchmark), a large-scale benchmark built on the public SUScape dataset, covering 41k frames with per-point non-rigid pedestrian annotations and a Rigid/Semi-Rigid/Non-Rigid evaluation protocol. Experiments show that fine-tuning on SFB reduces three-way endpoint error (EPE) by 2-3x, and brings non-rigid EPE to the same level as rigid EPE.
Chinese Translation
场景流估计对于许多自动驾驶（AD）和机器人应用至关重要，但获取非刚性物体（如行人）的真实场景流极为困难。我们提出了MeshBind，这是一种自动标注方法，利用网格模型为非刚性物体生成逐点场景流。本研究的创新之处有三：1）开发了一个多视角模块，从六个周围摄像头重建关节网格模型，并将其注册到激光雷达（LiDAR）帧；2）开发了一种局部框架绑定技术，通过每个三角形的切线法线框将每个点附着到网格上，该框架在刚性运动下保持不变，但对关节变形具有适应性；3）开发了一个三层流注册管道，结合自车运动补偿、框绑定和网格绑定，以生成全场景的逐点流。我们在SF-Studio中实现了MeshBind，这是一个交互式标注工具，能够可视化流并允许人工对网格进行修正及自动重新计算。我们还发布了SFB（SUScape-Flow基准），这是一个基于公共SUScape数据集的大规模基准，涵盖41,000帧的逐点非刚性行人标注以及刚性/半刚性/非刚性评估协议。实验表明，在SFB上进行微调可将三向端点误差（EPE）降低2-3倍，并使非刚性EPE达到与刚性EPE相同的水平。
cs.CV / 24 / 2610.12897
AD-SRF: Attack-Defense Semantic Reinforcement for Multi-Modal Image Fusion
AD-SRF：用于多模态图像融合的攻击-防御语义强化
Abstract
Existing image fusion methods struggle to balance visual quality and semantic attributes due to the inherent conflict between their optimization objectives. To address this challenge, we propose an attack-defense semantic reinforcement framework for multi-modal image fusion, termed AD-SRF. By decoupling the optimization of these two aspects and then losslessly reintegrating them, this framework injects rich semantic attributes into the fused image without compromising its visual quality. First, we develop a transformer-based restorative visual fusion network as a strong visual baseline. Equipped with the cross-modal retrieval module and the information restoration-fusion collaborative loss, it produces highly visible fused images even under degraded conditions. Second, we introduce an attack-defense game mechanism to strengthen the semantic attributes of the fused images. The attacker learns perturbation patterns that impair downstream segmentation, whereas the defender restores missing semantics from the original inputs and reintegrates them into the fused image through an imperceptible semantic map. Through iterative adversarial interaction, this mechanism drives the defender to continually strengthen the fused image's semantic attributes, boosting downstream segmentation performance without compromising visual quality. Extensive experiments demonstrate that AD-SRF achieves a win-win performance in both visual quality and semantic attributes. The code is publicly available at https://github.com/HaoZhang1018/AD-SRF.
Chinese Translation
现有的图像融合方法在视觉质量和语义属性之间难以取得平衡，因为它们的优化目标之间存在固有冲突。为了解决这一挑战，我们提出了一种用于多模态图像融合的攻击-防御语义强化框架，称为AD-SRF。通过解耦这两个方面的优化，然后无损地重新整合它们，该框架在不影响视觉质量的情况下，将丰富的语义属性注入到融合图像中。首先，我们开发了一种基于变换器的恢复视觉融合网络，作为强大的视觉基线。该网络配备了跨模态检索模块和信息恢复-融合协同损失，即使在退化条件下也能生成高度可见的融合图像。其次，我们引入了一种攻击-防御博弈机制，以增强融合图像的语义属性。攻击者学习扰动模式，以损害下游分割，而防御者则从原始输入中恢复缺失的语义，并通过不可感知的语义图将其重新整合到融合图像中。通过迭代的对抗互动，该机制促使防御者不断增强融合图像的语义属性，从而提升下游分割性能，而不影响视觉质量。大量实验表明，AD-SRF在视觉质量和语义属性方面实现了双赢性能。代码已公开发布于 https://github.com/HaoZhang1018/AD-SRF。
cs.CV / 25 / 2610.12899
A Lightweight Velocity Correction Framework for Improving Instruction Following in Few-Step Text-to-Image Generation
一种轻量级速度校正框架，用于改善少步文本到图像生成中的指令跟随
Abstract
Few-step text-to-image generators produce high-quality images with few network forward passes, yet struggle with complex prompts involving multiple subjects, interactions, spatial relations, counting, and attribute binding. Can these errors be reduced without changing the generator's parameters or increasing its forward-pass count? We propose a lightweight correction framework for a frozen 6B-parameter generator. The generator predicts velocities that determine how its intermediate image representation is updated; a corrector with fewer than 20M parameters learns residual adjustments at selected sampling steps. Because earlier corrections change the states encountered at later steps, our On-Trajectory Flow-Bridge Supervision derives residual targets at the states actually reached during corrected generation. A single corrector is shared across selected steps and runs in parallel with the generator. Only the corrector is trained, requiring neither reward optimization nor re-distillation. To our knowledge, this is the first framework for lightweight velocity-residual adaptation of fully frozen few-step text-to-image generators that preserves their original sampling schedule and forward-pass count. On our target complex-scene distribution, the framework achieves a +17.60-percentage-point NetWin over frozen Z-Image-Turbo in pairwise instruction-following evaluation, with only 26.9 ms (0.84%) additional end-to-end latency per image and no increase in backbone evaluations.
Chinese Translation
少步文本到图像生成器通过少量网络前向传播生成高质量图像，但在处理涉及多个主体、交互、空间关系、计数和属性绑定的复杂提示时存在困难。是否可以在不改变生成器参数或增加前向传播次数的情况下减少这些错误？我们提出了一种轻量级校正框架，适用于冻结的6B参数生成器。该生成器预测速度，以决定其中间图像表示的更新方式；一个参数少于2000万的校正器在选定的采样步骤中学习残差调整。由于早期的校正会改变后续步骤所遇到的状态，我们的轨迹流桥监督（On-Trajectory Flow-Bridge Supervision）在实际达到的状态中推导残差目标。一个校正器在选定步骤之间共享，并与生成器并行运行。只有校正器被训练，不需要奖励优化或再蒸馏。根据我们的了解，这是第一个轻量级速度残差适应框架，适用于完全冻结的少步文本到图像生成器，且保持其原始采样调度和前向传播次数。在我们的目标复杂场景分布上，该框架在成对指令跟随评估中实现了相较于冻结的Z-Image-Turbo提高17.60个百分点的净胜率，每幅图像仅增加26.9毫秒（0.84%）的端到端延迟，且没有增加主干评估次数。
cs.CV / 26 / 2610.12901
JMT-YOLO: Multi-task Domain Adaptation via Segmentation-Guided Consistency Learning
JMT-YOLO：通过分割引导的一致性学习实现多任务领域适应
Abstract
Domain adaptive object detection (DAOD) aims to reduce performance degradation due to distribution shifts between training (source) and testing (target) datasets. However, owing to insufficient modeling of fine-grained features and limited instance-level alignment, existing approaches become less effective under complex environmental conditions, such as extreme weather conditions and object occlusion. We overcome these challenges by proposing JMT-YOLO, a novel joint multi-task learning framework for DAOD. JMT-YOLO is based on the Mean Teacher architecture. It introduces an auxiliary semantic segmentation task and a masking mechanism to enhance the modeling of foreground-background semantic boundaries in the target domain. Specifically, fine-grained feature degradation is reduced by introducing a multiscale context-aware segmentation head equipped with a dense connection structure to effectively capture local and global contextual features. A mask-consistency loss is formulated by applying random masking operations to target images, thereby guiding the network to learn structural context and improving instance-level feature alignment. Extensive experiments on three benchmark domain adaptation tasks (Pascal VOC to Clipart1k, Cityscapes to Foggy Cityscapes, and KITTI to Cityscapes) demonstrate that JMT-YOLO achieves detection performances of 46.9%, 57.3%, and 61.0%, respectively. These results demonstrate superior performance over existing state-of-the-art and baseline methods, thereby validating the effectiveness and generalization of JMT-YOLO.
Chinese Translation
领域自适应目标检测（DAOD）旨在减少由于训练（源）数据集与测试（目标）数据集之间的分布变化而导致的性能下降。然而，由于对细粒度特征建模不足和实例级对齐有限，现有方法在复杂环境条件下（如极端天气和物体遮挡）变得不够有效。我们通过提出JMT-YOLO，克服了这些挑战，这是一种用于DAOD的新颖联合多任务学习框架。JMT-YOLO基于Mean Teacher架构，引入了辅助语义分割任务和掩膜机制，以增强目标领域中前景-背景语义边界的建模。具体而言，通过引入配备密集连接结构的多尺度上下文感知分割头，减少了细粒度特征的退化，从而有效捕捉局部和全局上下文特征。通过对目标图像应用随机掩膜操作，制定了掩膜一致性损失，从而引导网络学习结构上下文并改善实例级特征对齐。在三个基准领域适应任务（Pascal VOC到Clipart1k，Cityscapes到Foggy Cityscapes，以及KITTI到Cityscapes）上的大量实验表明，JMT-YOLO分别实现了46.9%、57.3%和61.0%的检测性能。这些结果显示出优于现有最先进和基线方法的性能，从而验证了JMT-YOLO的有效性和泛化能力。
cs.CV / 27 / 2610.12907
BLEND: Balancing Personalization vs. Generalization in Federated Vision-Language Models
BLEND：在联邦视觉-语言模型中平衡个性化与泛化
Abstract
Federated vision-language models (FedVLMs) represent one of the emerging frontiers of machine learning, aiming to integrate federated learning within the fine-tuning pipelines of vision-language models (VLMs). In this work, we introduce BLEND, a new framework designed to jointly enhance the personalization and generalization capabilities of FedVLMs. BLEND proposes a selective VLM parameter fine-tuning and aggregation strategy that consists of (i) global vision and text adapters shared across clients, and (ii) a personalized vision projection adapter tailored to each client. BLEND further introduces a novel loss function that blends the outputs of the global and personalized adapters to achieve personalization, while simultaneously promoting generalization through a principled anchor design implemented via the projection heads of the zero-shot VLM. Extensive experiments demonstrate that BLEND consistently outperforms state-of-the-art baselines, highlighting its ability to effectively balance personalization and generalization across diverse system settings. (Code is available at: https://github.com/payamsiabd/BLEND.git)
Chinese Translation
联邦视觉-语言模型（FedVLMs）代表了机器学习的一个新兴前沿，旨在将联邦学习整合到视觉-语言模型（VLMs）的微调流程中。在本研究中，我们提出了BLEND，一个旨在共同增强FedVLMs个性化和泛化能力的新框架。BLEND提出了一种选择性VLM参数微调和聚合策略，包括（i）在客户端之间共享的全局视觉和文本适配器，以及（ii）为每个客户端量身定制的个性化视觉投影适配器。BLEND进一步引入了一种新颖的损失函数，该函数将全局适配器和个性化适配器的输出进行融合，以实现个性化，同时通过在零样本VLM的投影头中实施的原则性锚设计促进泛化。大量实验表明，BLEND在各种系统设置中始终优于最新的基线，突显了其在有效平衡个性化与泛化方面的能力。（代码可在：https://github.com/payamsiabd/BLEND.git获取）
cs.CV / 28 / 2610.12926
What Egocentric Video Embeddings Encode and How to Combine Them with Inertial Sensors for Human Activity Recognition
自我中心视频嵌入编码的内容及其与惯性传感器结合用于人类活动识别的方法
Abstract
Egocentric video and body-worn inertial sensors are increasingly recorded together, and the video is usually encoded by pretrained models. What these embeddings represent when the camera sits on the head, and how they should be combined with inertial sensing, has received little attention. Across VideoMAEv2, CLIP, DINOv2 and the egocentric VC-1, the leading principal directions of the embedding are dominated by where the camera looks. Measured head orientation explains them on Ego-Exo4D for egocentric (R^2 0.49-0.70) but not exocentric cameras, and the scene layout explains them on WEAR. The part that the limbs' inertial signals predict separates activities across people. In a set-level probe, the rest separates exercise sets within a participant but not activities across people. This is consistent with separate roles: the video finds the clips of the same set, and the inertial classifier labels them. On WEAR, with temporal order removed and one random limb per one-second clip, we propagate inertial predictions over a within-participant video graph whose leading components are removed. This raises subject-wise cross-validated macro-F1 from 0.648 to 0.852. In our comparisons, propagation helps only over the video graph, and fusing the modalities in one classifier, or adding inertial edges to the graph, performs worse.
Chinese Translation
自我中心视频和佩戴式惯性传感器越来越多地一起录制，而视频通常由预训练模型进行编码。当相机位于头部时，这些嵌入表示的内容以及它们如何与惯性传感器结合，尚未受到足够关注。在 VideoMAEv2、CLIP、DINOv2 和自我中心 VC-1 中，嵌入的主要主方向主要由相机的视角决定。测量的头部朝向在自我中心（R^2 0.49-0.70）上能够解释这些嵌入，但对外部中心相机则不然，而场景布局在 WEAR 上能够解释这些嵌入。肢体的惯性信号预测的部分能够区分不同个体的活动。在一个集水平探测中，其余部分能够区分同一参与者的锻炼集，但无法区分不同个体的活动。这与各自的角色一致：视频找到同一集合的片段，而惯性分类器为其标记。在 WEAR 上，去除时间顺序并对每秒片段随机选择一个肢体，我们在一个去除主要成分的参与者内部视频图上传播惯性预测。这将受试者交叉验证的宏观 F1 从 0.648 提升至 0.852。在我们的比较中，传播仅在视频图上有效，而在一个分类器中融合模态或向图中添加惯性边缘的效果较差。
cs.CV / 29 / 2610.12941
UniDeblur: Unified Motion Deblurring via Fourier Convolution Learning
UniDeblur：通过傅里叶卷积学习实现统一运动去模糊
Abstract
Image deblurring is a fundamental task in computer vision, yet existing methods mainly specialize in isolated scenarios. Consequently, the deblurring performance of these models drops significantly across scenarios with different cameras and image contents. This motivates the development of a unified motion deblurring framework that can handle diverse blur patterns. To this end, we construct a multi-source dataset and develop a data pruning and quality control pipeline. Then, to restore blur images with various motion blur kernels across different scenarios, we propose a unified motion deblurring framework via Fourier Convolution Learning. Existing deblurring networks usually rely on pixel-level mappings and overlook the convolutional nature of blur. Thus, the network mostly fits the image content itself rather than learning the convolution process, making it difficult to handle multi-source blur images. Instead of relying on spatial mapping, we perform blur-adaptive global filter learning in Fourier space, which avoids the limited spatial support of kernel estimation. Extensive experiments demonstrate that our unified model effectively handles motion blurs across diverse sources and scenes, achieving competitive performance across both in-domain and OOD benchmarks. Code will be released at github.com/DeepMed-Lab-ECNU/Single-Image-Deblur.
Chinese Translation
图像去模糊是计算机视觉中的一项基础任务，但现有方法主要专注于孤立场景。因此，这些模型在不同相机和图像内容的场景中的去模糊性能显著下降。这促使我们开发一个能够处理多样模糊模式的统一运动去模糊框架。为此，我们构建了一个多源数据集，并开发了数据修剪和质量控制管道。然后，为了恢复具有不同运动模糊核的模糊图像，我们提出了一种通过傅里叶卷积学习实现的统一运动去模糊框架。现有的去模糊网络通常依赖于像素级映射，忽视了模糊的卷积特性。因此，网络主要适应图像内容本身，而不是学习卷积过程，使得处理多源模糊图像变得困难。我们不依赖空间映射，而是在傅里叶空间中执行模糊自适应全局滤波器学习，这避免了核估计的有限空间支持。大量实验表明，我们的统一模型能够有效处理来自不同源和场景的运动模糊，在领域内和OOD基准测试中均实现了具有竞争力的性能。代码将发布在github.com/DeepMed-Lab-ECNU/Single-Image-Deblur。
cs.CV / 30 / 2610.12942
In-Context Learning Can Help Vision-Language Models Overcome Training Priors
上下文学习可以帮助视觉-语言模型克服训练先验
Abstract
Vision-language models (VLMs) learn strong statistical regularities during training, which can make them fail to properly respond to visual evidence when input images violate those regularities. It is challenging to disambiguate these prior-induced errors from errors caused by the model missing the underlying visual capabilities. In this work, we use controlled visual in-context learning to encourage the model to focus on the visual evidence, and demonstrate that this alone is frequently sufficient to correct the errors -- suggesting that the models in fact do possess the relevant visual capabilities. Specifically, we show that across five vision-centric tasks, demonstrations yield large gains on prior-conflicting counterfactual examples ($+27.2\%$ for Qwen3-VL 235B and $+15.1\%$ for Qwen3-VL 32B) while leaving accuracy on prior-aligned examples nearly unchanged. These findings generalize across five VLM families with gains from $+9.7\%$ to $+30.3\%$. We analyze visual representations and find that demonstrations indeed redirect attention toward grounding cues while reducing reliance on cues that support the learned prior. Finally, we find that in-context learning is limited when the underlying visual capability is weak or absent. Together, these results show that current VLMs possess visual grounding capabilities that standard evaluations may overlook, underscoring the need for controlled evaluation design in diagnosing and improving model capability.
Chinese Translation
视觉-语言模型（VLMs）在训练过程中学习到强烈的统计规律，这可能导致它们在输入图像违反这些规律时无法正确响应视觉证据。区分这些由先验引起的错误与模型缺乏基本视觉能力导致的错误是具有挑战性的。在本研究中，我们使用受控的视觉上下文学习来鼓励模型关注视觉证据，并证明仅此一项通常足以纠正错误——这表明模型实际上具备相关的视觉能力。具体而言，我们展示了在五个以视觉为中心的任务中，演示在与先验相悖的反事实示例上产生了显著的增益（Qwen3-VL 235B 增加了 $+27.2\%$，Qwen3-VL 32B 增加了 $+15.1\ rac{1}{2}$），而对与先验一致的示例的准确性几乎没有影响。这些发现跨五个 VLM 家族具有普遍性，增益范围从 $+9.7\ rac{1}{2}$ 到 $+30.3 rac{1}{2}$。我们分析了视觉表征，发现演示确实将注意力重新引导到基础线索上，同时减少对支持学习先验的线索的依赖。最后，我们发现当基本视觉能力较弱或缺失时，上下文学习是有限的。综合来看，这些结果表明当前的 VLMs 具备视觉基础能力，而标准评估可能会忽视这一点，强调了在诊断和提升模型能力时需要进行受控评估设计。
cs.CV / 31 / 2610.12959
Video Codec-Guided Training-Free Token Compression for Video Large Language Models
视频编解码引导的无训练标记压缩方法用于视频大型语言模型
Abstract
Video large language models (VLLMs) rely on long visual token sequences to understand complex temporal content, which makes visual token compression important for efficient inference. Existing training-free compression methods mainly reduce the number of tokens sent to the language model, but the compression process itself can still require many visual feature comparisons, especially for intra-frame token protection and inter-frame temporal token merging. This paper proposes a video codec-guided training-free token compression method for VLLMs that uses coding information already available in compressed video as a low-cost search prior. The method aligns reference-frame dependencies, motion vectors, and residuals with the visual token grid. Residual information forms compact candidate pools for token protection, while motion information is propagated along reference-frame chains to construct localized candidates for temporal token merging between sparsely sampled frames. Local visual feature similarity comparison is then used within these candidates to make the final compression decisions, without modifying the base VLLM. Experiments on three representative VLLMs and four video-understanding benchmarks show that the proposed method achieves performance comparable to the state-of-the-art training-free baseline FlashVID, and in some settings surpasses it, while substantially reducing the feature comparisons required by token protection and temporal matching. These results demonstrate that codec information can effectively improve the accuracy-complexity tradeoff of video token compression.
Chinese Translation
视频大型语言模型（VLLMs）依赖于长时间视觉标记序列来理解复杂的时间内容，这使得视觉标记压缩对于高效推理变得至关重要。现有的无训练压缩方法主要减少发送到语言模型的标记数量，但压缩过程本身仍可能需要进行大量的视觉特征比较，尤其是在帧内标记保护和帧间时间标记合并时。本文提出了一种视频编解码引导的无训练标记压缩方法，该方法利用压缩视频中已有的编码信息作为低成本的搜索先验。该方法将参考帧依赖关系、运动矢量和残差与视觉标记网格对齐。残差信息形成紧凑的候选池以进行标记保护，而运动信息则沿参考帧链传播，以构建稀疏采样帧之间的时间标记合并的局部候选。然后在这些候选中使用局部视觉特征相似性比较来做出最终的压缩决策，而无需修改基础的VLLM。在三个具有代表性的VLLM和四个视频理解基准上的实验表明，所提出的方法在性能上可与最先进的无训练基线FlashVID相媲美，并在某些设置中超越了它，同时显著减少了标记保护和时间匹配所需的特征比较。这些结果表明，编解码信息可以有效改善视频标记压缩的准确性与复杂性之间的权衡。
cs.CV / 32 / 2610.12961
MVGraph: Benchmarking LVLMs on Evidence Composition across Multi-View Visual Graph
MVGraph：在多视角视觉图上评估大型视觉语言模型的证据组合基准
Abstract
We introduce MVGraph, a benchmark for evaluating large vision-language models (LVLMs) on multi-graph reasoning where the evidence required by a query is distributed across multiple interrelated visual graph views. Although such evidence commonly resides in source-specific graphs, temporal snapshots, or local observations, existing graph reasoning benchmarks remain confined to a single self-contained structure, assessing only graph perception and reasoning while leaving the alignment and composition of evidence across views underexplored. MVGraph comprises 8k instances spanning 8 \textit{graph-theoretic} tasks, abstracted from two multi-graph scenarios (partial and temporal graph views). Each problem is programmatically constructed and verified to be unsolvable from a single view. Beyond end-to-end evaluation, we decompose multi-graph reasoning into four complementary capabilities---Perception, Alignment, Composition, and Reasoning---and design 29 diagnostic tasks with 64k instances to probe each one, along with golden intermediate representations for evaluating the intermediate process. Across 15 LVLMs, we reveal the difficulty of multi-graph reasoning and identify cross-view edge alignment and relation-level composition as prominent bottlenecks through stage-oriented diagnostics. Meanwhile, temporal constraints and increasing view counts expose further capability boundaries. Code and data are available at https://github.com/AAAndy-Zhu/MVGraph.
Chinese Translation
我们介绍了MVGraph，这是一个用于评估大型视觉语言模型（LVLMs）在多图推理中的基准，其中查询所需的证据分布在多个相互关联的视觉图视图中。尽管这些证据通常存在于特定源的图形、时间快照或局部观察中，但现有的图推理基准仍然局限于单一自包含结构，仅评估图的感知和推理，而对跨视图的证据对齐和组合的研究则相对不足。MVGraph包含8000个实例，涵盖8个 extit{图论}任务，抽象自两种多图场景（部分和时间图视图）。每个问题都是程序化构建的，并经过验证从单一视图无法解决。除了端到端评估外，我们将多图推理分解为四个互补能力——感知、对齐、组合和推理——并设计了29个诊断任务，包含64,000个实例，以探测每个能力，同时提供黄金中间表示以评估中间过程。在15个LVLMs中，我们揭示了多图推理的困难，并通过阶段性诊断识别出跨视图边对齐和关系级组合作为显著瓶颈。同时，时间约束和视图数量的增加暴露了进一步的能力边界。代码和数据可在https://github.com/AAAndy-Zhu/MVGraph获取。
cs.CV / 33 / 2610.12963
MotionCoach: Constructing Multimodal Guidance for Video Professional Motion Editing
MotionCoach：构建视频专业动作编辑的多模态指导
Abstract
Video editing models achieve strong performance in modifying appearance, backgrounds, styles, and motions, empowered by large-scale training data. Beyond general scenarios, professional motions are significant in performance, sports, and martial arts. This paper focuses on professional motion editing, which involves fine-grained motion technique and temporal structure. Given professional motion names but lacking detailed descriptions, existing models often fail to generate professional motions, such as incorrectly lifting the front wheel under a Stoppie instruction. To address this challenge, we introduce MotionCoach, a trained multimodal agent that uses multi-turn tools and performs self-verification, producing a grounded prompt and a reference clip as reliable and sufficient multimodal guidance to inject external professional motion knowledge into video editing models. To supply training data, we build a data pipeline to construct trajectories. We cold-start the agent on MotionCoach-SFT-10k and apply Group Relative Policy Optimization (GRPO) on MotionCoach-RL-2k with rewards on the motion guidance prompt, the motion outcome video, and the format. For evaluation, we introduce VPMEBench, comprising 314 professional motion editing tasks with fine-grained references and annotations. Extensive experiments show that MotionCoach improves the overall score from 55.1 to 65.6 on Seedance 2.0 and from 51.2 to 64.5 on MiniMax-H3. Learning to construct multimodal guidance enables precise professional motion editing without retraining video editing models.
Chinese Translation
视频编辑模型在修改外观、背景、风格和动作方面表现出色，这得益于大规模的训练数据。除了常规场景外，专业动作在表演、体育和武术中具有重要意义。本文聚焦于专业动作编辑，这涉及到细粒度的动作技巧和时间结构。由于专业动作名称缺乏详细描述，现有模型往往无法生成专业动作，例如在 Stoppie 指令下错误地抬起前轮。为了解决这一挑战，我们提出了 MotionCoach，一个经过训练的多模态智能体，它使用多轮工具并进行自我验证，生成一个有根据的提示和一个参考片段，作为可靠且充分的多模态指导，将外部专业动作知识注入视频编辑模型中。为了提供训练数据，我们建立了一个数据管道来构建轨迹。我们在 MotionCoach-SFT-10k 上对智能体进行冷启动，并在 MotionCoach-RL-2k 上应用群体相对策略优化（Group Relative Policy Optimization, GRPO），奖励包括动作指导提示、动作结果视频和格式。为了评估，我们引入了 VPMEBench，包含 314 个专业动作编辑任务，配有细粒度的参考和注释。大量实验表明，MotionCoach 在 Seedance 2.0 上的整体得分从 55.1 提高到 65.6，在 MiniMax-H3 上从 51.2 提高到 64.5。学习构建多模态指导使得在不重新训练视频编辑模型的情况下实现精确的专业动作编辑成为可能。
cs.CV / 34 / 2610.12978
Learning Representations to Guide Pixel Diffusion Transformers
学习表示以引导像素扩散变换器
Abstract
Visual representations are largely developed for understanding, yet what makes them useful for generation remains unclear. We study this question in pixel diffusion transformers and uncover two key insights. First, better representation quality does not necessarily translate into greater generative utility, as externally aligned semantics can be progressively lost during pixel synthesis. Second, architectural separation with specialized pixel decoders mainly redistributes semantic and spatial modeling, without necessarily yielding representations better suited to generation. These insights suggest that a useful generative representation must be both properly formed and actively exploited. Motivated by these insights, we introduce Representation Guidance (RG), which directly forms and exploits internal representations while retaining a plain Transformer architecture. A low-dimensional bottleneck forces generation-relevant information into a compact representation, which is then explicitly used to guide decoder generation by modifying the flow-matching target according to its induced prediction offset. RG learns clearer semantic and spatial structure while scaling efficiently in both parameters and computation. On ImageNet 256x256, RG-L/16 with only 459M parameters outperforms substantially larger pixel diffusion transformers, while RG-XL/16 achieves an FID of 1.55. RG further reaches an FID of 1.74 at 512x512 and extends to text-to-image generation, achieving 0.89 on GenEval and 87.1 on DPG-Bench.
Chinese Translation
视觉表示主要是为了理解而发展，但它们在生成中的实用性尚不清楚。我们在像素扩散变换器中研究了这个问题，并揭示了两个关键见解。首先，更好的表示质量并不一定转化为更大的生成效用，因为在像素合成过程中，外部对齐的语义可能会逐渐丧失。其次，采用专门的像素解码器进行架构分离主要是重新分配语义和空间建模，而不一定产生更适合生成的表示。这些见解表明，有效的生成表示必须既形成得当，又被积极利用。基于这些见解，我们引入了表示引导（Representation Guidance, RG），它在保持简单变换器架构的同时，直接形成并利用内部表示。一个低维瓶颈将与生成相关的信息压缩到一个紧凑的表示中，然后通过根据其诱导的预测偏移修改流匹配目标，明确地用于引导解码器生成。RG在参数和计算上都能有效扩展，同时学习到更清晰的语义和空间结构。在ImageNet 256x256上，RG-L/16仅用459M参数就显著超越了更大规模的像素扩散变换器，而RG-XL/16达到了1.55的FID。RG在512x512上进一步达到1.74的FID，并扩展到文本到图像生成，在GenEval上取得了0.89，在DPG-Bench上达到了87.1。
cs.CV / 35 / 2610.12981
The Evidence Anchoring Boundary: Where Visual Interventions Matter and When They Help in Multimodal LLMs
证据锚定边界：视觉干预的重要性及其在多模态大型语言模型中的作用
Abstract
Visual interventions provide a training-free way to reduce hallucinations in multimodal large language models. However, an intervention may have little leverage at some layers and may damage a correct answer even when it changes the output. We investigate where visual interventions retain leverage and when they improve predictions. First, we apply the same visual-residual edit at successive decoder layers, matching its RMS strength across depth. Across six checkpoints from three model families, we observe a persistent drop in intervention effect and define its operational onset for this edit as the \textbf{Evidence Anchoring Boundary (EAB)}. With boundaries fixed on calibration queries, held-out effects on disjoint evaluation queries before EAB are $13.78\times$--$26.10\times$ larger than those after it. Comparison with intermediate readouts shows that the edit can lose leverage before answer preferences stabilize. Second, we compare counterfactual subtraction across presence, attribute, and relation questions. Subtraction can correct a misleading preference, leave a decision unchanged, or remove a preference supporting the correct answer. Based on these observations, we develop calibrated selective routing, which preserves the original prediction unless calibration supports an intervention for the question type. It improves full-set attribute accuracy by $1.56$--$1.81$ percentage points across three checkpoints, while universal subtraction is neutral or harmful.
Chinese Translation
视觉干预提供了一种无需训练的方法来减少多模态大型语言模型中的幻觉。然而，在某些层次上，干预可能几乎没有影响，甚至可能在改变输出的情况下损害正确答案。我们研究了视觉干预在何处保持影响力，以及何时能改善预测。首先，我们在连续的解码器层应用相同的视觉残差编辑，匹配其在深度上的均方根强度。在来自三个模型家族的六个检查点中，我们观察到干预效果持续下降，并将该编辑的操作起始点定义为 extbf{证据锚定边界（EAB）}。在校准查询上固定边界时，EAB之前在不相交的评估查询上的持出效果比EAB之后大$13.78 imes$--$26.10 imes$。与中间读出结果的比较表明，在答案偏好稳定之前，编辑可能会失去影响力。其次，我们比较了在存在性、属性和关系问题上的反事实减法。减法可以纠正误导性的偏好，保持决策不变，或移除支持正确答案的偏好。基于这些观察，我们开发了校准选择性路由，除非校准支持该问题类型的干预，否则保留原始预测。它在三个检查点上提高了全集属性准确率$1.56$--$1.81$个百分点，而通用减法则是中性的或有害的。
cs.CV / 36 / 2610.13012
Learning Features, Not Layouts: Feature-Axis Attention as a Transfer Prior for Medical Imaging
学习特征，而非布局：特征轴注意力作为医学成像的迁移先验
Abstract
Natural-image-pretrained foundation models are increasingly adapted to medical imaging, yet transfer remains unreliable under domain shift, modality heterogeneity, and label scarcity. We trace part of this fragility to an overlooked design choice. Standard vision transformers place all global modeling on self-attention over spatial tokens, an object-layout prior well suited to natural scenes but not to medical images, where the discriminative signal lies in feature correlations and texture that recur across scanners while layout does not. We ask whether the axis of global attention, spatial versus feature, is itself a lever for transferable medical representations. Existing channel-aware hybrids confound this question, adding channel attention to a spatially-mixed backbone and leaving the axes coupled and the feature axis unmeasurable in isolation. We design FAD-ViT, a backbone that strictly decouples the two axes, making feature-axis attention the primary global operator and the attention axis a controlled variable. Evaluating comprehensively across 23 classification, segmentation, and low-data tasks under matched tokenization, capacity, and protocol, we find that shifting global attention from the spatial to the feature axis leaves from-scratch performance essentially unchanged, yet markedly strengthens transfer. Identical ImageNet pretraining yields up to roughly three times the transfer gain, with consistent improvements on every held-out segmentation task at up to a quarter the parameter count and far lower compute than standard and channel-attention baselines. The advantage is largest where texture rather than geometry is discriminative, tracking greater reuse of pretrained features. The axis of global attention is thus a design lever for medical transfer, and making it channel-primary is a simple, cheap, and principled choice.
Chinese Translation
自然图像预训练的基础模型正越来越多地被应用于医学成像，但在领域转移、模态异质性和标签稀缺的情况下，迁移的可靠性仍然不足。我们将这种脆弱性的一部分归因于一个被忽视的设计选择。标准的视觉变换器将所有全局建模置于对空间标记的自注意力之上，这种对象布局先验非常适合自然场景，但不适用于医学图像，因为在医学图像中，区分信号存在于特征相关性和纹理中，这些在不同扫描仪之间反复出现，而布局则不然。我们探讨全局注意力的轴，即空间与特征，是否本身是可迁移医学表征的一个杠杆。现有的通道感知混合模型使这个问题变得复杂，它们在空间混合的主干上添加通道注意力，导致轴耦合，且特征轴无法单独测量。我们设计了FAD-ViT，一个严格解耦这两个轴的主干，使特征轴注意力成为主要的全局操作符，而注意力轴则成为一个可控变量。在匹配的标记化、容量和协议下，对23个分类、分割和低数据任务进行全面评估，我们发现将全局注意力从空间轴转移到特征轴时，从头开始的性能基本保持不变，但显著增强了迁移效果。相同的ImageNet预训练在迁移增益上可达到大约三倍的提升，并且在每个保留的分割任务上都有一致的改善，参数数量最多为标准和通道注意力基线的四分之一，计算量也远低于它们。优势在于纹理而非几何特征具有区分性时最大，追踪到预训练特征的更大重用。因此，全局注意力的轴是医学迁移的设计杠杆，而将其设置为通道优先是一种简单、经济且有原则的选择。
cs.CV / 37 / 2610.13023
Unified Generative Video Compression with Autoregressive and Bidirectional Video Diffusion
统一生成视频压缩：自回归与双向视频扩散
Abstract
Video diffusion models can reconstruct visually compelling content from severely compressed representations at ultra-low bitrates, but they impose a rigid deployment trade-off. Bidirectional (BI) models achieve high quality through dense full-sequence attention, yet require the entire sequence before reconstruction and incur substantial latency. Autoregressive (AR) models enable low-delay streaming, but their restricted temporal context reduces compression efficiency. This tension is particularly relevant when a video stream must be decoded in real time yet is also stored for later, quality-oriented playback, making AR and BI models natural choices for the live and replay stages, respectively. To support both stages, we present UniGVC, a Unified Generative Video Codec that supports two decoding modes from a single bitstream. UniGVC-BI employs a BI-DiT with dense attention for quality-oriented random-access decoding, whereas UniGVC-AR uses windowed AR-DiT for low-delay streaming. Both paths share a hierarchical latent codec that recurrently produces the same conditions with ultra-low bitrates, allowing UniGVC-BI to serve as a conditional teacher for single-step distillation of UniGVC-AR. A tiny conditional video decoder further reduces streaming latency while preserving most of the reconstruction quality. On standard 1080p benchmarks, UniGVC-BI achieves the best overall rate-distortion-perception trade-off among the evaluated methods and produces temporally stable video at rates as low as 0.0005 bits per pixel. Meanwhile, UniGVC-AR remains competitive while decoding 1080p video at 36.85 FPS, over $250\times$ faster than UniGVC-BI. Code and models will be released later.
Chinese Translation
视频扩散模型能够从严重压缩的表示中重建出视觉上引人注目的内容，适用于超低比特率，但它们在部署上存在严格的权衡。双向（BI）模型通过密集的全序列注意力实现高质量，但在重建之前需要整个序列，并且会产生显著的延迟。自回归（AR）模型支持低延迟流媒体，但其受限的时间上下文降低了压缩效率。这种矛盾在视频流必须实时解码的同时又需存储以便后续高质量播放时尤为明显，使得AR和BI模型分别成为直播和回放阶段的自然选择。为了支持这两个阶段，我们提出了UniGVC，一种统一生成视频编解码器，支持从单一比特流中进行两种解码模式。UniGVC-BI采用密集注意力的BI-DiT进行以质量为导向的随机访问解码，而UniGVC-AR则使用窗口化的AR-DiT进行低延迟流媒体传输。两条路径共享一个层次化的潜在编解码器，该编解码器以超低比特率重复产生相同的条件，使得UniGVC-BI能够作为UniGVC-AR单步蒸馏的条件教师。一个小型条件视频解码器进一步减少了流媒体延迟，同时保留了大部分重建质量。在标准的1080p基准测试中，UniGVC-BI在评估的方法中实现了最佳的整体率-失真-感知权衡，并在低至0.0005比特每像素的比率下生成时间稳定的视频。同时，UniGVC-AR在以36.85 FPS解码1080p视频时仍具竞争力，速度超过UniGVC-BI的$250 imes$。代码和模型将稍后发布。
cs.CV / 38 / 2610.13024
What Visual Representations Do World Action Models Need?
世界行动模型需要什么样的视觉表示？
Abstract
World Action Models (WAMs) inherit powerful priors from pretrained video generation models, but their variational autoencoder (VAE) representations are optimized for visual reconstruction rather than manipulation. External visual representations can improve manipulation, but existing methods couple representation choices with different integration mechanisms, obscuring their individual effects. We systematically compare visual representations, five integration mechanisms, and feature combinations under controlled settings. Our study yields three main findings. First, strong general-purpose visual representations provide broader robustness gains and can match or outperform representations pretrained for specific visual capabilities. Second, channel-wise joint prediction achieves the highest average success rate; further analyses suggest that future-feature prediction and the action expert's effective use of added visual information contribute to its gains. Third, additional sources should be selected for the task-relevant information they contribute beyond existing representations, as the same source can help one combination but harm another. However, even the best evaluated composition provides a smaller additional gain than the initial improvement from adding a strong general-purpose representation to the VAE-only baseline. We instantiate the resulting Representation Recipe as R$^2$-WAM. It achieves state-of-the-art performance on RoboCasa GR1 Tabletop with an 84.7% success rate and reaches 92.8% on RoboTwin 2.0, improving over the VAE-only baselines by 5.9 and 2.2 percentage points, respectively. Taken together, these findings suggest that future work should prioritize stronger general-purpose visual representations and more effective mechanisms for integrating them into WAMs, with feature composition serving as a supplementary route to improvement.
Chinese Translation
世界行动模型（WAMs）从预训练的视频生成模型中继承了强大的先验知识，但它们的变分自编码器（VAE）表示是针对视觉重建而优化的，而非操作。外部视觉表示可以改善操作，但现有方法将表示选择与不同的集成机制结合在一起，模糊了它们各自的影响。我们在受控环境下系统地比较了视觉表示、五种集成机制和特征组合。我们的研究得出了三个主要发现。首先，强大的通用视觉表示提供了更广泛的鲁棒性提升，并且可以与为特定视觉能力预训练的表示相匹配或超越。其次，通道级联合预测实现了最高的平均成功率；进一步分析表明，未来特征预测和行动专家有效利用附加视觉信息的能力有助于其提升。第三，额外来源应根据其对任务相关信息的贡献进行选择，因为同一来源可能对一种组合有帮助，但对另一种组合有害。然而，即使是评估的最佳组合，其额外增益也小于将强大的通用表示添加到仅使用VAE的基线所带来的初始改进。我们将结果表示配方实例化为R$^2$-WAM。它在RoboCasa GR1桌面上达到了84.7%的成功率，并在RoboTwin 2.0上达到了92.8%，分别比仅使用VAE的基线提高了5.9和2.2个百分点。综合来看，这些发现表明，未来的工作应优先考虑更强大的通用视觉表示以及更有效的集成机制，将特征组合作为改进的补充途径。
cs.CV / 39 / 2610.13027
Dataset-Specific Encoding of Image Content Underlies Cross-Dataset Failure in Brain MRI Foundation Models
数据集特定的图像内容编码导致脑MRI基础模型在跨数据集中的失败
Abstract
Frozen features of brain MRI foundation models lose accuracy on new datasets, a failure usually blamed on the features encoding where each scan came from. We show that the failure lies elsewhere. Across 17 public datasets and eight frozen representations, every representation recognizes the MRI sequence within each dataset, yet most fail on an unseen one. The cause is neither dataset encoding nor a shift of each dataset: the content itself is encoded along different directions in each dataset. This explains which datasets fail, for the sequence and for sex, and why removing the dataset signal with INLP, LEACE, CORAL or adversarial training never improves sequence transfer. Only the broadly pretrained 3DINO and SAM-Med3D encode the sequence consistently enough to transfer; the brain-specific BrainIAC and BrainFM do no better than an untrained network. Code is available at https://github.com/BrainFM/brainfm-profiler/tree/main/representation_utility.
Chinese Translation
脑MRI基础模型的冻结特征在新数据集上失去准确性，这种失败通常归因于特征编码的来源。我们表明，失败的原因在于其他方面。在17个公共数据集和八种冻结表示中，每种表示都能识别每个数据集中的MRI序列，但大多数在未见过的数据集上失败。原因既不是数据集编码，也不是每个数据集的偏移：内容本身在每个数据集中沿不同方向编码。这解释了哪些数据集在序列和性别上失败，以及为什么使用INLP、LEACE、CORAL或对抗训练去除数据集信号从未改善序列转移。只有广泛预训练的3DINO和SAM-Med3D能够一致地编码序列以实现转移；而特定于脑部的BrainIAC和BrainFM的表现不比未训练的网络好。代码可在https://github.com/BrainFM/brainfm-profiler/tree/main/representation_utility获取。
cs.CV / 40 / 2610.13035
GeoGR: Spatial-Semantic Token Compression and Group Routing for Efficient 3D VLMs
GeoGR：高效3D视觉语言模型的空间-语义标记压缩与组路由
Abstract
Multi-view representations enable pretrained image and video vision-language models to understand 3D scenes, but their long visual token sequences make inference costly. Redundancy arises both from repeated observations of the same scene regions across views and from processing all visual tokens at every LLM layer, regardless of their relevance to the current computation. We propose GeoGR, a two-stage framework that jointly compresses scene representations and visual computation. At the projector level, GeoSemZip consolidates cross-view observations into 3D voxel tokens and combines semantic saliency with spatial coverage to construct a compact scene representation. Unselected voxel features are aggregated into contextual anchors to retain scene information that would otherwise be discarded. At the LLM level, we identify an effective visual computation window and observe that query-conditioned token relevance remains locally stable but changes over longer layer intervals. Guided by these findings, GroupRoute combines late entry and early exit with recoverable group-wise skipping. Low-relevance tokens bypass complete LLM layers while retaining their hidden states and can rejoin the active group at subsequent anchor layers selected under a computation budget. For compression-aware post-training, we construct a compact dataset of 67K samples by retaining 30% of the original training pool. With this protocol, GeoGR retains 95.2% and 96.8% of the dense post-trained models' performance on LLaVA-OneVision-7B and Video-3D LLM-7B, respectively, while accelerating LLM prefill by $6.8\times$ and $6.1\times$. These results demonstrate the consistent effectiveness of GeoGR across 3D VLMs with and without explicit 3D spatial encoding, highlighting a favorable performance-efficiency trade-off. The code is available at https://github.com/EIT-NLP/GeoGR.
Chinese Translation
多视角表示使得预训练的图像和视频视觉语言模型能够理解3D场景，但其长视觉标记序列使得推理成本高昂。冗余来自于在不同视角中对相同场景区域的重复观察，以及在每个大语言模型（LLM）层中处理所有视觉标记，而不考虑它们与当前计算的相关性。我们提出了GeoGR，一个两阶段框架，联合压缩场景表示和视觉计算。在投影器层面，GeoSemZip将跨视角观察整合为3D体素标记，并结合语义显著性与空间覆盖构建紧凑的场景表示。未选择的体素特征被聚合为上下文锚点，以保留本应被丢弃的场景信息。在LLM层面，我们识别出一个有效的视觉计算窗口，并观察到查询条件下的标记相关性在局部保持稳定，但在较长的层间隔内发生变化。在这些发现的指导下，GroupRoute结合了晚进入和早退出的策略，并实现了可恢复的组跳过。低相关性的标记可以绕过完整的LLM层，同时保留其隐藏状态，并可以在后续根据计算预算选择的锚层中重新加入活跃组。为了进行压缩感知的后训练，我们通过保留原始训练池的30%构建了一个包含67K样本的紧凑数据集。通过这一协议，GeoGR在LLaVA-OneVision-7B和Video-3D LLM-7B上分别保留了95.2%和96.8%的密集后训练模型性能，同时加速了LLM预填充速度，分别达到6.8倍和6.1倍。这些结果展示了GeoGR在具有和不具有显式3D空间编码的3D视觉语言模型中的一致有效性，突显了良好的性能与效率的权衡。代码可在https://github.com/EIT-NLP/GeoGR获取。
cs.CV / 41 / 2610.13045
Redirecting Teacher Supervision: Temporal-Response-Guided Latent Distillation for Video-Language Models
重定向教师监督：基于时间响应引导的潜在蒸馏用于视频语言模型
Abstract
On-policy distillation (OPD) transfers capabilities through teacher predictions on student-generated sequences, as studied by GKD and MiniLLM for language models and Video-OPD for video tasks. We ask how internal supervision can further support temporal learning. Direct hidden-state alignment does not explicitly use the student's temporal responses to choose a supervision direction. We propose Temporal-Response Guidance (TRG), which measures representation responses to local temporal perturbations, adds the teacher correction's projection onto their span back to the original correction, and restores its original norm. The projection is not normalized separately, and correction components outside the response space are retained. With an orthogonal tangent projector and a detached target, TRG changes the supervision direction while preserving the cosine-loss value and aligned-representation gradient norm at target construction. On TempCompass, Video-MME, MVBench, and Video-MMMU, TRG improves the 4B student's macro accuracy by 2.86 percentage points over the reported output-only OPD baseline.
Chinese Translation
在政策蒸馏（On-policy distillation, OPD）中，通过教师对学生生成序列的预测来转移能力，GKD 和 MiniLLM 在语言模型中进行了相关研究，而 Video-OPD 则用于视频任务。我们探讨内部监督如何进一步支持时间学习。直接的隐藏状态对齐并未明确利用学生的时间响应来选择监督方向。我们提出了时间响应引导（Temporal-Response Guidance, TRG），该方法测量对局部时间扰动的表示响应，将教师修正的投影添加回其原始修正，并恢复其原始范数。投影不单独进行归一化，且保留响应空间外的修正成分。通过正交切线投影器和独立目标，TRG 在保持余弦损失值和目标构建时对齐表示梯度范数的同时改变监督方向。在 TempCompass、Video-MME、MVBench 和 Video-MMMU 上，TRG 将 4B 学生的宏观准确率提高了 2.86 个百分点，相较于报告的仅输出 OPD 基线。
cs.CV / 42 / 2610.13050
NOFACE3D: NOise-scheduled Focused Anchoring and Controlled Enhancement for 3D Generation
NOFACE3D：噪声调度聚焦锚定与控制增强的3D生成
Abstract
Recent image-to-3D models generate complete, high-quality assets from a single image, with global shape and appearance that are already convincing; what remains is the fine scale, where small, thin, elongated, or detail-rich regions are reconstructed softly and under-resolved. This gap can arise in generators with fixed-resolution spatial grids: the same spatial sampling density is used for a large, smooth, low-curvature surface and for a thin, high-curvature structure that carries most of the perceived detail. To this end, we introduce NOFACE3D, a training-free framework that refines these detail-critical regions while preserving the rest of the asset. Given a generated mesh, its input image, and a refinement box, NOFACE3D extracts the selected submesh, scales it up to fill the generator's input volume, and re-encodes it into the full-resolution latent grid. Conditioned on the image crop, it refines the region through flow-based sampling with a spatially varying noise schedule that anchors boundary latents. The refined region is then integrated back into the original asset, preserving the surrounding geometry and appearance. We implement NOFACE3D on TRELLIS.2 and its sparse voxel representation; all of it operates purely at inference time on the frozen pretrained generator. We evaluate NOFACE3D on PAniC-3D, TexVerse and BigDetail, a new benchmark of 339 assets with annotated detail-critical regions. On all three, it generates sharper geometry and finer texture detail than the baselines. Project page: https://idkcouscous.github.io/NoFACE3D/
Chinese Translation
近期的图像到3D模型能够从单张图像生成完整的高质量资产，其全球形状和外观已经相当令人信服；而剩下的问题是精细尺度，在这一尺度上，小的、细长的或富有细节的区域被柔和且不足以解析地重建。这一差距可能出现在具有固定分辨率空间网格的生成器中：对于一个大而平滑、低曲率的表面和一个细而高曲率的结构，使用的是相同的空间采样密度，而后者承载了大部分感知细节。为此，我们提出了NOFACE3D，这是一个无训练框架，能够在保留资产其余部分的同时，精细化这些关键细节区域。给定生成的网格、其输入图像和一个精细化框，NOFACE3D提取所选的子网格，将其放大以填充生成器的输入体积，并重新编码到全分辨率的潜在网格中。基于图像裁剪，它通过流基采样和空间变化的噪声调度来精细化该区域，并锚定边界潜变量。然后，精细化的区域被重新整合回原始资产中，保留周围的几何形状和外观。我们在TRELLIS.2及其稀疏体素表示上实现了NOFACE3D；所有操作均在冻结的预训练生成器上进行推理时完成。我们在PAniC-3D、TexVerse和BigDetail（一个包含339个带注释的关键细节区域的新基准）上评估NOFACE3D。在这三个基准上，它生成的几何形状更为锐利，纹理细节更为精细。项目页面：https://idkcouscous.github.io/NoFACE3D/
cs.CV / 43 / 2610.13053
UniMergeTrack: Hierarchical Model Merging for General RGB-X Visual Object Tracking
UniMergeTrack：用于通用 RGB-X 视觉目标跟踪的层次模型合并
Abstract
RGB-X tracking leverages complementary modalities, such as thermal infrared, event, or depth, to improve robustness under challenging conditions. Existing general-purpose RGB-X trackers typically follow two paradigms: independent task-specific training and joint training across modality combinations. However, the former suffers from knowledge isolation and model redundancy, while the latter introduces optimization interference across heterogeneous tasks. In this work, we present UniMergeTrack, a new model merging paradigm that preserves task-specific optimization while enabling effective cross-task knowledge integration. UniMergeTrack first trains modality-specific trackers independently and then performs hierarchical model merging. Specifically, transferable backbones are merged across RGB-X tasks via high-pass-filtered merging, and complemented by lightweight task-specific experts that preserve modality-specific cues. Tracking heads are merged within each task using optimal transport-guided channel alignment to mitigate hidden channel mismatch. A subsequent distribution alignment fine-tuning stage further improves compatibility between the unified backbone and task-specific heads. Extensive experiments on the LasHeR, RGBT234, VisEvent, VOT-RGBD22, COESOT, and CDTB benchmarks demonstrate that UniMergeTrack achieves state-of-the-art or competitive performance across RGB-Thermal, RGB-Event, and RGB-Depth tracking tasks, while exhibiting strong generalization to unseen datasets. The source code will be released upon acceptance. The source code of this paper will be released on https://github.com/Event-AHU/OpenEvTracking.
Chinese Translation
RGB-X 跟踪利用热红外、事件或深度等互补模态，在挑战性条件下提高鲁棒性。现有的通用 RGB-X 跟踪器通常遵循两种范式：独立的任务特定训练和跨模态组合的联合训练。然而，前者存在知识孤立和模型冗余的问题，而后者则引入了异构任务之间的优化干扰。在本研究中，我们提出了 UniMergeTrack，一种新的模型合并范式，既保留了任务特定的优化，又实现了有效的跨任务知识整合。UniMergeTrack 首先独立训练模态特定的跟踪器，然后进行层次模型合并。具体而言，通过高通滤波合并跨 RGB-X 任务合并可迁移的主干，并辅以轻量级的任务特定专家，以保留模态特定线索。跟踪头在每个任务内通过最优运输引导的通道对齐进行合并，以减轻隐藏通道不匹配的问题。随后，分布对齐微调阶段进一步提高了统一主干与任务特定头之间的兼容性。在 LasHeR、RGBT234、VisEvent、VOT-RGBD22、COESOT 和 CDTB 基准上的大量实验表明，UniMergeTrack 在 RGB-热红外、RGB-事件和 RGB-深度跟踪任务中实现了最先进或具有竞争力的性能，同时对未见数据集展现出强大的泛化能力。源代码将在论文接受后发布。本文的源代码将发布在 https://github.com/Event-AHU/OpenEvTracking 上。
cs.CV / 44 / 2610.13056
A4-MR: Abnormality-aware Anatomical Alignment for MR
A4-MR：面向异常的解剖对齐用于磁共振成像
Abstract
Recent advances in large-scale foundation models have enabled strong generalization across a wide range of clinical and research tasks in magnetic resonance imaging (MRI). However, existing vision-language MRI models remain limited in anatomy-specific localization and pathology-aware representation, particularly in real-world settings. We introduce A4-MR, a 3D vision-language foundation model trained on approximately 50,000 MRI studies comprising 491,000 series, roughly half of which are paired with radiology reports. A4-MR is designed to: (1) learn anatomy localization through large-scale pretraining with noisy labels; (2) align image and text representations at the anatomical level; and (3) learn distinct representations of healthy and abnormal anatomy. Across a diverse set of downstream tasks, A4-MR outperforms existing foundation models, including a +1.7 AUROC improvement over MedSigLIP for abnormality detection. Using a frozen encoder, A4-MR also surpasses an end-to-end task-specific MedNeXt model for knee segmentation by +1.4 Dice and performs comparably to the best task-specific model for abdominal organ segmentation on AMOS, within 0.4 Dice. Finally, without anatomical prompts, A4-MR localizes 3D anatomy comparably to MedSAM2 provided with ground-truth bounding boxes. Together, these results demonstrate that anatomy-level multimodal alignment and abnormality-aware representation learning can enable a single MRI foundation model to support localization, segmentation, and pathology-related tasks.
Chinese Translation
近期大型基础模型的进展使得在磁共振成像（MRI）领域的广泛临床和研究任务中实现了强大的泛化能力。然而，现有的视觉-语言MRI模型在解剖特定的定位和病理意识的表征方面仍然有限，尤其是在现实世界的应用中。我们提出了A4-MR，这是一个基于3D视觉-语言的基础模型，训练于约50,000个MRI研究，涵盖491,000个系列，其中大约一半与放射学报告配对。A4-MR旨在：(1) 通过带噪声标签的大规模预训练学习解剖定位；(2) 在解剖层面对齐图像和文本表征；(3) 学习健康与异常解剖的不同表征。在一系列多样的下游任务中，A4-MR超越了现有的基础模型，包括在异常检测上比MedSigLIP提高了1.7 AUROC。使用冻结编码器，A4-MR还在膝关节分割任务中超越了一个端到端的任务特定MedNeXt模型，提升了1.4 Dice，并且在AMOS的腹部器官分割任务中与最佳任务特定模型的表现相当，差距仅为0.4 Dice。最后，在没有解剖提示的情况下，A4-MR的3D解剖定位与提供真实边界框的MedSAM2相当。这些结果共同表明，解剖层面的多模态对齐和面向异常的表征学习能够使单一的MRI基础模型支持定位、分割和病理相关任务。
cs.CV / 45 / 2610.13066
Evolving Creatives with Clicks: CTR-Driven Advertising image Editing via Attribute Disentanglement and Preference Alignment
通过点击演化创意：基于点击率驱动的广告图像编辑方法，结合属性解耦与偏好对齐
Abstract
The optimization objective for e-commerce advertising creatives has shifted from aesthetic quality to online business metrics like click-through rate (CTR). Recent methods generate appealing ad images from product images by constructing CTR-based reward signals and refining prompt generation via Reinforcement Learning. Such paradigms equally penalize all attributes within an underperforming image, even though some are actually beneficial to clicks. In this work, we propose Evolving Creatives with Clicks (ECC), which reformulates creative optimization as fine-grained editing rather than generation from scratch. ECC selectively refines attributes that hurt CTR while preserving those that contribute positively. Specifically, faced with a rich attribute space across massive images, ECC first identifies 28 attributes significantly correlated with CTR via ANOVA and Random Forest regression. Based on attribute contrasts between images with disparate CTRs, we instruction-tune a prompt model to generate editing suggestions, then refine it with real click preferences since reference-derived suggestions are not uniformly beneficial. Furthermore, we introduce Consistent Preference Optimization (CPO) as ECC's preference optimization strategy, which enforces consistency across semantically equivalent shuffled suggestions to mitigate DPO overfitting and improve preference generalization. Extensive experiments demonstrate that ECC significantly improves the CTR of ad images, outperforming state-of-the-art models by a substantial margin. The complete data pipeline, including attribute extraction and analysis, the resulting dataset, and source code will be released upon acceptance.
Chinese Translation
电子商务广告创意的优化目标已从美学质量转向在线业务指标，如点击率（CTR）。最近的方法通过构建基于CTR的奖励信号并通过强化学习优化提示生成，从产品图像生成吸引人的广告图像。这些范式对表现不佳的图像中的所有属性施加相同的惩罚，尽管其中一些实际上对点击是有益的。在本研究中，我们提出了通过点击演化创意（ECC），将创意优化重新定义为精细编辑，而不是从头生成。ECC选择性地优化那些对CTR有负面影响的属性，同时保留那些积极贡献的属性。具体而言，面对庞大图像中的丰富属性空间，ECC首先通过方差分析（ANOVA）和随机森林回归识别出与CTR显著相关的28个属性。基于不同CTR图像之间的属性对比，我们对提示模型进行指令调优，以生成编辑建议，然后根据真实点击偏好进行优化，因为参考生成的建议并不总是有益。此外，我们引入了一致性偏好优化（CPO）作为ECC的偏好优化策略，强制在语义上等价的随机建议之间保持一致性，以减轻DPO过拟合并提高偏好的泛化能力。大量实验表明，ECC显著提高了广告图像的CTR，远超最先进的模型。完整的数据管道，包括属性提取与分析、生成的数据集和源代码将在接受后发布。
cs.CV / 46 / 2610.13088
Volume Rendering with Metric--Phase Fields for Thin Structures and Sign-Ambiguous Geometry
基于度量-相位场的薄结构和符号模糊几何体的体积渲染
Abstract
Neural implicit surface methods that couple volume rendering with signed distance functions (SDFs), such as VolSDF and NeuS, achieve high-quality multi-view reconstruction for watertight objects. However, these approaches rely on a globally consistent inside-outside partition of space. This assumption becomes fragile for thin structures (e.g., sheets, tubes, and thin walls) and can break down for sign-ambiguous geometry (e.g., surfaces with boundary, non-manifold junctions, and closely spaced surface components), often resulting in spurious closures or loss of fine structures. Unsigned distance field (UDF) formulations relax the global sign requirement and better support non-watertight geometry, but are substantially harder to optimize. We propose PhaseSurf, a neural volume rendering framework built on Metric-Phase Fields (MPFs). MPFs decompose geometry into a nonnegative metric field capturing unsigned proximity and a continuous phase field providing a soft sign cue. By coupling the two through residual phase injection, MPFs yield a signed composite implicit function with stable near-surface gradients while retaining UDF-like robustness in sign-ambiguous regions. This design enables direct iso-surface extraction using standard Marching Cubes and avoids the failure modes of both pure SDF- and UDF-based formulations. We further introduce MPF-specific training losses that enforce metric consistency, stabilize phase transitions near the interface, and suppress spurious geometry in free space. Experiments on multi-view datasets with thin structures and sign-ambiguous geometry show that PhaseSurf produces more stable reconstructions and higher-quality geometry than UDF-based methods and recent 3DGS pipelines, while avoiding artifacts typical of SDF-based approaches.
Chinese Translation
将体积渲染与有符号距离函数（SDF）相结合的神经隐式表面方法，如VolSDF和NeuS，能够为密闭物体实现高质量的多视图重建。然而，这些方法依赖于空间的全局一致的内外部分划分。这一假设在薄结构（例如，薄片、管道和薄壁）中变得脆弱，并且在符号模糊几何体（例如，带边界的表面、非流形连接和紧密间隔的表面组件）中可能会失效，常常导致虚假闭合或细结构的丢失。无符号距离场（UDF）公式放宽了全局符号要求，更好地支持非密闭几何体，但优化难度显著增加。我们提出了PhaseSurf，一个基于度量-相位场（MPF）的神经体积渲染框架。MPF将几何体分解为一个捕捉无符号接近度的非负度量场和一个提供软符号线索的连续相位场。通过残差相位注入将两者耦合，MPF生成具有稳定近表面梯度的有符号复合隐式函数，同时在符号模糊区域保持类似UDF的鲁棒性。该设计使得可以使用标准的Marching Cubes直接提取等值面，并避免了纯SDF和UDF公式的失效模式。我们进一步引入了特定于MPF的训练损失，以强制度量一致性，稳定界面附近的相位转变，并抑制自由空间中的虚假几何体。在具有薄结构和符号模糊几何体的多视图数据集上的实验表明，PhaseSurf产生比基于UDF的方法和最近的3DGS管线更稳定的重建和更高质量的几何体，同时避免了基于SDF方法的典型伪影。
cs.CV / 47 / 2610.13104
Robust Vision-Language Navigation with Awareness of Instruction-Execution Progress
具有指令执行进度意识的鲁棒视觉-语言导航
Abstract
In Vision-Language Navigation (VLN) tasks involving multi-step instructions as guidance, an agent must associate its observations with the relevant instruction content at each decision point. Most graph-based methods commonly match candidate trajectories with the entire instruction, allowing already executed content to interfere with subsequent decisions. Although instruction-execution progress provides a direct cue for separating executed and pending content, existing progress estimation approaches often rely on explicit progress supervision, sophisticated progress-learning stages, or costly online reasoning with LLMs. In this paper, we propose a robust VLN framework with Awareness of Instruction-Execution Progress (VLN-AIEP), which couples progress estimation with navigation decision-making. VLN-AIEP associates the visited trajectory with the instruction to suppress embeddings corresponding to executed content, and then matches candidate locations with the pending content. This coupling makes correct navigation decisions depend on accurate progress estimation, allowing the progress estimator to learn directly from navigation decision labels without explicit progress annotations. To support progress estimation, we construct a landmark-anchored trajectory representation that encodes the semantic labels and viewpoint-relative geometry of instruction-referred landmarks. These landmarks serve as spatial references for associating the visited trajectory with executed instruction content. Extensive experiments on R2R, RxR, CVDN, and R2R-CE demonstrate that the proposed method achieves favorable performance across multiple metrics.
Chinese Translation
在涉及多步骤指令作为指导的视觉-语言导航（VLN）任务中，代理必须在每个决策点将其观察与相关的指令内容关联起来。大多数基于图的方法通常将候选轨迹与整个指令匹配，这使得已执行的内容干扰后续决策。尽管指令执行进度为区分已执行和待执行内容提供了直接线索，但现有的进度估计方法往往依赖于显式的进度监督、复杂的进度学习阶段或与大型语言模型（LLMs）进行成本高昂的在线推理。在本文中，我们提出了一种具有指令执行进度意识的鲁棒视觉-语言导航框架（VLN-AIEP），将进度估计与导航决策相结合。VLN-AIEP将访问的轨迹与指令关联，以抑制与已执行内容对应的嵌入，然后将候选位置与待执行内容匹配。这种结合使得正确的导航决策依赖于准确的进度估计，从而使进度估计器能够直接从导航决策标签中学习，而无需显式的进度注释。为了支持进度估计，我们构建了一种基于地标的轨迹表示，编码了指令所提及地标的语义标签和视点相对几何。这些地标作为空间参考，用于将访问的轨迹与已执行的指令内容关联。我们在R2R、RxR、CVDN和R2R-CE上的大量实验表明，所提出的方法在多个指标上都取得了良好的性能。
cs.CV / 48 / 2610.13111
What Do Visual-Head Discovery Scores Actually Find? Attention Source, Necessity, and Random-Control Pitfalls in Vision-Language Models
视觉头发现评分究竟发现了什么？视觉-语言模型中的注意力来源、必要性和随机控制的陷阱
Abstract
Attention heads that attend to image regions relevant to a question are increasingly used to steer, ground and interpret vision-language models (VLMs) and to reduce their hallucinations. They are mostly found by ranking heads with an attention- pattern score and validated by steering or masking them. We ask how the choices inside such a discovery score shape the heads it returns, and which of the two validations identifies them. On Qwen3-VL-8B-Instruct we vary the query type, the discovery corpus and the token positions whose attention is read (the attention source) in a full 2 x 3 x 2 design, and test all 12 head-sets for steerability (CERS: Contrastive Evidence-Region Steering score) and necessity (NCS: Necessity Con- trast Score) against layer-matched random head-sets. (1) The attention source and the query type reshape the selected heads far more than the corpus: changing the source keeps a median of 49 of the top-100 heads, the query 55 and the cor- pus 84, while two halves of the same data share 96-99. (2) Necessity follows the attention source: head-sets discovered from query-span attention are more nec- essary than last-token ones (difference 0.083, p=6 x 10^-10), and the two re- ferring + query-span sets lie 17-19 standard deviations above 250 layer-matched random sets (Holm-adjusted p=0.048). The referring + query-span effect re- plicates on several recent VLMs. (3) Steerability is not specific to the discovered heads: no discovered head-set steers better than its layer-matched random sets (rank p >= 0.14), and random sets drawn from the same or from any layer span the discovered sets' scores. Necessity therefore separates the discovery settings, while steering at this strength does not. We turn these findings into a protocol for reporting visual-head discoveries.
Chinese Translation
关注与问题相关的图像区域的注意力头越来越多地用于引导、定位和解释视觉-语言模型（VLMs），并减少它们的幻觉。它们主要通过对头部进行注意力模式评分并通过引导或屏蔽来验证。我们探讨这种发现评分中的选择如何影响返回的头部，以及哪种验证能够识别它们。在 Qwen3-VL-8B-Instruct 上，我们在一个完整的 2 x 3 x 2 设计中变化查询类型、发现语料库和读取注意力的标记位置（注意力来源），并测试所有 12 个头部集的引导性（CERS：对比证据区域引导评分）和必要性（NCS：必要性对比评分），与层匹配的随机头部集进行比较。（1）注意力来源和查询类型对所选头部的重塑远远超过语料库：改变来源保持了前 100 个头部中的中位数 49 个，查询为 55 个，语料库为 84 个，而同一数据的两个部分共享 96-99。（2）必要性遵循注意力来源：从查询跨度注意力发现的头部集比最后标记的头部集更为必要（差异 0.083，p=6 x 10^-10），而两个引用 + 查询跨度集的标准差在 250 个层匹配的随机集之上高出 17-19 个标准差（Holm 调整后的 p=0.048）。引用 + 查询跨度效应在几种最近的 VLMs 上得到了复制。（3）引导性并不特定于发现的头部：没有发现的头部集的引导性优于其层匹配的随机集（排名 p >= 0.14），而从同一层或任何层中抽取的随机集跨越了发现集的评分。因此，必要性区分了发现设置，而这种强度的引导性则没有。我们将这些发现转化为报告视觉头发现的协议。
cs.CV / 49 / 2610.13120
Replaying the Movie in Your Mind Decoding Dynamic Visual Perception and Recall from EEG
在脑海中重放电影：从脑电图解码动态视觉感知与回忆
Abstract
Decoding dynamic visual content from EEG has advanced rapidly for visual perception, yet it remains unclear which aspects of neural representations are preserved when people internally replay dynamic scenes. We introduce PAIR, an EEG benchmark collected from 20 participants using an interleaved, clip-wise paradigm that pairs naturalistic video viewing with immediate visual recall. This design enables a comparison between externally driven perception and internally generated recall for the same stimulus content within a fixed trial sequence. We benchmark multiple decoding models on hierarchical semantic classification at both fine-grained and coarse-grained levels, meta-feature decoding tasks covering motion, color, object number, face, and human presence, and a state-decoding task that discriminates perception from recall. Within-subject validation results show numerically similar performance for several model--task combinations, alongside higher perception accuracy in two coarse-grained comparisons after multiple-comparison correction. In contrast, perception and recall are highly separable at the state level under the present validation protocol. State accuracy remains high after alpha-band ablation, although this does not exclude ocular or other task-related contributions.
Chinese Translation
从脑电图解码动态视觉内容在视觉感知领域取得了快速进展，但当人们在内部重放动态场景时，神经表征的哪些方面得以保留仍不清楚。我们引入了PAIR，这是一个从20名参与者收集的脑电图基准数据集，采用交错的剪辑范式，将自然视频观看与即时视觉回忆相结合。该设计使得在固定试验序列中比较外部驱动的感知与内部生成的回忆成为可能。我们在细粒度和粗粒度层面的层次语义分类、涵盖运动、颜色、物体数量、面孔和人类存在的元特征解码任务，以及一个区分感知与回忆的状态解码任务上对多个解码模型进行了基准测试。被试内验证结果显示，多个模型-任务组合的性能数值相似，同时在经过多重比较校正后，两个粗粒度比较中的感知准确率更高。相反，在当前验证协议下，感知与回忆在状态层面上高度可分。在进行α波段消融后，状态准确率仍然较高，尽管这并不排除眼动或其他任务相关贡献的可能性。
cs.CV / 50 / 2610.13122
GuidedNav: Shaping Vision-Language Navigation Representations through Pre-action Attention and Spatial Guidance
GuidedNav：通过预行动注意力和空间引导塑造视觉-语言导航表征
Abstract
Vision-and-language navigation (VLN) requires an agent to track progress through an extended instruction, identify landmarks relevant to its current objective, and understand the surrounding spatial structure to act. End-to-end navigation policies typically rely on action supervision to implicitly learn these capabilities. Without explicit guidance on the evidence underlying each decision, however, policies can overfit to correlations predictive of training actions rather than learn the task-relevant grounding and geometry needed for generalization. Our insight is to guide the representations that connect instructions to actions: identifying what remains to be done determines which visual evidence matters, while understanding the geometry around that evidence informs how to move. We introduce GuidedNav, a unified framework that explicitly shapes these representations through training-time guidance. Subtask and landmark guidance steer native pre-action attention toward the active instruction segment and relevant visual regions, grounding action prediction in the current objective. Spatial guidance aligns the policy's internal representations with geometric features that capture long-horizon context and local spatial structure. Together, these objectives strengthen navigation capabilities while avoiding the inference overhead of intermediate textual reasoning and online geometric encoding. We also introduce Guided-R2R and Guided-RxR, two annotation-enriched navigation datasets that support more interpretable VLN research. GuidedNav establishes state-of-the-art performance among the compared monocular RGB policies, improving success rates by 8.1% and 13.1% relative to the strongest prior results on R2R-CE and RxR-CE val-unseen, respectively, while achieving faster inference. Extensive experiments further demonstrate the effectiveness and real-world transferability of our approach.
Chinese Translation
视觉与语言导航（VLN）要求代理在扩展指令中跟踪进展，识别与当前目标相关的地标，并理解周围的空间结构以便采取行动。端到端导航策略通常依赖于行动监督来隐式学习这些能力。然而，在缺乏对每个决策背后证据的明确指导的情况下，策略可能会过拟合于与训练行动相关的预测性关联，而不是学习任务相关的基础和几何结构，这对于泛化是必需的。我们的见解是引导连接指令与行动的表征：确定尚需完成的任务决定了哪些视觉证据是重要的，而理解该证据周围的几何结构则告知如何移动。我们提出了GuidedNav，这是一个统一框架，通过训练时的指导明确塑造这些表征。子任务和地标指导将本地预行动注意力引导至活跃的指令片段和相关的视觉区域，将行动预测与当前目标相结合。空间引导将策略的内部表征与捕捉长时间范围上下文和局部空间结构的几何特征对齐。综合这些目标，增强了导航能力，同时避免了中间文本推理和在线几何编码的推理开销。我们还引入了Guided-R2R和Guided-RxR，这两个注释丰富的导航数据集支持更具可解释性的VLN研究。GuidedNav在比较的单目RGB策略中建立了最先进的性能，相较于R2R-CE和RxR-CE val-unseen上最强的先前结果，成功率分别提高了8.1%和13.1%，同时实现了更快的推理。大量实验进一步证明了我们方法的有效性和现实世界的可转移性。
cs.CV / 51 / 2610.13143
Measurable Semantic 3D Gaussian Splatting with Joint Region and Depth Supervision for Interactive Building Inspection
可测量的语义三维高斯点云渲染与联合区域和深度监督用于交互式建筑检查
Abstract
Image-based 3D reconstruction provides a scalable representation for building inspection and maintenance. However, photorealistic reconstruction alone does not fully provide the geometric measurability, semantic structure, and interactive information access required for engineering applications. This study proposes a measurable semantic 3D Gaussian Splatting (3DGS) framework with region and depth supervision for interactive building inspection from multi-view imagery. Mask-based region supervision suppresses irrelevant backgrounds and focuses optimization on target regions. Point-wise depth supervision guides Gaussian centers toward physical surfaces using reference geometry derived from multi-view stereo (MVS) point clouds, providing geometric anchors for measuring physical dimensions and areas. A semantic Gaussian field associates learnable class features with individual Gaussian primitives, enabling semantic rendering and component-level inspection. These capabilities are operationalized through AEC-SplatTwin, an interactive building inspection platform supporting real-time visualization, semantic analysis, quantitative measurement, and multimodal large language model (MLLM)-assisted interaction. On the synthetic UAV-inspection QuakeCity and smartphone-captured HKUPB datasets, the framework reduces Gaussian-center depth MAE by 58.1% and 86.8%, while its semantic field achieves mIoU values of 0.8529 and 0.8642. On HKUPB, the mean relative errors for height, width, and area measurements are 0.33%, 1.07%, and 0.60%. These results show that the proposed framework extends 3DGS beyond visual reconstruction into a measurable, semantically structured, and interactive engineering representation for building inspection.
Chinese Translation
基于图像的三维重建为建筑检查和维护提供了一种可扩展的表示。然而，光真实的重建并不能完全满足工程应用所需的几何可测量性、语义结构和交互信息访问。本研究提出了一种可测量的语义三维高斯点云渲染（3DGS）框架，结合区域和深度监督，旨在通过多视角图像进行交互式建筑检查。基于掩膜的区域监督抑制无关背景，集中优化于目标区域。逐点深度监督利用从多视角立体（MVS）点云中提取的参考几何，引导高斯中心朝向物理表面，为测量物理尺寸和面积提供几何锚点。语义高斯场将可学习的类别特征与单个高斯原语关联，使得语义渲染和组件级检查成为可能。这些功能通过AEC-SplatTwin这一交互式建筑检查平台实现，支持实时可视化、语义分析、定量测量和多模态大语言模型（MLLM）辅助交互。在合成的无人机检查QuakeCity和智能手机捕获的HKUPB数据集上，该框架将高斯中心深度的平均绝对误差（MAE）分别降低了58.1%和86.8%，而其语义场的mIoU值达到了0.8529和0.8642。在HKUPB数据集上，测量高度、宽度和面积的平均相对误差分别为0.33%、1.07%和0.60%。这些结果表明，所提出的框架将3DGS的应用扩展到可测量、语义结构化和交互式的建筑检查工程表示中。
cs.CV / 52 / 2610.13154
From Glance to Focus: Mitigating Recognition Laziness in MLLMs for Fine-Grained Recognition
从一瞥到聚焦：缓解多模态大语言模型在细粒度识别中的识别懒惰
Abstract
Multimodal Large Language Models (MLLMs) generally fall short of expectations in Fine-Grained Visual Recognition (FGVR), although they exhibit promising FGVR foundations, i.e., visual perception and broad category knowledge. Most existing studies seek to remedy this deficiency by adopting domain-specific post-training, at the risk of compromising generalization to unseen domains. Different from these efforts, in this paper, we revisit the FGVR mechanism underlying MLLMs from both the empirical and theoretical perspectives and uncover an inference-time phenomenon termed FGVR Laziness. In brief, FGVR Laziness refers to the insufficient association between visual cues and category concepts under certain candidate category lists and recognition guidance, leaving the FGVR foundations of MLLMs largely under-utilized. To mitigate FGVR Laziness, we propose Laziness-resIstant Visual rEcognition (LIVE), a training-free framework which jointly refines candidate lists and recognition guidance to elicit the latent FGVR capability of MLLMs. Specifically, LIVE first performs query-adaptive candidate inference to derive a compact yet reliable candidate list. Meanwhile, LIVE constructs category-specific FGVR skills by mining the intra-category characteristics and inter-category differences. Finally, LIVE selectively invokes detailed skill guidance according to recognition uncertainty and category confusion, shifting MLLM-based FGVR from a cursory glance to a focus on associating discriminative visual cues with category concepts. Extensive experiments on 13 benchmarks against 24 state-of-the-art methods spanning fine-grained category recognition, action recognition, and general category recognition, verify the effectiveness and generalizability of LIVE across different MLLM backbones. The code will be released upon acceptance.
Chinese Translation
多模态大语言模型（MLLMs）在细粒度视觉识别（FGVR）方面通常未能达到预期，尽管它们在视觉感知和广泛类别知识方面展现出有希望的FGVR基础。大多数现有研究试图通过采用特定领域的后训练来弥补这一不足，但这可能会影响对未见领域的泛化。与这些努力不同，本文从经验和理论的角度重新审视MLLMs背后的FGVR机制，并揭示了一种称为FGVR懒惰的推理时现象。简而言之，FGVR懒惰指的是在某些候选类别列表和识别指导下，视觉线索与类别概念之间的关联不足，导致MLLMs的FGVR基础未能得到充分利用。为了缓解FGVR懒惰，我们提出了一种无训练框架——懒惰抗性视觉识别（LIVE），该框架共同优化候选列表和识别指导，以引发MLLMs潜在的FGVR能力。具体而言，LIVE首先进行查询自适应候选推理，以得出一个紧凑而可靠的候选列表。同时，LIVE通过挖掘类别内特征和类别间差异构建类别特定的FGVR技能。最后，LIVE根据识别不确定性和类别混淆选择性地调用详细的技能指导，将基于MLLM的FGVR从粗略的观察转向聚焦于将区分性视觉线索与类别概念关联起来。在13个基准数据集上与24种最先进的方法进行的广泛实验，验证了LIVE在不同MLLM骨干网络上的有效性和泛化能力。代码将在接受后发布。
cs.CV / 53 / 2610.13174
PreCoG: Preview-Edit for Efficient Video Correction in Generation-Time
PreCoG：生成时高效视频校正的预览编辑
Abstract
Iterative text-to-video generation relies on feedback to correct compositional and temporal failures. However, existing methods typically require another full generation run for each correction, repeating computation and potentially disrupting already correct content. We introduce PreCoG, an iterative framework for correcting semantic failures during video generation. PreQA diagnoses failures from low-cost previews and constructs corrective conditions, while PreEdit edits cached generation trajectories under these conditions, attenuating orthogonal changes to limit disruption of already correct content. PreCoG completes generation only for the selected candidate, reusing its cached trajectory to reduce redundant computation and help preserve semantic consistency between the preview and the final video. Experiments show that PreCoG achieves the highest VBench and T2V-CompBench scores and approaches VSR's leading VS2 score at $0.56\times$ its inference time. It also outperforms GENMAC on all three quality and alignment metrics at $0.86\times$ its inference time.
Chinese Translation
迭代文本到视频生成依赖反馈来纠正组合和时间上的失败。然而，现有方法通常需要为每次修正进行一次完整的生成过程，重复计算并可能干扰已经正确的内容。我们提出了PreCoG，这是一个在视频生成过程中纠正语义失败的迭代框架。PreQA通过低成本预览诊断失败并构建修正条件，而PreEdit则在这些条件下编辑缓存的生成轨迹，减轻正交变化以限制对已正确内容的干扰。PreCoG仅为选定的候选项完成生成，重用其缓存轨迹以减少冗余计算，并帮助保持预览与最终视频之间的语义一致性。实验表明，PreCoG在VBench和T2V-CompBench得分上均达到最高，并在$0.56 imes$推理时间内接近VSR的领先VS2得分。在所有三个质量和对齐指标上，PreCoG也在$0.86 imes$推理时间内超越了GENMAC。
cs.CV / 54 / 2610.13176
Fix the Match, Not the Solver: STITCH -- a Self-Supervised Compatibility Measure for Eroded Visual Puzzles
修复匹配，而非求解器：STITCH——一种自监督的兼容性度量用于侵蚀视觉拼图
Abstract
Unlike the clean and pixel-aligned complete pieces of standard jigsaw benchmarks, real-world puzzle fragments lose their boundary regions to erosion, destroying the signal that classical compatibility measures (CMs) rely on. We first re-establish, via oracle experiments, that the CM, not the solver, is the performance bottleneck: with a perfect CM, even a simple greedy solver achieves perfect reconstruction. This motivates STITCH, a self-supervised CM that sidesteps missing boundaries by comparing frozen ViT patch tokens from piece interiors via a lightweight cross-attention head (~249K trainable parameters), trained without manual annotation. Plugged into unchanged compatibility-based solvers, STITCH reaches .992/.954 Abs/Perf on JPwLEG-5 (+.629 Perf over the best prior method), nearly saturating the benchmark. On a new ImageNet 7x7 benchmark, it outperforms all alternatives, including inpaint-then-match pipelines built on modern generative models. Importantly, performance remains robust across erosion levels and generalizes to puzzles with over 40% missing pieces without retraining. Trained on the recent GAP datasets of irregular, heavily eroded fragments, the same CM recipe enables compatibility-based solvers to surpass recent learning-based solvers on GAP-3 and GAP-5. Since this leap in performance is achieved by a pluggable CM module, the takeaway is clear: fix the match, not the solver.
Chinese Translation
与标准拼图基准中干净且像素对齐的完整拼图块不同，现实世界中的拼图碎片由于侵蚀而失去了边界区域，破坏了经典兼容性度量（CM）所依赖的信号。我们首先通过oracle实验重新确认，CM而非求解器是性能瓶颈：在完美的CM下，即使是简单的贪心求解器也能实现完美重建。这激励了STITCH的提出，一种自监督的CM，通过比较来自拼图内部的冻结ViT补丁令牌，绕过缺失边界的问题，采用轻量级的交叉注意力头（约249K可训练参数），且无需人工标注进行训练。STITCH嵌入不变的基于兼容性的求解器中，在JPwLEG-5上达到0.992/0.954的绝对/性能分数（比之前最佳方法提高0.629的性能），几乎饱和了基准。在新的ImageNet 7x7基准上，它超越了所有替代方案，包括基于现代生成模型构建的修复-匹配管道。重要的是，性能在不同的侵蚀水平下保持稳健，并且在缺失超过40%拼图的情况下无需重新训练即可泛化。基于最近的GAP不规则重度侵蚀碎片数据集训练的相同CM配方，使得基于兼容性的求解器在GAP-3和GAP-5上超越了最近的学习型求解器。由于这一性能飞跃是通过可插拔的CM模块实现的，因此结论明确：修复匹配，而非求解器。
cs.CV / 55 / 2610.13177
CompactSplat: Spatially Adaptive Gaussian Distribution for Feedforward Scene Reconstruction
CompactSplat：用于前馈场景重建的空间自适应高斯分布
Abstract
Reconstructing 3D scenes from sparse views without per-scene optimization remains highly challenging, especially for recovering accurate geometry and fine textures. While recent feedforward approaches leverage generalizable 3D Gaussian Splatting (3DGS) for scene generation, they typically assign one or multiple Gaussians to each pixel. Such uniform allocation not only produces highly redundant representations by wasting primitives in homogeneous areas, but also fails to exploit the inherent geometric priors of planar surfaces, often resulting in sub-optimal structural modeling. To address these limitations, we present CompactSplat, a novel feedforward framework that enables a content-aware allocation of 3D Gaussians. By integrating texture-guided spatial partitioning with hierarchical sampling, CompactSplat adaptively distributes primitives - concentrating dense Gaussians in complex regions while tiling flat areas with expanded, scale-modulated primitives. Moreover, our framework inherently supports dynamic sparsity control at inference time, enabling seamless trade-offs between rendering fidelity and memory footprint without requiring network retraining. Extensive experiments on RealEstate10K, DL3DV, and ScanNet demonstrate that CompactSplat consistently outperforms prior methods on both standard metrics and high-resolution rendering consistency, achieving high-fidelity reconstructions with significantly fewer primitives.
Chinese Translation
从稀疏视图重建三维场景而无需对每个场景进行优化仍然具有很高的挑战性，尤其是在恢复准确的几何形状和细致纹理方面。虽然最近的前馈方法利用可泛化的三维高斯点云（3D Gaussian Splatting, 3DGS）进行场景生成，但它们通常将一个或多个高斯分配给每个像素。这种均匀分配不仅通过在同质区域浪费原始元素而产生高度冗余的表示，而且未能利用平面表面的固有几何先验，常常导致次优的结构建模。为了解决这些局限性，我们提出了CompactSplat，一种新颖的前馈框架，能够实现内容感知的三维高斯分配。通过将纹理引导的空间分区与层次采样相结合，CompactSplat自适应地分配原始元素——在复杂区域集中分布密集的高斯，而在平坦区域则使用扩展的、尺度调制的原始元素。此外，我们的框架在推理时本质上支持动态稀疏控制，实现渲染保真度与内存占用之间的无缝权衡，而无需重新训练网络。在RealEstate10K、DL3DV和ScanNet上的大量实验表明，CompactSplat在标准指标和高分辨率渲染一致性方面始终优于先前的方法，以显著更少的原始元素实现高保真重建。
cs.CV / 56 / 2610.13179
High-Throughput Generative Video Super-Resolution on a Single Consumer GPU
单个消费者GPU上的高通量生成视频超分辨率
Abstract
Generative video super-resolution (VSR) can recover visually compelling high-frequency detail by leveraging diffusion priors, but iterative sampling and heavy spatio-temporal coupling make practical deployments difficult. We study a different operating point: practical HD generative VSR on a single high-end consumer GPU without end-to-end retraining. To this end, we present a zero-shot framework built entirely from pretrained components. The proposed system decouples spatial enhancement from temporal stabilization. A time-distilled one-step diffusion backbone handles spatial upscaling, while temporal coherence is promoted by a lightweight pretrained flow-based interpolator that synthesizes intermediate frames between super-resolved keyframes. To further improve scalability, we introduce a resolution-aware hybrid attention scheme that, in a zero-shot manner, replaces selected high-resolution global-attention layers with neighborhood attention while preserving global attention in semantically critical low-resolution layers, while also improving temporal consistency. Experiments show that the proposed framework reaches up to ~32 FPS at 720p resolution under interleaved inference on a single NVIDIA RTX 4090. At 4K, the proposed attention substitution can yield up to a 44% speedup over the original attention implementation. Across standard VSR benchmarks, the method consistently shifts the speed-quality trade-off decisively toward runtime efficiency, trading reconstruction fidelity for strong no-reference perceptual quality, while remaining reasonably competitive against generative approaches in temporal consistency. It also exposes clear limitations in handling more challenging motion, which a simple motion gate partially mitigates. Code will be released upon acceptance.
Chinese Translation
生成视频超分辨率（VSR）可以通过利用扩散先验恢复视觉上引人注目的高频细节，但迭代采样和重度时空耦合使得实际部署变得困难。我们研究了一个不同的操作点：在单个高端消费者GPU上进行实用的高清生成VSR，而无需端到端的重新训练。为此，我们提出了一个完全由预训练组件构建的零-shot框架。所提出的系统将空间增强与时间稳定解耦。一个经过时间蒸馏的一步扩散主干处理空间上采样，而一个轻量级的预训练基于流的插值器则在超分辨率关键帧之间合成中间帧，以促进时间一致性。为了进一步提高可扩展性，我们引入了一种分辨率感知的混合注意力机制，该机制以零-shot的方式用邻域注意力替换选定的高分辨率全局注意力层，同时在语义关键的低分辨率层中保留全局注意力，从而改善时间一致性。实验表明，所提出的框架在单个NVIDIA RTX 4090上以720p分辨率在交错推理下可达到约32 FPS。在4K分辨率下，所提出的注意力替换可以比原始注意力实现高达44%的加速。在标准VSR基准测试中，该方法始终将速度-质量权衡明显向运行时效率倾斜，以重建保真度换取强大的无参考感知质量，同时在时间一致性方面仍然与生成方法保持合理的竞争力。它还暴露出在处理更具挑战性的运动时的明显局限性，而一个简单的运动门部分缓解了这一问题。代码将在接受后发布。
cs.CV / 57 / 2610.13181
MixTok: Adaptive Visual Tokenization with Global 1D Tokens and Sparse 2D Refinement
MixTok：具有全局一维标记和稀疏二维细化的自适应视觉标记化
Abstract
Efficient visual tokenization requires balancing representation compactness with the preservation of fine-grained image content. Conventional grid-based tokenizers allocate tokens uniformly across spatial locations despite non-uniform visual complexity, whereas compact 1D tokenizers aggregate global information but can struggle to preserve local details. We introduce MixTok, a hybrid visual tokenizer that combines a compact 1D global representation with sparse 2D tokens for local refinement. A content-adaptive router ranks candidate locations by refinement priority and determines the number of 2D tokens for each image under a target average token budget. At the highest-scoring locations, 2D token features replace the spatial features recovered from the 1D representation, and a shared decoder reconstructs the image from the resulting mixed features. On ImageNet 256 x 256, MixTok achieves highly competitive reconstruction and generation performance, with an rFID of 1.195 and a gFID of 1.95 at respective average budgets of approximately 128 and 131 content tokens per image.
Chinese Translation
高效的视觉标记化需要在表示紧凑性与细粒度图像内容的保留之间取得平衡。传统的基于网格的标记器在空间位置上均匀分配标记，尽管视觉复杂性并不均匀，而紧凑的一维标记器则聚合全局信息，但在保留局部细节方面可能存在困难。我们提出了MixTok，一种混合视觉标记器，结合了紧凑的一维全局表示和用于局部细化的稀疏二维标记。内容自适应路由器根据细化优先级对候选位置进行排名，并在目标平均标记预算下确定每幅图像的二维标记数量。在得分最高的位置，二维标记特征替代从一维表示中恢复的空间特征，共享解码器从生成的混合特征中重建图像。在ImageNet 256 x 256上，MixTok实现了高度竞争的重建和生成性能，分别在每幅图像约128和131个内容标记的平均预算下，获得了1.195的rFID和1.95的gFID。
cs.CV / 58 / 2610.13187
Destination-Conditioned Adaptation with Semantic-Geometric Alignment for Generative Inbetweening
基于语义-几何对齐的目标条件适应生成中间帧
Abstract
Generative inbetweening aims to synthesize temporally coherent intermediate frames between a start frame and a target frame. Although pretrained image-to-video diffusion models provide strong priors for large-motion inbetweening, they are not explicitly designed to reach a specified target frame. They may therefore generate realistic individual frames while drifting from the target or producing unstable motion under large displacements and non-rigid dynamics. We propose DSG-GI, a framework that adapts pretrained latent video diffusion models through \textbf{D}estination-Conditioned Adaptation and \textbf{S}emantic-\textbf{G}eometric Alignment. Destination-Conditioned Adaptation treats the target frame as the destination of the sequence and provides temporally progressive guidance along the generated trajectory. Semantic-Geometric Alignment aligns the token-wise spatio-temporal relational geometry of backbone features with that of a frozen video foundation encoder and applies a flow-matching-aware latent flow constraint to samples near the data endpoint of the probability path. Together, these components strengthen destination guidance and preserve coherent semantic-geometric motion structure. Extensive experiments on standard and large-motion inbetweening benchmarks demonstrate improvements in temporal stability, target-frame faithfulness, and perceptual quality. Our code will be made publicly available.
Chinese Translation
生成中间帧旨在合成起始帧与目标帧之间时间上连贯的中间帧。尽管预训练的图像到视频扩散模型为大幅度中间帧生成提供了强大的先验，但它们并未明确设计为达到指定的目标帧。因此，它们可能会生成逼真的单帧图像，但在目标帧上漂移，或者在大位移和非刚性动态下产生不稳定的运动。我们提出了DSG-GI，一个通过目标条件适应（Destination-Conditioned Adaptation）和语义-几何对齐（Semantic-Geometric Alignment）来适应预训练潜在视频扩散模型的框架。目标条件适应将目标帧视为序列的目的地，并在生成的轨迹上提供时间上渐进的指导。语义-几何对齐将主干特征的逐标记时空关系几何与冻结的视频基础编码器的几何进行对齐，并对概率路径数据端点附近的样本应用流匹配感知的潜在流约束。这些组件共同增强了目标指导，并保持了连贯的语义-几何运动结构。在标准和大幅度中间帧生成基准上的广泛实验表明，在时间稳定性、目标帧忠实度和感知质量方面都有所改善。我们的代码将公开发布。
cs.CV / 59 / 2610.13217
Golden Anchor for Video Generation
视频生成的黄金锚点
Abstract
Video generation supports digital content creation, with applications in virtual avatars, AI-driven filmmaking, interactive game engines, and world models. State-of-the-art systems achieve controllability primarily through image-to-video (I2V) generation, in which an injected anchor image determines the identity, composition, and style of the resulting video. In modern pipelines, this anchor is synthesized by a text-to-image (T2I) generator. This raises a largely overlooked question: Are images obtained directly from the T2I model truly suitable for current video models? Our analysis shows that anchor aesthetic and quality scores are weakly correlated with downstream video quality. The structure-and-texture score is negatively correlated with I2V video quality in both the visual and motion dimensions. Fine-tuning the T2I model on an image reward accordingly yields anchors with visibly denser structural and character detail: they score higher on image metrics, yet lower the video reward on every I2V model we test and change downstream motion unpredictably. We therefore introduce Golden Anchor Learning (GAL), to our knowledge the first framework for learning golden anchors by optimizing a T2I generator with downstream video feedback. GAL evaluates candidate anchors through the videos they induce under a frozen I2V model and converts video rewards into image-level preferences for DPO fine-tuning of the T2I generator. On all four I2V models, GAL raises the VBench average by $2.97\%$--$8.46\%$ over the untuned generator. The rendered clips also score higher under VideoReward-Overall, and their extracted frames under HPSv3 and UniPercept. GAL yields consistent motion gains across all four I2V models, as measured by a $5.62\%$--$38.24\%$ increase in VBench dynamic degree. Although GAL is trained only on HunyuanVideo-1.5 renders, these gains extend to the other three I2V models.
Chinese Translation
视频生成支持数字内容创作，应用于虚拟化身、人工智能驱动的电影制作、互动游戏引擎和世界模型。最先进的系统主要通过图像到视频（I2V）生成实现可控性，其中注入的锚点图像决定了生成视频的身份、构图和风格。在现代流程中，这个锚点是由文本到图像（T2I）生成器合成的。这引发了一个被广泛忽视的问题：直接从T2I模型获得的图像是否真的适合当前的视频模型？我们的分析表明，锚点的美学和质量评分与下游视频质量的相关性较弱。结构和纹理评分在视觉和运动维度上与I2V视频质量呈负相关。相应地，对T2I模型进行图像奖励的微调会产生在结构和角色细节上明显更密集的锚点：它们在图像指标上得分更高，但在我们测试的每个I2V模型上降低了视频奖励，并且不可预测地改变了下游运动。因此，我们引入了黄金锚点学习（Golden Anchor Learning, GAL），据我们所知，这是第一个通过优化T2I生成器并结合下游视频反馈来学习黄金锚点的框架。GAL通过在冻结的I2V模型下评估候选锚点所引发的视频，并将视频奖励转化为图像级偏好，以进行TPO微调T2I生成器。在所有四个I2V模型中，GAL使VBench平均值提高了$2.97\%$--$8.46\\%$，相较于未调优的生成器。渲染的剪辑在VideoReward-Overall下得分也更高，其提取的帧在HPSv3和UniPercept下得分更高。GAL在所有四个I2V模型中实现了一致的运动增益，VBench动态度提高了$5.62\\%$--$38.24\\%$。尽管GAL仅在HunyuanVideo-1.5渲染上进行训练，但这些增益扩展到其他三个I2V模型。
cs.CV / 60 / 2610.13219
Complexity-Calibrated Reconstruction Discrepancy for No-Reference Point Cloud Quality Assessment
复杂度校准重建差异用于无参考点云质量评估
Abstract
The discrepancy induced by surface reconstruction in a distorted point cloud is jointly influenced by its intrinsic structural complexity and signal degradation. This observation motivates a referencefree point cloud quality assessment (PCQA) strategy that estimates and compensates for the complexity-induced component. We implement this strategy through successive reconstruction. Specifically, reconstructing the distorted point cloud produces an input-toreconstruction discrepancy that combines the effects of signal degradation and content complexity. Reapplying the reconstruction operator with stronger smoothing produces a discrepancy between successive reconstructions, which serves as a proxy for the complexityinduced component. This proxy is then used to calibrate the inputto-reconstruction discrepancy. We apply the same strategy to 3D geometry patches and their aligned 2D regions, enabling local comparison of the resulting discrepancies. The calibrated geometric and projected-view discrepancies are then fused into a final degradation score. For both branches, the discrepancies are measured using frozen pretrained perceptual representations. Our method does not require PCQA-specific training or fine-tuning. Experiments on three public databases show that the proposed method consistently outperforms most generic baselines and achieves performance comparable to several full-reference metrics.The code is available at https://github.com/HC-W-whc/CCRD-NR-PCQA.git.
Chinese Translation
在扭曲点云中，由表面重建引起的差异受到其内在结构复杂性和信号退化的共同影响。这一观察促使我们提出了一种无参考点云质量评估（PCQA）策略，该策略估计并补偿由复杂性引起的成分。我们通过连续重建实现这一策略。具体而言，重建扭曲的点云会产生输入到重建的差异，该差异结合了信号退化和内容复杂性的影响。再次应用更强平滑的重建算子会产生连续重建之间的差异，这可以作为复杂性引起的成分的代理。然后使用该代理来校准输入到重建的差异。我们将相同的策略应用于3D几何补丁及其对齐的2D区域，从而实现结果差异的局部比较。校准后的几何和投影视图差异随后融合为最终的退化评分。在两个分支中，差异使用冻结的预训练感知表示进行测量。我们的方法不需要针对PCQA的特定训练或微调。在三个公共数据库上的实验表明，所提出的方法始终优于大多数通用基线，并且其性能与若干全参考指标相当。代码可在 https://github.com/HC-W-whc/CCRD-NR-PCQA.git 获取。
cs.CV / 61 / 2610.13220
Multi-Modal Graph Neural Network-Inspired Vessel Identification and Tracking for UAV-Based Maritime Surveillance
基于多模态图神经网络的无人机海洋监视中的船舶识别与跟踪
Abstract
The unmanned aerial vehicle (UAV)-based maritime surveillance is becoming an important part of maritime Internet of Things (IoT) and intelligent transportation systems. It provides a flexible and cost-effective way to large-scale maritime traffic monitoring. However, achieving robust situational awareness remains challenging in practical maritime environments. The main challenges arise from heterogeneous sensing modalities, adverse imaging conditions, and spatial misalignment caused by UAV motion. To address these problems, this paper proposes an airborne multi-modal graph learning-driven framework, termed AMGT, for vessel identification and tracking in UAV-assisted maritime IoT. Specifically, visible and infrared images are first aligned and fused through a multi-parameter perspective transformation model. This improves visual perception under low visibility and complex illumination conditions. Automatic identification system (AIS) messages are then temporally synchronized using interpolation and linear prediction. To detect small and oriented vessels, we develop a mixed-modality MAir-YOLOv8-OBB detector, followed by the MAir-SORT algorithm for stable vessel tracking. We further design a graph learning-driven association network to model temporal dynamics and spatial dependencies among vessels and between vessels and UAVs. A motion-aware coordinate unification (MCU) strategy is also introduced to reduce nonlinear misalignment caused by UAV motion. Experiments on the MAP-M2UAV50 dataset show that AMGT achieves high accuracy and efficiency. The results demonstrate its potential for UAV-assisted maritime IoT applications, including vessel monitoring, traffic situational awareness, and intelligent maritime transportation services. The dataset and code will be available at https://github.com/MAPGroup/AMGT.
Chinese Translation
基于无人机（UAV）的海洋监视正成为海洋物联网（IoT）和智能交通系统的重要组成部分。它提供了一种灵活且具有成本效益的大规模海洋交通监测方式。然而，在实际的海洋环境中，实现稳健的态势感知仍然面临挑战。主要挑战来自于异构传感模态、不利的成像条件以及无人机运动引起的空间错位。为了解决这些问题，本文提出了一种基于空中多模态图学习的框架，称为AMGT，用于无人机辅助的海洋物联网中的船舶识别与跟踪。具体而言，首先通过多参数透视变换模型对可见光和红外图像进行对齐和融合。这在低能见度和复杂照明条件下改善了视觉感知。然后，利用插值和线性预测对自动识别系统（AIS）消息进行时间同步。为了检测小型和定向船舶，我们开发了一种混合模态的MAir-YOLOv8-OBB检测器，随后使用MAir-SORT算法进行稳定的船舶跟踪。我们进一步设计了一个基于图学习的关联网络，以建模船舶之间及船舶与无人机之间的时间动态和空间依赖关系。同时，引入了一种运动感知坐标统一（MCU）策略，以减少无人机运动引起的非线性错位。在MAP-M2UAV50数据集上的实验表明，AMGT实现了高准确性和高效率。结果展示了其在无人机辅助海洋物联网应用中的潜力，包括船舶监测、交通态势感知和智能海洋运输服务。数据集和代码将可在 https://github.com/MAPGroup/AMGT 获取。
cs.CV / 62 / 2610.13224
Automated Detection and Encoding of Cartographic Signs in Historical Maps
历史地图中制图符号的自动检测与编码
Abstract
Symbols and icons in archival maps represent an expansive source of evidence for historical geography. In this paper, we present a two-stage framework for (1) detecting cartographic signs in historical maps and (2) generating interpretable sign embeddings that capture both semantics and stylistic variations. We also introduce CaSiDD, an open benchmark dataset spanning multiple sign categories and diverse cartographic styles, designed for generic sign detection. Using YOLOv10-X, we demonstrate the practicality of automated sign detection on historical maps. We then present strategies for creating contrasted, informative embeddings of cartographic signs. We approach embedding design systematically, maximizing expressiveness while suppressing digitization or background material confounders. Our final method, based on an adapted RADIOv4 backbone, effectively encodes both sign form and semantics, while producing interpretable representations. This work lays a foundation for scalable evidence extraction for the comparative study of map signs within cartographic culture and geohistorical research.
Chinese Translation
档案地图中的符号和图标代表了历史地理学的丰富证据来源。本文提出了一个两阶段框架，(1) 检测历史地图中的制图符号，(2) 生成可解释的符号嵌入，以捕捉语义和风格的变化。我们还介绍了CaSiDD，这是一个开放的基准数据集，涵盖多个符号类别和多样的制图风格，旨在用于通用符号检测。使用YOLOv10-X，我们展示了在历史地图上自动检测符号的实用性。接着，我们提出了创建对比鲜明、信息丰富的制图符号嵌入的策略。我们系统地设计嵌入，最大限度地提高表达能力，同时抑制数字化或背景材料的干扰。我们最终的方法基于改进的RADIOv4骨干网络，有效地编码符号的形式和语义，同时生成可解释的表示。这项工作为在制图文化和地理历史研究中进行地图符号的比较研究奠定了可扩展的证据提取基础。
cs.CV / 63 / 2610.13232
FPD-Drive: Future Policy Distillation for End-to-End Autonomous Driving
FPD-Drive：端到端自主驾驶的未来策略蒸馏
Abstract
End-to-end autonomous driving models are typically trained on real-world driving logs, which record only the trajectory actually executed in each scene. This provides limited supervision for learning alternative valid behaviors. Multi-target supervision can mitigate this limitation, but broader trajectory coverage makes safe candidate selection more challenging. We propose FPD-Drive, an end-to-end planner that distills privileged future knowledge into a visual driving policy to improve both multi-trajectory supervision and safety selection. Specifically, we introduce Future-Privileged Multi-Target Distillation to transfer diverse, high-quality trajectory targets from a teacher with access to recorded future observations to the student. These targets extend supervision beyond the single recorded human trajectory. For candidate selection, we further introduce Safety-Aware Score Correction to correct ranking errors between nearby safe and unsafe candidates and raise underestimated safety scores of safe candidates. At deployment, the student requires neither future observations nor teacher inference. Our base model achieves 94.9 PDMS on NAVSIM v1 navtest and 55.9 EPDMS on NAVSIM v2 navhard. The scaled FPD-Drive-Pro improves these scores to 95.5 PDMS and 58.1 EPDMS, respectively, achieving SOTA performance among the compared methods on both benchmarks. Zero-shot closed-loop evaluation on HUGSIM further yields an HD-Score of 45.1 and RC of 55.3.
Chinese Translation
端到端自主驾驶模型通常是在真实世界驾驶日志上训练的，这些日志仅记录了每个场景中实际执行的轨迹。这为学习替代有效行为提供了有限的监督。多目标监督可以缓解这一限制，但更广泛的轨迹覆盖使得安全候选选择变得更加困难。我们提出了FPD-Drive，这是一种端到端规划器，将特权未来知识蒸馏为视觉驾驶策略，以改善多轨迹监督和安全选择。具体而言，我们引入了未来特权多目标蒸馏（Future-Privileged Multi-Target Distillation），以将来自能够访问记录未来观察的教师的多样化、高质量轨迹目标转移到学生。这些目标将监督扩展到单一记录的人类轨迹之外。对于候选选择，我们进一步引入了安全意识评分修正（Safety-Aware Score Correction），以纠正安全和不安全候选之间的排名错误，并提高安全候选的低估安全评分。在部署时，学生既不需要未来观察也不需要教师推理。我们的基础模型在NAVSIM v1 navtest上达到了94.9的PDMS，在NAVSIM v2 navhard上达到了55.9的EPDMS。经过扩展的FPD-Drive-Pro将这些分数提高到95.5 PDMS和58.1 EPDMS，分别在两个基准测试中实现了与比较方法相比的SOTA性能。在HUGSIM上的零-shot闭环评估进一步得到了45.1的HD-Score和55.3的RC。
cs.CV / 64 / 2610.13239
Jag: Direct Box Prediction for Efficient Visual Grounding
Jag：高效视觉定位的直接框预测
Abstract
Visual grounding combines image--language understanding with geometric prediction, yet autoregressive multimodal models often express a box as a coordinate sequence. Sequential decoding adds computation to the multimodal processing needed for localization. Adapting these representations to box geometry offers a route to direct prediction. We present Jag, which adapts a compact pretrained multimodal model for direct continuous box prediction. A lightweight regression head reads the last valid input-token state, conditioned on the image and referring expression, and predicts all four box coordinates together. Geometric supervision jointly adapts the language backbone, visual merger, and regression head, allowing localization losses to shape both the multimodal representations and the coordinate mapping. Inference retains image--language processing and returns the box in one forward pass, without coordinate generation or text parsing. Across benchmarks, Jag consistently improves grounding accuracy over its base model across the observed target sizes and expression groups. Component ablation show that jointly adapting the representations and head substantially improves accuracy over training the head alone. Against existing specialized grounding models, it achieves competitive accuracy with lower single-request latency and GPU memory use. These results show how supervision tailored to the geometric answer can specialize a compact pretrained multimodal model for accurate grounding and efficient direct prediction. Project page: https://xyzzzh.github.io/Jag/
Chinese Translation
视觉定位结合了图像-语言理解与几何预测，但自回归多模态模型通常将框表示为坐标序列。顺序解码为定位所需的多模态处理增加了计算负担。将这些表示适应于框几何提供了一条直接预测的途径。我们提出了Jag，它将一个紧凑的预训练多模态模型调整为直接连续框预测。一个轻量级回归头读取最后一个有效输入标记状态，基于图像和指代表达进行条件预测，并同时预测四个框坐标。几何监督共同调整语言主干、视觉合并和回归头，使得定位损失能够塑造多模态表示和坐标映射。在推理过程中，保留图像-语言处理，并在一次前向传播中返回框，无需坐标生成或文本解析。在各基准测试中，Jag在观察到的目标大小和表达组上，始终提高了相较于其基础模型的定位准确性。组件消融实验表明，共同调整表示和回归头显著提高了准确性，相较于单独训练回归头。与现有的专用定位模型相比，它在单次请求延迟和GPU内存使用上实现了竞争力的准确性。这些结果表明，针对几何答案量身定制的监督如何使紧凑的预训练多模态模型专门化，以实现准确的定位和高效的直接预测。项目页面：https://xyzzzh.github.io/Jag/
cs.CV / 65 / 2610.13242
Deep Learning Approach for RealTime Contraction Detection in Laboring Dairy Cattle
基于深度学习的奶牛分娩实时收缩检测方法
Abstract
Currently, the animal production sector is undergoing rapid development of techniques that allow automatic monitoring of animal behaviour and health. This paper presents a computer vision-based approach for the automatic recognition of labor contractions in cows. During the course of the study, videos depicting 100 cow births were collected. From the obtained videos, 8500 cow contraction actions and about 25 000 non-contraction sequences were extracted which constituted the dataset. Leveraging image processing techniques and deep learning algorithms like Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), our methodology extracts relevant features from video frames to detect and classify contractions. Our study demonstrates a high accuracy rate of over 95% in detecting labor contractions. This automated system has the potential to enhance productivity and animal welfare in the dairy industry by providing real-time monitoring and facilitating timely intervention during difficult calvings. The video demonstration of the system can be found at the following URL: https://www.youtube.com/watch?v=8fTSG5D7dZ0
Chinese Translation
目前，动物生产领域正在迅速发展能够自动监测动物行为和健康的技术。本文提出了一种基于计算机视觉的自动识别奶牛分娩收缩的方法。在研究过程中，收集了100个奶牛分娩的视频。从获得的视频中提取了8500个奶牛收缩动作和约25000个非收缩序列，构成了数据集。利用图像处理技术和深度学习算法，如卷积神经网络（CNNs）和视觉变换器（ViTs），我们的方法从视频帧中提取相关特征以检测和分类收缩。我们的研究显示在检测分娩收缩方面具有超过95%的高准确率。该自动化系统有潜力通过提供实时监测和促进在困难分娩时的及时干预，来提高奶牛行业的生产力和动物福利。系统的视频演示可以在以下网址找到：https://www.youtube.com/watch?v=8fTSG5D7dZ0
cs.CV / 66 / 2610.13252
Vision Transformers Learn Word Meaning from Images Alone
视觉变换器仅通过图像学习词义
Abstract
Can a vision model learn what words mean from pixels alone? We show that the two largest DINOv3 vision transformers organize words rendered in images by word meaning, despite being pretrained using self-supervision on images only, without paired captions, class labels, or other language supervision. We derive word representations from embedding differences between word images and matched controls, then test whether semantically related words are neighbors within each model's embedding space, while controlling for spelling similarity. We find that brand names in the same industry group together, as do nouns from the same semantic domain. Country-capital pairs exhibit consistent vector offsets, similarly to word embeddings trained on text. These word representations also align with the representations of the visual object categories they name. DINOv3-H and DINOv3-7B are, to our knowledge, the first self-supervised vision models to show substantial semantic structure for rendered words. Comparisons with text and speech encoders reveal shared structure among representations of the same words. Contrastive alignment with either encoder further strengthened DINOv3-H's word-level semantic structure, while leaving the other finetuned vision models largely unchanged. Image-only training can therefore teach vision models not just what words look like, but also how their meanings relate.
Chinese Translation
视觉模型能否仅通过像素学习词义？我们展示了两个最大的 DINOv3 视觉变换器如何根据词义组织图像中呈现的词，尽管它们仅在图像上进行了自监督预训练，没有配对的标题、类别标签或其他语言监督。我们通过词图像与匹配对照之间的嵌入差异推导出词表示，然后测试语义相关的词是否在每个模型的嵌入空间中相邻，同时控制拼写相似性。我们发现，来自同一行业的品牌名称会聚在一起，来自同一语义领域的名词也是如此。国家与首都的配对表现出一致的向量偏移，类似于在文本上训练的词嵌入。这些词表示与它们所命名的视觉对象类别的表示也相一致。根据我们的知识，DINOv3-H 和 DINOv3-7B 是首批展示渲染词显著语义结构的自监督视觉模型。与文本和语音编码器的比较揭示了相同词的表示之间的共享结构。与任一编码器的对比对齐进一步增强了 DINOv3-H 的词级语义结构，而对其他微调的视觉模型影响不大。因此，仅通过图像训练可以教会视觉模型不仅了解词的外观，还了解其意义之间的关系。
cs.CV / 67 / 2610.13259
Beyond a Single Crop: Contextualized Embeddings for Player Re-Identification
超越单一裁剪：用于玩家重识别的上下文化嵌入
Abstract
Person re-identification is typically addressed by learning identity-discriminative embeddings from individual person crops. However, player re-identification in broadcast sports videos presents a unique challenge: players from the same team often share similar appearances, while video sequences provide repeated observations of the same players under different viewpoints and surrounding contexts. Existing crop-based approaches largely ignore this contextual information, which can be critical for resolving visually similar identities. Here, we propose ContextReID, a contextualized embedding framework that represents a player by jointly considering other players in the same frame and observations from different frames of the broadcast sequence. By incorporating both co-occurring players and cross-frame observations, the model learns identity features relative to the surrounding visual context rather than relying only on isolated appearance cues. We implement this framework with a vision-language model that takes sets of player crops as input and generates context-aware embeddings for individual players. A combined contextual learning and retrieval objective encourages the model to capture cross-frame identity relationships while maintaining discriminative embeddings for player matching. Experiments across three multi-object tracking benchmarks show that ContextReID improves player retrieval over conventional single-crop embedding approaches in both Rank-1 and mAP. These results highlight the utility of spatial and temporal context for reliable player identification in broadcast sports videos.
Chinese Translation
人员重识别通常通过从单个个体裁剪中学习身份区分嵌入来解决。然而，在广播体育视频中，玩家重识别面临着独特的挑战：同一支球队的玩家往往具有相似的外观，而视频序列提供了在不同视角和周围环境下对同一玩家的重复观察。现有的基于裁剪的方法在很大程度上忽视了这些上下文信息，而这些信息对于解决视觉上相似的身份至关重要。在此，我们提出了ContextReID，一个上下文化嵌入框架，通过联合考虑同一帧中的其他玩家和广播序列中不同帧的观察来表示玩家。通过结合共现玩家和跨帧观察，该模型学习相对于周围视觉上下文的身份特征，而不仅仅依赖于孤立的外观线索。我们使用一个视觉-语言模型实现该框架，该模型将一组玩家裁剪作为输入，并为个别玩家生成上下文感知嵌入。结合的上下文学习和检索目标鼓励模型捕捉跨帧身份关系，同时保持用于玩家匹配的区分性嵌入。在三个多目标跟踪基准上的实验表明，ContextReID在Rank-1和mAP方面均优于传统的单裁剪嵌入方法。这些结果突显了空间和时间上下文在广播体育视频中可靠玩家识别中的重要性。
cs.CV / 68 / 2610.13263
SheafPrune: Sheaf-Guided Graph Propagation for Visual Token Pruning
SheafPrune：基于Sheaf引导的图传播用于视觉令牌修剪
Abstract
Multimodal large language models process the complete visual token sequence, increasing prefill cost even when many tokens are irrelevant to the instruction. Token-wise saliency captures limited inter-token dependence; graph-based selection instead depends on how relations govern importance propagation and ranking. We study support-constrained propagation with node-degree correction. Inspired by sheaf geometry, SheafPrune uses modality-specific maps and a sparse text--visual graph to transfer relevance, then propagates scores on a separate visual graph and corrects them for degree. It retains a hard top-$k$ subset once before the frozen language model. The selector has about $4.5$M parameters and is trained end-to-end through the frozen LLaVA-NeXT-8B backbone using only the language-modeling objective. Explicit retention-ratio conditioning lets one checkpoint serve every budget. Across four backbones, ten benchmarks and seven pruning ratios, with all baselines re-implemented under one protocol, SheafPrune achieves the best retention at $22$ of $28$ backbone--budget settings, with monotonically growing margins in the moderate-to-aggressive regime. On LLaVA-NeXT, retaining $10\%$ of visual tokens yields a $5.15\times$ net prefill speedup and an $84.0\%$ end-to-end FLOPs reduction. Together with component ablations, these results support graph-constrained importance propagation and degree correction as a design for visual token selection across the evaluated backbones and tokenization schemes, improving the trade-off between accuracy retention and inference cost. code:https://github.com/Bob0817912/SheafPrune_ICLR_2027
Chinese Translation
多模态大型语言模型处理完整的视觉令牌序列，即使许多令牌与指令无关，也会增加预填充成本。基于令牌的显著性捕捉了有限的令牌间依赖关系；而基于图的选择则依赖于关系如何支配重要性传播和排序。我们研究了支持约束传播与节点度校正。受到Sheaf几何的启发，SheafPrune使用特定模态的映射和稀疏的文本-视觉图来传递相关性，然后在单独的视觉图上传播分数，并对其进行度校正。在冻结的语言模型之前，它保留了一次硬性前$k$子集。选择器大约有$4.5$M参数，并通过冻结的LLaVA-NeXT-8B主干进行端到端训练，仅使用语言建模目标。显式的保留比率条件使得一个检查点可以满足每个预算。在四个主干、十个基准和七个修剪比率下，所有基线在一个协议下重新实现，SheafPrune在$28$个主干-预算设置中实现了$22$个的最佳保留，并在中等到激进的范围内以单调增长的边际表现。在LLaVA-NeXT上，保留$10\%$的视觉令牌带来了$5.15 imes$的净预填充加速和$84.0\%$的端到端FLOPs减少。结合组件消融实验，这些结果支持基于图的约束重要性传播和度校正作为视觉令牌选择的设计，改善了准确性保留与推理成本之间的权衡。代码：https://github.com/Bob0817912/SheafPrune_ICLR_2027
cs.CV / 69 / 2610.13264
CV-OPD: Contrastive Vision On-policy Distillation
CV-OPD：对比视觉在线蒸馏
Abstract
On-policy distillation (OPD) reveals versatility and convenience during VLM post-training, providing dense supervision for student from teacher. However, we observed an anomaly that the distilled VLM could still answer the questions accurately even when the related visual regions were masked or replaced with noise. Building on this observation, our quantitative and qualitative experiments demonstrate that OPD, which uses the answer as a hint, has failed to teach students to genuinely seek out and rely on visual evidence. Our theoretical analysis suggests that this failure arises from a shortcut enabled by the lack of visual constraints during distillation. To address the issue, we introduce Contrastive Vision On-policy Distillation (CV-OPD), which compares teacher distributions induced by positive and negative crops while conditioning on the same student prefix and augments optimization with a margin-based contrastive constraint, which is modulated with discrepancy gating across token and sample level. Specifically, CV-OPD first introduces a negative teacher unrelated to the answer and designs a Contrastive Divergence Loss (CDL) that constrains the feature distribution of student by contrasting the positive and negative teachers, correcting how the student exploits visual regions. The CV-OPD introduces a Discrepancy Aware Gate (DAG) that uses token-level and sample-level discrepancies to select visually relevant tokens for optimization, directing the student's attention toward visual content. Extensive experiments across a broad range of fine-grained visual benchmarks demonstrate that CV-OPD substantially alleviates shortcut and consistently improves the visual reasoning capabilities of existing VLMs.
Chinese Translation
在线蒸馏（On-policy distillation, OPD）在视觉语言模型（Visual Language Model, VLM）后训练中展现了多样性和便利性，为学生提供了来自教师的密集监督。然而，我们观察到一个异常现象，即使在相关视觉区域被遮蔽或替换为噪声的情况下，蒸馏后的 VLM 仍然能够准确回答问题。基于这一观察，我们的定量和定性实验表明，使用答案作为提示的 OPD 未能教会学生真正寻求和依赖视觉证据。我们的理论分析表明，这一失败源于蒸馏过程中缺乏视觉约束所导致的捷径。为了解决这一问题，我们引入了对比视觉在线蒸馏（Contrastive Vision On-policy Distillation, CV-OPD），该方法在条件相同的学生前缀下比较由正负样本诱导的教师分布，并通过基于边际的对比约束增强优化，该约束通过标记和样本级别的差异门控进行调节。具体而言，CV-OPD 首先引入一个与答案无关的负教师，并设计了一种对比散度损失（Contrastive Divergence Loss, CDL），通过对比正负教师来约束学生的特征分布，从而纠正学生对视觉区域的利用方式。CV-OPD 引入了一种差异感知门（Discrepancy Aware Gate, DAG），利用标记级和样本级的差异选择与视觉相关的标记进行优化，引导学生关注视觉内容。在广泛的细粒度视觉基准测试中进行的广泛实验表明，CV-OPD 显著缓解了捷径问题，并持续提升了现有 VLM 的视觉推理能力。
cs.CV / 70 / 2610.13268
ConClick: Element-Wise Contrastive Learning Improves GUI Grounding
ConClick：逐元素对比学习提升GUI定位
Abstract
GUI grounding, the task of locating the interface element specified by a user instruction, is a fundamental capability of GUI agents. However, existing grounding models remain error-prone, often causing downstream task failures. We systematically evaluate 10 grounding models across five datasets and identify a major error pattern: most incorrect clicks hit non-target interface elements rather than the background, with different models repeatedly selecting the same distractors. Motivated by these findings, we propose ConClick, a contrastive learning framework that teaches models to distinguish the target element from plausible distractors. We construct ConClick-Data, a dataset containing 4,292 challenging target-distractor pairs, and train the model using complementary listwise ranking and pairwise margin objectives. ConClick improves the grounding accuracy of Qwen3-VL-8B from 54.71% to 66.03% on ScreenSpot-Pro, from 83.72% to 86.28% on MMBench-GUI L2, and from 63.33% to 70.20% on OSWorld-G, outperforming prior methods built on the same or similarly sized backbones. We further introduce ReClick, an inference-time strategy that revisits promising regions and compares predictions across competing candidates, yielding an additional 2.96 percentage-point improvement in average accuracy across the three benchmarks. Together, our analysis, dataset, and methods provide both empirical insights and practical tools for building more reliable GUI agents. Data and code are available at https://github.com/zhangruotian/ConClick.
Chinese Translation
GUI定位是根据用户指令定位界面元素的任务，是GUI代理的一项基本能力。然而，现有的定位模型仍然容易出错，常常导致下游任务失败。我们系统地评估了10个定位模型在五个数据集上的表现，并识别出一个主要的错误模式：大多数错误点击击中的是非目标界面元素，而不是背景，不同模型反复选择相同的干扰项。基于这些发现，我们提出了ConClick，一个对比学习框架，旨在教会模型区分目标元素和合理的干扰项。我们构建了ConClick-Data，一个包含4,292对具有挑战性的目标-干扰项对的数据集，并使用互补的列表排名和成对边际目标对模型进行训练。ConClick使Qwen3-VL-8B在ScreenSpot-Pro上的定位准确率从54.71%提升至66.03%，在MMBench-GUI L2上的准确率从83.72%提升至86.28%，在OSWorld-G上的准确率从63.33%提升至70.20%，超越了基于相同或相似规模骨干网络的先前方法。我们进一步引入了ReClick，一种推理时策略，重新审视有前景的区域并比较竞争候选的预测，在三个基准测试中平均准确率额外提高了2.96个百分点。总之，我们的分析、数据集和方法为构建更可靠的GUI代理提供了经验见解和实用工具。数据和代码可在https://github.com/zhangruotian/ConClick获取。
cs.CV / 71 / 2610.13272
OWLET: On-Policy Weighted Learning via Evidence Transfer for Video Large Language Model Post-Training
OWLET：通过证据转移进行视频大型语言模型后训练的在线加权学习
Abstract
Post-training has become a key stage for eliciting reasoning capabilities in Video Large Language Models (Video-LLMs). Existing post-training methods mainly follow two paradigms: reinforcement learning (RL) and on-policy distillation (OPD). However, they do not fully exploit visual evidence for effective training. RL supervises the policy with a single trajectory-level reward, which is too coarse-grained for video reasoning: the same response-level advantage is applied to every generated token, disregarding that the tokens of a video response are grounded in the visual stream to very different degrees. OPD provides token-level supervision through output-distribution matching, but lacks direct supervision of the spatiotemporal representations that produce those outputs. We observe across extensive experiments that the same model reasons more accurately when its video view combines global temporal context with question-relevant details. Comparing the same student-generated response under this view and uniform sampling reveals changes in token likelihoods and internal representations, providing a training signal for evidence use. Building on this observation, we propose OWLET, an evidence-privileged RL method that derives training signals from a richer teacher view. This view couples uniform global context with question-conditioned local windows, providing visual evidence without requiring a larger model. To better exploit this privileged visual evidence, OWLET introduces (i) \textbf{evidence-aware credit redistribution}, which compares token likelihoods across the two views and redistributes positive response-level advantages according to their relative evidence support; and (ii) \textbf{spatiotemporal state alignment}, which aligns the two views' hidden states on verifier-accepted responses.
Chinese Translation
后训练已成为激发视频大型语言模型（Video-LLMs）推理能力的关键阶段。现有的后训练方法主要遵循两种范式：强化学习（RL）和在线策略蒸馏（OPD）。然而，它们并未充分利用视觉证据进行有效训练。RL通过单一轨迹级奖励来监督策略，这对于视频推理来说过于粗糙：相同的响应级优势应用于每个生成的标记，忽视了视频响应的标记在视觉流中被赋予的不同程度的基础。OPD通过输出分布匹配提供标记级监督，但缺乏对生成这些输出的时空表示的直接监督。我们在大量实验中观察到，当模型的视频视图结合全球时间上下文与问题相关细节时，其推理更为准确。在这种视图和均匀采样下比较相同的学生生成响应揭示了标记可能性和内部表示的变化，为证据使用提供了训练信号。基于这一观察，我们提出了OWLET，一种特权证据的RL方法，从更丰富的教师视图中推导训练信号。该视图将均匀的全球上下文与问题条件的局部窗口结合，提供视觉证据而无需更大的模型。为了更好地利用这一特权视觉证据，OWLET引入了（i） extbf{证据感知的信用重分配}，比较两种视图的标记可能性，并根据其相对证据支持重新分配正响应级优势；以及（ii） extbf{时空状态对齐}，对齐两种视图在验证者接受的响应上的隐藏状态。
cs.CV / 72 / 2610.13281
QCLR: A Plug-in Robustification Module via Quantization for Contrastive Learning in Self-Supervised Skeleton-Based Human Action Recognition
QCLR：一种通过量化实现的对比学习鲁棒性插件模块，用于自监督骨架基础的人类动作识别
Abstract
Skeleton-based human action recognition has gained increasing attention due to the stability of skeleton data across diverse environments and advances in motion capture technologies. Following the success of supervised learning approaches, self-supervised learning approaches, particularly contrastive learning ones, have achieved impressive performance. However, they remain vulnerable to adversarial attacks. In this work, we propose a plug-in robustification module via quantization for contrastive learning in self-supervised skeleton-based human action recognition (QCLR), which enhances the robustness of representations against adversarial attacks while preserving their representational capability. Specifically, we employ a dual-level quantization scheme across the encoding pipeline, which breaks the gradient propagation and absorbs small adversarial perturbations within the quantization bins. To mitigate the impact of inevitable quantization error, we further design an attention-guided multi-faceted quantization strategy to maintain the representation quality. Experiments on several benchmark datasets demonstrate that integrating QCLR into a progressive lineage of mainstream contrastive learning models substantially improves robustness against both gradient-based white-box and gradient-free black-box adversarial attacks, while preserving highly competitive classification accuracy.
Chinese Translation
基于骨架的人类动作识别因其在多样环境中的数据稳定性以及运动捕捉技术的进步而受到越来越多的关注。在监督学习方法取得成功之后，自监督学习方法，尤其是对比学习方法，表现出令人印象深刻的性能。然而，它们仍然容易受到对抗攻击的影响。在本研究中，我们提出了一种通过量化实现的对比学习鲁棒性插件模块（QCLR），用于自监督骨架基础的人类动作识别，该模块在增强表示对抗攻击的鲁棒性的同时，保持其表示能力。具体而言，我们在编码流程中采用了双层量化方案，这一方案打破了梯度传播并吸收了量化箱内的小幅对抗扰动。为了减轻不可避免的量化误差的影响，我们进一步设计了一种基于注意力的多面向量化策略，以保持表示质量。在多个基准数据集上的实验表明，将QCLR集成到主流对比学习模型的渐进序列中，显著提高了对基于梯度的白盒和无梯度的黑盒对抗攻击的鲁棒性，同时保持了高度竞争的分类准确性。
cs.CV / 73 / 2610.13299
Decoding Register Tokens Preserves Global Scene Details
解码注册标记保留全球场景细节
Abstract
Pretrained vision transformers learn structured image representations spanning spatial patch tokens and non-spatial tokens. Recent analyses suggest that non-spatial register tokens encode global scene properties such as illumination, blur, and style. Yet patch-based image decoders discard these outputs, leaving the decoder to recover image appearance from patches alone. We investigate whether retaining register features improves decoding when the complete spatial patch representation is already available. We study this question using representation autoencoders (RAEs) with frozen register-equipped encoders. Our decoder processes four register tokens together with all patch tokens through joint self-attention. Appearance probes, paired reconstruction-error analysis, and fixed-decoder register substitutions examine which global image properties are accessible from registers and how their values affect reconstructed pixels. For generation, we train a multi-head attention pooling (MHAP) predictor to recover register features from sampled patches, keeping the pretrained diffusion model and its samples unchanged. On ImageNet-1K, our ViT-B decoders achieve mean reconstruction FIDs of 0.472 with DINOv2 and 0.442 with DINOv3, reductions of 9.5% and 5.2% relative to their respective patch-only baselines. The corresponding PSNRs are 19.73 and 20.54 dB. For generation, our DINOv2 models achieve a mean FID of 1.641, 9.8% below the patch-only baseline on identical sampled patches.
Chinese Translation
预训练的视觉变换器学习跨越空间补丁标记和非空间标记的结构化图像表示。最近的分析表明，非空间注册标记编码了诸如光照、模糊和风格等全球场景属性。然而，基于补丁的图像解码器会丢弃这些输出，使得解码器只能依赖补丁来恢复图像外观。我们研究在完整的空间补丁表示已经可用的情况下，保留注册特征是否能改善解码。我们使用带有冻结注册编码器的表示自编码器（RAEs）来研究这个问题。我们的解码器通过联合自注意力处理四个注册标记和所有补丁标记。外观探测、配对重建误差分析和固定解码器注册替代分析考察了哪些全球图像属性可以从注册中访问，以及它们的值如何影响重建像素。在生成方面，我们训练了一个多头注意力池化（MHAP）预测器，从采样补丁中恢复注册特征，同时保持预训练的扩散模型及其样本不变。在 ImageNet-1K 上，我们的 ViT-B 解码器在使用 DINOv2 时实现了 0.472 的平均重建 FID，而使用 DINOv3 时为 0.442，分别比各自的仅补丁基线减少了 9.5% 和 5.2%。相应的 PSNR 分别为 19.73 和 20.54 dB。在生成方面，我们的 DINOv2 模型在相同采样补丁上实现了 1.641 的平均 FID，比仅补丁基线低 9.8%。
cs.CV / 74 / 2610.13300
Sonic-AMG: Streaming Human Motion Generation from Spatial Audio
Sonic-AMG：基于空间音频的流式人类动作生成
Abstract
Modeling human reactions to spatial audio requires understanding both what is heard and where it comes from. Prior work generates motion offline from complete acoustic and spatial input sequences. Humans, however, react to sounds without knowing what they will hear next. We present Sonic-AMG, to our knowledge the first framework for streaming human motion generation from spatial audio. The model encodes acoustic content and body-relative source geometry as separate tokens. Conditioned on these tokens, a causal transformer with a flow head generates continuous motion latents, from which a causal decoder reconstructs motion segments. Standard teacher-forced training derives motion histories and body-relative source locations from ground-truth motion, whereas streaming inference relies on generated motion for both, resulting in a training-inference mismatch. To reduce this history discrepancy, we introduce rehearsal training, which combines full-sequence teacher-forced supervision with supervised prediction along short autoregressive self-rollouts. To keep spatial conditioning consistent with generated motion, we propose spatial feedback, which recomputes body-relative source locations from generated root poses during both rehearsal training and inference. On SAM, Sonic-AMG achieves state-of-the-art FID and R-precision. It further supports real-time online generation of coherent reactions across sound events and source locations. Project page: https://jingchengni.com/sonic-amg/.
Chinese Translation
建模人类对空间音频的反应需要理解所听到的内容及其来源。之前的研究通过完整的声学和空间输入序列离线生成动作。然而，人类在不知道接下来会听到什么的情况下对声音做出反应。我们提出了Sonic-AMG，作为我们所知的第一个基于空间音频的流式人类动作生成框架。该模型将声学内容和相对身体的源几何信息编码为独立的标记。在这些标记的条件下，一个带有流头的因果变换器生成连续的动作潜变量，随后因果解码器从中重建动作片段。标准的教师强制训练从真实动作中推导动作历史和相对身体的源位置，而流式推理则依赖于生成的动作来获取这两者，导致训练与推理之间的不匹配。为了减少这种历史差异，我们引入了复习训练，它结合了完整序列的教师强制监督与沿短期自回归自回滚的监督预测。为了保持空间条件与生成的动作一致，我们提出了空间反馈，它在复习训练和推理过程中根据生成的根姿势重新计算相对身体的源位置。在SAM上，Sonic-AMG达到了最先进的FID和R-precision。此外，它还支持在声音事件和源位置之间实时生成一致的反应。项目页面：https://jingchengni.com/sonic-amg/
cs.CV / 75 / 2610.13301
PAGE: Parts-Aware Guidance for Co-Speech Gesture Portrait Video Generation
PAGE：基于部件感知的共语手势肖像视频生成指导
Abstract
Co-speech gesture video generation aims to synthesize realistic, audio-driven portrait videos from a single reference image and a speech signal. Despite advances in visual quality, diffusion-based approaches still suffer from degraded local details and imperfect audio-visual alignment. We present PAGE (Parts-Aware GuidancE), a general, backbone-agnostic optimization framework for diffusion-based generation that enhances local spatial fidelity and global audio-visual consistency at both the training-objective and denoising-trajectory levels. During training, we propose Parts-Specific Reweighting (PSR), which adaptively increases supervision in semantically important regions such as lips, eyes, and hands, improving visual fidelity in perceptually critical areas. During inference, we introduce Consistency Classifier Guidance (CCG), an audio-conditioned latent classifier that identifies and corrects audio-visual inconsistencies in the generated video. We further design Differential Guidance (DG), an efficient inference strategy that refines facial and body motion separately while avoiding cross-region interference. To enable large-scale evaluation, we construct the Chinese Announcer Speech (CAS) dataset, the first comprehensive Chinese benchmark for co-speech gesture portrait video generation. Compared with the widely used PATS dataset, CAS includes substantially more identities, video clips, and scene diversity. Extensive experiments on both PATS and CAS demonstrate that PAGE achieves state-of-the-art performance in perceptual quality, gesture diversity, and audio-visual alignment. Project page: https://aigc-explorer.github.io/PAGE/.
Chinese Translation
共语手势视频生成旨在从单一参考图像和语音信号合成真实的、音频驱动的肖像视频。尽管在视觉质量上取得了进展，基于扩散的方法仍然存在局部细节退化和音视频对齐不完美的问题。我们提出了PAGE（基于部件感知的指导），这是一个通用的、与主干无关的优化框架，旨在增强扩散生成的局部空间保真度和全局音视频一致性，涵盖训练目标和去噪轨迹两个层面。在训练过程中，我们提出了部件特定重加权（Parts-Specific Reweighting, PSR），该方法自适应地增加在语义重要区域（如嘴唇、眼睛和手）中的监督，从而提高感知关键区域的视觉保真度。在推理阶段，我们引入了一种一致性分类器指导（Consistency Classifier Guidance, CCG），这是一种音频条件的潜在分类器，用于识别和修正生成视频中的音视频不一致性。我们进一步设计了差异指导（Differential Guidance, DG），这是一种高效的推理策略，能够分别优化面部和身体运动，同时避免跨区域干扰。为了实现大规模评估，我们构建了中文播音员语音（Chinese Announcer Speech, CAS）数据集，这是第一个全面的中文共语手势肖像视频生成基准。与广泛使用的PATS数据集相比，CAS包含了显著更多的身份、视频片段和场景多样性。在PATS和CAS上的大量实验表明，PAGE在感知质量、手势多样性和音视频对齐方面达到了最先进的性能。项目页面：https://aigc-explorer.github.io/PAGE/
cs.CV / 76 / 2610.13309
NovaSplat: Model-Free 6D Object Pose Estimation via 3D Gaussian Splatting Onboarding and Refinement
NovaSplat：通过3D高斯点云的引导和优化实现无模型的6D物体姿态估计
Abstract
6D pose estimation of novel objects without CAD models remains an open challenge: model-free methods must build the object representation from scratch during onboarding and have long trailed their model-based counterparts. We present NovaSplat, a model-free pipeline that closes this gap. From a short smartphone video captured at onboarding, we train a 3D Gaussian Splatting (3DGS) model that serves both as a template source for detection and coarse pose estimation and as a differentiable backbone for pose refinement. Our refiner optimizes the object-to-camera pose with combined photometric and geometric losses under a continuous rotation parametrization free of gradient singularities. On a household-object benchmark, NovaSplat surpasses all model-based methods. On a challenging industrial benchmark with textureless and specular surfaces, it outperforms all model-free competitors while substantially narrowing the gap to model-based ones. We further introduce NovaSplat-Lite, an inference variant that is $3.1\times$ faster at negligible accuracy cost. Code will be made available at \href{https://github.com/POSE-Lab/novasplat}{https://github.com/POSE-Lab/novasplat}.
Chinese Translation
在没有CAD模型的情况下对新颖物体进行6D姿态估计仍然是一个开放的挑战：无模型方法必须在引导过程中从头构建物体表示，且长期落后于基于模型的方法。我们提出了NovaSplat，这是一种无模型的管道，填补了这一空白。通过在引导阶段捕获的短手机视频，我们训练了一个3D高斯点云（3D Gaussian Splatting, 3DGS）模型，该模型既作为检测和粗略姿态估计的模板源，也作为姿态优化的可微分骨干。我们的优化器在没有梯度奇点的连续旋转参数化下，通过结合光度损失和几何损失来优化物体到相机的姿态。在家庭物体基准测试中，NovaSplat超越了所有基于模型的方法。在一个具有无纹理和镜面表面的挑战性工业基准测试中，它超越了所有无模型竞争者，同时显著缩小了与基于模型方法的差距。我们还推出了NovaSplat-Lite，这是一种推理变体，其速度提高了$3.1 imes$，而准确性损失微乎其微。代码将发布在 exttt{https://github.com/POSE-Lab/novasplat}。
cs.CV / 77 / 2610.13316
ThinkDrag: Semantic Drag-Based Image Editing with Visual Reasoning
ThinkDrag：基于语义的拖拽图像编辑与视觉推理
Abstract
Drag-based image editing provides an intuitive interface for spatially controlled image manipulation by allowing users to specify handle and target points. Existing methods have made substantial progress through optimization-based, one-step, and feed-forward formulations, but they often interpret drag instructions as geometric displacements. This limits their effectiveness when the desired edit depends on interpreting image content and inferring a plausible semantic transformation from sparse point constraints. We introduce ThinkDrag, a unified multimodal framework for drag-based image editing with visual reasoning. ThinkDrag is trained to associate point constraints with meaningful object- and scene-level transformations, and can optionally produce an explicit reasoning trace that interprets the intended edit before image generation. This reasoning trace improves interpretability, particularly for ambiguous drags. Drag Endpoint Markers, motivated by an analysis of cross-image attention, provide an explicit correspondence signal that grounds each drag in image space. To support this framework, we construct a supervised dataset of semantic drag transformations paired with reasoning traces and introduce DragBench++, a benchmark targeting challenging drag-based editing scenarios with reference edit solutions. Experiments show that ThinkDrag achieves state-of-the-art performance, improving generation quality and plausibility while maintaining competitive point-following precision.
Chinese Translation
基于拖拽的图像编辑通过允许用户指定控制点和目标点，提供了一种直观的空间控制图像操作界面。现有方法通过基于优化的一步和前馈形式取得了显著进展，但它们通常将拖拽指令解释为几何位移。这限制了它们在所需编辑依赖于解释图像内容并从稀疏点约束中推断出合理语义变换时的有效性。我们提出了ThinkDrag，一个统一的多模态框架，用于基于拖拽的图像编辑与视觉推理。ThinkDrag经过训练，将点约束与有意义的对象和场景级变换关联，并可以选择性地生成一个明确的推理轨迹，在图像生成之前解释预期的编辑。这一推理轨迹提高了可解释性，特别是对于模糊的拖拽。拖拽端点标记（Drag Endpoint Markers）是基于对跨图像注意力的分析而提出的，提供了一个明确的对应信号，将每个拖拽与图像空间中的位置相结合。为了支持这一框架，我们构建了一个配有推理轨迹的语义拖拽变换的监督数据集，并引入了DragBench++，一个针对具有参考编辑解决方案的挑战性拖拽编辑场景的基准测试。实验表明，ThinkDrag实现了最先进的性能，提高了生成质量和合理性，同时保持了竞争性的点跟踪精度。
cs.CV / 78 / 2610.13322
CoVA: Composing VAE Representations for Generative Learning
CoVA：用于生成学习的VAE表示组合
Abstract
Representation alignment with pretrained visual encoders improves diffusion transformer training but requires auxiliary encoders and extra training overhead. Recent approaches instead exploit representations from the generative model's own variational autoencoder (VAE). However, reconstruction-oriented VAE features are not optimized for the semantic discrimination and spatial structure needed for generative learning. We introduce \textbf{CoVA} (\textbf{Co}mposed \textbf{VA}E Representations), a lightweight framework that composes VAE features online into stage-wise guidance for flow-matching training. A small latent composer is jointly trained with class supervision, masked latent reconstruction, and horizontal-flip consistency, yielding semantically discriminative, spatially structured dense features and sample-dependent class posteriors. CoVA aligns intermediate SiT features with the composed representations early in training, then removes the increasingly detrimental alignment and reuses detached composer posteriors for sample-dependent soft label conditioning. We further uncover a hidden confound in soft label conditioning: averaging class embeddings attenuates their magnitude, coupling semantic interpolation with conditioning strength. This attenuation alone can substantially improve FID while degrading Recall, revealing a non-semantic source of the apparent gain. CoVA corrects this confound by restoring the hard-label magnitude, isolating posterior-dependent directional semantics. CoVA adds only 8.5M training-only auxiliary parameters (1.3\% of SiT-XL/2) and introduces no inference-time overhead. On ImageNet 256$\times$256, CoVA reduces SiT-XL/2 FID from 17.2 to 7.4 (\textbf{57\%}) in 400K steps. With classifier-free guidance, it achieves an FID of \textbf{2.03} in only 400 epochs, versus 2.06 after 1,400 epochs for vanilla SiT-XL/2.
Chinese Translation
与预训练视觉编码器的表示对齐可以改善扩散变换器的训练，但需要辅助编码器和额外的训练开销。最近的方法则利用生成模型自身的变分自编码器（VAE）中的表示。然而，面向重构的VAE特征并未针对生成学习所需的语义区分和空间结构进行优化。我们提出了 extbf{CoVA}（ extbf{Co}mposed extbf{VA}E Representations），一个轻量级框架，能够在线组合VAE特征，形成阶段性指导以用于流匹配训练。一个小型潜在组合器与类别监督、掩码潜在重构和水平翻转一致性共同训练，从而产生语义上具有区分性、空间结构化的稠密特征和样本依赖的类别后验。CoVA在训练早期将中间的SiT特征与组合表示对齐，然后去除日益有害的对齐，并重用分离的组合后验进行样本依赖的软标签调节。我们进一步发现了软标签调节中的一个隐藏混淆因素：平均类别嵌入会减弱其幅度，将语义插值与调节强度耦合在一起。仅此减弱就可以显著改善FID，同时降低召回率，揭示了表面增益的非语义来源。CoVA通过恢复硬标签幅度来纠正这一混淆，隔离后验依赖的方向语义。CoVA仅增加了8.5M仅用于训练的辅助参数（占SiT-XL/2的1.3%），且在推理时没有引入额外开销。在ImageNet 256$ imes$256上，CoVA将SiT-XL/2的FID从17.2降低到7.4（ extbf{57\%}），历时400K步。使用无分类器引导时，仅需400个周期便达到了 extbf{2.03}的FID，而对于原始SiT-XL/2在1400个周期后则为2.06。
cs.CV / 79 / 2610.13333
LETHE: Defending against Unauthorized Image-to-Video Generation via Value Diversity Collapse
LETHE：通过价值多样性崩溃防御未经授权的图像到视频生成
Abstract
Image-to-video (I2V) diffusion models can animate a single reference image into high-quality, subject-consistent videos, posing threats of misuse as anyone can turn personal photos into fabricated videos without consent. A promising solution is to proactively add an adversarial perturbation to the image before release, such that videos generated from it would be unfaithful or visually corrupted. However, due to an insufficient understanding of what governs how subsequent frames read the reference image in modern I2V models, existing protections struggle to offer strong protection under common perturbation budgets. In this paper, we approach the problem by exploring this underlying reading mechanism. Through a convex geometric analysis, we theoretically identify the spread of the reference value vectors as a key factor governing this reading process, and empirically find that even a mild contraction of this spread can severely degrade the generated video, making it an efficient lever for protection. Guided by this insight, we propose LETHE, a novel protection method that directly disrupts this reading process. LETHE pairs two complementary objectives, where a value collapse objective drives the reference value vectors toward uniformity, while an attention attraction objective draws subsequent frames toward these collapsed tokens to amplify corruption. We further construct two human-centric I2V protection benchmarks and conduct extensive experiments with multiple modern I2V models, which demonstrate the state-of-the-art protection performance and versatility of LETHE.
Chinese Translation
图像到视频（I2V）扩散模型能够将单一参考图像动画化为高质量、主题一致的视频，这带来了滥用的威胁，因为任何人都可以在未获同意的情况下将个人照片转化为伪造视频。一种有前景的解决方案是在发布之前主动向图像添加对抗性扰动，以使从中生成的视频不真实或在视觉上受到损坏。然而，由于对现代 I2V 模型中后续帧如何读取参考图像的机制理解不足，现有的保护措施在常见扰动预算下难以提供强有力的保护。在本文中，我们通过探索这一基础读取机制来解决该问题。通过凸几何分析，我们理论上确定了参考值向量的扩散作为支配这一读取过程的关键因素，并实证发现，即使是轻微的收缩也会严重降低生成视频的质量，从而成为一种有效的保护杠杆。在这一见解的指导下，我们提出了 LETHE，这是一种直接干扰这一读取过程的新型保护方法。LETHE 结合了两个互补的目标，其中价值崩溃目标驱动参考值向量趋向均匀，而注意力吸引目标则将后续帧吸引向这些崩溃的标记以增强损坏。我们进一步构建了两个以人为本的 I2V 保护基准，并对多个现代 I2V 模型进行了广泛实验，证明了 LETHE 的先进保护性能和多样性。
cs.CV / 80 / 2610.13337
Exploring Attention Sinks in Text-to-Image Transformers from Diffusion and Autoregressive Models
探索扩散模型和自回归模型中文本到图像变换器的注意力汇聚现象
Abstract
Attention sinks are tokens or positions that consistently receive disproportionate attention mass, often beyond what their semantic content would suggest. Despite extensive study in large language models, their behavior and functional relevance in text-to-image (T2I) Transformers remain largely unexplored. To fill this gap and better understand how text-to-image Transformers allocate and route attention, we explore this phenomenon of text-to-image transformers across bidirectional diffusion models and causal autoregressive models. Specifically, we explore whether attention sinks exist, where they emerge, and how they functionally influence generation. Through systematic attention analysis and targeted interventions, we identify that attention sinks emerge in both bidirectional diffusion and autoregressive models but exhibit distinct patterns. The autoregressive model concentrates attention on the first and global-scale tokens, whereas the bidirectional models develop sinks near modality boundaries, within text regions, and at spatial corners. Moreover, these patterns vary across generation stages and are localized to a sparse subset of layers and heads. To assess their functional relevance, we perturb sink-related heads and token positions. It turns out that perturbing sink-ranked heads produces larger average changes than non-sink comparisons, while some sink-position perturbations have limited effects on bidirectional model scores. We further introduce register tokens to redistribute attention sinks and find that several configurations preserve or modestly improve generation performance. Overall, attention sinks are architecture-, inference-stage-, and component-dependent phenomena, providing a basis for mechanistic analysis and principled attention design in text-to-image transformers.
Chinese Translation
注意力汇聚是指那些始终接收不成比例注意力质量的标记或位置，通常超出其语义内容所暗示的范围。尽管在大型语言模型中进行了广泛研究，但在文本到图像（T2I）变换器中，它们的行为和功能相关性仍然基本未被探索。为填补这一空白并更好地理解文本到图像变换器如何分配和引导注意力，我们在双向扩散模型和因果自回归模型中探讨了这一现象。具体而言，我们研究注意力汇聚是否存在、它们的出现位置以及它们如何在功能上影响生成。通过系统的注意力分析和针对性干预，我们发现注意力汇聚在双向扩散模型和自回归模型中均会出现，但表现出不同的模式。自回归模型将注意力集中在首个和全局范围的标记上，而双向模型则在模态边界附近、文本区域内和空间角落发展出汇聚现象。此外，这些模式在生成阶段之间有所不同，并且局限于稀疏的层和头部子集。为了评估它们的功能相关性，我们对与汇聚相关的头部和标记位置进行了扰动。结果表明，扰动汇聚排名的头部产生的平均变化大于非汇聚比较，而一些汇聚位置的扰动对双向模型得分的影响有限。我们进一步引入注册标记以重新分配注意力汇聚，发现几种配置能够保持或适度改善生成性能。总体而言，注意力汇聚是依赖于架构、推理阶段和组件的现象，为文本到图像变换器的机制分析和原则性注意力设计提供了基础。
cs.CV / 81 / 2610.13338
VideoAWR: Reinforcing Video Flow Models with Decoupled-Timestep Advantage-Weighted Rectification
VideoAWR：通过解耦时间步长优势加权修正增强视频流模型
Abstract
Regression-based RL for flow models mandates sampling \emph{without} classifier-free guidance (CFG) to align the rectification target with the reference marginal velocity. Extending Advantage-Weighted Rectification (AWR), which dynamically scales this regression target based on reward, to video generation encounters a cold-start pathology: unguided rollouts collapse into low-fidelity regions where intra-group reward gaps remain wide but no longer track perceptual quality, rendering the optimization target arbitrary. Conversely, utilizing guided rollouts to restore fidelity, while keeping the regression target tied to the guided sampling distribution, provably triggers a geometric guidance blow-up under exponential-moving-average (EMA) updates. We propose \textbf{VideoAWR}, which resolves this tension by untying sampling from the optimization anchor. An EMA reference policy generates \emph{guided} rollouts to restore reward separability, while the trainable policy is rectified against the \emph{unguided} conditional anchor. We prove this decoupled objective is mathematically exact: the centered advantage cancels the anchor in expectation, filtering out the reward-neutral guidance offset and retaining only the reward-correlated covariance increment. Consequently, inference runs single-branch and CFG-free, yet inference-time CFG preserves its original semantics. Furthermore, we introduce \emph{Decoupled Timestep Rectification}, supervising each rollout at an independently sampled timestep without bias to improve sample efficiency. Experiments on Wan2.1-1.3B and 14B demonstrate stable optimization and superior visual and motion quality in \emph{both} CFG-free and CFG-enabled regimes. Code is available at \pdfstartlink attr{/Border [0 0 0]} user{/Subtype /Link /A << /S /URI /URI (https://github.com/Fuzm7/FlowAWR) >>}% \textcolor{pearDark}{VideoAWR}\pdfendlink.
Chinese Translation
基于回归的强化学习（RL）对于流模型要求在没有分类器无引导（CFG）的情况下进行采样，以将修正目标与参考边际速度对齐。将动态根据奖励缩放的优势加权修正（AWR）扩展到视频生成时遇到了冷启动病态：无引导的展开会崩溃到低保真度区域，在这些区域内组内奖励差距依然较大，但不再跟踪感知质量，从而使优化目标变得任意。相反，利用有引导的展开来恢复保真度，同时将回归目标与引导采样分布绑定，证明在指数移动平均（EMA）更新下会触发几何引导爆炸。我们提出了 extbf{VideoAWR}，通过将采样与优化锚点解耦来解决这一矛盾。EMA参考策略生成 extit{有引导}的展开以恢复奖励可分性，而可训练策略则针对 extit{无引导}的条件锚点进行修正。我们证明这一解耦目标在数学上是精确的：中心优势在期望上抵消了锚点，过滤掉奖励中性引导偏移，仅保留与奖励相关的协方差增量。因此，推理过程运行单分支且无CFG，但推理时的CFG保留其原始语义。此外，我们引入了 extit{解耦时间步长修正}，在独立采样的时间步长上监督每个展开而不引入偏差，以提高样本效率。在Wan2.1-1.3B和14B上的实验表明，在 extit{无CFG}和 extit{有CFG}的情况下均实现了稳定的优化和优越的视觉及运动质量。代码可在此获取： extcolor{pearDark}{VideoAWR}。
cs.CV / 82 / 2610.13344
Why DiT Models Underperform as Representation Learners without Long Skip Connections
为何DiT模型在没有长跳跃连接时表现不佳作为表征学习者
Abstract
Diffusion feature, i.e., intermediate activations extracted from diffusion model backbones, has emerged as a promising approach for unifying generation and vision understanding (discrimination). Despite the advancement of diffusion backbones into Diffusion Transformers (DiT), it has been found in previous literature that DiT models tend to yield weaker features despite their stronger generation capability. This poses a threat to good generation-understanding unification. In this work, we show via controlled from-scratch training that long skip connections (LSCs) affect feature quality: a DiT with LSCs can produce better features. Furthermore, through detailed diagnosis, we reveal the mechanism causing this effect. Specifically, the input to a diffusion backbone contains both the latent image and the diffusion noise, where the diffusion noise, though appearing in the form of noises, actually reflects the information of the diffusion trajectory. To make the final prediction of a noisy variable ($v$ or $\epsilon$), the diffusion backbone needs both signals. However, carrying both signals through the main backbone effectively "pollutes" the image signal with noises, leading to poor feature quality. In comparison, LSCs can shortcut diffusion noises directly to where they are needed, leaving much weaker noise signals in the main backbone, thus improving feature quality. To name the broader impact, this reveals that LSCs are not a mere trick but closely related to the diffusion objective. The code is at https://github.com/Darkbblue/dit-lsc.
Chinese Translation
扩散特征，即从扩散模型主干提取的中间激活，已成为统一生成与视觉理解（区分）的一种有前景的方法。尽管扩散主干已发展为扩散变换器（Diffusion Transformers, DiT），但先前文献发现，尽管DiT模型具有更强的生成能力，但往往产生较弱的特征。这对良好的生成与理解统一构成了威胁。在本研究中，我们通过从头开始的受控训练表明，长跳跃连接（Long Skip Connections, LSCs）影响特征质量：带有LSCs的DiT能够产生更好的特征。此外，通过详细的诊断，我们揭示了导致这一效应的机制。具体而言，扩散主干的输入包含潜在图像和扩散噪声，其中扩散噪声虽然以噪声的形式出现，但实际上反映了扩散轨迹的信息。为了对一个噪声变量（$v$或$ ext{ε}$）做出最终预测，扩散主干需要这两种信号。然而，通过主干传递这两种信号会有效地“污染”图像信号，导致特征质量下降。相比之下，LSCs可以直接将扩散噪声快捷地传递到需要的地方，从而在主干中留下更弱的噪声信号，从而改善特征质量。更广泛地说，这揭示了LSCs不仅仅是一个技巧，而是与扩散目标密切相关。代码可在 https://github.com/Darkbblue/dit-lsc 获取。
cs.CV / 83 / 2610.13412
TextVBench: Evaluating Text Rendering and Preservation in Video Generation
TextVBench：评估视频生成中的文本渲染与保留
Abstract
Modern video generators can synthesize convincing scenes while rendering a few characters incorrectly, omitting requested text, or changing it across frames. Measuring these failures is harder than applying OCR independently to each frame: when a scene contains multiple text targets and multiple noisy OCR regions, the evaluator must first determine which observation belongs to which intended text. We introduce TextVBench, a benchmark and evaluation framework that formulates visual-text evaluation as a multi-target, multi-instance spatiotemporal correspondence problem. TextVBench contains two complementary tracks: TextVBench-T2V measures the creation and maintenance of prompt-specified text, while TextVBench-Ref2V measures preservation of text from reference images. Both tracks use target-conditioned candidate selection rather than unconstrained VLM transcription, support one target mapping to multiple OCR regions, and separate frame-level spatial fidelity from cross-frame temporal fidelity. A length-weighted harmonic score summarizes these factors, while extra-text diagnostics expose failure modes hidden by a single OCR score. The benchmark is organized around controlled text, layout, carrier, and motion attributes so that aggregate scores can be traced back to specific capability limits. Across the evaluated model runs, MiniMax-H3 reaches a T2V TextVScore of 68.32, while the strongest Ref2V score is 56.51; attribute slices reveal a pronounced drop as the number of simultaneous reference text regions increases.
Chinese Translation
现代视频生成器能够合成令人信服的场景，但在渲染时可能会错误地显示一些字符、遗漏请求的文本或在不同帧中更改文本。测量这些失败的难度超过了对每一帧独立应用光学字符识别（OCR）：当一个场景包含多个文本目标和多个嘈杂的OCR区域时，评估者必须首先确定哪个观察结果属于哪个预期文本。我们提出了TextVBench，一个基准和评估框架，将视觉-文本评估公式化为一个多目标、多实例的时空对应问题。TextVBench包含两个互补的轨道：TextVBench-T2V用于测量根据提示指定文本的创建和维护，而TextVBench-Ref2V则用于测量参考图像中文本的保留。这两个轨道使用目标条件的候选选择，而不是不受限制的视觉语言模型（VLM）转录，支持一个目标映射到多个OCR区域，并将帧级空间保真度与跨帧时间保真度分开。一个长度加权的调和得分总结了这些因素，而额外的文本诊断则揭示了单一OCR得分所隐藏的失败模式。该基准围绕受控文本、布局、载体和运动属性进行组织，以便聚合得分可以追溯到特定的能力限制。在评估的模型运行中，MiniMax-H3达到了68.32的T2V TextVScore，而最强的Ref2V得分为56.51；属性切片显示，随着同时参考文本区域数量的增加，得分显著下降。
cs.CV / 84 / 2610.13422
SADL: A Size-Adaptive Decision Layer for Chronic Stroke Lesion Segmentation in ISLES 2026
SADL：一种适应大小的决策层用于ISLES 2026中的慢性中风病灶分割
Abstract
Chronic ischemic stroke lesions are difficult to segment on T1-weighted MRI. Their appearance overlaps with healthy tissue, and their volumes span four orders of magnitude, so voxel-level and lesion level metrics often disagree. We build an eleven-member nnU-Net ensemble that covers five design axes: a baseline dice + top-K loss, heavy augmentation, per-lesion volume loss re-weighting, stochastic weight averaging, and a sampling-and-architecture block that pairs cohort-balanced sampling with a residual-encoder backbone. Weighted soft-map fusion of the members feeds a tuned decision layer, SADL, that estimates lesion volume, applies an adaptive per-lesion-size threshold, prunes components by voxel count and peak probability, and adds a never-empty rescue. All fusion weights and per-bucket parameters are fitted on out-of-fold (OOF) predictions, so tuning never touches the held-out test set. On a site-disjoint 91-case holdout, SADL reaches Dice 0.7322, lesion-F1 0.7461 and PR-AUC 0.8430 when scored online with the challenge's released evaluation code. A paired non-parametric bootstrap against a reference single ensemble member confirms the lesion-F1 gain at p=0.024, and the gain is largest in the small-lesion regime, where the rescue step recovers cases that a fixed threshold would silently discard.
Chinese Translation
慢性缺血性中风病灶在T1加权MRI上难以进行分割。它们的外观与健康组织重叠，且其体积跨越四个数量级，因此体素级和病灶级指标往往存在不一致。我们构建了一个由十一名成员组成的nnU-Net集成，涵盖五个设计轴：基线Dice + top-K损失、重度增强、每个病灶体积损失重加权、随机权重平均以及一个将群体平衡采样与残差编码器主干相结合的采样与架构模块。成员的加权软图融合为一个调优的决策层SADL提供输入，该层估计病灶体积，应用适应性每个病灶大小的阈值，通过体素计数和峰值概率修剪组件，并添加一个永不为空的救援步骤。所有融合权重和每个桶参数均在折叠外（OOF）预测上进行拟合，因此调优从未接触到保留的测试集。在一个与站点不重叠的91例保留集中，当使用挑战发布的评估代码在线评分时，SADL达到了Dice 0.7322，病灶-F1 0.7461和PR-AUC 0.8430。与参考单一集成成员进行的配对非参数自助法确认了病灶-F1的增益，p=0.024，并且增益在小病灶范围内最大，在该范围内，救援步骤恢复了固定阈值会默默丢弃的病例。
cs.CV / 85 / 2610.13465
When MLLMs miss what humans read: diagnosing and learning to read squint text
当多模态大语言模型无法识别人类阅读的文本时：诊断和学习阅读眯眼文本
Abstract
Multimodal LLMs can fail to read text embedded in an image's coarse structure (text a human recovers by squinting) and report that no text is present. We name this class Squint Text and study it as a recognition deficiency, motivated by cases from a production moderation pipeline. We introduce SquintBench, comprising 1076 production-derived moderation images, and SquintCtrl, a controlled synthetic suite for mechanism diagnosis. Frozen-weight analysis shows that hidden text leaves a coarse signal while glyph identity is weakly represented; identity becomes substantially more accessible when the correct spatial regions are supplied, implicating spatial selection as a major bottleneck. Suppressing high-frequency content also improves reading at a fixed visual-token budget, while visual token count may also contribute. These findings motivate Squint OPSD: a frozen teacher receives a downscaled view and the ground-truth transcription as privileged context, and its token-level distributions supervise a student along on-policy trajectories. At inference, the student reads the original image alone. Squint OPSD reaches 88.61 ANLS on SquintBench, versus 26.50 for the unadapted backbone, 24.44 for the best general-purpose model on the original view, and 68.60 at best for an off-the-shelf model under a downscaled view. These results substantially narrow the gap between available visual evidence and its generated read-out without auxiliary views at inference.
Chinese Translation
多模态大语言模型（MLLMs）可能无法读取嵌入在图像粗糙结构中的文本（人类通过眯眼恢复的文本），并报告没有文本存在。我们将这一类文本称为眯眼文本（Squint Text），并将其作为一种识别缺陷进行研究，研究动机源自生产审核流程中的案例。我们引入了SquintBench，包含1076个源自生产的审核图像，以及SquintCtrl，一个用于机制诊断的受控合成套件。冷冻权重分析表明，隐藏文本留下了粗糙信号，而字形身份的表示较弱；当提供正确的空间区域时，身份变得显著更容易获取，这表明空间选择是一个主要瓶颈。抑制高频内容也在固定视觉标记预算下改善了阅读效果，而视觉标记数量可能也有助于此。这些发现促使我们提出Squint OPSD：一个冷冻教师接收降尺度视图和真实转录作为特权上下文，其标记级分布在策略轨迹上监督学生。在推理阶段，学生仅读取原始图像。Squint OPSD在SquintBench上达到了88.61的ANLS，相比之下，未适配的主干模型为26.50，最佳通用模型在原始视图下为24.44，降尺度视图下的现成模型最佳为68.60。这些结果显著缩小了可用视觉证据与其生成的读取结果之间的差距，而无需在推理时使用辅助视图。
cs.CV / 86 / 2610.13466
A Real-World CAD Segmentation Benchmark: Comparing Rule-Based and Learning-Based Approaches through Hole Detection
一个真实世界的CAD分割基准：通过孔检测比较基于规则和基于学习的方法
Abstract
CAD segmentation datasets are predominantly synthetic or automatically labelled, leaving it unclear how well existing methods transfer to real-world engineering tasks. Fastener-relevant hole detection represents one such task, requiring human annotation and exhibiting greater geometric complexity than commonly used benchmarks. We address this gap through three main contributions. First, we develop two task-specific rule-based methods, one graph-wire-search and one multi-view based, and complement them with a human-annotated benchmark dataset derived from engineering CAD data that also enables training of UV-Net and AAGNet, representative learning-based approaches for B-Rep segmentation. Second, we conduct an extensive experimental study including parameter optimisation, evaluation on our benchmark dataset, cross-domain generalisation on a synthetic dataset, and an assessment of the feasibility of the approaches at varying dataset sizes and IoU thresholds. Third, we provide evidence-based insights into the strengths and limitations of different approaches: rule-based methods generalise well without training data, with the graph-wire-search method favouring recall and the multi-view method favouring precision, whereas learning-based methods achieve strong performance with sufficient annotations but exhibit reduced robustness under domain shift. These findings highlight trade-offs between data requirements, and robustness, and provide guidance not only for hole detection, but more broadly for real-world CAD segmentation tasks.
Chinese Translation
CAD分割数据集主要是合成的或自动标注的，这使得现有方法在真实工程任务中的迁移效果尚不明确。与紧固件相关的孔检测就是这样一个任务，它需要人工标注，并且展现出比常用基准更大的几何复杂性。我们通过三个主要贡献来填补这一空白。首先，我们开发了两种特定任务的基于规则的方法，一种是图-线搜索（graph-wire-search），另一种是基于多视图的方法，并且补充了一个源自工程CAD数据的人工标注基准数据集，该数据集还支持UV-Net和AAGNet的训练，这些是B-Rep分割的代表性基于学习的方法。其次，我们进行了广泛的实验研究，包括参数优化、在我们的基准数据集上的评估、在合成数据集上的跨领域泛化，以及在不同数据集大小和IoU阈值下对方法可行性的评估。第三，我们提供了基于证据的见解，揭示了不同方法的优缺点：基于规则的方法在没有训练数据的情况下泛化良好，图-线搜索方法更偏向于召回率，而多视图方法则更偏向于精确度；而基于学习的方法在有足够标注的情况下表现强劲，但在领域迁移下表现出降低的鲁棒性。这些发现突显了数据需求与鲁棒性之间的权衡，并为孔检测以及更广泛的真实世界CAD分割任务提供了指导。
cs.CV / 87 / 2610.13470
GraftOOD: Grafting Out-of-Distribution Exemplars into Advantage-Weighted Rectification for Video Reward Optimization
GraftOOD：将分布外示例嫁接到优势加权校正中以优化视频奖励
Abstract
Video reward optimization can increase proxy rewards while reducing video dynamics, a mismatch we identify as \emph{motion collapse}, including under a video-level motion-quality reward. We propose \textbf{GraftOOD}, which replaces part of the online rollouts in selected prompt groups with out-of-distribution video exemplars. A shared velocity-regression objective combines group-relative reward rectification with sequence-level exemplar supervision, requiring neither exemplar generation trajectories nor an additional supervised-loss coefficient. We characterize its pointwise optimal field as a source-posterior-weighted combination of the two source-specific target fields. We construct \textbf{GraftSet-Video}, containing 9,588 format-matched exemplars for 2,397 prompts, to supply external supervision. On Wan2.1-1.3B, GraftOOD raises Dynamic Degree from 4.86 to 44.41 relative to reward-only FlowAWR, while HPSv3 decreases from 8.10 to 7.32 and VisionReward remains close. At the main setting, GraftOOD leads nine of ten metrics among three exemplar-augmented methods. Ablations of exemplar frequency and within-group count reveal dynamics--reward trade-offs rather than uniform gains from increasing exemplar usage. These results support external exemplar supervision as a complement to scalar reward feedback, without equating higher measured dynamics with coherent, visually plausible motion in the generated videos. Code is available at \href{https://github.com/Fuzm7/FlowAWR}{GraftOOD}.
Chinese Translation
视频奖励优化可以在减少视频动态的同时增加代理奖励，我们将这种不匹配称为 extit{运动崩溃}，即使在视频级运动质量奖励下也是如此。我们提出了 extbf{GraftOOD}，它用分布外视频示例替换选定提示组中的部分在线回放。一个共享的速度回归目标将组相对奖励校正与序列级示例监督结合在一起，既不需要示例生成轨迹，也不需要额外的监督损失系数。我们将其逐点最优场描述为两个源特定目标场的源后验加权组合。我们构建了 extbf{GraftSet-Video}，包含9,588个格式匹配的示例，针对2,397个提示，以提供外部监督。在Wan2.1-1.3B上，GraftOOD使动态度从4.86提高到44.41，相较于仅奖励的FlowAWR，而HPSv3则从8.10降至7.32，VisionReward保持接近。在主要设置中，GraftOOD在三种增强示例的方法中领先十个指标中的九个。示例频率和组内计数的消融实验揭示了动态与奖励之间的权衡，而不是通过增加示例使用获得的均匀收益。这些结果支持外部示例监督作为标量奖励反馈的补充，而不将更高的测量动态等同于生成视频中连贯且视觉上合理的运动。代码可在 exttt{https://github.com/Fuzm7/FlowAWR}找到。
cs.CV / 88 / 2610.13502
Stride: Scale-Aware Multi-Agent Evolutionary Refinement for Infographic Generation
Stride：一种规模感知的多智能体进化精炼方法用于信息图生成
Abstract
Infographic generation challenges text-to-image models because a single image must satisfy numerous visual, textual, and semantic constraints. Iterative agentic refinement with a multimodal verifier is a natural remedy, but existing methods face a scale mismatch: local edits cannot correct global composition errors, while full regeneration may destroy already-correct regions. We introduce Stride, a constraint-grounded multi-agent framework that treats refinement over any frozen generator as a history-aware evolution strategy. Stride separates constraint evaluation - whether each requirement is satisfied - from visual grounding - where and how it is violated. It then routes global revisions to a composer subagent and local corrections to a retoucher subagent. Each subagent conducts internal planning and generates multiple candidates, while Stride accepts a candidate only when it improves the verifier score, reducing regression across iterations. Because stochastic generators and imperfect verifiers make any single step unreliable, it is this combination of scale-aware routing, per-step redundancy, and monotonic acceptance that yields stable improvement across iterations. With fully open-weight generator-verifier systems, Stride outperforms proprietary systems including Nano-Banana-2.0 and Seedream-5.0-Pro on BizGenEval, matches GPT-Image-2.0 and surpasses Seedream-5.0-Pro on IGenBench, and achieves a new state-of-the-art score on WISE, while substantially outperforming existing agentic methods across all evaluated benchmarks.
Chinese Translation
信息图生成对文本到图像模型提出了挑战，因为单一图像必须满足众多视觉、文本和语义约束。采用多模态验证器的迭代智能体精炼是一种自然的解决方案，但现有方法面临规模不匹配的问题：局部编辑无法纠正全局构图错误，而完全重生可能会破坏已经正确的区域。我们提出了Stride，一个基于约束的多智能体框架，将对任何冻结生成器的精炼视为一种历史感知的进化策略。Stride将约束评估（即每个要求是否得到满足）与视觉基础（即在哪里以及如何被违反）分开。然后，它将全局修订路由到作曲子智能体，将局部修正路由到修饰子智能体。每个子智能体进行内部规划并生成多个候选项，而Stride仅在候选项提高验证器分数时接受该候选项，从而减少迭代过程中的回归。由于随机生成器和不完美的验证器使得任何单一步骤都不可靠，因此这种规模感知路由、每步冗余和单调接受的组合在迭代过程中实现了稳定的改进。在完全开放权重的生成器-验证器系统中，Stride在BizGenEval上超越了包括Nano-Banana-2.0和Seedream-5.0-Pro在内的专有系统，在IGenBench上与GPT-Image-2.0持平，并在WISE上取得了新的最先进分数，同时在所有评估基准上大幅超越现有的智能体方法。
cs.CV / 89 / 2610.13503
SplatQ: Interaction-Aware Pruning of 3D Gaussian Splatting via Quadratic Selection
SplatQ：基于二次选择的3D高斯点云交互感知剪枝
Abstract
Existing 3D Gaussian Splatting (3DGS) pruning methods primarily rely on global Top-K selection based on independent importance scores, lacking regional budget control and ignoring joint deletion effects. Consequently, they cannot ensure balanced reconstruction quality across a scene. We propose SplatQ, a combinatorial pruning algorithm that coordinates regional budget allocation and joint Gaussian selection under a fixed budget. SplatQ partitions Gaussians by spatial proximity and appearance similarity into size-limited blocks, decomposing global selection into subproblems of a specific size. It then smooths importance-based block retention quotas to prevent excessive budget concentration. Finally, it constructs a quadratic approximation of joint deletion effects using the responses of representative pixel colors to changes in Gaussian opacity. Individual responses, pairwise interactions, and quota penalties are incorporated into a quadratic unconstrained binary optimization (QUBO) objective, solved using simulated annealing (SA) or a coherent Ising machine (CIM) to jointly select Gaussians. Experiments on 5 benchmarks demonstrate that SplatQ outperforms state-of-the-art pruning methods in aggregate reconstruction quality, spatial quality balance, and rendering speed. Ablations confirm that budget smoothing improves local detail reconstruction, while interaction modeling improves Gaussian selection quality under a fixed budget.
Chinese Translation
现有的3D高斯点云（3DGS）剪枝方法主要依赖于基于独立重要性评分的全局Top-K选择，缺乏区域预算控制，并忽视联合删除效应。因此，它们无法确保场景中重建质量的平衡。我们提出了SplatQ，这是一种组合剪枝算法，在固定预算下协调区域预算分配和联合高斯选择。SplatQ根据空间接近性和外观相似性将高斯分区为大小有限的块，将全局选择分解为特定大小的子问题。然后，它平滑基于重要性的块保留配额，以防止预算过度集中。最后，它利用代表性像素颜色对高斯不透明度变化的响应构建联合删除效应的二次近似。个体响应、成对交互和配额惩罚被纳入二次无约束二进制优化（QUBO）目标，通过模拟退火（SA）或相干伊辛机（CIM）解决，以联合选择高斯。在5个基准测试上的实验表明，SplatQ在整体重建质量、空间质量平衡和渲染速度上优于最先进的剪枝方法。消融实验确认，预算平滑改善了局部细节重建，而交互建模在固定预算下提高了高斯选择质量。
cs.CV / 90 / 2610.13527
S-Space: Exploring Spatial Workspace in Multimodal Language Models
S-Space：探索多模态语言模型中的空间工作空间
Abstract
Multimodal language models (MLLMs) exhibit promising spatial reasoning capabilities, yet the internal mechanisms supporting them remain poorly understood. We uncover S-Space, a linear representation subspace in which object-token representations encode continuous coordinates along horizontal, vertical, and distance axes. Although we identify these axes by differentiating spatial-answer scores with respect to object-token activations, S-Space is intrinsic rather than induced by specific data or prompts, remaining readable under non-spatial questions. We show that S-Space functions as an internal spatial workspace of MLLMs. Perception encodes object locations into it, forming spatial representations that can be directly read out to support accurate spatial judgments. Internal reasoning transforms it for changing viewpoints. Verbal responses selectively access it, evidenced by causal interventions that alter spatial judgments while largely preserving non-spatial responses. Case studies further show that S-Space assigns locations to unseen entities and the observer, and connects physical directions with abstract linguistic meanings, suggesting both the capabilities and limitations of large-scale language training. Finally, we use S-Space as a diagnostic tool for spatial reasoning failures. Direction-word readouts reveal systematic coupling between cardinal and image-relative reference frames. More importantly, a perception-computation decomposition exposes a key bottleneck: MLLMs can form capable spatial representations, yet fall short in the spatial transformations required to reason reliably across viewpoint changes.
Chinese Translation
多模态语言模型（MLLMs）展现出令人期待的空间推理能力，但支持这些能力的内部机制仍然不甚明了。我们揭示了 S-Space，这是一个线性表示子空间，其中对象标记表示沿水平、垂直和距离轴编码连续坐标。尽管我们通过区分与对象标记激活相关的空间答案分数来识别这些轴，S-Space 是内在的，而非由特定数据或提示引起的，并且在非空间问题下仍然可读。我们展示了 S-Space 作为 MLLMs 的内部空间工作空间的功能。感知将对象位置编码到其中，形成可以直接读取的空间表示，以支持准确的空间判断。内部推理对其进行变换，以适应不同的视角。语言响应选择性地访问它，因果干预的证据表明，这种干预改变了空间判断，同时在很大程度上保留了非空间响应。案例研究进一步表明，S-Space 为未见实体和观察者分配位置，并将物理方向与抽象语言意义联系起来，暗示了大规模语言训练的能力和局限性。最后，我们使用 S-Space 作为空间推理失败的诊断工具。方向词的读出揭示了基于基准和图像相对参考框架之间的系统耦合。更重要的是，感知-计算分解揭示了一个关键瓶颈：MLLMs 可以形成有效的空间表示，但在跨视角变化进行可靠推理所需的空间变换方面却显得不足。
cs.CV / 91 / 2610.13535
Knowing Whether, Not Just Where: Anchor-Probing Distillation for Open-Set Remote Sensing Visual Grounding
不仅要知道位置，更要知道是否存在：开放集遥感视觉定位的锚点探测蒸馏
Abstract
Open-set visual grounding in remote sensing requires models to determine not only where a queried object is located, but also whether the image provides sufficient evidence for the query at all. Existing closed-set grounding models, however, tend to return plausible boxes even for absent objects, blurring localization confidence with genuine Visual Support. We introduce AnchorProbe-Distill (APD), a teacher-student framework that recasts open-set grounding as Visual Support learning. APD builds on the insight that supported queries should induce coherent spatial anchors, whereas unsupported or mismatched queries tend to produce unstable or misleading localization patterns. APD-T therefore uses the Original Query as a Reference Anchor to interpret query perturbations: Semantic-Preserving Queries retain the annotated target, Near-Category Ambiguous Queries are resolved according to whether they form a reliable alternative anchor or collapse onto the Reference Anchor, and Far-Category Counterfactual Queries provide rejection supervision. This reference-aware reasoning is distilled into APD-S, enabling unified localization and rejection from a single image--query pair. Experiments show that APD enables the resulting model to reject absent targets while also improving its localization of present targets.
Chinese Translation
开放集视觉定位在遥感中要求模型不仅要确定查询对象的位置，还要判断图像是否提供了足够的证据来支持该查询。然而，现有的闭集定位模型往往即使在缺失对象的情况下也会返回合理的框，这使得定位的可信度与真实的视觉支持混淆。我们提出了 AnchorProbe-Distill (APD)，一个将开放集定位重构为视觉支持学习的教师-学生框架。APD 基于以下见解：被支持的查询应诱导出一致的空间锚点，而不被支持或不匹配的查询往往会产生不稳定或误导的定位模式。因此，APD-T 使用原始查询作为参考锚点来解释查询扰动：语义保留查询保持注释目标，近类模糊查询根据它们是否形成可靠的替代锚点或崩溃到参考锚点来解决，而远类反事实查询提供拒绝监督。这种参考感知推理被蒸馏为 APD-S，使得从单个图像-查询对中实现统一的定位和拒绝。实验表明，APD 使得生成的模型能够拒绝缺失的目标，同时改善对存在目标的定位。
cs.CV / 92 / 2610.13541
Efficient Confidence-Aware Densification for 3DGS
高效的置信度感知稠密化用于3DGS
Abstract
3DGS-MCMC has replaced the gradient-driven density control of 3D Gaussian Splatting with opacity-based sampling and relocation, and is now the de facto standard for prior-free densification under a fixed primitive budget. Because opacity stays high where the scene is already well reconstructed, this sampling sends most new primitives to the foreground and leaves the background under-reconstructed. At low budgets, prior-based methods such as ConeGS and EDGS outperform it, at the cost of a pre-trained radiance field or dense image matches and a separate initialization stage. We argue for a comeback of error-driven densification within the fixed-budget MCMC framework. Our confidence-aware densification keeps the fixed budget and gradual growth of 3DGS-MCMC, but places primitives by the image error instead of the opacity. New primitives are drawn mostly where screen-space gradients are large. We favor primitives with a low learned confidence, which marks regions the reconstruction still explains poorly, and relocate the primitives that contribute least to the rendered images to these places. Averaged over 24 scenes, our method outperforms 3DGS-MCMC in PSNR, SSIM and LPIPS at every budget from 100k to 2M primitives. At 100k, without a prior or an initialization stage, it recovers 0.84dB of the 1.05dB PSNR by which ConeGS leads 3DGS-MCMC, and from 500k on, it has the highest PSNR and SSIM of all fixed-budget methods we compare with, including ConeGS and EDGS. While inherently prior-free, our densification can also exploit priors: from the ConeGS or EDGS initialization, it matches ConeGS in PSNR and SSIM at 100k and is best in all three metrics at 500k.
Chinese Translation
3DGS-MCMC用基于不透明度的采样和重定位替代了3D高斯喷溅的梯度驱动密度控制，现已成为在固定原始预算下无先验稠密化的事实标准。由于在场景已被良好重建的地方不透明度保持较高，这种采样将大多数新原始体发送到前景，而使背景重建不足。在低预算下，基于先验的方法如ConeGS和EDGS在性能上优于它，但代价是需要预训练的辐射场或密集图像匹配以及单独的初始化阶段。我们主张在固定预算的MCMC框架内重新引入基于误差的稠密化。我们的置信度感知稠密化保持了固定预算和3DGS-MCMC的逐步增长，但通过图像误差而非不透明度来放置原始体。新的原始体主要在屏幕空间梯度较大的地方生成。我们倾向于选择低学习置信度的原始体，这标记了重建仍然解释不佳的区域，并将对渲染图像贡献最小的原始体重新定位到这些地方。在24个场景的平均结果中，我们的方法在每个预算（从100k到2M原始体）上都在PSNR、SSIM和LPIPS上优于3DGS-MCMC。在100k时，无需先验或初始化阶段，它恢复了ConeGS领先于3DGS-MCMC的1.05dB PSNR中的0.84dB，从500k开始，它在所有我们比较的固定预算方法中，包括ConeGS和EDGS，具有最高的PSNR和SSIM。尽管本质上是无先验的，我们的稠密化也可以利用先验：从ConeGS或EDGS初始化开始，在100k时其PSNR和SSIM与ConeGS相匹配，并在500k时在所有三个指标中表现最佳。
cs.CV / 93 / 2610.13547
MOSAIC: Multi-Scale Object-Centric Slot Adaptation with Image-Guided Prototype Conditioning for Remote Sensing Image Super-Resolution
MOSAIC：基于图像引导原型调节的多尺度对象中心插槽适应用于遥感图像超分辨率
Abstract
Pretrained diffusion Transformers provide powerful generative priors for image restoration, yet adapting them to structure-sensitive tasks remains challenging. Existing methods mainly rely on dense low-resolution features as conditions, which often entangle salient objects with dominant backgrounds and provide limited abstraction for recurring structures across scales and samples. We propose MOSAIC, an object-centric representation learning framework that learns compact structural conditions for generative model adaptation. MOSAIC organizes degraded observations into structural slots that capture complementary object-level and layout-level evidence, integrates them across resolutions, and retrieves recurring priors from a shared prototype memory to complement incomplete image-specific information. The resulting representations are injected into selected layers of a frozen diffusion Transformer through lightweight residual modulation, preserving its pretrained generative prior. We instantiate MOSAIC for remote sensing image super-resolution, where small objects, repetitive structures, and large-scale variation provide a challenging testbed. Experiments on UCMerced, AID, and RSSCN7 show consistent PSNR improvements across $\times 2$ and $\times 4$ settings and favorable LPIPS results at $\times 4$. Qualitative results further demonstrate improved preservation of small objects, structural boundaries, and repetitive spatial patterns.
Chinese Translation
预训练的扩散变换器为图像恢复提供了强大的生成先验，但将其适应于结构敏感任务仍然具有挑战性。现有方法主要依赖于稠密的低分辨率特征作为条件，这往往将显著对象与主导背景纠缠在一起，并为跨尺度和样本的重复结构提供有限的抽象。我们提出了MOSAIC，一个以对象为中心的表示学习框架，学习用于生成模型适应的紧凑结构条件。MOSAIC将退化的观测组织成结构插槽，捕捉互补的对象级和布局级证据，跨分辨率整合它们，并从共享的原型记忆中检索重复的先验，以补充不完整的图像特定信息。生成的表示通过轻量级的残差调制注入到冻结的扩散变换器的选定层中，保留其预训练的生成先验。我们将MOSAIC实例化用于遥感图像超分辨率，其中小对象、重复结构和大尺度变化提供了一个具有挑战性的测试平台。在UCMerced、AID和RSSCN7上的实验显示，在$ imes 2$和$ imes 4$设置下，PSNR一致提高，并且在$ imes 4$下LPIPS结果良好。定性结果进一步证明了小对象、结构边界和重复空间模式的保留得到了改善。
cs.CV / 94 / 2610.13548
Auditing VLM Benchmarks from Existing Evaluation Runs
审计现有评估运行的视觉语言模型基准
Abstract
Every evaluation run of a vision-language model (VLM) already produces per-question artifacts (predictions, judged correctness, metadata) that leaderboards reduce to a single score and then discard. We reuse those artifacts to audit the benchmark itself, and add an audit layer to VLMEvalKit: six detectors that run as post-processing over saved outputs, plus one additional text-only rerun with visual inputs removed, emitting a per-question report. The layer reports per-question findings together with aggregate audit statistics; for the headline quantities, it attaches permutation-null excess where applicable, interval estimates for proportions, and sensitivity to the model pool. Auditing MMStar, a benchmark curated specifically so that the image is indispensable, with three models, we find zero aggregate accuracy gain from the image on 20.2% of its questions. An item-matched permutation null accounts for 17.0%, leaving a +3.2 pp excess (corrected permutation p = 0.001). This excess is concentrated in the mathematics and science categories and is absent from coarse perception, a localization consistent with knowledge substituting for visual evidence; confirming that mechanism would require item-level review. A second detector flags 134 questions on which every model that answered in parseable form picked the same non-gold option, and these candidates peak in the perception categories, whereas the zero-gain excess is largest in mathematics and science & technology. Replacing the LLM judge with exact string matching changes no conclusion, and every scoring discrepancy between the two traces to one model whose free-form answers a parser cannot read. On a binary yes/no benchmark the detectors measurably fall out of calibration, a scope boundary that we measure and report. The audit layer, the prediction artifacts and the generated reports accompany this submission.
Chinese Translation
每次视觉语言模型（VLM）的评估运行都会产生每个问题的工件（预测、判断正确性、元数据），而排行榜将这些工件简化为一个分数后便被丢弃。我们重新利用这些工件来审计基准，并在 VLMEvalKit 中增加一个审计层：六个检测器作为后处理运行在保存的输出上，以及一个额外的仅文本重跑，去除视觉输入，生成每个问题的报告。该层报告每个问题的发现以及汇总审计统计数据；对于主要数量，它附加了适用时的置换-零过剩、比例的区间估计和对模型池的敏感性。审计 MMStar，这是一个专门策划的基准，使得图像不可或缺，使用三种模型，我们发现其 20.2% 的问题在图像上没有整体准确率提升。一个与项目匹配的置换零解释了 17.0%，留下 +3.2 个百分点的过剩（校正置换 p = 0.001）。这种过剩集中在数学和科学类别中，而在粗略感知中则不存在，这与知识替代视觉证据的定位一致；确认该机制需要逐项审查。第二个检测器标记了 134 个问题，在这些问题上，所有以可解析形式回答的模型都选择了相同的非金标准选项，这些候选项在感知类别中达到峰值，而零增益过剩在数学和科学与技术中最大。用精确字符串匹配替换 LLM 判断不会改变任何结论，两个评分之间的每个差异都追溯到一个模型，其自由格式答案无法被解析器读取。在一个二元是/否基准上，检测器的校准明显失效，这是我们测量和报告的范围边界。审计层、预测工件和生成的报告随此提交一起附上。
cs.CV / 95 / 2610.13549
ELMA: Benchmarking Anaphoric Compositions for Long-Term Text-to-Motion Generation
ELMA：用于长期文本到运动生成的指代组合基准测试
Abstract
Text-guided long-term human motion generation is essential for applications such as continuously operating digital avatars. Despite recent progress on short clips, existing models struggle with multi-segment instructions that involve anaphora, where constraints established in earlier segments must be preserved over time. In this paper, we introduce a novel task of long-term motion generation from such anaphoric compositions. To mitigate this, we present ELMA, an episodic long-term motion dataset with dense segment-level annotations that preserve cross-segment dependencies. Built via a fully automated collection and filtering pipeline, ELMA extracts high-quality, continuous 30-second clips from in-the-wild videos, capturing the episodic narratives necessary for long-range state modeling. Leveraging this dataset, we fine-tune state-of-the-art motion generation models to incorporate long-range context. Extensive experiments reveal fundamental shortcomings in current models when handling context-dependent instructions. We demonstrate that ELMA enables significant improvements in both global physical realism and semantic alignment, setting a robust new baseline for long-term motion synthesis.
Chinese Translation
文本引导的长期人类运动生成对于持续运行的数字化身等应用至关重要。尽管近期在短片段生成方面取得了一定进展，但现有模型在处理涉及指代的多段指令时仍然面临困难，这些指令要求在时间上保持早期段落中建立的约束。在本文中，我们提出了一项新的任务，即从这种指代组合中生成长期运动。为了解决这一问题，我们提出了ELMA，一个具有密集段级注释的情节长期运动数据集，能够保留跨段依赖关系。ELMA通过完全自动化的收集和过滤流程构建，从野外视频中提取高质量、连续的30秒片段，捕捉进行长期状态建模所需的情节叙事。利用该数据集，我们对最先进的运动生成模型进行了微调，以融入长期上下文。大量实验揭示了当前模型在处理依赖上下文的指令时的基本缺陷。我们证明ELMA在全球物理现实性和语义对齐方面实现了显著改进，为长期运动合成设定了一个稳健的新基准。
cs.CV / 96 / 2610.13553
Learning Holistic Scene Layout against View Shifts in Cross-View Geo-Localization
针对视角变化的全局场景布局学习在跨视角地理定位中的应用
Abstract
Cross-view geo-localization remains challenging due to drastic appearance variations and geometric distortions caused by viewpoint changes. Existing methods primarily rely on discriminative feature learning or explicit geometric alignment, yet often overlook an important observation: humans tend to match scenes by integrating the spatial context of multiple landmarks rather than relying solely on isolated local visual cues. Inspired by this, we propose a Spatial Context Modeling (SCM) module, which traverses feature maps along four orthogonal directions and models interactions among the resulting directional representations at each spatial position. By aggregating complementary directional observations, SCM constructs richer scene layout representations. Since reliable spatial context depends on robust visual features, we introduce a High-Frequency Information Guidance (HFIG) module that extracts stable structural cues from the frequency domain. By emphasizing edges and contours that are less sensitive to appearance variations, HFIG encourages the model to learn more robust representations. Extensive experiments demonstrate that our method achieves competitive performance on standard cross-view geo-localization benchmarks while exhibiting strong robustness under more challenging scenarios with unknown orientations and limited fields of view (FoVs).
Chinese Translation
跨视角地理定位仍然面临挑战，主要由于视角变化引起的外观剧烈变化和几何扭曲。现有方法主要依赖于判别特征学习或显式几何对齐，但往往忽视了一个重要观察：人类倾向于通过整合多个地标的空间上下文来匹配场景，而不是仅仅依赖孤立的局部视觉线索。受此启发，我们提出了一种空间上下文建模（Spatial Context Modeling，SCM）模块，该模块沿四个正交方向遍历特征图，并在每个空间位置建模生成的方向表示之间的交互。通过聚合互补的方向观察，SCM 构建了更丰富的场景布局表示。由于可靠的空间上下文依赖于稳健的视觉特征，我们引入了一种高频信息引导（High-Frequency Information Guidance，HFIG）模块，从频域中提取稳定的结构线索。通过强调对外观变化不太敏感的边缘和轮廓，HFIG 鼓励模型学习更稳健的表示。大量实验表明，我们的方法在标准跨视角地理定位基准测试中表现出竞争力，同时在未知方向和有限视场（FoVs）的更具挑战性的场景下展现出强大的鲁棒性。
cs.CV / 97 / 2610.13558
What We Talk About When We Talk About Emotion Editing
我们谈论情感编辑时谈论的是什么
Abstract
Emotional image editing aims to change the emotion a scene conveys while preserving its essential content. Because the same scene can support different emotional interpretations, achieving an emotional goal requires visual changes grounded in the source context. Motivated by the Component Process Model, which links emotional responses to event appraisal, we investigate neural representations as a reference for these editing decisions. Using TRIBE v2, an fMRI encoding model, we analyze 400 source-target pairs across eight emotion categories. Hidden states from EmoEdit, EmoEditor, and AffectDelta correspond more strongly to predicted target-image responses in visual regions than in affect-related regions. This imbalance motivates ROI-ReFT, which combines source visual responses as a content reference with a predicted affective response change as an editing direction. Two interacting branches use both conditions to guide a frozen editor throughout generation. Predicted responses of the generated image supervise affective change and constrain visual drift. On a post hoc selected subset of 400 editing requests, ROI-ReFT achieves 91.75% target-emotion accuracy and 83.25% joint success, with the same output both matching the target emotion and reaching source-output DINO-I of at least 0.80. Our code and dataset will be publicly released.
Chinese Translation
情感图像编辑旨在改变场景传达的情感，同时保留其基本内容。由于同一场景可以支持不同的情感解读，实现情感目标需要基于源上下文的视觉变化。受到成分过程模型（Component Process Model）的启发，该模型将情感反应与事件评估联系起来，我们研究神经表征作为这些编辑决策的参考。使用TRIBE v2，这是一种功能性磁共振成像（fMRI）编码模型，我们分析了跨越八个情感类别的400对源-目标图像。来自EmoEdit、EmoEditor和AffectDelta的隐藏状态在视觉区域与预测的目标图像反应之间的对应关系比在情感相关区域更强。这种不平衡促使我们提出ROI-ReFT，它将源视觉反应作为内容参考，并将预测的情感反应变化作为编辑方向。两个相互作用的分支利用这两种条件来指导一个冻结的编辑器进行生成。生成图像的预测反应监督情感变化并限制视觉漂移。在后期选择的400个编辑请求的子集上，ROI-ReFT实现了91.75%的目标情感准确率和83.25%的联合成功率，且输出同时匹配目标情感并达到源输出DINO-I至少为0.80。我们的代码和数据集将公开发布。
cs.CV / 98 / 2610.13604
Thyroid Nodule Segmentation in Ultrasound Images with Conformal Uncertainty Estimation
超声图像中甲状腺结节分割与保形不确定性估计
Abstract
Thyroid nodules affect a large proportion of the general population. Although most are benign, a non-negligible fraction is malignant. Ultrasound imaging is the primary diagnostic modality, and accurate nodule segmentation is a prerequisite for CAD systems to classify malignancy and assess metastatic risk.\\ In this work, we propose a thyroid nodule segmentation framework that integrates a DeepLabV3+, channel-attention modules and post-hoc conformal prediction, providing both accurate mask and pixel-level uncertainty estimates. We address thyroid ultrasound nodule segmentation as the first stage of a pipeline that also includes malignancy classification and metastatic risk prediction. We adopt DeepLabV3+, whose Atrous Spatial Pyramid Pooling (ASPP) captures multi-scale context, with an EfficientNet-B5 backbone for its accuracy-efficiency trade-off. Additionally, we test three channel-attention modules: Squeeze-and-Excitation, CBAM and Grouped Channel Attention (GCA), paired with two loss formulations (Dice-BCE and Boundary).\\ Since binary masks do not indicate prediction confidence, we wrap the trained model with a split conformal prediction layer that produces pixel-level uncertainty maps with statistical coverage guarantees.\\ We evaluate all configurations on the TN3K public benchmark using a fixed split and directly compare them with state-of-the-art methods. The plain backbone already improves upon previously reported results, with CBAM + Dice-BCE emerging as the best overall configuration.
Chinese Translation
甲状腺结节影响着大量普通人群。尽管大多数结节是良性的，但仍有相当一部分是恶性的。超声成像是主要的诊断手段，而准确的结节分割是计算机辅助诊断（CAD）系统分类恶性和评估转移风险的前提。在本研究中，我们提出了一种甲状腺结节分割框架，该框架集成了DeepLabV3+、通道注意模块和后验保形预测，提供准确的掩膜和像素级不确定性估计。我们将甲状腺超声结节分割视为一个管道的第一阶段，该管道还包括恶性分类和转移风险预测。我们采用DeepLabV3+，其空洞空间金字塔池化（Atrous Spatial Pyramid Pooling, ASPP）能够捕捉多尺度上下文，并使用EfficientNet-B5作为骨干网络，以实现准确性与效率的平衡。此外，我们测试了三种通道注意模块：Squeeze-and-Excitation、CBAM和分组通道注意（Grouped Channel Attention, GCA），并配合两种损失公式（Dice-BCE和Boundary）。由于二进制掩膜无法指示预测置信度，我们将训练好的模型与一个分裂保形预测层结合，生成具有统计覆盖保证的像素级不确定性图。我们在TN3K公共基准上评估所有配置，使用固定分割并与最先进的方法进行直接比较。简单的骨干网络已经改善了先前报告的结果，其中CBAM + Dice-BCE被认为是最佳整体配置。
cs.CV / 99 / 2610.13610
CUBE: Class-conditioned Unknown Boundary Evidence for Open-Set Insect Species Recognition
CUBE：用于开放集昆虫物种识别的类别条件未知边界证据
Abstract
Open-set insect species recognition faces significant challenges when deployed in real-world biodiversity scenarios, particularly when encountering unknown species that are visually close to known training classes. Unlike generic out-of-distribution samples, fine-grained unknown insects are often not visually anomalous. They instead tend to appear near the decision boundary of specific known species, leading to unreliable closed-set predictions. To address these challenges, we propose \textbf{Class-conditioned Unknown Boundary Evidence (CUBE)}, a novel evidence-based framework that models unknownness around each known species rather than as a single global unknown class. Our key idea is to evaluate a test sample relative to the known species it most closely resembles, enabling local rejection for visually confusable unknown insects. CUBE learns class-conditioned boundary evidence from auxiliary species, preserves auxiliary species as shadow references during inference, and incorporates taxonomic candidate calibration to verify biological plausibility. When ecological metadata is available, we further introduce CUBE-E, which adds a lightweight location-consistency term without modifying the visual backbone. Experiments on the Open-Insect C-America benchmark show that CUBE achieves superior performance, outperforming state-of-the-art baselines with 95.8\%, 98.7\%, and 99.1\% AUROC on local, non-local, and non-moth unknown splits, respectively, while maintaining 94.9\% closed-set accuracy. With ecological metadata, CUBE-E further improves the results to 96.0\%, 99.7\%, and 99.5\% AUROC with 95.2\% closed-set accuracy.
Chinese Translation
开放集昆虫物种识别在实际生物多样性场景中面临重大挑战，尤其是在遇到与已知训练类别视觉上相近的未知物种时。与一般的分布外样本不同，细粒度的未知昆虫往往并不具有视觉异常性。相反，它们通常出现在特定已知物种的决策边界附近，从而导致不可靠的闭集预测。为了解决这些挑战，我们提出了 extbf{类别条件未知边界证据（CUBE）}，这是一个新颖的基于证据的框架，它围绕每个已知物种建模未知性，而不是将其视为单一的全局未知类别。我们的关键思想是相对于与之最相似的已知物种评估测试样本，从而实现对视觉上易混淆的未知昆虫的局部拒绝。CUBE从辅助物种中学习类别条件边界证据，在推理过程中将辅助物种保留为影子参考，并结合分类候选校准以验证生物学的合理性。当生态元数据可用时，我们进一步引入CUBE-E，它在不修改视觉骨干的情况下增加了轻量级的位置一致性项。在Open-Insect C-America基准上的实验表明，CUBE表现优越，在局部、非局部和非蛾类未知分割上分别达到了95.8\%、98.7\%和99.1\%的AUROC，同时保持了94.9\%的闭集准确率。结合生态元数据，CUBE-E进一步将结果提高到96.0\%、99.7\%和99.5\%的AUROC，闭集准确率为95.2\%。
cs.CV / 100 / 2610.13636
Benchmarking Answer-Grounding Consistency in Remote Sensing Visual Question Answering
遥感视觉问答中答案基础一致性的基准测试
Abstract
In grounded remote sensing visual question answering, a model produces a textual answer and identifies relevant objects in the image as grounding evidence. Existing evaluations measure answer correctness and localization quality, but do not directly assess answer-grounding consistency. We propose the Answer-Evidence Support Protocol (AESP) to assess this consistency through evidence support: whether the grounding evidence supports the model's own answer. AESP applies task-specific rules to existence, counting, and comparison questions. We further construct the Grounded Remote Sensing Visual Question Answering (G-RSVQA) benchmark, containing 15,027 samples with questions, reference answers, and all related object-level annotations. Experiments on seven grounded vision--language models reveal failures of evidence support across all three question types. Up to 50.82% of correct answers in the evidence evaluation subset are not supported by the provided grounding evidence.
Chinese Translation
在基础遥感视觉问答中，模型生成文本答案并识别图像中相关对象作为基础证据。现有评估方法测量答案的正确性和定位质量，但未直接评估答案基础一致性。我们提出了答案证据支持协议（Answer-Evidence Support Protocol, AESP）来通过证据支持评估这种一致性：即基础证据是否支持模型自身的答案。AESP 针对存在性、计数和比较问题应用特定任务规则。我们进一步构建了基础遥感视觉问答（Grounded Remote Sensing Visual Question Answering, G-RSVQA）基准，包含 15,027 个样本，涵盖问题、参考答案及所有相关的对象级注释。在七个基础视觉-语言模型上的实验揭示了所有三种问题类型中证据支持的失败。在证据评估子集中，最高可达 50.82% 的正确答案未得到提供的基础证据的支持。
cs.CV / 101 / 2610.13681
PhysForge: Material-Controllable 3D Reconstruction with Separable Components
PhysForge：可控材料的可分离组件3D重建
Abstract
Recent advances in physical 3D generation have enabled visually faithful assets with simulatable mechanical behavior. However, existing methods provide limited control over material behavior beyond appearance-based inference and struggle to estimate physical properties from complex interactions between objects and semantic parts. Therefore, we present PhysForge, a Video Diffusion Model (VDM)-based framework for addressing these challenges, comprising two complementary modules for material-controllable physical fitting and componentlevel interaction modeling. The first, PhysFit, leverages material-conditioned VDM dynamics as references for physical-property fitting, introducing windowed fitting to focus on material-revealing interactions and Hallucinated Forcing to disentangle nonphysical motion hallucinated by the VDM from material responses. The second, PhysComp decomposes complex interactions into individual objects and semantic components, enabling their physical parameters to be optimized independently and flexibly composed for joint simulation. Experiments on PhysAssets and Google Scanned Objects demonstrate that, compared with existing methods, PhysForge provides stronger control over material-dependent physical responses while supporting component-level physical modeling for complex interactions. It achieves the best CLIP and physical controllability scores on both common and uncommon materials, with qualitative comparisons, component-level ablations, and a user study further supporting these improvements. Our project page is available at https://masterwanna.github.io/PhysForge.html .
Chinese Translation
近年来，物理3D生成的进展使得能够生成视觉上真实的资产，并具备可模拟的机械行为。然而，现有方法在材料行为的控制方面仅限于基于外观的推断，并且在从对象与语义部分之间复杂交互中估计物理属性时面临困难。因此，我们提出了PhysForge，一个基于视频扩散模型（VDM）的框架，以应对这些挑战，该框架包含两个互补模块，用于可控材料的物理拟合和组件级交互建模。第一个模块PhysFit利用材料条件的VDM动态作为物理属性拟合的参考，引入了窗口拟合以关注揭示材料的交互，并采用幻觉强制（Hallucinated Forcing）来将VDM幻觉出的非物理运动与材料响应分离。第二个模块PhysComp将复杂的交互分解为单独的对象和语义组件，使其物理参数能够独立优化并灵活组合以进行联合模拟。在PhysAssets和Google扫描对象上的实验表明，与现有方法相比，PhysForge在材料依赖的物理响应控制方面提供了更强的能力，同时支持复杂交互的组件级物理建模。它在常见和不常见材料上都达到了最佳的CLIP和物理可控性评分，定性比较、组件级消融实验以及用户研究进一步支持了这些改进。我们的项目页面可访问：https://masterwanna.github.io/PhysForge.html 。
cs.CV / 102 / 2610.13687
X-AVDD: Cross-Attentive Audio-Visual Dataset Distillation
X-AVDD：跨注意力音视频数据集蒸馏
Abstract
Audio-visual dataset distillation aims to synthesize a small dataset from a large paired audio-visual dataset, while nearly preserving the training utility for neural networks. Existing approaches such as DM and AVDD face two critical limitations: 1) they are memory-intensive, making them difficult to scale to complex architectures and higher samples-per-class (SPC) settings, and 2) they often fail to preserve strong cross-modal semantic correspondence. To tackle these issues, we propose X-AVDD, a cross-attention-based distillation method that explicitly captures inter-modal relationships during synthesis. By decoupling cross-modal alignment from joint optimization, X-AVDD is substantially more memory-efficient, reducing end-to-end synthesis cost by $\sim 8\times$ GPU-hours, while producing highly aligned synthetic data. Empirically, X-AVDD significantly outperforms other audio-visual distillation methods. On VGGSound, an audio-visual dataset of short YouTube clips, X-AVDD improves top-1 test accuracy from $8.2\%\rightarrow24.0\%$ at SPC=$10$ and $9.8\%\rightarrow26.5\%$ at SPC=$20$. Furthermore, X-AVDD improves cross-architecture generalization (e.g., $25.5\%\rightarrow35.3\%$ for a ViT target on AVE dataset at SPC $=50$) and successfully scales to complex architectures such as AudioCLIP, achieving $62.3\%$ accuracy using only $5.65\%$ of the full dataset, against the full dataset accuracy of $72.4\%$ while completing in $9\times$ lower training time compared to the full dataset.
Chinese Translation
音视频数据集蒸馏旨在从大型配对音视频数据集中合成一个小型数据集，同时几乎保留对神经网络的训练效用。现有方法如 DM 和 AVDD 面临两个关键限制：1）它们占用大量内存，使得难以扩展到复杂架构和更高的每类样本数（SPC）设置；2）它们往往无法保持强烈的跨模态语义对应关系。为了解决这些问题，我们提出了 X-AVDD，一种基于跨注意力的蒸馏方法，明确捕捉合成过程中的模态间关系。通过将跨模态对齐与联合优化解耦，X-AVDD 在内存效率上显著提高，减少了端到端合成成本约 $ ext{8} imes$ GPU 小时，同时生成高度对齐的合成数据。从实证结果来看，X-AVDD 显著优于其他音视频蒸馏方法。在 VGGSound 上，一个短 YouTube 视频剪辑的音视频数据集，X-AVDD 将顶级测试准确率从 $8.2\%
ightarrow 24.0\%$（SPC=$10$）和 $9.8\
ightarrow 26.5\\%$（SPC=$20$）提高。此外，X-AVDD 改善了跨架构泛化（例如，在 SPC $=50$ 的 AVE 数据集上，ViT 目标的准确率从 $25.5\
ightarrow 35.3\\%$），并成功扩展到复杂架构如 AudioCLIP，仅使用 $5.65\\%$ 的完整数据集便达到了 $62.3\\%$ 的准确率，而完整数据集的准确率为 $72.4\\%$，同时训练时间比完整数据集低 $9 imes$。
cs.CV / 103 / 2610.13697
Improving Semantic Coverage in AI Test & Evaluation through Generative Augmentation
通过生成增强改善人工智能测试与评估中的语义覆盖
Abstract
In test and evaluation (T&E) of AI models, evaluators typically have access only to a held-out test dataset and no knowledge of the training data or its semantic composition. If critical input conditions---such as time-of-day, weather, or scene location---are underrepresented in this test dataset, T&E results can be misleadingly due to this semantic coverage gap. We propose a three-stage generative AI pipeline to detect and fill such semantic gaps without requiring ground-truth labels of these semantics. First, vision-language models (VLMs) perform zero-shot binary labeling of semantic factors, and an embedding-space clustering step identifies which semantics most affect downstream model behavior. Second, diffusion-based image-to-image models transform existing covered images into the underrepresented semantic domain, preserving scene content while augmenting high-level appearance. Third, statistical analysis of the augmented image embeddings verifies that the synthetic imagery improves dataset coverage before it is incorporated. Crucially, the synthetic data is used only to enrich the test set and no model retraining is performed for T&E. We validate the full pipeline on both image classification and object detection under day/night semantic gaps, showing that the augmented test datasets better distinguish models with and without exposure to the missing semantic, thereby yielding T&E conclusions that more closely match those of an ideal, fully covered dataset. Code available at https://github.com/Hashe037/DEDUCE
Chinese Translation
在人工智能模型的测试与评估（T&E）中，评估者通常只能访问一个保留的测试数据集，而对训练数据或其语义组成没有任何了解。如果关键输入条件——例如时间、天气或场景位置——在该测试数据集中表现不足，则由于这一语义覆盖差距，T&E结果可能会产生误导。我们提出了一种三阶段的生成性人工智能流程，以检测并填补此类语义空白，而无需这些语义的真实标签。首先，视觉-语言模型（VLMs）对语义因素进行零样本二元标记，嵌入空间聚类步骤识别出哪些语义对下游模型行为影响最大。其次，基于扩散的图像到图像模型将现有的覆盖图像转换为不足代表的语义领域，保留场景内容的同时增强高层次外观。第三，对增强图像嵌入的统计分析验证了合成图像在纳入之前改善了数据集的覆盖。重要的是，合成数据仅用于丰富测试集，并且在T&E过程中不进行模型重训练。我们在日/夜语义差距下对图像分类和目标检测的完整流程进行了验证，显示增强的测试数据集能够更好地区分接触与未接触缺失语义的模型，从而得出更接近理想、完全覆盖数据集的T&E结论。代码可在 https://github.com/Hashe037/DEDUCE 获取。
cs.CV / 104 / 2610.13706
GLIMPSE: Global-Local Indexing with Multi-Precision Spectral Encoding for Explorable Video World Models
GLIMPSE：具有多精度谱编码的全球-局部索引用于可探索的视频世界模型
Abstract
A camera-controlled video world model must keep a scene consistent when the camera returns to it. Memories that render stored features into the target frame can turn depth and pose errors into ghosting or duplicated structure. Positional transport avoids that resampling by relocating a historical patch's address rather than its content, but assigns full positional precision even to inaccurate projections. We present GLIMPSE, a global-local memory interface with band-wise multi-precision coordinates and geometry-conditioned spectral attenuation to reduce sensitivity to inaccurate projected addresses. Local patch memory supplies view-specific appearance, while recurrent global memory provides scene context. Footprint-guided retrieval selects historical patches under a fixed local read budget. Under matched retrieval and read budgets, GLIMPSE improves geometric correspondence and photometric consistency over deterministic positional transport.
Chinese Translation
一个由相机控制的视频世界模型必须在相机返回时保持场景的一致性。将存储特征渲染到目标帧的记忆可能会将深度和姿态误差转化为重影或重复结构。位置传输通过重新定位历史补丁的地址而非其内容来避免这种重采样，但即使在不准确的投影中也会分配完整的位置精度。我们提出了GLIMPSE，一种具有带状多精度坐标和几何条件谱衰减的全球-局部记忆接口，以降低对不准确投影地址的敏感性。局部补丁记忆提供视图特定的外观，而递归全球记忆提供场景上下文。足迹引导检索在固定的局部读取预算下选择历史补丁。在匹配的检索和读取预算下，GLIMPSE在几何对应性和光度一致性方面优于确定性位置传输。
cs.CV / 105 / 2610.13720
State-Isolated Vision Mamba via Update-Tendency Routing for Underwater Image Enhancement
通过更新倾向路由实现状态隔离视觉Mamba的水下图像增强
Abstract
Vision Mamba has recently emerged as an efficient alternative to convolutional and Transformer-based architectures by modeling long-range spatial dependencies through recurrent state evolution. Existing studies primarily improve token serialization, scanning strategies, or expert routing, yet largely overlook a state level question question: how should recurrent states be organized for spatially heterogeneous visual restoration? In this paper, we show that forcing tokens with heterogeneous update behaviors to recurrently modify a shared hidden state leads to persistent cross-content interference, which we term \textbf{State Contention}. We further argue that effective recurrent modeling should organize state evolution according to token compatibility rather than token ordering. Based on this observation, we introduce \textbf{State Isolation}, a new recurrent state organization principle for Vision Mamba,tokens with compatible state-update tendencies evolve within independent recurrent state streams. To instantiate this principle, we propose \textbf{IsoMamba}, a $\Delta$-guided architecture that constructs model-intrinsic update representations, discovers update-compatible token groups through label-free clustering, and routes them to isolated Mamba experts following a \textbf{cluster--route--isolate} paradigm. In addition, we develop a dynamic channel-wise reweighting objective to compensate for wavelength-dependent reconstruction difficulty in underwater images. Extensive experiments demonstrate that IsoMamba consistently achieves state-of-the-art performance on multiple underwater image enhancement benchmarks while providing cleaner hidden state responses, validating the effectiveness of state isolated recurrent modeling.
Chinese Translation
视觉Mamba最近作为卷积和基于Transformer架构的高效替代方案出现，通过递归状态演化建模长距离空间依赖关系。现有研究主要改善令牌序列化、扫描策略或专家路由，但在很大程度上忽视了一个状态层面的问题：如何组织递归状态以实现空间异质视觉恢复？在本文中，我们展示了强迫具有异质更新行为的令牌递归地修改共享隐状态会导致持续的跨内容干扰，我们称之为 extbf{状态争用}。我们进一步认为，有效的递归建模应根据令牌兼容性而非令牌顺序来组织状态演化。基于这一观察，我们引入了 extbf{状态隔离}，这是一种新的视觉Mamba递归状态组织原则，具有兼容状态更新倾向的令牌在独立的递归状态流中演化。为了实现这一原则，我们提出了 extbf{IsoMamba}，一种$ riangle$引导架构，构建模型内在的更新表示，通过无标签聚类发现更新兼容的令牌组，并按照 extbf{聚类--路由--隔离}范式将它们路由到隔离的Mamba专家。此外，我们开发了一种动态通道加权目标，以补偿水下图像中波长依赖的重建难度。大量实验表明，IsoMamba在多个水下图像增强基准测试中始终实现了最先进的性能，同时提供了更清晰的隐状态响应，验证了状态隔离递归建模的有效性。
cs.CV / 106 / 2610.13732
LAPT-Scale: Scaling Embodied Policies with Latent Action Pre-Training on Human Videos
LAPT-Scale：基于人类视频的潜在动作预训练扩展具身策略
Abstract
Scaling generalist robotic policies is limited by the high cost of collecting embodied interaction data. Large-scale egocentric human videos offer a rich alternative, but they lack robot action labels, and existing action annotation methods are often noisy under real-world visual complexity and diverse manipulation dynamics. We introduce LAPT-Scale, a scalable latent action pre-training paradigm with two components: LAPT-Scale-LAM learns continuous latent actions by explaining frame-to-frame visual transitions in human videos and uses them as human-derived supervision, while LAPT-Scale-Policy trains a VLA policy through human-video latent-action pre-training, target-robot embodiment alignment, and task-specific post-training. Unlike prior vector-quantized LAMs, LAPT-Scale-LAM uses a continuous latent space to preserve fine-grained motion and interaction dynamics. We further combine perceptual, semantic, and optical-flow alignment signals to improve transition-level supervision beyond pixel reconstruction. Trained on 23.3K hours of egocentric video, LAPT-Scale-LAM provides implicit action supervision for policy pre-training, yielding consistent scaling gains and improving real-robot success from 63.3% to 85.0% as annotated human video scales from 1K to 5K hours. The learned policy also generalizes to unseen visual conditions.
Chinese Translation
扩展通用机器人策略受到收集具身交互数据高成本的限制。大规模自我中心人类视频提供了一种丰富的替代方案，但它们缺乏机器人动作标签，现有的动作注释方法在现实世界的视觉复杂性和多样化操作动态下往往存在噪声。我们提出了LAPT-Scale，一种可扩展的潜在动作预训练范式，包含两个组成部分：LAPT-Scale-LAM通过解释人类视频中的帧间视觉过渡来学习连续的潜在动作，并将其作为人类派生的监督，而LAPT-Scale-Policy通过人类视频潜在动作预训练、目标机器人具身对齐和任务特定后训练来训练VLA策略。与之前的向量量化LAM不同，LAPT-Scale-LAM使用连续潜在空间来保留细粒度的运动和交互动态。我们进一步结合感知、语义和光流对齐信号，以改善超越像素重建的过渡级监督。在23.3K小时的自我中心视频上训练后，LAPT-Scale-LAM为策略预训练提供隐式动作监督，产生一致的扩展增益，并将真实机器人成功率从63.3%提高到85.0%，当注释的人类视频从1K小时扩展到5K小时时。所学习的策略也能够在未见过的视觉条件下进行泛化。
cs.CV / 107 / 2610.13733
KVeil: First-Place Solution to ECCV 2026 LifeGenIP Challenge on Unlearnable Videos
KVeil：ECCV 2026 LifeGenIP 挑战赛关于不可学习视频的第一名解决方案
Abstract
The rapid development of personalized video generation has raised serious concerns over identity privacy, as a few facial videos of a person now suffice to reproduce their identity in synthetic videos. This report presents the first-place solutions to both phases of the ECCV 2026 LifeGenIP Challenge on Unlearnable Videos, which aims to protect facial videos from unauthorized customization under a limited and imperceptible perturbation budget. For Phase 1, we introduce Reference Capacity Collapse, a pair of complementary losses that degrade the information carried by the reference condition while keeping subsequent video representations dependent on it, together with a Motion-Guided EOT mechanism that improves robustness across unknown inference conditions. Phase 2 further considers tuning-based customization and unseen generative models. To address both threats within a single perturbation, we design a two-layer framework that protects the facial and background regions separately, constructing structure-aligned perturbations from a synthetic target identity in the former and strengthening the white-box disruption of Phase 1 in the latter. Our method achieves nearly perfect protection under white-box reference-based customization, and remains effective under other customization paradigms, various models, and temporal attacks. Both solutions are built upon our recent work LETHE. We hope our solutions can inspire future work on protecting facial videos against unauthorized customization. Our code is available at https://github.com/AntigoneRandy/KVeil.
Chinese Translation
个性化视频生成的快速发展引发了对身份隐私的严重关注，因为现在只需少量某人的面部视频即可在合成视频中重现其身份。本报告展示了ECCV 2026 LifeGenIP 挑战赛关于不可学习视频的两个阶段的第一名解决方案，该挑战旨在在有限且不可察觉的扰动预算下保护面部视频免受未经授权的定制。在第一阶段，我们引入了参考容量崩溃（Reference Capacity Collapse），这是一对互补损失，旨在降低参考条件所携带的信息，同时保持后续视频表示依赖于该条件，并结合了一种运动引导的EOT（End of Transmission）机制，以提高在未知推理条件下的鲁棒性。第二阶段进一步考虑基于调优的定制和未见的生成模型。为了在单一扰动内应对这两种威胁，我们设计了一个两层框架，分别保护面部和背景区域，从前者的合成目标身份构建结构对齐的扰动，并在后者中增强第一阶段的白盒干扰。我们的方法在基于白盒参考的定制下实现了几乎完美的保护，并在其他定制范式、各种模型和时间攻击下仍然有效。这两个解决方案均基于我们最近的工作LETHE。我们希望我们的解决方案能够激励未来在保护面部视频免受未经授权定制方面的研究。我们的代码可在 https://github.com/AntigoneRandy/KVeil 获取。
cs.CV / 108 / 2610.13757
Gaussian Splatting with Reflections Guided by What is Seen
基于可见信息引导的高斯点云与反射
Abstract
While novel view synthesis methods like Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have shown impressive performance on common scenes, they often struggle with capturing view-dependent specular effects, particularly near-field reflections. Existing methods utilize environment maps, which only represent lights from infinity, or environment Gaussians, which rely solely on indirect supervision from reflections, are insufficient for accurately modeling such effects. Here, we propose a framework based on 2D Gaussian Splatting (2DGS) that leverages directly observable environment information in input images. Our approach also integrates a hybrid reflection appearance representation and view-dependent specular modeling. Experiments on public benchmarks and a newly curated near-field reflection dataset demonstrate that our method achieves superior rendering quality and geometric accuracy.
Chinese Translation
尽管像神经辐射场（Neural Radiance Fields, NeRF）和三维高斯点云（3D Gaussian Splatting, 3DGS）等新颖视图合成方法在常见场景中表现出色，但它们在捕捉视角依赖的高光效果，特别是近场反射方面常常面临挑战。现有方法利用环境贴图，这仅能表示来自无限远的光源，或依赖于反射的间接监督的环境高斯，这些方法不足以准确建模此类效果。在此，我们提出了一种基于二维高斯点云（2D Gaussian Splatting, 2DGS）的框架，该框架利用输入图像中直接可观察的环境信息。我们的方法还集成了混合反射外观表示和视角依赖的高光建模。在公共基准测试和新近整理的近场反射数据集上的实验表明，我们的方法在渲染质量和几何准确性方面均表现优越。
cs.CV / 109 / 2610.13763
Neurons++: A Unified Visual-Stream Framework for Image and Video Reconstruction from fMRI with Neural-to-Visual Interpretability
Neurons++：一种统一的视觉流框架，用于从fMRI重建图像和视频，并具备神经到视觉的可解释性
Abstract
Decoding visual experience from fMRI is important for understanding human perception and developing brain-computer interfaces. Recent generative decoding methods can reconstruct plausible images and videos, but visual plausibility alone does not ensure that the output is faithfully grounded in brain activity. In this work, we propose Neurons++, a visual-stream fMRI reconstruction framework for both static image and dynamic video reconstruction. Compared with our conference version Neurons, Neurons++ extends the paradigm from video-only reconstruction to visual-stream reconstruction for both images and videos, and replaces parallel task conditioning with serial cortex-inspired decoding. The model first uses image-fMRI pretraining to obtain a stable spatial-semantic initialization, and then performs subject-specific fine-tuning for image and video reconstruction separately. The serial design organizes fMRI-derived representations into explicit spatial, semantic, language-level, and temporal cues. Extensive experiments show that Neurons++ achieves state-of-the-art video reconstruction and competitive image reconstruction across standard benchmarks. Beyond reconstruction fidelity, neural-to-visual analyses indicate that Neurons++ exhibits qualitative directional controllability under neural perturbation, that its fMRI encoder read-out is sensitive to stimulus-relevant cortical features, and that it supports ROI-guided semantic modulation. Code and model weights are available at: https://github.com/xmed-lab/NeuronsPlusPlus.
Chinese Translation
从fMRI解码视觉体验对于理解人类感知和开发脑机接口至关重要。近期的生成解码方法能够重建出合理的图像和视频，但单凭视觉合理性并不能确保输出忠实于脑活动。在本研究中，我们提出了Neurons++，一种用于静态图像和动态视频重建的视觉流fMRI重建框架。与我们的会议版本Neurons相比，Neurons++将范式从仅视频重建扩展到图像和视频的视觉流重建，并将并行任务条件替换为串行的皮层启发式解码。该模型首先使用图像-fMRI预训练以获得稳定的空间-语义初始化，然后分别对图像和视频重建进行特定于受试者的微调。串行设计将fMRI衍生的表征组织为明确的空间、语义、语言层面和时间线索。大量实验表明，Neurons++在标准基准测试中实现了最先进的视频重建和具有竞争力的图像重建。除了重建的保真度外，神经到视觉的分析表明，Neurons++在神经扰动下表现出定性的方向可控性，其fMRI编码器的读出对与刺激相关的皮层特征敏感，并支持ROI引导的语义调制。代码和模型权重可在以下网址获取：https://github.com/xmed-lab/NeuronsPlusPlus。
cs.CV / 110 / 2610.13766
Shortcut Learning in Geospatial Foundation Model Embeddings for Cross-Disaster Damage Assessment
地理空间基础模型嵌入中的捷径学习用于跨灾害损失评估
Abstract
Geospatial foundation models can support disaster damage assessment, but high accuracy on familiar disasters may hide shortcut learning: prediction from event or scene characteristics rather than evidence that transfers to a new disaster. We study embeddings from TerraMind, Prithvi-EO-2.0, DOFA, and DINOv3 on the multi-disaster datasets xView2 and BRIGHT. Holding out an entire disaster, instead of randomly splitting image tiles, reduces mean balanced accuracy from 0.807 to 0.638 on xView2 and from 0.818 to 0.568 on BRIGHT. Four sets of tests examine what information supports prediction: event and acquisition context, before/after image inputs, building count, and correct image pairing. Pre-event images remain predictive, and a classifier given only the annotated number of buildings reaches balanced accuracy of 0.629 and 0.653. Combining before/after features performs best on average on xView2, but does not improve over pre-only features on BRIGHT. Shuffling image pairs within the same disaster and damage class causes modest losses. Motivated by these findings, we test whether feature transformations learned only from training disasters improve generalization to unseen disasters. These transformations improve performance, with supervised contrastive learning performing best on xView2 and PCA on BRIGHT. Our results highlight the importance of evaluation on unseen domains for assessing model robustness and show that simple feature-level transformations can improve generalization. Code and experiments can be found at https://github.com/alishibli97/gfm-shortcut-learning.
Chinese Translation
地理空间基础模型可以支持灾害损失评估，但在熟悉的灾害上高准确率可能掩盖了捷径学习：即从事件或场景特征进行预测，而不是转移到新灾害的证据。我们研究了TerraMind、Prithvi-EO-2.0、DOFA和DINOv3在多灾害数据集xView2和BRIGHT上的嵌入。将整个灾害保留，而不是随机拆分图像瓦片，导致xView2的平均平衡准确率从0.807降至0.638，BRIGHT的平均平衡准确率从0.818降至0.568。四组测试考察了哪些信息支持预测：事件和获取背景、前后图像输入、建筑物数量和正确的图像配对。前事件图像仍然具有预测性，仅给定标注建筑物数量的分类器达到了0.629和0.653的平衡准确率。在xView2上，结合前后特征的表现平均最佳，但在BRIGHT上并未优于仅前特征。对同一灾害和损害类别内的图像对进行洗牌会造成适度损失。基于这些发现，我们测试了仅从训练灾害中学习的特征变换是否能改善对未见灾害的泛化。这些变换提高了性能，其中监督对比学习在xView2上表现最佳，而主成分分析（PCA）在BRIGHT上表现最佳。我们的结果强调了在未见领域进行评估对于评估模型鲁棒性的重要性，并表明简单的特征级变换可以改善泛化。代码和实验可以在https://github.com/alishibli97/gfm-shortcut-learning找到。
cs.CV / 111 / 2610.13776
READ: Reading the Selective Scan for Faithful Video Attribution
READ：针对可信视频归因的选择性扫描阅读
Abstract
Selective state-space models are an efficient approach to long-sequence video modeling. However, their recurrent computation makes it difficult to determine which video tokens drive a prediction. Existing attribution methods often require additional backward passes or repeated model evaluations, making them substantially more expensive than the forward inference they aim to explain. In this work, we introduce READ, a parameter-free attribution method for selective state-space models that unrolls the recurrence at the position read by the classifier. READ traces the contribution accumulated in the recurrent state back to the video tokens that contribute to the final prediction. By anchoring attribution at the classifier's readout, it produces a single spatiotemporal relevance map from which frame- and region-level relevance can be obtained directly. Despite requiring only one forward pass, READ achieves competitive faithfulness and localisation compared with substantially more expensive attribution methods. Experimental results show that READ avoids the systematic relevance biases of existing methods while recovering meaningful temporal and class-specific structure across datasets and backbones. More broadly, our results show that the recurrence underlying selective state-space models is not only an efficient mechanism for inference, but also a direct source of interpretable evidence for their predictions.
Chinese Translation
选择性状态空间模型是一种高效的长序列视频建模方法。然而，它们的递归计算使得确定哪些视频标记驱动预测变得困难。现有的归因方法通常需要额外的反向传递或重复的模型评估，这使得它们的计算成本显著高于它们所试图解释的前向推理。在本研究中，我们引入了READ，一种针对选择性状态空间模型的无参数归因方法，该方法在分类器读取的位置展开递归。READ将递归状态中积累的贡献追溯到对最终预测有贡献的视频标记。通过将归因锚定在分类器的读取输出上，READ生成一个单一的时空相关性图，从中可以直接获得帧级和区域级的相关性。尽管仅需一次前向传递，READ在可信度和定位方面与显著更昂贵的归因方法相比仍然具有竞争力。实验结果表明，READ避免了现有方法的系统性相关性偏差，同时在不同数据集和骨干网络中恢复了有意义的时间和类别特定结构。更广泛地说，我们的结果表明，选择性状态空间模型背后的递归不仅是推理的高效机制，也是其预测的可解释证据的直接来源。
cs.CV / 112 / 2610.13780
Detection of extremely rare dead cells in fluorescence microscopy images
荧光显微镜图像中极其稀有死亡细胞的检测
Abstract
Dead cell detection in fluorescence microscopy is essential for phenotypic screening and drug efficacy assessment. However, automated identification remains challenging due to signal variability, optical artifacts, and colocalization ambiguities across multiple channels. Markers that define nuclear boundaries do not neatly coincide spatially or temporally with apoptotic markers. Furthermore, marker accumulation kinetics vary across individual cells and apoptosis stages, making universal thresholds unreliable. We present a systematic comparison of several computational approaches for detecting apoptotic cells in nucleus-stained specimens: (1) heuristic methods using intensity thresholds, distributions, correlations, and domain guidance; (2) classification using numerical features (Logistic Regression, Random Forest, XGBoost and LightGBM) and image features (ResNet-18 and EfficientNet-B0); and (3) anomaly detection, also using numerical features (Isolation Forest, one-class SVM and PCA reconstruction) and image features (convolutional autoencoders and Deep SVDD). Using a dataset of 258 images containing 65,527 cells/nuclei across multiple scenes with 243 manually annotated ground truth dead cells, we evaluate each method on metrics such as PR-AUC, Precision, Recall, F1-score, and macro-F1. Our results indicate that classification using image features (deep learning) and numerical features (traditional) achieved the best performance, with ResNet-18 and XGBoost significantly outperforming manually designed heuristics and anomaly detection methods. Our findings demonstrate that such biomedical problems remain highly challenging to solve even with the state-of-the-art computational and learning techniques. We aim to provide an effective decision-support tool to researchers while tackling such problems either fully-automatically or in a human-in-the-loop setting as an aide to domain scientists.
Chinese Translation
在荧光显微镜中检测死亡细胞对于表型筛选和药物疗效评估至关重要。然而，由于信号变异性、光学伪影以及多个通道之间的共定位模糊性，自动识别仍然具有挑战性。定义核边界的标记在空间和时间上与凋亡标记并不完全重合。此外，标记的积累动力学在个体细胞和凋亡阶段之间有所不同，使得通用阈值不可靠。我们对几种计算方法在染核标本中检测凋亡细胞进行了系统比较：（1）使用强度阈值、分布、相关性和领域指导的启发式方法；（2）使用数值特征（逻辑回归、随机森林、XGBoost 和 LightGBM）和图像特征（ResNet-18 和 EfficientNet-B0）的分类；（3）异常检测，同样使用数值特征（孤立森林、单类支持向量机和主成分分析重构）和图像特征（卷积自编码器和深度 SVDD）。使用包含 65,527 个细胞/核的 258 张图像的数据集，其中有 243 个手动标注的真实死亡细胞，我们在 PR-AUC、精确度、召回率、F1 分数和宏 F1 等指标上评估了每种方法。我们的结果表明，使用图像特征（深度学习）和数值特征（传统方法）的分类方法表现最佳，其中 ResNet-18 和 XGBoost 的表现显著优于手动设计的启发式方法和异常检测方法。我们的研究结果表明，即使使用最先进的计算和学习技术，这类生物医学问题仍然极具挑战性。我们旨在为研究人员提供一个有效的决策支持工具，同时在完全自动化或人机协作的环境中帮助领域科学家解决此类问题。
cs.CV / 113 / 2610.13781
SimVTA: Simulation-Grounded Video-Trajectory Alignment for Driving World Action Models
SimVTA：基于仿真的视频轨迹对齐用于驾驶世界动作模型
Abstract
Driving world action models (WAMs) learn from paired trajectories and future observations. Post-training with reinforcement learning allows the policy to explore alternative trajectories and improve planning using driving rewards. However, driving logs contain future videos only for the recorded trajectories, leaving these alternatives without corresponding visual supervision. Thus, we introduce SimVTA, a simulation-grounded post-training framework that combines trajectory optimization with video supervision. First, we sample candidate trajectories for each observation and use a real-world simulator (e.g., World Engine) to render future videos for high-reward candidates in real-world scenes reconstructed with 3D Gaussian Splatting (3DGS). This provides trajectory-matched supervision independent of the model's own predictions. Under a limited rendering budget, we prioritize video supervision for higher-error futures among these rendered candidates. Finally, we optimize trajectory planning with Group Relative Policy Optimization (GRPO) and use a flow-matching loss on the matched videos to update shared representations. At inference time, the model predicts trajectories directly, without simulator calls. On NAVSIM v1, applying SimVTA to a WAM initialization improves PDMS from 86.2 to 91.3 (+5.1). It achieves 88.2 EPDMS on NAVSIM v2 and an average HD-Score of 38.8 in closed-loop evaluation on HUGSIM. Ablations show that updating shared representations with video supervision improves planning, while prioritizing higher-error futures among high-reward trajectories further improves performance.
Chinese Translation
驾驶世界动作模型（WAMs）通过配对轨迹和未来观察进行学习。使用强化学习进行后训练使得策略能够探索替代轨迹，并利用驾驶奖励改善规划。然而，驾驶日志仅包含记录轨迹的未来视频，导致这些替代轨迹缺乏相应的视觉监督。因此，我们提出了SimVTA，一种基于仿真的后训练框架，将轨迹优化与视频监督相结合。首先，我们为每个观察样本生成候选轨迹，并使用真实世界模拟器（如World Engine）在通过3D高斯点云重建的真实场景中为高奖励候选轨迹渲染未来视频。这提供了与模型自身预测无关的轨迹匹配监督。在有限的渲染预算下，我们优先考虑对这些渲染候选中高误差未来的视觉监督。最后，我们使用群体相对策略优化（GRPO）优化轨迹规划，并在匹配视频上使用流匹配损失更新共享表示。在推理时，模型直接预测轨迹，无需调用模拟器。在NAVSIM v1上，将SimVTA应用于WAM初始化将PDMS从86.2提高到91.3（+5.1）。在NAVSIM v2上实现88.2的EPDMS，并在HUGSIM的闭环评估中获得38.8的平均HD-Score。消融实验表明，通过视频监督更新共享表示改善了规划，而在高奖励轨迹中优先考虑高误差未来进一步提升了性能。
cs.CV / 114 / 2610.13790
OccludeBench: Can VLMs Find What's Missing?
OccludeBench：视觉语言模型能否找到缺失的内容？
Abstract
How well can vision-language models (VLMs) reason about what is missing from an image? We study this question with OccludeBench, a manually curated benchmark for recovering fully occluded objects from the visual evidence that remains. OccludeBench distinguishes direct cases, where the hidden object must be recovered from a specific physical cue such as a shadow or reflection, from inferential cases, where it can be inferred from broader scene context. Across five VLMs, accuracy averages 42.7% on direct cases versus 62.7% on inferential cases, exhibiting a 20.0-point gap. Humans attain higher accuracies of 63.4% versus 68.5% for direct and inferential images, respectively, with no significant gap. To locate this failure, we introduce OccludeTrace, a four-stage diagnostic that isolates where the visual reasoning process breaks down. The bottleneck emerges when models must recover the relevant object from the projection within the full image: they succeed on only 50% of failed examples, compared with 91% of examples they answer correctly. Extended reasoning, tool use, and few-shot prompting yield inconsistent improvements and do not reliably eliminate the gap. In contrast, an inpaint-then-confront intervention, which prompts models to first generate a candidate hidden object and then evaluate that hypothesis against the visible physical cue, reduces the gap to a statistically non-significant difference for both models with image-generation capabilities. Together, our results establish a tractable, controlled diagnostic for isolating how VLMs reason about visual content that is missing from an image.
Chinese Translation
视觉语言模型（VLMs）在推理图像中缺失内容的能力如何？我们通过OccludeBench这一手动策划的基准来研究这个问题，该基准旨在从剩余的视觉证据中恢复完全被遮挡的物体。OccludeBench区分了直接案例，即必须从特定的物理线索（如阴影或反射）中恢复隐藏物体的情况，以及推理案例，即可以从更广泛的场景上下文中推断出隐藏物体的情况。在五个VLMs中，直接案例的平均准确率为42.7%，而推理案例的准确率为62.7%，显示出20.0个百分点的差距。人类在直接和推理图像中的准确率分别为63.4%和68.5%，且没有显著差距。为了定位这一失败，我们引入了OccludeTrace，这是一种四阶段诊断方法，用于隔离视觉推理过程中的故障点。当模型必须从完整图像中的投影中恢复相关物体时，瓶颈就出现了：它们在仅有50%的失败示例中成功，而在91%的正确答案示例中成功。扩展推理、工具使用和少量提示带来了不一致的改善，并未可靠地消除差距。相反，一种先生成候选隐藏物体再根据可见物理线索评估该假设的“先填充再对比”干预，能够将差距缩小到对具有图像生成能力的模型而言统计上不显著的差异。总的来说，我们的结果建立了一种可操作的、受控的诊断方法，以隔离VLMs如何推理图像中缺失的视觉内容。
cs.CV / 115 / 2610.13832
SpatialASM: Internalizing Active Mentalization for Spatial Reasoning
SpatialASM：内化主动心理化以进行空间推理
Abstract
When solving complex spatial reasoning tasks, ordinary humans naturally revisit the scenes from task-relevant viewpoints and mentally reconstruct the spatial layouts from those viewpoints. But can Large Vision-Language Models (LVLMs) proactively think in the same way? We term this ability Active Spatial Mentalization (ASM), an intrinsic component of visual intelligence that remains substantially underdeveloped in current LVLMs. To bridge this gap, we introduce SpatialASM, a scalable framework to internalize ASM into LVLMs through training-time spatial simulation. SpatialASM employs an allocentric novel-view generator to ground hypothetical viewpoints in visual evidence and produces two complementary forms of supervision: i) explicit ASM trajectories that teach LVLMs when and how to adaptively invoke view planning and imagination during spatial reasoning, and ii) automatically-constructed perspective-taking tasks that further strengthens viewpoint-conditioned scene imagination. On seven benchmarks spanning relatively perception-oriented spatial understanding to more challenging tasks requiring multi-image, cross-view and viewpoint-transformation reasoning, SpatialASM yields performance gain on LVLMs of four LVLM backbones in spatial reasoning while retaining their original general vision capabilities. The experimental results demonstrate that our approach improves average accuracy across the seven spatial benchmarks. And the synthesis of ASM trajectories provides a scalable pathway to alleviate the scarcity of spatial Chain-of-Thought (CoT) supervision for eliciting active viewpoint exploration and mental scene simulation.
Chinese Translation
在解决复杂的空间推理任务时，普通人类自然会从与任务相关的视角重新审视场景，并从这些视角中心理重建空间布局。但大型视觉语言模型（LVLMs）能否以相同的方式主动思考？我们将这种能力称为主动空间心理化（Active Spatial Mentalization，ASM），它是视觉智能的一个内在组成部分，但在当前的LVLMs中仍然相对欠发达。为了弥补这一差距，我们提出了SpatialASM，这是一个可扩展的框架，通过训练时的空间模拟将ASM内化到LVLMs中。SpatialASM采用了一种以分配中心视角的新视图生成器，将假设视角与视觉证据相结合，并产生两种互补的监督形式：i) 明确的ASM轨迹，教导LVLMs在空间推理过程中何时以及如何自适应地调用视图规划和想象；ii) 自动构建的视角转换任务，进一步增强基于视角的场景想象。在七个基准测试中，从相对以感知为导向的空间理解到更具挑战性的任务，涉及多图像、跨视角和视角转换推理，SpatialASM在四个LVLM骨干网络的空间推理中提高了性能，同时保留了它们原有的通用视觉能力。实验结果表明，我们的方法提高了七个空间基准的平均准确率。而ASM轨迹的综合提供了一条可扩展的途径，以缓解激发主动视角探索和心理场景模拟所需的空间链式思维（Chain-of-Thought，CoT）监督的稀缺性。
cs.CV / 116 / 2610.13856
Event-based Action Recognition by Pre-Training on Multi-Frame Representations
基于事件的动作识别通过多帧表示的预训练
Abstract
Action recognition is a crucial task in domains like robotics, patient monitoring, or surveillance. Recently, event-based action recognition has attracted much attention due to the event camera's ability to capture fast motions and its improved privacy over conventional vision cameras. While frame-based event representations work well for short event sequences, their discriminative ability declines sharply with longer and more complex event sequences. In this work, we introduce a multi-frame representation that greatly improves recognition performance over its single-frame counterpart. We adopt a masked auto-encoding framework to pre-train a Vision Transformer using a sparsity-aware masking algorithm and token selection mechanism to improve downstream performance and control computational complexity. We also introduce multiple mask tokens for the decoder with a position-aware attention mechanism, leading to further improved performance. Our proposed method shows impressive performance for challenging long-duration and complex event sequences, yielding state-of-the-art performance on THU-EACT-50 CHL and DailyDVS-200.
Chinese Translation
动作识别是机器人技术、患者监测和监控等领域中的一项关键任务。近年来，基于事件的动作识别因事件相机能够捕捉快速运动以及相较于传统视觉相机更好的隐私保护而受到广泛关注。虽然基于帧的事件表示在短事件序列中表现良好，但随着事件序列变长和变得更加复杂，其区分能力急剧下降。在本研究中，我们引入了一种多帧表示，显著提高了识别性能，相较于单帧表示。我们采用了一种掩码自编码框架，通过稀疏感知掩码算法和标记选择机制对视觉变换器（Vision Transformer）进行预训练，以提升下游性能并控制计算复杂度。我们还为解码器引入了多个掩码标记，并结合位置感知的注意机制，从而进一步提高性能。我们提出的方法在处理具有挑战性的长时长和复杂事件序列时表现出色，在 THU-EACT-50 CHL 和 DailyDVS-200 数据集上实现了最先进的性能。
cs.CV / 117 / 2610.13857
Revisiting the Task of Building Damage Assessment by Satellite Imaging
重新审视通过卫星影像进行建筑损伤评估的任务
Abstract
Natural disasters such as earthquakes, tropical storms, and floods cause extensive damage to buildings, and rapid, reliable damage maps are essential to guide emergency response. Automating this task, referred to as building damage assessment (BDA), from satellite imagery, has become an active research topic at the intersection of remote sensing, computer vision, and machine learning, particularly since supervised deep learning methods were shown to outperform earlier handcrafted approaches. Yet, despite a decade of intense activity and the release of several large-scale datasets, the field lacks a shared foundation. This chapter revisits the task of BDA from first principles, organized around three pillars: data, evaluation, and methods. We first clarify the candidate task for- mulations and argue for a sensor-agnostic, instance-based definition that tolerates missing pre-event imagery. We then review existing datasets along the axes of sensing modality, temporal configuration, and geographic coverage. On the evaluation side, we show that prevailing practices, largely based on a single aggregate score, are theoretically fragile, and we lay the foundations of a more informative, ranking-based, multi-score protocol. We further review methods ranging from task-specific deep networks to recent Earth observation foundation models, noticing that generalization to unseen events and regions remains an open difficulty. Finally, we discuss how these elements can inform the organization of a BDA challenge, including cross-validation, challenge, and private data-sharing setups designed to build trust between developers and users.
Chinese Translation
自然灾害如地震、热带风暴和洪水对建筑物造成了广泛的损害，快速可靠的损伤地图对于指导应急响应至关重要。将这一任务（称为建筑损伤评估，Building Damage Assessment，BDA）从卫星影像中自动化，已成为遥感、计算机视觉和机器学习交叉领域的一个活跃研究主题，特别是在监督深度学习方法被证明优于早期手工制作的方法之后。然而，尽管经历了十年的密集研究和多个大规模数据集的发布，该领域仍缺乏一个共享的基础。本章从基本原理重新审视BDA任务，围绕数据、评估和方法三个支柱进行组织。我们首先澄清候选任务的定义，并主张采用一种与传感器无关的、基于实例的定义，以容忍缺失的事件前影像。接着，我们沿着传感模式、时间配置和地理覆盖的轴线回顾现有数据集。在评估方面，我们展示了当前的实践（主要基于单一的综合评分）在理论上是脆弱的，并为更具信息量的基于排名的多评分协议奠定了基础。我们进一步回顾了从任务特定的深度网络到最近的地球观测基础模型的方法，注意到对未见事件和区域的泛化仍然是一个未解决的难题。最后，我们讨论了这些元素如何为BDA挑战的组织提供信息，包括交叉验证、挑战和旨在建立开发者与用户之间信任的私有数据共享设置。
cs.CV / 118 / 2610.13858
LapMerge: Differentiable Exact-Budget Token Merging in Linear Time
LapMerge：线性时间内可微分的精确预算令牌合并
Abstract
Modern vision and language models process ever longer token sequences, and reducing them is the main lever on downstream cost. Token compression, however, faces a fundamental trade-off: selecting tokens discards information, while existing merging methods preserve it only through pairwise token-to-token matching or $MN$ token-to-output interactions. We introduce LapMerge, a differentiable merging operator that resolves this trade-off for ordered sequences: a learned mass distribution partitions the sequence into exactly $N$ equal-mass segments under a Laplace-kernel density, and prefix/suffix scans aggregate each segment, so both the forward and the backward pass run in $\mathcal{O}(M+N)$ time. This combination - an exact output budget, end-to-end differentiability, and linear cost - is precisely what previous reducers trade off against one another. The budget can be changed without changing operator parameters, and the learned mass doubles as a token-level map of where the budget is allocated. Across vision and vision-language tasks, LapMerge matches or improves on strong token reducers while being up to two orders of magnitude faster at large sequence lengths; it is particularly effective when token order reflects spatial or temporal locality, and works both as a drop-in reduction module and as a trainable architectural component.
Chinese Translation
现代视觉和语言模型处理越来越长的令牌序列，而减少这些序列是降低下游成本的主要手段。然而，令牌压缩面临着一个基本的权衡：选择令牌会丢失信息，而现有的合并方法仅通过成对的令牌匹配或 $MN$ 令牌与输出的交互来保留信息。我们提出了 LapMerge，这是一种可微分的合并算子，解决了有序序列的这一权衡：一个学习的质量分布在拉普拉斯核密度下将序列划分为恰好 $N$ 个等质量段，前缀/后缀扫描聚合每个段，因此前向和后向传递均在 $ ext{O}(M+N)$ 时间内运行。这种组合——精确的输出预算、端到端的可微性和线性成本——正是之前的压缩方法所相互权衡的。预算可以在不改变算子参数的情况下进行调整，而学习到的质量也可以作为预算分配的令牌级映射。在视觉和视觉-语言任务中，LapMerge 的性能与强大的令牌压缩器相当或更好，同时在处理长序列时速度快达两个数量级；当令牌顺序反映空间或时间局部性时，它特别有效，并且既可以作为一个直接的压缩模块，也可以作为一个可训练的架构组件。
cs.CV / 119 / 2610.13860
Make It Flow: Animating Water in 3D Gaussian Splatting Scenes
流动起来：在 3D 高斯点云场景中动画水流
Abstract
Explicit 3D scene representations such as 3D Gaussian Splatting (3DGS) have recently enabled high-quality, real-time novel view synthesis. However, these methods assume static scenes, limiting realism in the presence of dynamic phenomena ubiquitous in real-world environments. While coherent motion (e.g., humans or vehicles) can be captured and reconstructed, high-frequency effects such as flowing water remain intractable under casual capture conditions, where dense multi-view, high-speed acquisition is not feasible. We present a method to animate water in static 3DGS scenes, enabling visually plausible motion without requiring dynamic capture or physical simulation. Our approach estimates 2D optical flow from reference images or model renderings and lifts it into a consistent 3D Eulerian motion field. We propose a cinemagraph-inspired algorithm that applies this motion field to deform Gaussian primitives, producing temporally seamless looping animations. The resulting representation retains real-time rendering performance, significantly enhances perceptual realism, and preserves free-viewpoint navigation within the scene. We evaluate our method on a newly captured dataset of in-the-wild waterfall scenes and demonstrate clear improvements over prior work in dynamic Gaussian splatting and 3D cinemagraph generation in visual quality and computational efficiency.
Chinese Translation
显式的 3D 场景表示方法，如 3D 高斯点云 (3DGS)，最近使高质量、实时的新视图合成成为可能。然而，这些方法假设场景是静态的，这在现实环境中存在动态现象时限制了真实感。虽然可以捕捉和重建一致的运动（例如，人类或车辆），但在随意捕捉条件下，高频效应如流动的水仍然难以处理，因为在这种情况下密集的多视角、高速采集并不可行。我们提出了一种在静态 3DGS 场景中动画水流的方法，使得在不需要动态捕捉或物理模拟的情况下实现视觉上合理的运动。我们的方法从参考图像或模型渲染中估计 2D 光流，并将其提升为一致的 3D 欧拉运动场。我们提出了一种受动态影像启发的算法，将该运动场应用于变形高斯原始体，生成时间上无缝的循环动画。最终的表示保留了实时渲染性能，显著增强了感知真实感，并保持了在场景中的自由视角导航。我们在新捕获的野外瀑布场景数据集上评估了我们的方法，并在动态高斯点云和 3D 动态影像生成方面在视觉质量和计算效率上展示了明显的改进。
cs.CV / 120 / 2610.13872
Tracking the Invisible: Self-Supervised Wheel Track Detection with Thermal-RGB Fusion
追踪隐形轨迹：基于热成像与RGB融合的自监督轮轨检测
Abstract
Accurate detection of drivable area is essential for autonomous driving and advanced driver assistance systems. While state-of-the-art methods focus on general traversability estimation, they prove inadequate for snow-covered roads where following existing wheel tracks offers crucial advantages: improved traction, predictable vehicle dynamics, and verified path stability. To address this, we propose a low-cost, self-supervised approach detecting safety-critical wheel tracks using synchronized RGB-I (RGB + Long-Wave Infrared) imagery. Our method automatically generates training labels by projecting future wheel trajectories from recorded GNSS data. A neural network is designed to identify wheel tracks with the 4-channel RGB-I images from challenging, white-out road surfaces. Experimental results demonstrate the effectiveness of our approach in visually monotonous and featureless adverse environment, offering a robust and cost-effective alternative for winter autonomy. The newly introduced ThermalTrack dataset is available at: https://digitalcommons.mtu.edu/thermaltrack/.
Chinese Translation
准确检测可行驶区域对于自动驾驶和高级驾驶辅助系统至关重要。尽管最先进的方法专注于一般的可通行性评估，但它们在雪覆盖道路上的表现却显得不足，因为在这种情况下，遵循现有的轮轨提供了重要的优势：改善的牵引力、可预测的车辆动态以及经过验证的路径稳定性。为了解决这一问题，我们提出了一种低成本的自监督方法，通过同步的RGB-I（RGB + 长波红外）图像检测安全关键的轮轨。我们的方法通过从记录的GNSS数据中投影未来的轮轨迹自动生成训练标签。设计了一种神经网络，以识别来自具有挑战性的白色道路表面的4通道RGB-I图像中的轮轨。实验结果表明，我们的方法在视觉单调和特征缺乏的不利环境中有效，提供了一种稳健且具有成本效益的冬季自主替代方案。新引入的ThermalTrack数据集可在以下网址获取：https://digitalcommons.mtu.edu/thermaltrack/
cs.CV / 121 / 2610.13891
Beyond the Snowflakes: Disentangling Weather Effects from Confounding Meta-Factors in LiDAR Object Detection
超越雪花：剔除激光雷达目标检测中的天气影响与混杂元因素
Abstract
Light Detection and Ranging (LiDAR) is critical for autonomous perception, yet its performance is known to degrade in adverse weather. While existing literature models snow as atmospheric noise that degrades performance, our granular evaluation on the WADS-3D and CADC datasets reveals a counter-intuitive phenomenon: detection accuracy consistently improves as snow intensity increases. Through ablation study and meta-information analysis, we reconcile this by confirming that while falling snow acts as noise, its negative impact is outweighed by a dominant, concurrent factor: significantly increased object point count. We hypothesize that this object point count boost acts as a geometric enhancement, driven by snow accumulation increasing vehicle surface area and a shift from specular to diffuse reflectance. Our findings demonstrate that evaluating perception systems using broad weather-severity bins without accounting for correlated object-level data can lead to misleading conclusions about detector robustness. We conclude that accurate evaluation requires decoupling atmospheric noise from surface accumulation effects and outline future research to validate these physical models. The WADS-3D dataset is available at: https://digitalcommons.mtu.edu/wads-3d/.
Chinese Translation
光探测与测距（LiDAR）对自主感知至关重要，但其性能在恶劣天气条件下会下降。现有文献将雪视为降低性能的气象噪声，而我们对WADS-3D和CADC数据集的细致评估揭示了一个反直觉的现象：随着雪强度的增加，检测准确率持续提高。通过消融研究和元信息分析，我们确认，尽管降雪作为噪声存在，其负面影响被一个主导的、同时发生的因素所抵消：显著增加的目标点数量。我们假设，这一目标点数量的提升作为几何增强，是由于雪的积累增加了车辆的表面积，并导致从镜面反射转向漫反射。我们的研究结果表明，在评估感知系统时，如果不考虑相关的目标级数据，仅使用广泛的天气严重性区间，可能会导致对检测器鲁棒性的误导性结论。我们得出结论，准确的评估需要将大气噪声与表面积累效应解耦，并概述了未来研究以验证这些物理模型的方向。WADS-3D数据集可在以下链接获取：https://digitalcommons.mtu.edu/wads-3d/
cs.CV / 122 / 2610.13915
Where to Look Matters: Rethinking Sub-Volume Sampling in 3D Medical Self-supervised Learning
观察位置的重要性：重新思考3D医学自监督学习中的子体积采样
Abstract
Self-supervised learning on 3D medical images commonly relies on random sub-volume sampling during pretraining. However, random cropping implicitly treats all spatial regions as equally worth observing, despite anatomical priors and spatial redundancy that make regions inherently unequal in their value for representation learning. To this end, we first show through controlled crop-quality manipulations that better sub-volumes lead to better downstream transfer, revealing the observation policy as an overlooked bottleneck in 3D medical SSL. Motivated by this finding, we revisit sub-volume sampling as a \emph{where-to-look} problem. We propose \textbf{\method}, a plug-and-play framework that replaces random cropping with an offline coarse-to-fine probe bank guided by \textit{anatomy-awareness}, \textit{informativeness}, and \textit{diversity}. We instantiate \method as a lightweight \lite variant using geometric and image-statistical cues, and a diagnostic \feat variant using frozen visual features, to examine whether lightweight volume-intrinsic priors suffice or richer semantic cues reliably provide additional value. Extensive experiments across multiple 3D medical SSL backbones and downstream tasks demonstrate that \method-\lite consistently improves transfer performance over random sampling. Mechanistic analyses further attribute these gains to more anatomically plausible and informative observations, reduced spatial redundancy, more efficient learning dynamics, and more structured representations.
Chinese Translation
在3D医学图像上进行自监督学习通常依赖于预训练期间的随机子体积采样。然而，随机裁剪隐含地将所有空间区域视为同等值得观察，尽管解剖先验和空间冗余使得这些区域在表示学习的价值上本质上是不平等的。为此，我们首先通过控制裁剪质量的操作展示了更好的子体积能够带来更好的下游迁移，揭示了观察策略作为3D医学自监督学习中的一个被忽视的瓶颈。受到这一发现的启发，我们将子体积采样重新审视为一个“观察位置”问题。我们提出了 extbf{ extit{method}}，一个即插即用的框架，用以用一个离线的粗到细探测库替代随机裁剪，该探测库由 extit{解剖意识}、 extit{信息量}和 extit{多样性}引导。我们将 extit{method}实例化为一个轻量级的 extit{lite}变体，利用几何和图像统计线索，以及一个诊断性的 extit{feat}变体，使用冻结的视觉特征，以检验轻量级体积内在先验是否足够，或更丰富的语义线索是否可靠地提供额外价值。针对多个3D医学自监督学习骨干网络和下游任务的广泛实验表明， extit{method-lite}在迁移性能上始终优于随机采样。机制分析进一步将这些提升归因于更具解剖学合理性和信息量的观察、减少的空间冗余、更高效的学习动态和更结构化的表示。
cs.CV / 123 / 2610.13925
Can machines reason non-verbally?
机器能够进行非语言推理吗？
Abstract
Modern machines can solve equations, follow logic and reason through language, yet they still struggle with problems that require visual non-verbal reasoning. This is evident even in advanced systems such as DeepSeek, ChatGPT and Gemini, which remain limited on visual tasks such as completing abstract patterns or delineating ambiguous regions-of-interest. Unlike explicit intermediate steps in verbal reasoning which can be verified through logic or automated tests, visual reasoning involves implicit intermediate steps with no clear methods for verification. Non-verbal reasoning is therefore challenging to model and, as a result, remains a fundamental limitation of current machine intelligence. Here we introduce GenVerOpt, a general framework that unifies generation, verification and optimisation into a reasoning pipeline for machines. By generating candidate solutions, verifying their correctness, and iteratively refining them, GenVerOpt enables per-sample optimisation rather than relying solely on dataset-level trends captured during model training. This process mirrors human problem-solving where multiple possibilities are imagined, evaluated, and refined until a correct solution emerges. Across diverse domains, GenVerOpt demonstrates capabilities beyond existing systems: it surpasses humans on abstract IQ puzzles, matches clinicians in prostate cancer segmentation, exceeds experts in liver vessel reconstruction and outperforms state-of-the-art approaches in tumour boundary alignment. These results show that machine reasoning can extend beyond language, establishing visual non-verbal reasoning as a foundation for more flexible and generalisable intelligence.
Chinese Translation
现代机器可以解方程、遵循逻辑并通过语言进行推理，但它们在需要视觉非语言推理的问题上仍然面临挑战。这一点在诸如 DeepSeek、ChatGPT 和 Gemini 等先进系统中尤为明显，这些系统在完成抽象模式或划定模糊兴趣区域等视觉任务上仍然有限。与可以通过逻辑或自动化测试验证的语言推理中的明确中间步骤不同，视觉推理涉及隐含的中间步骤，缺乏明确的验证方法。因此，非语言推理的建模具有挑战性，成为当前机器智能的一个基本限制。在此，我们介绍 GenVerOpt，这是一个将生成、验证和优化统一为机器推理管道的通用框架。通过生成候选解决方案、验证其正确性并进行迭代优化，GenVerOpt 实现了逐样本优化，而不是仅依赖于模型训练期间捕获的数据集级趋势。这个过程类似于人类的问题解决，其中想象、评估和优化多个可能性，直到出现正确的解决方案。在多个领域中，GenVerOpt 展现出超越现有系统的能力：在抽象智商谜题上超越人类，在前列腺癌分割上与临床医生相匹配，在肝脏血管重建中超过专家，并在肿瘤边界对齐方面优于最先进的方法。这些结果表明，机器推理可以超越语言，将视觉非语言推理确立为更灵活和可推广智能的基础。
cs.CV / 124 / 2610.13935
Hide and Seek: HIDE Concepts to SEEK explanations for black box models
藏与寻：HIDE概念以寻求黑箱模型的解释
Abstract
Interpreting machine learning model predictions is a central challenge, particularly for black-box vision models. Concept-based attribution methods interpret models using human-understandable concepts. Existing concept-based methods can only interpret classification and regression models. They require access to model weights or large concept-annotated datasets, and most methods cannot provide image-level attribution. We propose Hide and Seek (HnS), a new method for concept-level attribution that does not require access to model weights or labeled concept datasets. HnS can interpret a wide range of vision models. Given a concept and image set, HnS retrieves concept-aligned images using a vision-language model, localizes the concept within each image, masks the localized region, and measures attribution from the change in the black-box model's output between the original and masked images. This enables faithful attribution by directly measuring the model's behavior and supporting local attributions. We validated HnS on CelebA and COCO-Stuff. Compared to existing methods, HnS showed better agreement with ground-truth attributions. We demonstrated the applicability of HnS on diverse tasks, including image captioning and satellite image segmentation, where no existing concept attribution method can be used. We provided our anonymous code in the supplementary material.
Chinese Translation
解释机器学习模型的预测是一个核心挑战，尤其对于黑箱视觉模型而言。基于概念的归因方法使用人类可理解的概念来解释模型。现有的基于概念的方法仅能解释分类和回归模型。它们需要访问模型权重或大型概念标注数据集，并且大多数方法无法提供图像级别的归因。我们提出了Hide and Seek（HnS），一种新的概念级别归因方法，不需要访问模型权重或标注的概念数据集。HnS能够解释广泛的视觉模型。给定一个概念和图像集，HnS使用视觉-语言模型检索与概念对齐的图像，在每个图像中定位该概念，掩蔽定位区域，并通过比较原始图像和掩蔽图像之间黑箱模型输出的变化来测量归因。这种方法通过直接测量模型的行为并支持局部归因，实现了真实的归因。我们在CelebA和COCO-Stuff上验证了HnS。与现有方法相比，HnS在与真实归因的一致性上表现更佳。我们展示了HnS在多种任务中的适用性，包括图像描述和卫星图像分割，而现有的概念归因方法无法应用。我们在补充材料中提供了匿名代码。
cs.CV / 125 / 2610.13941
Images Talk Back: Guiding MMDiTs by Suppressing Image-to-Text Feedback
图像回应：通过抑制图像到文本反馈引导多模态扩散变换器
Abstract
Multimodal diffusion transformers (MMDiTs) allow the evolving image to update internal text representations through joint attention. These representations then guide subsequent image updates, raising a question: does this feedback help correct deviations from the prompt, or preserve them? We find that image-to-text (I2T) feedback can preserve the objects and attributes already represented in the evolving image, including errors. Transplanting I2T messages can rescue failures or disrupt successes. At matched intervention strengths, suppressing the influence of I2T feedback improves generation, whereas opposite controls degrade it. These findings motivate Semantic Feedback Guidance (SFG), which suppresses the influence of I2T feedback by guiding away from an auxiliary branch that strengthens this feedback and weakens text-to-image conditioning. The in-place variant, iSFG, directly suppresses I2T feedback and strengthens text-to-image conditioning, requiring no extra model evaluations. Temporal interventions motivate running the SFG branch during the first 20% of image sampling steps, requiring about $1.2\times$ the base NFE without classifier-free guidance (CFG). Experiments across three image-generation backbones show gains in alignment and preference-model scores. On FLUX, adding SFG to CFG raises GenEval from $0.6661$ to $0.7479$; on SD3.5M, iSFG adds $+1.33$ HPSv2.1 over CFG at zero extra NFE. The control also transfers to text-to-video and image-to-video generation, improving VBench scores. Code is available at https://github.com/yyc1215lll-creator/SFG.
Chinese Translation
多模态扩散变换器（MMDiTs）允许不断演变的图像通过联合注意力更新内部文本表示。这些表示随后引导后续的图像更新，提出了一个问题：这种反馈是帮助纠正与提示的偏差，还是保留这些偏差？我们发现，图像到文本（I2T）反馈可以保留已经在演变图像中表示的对象和属性，包括错误。移植 I2T 消息可以挽救失败或破坏成功。在相同的干预强度下，抑制 I2T 反馈的影响可以改善生成，而相反的控制则会降低生成质量。这些发现促使了语义反馈引导（Semantic Feedback Guidance, SFG）的提出，该方法通过引导远离一个增强此反馈并削弱文本到图像条件的辅助分支来抑制 I2T 反馈的影响。就地变体 iSFG 直接抑制 I2T 反馈并增强文本到图像的条件，无需额外的模型评估。时间干预促使在图像采样步骤的前 20% 运行 SFG 分支，所需的计算量约为基础 NFE 的 $1.2 imes$，而无需分类器自由引导（CFG）。在三个图像生成基础模型上的实验显示了对齐和偏好模型评分的提升。在 FLUX 上，将 SFG 添加到 CFG 使 GenEval 从 $0.6661$ 提升至 $0.7479$；在 SD3.5M 上，iSFG 在零额外 NFE 的情况下增加了 $+1.33$ HPSv2.1。该控制也可转移到文本到视频和图像到视频生成，提升 VBench 评分。代码可在 https://github.com/yyc1215lll-creator/SFG 获取。
cs.CV / 126 / 2610.13944
InstantFusion: A Shared Latent Interface for Heterogeneous Image Generators
InstantFusion：异构图像生成器的共享潜在接口
Abstract
Diffusion/flow models have emerged as the de facto paradigms for image generation, giving rise to a diverse family of modern generators with increasingly heterogeneous architectures and representations. Despite their remarkable generation capabilities, the cross-model alignment of latent representations and the consistency of their underlying denoising dynamics remain underexplored. In this work, we uncover an intriguing cross-model compatibility: independently trained generators exhibit similar latent spatial organization, while directly handing off intermediate states between unaligned models can still preserve coarse visual content. Motivated by these observations, we propose InstantFusion, a lightweight shared latent interface that connects the denoising trajectories of heterogeneous image generators. Specifically, to enable flexible cross-model state translation, model-specific Latent AutoEncoders (LAEs) are introduced to project intermediate states from different generators into a shared latent space and decode them into the latent space of a target generator. This shared space is constrained by two complementary objectives: aligning intermediate states derived from the same image at matched noise levels and synchronizing their corresponding local denoising velocities. Once trained, InstantFusion allows denoising to seamlessly transition across frozen generators without requiring a dedicated mapping for every model pair. The versatility of InstantFusion is demonstrated across acceleration, multi-preference composition, cross-architecture on-policy distillation, and cross-model composition, achieving competitive performance across these diverse settings.
Chinese Translation
扩散/流模型已成为图像生成的事实标准，催生了一系列具有日益异构架构和表示的现代生成器。尽管它们在生成能力上表现出色，但潜在表示的跨模型对齐及其基础去噪动态的一致性仍然未被充分探索。在本研究中，我们揭示了一种有趣的跨模型兼容性：独立训练的生成器表现出相似的潜在空间组织，而在未对齐模型之间直接传递中间状态仍然可以保留粗略的视觉内容。基于这些观察，我们提出了InstantFusion，一种轻量级的共享潜在接口，连接异构图像生成器的去噪轨迹。具体而言，为了实现灵活的跨模型状态转换，引入了模型特定的潜在自编码器（Latent AutoEncoders, LAEs），将来自不同生成器的中间状态投影到共享潜在空间，并将其解码到目标生成器的潜在空间。该共享空间受到两个互补目标的约束：在匹配噪声水平下对齐来自同一图像的中间状态，以及同步其对应的局部去噪速度。一旦训练完成，InstantFusion允许去噪在冻结的生成器之间无缝过渡，而无需为每对模型提供专门的映射。InstantFusion的多功能性在加速、多偏好组合、跨架构在线蒸馏和跨模型组合等多种设置中得到了验证，并在这些多样化的环境中实现了竞争力的性能。
cs.CV / 127 / 2610.13947
Towards Physically Grounded Articulated 3D Assets: A Multi-Agent Pipeline and Benchmark
朝向物理基础的关节3D资产：一个多智能体管道和基准测试
Abstract
In this paper, we study physically grounded articulated 3D asset generation---moving the goal from reproducing how objects look to capturing how they move and function. Inspired by the literature on procedural and agentic generation, we present Artisan, a system that extends the latest method Articraft by introducing a multi-agent pipeline with physical simulation feedback, enabling comprehensive attributes beyond geometry, appearance, and physically plausible articulation, such as collision-free motion, stability under gravity, and accurate mass and material properties. Furthermore, we construct a unified benchmark spanning four key aspects: spatial fidelity, physical plausibility, downstream task performance, and overall quality. The benchmark encompasses 100 reference assets from ArtVIP and LightWheel, 50 stability prompts, and 50 ABO objects with reference masses, with protocols evaluating various metrics. Through extensive evaluation of more than 10 representative models, Artisan achieves competitive spatial fidelity while attaining the lowest collision rates and highest stability rates among the evaluated methods. Together, our system and benchmark establish insights for advancing 3D generation towards physically plausible, functionally usable assets.
Chinese Translation
在本文中，我们研究物理基础的关节3D资产生成——将目标从再现物体的外观转向捕捉物体的运动和功能。受到程序生成和智能生成文献的启发，我们提出了Artisan，一个通过引入具有物理仿真反馈的多智能体管道来扩展最新方法Articraft的系统，使其能够提供超越几何、外观和物理上合理的关节的全面属性，例如无碰撞运动、重力下的稳定性，以及准确的质量和材料属性。此外，我们构建了一个统一的基准，涵盖四个关键方面：空间保真度、物理合理性、下游任务性能和整体质量。该基准包括来自ArtVIP和LightWheel的100个参考资产、50个稳定性提示和50个具有参考质量的ABO对象，评估各种指标的协议。通过对10多个代表性模型的广泛评估，Artisan在实现竞争性的空间保真度的同时，在评估的方法中达到了最低的碰撞率和最高的稳定率。我们的系统和基准共同为推动3D生成朝向物理合理、功能可用的资产提供了见解。
cs.AI / 1 / 2610.12493
Faithfulness Without Anchors: A Position Paper on the Structural Misalignments of Explainable AI Faithfulness Metrics for Feature Importance
无锚的忠实性：关于可解释人工智能忠实性度量在特征重要性上的结构性失调的立场论文
Abstract
Explainable Artificial Intelligence (XAI) tools have often been presented as methods to increase trust in machine learning models by making aspects of their algorithms more human-understandable. However, evidence accumulated over the last decade points to an overall failure of this objective, instead promoting overreliance on incorrect predictions, which can subsequently lead to algorithmic aversion. Rather, XAI appears to function primarily as a debugging tool, useful for obtaining anecdotal insights into, for example, features learned by the model. One of the main issues underlying this limitation is that XAI methods often oversimplify the inner workings of the model, resulting in misalignment between the explanation and the model itself - a behavior referred to as unfaithfulness or infidelity. Faithfulness can be interpreted as the degree of approximation between an explanation and the model. Computing faithfulness in a rigorous manner would thus make it possible to quantify one of the various elements of trustworthiness of an explanation. However, existing metrics present significant limitations, making this quantification either infeasible or computationally expensive. In this position paper, we outline five major shortcomings of current faithfulness metrics and propose four possible avenues for research to bolster their reliability.
Chinese Translation
可解释人工智能（XAI）工具常被视为提高对机器学习模型信任的方法，通过使其算法的某些方面更易于人类理解。然而，过去十年积累的证据表明，这一目标总体上未能实现，反而促进了对错误预测的过度依赖，这可能导致算法厌恶。相反，XAI似乎主要作为调试工具，能够为模型学习的特征等提供轶事性见解。造成这一局限性的主要问题之一是，XAI方法常常过于简化模型的内部工作原理，导致解释与模型之间的不一致，这种行为被称为不忠实性或不可靠性。忠实性可以被理解为解释与模型之间的近似程度。因此，以严格的方式计算忠实性将使得量化解释的各种可信度元素成为可能。然而，现有的度量标准存在显著的局限性，使得这种量化要么不可行，要么计算成本高昂。在本立场论文中，我们概述了当前忠实性度量的五个主要缺陷，并提出了四个可能的研究方向，以增强其可靠性。
cs.AI / 2 / 2610.12536
EliteTable: Self-Improving Program Search for Executable Control Policies
EliteTable：自我改进的可执行控制策略程序搜索
Abstract
EliteTable is a self-improving program-search algorithm guided by large language models (LLMs) that uses a fixed-capacity table of high-scoring solutions to guide new proposals, edits of existing solutions, and remixes of several solutions. Performance was evaluated across six Gymnasium control tasks, each with ten generations and fifty candidate slots per generation. Each task was evaluated across ten independently initialized searches. Candidate-score medians rose between the first and final generations in all six tasks, and selected-winner search-score medians rose in five. In CartPole-v1, the selected winner reached the search-score ceiling in the first generation of all ten searches. All ten paired held-out endpoints improved in four of the six tasks. BipedalWalker-v3 and CarRacing-v3 improved without reaching their reporting targets. Six of ten Acrobot-v1 searches declined on held-out episodes, and the median paired change was negative. The results document improvement during program search and the transfer of selected-winner gains to new episodes in some tasks. Comparisons and ablations are needed to establish EliteTable's relative performance and the contributions of its components.
Chinese Translation
EliteTable是一种由大型语言模型（LLMs）指导的自我改进程序搜索算法，利用固定容量的高分解决方案表来指导新的提案、现有解决方案的编辑以及多个解决方案的混合。在六个Gymnasium控制任务中评估了性能，每个任务进行了十代，每代有五十个候选槽。每个任务在十次独立初始化的搜索中进行了评估。在所有六个任务中，候选分数的中位数在第一代和最后一代之间上升，而选定赢家的搜索分数中位数在五个任务中上升。在CartPole-v1中，选定的赢家在所有十次搜索的第一代中达到了搜索分数的上限。所有十对保留的端点在六个任务中的四个任务中有所改善。BipedalWalker-v3和CarRacing-v3在未达到报告目标的情况下有所改善。十次Acrobot-v1搜索中有六次在保留的回合中下降，且中位数变化为负。结果记录了程序搜索过程中的改进以及在某些任务中选定赢家的收益转移到新回合的情况。需要进行比较和消融实验，以确定EliteTable的相对性能及其各个组件的贡献。
cs.AI / 3 / 2610.12542
Equal-Budget Monte Carlo Tree Search versus Randomized-Restart List Scheduling for DAG Workflows: A Controlled Equal-Budget Computational Study
平等预算蒙特卡洛树搜索与随机重启列表调度在有向无环图工作流中的比较：一项受控的平等预算计算研究
Abstract
Monte Carlo Tree Search (MCTS) is often proposed for scheduling directed-acyclic-graph (DAG) workflows, yet most studies never run the comparison that could refute it - giving a much simpler method the same search effort. We run that comparison under a controlled protocol. All arms receive an identical budget - the same number of candidate schedules built and evaluated: an MCTS scheduling method, preregistered as the main arm; a randomized-restart list scheduler with no search tree as the control; and a textbook-UCT MCTS arm as a registered follow-up on the same domain. The analysis plan was locked before data collection; an independent checker confirmed all 768 confirmation-stage schedules valid with correct finish times. At the main budget of 2048 evaluated schedules, MCTS is 2.88% worse than the restarts (95% CI [-3.62%,-2.15%]; the 72 of 96 held-out instances with positive baseline-relative headroom). The deficit is not budget starvation: on the separate exploration pool restarts self-improve by 1.85% where MCTS gains 0.24%, and 700 of 720 of MCTS's first results are merely the warm start it was handed. The studied selection rule locks deterministically onto the first-expanded child, keeping its trials correlated while restarts draw gains from the tails; the causal link to makespan remains an inference. A structural map locates where the tree stops paying: restarts win on sparse irregular workflows, no difference is detectable on regular layered ones, and dense ones leave no room to improve. Swapping the rule for textbook UCT removes most of the deficit (-0.26%, 95% CI [-0.71%,+0.23%]) without reversing its sign. Results are confined to a controlled synthetic domain (N in {8,16,32} tasks, 2-8 processors, budgets up to 2048); we make no claim about larger budgets.
Chinese Translation
蒙特卡洛树搜索（MCTS）常被提议用于调度有向无环图（DAG）工作流，但大多数研究从未进行过可以反驳它的比较——即给予一个更简单的方法相同的搜索努力。我们在受控协议下进行了该比较。所有方法都获得相同的预算——相同数量的候选调度被构建和评估：MCTS调度方法作为主要方法进行预注册；没有搜索树的随机重启列表调度器作为对照；以及作为同一领域的注册后续的教科书式UCT MCTS方法。分析计划在数据收集之前锁定；独立检查员确认所有768个确认阶段的调度有效且完成时间正确。在主要预算2048个评估调度下，MCTS的表现比重启方法差2.88%（95%置信区间[-3.62%, -2.15%]；96个保留实例中有72个具有正基线相对空间）。这一差距并非预算饥饿：在单独的探索池中，重启方法自我改进了1.85%，而MCTS仅获得0.24%，且720个MCTS的首次结果中有700个仅是其获得的热启动。所研究的选择规则确定性地锁定在首次扩展的子节点上，使其试验相关，而重启方法则从尾部获得收益；与完成时间的因果关系仍然是一个推断。结构图定位了树停止付出的地方：重启方法在稀疏不规则工作流中获胜，在规则分层工作流中没有可检测的差异，而在密集工作流中没有改进的空间。将规则更换为教科书式UCT消除了大部分差距（-0.26%，95%置信区间[-0.71%, +0.23%]），但未能逆转其符号。结果限于一个受控的合成领域（任务数N为{8,16,32}，处理器数为2-8，预算高达2048）；我们不对更大预算做出任何声明。
cs.AI / 4 / 2610.12593
Know What You Don't Know: Agentic Data Synthesis by Evolving Generation Space for Closed-Loop Learning
了解你不知道的：通过演化生成空间实现闭环学习的自主数据合成
Abstract
The continued improvement of large language models increasingly depends on synthetic experience, yet effective data synthesis for interactive agents requires not only executable and verifiable trajectories, but also interactions that provide learning value for the current task agent. Existing methods increasingly incorporate environment feedback, data validation, or learner failures, but still face a feedback-to-decision gap: observed weaknesses do not directly specify how future synthesis should change, or whether they can be represented by the current synthesis space. To address this gap, we propose EvoSyn, a learning-feedback-driven multi-agent framework that organizes synthesis around a revisable state specifying what interaction factors can be deliberately generated and where synthesis effort should be allocated. A role-specialized multi-agent system translates learner feedback into synthesis revisions, while a Designer allocates synthesis directions and a Planner grounds them into executable experience. This process enables within-space adaptation toward weak or under-covered regions and synthesis-space expansion that promotes newly discovered factors into explicit generation controls. Experiments on $\tau^2$-Bench and CAR-Bench show that EvoSyn improves average task success by 13.3 points over the strongest synthesis baseline, while producing higher-quality and structurally broader synthetic experience and discovering controllable factors missed without learner feedback.
Chinese Translation
大型语言模型的持续改进日益依赖于合成经验，但有效的交互代理数据合成不仅需要可执行和可验证的轨迹，还需要为当前任务代理提供学习价值的交互。现有方法越来越多地纳入环境反馈、数据验证或学习者失败，但仍面临反馈与决策之间的差距：观察到的弱点并未直接指明未来合成应如何改变，或是否可以通过当前的合成空间表示。为了解决这一差距，我们提出了EvoSyn，这是一个以学习反馈为驱动的多代理框架，围绕可修订状态组织合成，该状态指定可以故意生成的交互因素以及合成努力应分配到何处。一个角色专门化的多代理系统将学习者反馈转化为合成修订，而设计者分配合成方向，规划者将其落实为可执行经验。这个过程使得在空间内适应弱或覆盖不足的区域成为可能，并促进合成空间的扩展，将新发现的因素纳入明确的生成控制。对$ au^2$-Bench和CAR-Bench的实验表明，EvoSyn在最强合成基线之上提高了平均任务成功率13.3个百分点，同时生成了更高质量和结构更广泛的合成经验，并发现了在没有学习者反馈的情况下被遗漏的可控因素。
cs.AI / 5 / 2610.12596
SpireBench: Evaluating Long-Horizon Strategic Play of Agentic Models with Slay the Spire
SpireBench：评估代理模型在《击败尖塔》中的长期战略玩法
Abstract
Agentic large language models now tackle complex tasks, yet strategic reasoning over long horizons is under-studied: most agentic benchmarks are short and episodic, and performance on long tasks is hard to dissect into stored knowledge, reasoning and luck. We introduce SpireBench, in which models play the complete commercial deck-builder Slay the Spire (StS), where early decisions decide long-term success. Across 20 large language models (LLMs) at the game's highest difficulty, frontier models beat heuristic baselines and match the average human player but remain far behind experts. This gap widens as the run lengthens: the strongest models win the first act as often as experts, then fall behind. What separates strong models from weak is how well they play each fight; all of them misjudge risk over a run, preserving health rather than spending it to grow stronger. A comprehensive, self-consistent renaming of every game term shows that part of their performance rests on prior knowledge of the game, and models do not improve from the memories they write across runs, which are often wrong and overly cautious. Our harness is the first to give a model the complete original game, and we compare it with existing harnesses. Overall, SpireBench offers an unsaturated long-horizon benchmark with tools to locate where and why agents fail.
Chinese Translation
代理大型语言模型现在能够处理复杂任务，但对长期战略推理的研究仍然不足：大多数代理基准测试都是短期和情节性的，而在长任务上的表现难以拆解为存储知识、推理和运气。我们引入了SpireBench，在该基准中，模型玩完整的商业卡牌构建游戏《击败尖塔》（Slay the Spire，StS），其中早期决策决定了长期成功。在游戏的最高难度下，20个大型语言模型（LLMs）超越了启发式基准，并与平均人类玩家相匹配，但仍远远落后于专家。随着游戏进程的延长，这一差距进一步扩大：最强的模型在第一幕的胜率与专家相当，但随后落后。强模型与弱模型的区别在于它们在每场战斗中的表现；所有模型在整个游戏过程中都错误评估风险，倾向于保留生命值而不是花费它来增强实力。对每个游戏术语进行全面、自洽的重命名表明，它们的部分表现依赖于对游戏的先前知识，而模型在多次游戏中所写的记忆并未带来改进，这些记忆往往是错误的且过于谨慎。我们的基准是第一个将完整原版游戏提供给模型的工具，并与现有基准进行了比较。总体而言，SpireBench提供了一个未饱和的长期基准，并配备了工具以定位代理失败的原因和位置。
cs.AI / 6 / 2610.12598
Beyond Binary Verifiable Rewards: Behavioral Self-Rewarding for SWE Agents
超越二元可验证奖励：软件工程代理的行为自我奖励
Abstract
Reinforcement learning with verifiable rewards (RLVR) provides reliable supervision for software engineering agents through executable tests, but binary rewards fail to distinguish trajectories with substantially different debugging quality. We investigate whether the current policy can judge its own trajectories and use these judgments to improve task performance without a separately trained reward model. To this end, we introduce Behavioral Self-Rewarding (BSR), a reward-refinement method that reuses the policy to compare observable debugging behavior among trajectories from the same task with the same verifier outcome. BSR aggregates pairwise preferences into task-local scores and converts them into bounded auxiliary rewards, introducing reward contrast while preserving the verifier-defined pass/fail ordering. This provides additional supervision from already-collected trajectories without requiring new environment interactions. Experiments on SWE-bench Verified and SWE-bench-Live show consistent improvements across the evaluated Qwen3-14B and Qwen3-32B configurations, supporting using the policy's own comparative judgments as a complementary learning signal for training software engineering agents.
Chinese Translation
具有可验证奖励的强化学习（RLVR）通过可执行测试为软件工程代理提供可靠的监督，但二元奖励无法区分调试质量显著不同的轨迹。我们研究当前策略是否能够判断其自身轨迹，并利用这些判断在不需要单独训练奖励模型的情况下提高任务性能。为此，我们引入了行为自我奖励（Behavioral Self-Rewarding，BSR），这是一种奖励精炼方法，它重用策略来比较来自同一任务且具有相同验证结果的轨迹中的可观察调试行为。BSR将成对偏好聚合为任务局部评分，并将其转换为有界辅助奖励，引入奖励对比，同时保留验证者定义的通过/失败顺序。这提供了来自已收集轨迹的额外监督，而无需新的环境交互。在SWE-bench Verified和SWE-bench-Live上的实验显示，在评估的Qwen3-14B和Qwen3-32B配置中，性能一致改善，支持使用策略自身的比较判断作为训练软件工程代理的补充学习信号。
cs.AI / 7 / 2610.12655
CLEAT: Learning Value-Guided Clarification for Interactive Agents
CLEAT：为交互代理学习价值引导的澄清
Abstract
LLM agents can execute increasingly complex tasks, but effective assistance also requires aligning their actions with goals and constraints that users may leave unstated. We study this problem as interactive intent alignment through clarification. Yet clarification is useful only when the agent can turn the answer into better decisions. As training changes how the agent handles incomplete information and uses additional answers, the same clarification can become more or less useful. We introduce CLEAT (CLarification through valuE-guided Agent co-Training), a framework that co-trains a clarification-value controller and the task policy from shared interaction outcomes. We formulate residual query value to measure the task utility gained by asking rather than proceeding under the current policy, and estimate it through paired rollouts from the same interaction history. These paired returns train the controller to predict clarification value and the task policy to select useful questions and act on the answers. The controller guides subsequent rollout collection, while outcomes from the updated agent provide new value labels that allow clarification guidance to adapt as the agent learns. With Qwen3-4B, CLEAT achieves the highest mean score in all ten evaluation settings across four benchmarks. Gains over the strongest baseline reach 13.2 and 6.5 percentage points on UserGym Travel and IN3 underspecification judgment accuracy, respectively. Ablations show that co-training improves the task policy even when the controller is removed at inference, while adapting clarification guidance to the evolving policy yields further gains.
Chinese Translation
大型语言模型（LLM）代理能够执行日益复杂的任务，但有效的辅助还需要将其行动与用户可能未明确说明的目标和约束对齐。我们将这个问题研究为通过澄清进行的交互意图对齐。然而，澄清只有在代理能够将答案转化为更好的决策时才是有用的。随着训练改变代理处理不完整信息的方式以及使用额外答案的方式，相同的澄清可能变得更有用或更无用。我们引入了CLEAT（通过价值引导的代理共同训练进行澄清），这是一个共同训练澄清-价值控制器和任务策略的框架，基于共享的交互结果。我们制定了残余查询价值的概念，以衡量通过提问而不是按照当前策略继续进行所获得的任务效用，并通过来自相同交互历史的配对回放进行估计。这些配对回报训练控制器预测澄清价值，并训练任务策略选择有用的问题并对答案进行响应。控制器指导后续的回放收集，而更新代理的结果提供新的价值标签，使澄清指导能够随着代理的学习而适应。在Qwen3-4B上，CLEAT在四个基准的十个评估设置中达到了最高的平均分数。在UserGym Travel和IN3的未充分说明判断准确性上，相较于最强基线的提升分别达到了13.2和6.5个百分点。消融实验表明，即使在推理时移除控制器，共同训练也能改善任务策略，而将澄清指导适应于不断演变的策略则带来了进一步的提升。
cs.AI / 8 / 2610.12659
MagicBeanStalk: Escaping the RLVR Zero Advantage Trap with Recursive Data Synthesis
MagicBeanStalk：通过递归数据合成逃离RLVR零优势陷阱
Abstract
Data-centric scaling of RLVR has focused on collecting or generating verifiable training problems. However, corpus size alone does not determine whether the dataset provides a useful learning signal for a target policy. To guide the synthesis of effective RLVR data, we experimentally characterize dataset effectiveness as a function of its problems, the target policy, and the training rollout budget. Under this characterization, we show that a large open-source math corpus pooled from widely used RLVR datasets falls almost entirely into a zero-advantage trap: for a fixed target policy and rollout budget, roughly 88% of its problems are solved in every rollout or failed in every rollout. Filtering concentrates what remains, but it can only select, discarding the hard problems. Effective synthetic problems should therefore require the reasoning operations and compositions required to solve those hard problems. Yet their difficulty must be calibrated, so that some rollouts succeed, some fail, and at least one correct solution fits within the training budget. Prompting a teacher to vary hard problems directly does not achieve this. Guided by these empirical principles, we introduce MagicBeanStalk, a top-down decomposition and bottom-up recursive data-synthesis algorithm that grows seed problems into families of substantively new, policy- and budget-calibrated training problems while preserving the seeds' reasoning blueprints. Across 8 math benchmarks, continual RLVR on MagicBeanStalk problems improves gpt-oss-120b from 61.2 to 69.8 (+8.6) and DeepSeek v4Flash from 72.4 to 77.9 (+5.4) on a 16K-token budget. Training on the full 78K source corpus improves by only +1.1 and its best filtered real data subset by +5.8. In finance, we synthesize over 14K verifiable problems from only 340 solutions and improve over base by 6.9 and 2.4 points on SecQue and FinanceBench, respectively.
Chinese Translation
以数据为中心的RLVR扩展关注于收集或生成可验证的训练问题。然而，仅仅依靠语料库的规模并不能决定数据集是否为目标策略提供有用的学习信号。为了指导有效RLVR数据的合成，我们通过实验将数据集的有效性表征为其问题、目标策略和训练回合预算的函数。在这种表征下，我们展示了一个来自广泛使用的RLVR数据集的大型开源数学语料库几乎完全陷入了零优势陷阱：对于固定的目标策略和回合预算，约88%的问题在每次回合中都被解决或在每次回合中都失败。过滤集中剩余的问题，但它只能选择，丢弃困难的问题。因此，有效的合成问题应当需要解决这些困难问题所需的推理操作和组合。然而，它们的难度必须经过校准，以便某些回合成功，某些回合失败，并且至少有一个正确的解决方案符合训练预算。直接提示教师变化困难问题并不能实现这一点。在这些经验原则的指导下，我们引入了MagicBeanStalk，这是一种自上而下的分解和自下而上的递归数据合成算法，能够将种子问题发展为实质性的新问题家族，这些问题经过策略和预算校准，同时保留种子的推理蓝图。在8个数学基准测试中，针对MagicBeanStalk问题的持续RLVR使gpt-oss-120b的性能从61.2提高到69.8（+8.6），使DeepSeek v4Flash的性能从72.4提高到77.9（+5.4），在16K令牌预算下。对完整的78K源语料库的训练仅提高了+1.1，而其最佳过滤的真实数据子集提高了+5.8。在金融领域，我们仅从340个解决方案合成了超过14K个可验证的问题，并在SecQue和FinanceBench上分别提高了6.9和2.4分。
cs.AI / 9 / 2610.12762
Faithful and Complete LLM-Assisted Automated Construction of Hierarchical Hardware Knowledge Graphs
忠实且完整的LLM辅助自动构建层次化硬件知识图谱
Abstract
Hardware datasheets contain engineering specifications essential for component selection, system integration, and hardware compatibility verification. However, manual inspection is tedious and error-prone because datasheets are lengthy and heterogeneous, with dense information fragmented across tables, figures, and text. Although large language models (LLMs) can automate this process, na\"ive one-shot processing omits relevant specifications even when the document fits within the model's context window. We present an automated LLM-driven pipeline that transforms hardware datasheets into general-purpose hierarchical hardware knowledge graphs through chunk-based property extraction, hierarchical routing, and iterative refinement. The framework uses LLMs for information extraction and semantic reasoning while preserving source provenance and applying graph updates through validated instructions and deterministic local execution. This staged design reduces per-query context and maintains traceable intermediate representations. The resulting knowledge graphs provide semantically organized, source-grounded hardware properties for downstream retrieval and reasoning. We evaluate the framework on 20 hardware datasheets spanning diverse component types and document lengths. We assess property faithfulness and introduce inter-graph completeness to measure relative semantic coverage without exhaustive ground truth. Our framework achieves 98.8% property faithfulness compared with 76.1% for the one-shot baseline, while fully covering 83.6% of faithful baseline properties; in the reverse direction, the baseline fully covers only 10.5% of ours. These improvements provide more reliable and comprehensive engineering information for downstream hardware design and verification, reducing the risk of overlooked constraints and incorrect decisions.
Chinese Translation
硬件数据表包含组件选择、系统集成和硬件兼容性验证所需的工程规格。然而，由于数据表冗长且异构，信息密集且分散在表格、图形和文本中，手动检查既繁琐又容易出错。尽管大型语言模型（LLMs）可以自动化这一过程，但简单的一次性处理即使在文档适合模型的上下文窗口时也会遗漏相关规格。我们提出了一种自动化的LLM驱动管道，通过基于块的属性提取、层次路由和迭代优化，将硬件数据表转化为通用的层次化硬件知识图谱。该框架利用LLMs进行信息提取和语义推理，同时保留源出处，并通过经过验证的指令和确定性的局部执行应用图更新。这种分阶段设计减少了每个查询的上下文，并保持可追溯的中间表示。生成的知识图谱为下游检索和推理提供了语义组织良好、源基础的硬件属性。我们在20个涵盖不同组件类型和文档长度的硬件数据表上评估了该框架。我们评估了属性的忠实度，并引入了图间完整性来测量相对语义覆盖，而无需详尽的真实情况。与一次性基线的76.1%相比，我们的框架实现了98.8%的属性忠实度，同时完全覆盖了83.6%的忠实基线属性；反之，基线仅完全覆盖了我们属性的10.5%。这些改进为下游硬件设计和验证提供了更可靠和全面的工程信息，减少了忽视约束和错误决策的风险。
cs.AI / 10 / 2610.12781
Representation Alignment from a Few Paired Data via Distilled Optimal Transport
通过蒸馏最优运输实现少量配对数据的表示对齐
Abstract
Aligning two independently pretrained unimodal encoders typically demands millions of paired examples. In this paper, we study the regime where only $N \sim 100$--$500$ pairs are available, while unpaired data on either side is abundant. Our method expresses every representation by its similarities to the $N$ paired anchors, which places both modalities in a common coordinate system; matches the two point clouds with a single entropic optimal transport problem solved over the entire pool, whose marginal constraints counteract the hubness that defeats nearest-neighbor matching; and distills the resulting plan into two lightweight encoders, so that the correspondence becomes inductive. We additionally put this marginal constraint in relation with the one-step hub corrections of the bilingual lexicon induction literature. Pairing is required only for the anchors: the unpaired collections may be drawn independently from the two marginals, so that no item outside the anchor set needs to have both of its modalities observed. We evaluate on image-text and molecule-text alignment, where our method outperforms existing approaches at the same anchor budget on Flickr30k, MS-COCO and ChEBI-20.
Chinese Translation
对齐两个独立预训练的单模态编码器通常需要数百万个配对样本。在本文中，我们研究了仅有 $N ext{ 约 } 100$--$500$ 对样本可用的情况，而任一侧的未配对数据则非常丰富。我们的方法通过与 $N$ 个配对锚点的相似性来表达每个表示，这将两种模态置于一个共同的坐标系统中；通过在整个池上解决单个熵最优运输问题来匹配两个点云，其边际约束抵消了导致最近邻匹配失败的中心性问题；并将得到的计划蒸馏为两个轻量级编码器，使得对应关系变得具有归纳性。我们还将这一边际约束与双语词典诱导文献中的一步中心修正进行了关联。配对仅对锚点是必需的：未配对的集合可以独立于两个边际进行抽取，因此锚点集之外的任何项目都不需要同时观察其两个模态。我们在图像-文本和分子-文本对齐任务上进行了评估，我们的方法在相同锚点预算下在 Flickr30k、MS-COCO 和 ChEBI-20 数据集上超越了现有方法。
cs.AI / 11 / 2610.12789
EvoBelief: Falsifiable World Models for Self-Improving LLM Agents
EvoBelief：可证伪的世界模型用于自我改进的LLM代理
Abstract
Self-evolving LLM agents typically improve how they act through accumulated experience, reflection, or skills. Yet even procedurally capable agents can fail when their beliefs about environment dynamics are incomplete, incorrect, or outdated. We study a complementary axis of self-evolution: evolving those beliefs during deployment. Such beliefs can support multiple tasks, but incorrect ones can also propagate across them. We introduce EvoBelief, a self-evolving world model that represents environment knowledge as falsifiable hypotheses. Unexpected outcomes and decision-relevant unknowns trigger hypothesis formation. Later observations and controlled experiments provide independent validation before a rule can guide predictions or plans within its supported scope. Counterexamples narrow or retire rules as the environment changes. The language model remains frozen, and each rule remains linked to its evidence and the decisions it changes. Across four benchmarks spanning static dynamics, hidden drift, and scientific discovery, EvoBelief improves task performance and reuses learned rules across tasks. On ALFWorld without actor-side transition memory, it improves success over the matched Frozen World Model by 23.9% and 31.3% with Claude Haiku 4.5 and GPT-5.4, respectively. On MiniGrid, it revises stale rules and recovers after unannounced dynamics changes. In DiscoveryWorld's Reactor Lab, it discovers and validates quantitative laws, completing 83.3% of Normal-tier and 58.3% of Challenge-tier instances, versus 0% for each evaluated baseline. These results demonstrate self-improvement through testable, revisable environment knowledge rather than model-parameter updates.
Chinese Translation
自我进化的LLM代理通常通过积累的经验、反思或技能来改善其行为。然而，即使是具备程序能力的代理，当其对环境动态的信念不完整、不正确或过时时，也可能会失败。我们研究自我进化的一个补充方向：在部署过程中演变这些信念。这些信念可以支持多种任务，但不正确的信念也可能在任务之间传播。我们提出了EvoBelief，这是一种自我进化的世界模型，将环境知识表示为可证伪的假设。意外结果和与决策相关的未知因素会触发假设的形成。后续观察和受控实验提供独立验证，以便在其支持的范围内，规则能够指导预测或计划。反例会随着环境的变化而缩小或撤回规则。语言模型保持不变，每个规则与其证据及其所改变的决策保持关联。在涵盖静态动态、隐藏漂移和科学发现的四个基准测试中，EvoBelief提高了任务性能，并在任务之间重用学习到的规则。在没有演员侧过渡记忆的ALFWorld中，它在与匹配的冻结世界模型相比，成功率提高了23.9%，与Claude Haiku 4.5和GPT-5.4分别提高了31.3%。在MiniGrid中，它修订了过时的规则，并在未宣布的动态变化后恢复。在DiscoveryWorld的反应堆实验室中，它发现并验证了定量法则，完成了83.3%的普通级和58.3%的挑战级实例，而每个评估的基线实例均为0%。这些结果表明，通过可测试、可修订的环境知识实现自我改进，而不是通过模型参数更新。
cs.AI / 12 / 2610.12813
VEPA: Verified Evolving Formal Policy for Lifelong LLM Agent Guardrails
VEPA：经过验证的演变形式政策，用于终身 LLM 代理的安全防护
Abstract
LLM agents increasingly operate over sensitive resources, such as private documents and mutable system states, which can create serious safety and security risks. Runtime guardrails should therefore make inspectable, reproducible, and deterministic allow or refuse decisions before potentially irreversible agent actions are executed. However, safety and security requirements are rarely complete before deployment, so guardrails need to keep adapting as new risks emerge. Existing adaptive guardrails learn from deployment experience, but they usually apply safety policies through LLM inference, which can lead to inconsistent decisions and ambiguous policy interpretations. Formal guardrails, on the other hand, enforce policies deterministically, but they usually rely on pre-deployment or task-specific policies and have no mechanism to adapt to newly observed risks. We introduce VEPA, a framework for the verified evolution of formal agent guardrails. VEPA represents the guardrail as a formal policy of executable symbolic rules over typed agent events. During deployment, we use an LLM guided by deployment feedback to propose policy patches that fix newly observed failures while preserving previously correct behavior. Then, a deterministic verifier only accepts policies that are structurally valid and conflict-free. Finally, an evidence-aware authority layer activates a verified rule only when it has enough supporting evidence, so uncertain updates cannot directly affect guardrail decisions. Across ASSEBench-Safety, ASSEBench-Security, and ATBench, VEPA consistently outperforms baselines while expanding the scope of deterministic formal-policy enforcement. These gains also hold under more challenging deployment conditions, including mixed streams, corrupted feedback, and longer horizons.
Chinese Translation
LLM 代理越来越多地在敏感资源上运行，例如私人文档和可变系统状态，这可能会带来严重的安全和保密风险。因此，运行时的安全防护应在潜在不可逆的代理操作执行之前，做出可检查、可复现和确定性的允许或拒绝决策。然而，安全和保密要求在部署前往往不够完整，因此安全防护需要随着新风险的出现而不断适应。现有的自适应安全防护从部署经验中学习，但它们通常通过 LLM 推理来应用安全政策，这可能导致不一致的决策和模糊的政策解释。另一方面，形式安全防护以确定性方式执行政策，但通常依赖于预部署或特定任务的政策，并且没有适应新观察到的风险的机制。我们提出了 VEPA，一个用于经过验证的形式代理安全防护演变的框架。VEPA 将安全防护表示为可执行符号规则的形式政策，适用于类型化的代理事件。在部署过程中，我们使用一个由部署反馈引导的 LLM 来提出政策补丁，以修复新观察到的故障，同时保持之前的正确行为。然后，确定性验证器只接受结构上有效且无冲突的政策。最后，一个关注证据的权威层仅在拥有足够支持证据时激活经过验证的规则，因此不确定的更新不能直接影响安全防护决策。在 ASSEBench-Safety、ASSEBench-Security 和 ATBench 上，VEPA 一直优于基线，同时扩展了确定性形式政策执行的范围。这些收益在更具挑战性的部署条件下也得以保持，包括混合流、损坏的反馈和更长的时间范围。
cs.AI / 13 / 2610.12822
Reward Hacking as a Transmissible Strategy in Multi-Agent Coding Systems
奖励黑客作为多智能体编码系统中的可传播策略
Abstract
Multi-agent systems are increasingly used to solve coding tasks by coordinating agents that share code and plans. These agents are often optimized with reinforcement learning (RL) using automated evaluators, but bypassing evaluation can earn reward without producing a genuine solution. Prior work has largely treated reward hacking as a single-agent failure. In this work, we show that a reward-hacking strategy elicited in one agent can propagate to others and affect system coordination. Across four open-weight backbones, coding RL elicits and amplifies reward hacking in two settings: explicit mechanism descriptions in the system prompt and mid-training that models upstream exposure to facts about the coding environment. Using post-RL agents from the mid-training setting as sources, we observe propagation through code and shared memory, and across roles through natural-language plans. Even agents based on small open-weight models induce reward hacking in stronger closed-weight targets. On hard tasks, propagation can intensify along a transmission chain after the original source leaves as the code acquires more persuasive justifications. We also theoretically analyze conditions under which reward hacking amplifies, persists, or decays across agent interactions. Beyond propagation, reward hacking can bias task routing toward reward-hacking agents and survive candidate aggregation. As potential mitigations, we evaluate code review by the coder or a separate verifier, both of which reduce delivered hacks. Together, these results establish reward hacking as a transmissible strategy whose effects extend beyond the agent in which it first emerges.
Chinese Translation
多智能体系统越来越多地用于通过协调共享代码和计划的智能体来解决编码任务。这些智能体通常使用强化学习（RL）和自动评估器进行优化，但绕过评估可以在不产生真实解决方案的情况下获得奖励。先前的研究主要将奖励黑客视为单智能体的失败。在本研究中，我们展示了在一个智能体中引发的奖励黑客策略可以传播到其他智能体并影响系统协调。在四个开放权重基础模型中，编码强化学习在两种设置中引发并放大了奖励黑客：系统提示中的明确机制描述和训练中期模型对编码环境事实的上游暴露。我们观察到，从训练中期设置的后RL智能体作为源，通过代码和共享内存传播，并通过自然语言计划在不同角色之间传播。即使是基于小型开放权重模型的智能体也会在更强的封闭权重目标中引发奖励黑客。在困难任务中，传播可以在原始源离开后沿着传输链加剧，因为代码获得了更具说服力的理由。我们还理论分析了奖励黑客在智能体交互中放大、持续或衰减的条件。除了传播，奖励黑客还可以使任务路由偏向奖励黑客智能体，并在候选聚合中存活。作为潜在的缓解措施，我们评估了由编码者或单独验证者进行的代码审查，这两者均减少了交付的黑客。综合这些结果，确立了奖励黑客作为一种可传播策略，其影响超出了首次出现的智能体。
cs.AI / 14 / 2610.12842
Neural networks are more modular than single neurons suggest
神经网络的模块化程度超出单个神经元的建议
Abstract
Neural networks routinely develop representations in which individual neurons contribute to multiple distinct functions, leading to the widespread view that such networks lack modular organization. We challenge this conclusion by introducing a formally nested hierarchy of modularity: unit-level, which requires distinct computations to occupy disjoint neurons; orthogonal, which asks whether a rotation reveals functionally independent subspaces; and linear, which asks whether a bounded linear map can separate task-relevant computations into independent subspaces. We develop a unified framework for learning decompositions at each level and evaluating them causally through selective ablation. Across vision models, language models and multi-task Recurrent Neural Networks (RNNs), we find that networks which lack unit-level modularity nonetheless often exhibit clear functional modularity at the subspace level. In vision and language models, content versus style and syntax versus semantics respectively dissociate under subspace ablation despite sharing overlapping populations of neurons. These learned subspaces also align differentially with brain data, with content and style subspaces predicting higher and lower visual cortex respectively, and semantic subspaces outperforming syntactic ones in predicting language cortex activity. In RNNs trained on multiple cognitive tasks and previously shown to lack modular structure, our decomposition recovers sharp double dissociations between tasks. The absence of unit-level modularity frequently coexists with clear subspace-level modularity across domains, enabling targeted intervention on specific computations even in networks where no individual neuron is specialized, with broad implications for mechanistic interpretability and for understanding functional organization in both biological and artificial neural systems.
Chinese Translation
神经网络通常会发展出个别神经元对多个不同功能的贡献，这导致普遍认为此类网络缺乏模块化组织。我们通过引入一个正式的模块化嵌套层次结构来挑战这一结论：单元级模块化，要求不同的计算占据不重叠的神经元；正交模块化，询问旋转是否揭示功能上独立的子空间；线性模块化，询问是否可以用有界线性映射将与任务相关的计算分隔到独立的子空间。我们开发了一个统一框架，用于学习每个层次的分解并通过选择性消融进行因果评估。在视觉模型、语言模型和多任务递归神经网络（RNN）中，我们发现缺乏单元级模块化的网络，通常在子空间层面上仍然表现出明显的功能模块化。在视觉和语言模型中，内容与风格、语法与语义在子空间消融下分别解耦，尽管它们共享重叠的神经元群体。这些学习到的子空间与脑数据也存在差异性对齐，内容和风格子空间分别预测更高和更低的视觉皮层，而语义子空间在预测语言皮层活动方面优于语法子空间。在训练于多个认知任务且之前被证明缺乏模块结构的RNN中，我们的分解恢复了任务之间的明显双重解耦。单元级模块化的缺失常常与跨领域的明确子空间级模块化共存，使得即使在没有单个神经元专门化的网络中，也能对特定计算进行有针对性的干预，这对机制可解释性和理解生物及人工神经系统的功能组织具有广泛的影响。
cs.AI / 15 / 2610.12845
Towards Defining, Measuring, and Leveraging Creativity in AI Applications
定义、测量和利用人工智能应用中的创造力
Abstract
Despite extensive research, computational creativity still lacks a conceptually grounded, practically relevant, and empirically applicable formal definition across contexts such as art, scientific research, and games. Existing approaches to creativity often focus on proposing informal conceptual definitions without providing formal computational concepts. Alternatively, they propose formal definitions without carefully examining their underlying assumptions or providing validation of these definitions by testing their performance in specific application settings. To address these problems, this paper proposes several formal postulates of creativity grounded in prior psychological work on creativity, especially as related to Boden's three core criteria of creativity: newness, value, and surprise. Furthermore, we propose a conceptual mathematical definition of creativity that aligns with these postulates. We also validate our proposed creativity formula in two applications: determining the creativity of particular chess moves, and assessing the creativity of paintings by some artists, such as Picasso, Renoir, and other painters. Experimental results demonstrate that our formula of creativity successfully captures creative patterns across both the chess and the art domains. This suggests that creativity can be effectively characterized through a unified mathematical structure based on three fundamental criteria, namely newness, value, and surprise, whose specific measurements are adapted to the particular domains.
Chinese Translation
尽管已有大量研究，但计算创造力在艺术、科学研究和游戏等不同背景下仍然缺乏一个概念上扎实、实际相关且可实证应用的正式定义。现有的创造力研究往往侧重于提出非正式的概念定义，而未提供正式的计算概念。或者，它们提出了正式定义，但未仔细审视其潜在假设或通过在特定应用环境中测试其性能来验证这些定义。为了解决这些问题，本文提出了几个基于先前心理学研究的创造力正式公设，特别是与博登（Boden）提出的创造力三大核心标准：新颖性、价值和惊讶相关的公设。此外，我们还提出了一个与这些公设相一致的创造力概念数学定义。我们在两个应用中验证了所提出的创造力公式：确定特定国际象棋走法的创造力，以及评估一些艺术家（如毕加索、雷诺阿等）的绘画作品的创造力。实验结果表明，我们的创造力公式成功捕捉了国际象棋和艺术领域的创造性模式。这表明，创造力可以通过基于新颖性、价值和惊讶这三个基本标准的统一数学结构有效地表征，其具体测量适应于特定领域。
cs.AI / 16 / 2610.12855
Mitigating Capability Degradation in LLM Self-Improvement via Representation Geometry Regularization
通过表示几何正则化减轻大型语言模型自我提升中的能力退化
Abstract
On-Policy Self-distillation is a self-improvement paradigm that enables large language models (LLMs) to improve their capabilities by leveraging the same model as both teacher and student under different conditioning contexts. However, we find that OPSD methods can progressively degrade previously acquired capabilities: queries that are correctly responded at earlier training stages may become incorrectly responded as optimization proceeds, even the overall performance continues to improve. Further analysis reveals that this degradation coincides with a deterioration of the geometry of hidden representations, characterized by reduced intra-cluster compactness and diminished inter-cluster separation. Motivated by this observation, we propose Representation Geometry Regularization (RGR), which preserves representation geometry during post-training by encouraging compactness within cluster and separation across cluster. RGR can be seamlessly integrated into existing OPSD methods without altering their core optimization procedures. Extensive experiments on multiple OPSD methods and benchmarks show that RGR generally alleviates capability degradation while retaining capability improvements. Our code is available in https://github.com/supermachinelearner/RGR.
Chinese Translation
在政策自蒸馏（On-Policy Self-distillation）中，作为一种自我提升范式，它使大型语言模型（LLMs）能够通过在不同条件上下文中将同一模型作为教师和学生来提高其能力。然而，我们发现，OPSD（On-Policy Self-Distillation）方法可能会逐渐退化先前获得的能力：在早期训练阶段正确响应的查询，随着优化的进行可能会变得响应错误，即使整体性能持续改善。进一步分析表明，这种退化与隐藏表示几何的恶化相吻合，表现为簇内紧凑性降低和簇间分离度减小。基于这一观察，我们提出了表示几何正则化（Representation Geometry Regularization, RGR），该方法通过鼓励簇内紧凑性和簇间分离性，在后训练阶段保持表示几何。RGR可以无缝集成到现有的OPSD方法中，而无需改变其核心优化过程。在多个OPSD方法和基准上的大量实验表明，RGR通常能够减轻能力退化，同时保持能力的提升。我们的代码可在 https://github.com/supermachinelearner/RGR 获取。
cs.AI / 17 / 2610.12872
Collaborative Theory of Mind: Learning to Self-Revise by Anticipating User Responses
协作心智理论：通过预测用户反应学习自我修正
Abstract
Large language models (LLMs) demonstrate strong individual task performance, yet still struggle to collaborate with users over extended, multi-turn interactions, for example by attempting solutions before understanding the user's needs or by failing to revise mistaken assumptions after receiving feedback. We propose Collaborative Theory of Mind (CToM), a framework that addresses these challenges by explicitly training LLMs to anticipate user responses and to use those predictions to improve their own behavior. CToM drafts a response, predicts how the user will respond, and uses this prediction to guide internal revision before presenting a response. During training, the user's reply to each draft serves as the target for predicting the next turn, and we train the assistant to minimize the negative log-likelihood of that reply. We combine this prediction objective with group-relative policy optimization (GRPO) to optimize both task performance and interaction efficiency. Across mathematical reasoning, coding, and writing tasks, CToM achieves higher collaborative task performance than all baselines; for example, on a collaborative mathematical reasoning task, it improves final accuracy from 74.1% to 79.6% compared with naive GRPO. Ablations show that predicting user responses also makes collaboration more efficient: under the same reward, adding prediction supervision reduces the number of user replies by 8% on coding and 17% on writing while improving task performance. Behavioral analyses of interaction trajectories show that user-response predictions become more accurate after training, and that more accurate predictions are associated with larger improvements in collaborative task performance through self-revision.
Chinese Translation
大型语言模型（LLMs）在单一任务表现上展现出强大的能力，但在与用户进行长时间、多轮的互动时仍然面临挑战，例如在理解用户需求之前就尝试解决方案，或在收到反馈后未能修正错误假设。我们提出了协作心智理论（Collaborative Theory of Mind, CToM），这是一个通过明确训练LLMs预测用户反应并利用这些预测来改善自身行为的框架。CToM起草响应，预测用户的反应，并利用这一预测在呈现响应之前指导内部修正。在训练过程中，用户对每个草稿的回复作为预测下一轮的目标，我们训练助手以最小化该回复的负对数似然。我们将这一预测目标与群体相对策略优化（Group-Relative Policy Optimization, GRPO）相结合，以优化任务表现和互动效率。在数学推理、编码和写作任务中，CToM的协作任务表现优于所有基线；例如，在协作数学推理任务中，与简单的GRPO相比，最终准确率从74.1%提高到79.6%。消融实验表明，预测用户反应也使协作更加高效：在相同奖励下，增加预测监督使编码任务的用户回复数量减少了8%，写作任务减少了17%，同时提高了任务表现。对互动轨迹的行为分析显示，用户反应预测在训练后变得更加准确，而更准确的预测与通过自我修正在协作任务表现上的更大提升相关。
cs.AI / 18 / 2610.12879
One-Step Text Generation by Seq-Drifting
通过序列漂移实现一步文本生成
Abstract
Diffusion language models generate tokens in parallel, but still require multiple sampling steps to produce high-quality text. In this paper, we investigate whether a model can learn to generate an entire sequence in a single forward pass. We propose Seq-Drifting, a framework that applies drifting to sequences of continuous token representations and directly learns a map from Gaussian noise to text. Unlike existing DLMs that derive the supervision at each position directly from the observed token in a training sequence, Seq-Drifting builds a drift field whose position-wise targets adapt to the current generated context. At each position, attraction guides each output toward a set of contextually plausible token representations, while repulsion discourages collapse across samples and repetition within sequences. The generator learns from this drift field during training and produces all tokens jointly in one pass at inference. Empirically, Seq-Drifting achieves a generative perplexity of 36.01 on LM1B in one step and performs competitively in both unconditional and conditional generation. Experiments on translation, summarization, and reasoning further demonstrate the applicability of the framework beyond open-ended generation. These results suggest that drifting over token sequences offers a promising path toward effective one-step text generation. Our code is available at https://github.com/jyliu-ai/Seq-Drifting.
Chinese Translation
扩散语言模型以并行方式生成标记，但仍需多个采样步骤才能生成高质量文本。本文研究了模型是否可以在单次前向传播中学习生成整个序列。我们提出了序列漂移（Seq-Drifting）框架，该框架将漂移应用于连续标记表示的序列，并直接学习从高斯噪声到文本的映射。与现有的扩散语言模型（DLMs）直接从训练序列中观察到的标记推导每个位置的监督不同，序列漂移构建了一个漂移场，其位置目标适应于当前生成的上下文。在每个位置，吸引力引导每个输出朝向一组上下文上合理的标记表示，而排斥力则阻止样本之间的崩溃和序列内的重复。生成器在训练过程中从这个漂移场中学习，并在推理时一次性联合生成所有标记。实证结果表明，序列漂移在 LM1B 上以一步生成的方式达到了 36.01 的生成困惑度，并在无条件和条件生成中表现出竞争力。关于翻译、摘要和推理的实验进一步证明了该框架在开放式生成之外的适用性。这些结果表明，标记序列的漂移为有效的一步文本生成提供了一条有前景的路径。我们的代码可在 https://github.com/jyliu-ai/Seq-Drifting 获取。
cs.AI / 19 / 2610.12881
Mind the Spatial Gap: A Benchmark for Geospatial Reasoning and Constrained Spatial Editing
注意空间差距：地理空间推理与受限空间编辑的基准测试
Abstract
Understanding and manipulating spatial relations is fundamental to AI systems that reason about physical environments and construct grounded world models. However, existing evaluations of multimodal large language models (MLLMs) for spatial reasoning rely predominantly on static question-answering (QA) over small-scale or synthetic scenes, rarely probing whether models can modify spatial configurations while satisfying explicit geometric constraints. We introduce GeoSpatialBench, a benchmark designed to evaluate an escalating progression of spatial cognition, ranging from spatial relation understanding and geometric reasoning to constrained spatial editing. Grounded in real-world urban data across six geographically diverse cities, GeoSpatialBench evaluates a comprehensive taxonomy of 28 topological, directional, metric, and ordering relations across structured GeoJSON and map image modalities, with ground-truth references and constraints verified via GIS operations. Our empirical evaluation across frontier MLLMs uncovers a stark spatial gap: structured GeoJSON consistently outperforms map images, with multimodal combinations offering no reliable gain over GeoJSON alone, and while models show reasonable competence on qualitative topological and ordering, performance drops on quantitative metric relations. Crucially, editing performance on ordering relations declines relative to QA, and most failed edits violate at least one specified spatial constraint. These results highlight a gap between passively recognizing spatial relations and actively manipulating geometric configurations. GeoSpatialBench provides an open, rigorous diagnostic testbed to benchmark these spatial limitations, laying essential empirical groundwork for physically grounded world models and embodied spatial AI.
Chinese Translation
理解和操控空间关系是人工智能系统推理物理环境和构建扎根世界模型的基础。然而，现有的多模态大型语言模型（MLLMs）在空间推理方面的评估主要依赖于对小规模或合成场景的静态问答（QA），很少探讨模型是否能够在满足明确几何约束的情况下修改空间配置。我们提出了GeoSpatialBench，这是一个旨在评估空间认知逐步提升的基准测试，涵盖从空间关系理解和几何推理到受限空间编辑的各个方面。GeoSpatialBench基于六个地理多样化城市的真实城市数据，评估了28种拓扑、方向、度量和排序关系的全面分类，涉及结构化的GeoJSON和地图图像模态，且通过GIS操作验证了真实参考和约束。我们对前沿MLLMs的实证评估揭示了一个明显的空间差距：结构化的GeoJSON始终优于地图图像，而多模态组合在GeoJSON单独使用时并未提供可靠的增益。尽管模型在定性拓扑和排序方面表现出合理的能力，但在定量度量关系上的表现却下降。重要的是，排序关系的编辑性能相较于问答有所下降，大多数失败的编辑违反了至少一个指定的空间约束。这些结果突显了被动识别空间关系与主动操控几何配置之间的差距。GeoSpatialBench提供了一个开放、严格的诊断测试平台，以基准测试这些空间限制，为物理扎根的世界模型和具身空间人工智能奠定了重要的实证基础。
cs.AI / 20 / 2610.12914
OGPO: Offline Goal-conditioned Policy Optimization for Recoverable Vision-Language-Action Models
OGPO：可恢复视觉-语言-动作模型的离线目标条件策略优化
Abstract
Vision-language-action (VLA) models have shown strong promise in robotic manipulation, yet their post-training still predominantly relies on trajectory-level supervised imitation over robot demonstrations. While effective for learning feasible actions, this paradigm supervises policies to reproduce expert actions at each step, biasing them toward demonstration-specific action paths and providing limited guidance for recovering toward task-progress states once execution deviates from demonstrated trajectories. To address this limitation, we propose OGPO, an Offline Goal-conditioned Policy Optimization framework for recoverable VLA post-training. Instead of treating offline demonstrations merely as step-wise action labels, OGPO adaptively relabels trajectories with semantic key states and constructs goal-conditioned process rewards to optimize the policy for reaching these task-progress states. In this way, OGPO shifts VLA post-training from trajectory-level action imitation to key-state reachability learning, enabling policies to acquire a more robust ability to complete goal states from feasible off-demonstration states without requiring additional online interaction. Experiments on LIBERO and MetaWorld show that OGPO consistently improves VLA policy performance over standard supervised fine-tuning. Moreover, evaluations on LIBERO-Plus and our proposed LIBERO-ReAct, a mid-execution perturbation benchmark, further demonstrate that OGPO achieves stronger robustness and goal-directed recoverability under distribution shifts and recoverable off-demonstration states.
Chinese Translation
视觉-语言-动作（VLA）模型在机器人操作中展现出强大的潜力，但其后训练仍主要依赖于对机器人演示的轨迹级监督模仿。尽管这种方法在学习可行动作方面有效，但它监督策略在每一步重现专家动作，导致策略偏向于特定演示的动作路径，并在执行偏离演示轨迹后对恢复任务进展状态的指导有限。为了解决这一局限性，我们提出了OGPO，一种用于可恢复VLA后训练的离线目标条件策略优化框架。OGPO并不只是将离线演示视为逐步动作标签，而是自适应地用语义关键状态重新标记轨迹，并构建目标条件过程奖励，以优化策略以达到这些任务进展状态。通过这种方式，OGPO将VLA后训练从轨迹级动作模仿转变为关键状态可达性学习，使策略能够从可行的非演示状态中获得更强的能力，以完成目标状态，而无需额外的在线交互。在LIBERO和MetaWorld上的实验表明，OGPO在标准监督微调的基础上，始终提高了VLA策略的性能。此外，在LIBERO-Plus和我们提出的LIBERO-ReAct（一个中途执行扰动基准）上的评估进一步表明，OGPO在分布变化和可恢复的非演示状态下实现了更强的鲁棒性和目标导向的可恢复性。
cs.AI / 21 / 2610.12916
DeFiRank: Federated Pairwise Ranking for Governance in Decentralized Financial Networks
DeFiRank：去中心化金融网络治理的联合成对排名
Abstract
Existing governance approaches in decentralized financial networks rely on manual review processes conducted by human committees, which introduces subjective bias, creates a scalability bottleneck under high network volume, and fails to capture complex behavioural patterns and evolving network dynamics. To address these limitations, we propose \name, a federated learning-based pairwise ranking framework in which multiple institutions collaboratively train a ranking model without sharing their private transaction data. This collaborative learning enables the model to capture network-wide behavioural patterns that signify strong network participation and directly contribute to the network effect, including diversity of counterparties, usage of multiple applications, and frequency of transactions. The proposed system ranks providers based on the authenticity of their actions, promoting genuine contributors and demoting those exhibiting suspicious behaviour. We evaluate \name\ on a real-world blockchain transaction dataset. Results demonstrate that the federated approach achieves ranking performance modestly below the centralized baseline (NDCG@15: $\approx$0.87 vs.\ $\approx$0.92) while preserving data privacy and matching its ranking stability (Kendall $\tau$: $\approx$0.70 for both regimes), producing reliable governance decisions, including provider promotion and demotion (648 and 39 unique providers, respectively), without requiring centralized access to sensitive transaction data.
Chinese Translation
现有的去中心化金融网络治理方法依赖于由人工委员会进行的手动审查过程，这引入了主观偏见，在高网络流量下造成了可扩展性瓶颈，并未能捕捉复杂的行为模式和不断演变的网络动态。为了解决这些局限性，我们提出了
ame，一种基于联邦学习的成对排名框架，多个机构可以在不共享其私有交易数据的情况下协同训练排名模型。这种协作学习使模型能够捕捉网络范围内的行为模式，这些模式表明强大的网络参与度，并直接促进网络效应，包括交易对手的多样性、多个应用的使用以及交易频率。该系统根据提供者行为的真实性进行排名，促进真正的贡献者，并降低表现出可疑行为的提供者的排名。我们在一个真实的区块链交易数据集上评估了
ame。结果表明，联邦方法的排名性能略低于集中基线（NDCG@15: $ ext{约}$0.87 vs. ext{ $ ext{约}$}0.92），同时保持数据隐私并匹配其排名稳定性（Kendall $ au$: $ ext{约}$0.70，对于两种模式均如此），产生可靠的治理决策，包括提供者的提升和降级（分别为648和39个独特提供者），而无需集中访问敏感交易数据。
cs.AI / 22 / 2610.12917
AI-Mediation in Retail Markets: Revenue Effects and Assortment Optimization
零售市场中的人工智能中介：收入效应与商品组合优化
Abstract
AI shopping assistants increasingly influence which products consumers consider. We study retail assortment planning under AI mediation, where a retailer chooses a catalog, an AI intermediary forms a customer-facing shortlist, and the customer chooses from that shortlist according to a multinomial logit model. We consider both deterministic mediation, in which the AI follows a fixed segment-specific ranking, and stochastic mediation based on the Plackett-Luce model. Empirically, we construct a dataset of product rankings from leading AI models, estimate Plackett-Luce ranking models, and evaluate their predictive performance on held-out assortments. Combining these estimates with synthetic customer preferences, we find that accounting for AI mediation can materially change optimal assortments and revenue. Across the tested capacity configurations, ignoring mediation yields mean revenue losses between 2.52% and 11.37% relative to the PL optimum. A much simpler deterministic plug-in, based only on the ordering of fitted AI utilities, performs substantially better, with mean losses of at most 0.53%. Motivated by this performance, we study assortment optimization under deterministic mediation. We characterize its computational complexity and show that, for any fixed number of customer segments, the problem admits a fully polynomial-time approximation scheme.
Chinese Translation
人工智能购物助手越来越多地影响消费者考虑的产品。我们研究在人工智能中介下的零售商品组合规划，其中零售商选择一个目录，人工智能中介形成一个面向客户的短名单，客户根据多项式逻辑模型从该短名单中选择。我们考虑两种中介方式：确定性中介，即人工智能遵循固定的特定细分排名；以及基于 Plackett-Luce 模型的随机中介。在实证研究中，我们构建了来自领先人工智能模型的产品排名数据集，估计 Plackett-Luce 排名模型，并评估其在保留商品组合上的预测性能。将这些估计与合成客户偏好结合，我们发现考虑人工智能中介可以显著改变最佳商品组合和收入。在测试的容量配置中，忽略中介导致的平均收入损失在 2.52% 到 11.37% 之间，相对于 Plackett-Luce 最优解。一个更简单的确定性插件，仅基于拟合的人工智能效用的排序，表现得更好，平均损失最多为 0.53%。基于这一表现，我们研究了在确定性中介下的商品组合优化。我们描述了其计算复杂性，并表明，对于任何固定数量的客户细分，该问题承认一个完全多项式时间近似方案。
cs.AI / 23 / 2610.12921
Teaching neural decoders with language
用语言指导神经解码器
Abstract
Brain-computer interfaces decode neural activity, yet class-label supervision (CLS) leaves relationships among classes implicit. We propose Language as Supervision (LaS), which aligns neural features with natural-language class descriptions through learned projections and predicts classes using candidate text prototypes. We evaluate seven encoder architectures with independently trained models for visual, motor, and speech decoding from scalp and intracranial recordings. LaS improves speech-decoding accuracy by up to 17.15~pp over matched CLS baselines. Its representations support candidate-set reuse and label-supervised subject and session adaptation, while LaS enables conditional unseen-frequency decoding in steady-state visual evoked potential (SSVEP) responses. Correct class-description correspondence supports this generalization, and dual-sided attribution visualization shows neural and textual contributions to the same class-matching score. These findings suggest that text descriptions can serve as general-purpose references for learning and interpretation in neural decoding.
Chinese Translation
脑-计算机接口解码神经活动，但类标签监督（CLS）使得类之间的关系隐含。我们提出了语言作为监督（LaS），通过学习的投影将神经特征与自然语言类描述对齐，并使用候选文本原型预测类别。我们评估了七种编码器架构，独立训练模型用于从头皮和颅内记录中进行视觉、运动和语言解码。LaS在与匹配的CLS基线相比，提升了语音解码的准确性，最高可达17.15个百分点。其表示支持候选集重用和标签监督的受试者及会话适应，同时LaS使得在稳态视觉诱发电位（SSVEP）反应中实现条件未见频率解码成为可能。正确的类描述对应关系支持了这种泛化，双侧归因可视化展示了神经和文本对相同类匹配分数的贡献。这些发现表明，文本描述可以作为神经解码学习和解释的通用参考。
cs.AI / 24 / 2610.12927
Collaborative Learning and Reasoning over Model Slices
基于模型切片的协同学习与推理
Abstract
Large language models (LLMs) excel across diverse tasks, benefiting from knowledge priors, in-context reasoning, and test-time computation. Despite these strengths, single-model reasoning remains vulnerable to error propagation, as later steps depend on the model's earlier outputs, leaving reasoning blind spots difficult to independently reassess. To address this limitation, we develop \textit{collaborative learning}, encompassing the construction of heterogeneous collaborators and the learning of effective cross-model collaboration. Specifically, \textit{model slicing} replaces a large model with smaller, heterogeneous models under a comparable parameter budget, enabling prior reasoning to be reassessed across distinct model parameterizations. \textit{Pyramid supervision} combines protocol distillation with step-level supervision to refine reasoning and interaction decisions throughout collaboration. We evaluate our approach across four task categories: multi-hop question answering, mathematical reasoning, code generation, and open-ended constrained generation. The trained heterogeneous models achieve comparable or better task performance than the original large model and established multi-agent baselines, including reinforcement-learning-based methods, while also achieving higher inference efficiency than the multi-agent baselines and exhibiting differentiated collaboration patterns across tasks. These findings show that learned collaboration turns model heterogeneity into complementary reasoning capabilities.
Chinese Translation
大型语言模型（LLMs）在多种任务中表现出色，得益于知识先验、上下文推理和测试时计算。尽管具有这些优势，单一模型的推理仍然容易受到错误传播的影响，因为后续步骤依赖于模型早期的输出，这使得推理中的盲点难以独立重新评估。为了解决这一局限性，我们开发了 extit{协同学习}，涵盖异构协作者的构建和有效跨模型协作的学习。具体而言， extit{模型切片}在可比的参数预算下，用较小的异构模型替代大型模型，从而使得先前的推理能够在不同的模型参数化中重新评估。 extit{金字塔监督}结合协议蒸馏与步骤级监督，以在协作过程中细化推理和互动决策。我们在四个任务类别上评估了我们的方法：多跳问答、数学推理、代码生成和开放式约束生成。训练后的异构模型在任务性能上达到了与原大型模型和已建立的多智能体基线（包括基于强化学习的方法）相当或更好的效果，同时在推理效率上也优于多智能体基线，并在不同任务中展现出差异化的协作模式。这些发现表明，学习到的协作将模型异构性转化为互补的推理能力。
cs.AI / 25 / 2610.12933
Learning More by Training Less: Uncertainty-Guided Progressive Contraction for Efficient Reasoning Post-Training
通过减少训练学习更多：基于不确定性的渐进收缩用于高效推理后训练
Abstract
Large language models have achieved remarkable advances in complex reasoning. However, scaling reasoning post-training with ever-larger corpora can substantially increase training cost without proportional performance gains. In this study, we therefore ask: how far can reasoning supervision be compressed without sacrificing capability? To this end, we propose an uncertainty-guided progressive contraction framework that concentrates supervision and optimization on the reasoning trajectories most valuable for learning. Specifically, rather than treating a teacher-generated chain-of-thought (CoT) as a single realized path, we construct uncertainty-profiled reasoning supervision by building Monte Carlo reasoning trees around intermediate reasoning states to expose process-level uncertainty. This uncertainty first contracts the original corpus into a compact teacher-informed pool. A second, learner-aware contraction then performs online routing during SFT, concentrating optimization according to the student's evolving learning state. Together, the two stages progressively contract both supervision and optimization, substantially reducing the student training budget. Experiments across diverse reasoning benchmarks and LLM families show that our method uses as little as 0.22% of the optimization budget of full-data SFT, over 450x less training exposure, while maintaining or improving reasoning performance. Notably, our framework generalizes well across heterogeneous student families, black-box teachers, and weak-to-strong transfer. This indicates that it is not only a way to accelerate training but can also enhance reasoning capability.
Chinese Translation
大型语言模型在复杂推理方面取得了显著进展。然而，随着语料库规模的不断扩大，推理后训练的成本可能会大幅增加，而性能提升却不成比例。因此，在本研究中，我们提出了一个问题：在不牺牲能力的情况下，推理监督可以压缩到什么程度？为此，我们提出了一种基于不确定性的渐进收缩框架，集中监督和优化于最有价值的推理轨迹。具体而言，我们并不将教师生成的思维链（Chain-of-Thought, CoT）视为单一的实现路径，而是通过在中间推理状态周围构建蒙特卡洛推理树来构造不确定性特征的推理监督，以揭示过程级的不确定性。这种不确定性首先将原始语料库收缩为一个紧凑的教师信息池。接下来，学习者感知的收缩在监督微调（SFT）过程中进行在线路由，根据学生不断变化的学习状态集中优化。两个阶段共同逐步收缩监督和优化，显著减少学生的训练预算。在多种推理基准和大型语言模型（LLM）系列上的实验表明，我们的方法仅使用了完整数据SFT优化预算的0.22%，训练暴露量减少了450倍，同时保持或提高了推理性能。值得注意的是，我们的框架在异构学生系列、黑箱教师以及从弱到强的迁移中表现良好。这表明它不仅是一种加速训练的方法，还可以增强推理能力。
cs.AI / 26 / 2610.12954
MathVis: Can AI Reason through Drawing? A Benchmark for Mathematical Figure Generation
MathVis：人工智能能通过绘图进行推理吗？一个数学图形生成的基准测试
Abstract
This paper introduces MathVis, a benchmark designed to evaluate LLMs' ability to visualize frontier mathematical problems and perform diagram-assisted reasoning. The benchmark consists of 512 tasks, each comprising a mathematical statement, a reference diagram, executable TikZ or SVG code, a Gold Intermediate Representation (IR), atomic rubrics, and error codes. Through rigorous evaluation, GPT-6-Astra, Claude Fable 5, and Gemini 3.1 Pro Preview achieve pass@1 scores of 31.25%, 26.56%, and 22.85%, respectively. We found that although the tasks require visualization, LLMs produce outputs dominated by formal languages, symbols, and graphics programs, rather than explicit diagram-mediated reasoning. Furthermore, while the solutions generated by LLMs may contain all the necessary elements, they are often visually confusing to humans. Ultimately, this benchmark aims to visualize mathematical problems and promote research on enabling LLMs to simplify their reasoning via visual diagrams.
Chinese Translation
本文介绍了MathVis，这是一个旨在评估大型语言模型（LLMs）可视化前沿数学问题和进行图示辅助推理能力的基准测试。该基准测试由512个任务组成，每个任务包括一个数学陈述、一个参考图、可执行的TikZ或SVG代码、一个黄金中间表示（IR）、原子评分标准和错误代码。通过严格评估，GPT-6-Astra、Claude Fable 5和Gemini 3.1 Pro Preview的通过率（pass@1）分别达到了31.25%、26.56%和22.85%。我们发现，尽管这些任务需要可视化，LLMs生成的输出却主要以形式语言、符号和图形程序为主，而非明确的图示中介推理。此外，尽管LLMs生成的解决方案可能包含所有必要元素，但对人类而言，它们往往在视觉上令人困惑。最终，该基准测试旨在可视化数学问题，并促进研究，使LLMs能够通过视觉图示简化其推理过程。
cs.AI / 27 / 2610.12975
SmartSalesBench: A Resistance-Aware Multi-Turn Benchmark for Evaluating LLM Sales Agents
SmartSalesBench：一种关注抵抗的多轮基准，用于评估大型语言模型（LLM）销售代理
Abstract
We present SmartSalesBench, the first multi-turn conversational benchmark explicitly designed to evaluate large language model (LLM) sales agents against psychologically grounded, resistant buyer personas. Existing task-oriented dialogue benchmarks assume cooperative users and measure task completion; SmartSalesBench inverts this assumption, placing agents against personas programmed to resist and challenge across conversations of up to twenty turns. The benchmark includes five agent profiles, twelve buyer personas, fifteen smartphone products across six price tiers, and four controlled experiments with 3 iterations totaling 13,680 sessions. Both sales-agent and buyer persona generations are grounded through retrieval-augmented generation (RAG). We introduce nine evaluation metrics grouped into three tiers: primary outcome metrics (CR, BRS, RIR), secondary quality metrics (SMS, PFS, ETI), and exploratory metrics planned for future reporting (ITS, ARI, SPI). Our evaluation combines automated LLM-as-judge scoring with internal human assessment (n = 50, \k{appa} = 0.71, Pearson r = 0.81 with BRS).
Chinese Translation
我们提出了SmartSalesBench，这是第一个专门设计用于评估大型语言模型（LLM）销售代理的多轮对话基准，针对心理学基础的、具有抵抗性的买家角色。现有的任务导向对话基准假设用户是合作的，并测量任务完成情况；而SmartSalesBench则颠覆了这一假设，将代理与被编程为在多达二十轮对话中进行抵抗和挑战的人物进行对抗。该基准包括五个代理配置、十二个买家角色、六个价格层级的十五款智能手机产品，以及四个控制实验，包含3次迭代，总计13,680个会话。销售代理和买家角色的生成均通过检索增强生成（RAG）进行支持。我们引入了九个评估指标，分为三个层级：主要结果指标（CR, BRS, RIR）、次要质量指标（SMS, PFS, ETI）和计划用于未来报告的探索性指标（ITS, ARI, SPI）。我们的评估结合了自动化的LLM作为评审的评分与内部人类评估（n = 50, ext{kappa} = 0.71, Pearson r = 0.81与BRS）。
cs.AI / 28 / 2610.12977
Beyond Semantic IDs: Finite-Beam Reachability and Continuous-Query Generation in Generative Recommendation
超越语义ID：生成推荐中的有限束可达性与连续查询生成
Abstract
Generative recommendation (GR) of social media platforms commonly retrieves items by autoregressively decoding discrete Semantic IDs (SIDs) with finite-width beam search. This work examines how this output interface constrains whether held-out targets enter finite returned lists, while distinguishing catalog-wide scoreability from empirical top-K support. We introduce two continuous-query paradigms: CQG-Single generates one retrieval query in a single pass, and CQG-AR autoregressively generates continuous residual updates, retaining step-wise decoding while removing the SID item-token vocabulary, code lookup, and discrete beam expansion. These designs thereby eliminate the SID-specific reachability constraint induced by finite-beam decoding and avoid the technical challenge of aligning a discrete SID vocabulary with the pretrained backbone's vector space. Both methods make every encoded item directly scoreable through dense or approximate-nearest-neighbor retrieval. Under a unified protocol with valid-prefix constrained decoding, CQG-Single exceeds the TIGER-style baseline in all seven runs while using 55.3% as many trainable parameters. CQG-AR remains comparable on ML-1M and is numerically stronger on Beauty and Sports datasets. A controlled depth ablation further shows that additional continuous refinement does not monotonically improve shallow accuracy. These findings motivate decoupling output representation, autoregressive modeling, and retrieval budget when evaluating generative recommendation systems.
Chinese Translation
社交媒体平台的生成推荐（GR）通常通过自回归解码离散的语义ID（SIDs）并使用有限宽度的束搜索来检索项目。本研究考察了这一输出接口如何限制被保留目标进入有限返回列表的可能性，同时区分目录范围内的可评分性与经验上的前K支持。我们引入了两种连续查询范式：CQG-Single在单次传递中生成一个检索查询，而CQG-AR则自回归地生成连续的残差更新，保留逐步解码的过程，同时去除SID项目标记词汇、代码查找和离散束扩展。这些设计因此消除了由有限束解码引起的特定于SID的可达性约束，并避免了将离散SID词汇与预训练主干的向量空间对齐的技术挑战。这两种方法使每个编码项目都可以通过密集或近似最近邻检索直接评分。在一个统一的协议下，使用有效前缀约束解码，CQG-Single在所有七次运行中超越了TIGER风格的基线，同时使用的可训练参数减少了55.3%。CQG-AR在ML-1M上保持了可比性，并在Beauty和Sports数据集上表现出更强的数值优势。控制深度的消融实验进一步表明，额外的连续细化并不单调地提高浅层准确性。这些发现激励我们在评估生成推荐系统时解耦输出表示、自回归建模和检索预算。
cs.AI / 29 / 2610.12984
Work2Skill: Capturing Human GUI Experience for Reusable Agent Skill Induction
Work2Skill：捕捉人类图形用户界面体验以实现可重用的智能体技能引导
Abstract
Computer-use agents can perform increasingly complex GUI tasks, yet unfamiliar software and workflows requiring human domain knowledge remain challenging. Human demonstrations convey procedural knowledge through the same GUI interaction space, but reuse requires interpreting actions in context and separating workflow rules from task-specific details. We thus introduce Work2Skill, a framework that captures human GUI experience for reusable agent skill induction. WorkRecorder provides the framework's infrastructure for collecting this experience, aligning screen recordings, mouse and keyboard events, and available accessibility-tree information. Using WorkRecorder, we construct WorkDemo-350, the first synchronized multimodal dataset of human desktop GUI demonstrations for reusable agent skill induction and downstream evaluation. Its 350 demonstrations are screened for completeness and task success. We then propose an agentic skill-induction method that inspects evidence on demand to produce reusable skills capturing workflow rules, step dependencies, and completion checks. When the same four execution models are equipped with skills induced by each method under shared execution settings, Work2Skill achieves the highest scores and success rates. Mean score improvements over the corresponding No Skill baseline are 4.67--10.54 points on MyPCBench Bounded Action and 3.11--7.50 points on a 286-task subset of OSWorld-Verified. Our code and data are available at https://github.com/china-qijizhifeng/Work2Skill.
Chinese Translation
计算机使用代理能够执行越来越复杂的图形用户界面（GUI）任务，但对不熟悉的软件和工作流程的需求，以及需要人类领域知识的任务，仍然具有挑战性。人类演示通过相同的GUI交互空间传达程序性知识，但重用这些知识需要在上下文中解释动作，并将工作流程规则与特定任务细节分离。因此，我们提出了Work2Skill，一个捕捉人类GUI体验以实现可重用智能体技能引导的框架。WorkRecorder提供了该框架的基础设施，用于收集这些体验，协调屏幕录制、鼠标和键盘事件以及可用的可访问性树信息。利用WorkRecorder，我们构建了WorkDemo-350，这是第一个用于可重用智能体技能引导和后续评估的人类桌面GUI演示的同步多模态数据集。该数据集包含350个演示，经过完整性和任务成功率筛选。然后，我们提出了一种智能体技能引导方法，该方法根据需求检查证据，以生成捕捉工作流程规则、步骤依赖关系和完成检查的可重用技能。当相同的四个执行模型在共享执行设置下配备由每种方法引导的技能时，Work2Skill实现了最高的得分和成功率。在MyPCBench Bounded Action上，相应的无技能基线的平均得分提高了4.67至10.54分，在OSWorld-Verified的286任务子集上提高了3.11至7.50分。我们的代码和数据可在https://github.com/china-qijizhifeng/Work2Skill获取。
cs.AI / 30 / 2610.12997
Benchmarking Embedding Geometries and Clustering Algorithms for Unsupervised Patent Claim Classification
无监督专利权利要求分类的嵌入几何和聚类算法基准测试
Abstract
Prior-art retrieval systems are structurally organized around classification codes, making cross-code discovery difficult. This paper asks if an unsupervised representation trained on patent text can recover relevant connections that code-restricted search cannot, and if hyperbolic space helps. We trained a hyperbolic and a matched Euclidean Variational Autoencoder (VAE) on 45,000 US patents (96 runs, 6 seeds). The hyperbolic model retrieves genuinely cross-category prior art 52% more often than the Euclidean model (18.76% vs. 12.34% of top-10 results, 3,000 queries). This represents a practical gain for the motivating search problem. The hyperbolic model agrees less with official classification (NMI 0.2392 vs. 0.3206), which we explain geometrically: the two measures capture different goals, and only one aligns with prior-art search needs. In a study of 10 technology domains (50,000 more patents), we find that a domain's classification breadth--the number of distinct CPC subgroups spanned--predicts approach performance (Pearson r=-0.642, p=0.045, N=10), and that patents per subgroup is an even stronger predictor (r=0.715, p=0.020). This provides a checkable signal for practitioners to determine where the method is most valuable. We report all findings with statistical transparency and release a complete, 91-test automated implementation.
Chinese Translation
现有的专利检索系统在结构上围绕分类代码组织，这使得跨代码发现变得困难。本文探讨了基于专利文本训练的无监督表示是否能够恢复代码限制搜索无法发现的相关连接，以及双曲空间是否有助于此。我们在45,000项美国专利上训练了一个双曲和一个匹配的欧几里得变分自编码器（Variational Autoencoder, VAE）（96次实验，6个种子）。双曲模型在检索真正跨类别的现有技术方面比欧几里得模型更频繁（比率为18.76%对12.34%，基于3,000个查询的前10个结果）。这为激励搜索问题提供了实际收益。双曲模型与官方分类的吻合度较低（NMI为0.2392对0.3206），我们从几何角度进行了解释：这两种度量捕捉了不同的目标，只有一个与现有技术搜索需求相一致。在对10个技术领域（额外50,000项专利）的研究中，我们发现一个领域的分类广度——所涵盖的不同CPC子组的数量——可以预测方法的表现（Pearson r=-0.642，p=0.045，N=10），而每个子组的专利数量是一个更强的预测因子（r=0.715，p=0.020）。这为从业者提供了一个可验证的信号，以确定该方法最有价值的地方。我们以统计透明度报告所有发现，并发布了一个完整的91测试自动化实现。
cs.AI / 31 / 2610.13000
Hindsight Structured Advice Distillation: Steering Executor Agents with Advisors Trained on Their Failures and Task Characteristics
事后结构化建议蒸馏：通过基于失败和任务特征训练的顾问引导执行代理
Abstract
Language-model agents are used for specialized work such as financial research, healthcare and law. Such agents are often capable enough for a task but still miss its details. In finance, for example, these details include which filing reports a figure or how a company discloses a metric. Retraining the agent to learn these details is expensive, and often impossible when the agent is a closed or shared system. Writing the missing know-how into the agent's prompt makes the prompt long and identical for every input, although different inputs need different guidance. We propose to learn to generate this guidance instead. A small advisor model reads each input and writes structured advice, which is appended to the prompt of a frozen executor. The advisor is trained with hindsight, following work on distillation with privileged information. On a training pool with graded answers, we analyse the executor's failures, run it under several strategies to see which one worked on each input, and let a teacher that sees the grades and the traces write the advice that would have helped. The advisor then learns, by supervised fine-tuning, to write that advice from the input alone. Because the student is a separate small model rather than the executor, it is cheap to train and adds little latency. We call this approach hindsight structured advice distillation (HSAD). On FrontierFinance, a public benchmark of expert-written financial research queries, an 8B advisor raises the qualification rate of Samaya's proprietary research agent, which already leads the benchmark's leaderboard, from 52.9% to 56.6%. This is above the agent's own high-effort configuration, at 37% lower cost per query. We study the generalization of the proposed approach, and in particular show that HSAD also improves results on WebShop, BIRD text-to-SQL and ExpertLongBench.
Chinese Translation
语言模型代理被用于金融研究、医疗保健和法律等专业工作。这些代理通常在某项任务上具有足够的能力，但仍然会遗漏细节。例如，在金融领域，这些细节包括某个数字所对应的报告文件或公司如何披露某个指标。重新训练代理以学习这些细节成本高昂，并且当代理是封闭或共享系统时，往往是不可能的。将缺失的知识写入代理的提示中会使提示变得冗长且对每个输入都相同，尽管不同的输入需要不同的指导。我们提出学习生成这种指导的方法。一个小型顾问模型读取每个输入并写出结构化建议，这些建议被附加到一个冻结的执行器的提示中。顾问通过事后训练，遵循使用特权信息进行蒸馏的工作。在一个带有分级答案的训练池中，我们分析执行器的失败，采用多种策略运行它，以查看哪种策略在每个输入上有效，并让一个能够看到分数和痕迹的教师写出能够提供帮助的建议。然后，顾问通过监督微调学习仅根据输入写出该建议。由于学生是一个独立的小模型而不是执行器，因此训练成本低且延迟很小。我们称这种方法为事后结构化建议蒸馏（Hindsight Structured Advice Distillation，HSAD）。在FrontierFinance这一公共基准的专家撰写金融研究查询中，一个8B的顾问将Samaya的专有研究代理的合格率从52.9%提高到56.6%，该代理已经在基准的排行榜上领先。这一提升是在每个查询的成本降低37%的情况下，超过了代理自身的高努力配置。我们研究了所提方法的泛化能力，特别表明HSAD也改善了WebShop、BIRD文本到SQL和ExpertLongBench的结果。
cs.AI / 32 / 2610.13002
Scaling Empowered Agents
扩展赋能智能体
Abstract
Empowerment measures the capacity of an agent to actively control its environment. In this work, we explain the connection between skill learning and empowerment. While skill learning is not empowerment, we introduce a simple scalable recipe for empowerment maximization: empowerment corresponds to the number of executable skills from a state, so maximizing empowerment corresponds to reaching states that are good starting points for many skills. Our recipe applies to arbitrary underlying skill learning methods. Whereas most prior work on empowerment looks at tabular settings, this recipe enables us to study empowerment at a scale where we can realize theoretically-predicted but empirically-elusive properties of empowerment (e.g., gathering resources, environment modification, survival). Experiments show that empowerment maximization unlocks interesting emergent properties, such as building a house in Crafter. Website and code at \href{https://empowerment-geometry.github.io/scaling-empowerment/}{this https url}.
Chinese Translation
赋能衡量了智能体主动控制其环境的能力。在本研究中，我们解释了技能学习与赋能之间的关系。虽然技能学习并不等同于赋能，但我们提出了一种简单的可扩展赋能最大化方法：赋能对应于从某一状态出发可以执行的技能数量，因此最大化赋能对应于达到那些适合多项技能的良好起始状态。我们的方法适用于任意基础的技能学习方法。尽管以往大多数关于赋能的研究集中在表格设置上，但该方法使我们能够在一个可以实现理论预测但在经验上难以捕捉的赋能特性（例如，资源收集、环境修改、生存）规模上进行研究。实验表明，赋能最大化解锁了有趣的涌现特性，例如在Crafter中建造房屋。相关网站和代码可在[此链接](https://empowerment-geometry.github.io/scaling-empowerment/)找到。
cs.AI / 33 / 2610.13008
Provenance-Aware Belief Updating in Human-AI Feedback Loops
关注来源的信念更新在人机反馈循环中的应用
Abstract
In repeated human-AI interaction, a single observation can be quoted, summarized, and returned to a model as multiple reports, making repetition look like independent corroboration. We study whether language models can use supplied report lineage to update according to evidence provenance rather than report count. We introduce Latent Quotient Evidence Training (LQET), a relational objective that combines sparse Bayesian supervision with constraints that encourage stable predictions when reports add no new evidence and consistent updates to the same new evidence across relay histories. On a controlled binary benchmark, LQET reduces new-evidence update RMSE by 41-44% and log-odds drift after eight relay steps by 89-91% relative to task-only training across Qwen3-4B and Phi-4-mini, with three paired training seeds per backbone. The gains extend to a closed-loop simulation in which model confidence controls how many relayed reports re-enter the model's context in later rounds. By grounding belief updates in supplied provenance, LQET offers a route toward dynamic human-AI systems in which influence follows new evidence, not the number of times the same evidence returns.
Chinese Translation
在重复的人机交互中，单一观察可以被引用、总结，并作为多个报告返回给模型，使得重复看起来像是独立的证实。我们研究语言模型是否能够利用提供的报告来源根据证据的来源而非报告数量进行更新。我们引入了潜在商证据训练（Latent Quotient Evidence Training, LQET），这是一种关系目标，结合了稀疏贝叶斯监督与约束，鼓励在报告未提供新证据时保持稳定的预测，并在不同的传递历史中对相同的新证据进行一致的更新。在一个受控的二元基准测试中，LQET相较于仅基于任务的训练，在Qwen3-4B和Phi-4-mini模型上减少了41-44%的新证据更新均方根误差（RMSE），并在经过八次传递步骤后减少了89-91%的对数几率漂移，每个基础模型有三个配对的训练种子。这些收益扩展到一个闭环模拟中，其中模型的置信度控制着在后续回合中多少传递报告重新进入模型的上下文。通过将信念更新基于提供的来源，LQET为动态人机系统提供了一条途径，使影响力跟随新证据，而不是相同证据返回的次数。
cs.AI / 34 / 2610.13021
EmbodiedCLBench: Evaluating Continual Learning for Self-Evolving Embodied Agents
EmbodiedCLBench：评估自我进化具身智能体的持续学习能力
Abstract
Continual learning requires models to accumulate past experience and transfer it to new tasks. Despite substantial research on continual learning, existing work remains largely confined to non-embodied settings, without extending to embodied environments. To address this gap, we introduce EmbodiedCLBench, the first benchmark in embodied simulation for evaluating the continual learning capability of self-evolving agents. With compositional generalization as the core evaluation principle, our benchmark evaluates whether agents can first learn from basic tasks and then recombine the acquired skills to solve complex, unseen advanced tasks. Based on EmbodiedCLBench, we conduct extensive experiments across multiple agent harnesses and learning methods, revealing consistent limitations and distinctive behaviors. Our results show that while advanced tasks can be hardly solved under zero-shot setting, the experience from basic tasks enables consistent improvement. Such benefit from the experience remains stable even if the task complexity increases. However, this improvement does not scale accordingly with the number of learning samples, and current agents cannot extract meaningful information from additional experience. Collectively, our findings highlight continual learning as a fundamental yet underexplored capability, and our benchmark offers valuable resources for designing effective self-evolving agents in embodied environments. Our code and dataset are available at https://pku-value-lab.github.io/EmbodiedCLBench-homepage.
Chinese Translation
持续学习要求模型积累过去的经验并将其转移到新任务上。尽管在持续学习方面已有大量研究，但现有工作仍主要局限于非具身环境，未能扩展到具身环境。为了解决这一空白，我们引入了EmbodiedCLBench，这是第一个用于评估自我进化智能体持续学习能力的具身仿真基准。以组合泛化作为核心评估原则，我们的基准评估智能体是否能够首先从基本任务中学习，然后重新组合所获得的技能以解决复杂的、未见过的高级任务。基于EmbodiedCLBench，我们在多个智能体框架和学习方法上进行了广泛实验，揭示了一致的局限性和独特的行为。我们的结果表明，尽管在零-shot 设置下几乎无法解决高级任务，但来自基本任务的经验能够带来持续的改进。这种经验带来的好处在任务复杂性增加时仍然保持稳定。然而，这种改进并未随着学习样本数量的增加而相应扩大，目前的智能体无法从额外的经验中提取有意义的信息。总体而言，我们的发现强调了持续学习作为一种基本但未被充分探索的能力，而我们的基准为在具身环境中设计有效的自我进化智能体提供了宝贵的资源。我们的代码和数据集可在 https://pku-value-lab.github.io/EmbodiedCLBench-homepage 获取。
cs.AI / 35 / 2610.13058
Travel Back To Move Forward: Retrospective Exploration and Evolving for Long Horizon Agents
回溯以前进：长时间跨度智能体的回顾性探索与演变
Abstract
Large language model agents increasingly operate in long-horizon environments where early mistakes can trigger cascading failures and trap agents in dead-end trajectories. Agents must therefore be able to proactively recover from such states and turn their exploratory experience into guidance for subsequent decisions. We introduce ReTRAIL, a framework that enables agents to rewind the environment to earlier states while preserving valid progress and carrying forward knowledge learned from abandoned exploration. The environment state moves backward while the agent's knowledge continues to move forward. ReTRAIL trains a shared policy with two complementary signals: State-Anchored GRPO compares alternative continuations from revisited states to improve execution, while a branch-advancement reward guides reflection through the progress of subsequent exploration. Experiments across four interactive benchmark MineSweeper, Sokoban, ALFWorld and ScienceWorld show that simply adding a rewind primitive provides limited benefit and can even degrade performance. In contrast, ReTRAIL enables agents to rewind from unproductive states and use insights from abandoned branches to guide subsequent exploration. These results demonstrate the value of coupling environment rewinding with experience reuse within an ongoing task for long-horizon task solving. The code is available at https://github.com/whi497/ReTRAIL.
Chinese Translation
大型语言模型智能体越来越多地在长时间跨度的环境中运行，在这些环境中，早期的错误可能会触发级联失败，并将智能体困在死胡同轨迹中。因此，智能体必须能够主动从这些状态中恢复，并将其探索经验转化为后续决策的指导。我们提出了ReTRAIL，一个框架，使智能体能够将环境回溯到早期状态，同时保留有效的进展并传递从放弃的探索中学到的知识。环境状态向后移动，而智能体的知识则继续向前发展。ReTRAIL通过两个互补信号训练共享策略：状态锚定的GRPO（State-Anchored GRPO）比较从回访状态出发的替代继续，以改善执行，而分支推进奖励（branch-advancement reward）则通过后续探索的进展引导反思。在四个交互式基准测试MineSweeper、Sokoban、ALFWorld和ScienceWorld上的实验表明，仅仅添加一个回溯原语的效果有限，甚至可能导致性能下降。相比之下，ReTRAIL使智能体能够从无效状态中回溯，并利用放弃分支的见解来指导后续探索。这些结果证明了在长时间跨度任务解决中，将环境回溯与经验重用结合的价值。代码可在 https://github.com/whi497/ReTRAIL 获取。
cs.AI / 36 / 2610.13075
LongHarness: Advancing Agentic Reasoning over 100M-Token Corpora via Stateful Evidence Management
LongHarness：通过状态证据管理推进超过1亿标记语料库的代理推理
Abstract
Corpus-level long-context reasoning requires agents to acquire and integrate evidence across document collections that exceed their context windows. Existing agent harnesses enable incremental access to these corpora, but acquired evidence and evolving conclusions often remain entangled in a growing interaction history. This implicit state management can lead to two coupled failure modes: unbounded exploration, where retrieval expands the interaction history without resolving decision-relevant information gaps, and evidence-state drift, where previously acquired evidence or subsequent corrections fail to inform the current conclusion. We present LongHarness, a model-agnostic harness that makes the interaction between acquired evidence and evolving conclusions explicit. Its core mechanism, Evidence-Conclusion Coevolution, uses unresolved conclusions to guide exploration toward missing information, while newly acquired evidence strengthens or revises the conclusions that directed the search. LongHarness couples this process with Evidence Lifecycle Management, which keeps accumulated evidence current and traceable throughout the interaction. Together, these mechanisms focus retrieval on remaining information needs and keep conclusions aligned with current evidence. We evaluate LongHarness on OfficeQA Pro V1 and V2, AA-LCR, and LongBench v2 using GLM-5.3-Flash and Qwen3.8-Flash. Compared with direct reasoning using the same backbone, LongHarness improves GLM-5.3-Flash exact match on OfficeQA Pro from 15.8% to 60.2% on V1 and from 4.4% to 61.1% on V2. Against state-of-the-art agent harnesses, it surpasses Claude Code by 12.0 percentage points on V1 and Codex by 18.9 points on V2. With Qwen3.8-Flash, LongHarness reaches 61.7% on V1, outperforming Claude Code by 8.3 points and Codex by 12.0 points.
Chinese Translation
语料级长上下文推理要求代理在超出其上下文窗口的文档集合中获取和整合证据。现有的代理工具使得对这些语料库的增量访问成为可能，但获取的证据和不断演变的结论往往在不断增长的交互历史中交织在一起。这种隐式状态管理可能导致两种相互关联的失败模式：无界探索，其中检索扩展了交互历史而未能解决与决策相关的信息缺口；以及证据状态漂移，其中先前获取的证据或后续修正未能为当前结论提供信息。我们提出了LongHarness，这是一种模型无关的工具，使得获取的证据与不断演变的结论之间的交互变得明确。其核心机制，证据-结论共演化，利用未解决的结论引导探索缺失的信息，而新获取的证据则加强或修正引导搜索的结论。LongHarness将这一过程与证据生命周期管理相结合，确保累积的证据在整个交互过程中保持最新和可追溯。通过这些机制，LongHarness将检索集中在剩余的信息需求上，并保持结论与当前证据的一致性。我们在OfficeQA Pro V1和V2、AA-LCR以及LongBench v2上使用GLM-5.3-Flash和Qwen3.8-Flash评估了LongHarness。与使用相同骨干网络的直接推理相比，LongHarness将GLM-5.3-Flash在OfficeQA Pro V1上的准确匹配率从15.8%提高到60.2%，在V2上从4.4%提高到61.1%。与最先进的代理工具相比，它在V1上超越Claude Code 12.0个百分点，在V2上超越Codex 18.9个百分点。使用Qwen3.8-Flash时，LongHarness在V1上达到了61.7%，超越Claude Code 8.3个百分点，超越Codex 12.0个百分点。
cs.AI / 37 / 2610.13078
Attention Sees, MLP Refuses: Tracing the Attention-to-MLP Safety Pathway in LLMs
注意力可见，MLP拒绝：追踪大型语言模型中的注意力到MLP的安全路径
Abstract
Safety-aligned large language models (LLMs) remain vulnerable to jailbreaks and can over-refuse benign requests. Understanding these failures requires distinguishing harmfulness recognition from the computation that writes refusal. We reframe this as a pathway question and make it measurable on residual-stream updates in decoder-only Transformers, separating the harmfulness information a module update exposes from the change it produces in readable refusal evidence. Across 18 Base-Instruct pairs spanning Qwen, Llama, and Gemma, harmfulness is decoded more readily from attention updates, particularly in early layers, whereas MLP updates contribute more strongly to harm-selective refusal writing in aligned models. Post-training strengthens MLP refusal writing in all 18 pairs without systematically improving harmfulness decodability, while positive MLP writing becomes more concentrated in deeper layers after post-training. Across six reasoning-enabled models, harmfulness remains decodable during reasoning while strong refusal writing concentrates at answer onset, a pattern that persists with reasoning disabled. Causal interventions further support this pathway: resetting downstream MLP updates to their clean values substantially reduces the refusal shift caused by upstream attention interventions. Building on these findings, we introduce Sentinel, which combines an attention-derived perception score P with an MLP-derived decision score D. Their joint profiles characterize jailbreaks and over-refusals, and conditional routing with both signals achieves a lower aggregate error rate than single-signal routing and evaluated fixed reminders across eight models.
Chinese Translation
安全对齐的大型语言模型（LLMs）仍然容易受到越狱攻击，并且可能过度拒绝良性请求。理解这些失败需要将有害性识别与写拒绝的计算区分开来。我们将其重新构建为一个路径问题，并在仅解码器的Transformer中通过残余流更新进行可测量，分离出模块更新所暴露的有害性信息与其在可读拒绝证据中产生的变化。在涵盖Qwen、Llama和Gemma的18对基础指令中，有害性更容易从注意力更新中解码，特别是在早期层，而MLP更新在对齐模型中更强烈地促进了有害选择性拒绝的写作。后训练在所有18对中增强了MLP拒绝写作，而没有系统性地改善有害性解码能力，同时正向MLP写作在后训练后变得更加集中于深层。对于六个启用推理的模型，在推理过程中有害性仍然可以解码，而强烈的拒绝写作则集中在答案开始时，这一模式在禁用推理时依然存在。因果干预进一步支持这一路径：将下游MLP更新重置为其干净值显著减少了上游注意力干预所造成的拒绝偏移。基于这些发现，我们引入了Sentinel，它结合了基于注意力的感知得分P与基于MLP的决策得分D。它们的联合特征描述了越狱和过度拒绝，并且使用这两个信号的条件路由实现了比单信号路由和评估固定提醒在八个模型中的总错误率更低。
cs.AI / 38 / 2610.13081
SWEAR: Section-Wise Efficient Audio Reasoning via Self-Rated Difficulty
SWEAR：基于自我评估难度的分段高效音频推理
Abstract
Recent work on large audio-language models (LALMs) explores adaptive reasoning to balance accuracy and efficiency. However, existing approaches often rely on coarse-grained difficulty estimation, overlooking the diverse factors that determine reasoning effort. We address this limitation by deriving five-dimensional difficulty scores and compressed reasoning traces from complementary offline assessments. Based on this characterization, we propose SWEAR, a difficulty-aware reasoning framework that learns difficulty signals during supervised fine-tuning and uses reinforcement learning to adapt reasoning allocation across reasoning stages. Experiments on MMAU-test-mini and MMAR show that SWEAR improves accuracy while substantially reducing generated tokens, shortening reasoning for easy instances without sacrificing performance on difficult ones.
Chinese Translation
近期关于大型音频语言模型（LALMs）的研究探讨了自适应推理，以平衡准确性和效率。然而，现有方法往往依赖于粗略的难度估计，忽视了决定推理努力的多样化因素。我们通过从互补的离线评估中推导出五维难度分数和压缩推理轨迹，来解决这一局限性。基于这一特征，我们提出了SWEAR，一个难度感知的推理框架，该框架在监督微调过程中学习难度信号，并利用强化学习在推理阶段之间调整推理分配。在MMAU-test-mini和MMAR上的实验表明，SWEAR在显著减少生成的标记数的同时提高了准确性，缩短了对简单实例的推理时间，而没有牺牲对困难实例的性能。
cs.AI / 39 / 2610.13092
When Faster Decisions Do Not Make Better Agents: An Empirical Study of Jev in Agent Workflows
快速决策并不总能提升代理性能：关于Jev在代理工作流程中的实证研究
Abstract
AI agents depend on repeated decisions to accomplish tasks, with large language models (LLMs) commonly serving as their core decision makers. The recent emergence of Jev offers an opportunity to make these decisions faster and at lower cost. However, it remains unclear whether such local savings can be translated into efficient and reliable task completion. In this paper, we investigate this efficiency transfer across five agents in two representative application settings, embodied tasks in ALFWorld and mobile app use in AndroidWorld. Across 4,500 task executions, we jointly evaluate task effectiveness, decision-level and task-level resource use, and execution behavior. Experimental results show that Jev integration substantially reduces decision cost and latency across all five agents, with task-level gains varying by workflow. Task success remains close to baseline levels on ALFWorld and declines more substantially in two mobile agents. Auxiliary reasoning and additional interactions absorb decision savings, while failed recovery and premature completion limit successful delivery. Mean cost per successful outcome increases in one agent even as spending per attempt falls. These findings highlight the importance of preserving progress, supporting recovery, and verifying completion when integrating efficient decision models into agent workflows. Their practical value should be assessed through successful task delivery, recovery behavior, and resources per delivered outcome.
Chinese Translation
人工智能代理依赖于重复决策来完成任务，大型语言模型（LLMs）通常作为其核心决策者。Jev的近期出现为加快决策速度和降低成本提供了机会。然而，目前尚不清楚这种局部节省是否能够转化为高效且可靠的任务完成。在本文中，我们研究了在两个代表性应用场景中，五个代理之间的效率转移，这两个场景分别是ALFWorld中的具身任务和AndroidWorld中的移动应用使用。在4500次任务执行中，我们共同评估了任务有效性、决策层级和任务层级的资源使用以及执行行为。实验结果表明，Jev的整合显著降低了所有五个代理的决策成本和延迟，但任务层级的收益因工作流程而异。在ALFWorld中，任务成功率接近基线水平，而在两个移动代理中则显著下降。辅助推理和额外交互吸收了决策节省，而失败的恢复和过早的完成限制了成功交付。在一个代理中，尽管每次尝试的支出减少，成功结果的平均成本却有所增加。这些发现强调了在将高效决策模型整合到代理工作流程中时，保持进展、支持恢复和验证完成的重要性。其实际价值应通过成功的任务交付、恢复行为和每个交付结果的资源使用进行评估。
cs.AI / 40 / 2610.13098
VETR: Execution-Grounded Self-Improving Agent for Symbolic Graphics Reasoning
VETR：基于执行的自我改进符号图形推理代理
Abstract
Visual agents for symbolic graphics reasoning depend on executable world representations, yet continued improvement remains bottlenecked by ambiguous supervision from final renders. Recent agentic systems refine scenes through repeated interaction, but they lack mechanisms to recursively extract actionable evidence from execution traces, overlooking which proposed symbolic edits survived, disappeared, or interfered during program execution. To address this, we propose VETR (Verified Execution Trajectories for Refinement), a framework that enables agentic recursive self-improvement over executable world representations for symbolic graphics reasoning. Within each task, VETR retains aligned program/render records with bounded runtime observations and introduces an execution refiner that compiles observed writes and state changes into constrained alternative programs. The refiner validates their endpoints and preservation conditions against captured runtime state; the ordinary visual verifier then assesses goal relevance. Repair construction makes no additional language-model calls, but can require additional executions. Evaluations on BlenderGym (280 tasks) and BlenderBench (27 released tasks) with GPT-5.6-luna and Qwen3.8-27B show that VETR consistently outperforms existing refinement strategies, including BlenderAlchemy and VIGA. Controlled ablations evaluate the contributions of execution refinement and role-asymmetric evidence organization under fixed generator-interaction caps.
Chinese Translation
用于符号图形推理的视觉代理依赖于可执行的世界表示，然而持续改进仍然受到最终渲染模糊监督的瓶颈。近期的代理系统通过反复交互来优化场景，但缺乏从执行轨迹中递归提取可操作证据的机制，忽视了在程序执行过程中哪些提议的符号编辑得以保留、消失或相互干扰。为了解决这一问题，我们提出了VETR（Verified Execution Trajectories for Refinement），一个框架，使得代理能够在可执行的世界表示上进行递归自我改进，以实现符号图形推理。在每个任务中，VETR保留了与有限运行时观察相对齐的程序/渲染记录，并引入了一个执行优化器，将观察到的写入和状态变化编译成受限的替代程序。该优化器根据捕获的运行时状态验证其端点和保留条件；普通的视觉验证器随后评估目标相关性。修复构建不需要额外的语言模型调用，但可能需要额外的执行。在BlenderGym（280个任务）和BlenderBench（27个已发布任务）上使用GPT-5.6-luna和Qwen3.8-27B进行的评估表明，VETR始终优于现有的优化策略，包括BlenderAlchemy和VIGA。控制消融实验评估了在固定生成器交互限制下执行优化和角色不对称证据组织的贡献。
cs.AI / 41 / 2610.13112
Does Location Matter? A Black-Box Causal Audit Reveals Model-Dependent Audio Grounding in LALM Hallucination
位置重要吗？黑箱因果审计揭示了LALM幻觉中模型依赖的音频定位
Abstract
Large audio-language models (LALMs) frequently exhibit object hallucination, affirming absent sounds. The dominant mitigation, Audio-Aware Decoding (AAD), contrasts logits under real audio against an all-zero "blank" waveform, flagged by prior review as out-of-distribution. We introduce a black-box causal test that isolates the claimed evidence itself: localize a claim's evidence region via CLAP similarity, remove it by spectral gating, and compare against a random-window control across a 2x2x2 grid on three LALMs. On Qwen2-Audio-7B, localized ablation's raw AUROC trails blank's (0.613 vs. 0.721); this is a duration confound, not a localization failure, since blank removes the entire clip versus the grid's 2-4 s. Holding duration fixed, localization beats random removal at every cell on both models. Raw detection value remains model-dependent, especially on Audio Flamingo 3. The same evidence corrects decoding: contrastive decoding combining blank and language-prior contrasts, gated and ensembled, raises Qwen2-Audio accuracy from 57.1% to 71.9%, beating AAD alone. A unified causal-grounding profile does not transfer between models, indicating different grounding regimes. The source code and a detailed report of our findings, including identified limitations, are available at https://github.com/khalooei/causa.
Chinese Translation
大型音频语言模型（LALMs）经常表现出物体幻觉，确认缺失的声音。主要的缓解方法是音频感知解码（Audio-Aware Decoding，AAD），它将真实音频下的logits与全零的“空白”波形进行对比，之前的审查将其标记为分布外。我们引入了一种黑箱因果测试，孤立所声称的证据：通过CLAP相似性定位证据区域，通过谱门控去除该区域，并在三个LALMs的2x2x2网格上与随机窗口对照进行比较。在Qwen2-Audio-7B上，局部消融的原始AUROC落后于空白（0.613对0.721）；这是一种持续时间混淆，而非定位失败，因为空白去除了整个片段，而网格仅为2-4秒。在固定持续时间的情况下，定位在两个模型的每个单元上均优于随机去除。原始检测值仍然依赖于模型，尤其是在Audio Flamingo 3上。同样的证据修正了解码：对比解码结合了空白和语言先验的对比，经过门控和集成，提升了Qwen2-Audio的准确率从57.1%到71.9%，超越了单独的AAD。统一的因果定位特征在模型之间并不转移，表明不同的定位机制。我们的源代码和详细的发现报告，包括识别的局限性，已在https://github.com/khalooei/causa上提供。
cs.AI / 42 / 2610.13116
DRYLAB-Bench: Benchmarking the Biosecurity Risk of Agentic LLM Mutation Design
Abstract
As large language models (LLMs) acquire advanced biological design capabilities, they offer substantial research benefits but simultaneously introduce severe dual-use risks, such as engineering enhanced pathogens or drug-resistant bacteria. While biosecurity frameworks recognize these threats, the actual risk of LLMs executing dangerous biological modifications remains unquantified, particularly when augmented with specialized biological design tools (BDTs). Existing evaluations primarily assess hazardous knowledge possession rather than practical application, overlook risk amplification from BDTs, and rely on unreliable, expensive, or hazardous validation methods. To address these gaps, we introduce DRYLAB-Bench (Design Risk Yielded by LLM Agent in Biodesign), the first comprehensive benchmark evaluating the high-risk biological mutation design capabilities of LLM agents. Moving beyond basic question-answering, DRYLAB-Bench tasks agents with 30 genuinely hazardous mutation-design scenarios distilled from biosecurity guidelines. It evaluates models in a multi-round, tool-augmented environment where LLMs orchestrate BDTs to iteratively refine candidates. Crucially, it scores designs directly against 187,695 empirical wet-lab measurements from ProteinGym and ViroGym using our proposed Biosecurity Risk Score (BRS), quantifying risk-associated effects while enabling scalable and safe ground-truth verification. Evaluations of 15 frontier models reveal substantial baseline threats: without tools, 67.6% of one-shot designs exceed the average measured variant, and 21.4% reach the highest-risk decile. Equipping models with BDTs significantly amplifies this risk, raising the pooled risk by \(\Delta BRS=+0.116\) and the top-decile share to 23.5%. Code: https://github.com/drylab-bio/DRYLAB-Bench
cs.AI / 43 / 2610.13128
Locate, Judge, Read: Role-Specialized Tool Agents over a Knowledge-Point Index for Open-Domain Multimodal Document QA
定位、判断、阅读：基于知识点索引的角色专用工具代理用于开放域多模态文档问答
Abstract
Question answering over large heterogeneous document collections requires first finding the right passages and then reading them, yet current retrieval-augmented agents give one model both kinds of tools and let it schedule itself. We present KnowForge, a pipeline separated by job: a hybrid-retrieval knowledge-point index; a Locator agent that can only search; a code-scheduled bridge judge that forces the missing second hop; a typed, size-bounded evidence pool that merges every search's ranking; a Deep Reader agent that can only read text, images and tables; and a synthesizer with a format checker. With a self-hosted Qwen3.6-35B-A3B (3B active), KnowForge reaches F1 47.5 on the full open-domain M3DocVQA (2,441 questions, 3,368 PDFs), +11.0 over the published M3DocRAG system and +7.1 over one-shot retrieve-and-answer on the same index and model, and 53.5 with DeepSeek-V4-Flash. On MultiHop-RAG it reaches 73.5 accuracy, above the strongest published RAG and GraphRAG results (71.2); on the single-document MMLongBench-Doc it lifts the same model from 34.8 to 47.0, level with GPT-4o reading page images (46.3). Per-job ablations on the full benchmarks attribute the gain: the Deep Reader is worth 3.8, 9.8 and 0 points on the three corpora, tracking how visual each is; merging both agents into one with all tools and twice the rounds loses 1.9 F1 because it keeps searching instead of reading; a round-budget sweep shows a step at four Locator rounds and a plateau beyond; query latency scales with model calls rather than corpus size. We serve the three indexed benchmarks as read-only knowledge bases (free registration) and release all prompts and the runner code: https://github.com/metask-ai/knowforge-bench
Chinese Translation
在大型异构文档集合上进行问答首先需要找到正确的段落，然后进行阅读，然而当前的检索增强代理将两种工具赋予同一个模型，并让其自行调度。我们提出了KnowForge，一个按任务分离的管道：一个混合检索知识点索引；一个只能搜索的定位代理；一个强制缺失第二步的代码调度桥接判断器；一个合并每次搜索排名的类型化、大小限制的证据池；一个只能阅读文本、图像和表格的深度阅读代理；以及一个带格式检查器的合成器。在自托管的Qwen3.6-35B-A3B（3B活跃）上，KnowForge在完整的开放域M3DocVQA（2,441个问题，3,368个PDF）上达到了F1 47.5，比已发布的M3DocRAG系统提高了11.0，比在同一索引和模型上进行的一次性检索与回答提高了7.1，并且在DeepSeek-V4-Flash上达到了53.5。在MultiHop-RAG上，它的准确率达到了73.5，超过了已发布的最强RAG和GraphRAG结果（71.2）；在单文档MMLongBench-Doc上，它将同一模型的表现从34.8提升至47.0，与阅读页面图像的GPT-4o（46.3）持平。对完整基准的逐任务消融分析归因于增益：深度阅读器在三个语料库上分别贡献了3.8、9.8和0点，跟踪其视觉程度；将两个代理合并为一个，使用所有工具和两倍的轮次则损失了1.9 F1，因为它继续搜索而不是阅读；轮次预算的扫查显示在四个定位轮次时达到一个步骤，之后则保持平稳；查询延迟与模型调用的数量而非语料库大小成比例。我们将这三个索引基准作为只读知识库提供（免费注册），并发布所有提示和运行代码： https://github.com/metask-ai/knowforge-bench
cs.AI / 44 / 2610.13131
Looking Ahead and Back: Anticipatory Generative Modeling for Road-Network Trajectories
展望与回顾：道路网络轨迹的预期生成建模
Abstract
Trajectory generation underpins urban applications such as traffic management and urban planning. Within road networks, vehicle movement along road segments is largely deterministic, whereas transitions at intersections are critical--multiple outgoing paths may appear equally plausible, yet they can lead to substantially different downstream routes. Existing generative methods such as diffusion models simulate location sequences and fail short of next transition selection, even producing trajectories that drift off the road network. To fill this gap, we advocate a fundamental shift: moving beyond location simulation toward behavior modeling on the road network. In this paper, we revisit trajectory generation through the lens of generative route planning, and introduce a new graph-constrained generative paradigm, Anticipatory Generative Modeling, which resolves next transition selection by incorporating the Anticipation of how the future route may unfold toward the destination. Concretely, we propose a novel generative model named Foresight that formulates a probabilistic model for an "anticipate--select--update--re-anticipate" loop that enables longer-term foresight while continuing sequential generation along the historical trajectory. Extensive experiments demonstrate that Foresight wins state-of-the-art baselines.
Chinese Translation
轨迹生成是城市应用的基础，例如交通管理和城市规划。在道路网络中，车辆沿道路段的运动在很大程度上是确定性的，而在交叉口的转变则至关重要——多个出发路径可能看似同样合理，但它们可能导致截然不同的下游路线。现有的生成方法，如扩散模型，模拟位置序列，但在下一个转变选择上存在不足，甚至会生成偏离道路网络的轨迹。为了填补这一空白，我们提倡一个根本性的转变：超越位置模拟，转向道路网络上的行为建模。在本文中，我们通过生成路线规划的视角重新审视轨迹生成，并引入一种新的图约束生成范式——预期生成建模（Anticipatory Generative Modeling），该方法通过结合对未来路线如何朝向目的地展开的预期来解决下一个转变选择问题。具体而言，我们提出了一种新颖的生成模型，命名为Foresight，它为“预期—选择—更新—再预期”循环制定了一个概率模型，使得在沿历史轨迹继续进行序列生成的同时，实现更长远的前瞻性。大量实验表明，Foresight在各项指标上超越了现有的最先进基准。
cs.AI / 45 / 2610.13132
DP-MIND: Differentially private memory-guided Long context Inference
DP-MIND：差分隐私记忆引导的长上下文推理
Abstract
LLM-as-a-Service has become a dominant paradigm for accessing large language models without requiring local deployment. However, the process raises severe privacy concerns when user queries may include sensitive or confidential contexts. Differential privacy (DP) provides a principled approach to protecting such inputs, but existing DP inference methods are primarily designed for short- or moderate-context settings and struggle with million-token-scale inputs due to limited context capacity and severe utility degradation under DP perturbation. In this paper, we propose DP-MIND, a memory-guided framework that scales differentially private inference to million-token contexts. DP-MIND enables private long-context inference by locally perturbing the input and processing it chunk by chunk with a compact memory that preserves task-relevant information across the sequence. To make this memory robust to DP noise, we optimize the memory construction policy directly on perturbed inputs using reinforcement learning, guided by both final answer correctness and memory consistency. We further recover perturbation-corrupted answer spans through a local denoising module without exposing the original context to the external model. We provide theoretical privacy analysis of DP-MIND and further conduct extensive empirical analysis across multiple long-context benchmarks with context lengths up to the million-token scale, demonstrating that DP-MIND consistently outperforms existing DP and long-context inference baselines in both scalability and utility.
Chinese Translation
LLM-as-a-Service 已成为访问大型语言模型的主流范式，无需本地部署。然而，当用户查询可能包含敏感或机密上下文时，这一过程引发了严重的隐私问题。差分隐私（DP）提供了一种保护此类输入的原则性方法，但现有的 DP 推理方法主要针对短或中等上下文设置，面对百万标记规模的输入时，由于上下文容量有限和在 DP 扰动下的严重效用下降而难以应对。本文提出了 DP-MIND，一种将差分隐私推理扩展到百万标记上下文的记忆引导框架。DP-MIND 通过局部扰动输入并逐块处理，利用紧凑的记忆在序列中保留与任务相关的信息，从而实现私密的长上下文推理。为了使该记忆对 DP 噪声具有鲁棒性，我们使用强化学习直接在扰动输入上优化记忆构建策略，指导因素包括最终答案的正确性和记忆的一致性。我们进一步通过局部去噪模块恢复扰动损坏的答案范围，而无需将原始上下文暴露给外部模型。我们提供了 DP-MIND 的理论隐私分析，并在多个长上下文基准上进行了广泛的实证分析，测试上下文长度高达百万标记，结果表明 DP-MIND 在可扩展性和效用方面始终优于现有的 DP 和长上下文推理基线。
cs.AI / 46 / 2610.13144
Competence-Aware Orchestration for LLM-Based Multi-Agent Systems
基于能力感知的LLM多智能体系统编排
Abstract
LLM-based multi-agent systems (MAS) have emerged as a promising paradigm for enhancing complex reasoning through role specialization and collaboration, yet a key challenge remains: how can collaboration be adapted to a backbone's competence on the current query? This problem matters because more collaboration is not always better--richer coordination may help on difficult cases, but it also incurs additional token cost and redundant interactions. We propose a two-stage competence-aware framework for automatic MAS orchestration. It first models the LLM backbone's task-conditioned competence for the current query, and then uses this signal to adapt role activation and the communication graph. Across six benchmarks, our method achieves an average score of 93.81%. Experiments with six LLM backbones of varying competence levels further show that our method adjusts collaboration according to the inferred competence: for higher-competence backbones, it improves average accuracy from 87.22 to 89.97 while reducing token cost by 36.26%; for lower-competence backbones, it yields a further average gain of 4.58 points over existing methods. The same trend also holds on unseen backbones.
Chinese Translation
基于LLM的多智能体系统（MAS）作为一种通过角色专业化和协作增强复杂推理的有前景的范式，已逐渐受到关注，但仍面临一个关键挑战：如何根据当前查询的主干能力调整协作？这个问题的重要性在于，更多的协作并不总是更好——更丰富的协调可能在困难情况下有所帮助，但也会带来额外的令牌成本和冗余交互。我们提出了一种两阶段的能力感知框架，用于自动化MAS编排。该框架首先对当前查询的LLM主干的任务条件能力进行建模，然后利用这一信号来调整角色激活和通信图。在六个基准测试中，我们的方法平均得分达到93.81%。与六个不同能力水平的LLM主干进行的实验进一步表明，我们的方法根据推断的能力调整协作：对于高能力主干，平均准确率从87.22提高到89.97，同时令牌成本降低了36.26%；对于低能力主干，较现有方法平均提高了4.58分。相同的趋势在未见过的主干上也得到了验证。
cs.AI / 47 / 2610.13145
VIGOR: Verifiable Industrial Geometry Optimization and Repair in Scene Programs
VIGOR：可验证的工业几何优化与场景程序修复
Abstract
Large language models (LLMs) increasingly generate structured layouts and executable scene programs from natural-language instructions. However, executability does not imply engineering feasibility: an industrial scene program may execute successfully while its final state still violates coupled boundary, collision, clearance, aisle, or structural constraints. We identify this mismatch as the executability-engineering-feasibility gap, where locally plausible placements can remain globally infeasible because corrections to one device may introduce violations elsewhere. To make this gap measurable in a controlled setting, we construct an execution-grounded industrial scene-program dataset with 1,749 constructible scenarios, complete executable programs, and 94 final-state geometric checks per scenario. We further propose VIGOR, which combines supervised scene-program generation with execution-grounded verification and staged geometric refinement. Supervised fine-tuning establishes complete program generation, local projection resolves independently correctable violations, and joint conflict-graph repair coordinates residual multi-object conflicts while preserving structural relations. Experiments show that VIGOR achieves 96.86% Complete Feasibility and an Overall score of 91.20, outperforming the strongest evaluated baseline by 10.42 points in Overall. These results show that, under controlled industrial layout conditions, reliable scene-program generation requires not only executable programs, but also explicit verification and coordinated resolution of coupled engineering constraints.
Chinese Translation
大型语言模型（LLMs）越来越多地从自然语言指令生成结构化布局和可执行的场景程序。然而，执行性并不意味着工程可行性：一个工业场景程序可能成功执行，但其最终状态仍然违反耦合边界、碰撞、间隙、通道或结构约束。我们将这种不匹配称为执行性-工程可行性差距，其中局部合理的放置可能在全局上不可行，因为对一个设备的修正可能会在其他地方引入违规。为了在受控环境中使这一差距可测量，我们构建了一个基于执行的工业场景程序数据集，包含1,749个可构建场景、完整的可执行程序以及每个场景的94个最终状态几何检查。我们进一步提出了VIGOR，它结合了监督的场景程序生成、基于执行的验证和分阶段的几何优化。监督微调建立了完整的程序生成，局部投影解决了独立可修正的违规，而联合冲突图修复协调了残余的多对象冲突，同时保持结构关系。实验表明，VIGOR实现了96.86%的完整可行性和91.20的总体得分，在总体上超越了最强评估基线10.42分。这些结果表明，在受控的工业布局条件下，可靠的场景程序生成不仅需要可执行程序，还需要明确的验证和耦合工程约束的协调解决。
cs.AI / 48 / 2610.13152
Quality, Not Just Outcome: A Structural Scoring Framework for Selecting Coding-Agent Trajectories for Supervised Fine-Tuning
质量，而不仅仅是结果：用于选择编码代理轨迹的结构评分框架以进行监督微调
Abstract
Open coding agents are commonly trained through supervised fine-tuning on interaction trajectories whose final patches pass the benchmark test suite. However, this outcome-level filtering criterion ignores process quality: trajectories that eventually succeed may still contain inefficient interaction patterns such as repetitive file inspection, unstable tool usage, or delayed modification behavior, all of which become part of the imitation target. We propose TQS (Trajectory Quality Score), a structural scoring framework for coding-agent trajectories that operates directly on trajectory structure, requires no LLM judge, and runs in linear time. TQS characterizes trajectory quality through structural signals organized around structured convergence: information focus, intent--action alignment, and policy dynamics. Using 155,144 successful trajectories from a public open-license corpus, we rank trajectories by TQS and fine-tune coding agents on filtered subsets. Models trained on top-ranked trajectories consistently outperform those trained on random or bottom-ranked subsets on SWE-Bench-Verified. Our results suggest that process-level trajectory quality is an important factor in coding-agent supervision and that structural trajectory filtering provides an effective and scalable alternative to LLM-based data selection.
Chinese Translation
开放编码代理通常通过在最终补丁通过基准测试套件的交互轨迹上进行监督微调来进行训练。然而，这种基于结果的过滤标准忽视了过程质量：最终成功的轨迹可能仍然包含低效的交互模式，例如重复的文件检查、不稳定的工具使用或延迟的修改行为，这些都成为模仿目标的一部分。我们提出了 TQS（轨迹质量评分），这是一个针对编码代理轨迹的结构评分框架，直接作用于轨迹结构，不需要 LLM（大型语言模型）评判，并且运行时间为线性。TQS 通过围绕结构收敛组织的结构信号来表征轨迹质量：信息聚焦、意图-行动对齐和策略动态。使用来自公共开放许可语料库的 155,144 条成功轨迹，我们通过 TQS 对轨迹进行排名，并在过滤后的子集上微调编码代理。在顶级轨迹上训练的模型在 SWE-Bench-Verified 上的表现始终优于在随机或底部排名子集上训练的模型。我们的结果表明，过程级轨迹质量是编码代理监督中的一个重要因素，而结构轨迹过滤提供了一种有效且可扩展的替代 LLM 基于数据选择的方法。
cs.AI / 49 / 2610.13172
RoleWorld: Knowledge Transfer in World Modeling with Functional Role Abstractions
RoleWorld：基于功能角色抽象的世界建模中的知识转移
Abstract
World models enable large language model (LLM) agents to anticipate action effects and make informed decisions. For generalization, it is desirable for such world models to transfer efficiently to new environments with similar interaction structure and underlying dynamics. Recent approaches encode world dynamics as symbolic rules or executable programs, whose compositional structure in principle affords transfer, but still organize such knowledge around environment-specific objects. To address this limitation, we introduce RoleWorld, a symbolic world modeling approach which uses reusable prototypes and environment-specific parameter values as functional role abstractions to represent how objects behave in interactions. From interaction trajectories, a teacher LLM extracts role prototypes and executable transition rules defined over role instances, together with a skill that guides the actor LLM to use role information in decision making. In a target world, without the teacher, the actor uses this skill together with its current role assignments and parameter estimates, and actively tests a role instance hypothesis through interaction when its uncertainty affects the current decision. Feedback from each action refines role assignments and estimates parameter values online, with no LLM parameter updates and the transition rules held fixed. Experiments on MARS benchmark show that RoleWorld enables robust generalization across environments with significantly altered object behavior, suggesting that functional role abstractions can serve as an effective component for knowledge transfer in world modeling. Project Page: https://future-item.github.io/RoleWorld.
Chinese Translation
世界模型使大型语言模型（LLM）代理能够预测行动效果并做出明智决策。为了实现泛化，这种世界模型应能够高效地转移到具有相似交互结构和潜在动态的新环境。近期的方法将世界动态编码为符号规则或可执行程序，其组合结构原则上支持转移，但仍然围绕环境特定对象组织这些知识。为了解决这一限制，我们引入了RoleWorld，这是一种符号世界建模方法，利用可重用原型和环境特定参数值作为功能角色抽象来表示对象在交互中的行为。从交互轨迹中，教师LLM提取角色原型和定义在角色实例上的可执行过渡规则，以及指导演员LLM在决策中使用角色信息的技能。在目标世界中，演员在没有教师的情况下，利用这一技能结合其当前的角色分配和参数估计，当不确定性影响当前决策时，通过交互主动测试角色实例假设。每个行动的反馈在线细化角色分配和参数值估计，无需更新LLM参数，并且过渡规则保持不变。在MARS基准上的实验表明，RoleWorld能够在对象行为显著改变的环境中实现稳健的泛化，表明功能角色抽象可以作为世界建模中知识转移的有效组成部分。项目页面：https://future-item.github.io/RoleWorld。
cs.AI / 50 / 2610.13184
AdaC-GRPO: Towards Mastering Long-Horizon Visual Sudoku Puzzles via Adaptive Curriculum Learning
AdaC-GRPO：通过自适应课程学习掌握长时间跨度的视觉数独难题
Abstract
While reinforcement learning with verifiable reward (RLVR) enhances reasoning in large language models, its efficacy in long-horizon multimodal tasks remains underexplored. Visual Sudoku puzzles pose a representative challenge, requiring multi-step deductive reasoning over both local constraints and global dependency chains. To this end, we introduce FG-Sudoku, a fine-grained benchmark categorized by grid size and required deductive steps. Evaluations reveal that frontier MLLMs fail on complex long-range dependencies (e.g., 9*9 puzzles). During post-training, this near-zero success rate causes severe reward sparsity, halting standard RLVR. We propose Adaptive Curriculum Group Relative Policy Optimization (AdaC-GRPO) to overcome this. Unlike rigid or prior-based schedules, AdaC-GRPO dynamically constructs training batches by tracking the model's "Zone of Proximal Development" via online rollout accuracy. Experiments demonstrate that AdaC-GRPO enables Qwen2.5-VL-7B to approach the performance of Gemini-2.5-Flash while bridging 62.2% of the overall accuracy gap to Gemini-2.5-Pro, with its generalizability further validated across InternVL-3.5 and Qwen3-VL.
Chinese Translation
尽管具有可验证奖励的强化学习（RLVR）增强了大型语言模型的推理能力，但其在长时间跨度的多模态任务中的有效性仍然未被充分探索。视觉数独难题构成了一个具有代表性的挑战，要求在局部约束和全局依赖链上进行多步推理。为此，我们引入了FG-Sudoku，这是一个按网格大小和所需推理步骤分类的细粒度基准。评估结果显示，前沿的多模态大型语言模型（MLLMs）在复杂的长距离依赖（例如，9*9难题）上表现不佳。在后训练阶段，这种接近零的成功率导致了严重的奖励稀疏性，阻碍了标准的RLVR。我们提出了自适应课程组相对策略优化（AdaC-GRPO）来克服这一问题。与僵化或基于先前的调度不同，AdaC-GRPO通过在线回滚准确性动态构建训练批次，跟踪模型的“最近发展区”。实验表明，AdaC-GRPO使Qwen2.5-VL-7B的性能接近Gemini-2.5-Flash，同时弥补了62.2%的整体准确性差距至Gemini-2.5-Pro，其通用性在InternVL-3.5和Qwen3-VL上得到了进一步验证。
cs.AI / 51 / 2610.13194
JANUS: Enhancing Long-Horizon Autonomous Research with Co-Evolving World Models
JANUS：通过共同演化的世界模型增强长时间跨度的自主研究
Abstract
Long-horizon autonomous research has shown promise in accelerating scientific progress through sustained cycles of hypothesis generation, implementation, and validation. However, this Generate-Execute-Feedback paradigm relies on reactive trial-and-error, where costly and time-consuming executions introduce severe feedback delays, throttling evolutionary progress under explosively expanding search space. To address this challenge, we propose JANUS, a framework that unifies future outcomes forecasting and past experience reflection through a Predict-Verify-Evolve loop driven by a co-evolving world model. Within an evolving graph, the research agent proposes candidates while world model scores them by forecasting value improvement, information gain, and exploration utility, then accordingly selects, refines, or prunes branches prior to execution. After receiving ground-truth feedback, JANUS distills experience into evolving skills: hierarchical skills for research agent, exploration-strategy skills for world model, and their coordination protocols. This mechanism enables continual co-evolution: the research agent sharpens candidate generation while the world model refines its predictive judgment over extended horizons. Extensive evaluations on MLS-Bench-Lite and real research tasks demonstrate that JANUS significantly outperforms state-of-the-art baselines, achieving both higher evolving efficiency and ceiling.
Chinese Translation
长时间跨度的自主研究在通过持续的假设生成、实施和验证循环加速科学进展方面显示出良好的前景。然而，这一生成-执行-反馈范式依赖于反应性的试错过程，其中高成本和耗时的执行引入了严重的反馈延迟，抑制了在快速扩展的搜索空间下的进化进程。为了解决这一挑战，我们提出了JANUS，一个通过共同演化的世界模型驱动的预测-验证-演化循环，统一未来结果预测和过去经验反思的框架。在一个不断演变的图中，研究代理提出候选项，而世界模型通过预测价值改善、信息增益和探索效用对其进行评分，然后在执行之前相应地选择、细化或修剪分支。在接收到真实反馈后，JANUS将经验提炼为不断演变的技能：研究代理的层次技能、世界模型的探索策略技能及其协调协议。这一机制使得持续的共同演化成为可能：研究代理在候选生成上不断提升，而世界模型在较长时间跨度内精炼其预测判断。在MLS-Bench-Lite和真实研究任务上的广泛评估表明，JANUS显著优于最先进的基线，达到了更高的演化效率和上限。
cs.AI / 52 / 2610.13267
Forgetting is Not Always Bad: A Neuro-Inspired Memory Repair Mechanism for Poisoned LLM Agents
遗忘并不总是坏事：一种神经启发的记忆修复机制用于被污染的LLM代理
Abstract
Large language model (LLM) agents are highly vulnerable to memory-poisoning attacks. Existing defenses primarily rely on external modules for either static memory isolation or continuous online auditing, resulting in low memory efficiency and high computational overhead. To address these limitations, inspired by neuroscientific mechanisms that weak reactivation can destabilize memories and induce decay in the absence of reinforcement, we propose an agent-intrinsic training-free memory repair framework, DREAM (Dynamic Reactivation for Engram Attenuation in Memory), that enables selective functional forgetting of poisoned memories. Specifically, DREAM implements a three-stage pipeline: perturbation-induced implicit reactivation to reveal the structural fragility of malicious memories, adaptive anomaly diagnosis based on multi-dimensional activation patterns to detect topological anomalies, and a dynamic memory repair module that selectively suppresses harmful memories while reinforcing benign ones. Extensive experiments across diverse attackers and real-world tasks demonstrate that DREAM reduces attack success rates by over 95\% against backdoor poisoning while preserving strong benign utility. DREAM also achieves competitive robustness against injection attacks. In terms of runtime and token consumption, DREAM achieves up to a 2.13$\times$ speedup and a 37.7\% reduction in token consumption compared with A-MemGuard. Furthermore, DREAM also achieves a task success rate of 92.96\% on poisoned multi-agent systems.
Chinese Translation
大型语言模型（LLM）代理对记忆污染攻击高度脆弱。现有的防御措施主要依赖外部模块进行静态记忆隔离或持续在线审计，导致记忆效率低下和计算开销高。为了解决这些局限性，我们受到神经科学机制的启发，提出了一种代理内在的无训练记忆修复框架DREAM（动态再激活以减弱记忆中的痕迹），该框架能够选择性地功能性遗忘被污染的记忆。具体而言，DREAM实现了一个三阶段的流程：扰动引发的隐式再激活以揭示恶意记忆的结构脆弱性，基于多维激活模式的自适应异常诊断以检测拓扑异常，以及一个动态记忆修复模块，选择性地抑制有害记忆，同时强化良性记忆。针对不同攻击者和现实任务的广泛实验表明，DREAM在抵御后门污染时将攻击成功率降低了超过95\%，同时保持了强大的良性效用。DREAM在抵御注入攻击方面也表现出竞争力的鲁棒性。在运行时间和令牌消耗方面，DREAM与A-MemGuard相比实现了最高2.13倍的加速和37.7\%的令牌消耗减少。此外，DREAM在被污染的多代理系统上也达到了92.96\\%的任务成功率。
cs.AI / 53 / 2610.13285
BoundaryEvo: Data-Model Co-Evolution at the Capability Boundary for Long-Chain Reasoning Distillation
BoundaryEvo：在能力边界下的数据-模型共演化用于长链推理蒸馏
Abstract
Distilling long-chain reasoning trajectories from large teacher models provides an effective way to improve the reasoning ability of small language models. However, existing methods typically select a fixed training set before distillation, overlooking that the usefulness of a reasoning example changes as the student evolves. This mismatch becomes especially costly when only a small number of teacher trajectories can be used for training. Besides, examples already mastered provide little new supervision, while examples far beyond the capability of student may be difficult to learn from. Training data should therefore adapt as model capability evolves, yet reliable data selection is difficult when the student is initially weak. To address these challenges, our key idea is to first establish the basic reasoning capability and then adapt the training data to the evolving capability boundary of the student. We propose BoundaryEvo, a two-stage data--model co-evolution framework. In Capability Priming, we directly distill a small set of teacher-generated long reasoning trajectories without using a separate model to judge and filter them, allowing the student to acquire an initial reasoning capability before capability-aware data selection. In Capability-Boundary Co-Evolution, we repeatedly select questions that the current student has not mastered but the expert can reliably solve, and update the training data as the student evolves. With only 1,000 training examples, BoundaryEvo consistently outperforms static and dynamic data selection baselines across 11 reasoning benchmarks, raising the original Qwen3.5-4B accuracy from 62.9 to 75.3 on IMO-AnswerBench and from 26.8 to 46.0 on AMO-Bench. The code and data will be made publicly available.
Chinese Translation
从大型教师模型中蒸馏长链推理轨迹为小型语言模型提升推理能力提供了一种有效的方法。然而，现有方法通常在蒸馏之前选择固定的训练集，忽视了随着学生模型的发展，推理示例的有效性会发生变化。当可用于训练的教师轨迹数量有限时，这种不匹配尤为昂贵。此外，已经掌握的示例提供的监督信息有限，而超出学生能力范围的示例可能难以学习。因此，训练数据应随着模型能力的演变而适应，但在学生模型初期较弱时，可靠的数据选择变得困难。为了解决这些挑战，我们的关键思想是首先建立基本的推理能力，然后将训练数据适应学生不断演变的能力边界。我们提出了BoundaryEvo，一个两阶段的数据-模型共演化框架。在能力预热阶段，我们直接蒸馏一小组教师生成的长推理轨迹，而不使用单独的模型进行判断和过滤，使学生在能力感知数据选择之前获得初步的推理能力。在能力边界共演化阶段，我们反复选择当前学生尚未掌握但专家能够可靠解决的问题，并随着学生的发展更新训练数据。在仅使用1,000个训练示例的情况下，BoundaryEvo在11个推理基准测试中始终优于静态和动态数据选择基线，将原始Qwen3.5-4B的准确率从62.9提升至75.3（在IMO-AnswerBench上），并从26.8提升至46.0（在AMO-Bench上）。代码和数据将公开发布。
cs.AI / 54 / 2610.13330
Masked Diffusion Language Agents for Tool-Integrated Chemical Reasoning
用于工具集成化化学推理的掩蔽扩散语言代理
Abstract
Designing tool-integrated reasoning agent systems for complex chemical tasks remains a fundamental challenge. Although recent chemical agents have demonstrated promising performance through the integration of external tools, they are still built on the autoregressive paradigm, which is inherently constrained by causal attention and left-to-right generation. As a result, they often struggle with bidirectional molecular understanding, global coordinated tool planning and multi-source evidence integration. To address this challenge, we introduce ChemDiffAgent, the first masked diffusion language agent for tool-integrated chemical reasoning. It reformulates chemical reasoning as iterative denoising over interleaved reasoning and tool-use trajectories, enabling tool-use decisions to be jointly determined under full-trajectory context and thereby better aligning the agent's decision boundary with its knowledge boundary. We further analyze how this paradigm benefits agentic reasoning through harder infilling subproblems during training and flexible arbitrary-order decoding during inference. Then we develop a two-stage post-training framework, comprising agentic supervised fine-tuning and variance-reduced preference optimization, to enhance tool-use and reasoning capabilities. We also construct a comprehensive benchmark covering six chemistry tasks across both single-turn and multi-turn tool-use scenarios. Under matched budgets and evaluation, ChemDiffAgent consistently outperforms autoregressive agents, achieving an overall improvement of approximately 20%. Notably, our 8B model achieves performance comparable to, in some cases surpassing, significantly larger frontier models, such as GPT 5.5 and Claude Opus 4.7. Code is available at https://github.com/bilemond/ChemDiffAgent.
Chinese Translation
设计用于复杂化学任务的工具集成推理代理系统仍然是一个基本挑战。尽管近期的化学代理通过集成外部工具展现了良好的性能，但它们仍然基于自回归范式，这一范式固有地受到因果注意力和从左到右生成的限制。因此，它们在双向分子理解、全局协调工具规划和多源证据整合方面常常面临困难。为了解决这一挑战，我们引入了ChemDiffAgent，这是首个用于工具集成化化学推理的掩蔽扩散语言代理。它将化学推理重新表述为在交错的推理和工具使用轨迹上进行迭代去噪，从而使工具使用决策能够在全轨迹上下文中共同确定，从而更好地将代理的决策边界与其知识边界对齐。我们进一步分析了这一范式如何通过在训练期间更困难的填充子问题和在推理期间灵活的任意顺序解码来促进代理推理。然后，我们开发了一个两阶段的后训练框架，包括代理监督微调和方差减少的偏好优化，以增强工具使用和推理能力。我们还构建了一个全面的基准，涵盖六个化学任务，涉及单轮和多轮工具使用场景。在匹配的预算和评估下，ChemDiffAgent始终优于自回归代理，整体性能提升约20%。值得注意的是，我们的8B模型在性能上可与一些情况下超越显著更大的前沿模型，如GPT 5.5和Claude Opus 4.7。代码可在https://github.com/bilemond/ChemDiffAgent获取。
cs.AI / 55 / 2610.13354
AlphaNIST: Emergent Assembly Skills from Binary Task Rewards
AlphaNIST：从二元任务奖励中涌现的组装技能
Abstract
The NIST Assembly Task Board #1 remains a challenging benchmark for learning-based methods in robotic manipulation. Over short horizons, it demands precise, contact-rich control of small, tight-fitting parts. Over long horizons, it requires recovery and coordination across extended action sequences. Demonstrations, task decomposition, and dense progress rewards have enabled research advances by providing guidance on how to assemble. But how much of the strategy must be specified? The progression from AlphaGo to AlphaGo Zero showed that RL at scale can discover superhuman strategies without prescribing how to win. AlphaNIST asks whether a sparse binary task reward and task diversity admitted by this task-agnostic reward can similarly let a robot discover how to assemble. In simulation, we train a single policy from scratch across all 20 assembly tasks on the NIST Assembly Task Board #1, reaching 98.2% overall training success without demonstrations, progress shaping, scripted primitives, or per-task tuning. The framework also trains policies to complete ten assembly pairs in a single episode without a prescribed order, with task success rewarded only when all ten pairs are simultaneously complete. The learned policies discover dexterous and planning-like behaviors: choosing grasps for later assembly, using surrounding geometry to reorient parts, and preparing one part, assembling another, then returning to finish the first without an explicit planner. For videos, see our website: https://alpha-nist.github.io/
Chinese Translation
NIST组装任务板#1仍然是机器人操作中基于学习的方法的一个具有挑战性的基准。在短时间内，它要求对小型、紧密配合的部件进行精确、接触丰富的控制。在长时间内，它需要在扩展的动作序列中进行恢复和协调。演示、任务分解和密集的进展奖励通过提供如何组装的指导，促进了研究的进展。但策略需要指定多少呢？从AlphaGo到AlphaGo Zero的进展表明，规模化的强化学习可以发现超人类策略，而无需规定如何获胜。AlphaNIST探讨了稀疏的二元任务奖励和该任务无关奖励所允许的任务多样性是否同样能够让机器人发现如何组装。在模拟中，我们从零开始训练一个单一策略，涵盖NIST组装任务板#1上的所有20个组装任务，达到98.2%的整体训练成功率，而无需演示、进展塑造、脚本化原语或逐任务调优。该框架还训练策略在单个回合中完成十对组装任务，而不规定顺序，只有在所有十对同时完成时才给予任务成功奖励。学习到的策略发现了灵巧和规划般的行为：选择后续组装的抓取，利用周围几何形状重新定向部件，以及准备一个部件、组装另一个部件，然后返回完成第一个部件，而无需显式的规划器。有关视频，请参见我们的网站：https://alpha-nist.github.io/
cs.AI / 56 / 2610.13356
SWE-Critix: Post-Training Generalist Verifiers for Assessing Coding-Agent Trajectories
SWE-Critix：用于评估编码代理轨迹的后训练通用验证器
Abstract
Coding agents are increasingly deployed for end-to-end software engineering tasks, where a readily available oracle evaluator that defines the success criteria is often absent. This motivates verifications for coding agents that can reliably judge and explain their behavior. However, verification is non-trivial as coding agents are increasingly capable of solving long-horizon software engineering tasks. Their execution trajectories can be extremely long, containing hundreds of iterative interactions with code repositories and external tools. Existing verifiers either rely on general-purpose LLMs without specialized training or train dedicated critics that predominantly focus on producing scalar rewards, highlighting the relatively limited exploration of training-based generalist verification. We introduce SWE-Critix, a post-training pipeline for training generalist verifiers of coding-agent trajectories. Given a task and its execution trajectory, the learned verifier predicts the completion label and generates a natural-language analysis that explains its judgment. Our pipeline combines supervised fine-tuning (SFT) with reinforcement learning (RL) and trains a series of 30B models on trajectories from CodeForge-Preview. On a held-out test set, SWE-Critix improves classification accuracy from 57% for the base Qwen3-30B-A3B reasoning model to 66% after SFT and 78% after RL. We release our models, datasets, and training code to facilitate further research.
Chinese Translation
编码代理在端到端软件工程任务中的应用日益增多，但通常缺乏一个可用的评估者来定义成功标准。这促使我们对编码代理进行验证，以可靠地判断和解释其行为。然而，验证并非易事，因为编码代理在解决长期软件工程任务方面的能力不断增强。它们的执行轨迹可能非常长，包含与代码库和外部工具的数百次迭代交互。现有的验证器要么依赖于没有专门训练的一般性大型语言模型（LLMs），要么训练专门的评估器，主要集中在产生标量奖励上，这突显了基于训练的通用验证探索相对有限。我们提出了SWE-Critix，一个用于训练编码代理轨迹的通用验证器的后训练管道。给定一个任务及其执行轨迹，学习到的验证器预测完成标签，并生成自然语言分析来解释其判断。我们的管道结合了监督微调（SFT）和强化学习（RL），并在CodeForge-Preview的轨迹上训练了一系列30B模型。在一个保留的测试集上，SWE-Critix将基础Qwen3-30B-A3B推理模型的分类准确率从57%提高到SFT后的66%，以及RL后的78%。我们发布了我们的模型、数据集和训练代码，以促进进一步的研究。
cs.AI / 57 / 2610.13399
RegulaTutor: A Theory-Informed Framework for Long-Term Personalized Learning
RegulaTutor：一个基于理论的长期个性化学习框架
Abstract
Long-term personalized tutoring requires language models to translate a learner's evolving history into appropriate instructional support. Existing LLM-based tutoring systems use memory and retrieval to personalize responses, but the pedagogical rationale linking historical evidence to teaching strategies is not always explicit. We introduce RegulaTutor, a theory-informed framework for long-term personalized tutoring that connects historical evidence, learner interpretation, and learner-facing instruction. Guided by self-regulated learning (SRL) research, its tutoring harness operationalizes a cognitive, affective, metacognitive, and motivational (CAMM) framework for deriving evidence-supported learner interpretations and teaching policies. The resulting policies retain links to their supporting evidence and are updated as new observations arrive. This staged workflow also supports systematic refinement through PairRoute, an offline procedure that uses intermediate execution records to distinguish harness, model, and mixed failures. PairRoute proposes and evaluates targeted harness revisions, then constructs reviewed stage-level supervision for confirmed model failures remaining under the selected harness. We focus this refinement on historical evidence acquisition, which underpins personalized instruction and allows outputs to be checked against learning records and reference answers. RegulaTutor improves historical-evidence accuracy, learner diagnosis, and judged teaching quality on average across evaluated model configurations. PairRoute results show an endpoint improvement in standalone-model historical-evidence accuracy over the refinement trajectory. Together, these results highlight RegulaTutor's potential to support lifelong personalized education through instructional policies that evolve with learner histories and targeted improvements to the tutoring system.
Chinese Translation
长期个性化辅导需要语言模型将学习者不断发展的历史转化为适当的教学支持。现有的基于大型语言模型（LLM）的辅导系统使用记忆和检索来个性化响应，但将历史证据与教学策略联系起来的教育理论依据并不总是明确。我们提出了RegulaTutor，一个基于理论的长期个性化辅导框架，它连接了历史证据、学习者的解读和面向学习者的教学。该框架受自我调节学习（SRL）研究的指导，其辅导机制运用认知、情感、元认知和动机（CAMM）框架来推导基于证据的学习者解读和教学政策。所产生的政策保留了与其支持证据的联系，并在新观察到达时进行更新。这一分阶段的工作流程还通过PairRoute支持系统化的改进，PairRoute是一种离线程序，利用中间执行记录来区分辅导机制、模型和混合失败。PairRoute提出并评估针对性的辅导机制修订，然后为在所选辅导机制下确认的模型失败构建经过审查的阶段级监督。我们将这一改进重点放在历史证据的获取上，这为个性化教学提供了基础，并允许将输出与学习记录和参考答案进行核对。RegulaTutor在评估的模型配置中平均提高了历史证据的准确性、学习者诊断和教学质量评估。PairRoute的结果显示，在改进过程中，独立模型的历史证据准确性在最终点上有所提升。这些结果共同突显了RegulaTutor通过与学习者历史相适应的教学政策以及对辅导系统的针对性改进来支持终身个性化教育的潜力。
cs.AI / 58 / 2610.13416
Web$hell: A Unified Interface for Agentic Web Access
Web$hell：一种统一的智能网络访问接口
Abstract
Large language model (LLM) agents are becoming direct users of the open web. As they take on increasingly complex and information-intensive tasks, effective web access is becoming a core capability for agentic problem solving. Existing approaches have advanced agentic web access through different access models, including search-augmented agents, browser-based agents, and, more recently, filesystem-based agents. Nevertheless, as users expect agents to complete long-horizon, end-to-end tasks, they increasingly need to conduct broad information gathering, perform browser actions, and revisit acquired evidence within a single trajectory. Yet our preliminary analyses reveal that existing access models struggle to meet these demands. They primarily support either information gathering or browser interaction, while combining their capabilities can involve substantial costs or performance trade-offs. To this end, we present Web\$hell, an interface that unifies web access under a single access model. Using a common schema defined by WebContract, WebCompiler compiles live webpages into addressable information and action objects. These objects are maintained in WebCache and handled through a shared set of operations, providing agents with consistent, operable, and persistent web access. Through extensive experiments across web search, web navigation, and deep research benchmarks, we show that Web\$hell achieves strong task performance at lower cost. Our results further indicate the importance of interface design for effective web access, with Web\$hell opening new possibilities for more capable and efficient LLM agents on the web.
Chinese Translation
大型语言模型（LLM）代理正在成为开放网络的直接用户。随着它们承担越来越复杂和信息密集的任务，有效的网络访问正成为智能问题解决的核心能力。现有方法通过不同的访问模型推动了智能网络访问，包括增强搜索的代理、基于浏览器的代理，以及最近的基于文件系统的代理。然而，随着用户期望代理能够完成长期、端到端的任务，它们越来越需要在单一轨迹中进行广泛的信息收集、执行浏览器操作并重新访问获取的证据。然而，我们的初步分析表明，现有的访问模型难以满足这些需求。它们主要支持信息收集或浏览器交互，而将其能力结合起来可能涉及巨大的成本或性能权衡。为此，我们提出了Web$hell，一个在单一访问模型下统一网络访问的接口。利用WebContract定义的通用模式，WebCompiler将实时网页编译为可寻址的信息和操作对象。这些对象保存在WebCache中，并通过一组共享操作进行处理，为代理提供一致、可操作和持久的网络访问。通过在网络搜索、网络导航和深度研究基准上的广泛实验，我们展示了Web$hell以更低的成本实现了强大的任务性能。我们的结果进一步表明，接口设计对于有效的网络访问至关重要，Web$hell为更强大和高效的LLM代理在网络上的应用开辟了新的可能性。
cs.AI / 59 / 2610.13437
Local Diagnostics of Learned Event Representations for Anomaly Detection at the HL-LHC
高亮度大型强子对撞机中学习事件表征的局部诊断用于异常检测
Abstract
Anomaly detection, the identification of collision events that deviate from Standard Model expectations, is one of the most promising model-agnostic strategies for new-physics searches at the Large Hadron Collider. Searching for rare events in learned low-dimensional representations removes the need to commit to hand-crafted observables a priori, but sacrifices the immediate physical interpretation those observables provide. For anomaly detection to enter a discovery pipeline, flagged events must therefore be diagnosed locally, both to distinguish potential new physics from mismodeling or detector effects and to identify observables for targeted follow-up searches. We evaluate supervised and self-supervised neural embedding strategies on COLLIDE-2V, a 750-million-event dataset simulated under realistic High-Luminosity LHC conditions, using the reconstruction error of an autoencoder trained in the learned latent space as the anomaly score. Across the configurations and benchmark signals considered, supervised objectives, which exploit simulation process labels during training, generally outperform their self-supervised counterparts and, in their best configurations, improve on an autoencoder trained directly on kinematic features, raising the signal efficiency at 10% background acceptance on all held-out processes. We further introduce a local diagnostic pipeline that fits a Gaussian Mixture Model to the SM latent space, profiles each component through high-level observables without process labels, localizes the flagged events, and ranks observables by their local discriminating power. The ranking recovers the defining kinematics of the held-out processes, while correlating the anomaly score with the same observables quantifies how closely the score tracks them, revealing cases where the most discriminating observable is not the one the score follows most closely.
Chinese Translation
异常检测，即识别偏离标准模型预期的碰撞事件，是在大型强子对撞机进行新物理搜索时最有前景的无模型策略之一。在学习的低维表征中搜索稀有事件，消除了事先承诺手工制作可观测量的需要，但牺牲了这些可观测量所提供的直接物理解释。因此，为了使异常检测进入发现流程，必须对标记事件进行局部诊断，以区分潜在的新物理与模型不匹配或探测器效应，并识别用于针对性后续搜索的可观测量。我们在COLLIDE-2V数据集上评估了监督和自监督神经嵌入策略，该数据集是在现实的高亮度LHC条件下模拟的，包含7.5亿个事件，使用在学习的潜在空间中训练的自编码器的重建误差作为异常分数。在考虑的配置和基准信号中，利用训练期间的模拟过程标签的监督目标通常优于其自监督对应物，并且在最佳配置中，改善了直接在运动学特征上训练的自编码器，提高了在所有保留过程上10%背景接受率下的信号效率。我们进一步引入了一个局部诊断流程，该流程将高斯混合模型拟合到标准模型潜在空间，通过高层可观测量对每个组件进行分析，而无需过程标签，定位标记事件，并根据其局部区分能力对可观测量进行排序。排序恢复了保留过程的定义运动学，同时将异常分数与相同可观测量相关联，量化了分数与它们的跟踪程度，揭示了最具区分性的可观测量并不是分数最紧密跟踪的量的情况。
cs.AI / 60 / 2610.13438
Rollout-Time Intervention: Unlocking Agentic Reinforcement Learning Beyond Policy Limits
回滚时间干预：超越策略限制的自主强化学习解锁
Abstract
Reinforcement learning (RL) has emerged as a key paradigm for improving large language model (LLM) agents, enabling stronger reasoning, planning, and tool-use in interactive environments. However, existing agent RL methods remain fundamentally constrained by an on-policy reasoning bottleneck: when tasks exceed the model's current capability boundary, effective trajectories become extremely sparse and training often collapses. We introduce Rollout-Time Intervention (RTI), a technique designed to elicit successful trajectories beyond the policy's intrinsic exploration capacity. RTI operates by leveraging reward-side evidence (i.e., evaluation rubrics) during rollout to convert erroneous trajectories into successful ones, providing effective learning signals for tasks that the policy rarely solves on its own. Although intervention trajectories are inherently off-policy, we find that they can effectively guide policy optimization when trajectories with excessive distributional shifts are filtered out. Extensive experiments across five benchmarks demonstrate that RTI consistently outperforms RL baselines, enabling the post-trained policy to approach the performance of substantially stronger contemporary LLMs and, more importantly, transcend its original capability boundary. The code is available at https://github.com/Alibaba-NLP/qqr.
Chinese Translation
强化学习（RL）已成为提升大型语言模型（LLM）代理的关键范式，使其在交互环境中具备更强的推理、规划和工具使用能力。然而，现有的代理RL方法在根本上受到在线策略推理瓶颈的限制：当任务超出模型当前能力边界时，有效轨迹变得极为稀疏，训练往往会崩溃。我们提出了回滚时间干预（RTI），这是一种旨在引导成功轨迹超越策略内在探索能力的技术。RTI通过在回滚过程中利用奖励侧证据（即评估标准），将错误轨迹转化为成功轨迹，从而为那些策略自身很少解决的任务提供有效的学习信号。尽管干预轨迹本质上是离线的，我们发现，当过滤掉过度分布偏移的轨迹时，它们可以有效指导策略优化。针对五个基准的广泛实验表明，RTI始终优于RL基线，使得后训练的策略能够接近当前更强大的LLM的性能，更重要的是，超越其原有能力边界。代码可在 https://github.com/Alibaba-NLP/qqr 获取。
cs.AI / 61 / 2610.13455
Agentic Large Language Models
自主大型语言模型
Abstract
Large language models (LLMs) have taken the world by storm. Their ability to communicate with us in our own language, and to reason about our world, has made them a favorite tool for writing, answering questions, and much more. Their answers already influence our decisions and our actions. What would happen if LLMs themselves could act in the world directly? What would happen if LLMs could interact with each other? Agentic LLMs combine the power of large language models with the ability to act in the real world. In this book we will discuss how agentic LLMs work, how they act, how they interact, and what this means for us, as users.
Chinese Translation
大型语言模型（LLMs）在全球范围内引起了轰动。它们能够用我们自己的语言与我们交流，并对我们的世界进行推理，这使它们成为写作、回答问题等多种用途的热门工具。它们的回答已经影响了我们的决策和行动。如果 LLMs 自身能够直接在世界中行动，会发生什么？如果 LLMs 能够相互互动，又会发生什么？自主 LLMs 将大型语言模型的强大能力与在现实世界中行动的能力结合在一起。在本书中，我们将讨论自主 LLMs 的工作原理、它们如何行动、如何互动，以及这对我们作为用户意味着什么。
cs.AI / 62 / 2610.13461
SqueezeBench: Saturation-Resilient Evaluation of Mathematical Capability
SqueezeBench：抗饱和的数学能力评估
Abstract
AI systems are making rapid progress in mathematics, but this progress is hard to measure. Benchmarks of solved problems saturate, and in benchmarks of open problems each problem can be resolved for the first time only once. We present SqueezeBench, a benchmark that is resilient to saturation: a success raises the bar instead of removing the problem. Each problem fixes an integer function $f(n)$ whose values are known only for small $n$, and the benchmark keeps a lower and an upper record on it. A record is a computable Lean function with a machine-checked proof that it bounds $f$. A submission proposes a new function and proves that it bounds $f$, that it is no worse than the current record at any $n$, and that it is strictly better at some $n$. Once accepted, it becomes the record that the next submission has to beat. The system thus chooses the bound as well as its proof, and it may improve the record at a single $n$ or its growth rate. At every $n$, the gap between the two records shows how far the certified bounds are from determining $f(n)$. Every proof is checked by two independent Lean kernels, so no referee needs to check the mathematics. The checks also reject a function that computes $f$ by exhaustive search at small $n$, and a maintainer only reads the submitter's report and looks for such a search at larger $n$. We release ten problems from extremal combinatorics with twenty proved initial records and the admission tools. As a baseline, agents based on a frontier language model, with at most one hour per record, had 82 submissions accepted and improved every record, without going beyond the published literature. Agents based on two newer models then started from these records, had 60 more submissions accepted and improved 19 of the 20 records again. For eight of the ten problems, at least one record is still asymptotically behind the best published bound.
Chinese Translation
人工智能系统在数学领域取得了快速进展，但这一进展难以衡量。已解决问题的基准存在饱和现象，而开放问题的基准中，每个问题只能被首次解决一次。我们提出了 SqueezeBench，这是一个抗饱和的基准：成功的解决方案提高了标准，而不是移除问题。每个问题固定一个整数函数 $f(n)$，其值仅在小的 $n$ 下已知，基准保持对该函数的上下记录。记录是一个可计算的 Lean 函数，并附有机器检查的证明，证明其界定了 $f$。提交者提出一个新函数，并证明其界定了 $f$，在任何 $n$ 下不比当前记录差，并且在某些 $n$ 下严格更优。一旦被接受，它就成为下一个提交者需要超越的记录。因此，系统选择界限及其证明，并可能在单个 $n$ 或其增长率上改善记录。在每个 $n$ 下，两个记录之间的差距显示了认证界限与确定 $f(n)$ 的距离。每个证明由两个独立的 Lean 内核检查，因此无需裁判检查数学。检查还会拒绝通过穷举搜索在小 $n$ 下计算 $f$ 的函数，维护者只需阅读提交者的报告，并在较大的 $n$ 下寻找此类搜索。我们发布了十个来自极端组合学的问题，提供了二十个已证明的初始记录和录取工具。作为基准，基于前沿语言模型的代理，在每个记录上最多花费一小时，接受了82个提交并改善了每个记录，且没有超出已发布的文献。基于两个更新模型的代理从这些记录开始，接受了60个额外的提交，并再次改善了20个记录中的19个。在十个问题中，至少有一个记录在渐近上仍落后于最佳已发布界限。
cs.AI / 63 / 2610.13479
A Local Agentic AI Framework to Generate Edge Case Scenarios for Safety Assurance
生成边缘案例场景的本地自主人工智能框架以确保安全性
Abstract
Generating edge case scenarios is crucial for training, testing and safety assuring autonomous systems. This identifies hazardous failures to increase system reliability and robustness, ensures the system can still function safely on the boundaries of its operational design domain and learns the expected system behaviour for these edge cases. The rapid advancement in the generation and analysis capabilities of language models (LMs) has expanded their applicability and recent work has investigated LM scenario generation. However, many recent approaches rely on human-in-the-loop or rule-based strategies. This restricts the LM's ability to generate diverse and edge case scenarios; the long tail of the distribution. Our work uses local LMs in a laptop-based agentic AI framework for adaptive scenario generation. This is cheap, fully-controlled, ensures data privacy and security. It uses think, act, observe, update, and continue loops with lightweight JSON encoding to generate scenarios. We use the system requirements, system description and environment description as knowledge bases in contextual prompts for an LM to ensure scenario realism. A second, independent LM critiques each scenario using domain specific criteria so it can be refined. The refined scenarios are run in simulation and evaluation metrics collected. These metric scores are analysed and similar scenarios are created for edge cases (exploitation) or novel scenarios are generated to explore the scenario space. Post-hoc clustering using metric scores and thematic-type analysis of the rich-text scenario descriptions produced by the LM pinpoints functional gaps and their triggering conditions. This creates traceable evidence for the safety assurance case. Our work highlights the potential of a flexible and adaptive local agentic AI framework for automating edge case generation and producing traceable safety assurance evidence.
Chinese Translation
生成边缘案例场景对于训练、测试和确保自主系统的安全性至关重要。这有助于识别危险故障，提高系统的可靠性和稳健性，确保系统在其操作设计域的边界上仍能安全运行，并学习这些边缘案例的预期系统行为。语言模型（LM）生成和分析能力的快速发展扩展了其适用性，近期的研究探讨了LM场景生成。然而，许多近期方法依赖于人机协作或基于规则的策略。这限制了LM生成多样化和边缘案例场景的能力，即分布的长尾。我们的工作在基于笔记本电脑的自主人工智能框架中使用本地LM进行自适应场景生成。这种方法成本低、完全可控，并确保数据隐私和安全。它使用思考、行动、观察、更新和继续循环，通过轻量级JSON编码生成场景。我们使用系统需求、系统描述和环境描述作为知识库，在上下文提示中为LM提供，以确保场景的真实感。第二个独立的LM使用特定领域的标准对每个场景进行批评，以便进行改进。经过改进的场景在仿真中运行，并收集评估指标。这些指标分数被分析，并为边缘案例创建相似场景（利用）或生成新场景以探索场景空间。使用指标分数进行后期聚类和对LM生成的丰富文本场景描述进行主题分析，能够指出功能缺口及其触发条件。这为安全保证案例创建了可追溯的证据。我们的工作突显了灵活和自适应的本地自主人工智能框架在自动生成边缘案例和提供可追溯的安全保证证据方面的潜力。
cs.AI / 64 / 2610.13481
Branching On-Policy Distillation
分支式在线策略蒸馏
Abstract
On-policy distillation (OPD) provides teacher supervision on student-generated trajectories, but a single trajectory realizes only one continuation at each decision, limiting supervision coverage under a finite sampling budget. We propose BranchOPD (Branching On-Policy Distillation), which selects informative decision points using student uncertainty and teacher-student disagreement, resamples student-policy continuations from shared prefixes, and applies token-level teacher supervision to both main trajectories and branches. BranchOPD operates at token-level decisions in mathematics and within-turn decisions in agent tasks. Shared-prefix reuse reduces student-generated token costs by approximately 19% on ALFWorld and 31% on WebShop relative to seven independent trajectories. We also introduce protocol-boundary repair (PBR), which uses targeted supervision under protocol constraints to mitigate repetition and wasted generation. BranchOPD outperforms Vanilla OPD and Multi-Rollout OPD (n=7) across all evaluated benchmarks, with gains over Vanilla OPD of 18.55 percentage points in mean mathematics accuracy over 16 sampled answers (avg@16) and 13.8 points in mean ALFWorld ID/OOD success. Mechanistic analyses and selection ablations show that branching at uncertain decisions expands subsequent behavior coverage and outperforms random selection and turn-initial branching.
Chinese Translation
在线策略蒸馏（On-policy distillation, OPD）为学生生成的轨迹提供教师监督，但单一轨迹在每个决策点仅实现一种延续，这在有限的采样预算下限制了监督覆盖范围。我们提出了分支式在线策略蒸馏（Branching On-Policy Distillation, BranchOPD），该方法利用学生的不确定性和教师-学生之间的分歧来选择信息丰富的决策点，从共享前缀重新采样学生策略的延续，并对主轨迹和分支应用令牌级别的教师监督。BranchOPD 在数学中的令牌级决策和代理任务中的回合内决策中运作。共享前缀的重用使得学生生成的令牌成本在 ALFWorld 上降低了约 19%，在 WebShop 上降低了 31%，相较于七条独立轨迹。我们还引入了协议边界修复（Protocol-Boundary Repair, PBR），该方法在协议约束下使用有针对性的监督以减轻重复和浪费生成。BranchOPD 在所有评估基准上均优于原始 OPD 和多次回合 OPD（n=7），在 16 个采样答案的平均数学准确率上比原始 OPD 提高了 18.55 个百分点，在 ALFWorld ID/OOD 成功率上提高了 13.8 个百分点。机制分析和选择消融实验表明，在不确定决策处进行分支扩展了后续行为的覆盖范围，并优于随机选择和回合初始分支。
cs.AI / 65 / 2610.13494
Humanize MLE Agents: Multi-Agent Session Alternation Beyond Single-Agent Limits
人性化MLE代理：超越单代理限制的多代理会话交替
Abstract
Existing LLM-based agent approaches for long-horizon machine learning engineering (MLE) often use context engineering, implemented through custom harnesses that manage tools, skills, and accumulated experience. However, we find that as single-session execution lengthens, the agent's reasoning intensity declines alongside slowing progress, summarized as long-horizon deceleration, and the benefits of these custom harnesses vanish as the underlying model improves, summarized as harness staleness. These findings suggest that their benefits may not persist across versions, which motivates a workflow-level approach in which agents build on one another's work across sessions. We model these alternating agent sessions as a context-reset semi-Markov decision process (CR-SMDP) and establish sufficient conditions for alternation to exceed single-agent reference limits, derive an explicit experiment-budget bound, and characterize when session caps preserve this advantage. Guided by this analysis, we introduce Humanize MLE Agents (HMA), which alternates two agents on a shared workspace, limits each session to at most k experiments and resets the context between sessions, without additional task-specific harness or prior knowledge supplements. Across 75 Kaggle tasks in MLE-bench, HMA reaches a state-of-the-art any-medal rate of 78.2%, exceeding the strongest open-source baseline by 8.0 points with a quarter of its time budget. Moreover, among six configurations spanning five heterogeneous agent pairs, HMA improves any-medal rate over the constituent baselines by 6.4 points on average, demonstrating cross-pair generality. A supplementary case on kernel optimization reaches 878 cycles, within 1.6% of the community leaderboard leader, demonstrating applicability beyond MLE.
Chinese Translation
现有基于大型语言模型（LLM）的长时间跨度机器学习工程（MLE）代理方法通常采用上下文工程，通过定制的工具管理框架来管理工具、技能和积累的经验。然而，我们发现，随着单会话执行时间的延长，代理的推理强度下降，进展减缓，这被总结为长时间跨度减速，且这些定制框架的优势随着基础模型的改进而消失，被总结为框架陈旧。这些发现表明，它们的优势可能不会在不同版本间持续，这激励我们采用一种工作流程级别的方法，使代理能够在会话间相互构建工作。我们将这些交替的代理会话建模为上下文重置半马尔可夫决策过程（CR-SMDP），并建立了交替超过单代理参考限制的充分条件，推导出明确的实验预算界限，并描述了何时会话上限能够保持这一优势。在这一分析的指导下，我们引入了人性化MLE代理（HMA），该方法在共享工作空间上交替两个代理，将每个会话限制为最多k个实验，并在会话之间重置上下文，而无需额外的任务特定框架或先前知识补充。在MLE-bench的75个Kaggle任务中，HMA达到了78.2%的最新任何奖牌率，超过了最强开源基线8.0个百分点，且仅用其时间预算的四分之一。此外，在涵盖五对异构代理的六种配置中，HMA在构成基线的基础上平均提高了6.4个百分点的任何奖牌率，展示了跨对的普适性。关于内核优化的补充案例达到了878个周期，距离社区排行榜的领先者仅1.6%，展示了其在MLE之外的适用性。
cs.AI / 66 / 2610.13510
Retrieval-Augmented Multimodal Language Model with Heterogeneous Evidence Integration for Clinical Prognosis
基于异构证据整合的检索增强多模态语言模型用于临床预后
Abstract
Clinical prognosis requires integrating structured measurements, physiological waveforms, longitudinal history, and population context. We introduce PRISM (Population-Retrieved Inference for Structured Multimodal Prognosis), a multimodal language model that combines per-feature numerical tokenization, recency-weighted aggregation of longitudinal ECGs, and a dual-modality population memory bank. Retrieval operates in representation space: the model receives mean-pooled embeddings of similar training-cohort encounters, while no individual retrieved record enters the prompt. On the 1,443-task MDS-ED benchmark, PRISM achieves AUROCs of 86.2% for diagnosis, 93.1% for deterioration, 91.8% for ICU admission, and 92.8% for mortality. These results use a 96-hour pre-arrival window unavailable to reported baselines. With benchmark-window inputs only, PRISM reaches 0.8895 AUROC, compared with 0.8937 for UniPACT and 0.8889 for MDS-ED. A matched-input compact fusion model using the same five evidence streams and frozen encoders reaches 0.7652 AUROC, versus 0.9092 for PRISM. Replacing numerical EHR encoding with text serialization reduces AUROC by 3.66 points; removing pre-arrival history reduces it by 1.97 points; and zeroing retrieval reduces it by 0.99 points. Genuine retrieval outperforms random-neighbor retrieval by 0.97 macro-average points, although a paired task-level sign test does not establish a uniform advantage (p=0.067). On the external MC-MED cohort, supervised fine-tuning reaches 0.8611 AUROC, compared with 0.8227 for locally trained gradient boosting and 0.7841 for source-only PRISM. These results support structure-aware multimodal integration and supervised external adaptation.
Chinese Translation
临床预后需要整合结构化测量、物理波形、纵向历史和人群背景。我们提出了PRISM（Population-Retrieved Inference for Structured Multimodal Prognosis），这是一种多模态语言模型，结合了逐特征数值标记、对纵向心电图（ECG）的近期加权聚合以及双模态人群记忆库。检索在表征空间中进行：模型接收相似训练队列遭遇的均值池化嵌入，而没有单个检索记录进入提示。在1,443任务的MDS-ED基准上，PRISM在诊断、恶化、ICU入院和死亡率方面分别达到了86.2%、93.1%、91.8%和92.8%的AUROC。这些结果使用了一个96小时的到达前窗口，而该窗口在报告的基准中不可用。仅使用基准窗口输入，PRISM的AUROC达到了0.8895，而UniPACT为0.8937，MDS-ED为0.8889。一个匹配输入的紧凑融合模型使用相同的五个证据流和冻结编码器达到了0.7652 AUROC，而PRISM为0.9092。用文本序列化替换数值电子健康记录（EHR）编码使AUROC降低了3.66点；去除到达前历史使其降低了1.97点；而将检索归零使其降低了0.99点。真实检索比随机邻居检索高出0.97个宏平均点，尽管配对任务级符号检验未能建立统一优势（p=0.067）。在外部MC-MED队列上，监督微调达到了0.8611 AUROC，而本地训练的梯度提升为0.8227，源仅PRISM为0.7841。这些结果支持结构感知的多模态整合和监督外部适应。
cs.AI / 67 / 2610.13520
Causal Effect Identifiability in Causal Abstractions: A Strict Hierarchy of Criteria
因果抽象中的因果效应可识别性：严格的标准层次结构
Abstract
Identifying causal effects from observational data typically requires a fully specified causal graph, an assumption that is often unrealistic in complex or high-dimensional settings. Causal abstractions address this limitation by retaining only partial causal information, thereby defining equivalence classes of compatible causal graphs. In this paper, we study causal effect identifiability directly at the level of these graph collections, rather than at the level of individual causal diagrams. We formalize several notions of identifiability in this setting and organize them into a hierarchy that clarifies their relationships and differences. In particular, we establish that a causal effect may be identified over a collection of graphs without admitting a single causal derivation that holds uniformly across all compatible graphs. These results help reveal gaps and clarify ambiguities in the existing literature.
Chinese Translation
从观察数据中识别因果效应通常需要一个完全指定的因果图，这一假设在复杂或高维环境中往往不切实际。因果抽象通过仅保留部分因果信息来解决这一限制，从而定义兼容因果图的等价类。本文直接在这些图集合的层面上研究因果效应的可识别性，而不是在单个因果图的层面上。我们在这一背景下形式化了几种可识别性的概念，并将其组织成一个层次结构，以阐明它们之间的关系和差异。特别地，我们建立了一个因果效应可以在一组图上被识别，而不需要承认在所有兼容图上均适用的单一因果推导。这些结果有助于揭示现有文献中的空白并澄清模糊之处。
cs.AI / 68 / 2610.13521
Beyond Domain-Routed Supervision: Cross-Teacher Guarding for Multi-Teacher On-Policy Distillation
超越领域路由监督：多教师在线蒸馏的跨教师保护
Abstract
Multi-teacher on-policy distillation (MOPD) extends on-policy distillation (OPD) by routing domain-specific data to specialized teachers, integrating their capabilities through dense token-level supervision on student-generated trajectories. However, domain routing only distinguishes the supervision sources for different domains; updates from all domains still act on shared parameters. When teachers differ substantially in response style and domain expertise, supervision that is effective in isolation may still lead to persistent gradient conflicts and capability degradation during joint optimization. To address this challenge, we propose Cross-Teacher Guarding for MOPD (Road MOPD), which introduces non-routed teachers as auxiliary cross-domain critics on the same student trajectories and coordinates multi-teacher supervision through token-level credit assignment and gradient updates. Specifically, cross-teacher-guided credit assignment uses teacher agreement to strengthen token-level corrections jointly supported by the teachers and attenuate conflicting signals; preventive multi-teacher gradient projection uses auxiliary gradients from the same trajectories to constrain potentially conflicting updates and adaptively controls projection strength to retain target-domain learning signals. In experiments, Road MOPD mitigates degradation in the disadvantaged domain, gains over 5 percentage points under high conflict, and improves overall performance in low-conflict and three-domain settings.
Chinese Translation
多教师在线蒸馏（MOPD）通过将特定领域的数据路由到专业教师，扩展了在线蒸馏（OPD），并通过对学生生成的轨迹进行密集的标记级监督来整合其能力。然而，领域路由仅区分不同领域的监督来源；来自所有领域的更新仍然作用于共享参数。当教师在响应风格和领域专业知识上存在显著差异时，单独有效的监督仍可能在联合优化过程中导致持续的梯度冲突和能力下降。为了解决这一挑战，我们提出了针对MOPD的跨教师保护（Road MOPD），该方法引入非路由教师作为同一学生轨迹上的辅助跨领域评论者，并通过标记级信用分配和梯度更新协调多教师监督。具体而言，跨教师指导的信用分配利用教师之间的一致性来加强教师共同支持的标记级修正，并减弱冲突信号；预防性多教师梯度投影使用来自同一轨迹的辅助梯度来约束潜在的冲突更新，并自适应控制投影强度以保留目标领域的学习信号。在实验中，Road MOPD减轻了劣势领域的能力下降，在高冲突情况下提高了超过5个百分点，并在低冲突和三领域设置中改善了整体性能。
cs.AI / 69 / 2610.13534
MAGMA-Bench: Evaluating Long-Horizon Embodied Agents under Dynamic Tasks and Execution Failures
MAGMA-Bench：在动态任务和执行失败下评估长时间地体现代理
Abstract
Long-horizon embodied agents must maintain task-relevant information, coordinate heterogeneous robots, and adapt their decisions as execution unfolds. Existing benchmarks investigate dynamic task execution and embodied multi-agent coordination, but provide limited evaluation of their intersection. We introduce MAGMA-Bench, an architecture-agnostic benchmark for persistent task reasoning in dynamic multi-robot environments. MAGMA-Bench evaluates agents across increasing task horizons and controlled online perturbations, including execution failures, task interruptions, and mission updates, with scenarios requiring coordination across robots with heterogeneous capabilities. It uses a structured skill-level interface to separate task-level reasoning from perception and low-level control. We evaluate locally deployable open-weight planners, multi-agent state-tracking strategies, and frontier models. Even the strongest evaluated models fail on a substantial fraction of clean episodes, and their success rates decrease further under online perturbations. Supervised adaptation greatly improves a small model, but performance remains substantially lower on longer and multi-robot missions. We release the benchmark, 261,646 task-level training examples, and an architecture-agnostic expert-trace generation pipeline.
Chinese Translation
长时间地体现代理必须维护与任务相关的信息，协调异构机器人，并在执行过程中调整其决策。现有基准测试研究动态任务执行和体现多智能体协调，但对其交集的评估有限。我们提出了MAGMA-Bench，这是一个与架构无关的基准，用于在动态多机器人环境中进行持续任务推理。MAGMA-Bench在不断增加的任务时间范围和受控的在线扰动下评估代理，包括执行失败、任务中断和任务更新，场景要求在具备异构能力的机器人之间进行协调。它使用结构化的技能级接口将任务级推理与感知和低级控制分离。我们评估了可本地部署的开放权重规划器、多智能体状态跟踪策略和前沿模型。即使是评估的最强模型在相当一部分干净的情节中也失败，并且在在线扰动下成功率进一步下降。监督适应显著提高了一个小模型的性能，但在更长和多机器人任务上的表现仍然显著较低。我们发布了该基准、261,646个任务级训练示例，以及一个与架构无关的专家轨迹生成管道。
cs.AI / 70 / 2610.13570
Constructing Pareto Frontiers of Optimal Reasoning Termination Methods for Overthinking in Reasoning Models
构建推理模型中过度思考的最佳推理终止方法的帕累托前沿
Abstract
Reasoning models can benefit from test-time scaling, but continued reasoning may become ineffective or harmful, a phenomenon termed overthinking. Evaluations against full thinking compare accuracy and token usage but lack references for overthinking severity and strategies' theoretical mitigation potential. To characterize overthinking at the model-task level, we first introduce Step-wise Truncation Probing (STP), repeatedly generating answers from multiple retained prefixes along each fixed sampled reasoning trajectory. Using each trajectory's earliest measured position attaining maximum empirical accuracy as a reference, we define Overthinking Ratio and Accuracy Degradation as the fraction of thinking tokens generated afterward and the accuracy decrease from this reference to full thinking, respectively. Secondly, we then formulate dynamic routing, fixed budgets, and dynamic early exit within a unified framework to quantify the theoretical mitigation potential of different termination strategies. Based on this unified framework, we use the complete STP measurements to construct a token-accuracy Pareto frontier for each strategy under a common normalized thinking-token budget. We further develop OTBench, a reusable benchmark built from the STP process, which produced 70.5B reasoning tokens and the corresponding probing measurements across six models and four benchmark groups.
Chinese Translation
推理模型可以从测试时缩放中受益，但持续推理可能变得无效或有害，这种现象被称为过度思考。针对完整思考的评估比较了准确性和令牌使用情况，但缺乏对过度思考严重性和策略理论缓解潜力的参考。为了在模型-任务层面上表征过度思考，我们首先引入逐步截断探测（Step-wise Truncation Probing, STP），通过在每个固定采样推理轨迹上重复生成来自多个保留前缀的答案。以每条轨迹上最早测得的达到最大经验准确性的点作为参考，我们定义了过度思考比率（Overthinking Ratio）和准确性下降（Accuracy Degradation），分别表示之后生成的思考令牌的比例和从此参考点到完整思考的准确性下降。其次，我们在统一框架内制定动态路由、固定预算和动态提前退出，以量化不同终止策略的理论缓解潜力。基于这一统一框架，我们利用完整的STP测量为每种策略在共同的标准化思考令牌预算下构建了一个令牌-准确性帕累托前沿。我们进一步开发了OTBench，这是一个基于STP过程构建的可重用基准，生成了70.5亿个推理令牌及其在六个模型和四个基准组中的相应探测测量。
cs.AI / 71 / 2610.13594
MENTOR: Learning from Failure with On-Policy Pedagogical Intervention
MENTOR：通过在线教学干预从失败中学习
Abstract
Reinforcement Learning from Verifiable Rewards (RLVR) is a key driver for Large Reasoning Models, but it is vulnerable to "zero-reward collapse", a training failure where a sudden increase in task difficulty causes the learning signal to vanish. We argue that the conventional approach of discarding these failed trajectories is wasteful, as they contain valuable diagnostic information about the model's cognitive gaps, a bright side that can be leveraged for recovery. Existing solutions, however, often regress to static, "off-policy-style" interventions like pre-defined hint libraries or perform post-hoc refinement, failing to provide the adaptive, forward-looking guidance needed to escape the collapse. To address this, we propose a paradigm shift towards on-policy pedagogical intervention and introduce MENTOR, a multi-agent framework that acts as a live mentor. MENTOR diagnoses the root cause of failure in real-time and dynamically generates targeted, instructional prompts to guide the learning agent's next attempt. As a plug-and-play module, MENTOR integrates seamlessly into existing policy optimization pipelines like GRPO without modifying the training infrastructure. Extensive experiments on math and code generation show that MENTOR effectively mitigates zero-reward collapse, outperforming state-of-the-art baselines. Notably, it achieves massive average relative gains of up to 78.3% on Llama-3.2-3B-Instruct over its base model.
Chinese Translation
可验证奖励的强化学习（Reinforcement Learning from Verifiable Rewards, RLVR）是大型推理模型的重要驱动力，但它容易受到“零奖励崩溃”的影响，这是一种训练失败现象，在任务难度突然增加时，学习信号消失。我们认为，传统的丢弃这些失败轨迹的方法是浪费的，因为它们包含关于模型认知缺口的宝贵诊断信息，这一积极方面可以被利用来实现恢复。然而，现有的解决方案通常退化为静态的“离策略风格”干预，如预定义提示库，或进行事后修正，未能提供逃离崩溃所需的自适应、前瞻性指导。为了解决这个问题，我们提出了向在线教学干预的范式转变，并引入了MENTOR，一个作为实时导师的多智能体框架。MENTOR实时诊断失败的根本原因，并动态生成有针对性的教学提示，以指导学习智能体的下一次尝试。作为一个即插即用模块，MENTOR可以无缝集成到现有的策略优化管道中，如GRPO，而无需修改训练基础设施。在数学和代码生成的广泛实验中，MENTOR有效地缓解了零奖励崩溃，超越了最先进的基线。值得注意的是，它在Llama-3.2-3B-Instruct模型上实现了高达78.3%的平均相对增益，显著优于其基础模型。
cs.AI / 72 / 2610.13632
Humanize Olympiad Agents: A Multi-Session Workflow for Olympiad Problem Solving with Solve, Review, Wiki, Verify
人性化奥林匹克代理：一种通过解决、审查、维基、验证的多会话工作流程用于奥林匹克问题解决
Abstract
Large language model (LLM) agents are rapidly improving on long-horizon scientific and olympiad tasks. Recent systems have produced a Lean-formalized resolution of the Navier--Stokes problem and autonomous solutions to four open questions in the Erd\H{o}s Conjectures database. Yet neither system releases both its underlying model and complete inference stack. More broadly, a single agent session remains brittle on difficult problems: early errors can propagate, missing domain knowledge can stall progress, and apparent success can mask a misinterpretation or weakening of the original task. We introduce HOA (Humanize Olympiads Agent), an open, flow-level framework that coordinates multiple agent sessions through three complementary feedback channels. First, an independent reviewer evaluates each candidate and provides actionable feedback. Second, a structured wiki supplies task-specific knowledge that may be missing or unreliable in model pretraining. Third, a verifier checks generated Lean proofs against trusted challenge statements. With GPT-5.6 Sol, HOA achieves perfect scores on the IMO, IPhO, and IOI 2026, near-full marks on the IChO 2026 problem set, and solves all 672 problems (100%) on PutnamBench. On the Lean-Eval leaderboard, HOA ranks first, solving 251 of 304 problems (82.6%). We further evaluate HOA with open models. Using Kimi-K3, HOA achieves full scores on the IMO, IPhO, and IOI, and a gold-level score of 97.1% on the IChO. These results closely match the closed-source GPT-5.6 Sol + /Goal baseline on the IMO and IPhO, while exceeding it on the IChO and IOI, demonstrating that systematic feedback can help bridge substantial differences in base-model capability.
Chinese Translation
大型语言模型（LLM）代理在长时间跨度的科学和奥林匹克任务上正在迅速提升。最近的系统已经产生了对Navier–Stokes问题的Lean形式化解决方案，并对Erdős猜想数据库中的四个开放问题提供了自主解决方案。然而，这两个系统都没有发布其基础模型和完整的推理堆栈。更广泛地说，单个代理会话在处理困难问题时仍然脆弱：早期错误可能会传播，缺失的领域知识可能会阻碍进展，而表面上的成功可能掩盖了对原始任务的误解或削弱。我们引入了HOA（人性化奥林匹克代理），这是一个开放的流级框架，通过三个互补的反馈渠道协调多个代理会话。首先，独立的审查员评估每个候选者并提供可操作的反馈。其次，结构化的维基提供任务特定的知识，这些知识在模型预训练中可能缺失或不可靠。第三，验证者检查生成的Lean证明与可信的挑战声明的一致性。使用GPT-5.6 Sol，HOA在IMO、IPhO和IOI 2026中获得满分，在IChO 2026问题集中接近满分，并在PutnamBench上解决了672个问题（100%）。在Lean-Eval排行榜上，HOA排名第一，解决了304个问题中的251个（82.6%）。我们进一步使用开放模型评估HOA。使用Kimi-K3，HOA在IMO、IPhO和IOI中获得满分，并在IChO中获得97.1%的金级分数。这些结果与闭源的GPT-5.6 Sol + /Goal基线在IMO和IPhO上的表现非常接近，同时在IChO和IOI上超过了该基线，证明系统反馈可以帮助弥补基础模型能力之间的重大差异。
cs.AI / 73 / 2610.13640
Learning Size Independent Constraint Models from Input Data and Positive and Negative Example Solutions
从输入数据及正负例解中学习与大小无关的约束模型
Abstract
In this paper we describe a Constraint Acquisition Tool that generates size independent constraint programs from input data and positive and negative example solutions of multiple problem instances. The system finds size specific models based on typical constraint patterns over global and binary constraints, then generalizes the models to a size independent model that can be run on unseen input data. The system produces MiniZinc code as well as a textual description of the model. We evaluate the tool on the example problems of the PTHG21 Challenge and show the feasibility of the approach.
Chinese Translation
在本文中，我们描述了一种约束获取工具，该工具能够从输入数据以及多个问题实例的正负例解中生成与大小无关的约束程序。该系统基于全局约束和二元约束的典型约束模式寻找特定大小的模型，然后将这些模型推广为可以在未见输入数据上运行的与大小无关的模型。该系统生成 MiniZinc 代码以及模型的文本描述。我们在 PTHG21 挑战的示例问题上评估了该工具，并展示了该方法的可行性。
cs.AI / 74 / 2610.13663
A Multi-Agent LLM System for Automated Tuning of Specialized Neural Network Models
用于自动调优专业神经网络模型的多智能体大语言模型系统
Abstract
Many machine learning approaches are two-level: an expert selects a class of methods, within this class the choice of key components is critical and usually manual. This paper addresses the automatic tuning of these components and proposes a framework transferable across models of different nature. The multi-agent large language model (LLM) framework has three roles---analyst, coder, and recorder~---with an experiment log and a sandboxed execution environment. Unlike systems that optimize the entire ML script or search architectures in a discrete space, the framework operates over an open but semantically narrow space: the agent generates code only for the specified components, leaving the overall network structure unchanged. The approach is validated in two directions: tuning a classification spiking neural network with correlation learning, and solving high-order differential equations with Physics-Informed Neural Networks (PINN). Across five tasks---spiking classification on MNIST and Digits, and boundary-value problems for the generalized nonlinear Schr\"odinger equation (GNLSE) and Kuramoto--Sivashinsky (KS) equations---the agent outperforms the baselines: classification accuracy is 13 percentage points (MNIST) and 6 percentage points (Digits) higher; the error against the analytical single-soliton solution is reduced approximately 2491-fold; the invariant-preservation error drops approximately 325-fold in the selected GNLSE run with a Gaussian perturbation and by more than 100-fold on average over repeated initializations; and the mass drift drops 14-fold for the KS problem with a Gaussian perturbation. The framework thus tunes key components on tasks of different nature without changing its internal structure.
Chinese Translation
许多机器学习方法采用两级结构：专家选择一类方法，在这一类中，关键组件的选择至关重要且通常是手动的。本文针对这些组件的自动调优进行了探讨，并提出了一个可跨不同性质模型迁移的框架。该多智能体大语言模型（LLM）框架具有三种角色——分析师、编码者和记录者——并配备实验日志和沙箱执行环境。与优化整个机器学习脚本或在离散空间中搜索架构的系统不同，该框架在一个开放但语义狭窄的空间中运行：智能体仅为指定组件生成代码，而不改变整体网络结构。该方法在两个方向上得到了验证：使用相关学习调优分类脉冲神经网络，以及利用物理信息神经网络（PINN）求解高阶微分方程。在五个任务中——在 MNIST 和 Digits 上的脉冲分类，以及广义非线性薛定谔方程（GNLSE）和 Kuramoto–Sivashinsky（KS）方程的边值问题——智能体的表现优于基线：分类准确率在 MNIST 上提高了 13 个百分点，在 Digits 上提高了 6 个百分点；与解析单孤子解的误差减少了约 2491 倍；在选定的 GNLSE 运行中，带有高斯扰动的不变性保持误差减少了约 325 倍，在多次初始化的平均值上减少了超过 100 倍；而在带有高斯扰动的 KS 问题中，质量漂移减少了 14 倍。因此，该框架在不同性质的任务上调优关键组件，而不改变其内部结构。
cs.AI / 75 / 2610.13669
A Step Harder, Not a Step Too Far: The Benefits and Limits of Curriculum Difficulty Progression in GRPO
更进一步，而非过于冒进：GRPO中课程难度进展的益处与局限
Abstract
Curriculum learning typically organizes training samples from easy to hard. In GRPO, medium-difficulty samples are attractive for online selection because they are more likely to yield informative group-relative learning signals, and prioritizing them can improve training efficiency. However, it remains unclear whether persistently selecting such samples also leads to the strongest final model performance, or whether training can benefit from progressing toward harder samples. Meanwhile, sample difficulty changes as the policy evolves, so fixed difficulty estimates can become misaligned with the current policy. We therefore re-estimate sample difficulty under the current policy and, inspired by the easy-to-hard principle of curriculum learning, design and evaluate a staged curriculum that progressively shifts the active difficulty band toward harder samples. Experiments on GSM8K show that persistent medium-difficulty selection does not achieve the strongest observed performance. Across three random seeds, progressing to a moderately harder difficulty band improves test accuracy to $0.6591 \pm 0.0077$, compared with $0.6449 \pm 0.0024$ for the fixed-medium curriculum. However, further progression to a harder difficulty band reduces accuracy to $0.6405 \pm 0.0053$. These results show that staged difficulty progression can outperform persistent medium-difficulty selection, but its benefit is non-monotonic. Effective GRPO curricula can therefore benefit from progressing beyond medium difficulty, while avoiding excessive progression toward overly difficult samples.}
Chinese Translation
课程学习通常将训练样本从简单到困难进行组织。在GRPO中，中等难度的样本因其更可能产生有信息量的群体相对学习信号而受到在线选择的青睐，优先选择这些样本可以提高训练效率。然而，目前尚不清楚持续选择这些样本是否也能导致最终模型性能的最佳表现，或者训练是否可以从向更困难样本的进展中受益。同时，随着策略的演变，样本的难度也在变化，因此固定的难度估计可能与当前策略不匹配。因此，我们在当前策略下重新估计样本的难度，并受到课程学习中从易到难原则的启发，设计并评估一个逐步课程，该课程逐渐将活动难度带向更困难的样本。在GSM8K上的实验表明，持续选择中等难度样本并未实现观察到的最佳性能。在三个随机种子下，向中等更困难的难度带进展将测试准确率提高至$0.6591 imes 0.0077$，而固定中等课程的准确率为$0.6449 imes 0.0024$。然而，进一步向更困难的难度带进展则将准确率降低至$0.6405 imes 0.0053$。这些结果表明，分阶段的难度进展可以优于持续的中等难度选择，但其益处并非单调的。因此，有效的GRPO课程可以从超越中等难度的进展中受益，同时避免向过于困难样本的过度进展。
cs.AI / 76 / 2610.13703
A Baseline Assessment for the Agentic Web Technology Stack
代理网络技术栈的基线评估
Abstract
The fast pace in which protocols, frameworks, and technologies are being proposed for LLM-based agents results in a complex scenario for anyone aiming to create state-of-the-art AI solutions. Without an assessment on how different protocols, frameworks, and libraries work together to compose an Agentic Web solution, developers and organizations face different challenges (e.g., skilling, governance, infrastructure) even before becoming part of the Agentic Web ecosystem. In this context, we propose a system for performing Agentic Web evaluations called MASTIF (Multi-Agent System TestIng Framework) and present a baseline assessment of Agentic Web technologies, including 3 LLMs, 3 protocols, and 4 agent implementation frameworks, against 300 Mind2Web benchmark tasks. Results suggest that selecting a task-specific technology stack may result in gains of task completion rate, efficiency score, and social value stability scores; and reductions in verbosity and time to complete tasks. With the presented results, we expect to present a baseline in terms of metrics and results for this complex and fast evolving scenario for those aiming at designing, assessing, and developing multi-agent systems.
Chinese Translation
在基于大型语言模型（LLM）的代理所提出的协议、框架和技术迅速发展的背景下，任何希望创建最先进的人工智能解决方案的人都面临着复杂的情境。在没有对不同协议、框架和库如何协同工作以构成代理网络解决方案进行评估的情况下，开发者和组织在成为代理网络生态系统的一部分之前就面临着不同的挑战（例如，技能培训、治理、基础设施）。在这种背景下，我们提出了一种用于执行代理网络评估的系统，称为MASTIF（多代理系统测试框架），并对代理网络技术进行了基线评估，包括3个LLM、3个协议和4个代理实现框架，针对300个Mind2Web基准任务进行评测。结果表明，选择特定任务的技术栈可能会提高任务完成率、效率评分和社会价值稳定性评分，同时减少冗长性和完成任务所需的时间。通过呈现的结果，我们期望为那些旨在设计、评估和开发多代理系统的人提供一个关于这一复杂且快速发展的情境的基线指标和结果。
cs.AI / 77 / 2610.13708
Adaptive DHiSparse: Speculative Decoding over a Load-Adaptive KV Hierarchy
自适应 DHiSparse：基于负载自适应 KV 层次结构的推测解码
Abstract
Long-context serving runs out of KV memory before the hardware reaches its useful throughput. HiSparse stores history in host memory and fetches only the selected entries, but it uses one cache budget at every load and does not handle draft-model speculative decoding such as DSpark and DFlash2. We present Adaptive DHiSparse, which supports speculative decoding over hierarchical sparse KV and adapts residency to load. It protects the required working set, retains additional history while GPU memory is spare, and recovers the rest through exact fused retrieval. DSpark and DFlash2 keep their proposal and acceptance rules: queries share missing-row retrieval, draft KV uses a window-bounded ring, and verification staging is reused across layer groups. At 100K context and concurrency 128, GLM-5.3 with DFlash2 delivers 793 output tokens/s per decode GPU, 1.45$\times$ HiSparse and 1.27$\times$ non-speculative adaptive serving. Held to the same latency bounds as HiSparse, it sustains up to 4$\times$ the concurrency and 2.3 to 3.7$\times$ the output, depending on the per-token bound. On DeepSeek-V4.1-Flash at 500K context and concurrency 512, Adaptive DHiSparse with DSpark delivers 1.85$\times$ the output of resident serving with DSpark. In a load step, optional residency reduces host fetches from 14.2% to 1.4% of selected rows at low load and releases memory under pressure.
Chinese Translation
长上下文服务在硬件达到有效吞吐量之前耗尽了 KV 内存。HiSparse 将历史存储在主机内存中，仅提取所选条目，但在每次加载时使用一个缓存预算，并且不处理诸如 DSpark 和 DFlash2 的草稿模型推测解码。我们提出了自适应 DHiSparse，它支持在层次稀疏 KV 上进行推测解码，并根据负载调整驻留时间。它保护所需的工作集，在 GPU 内存充足时保留额外的历史，并通过精确的融合检索恢复其余部分。DSpark 和 DFlash2 保持其提议和接受规则：查询共享缺失行检索，草稿 KV 使用窗口限制的环，验证阶段在层组之间重用。在 100K 上下文和并发 128 的情况下，使用 DFlash2 的 GLM-5.3 每个解码 GPU 提供 793 个输出标记/秒，比 HiSparse 高 1.45 倍，比非推测自适应服务高 1.27 倍。在与 HiSparse 相同的延迟限制下，它在并发性上可维持高达 4 倍，输出量可达 2.3 到 3.7 倍，具体取决于每个标记的限制。在 500K 上下文和并发 512 的 DeepSeek-V4.1-Flash 上，自适应 DHiSparse 与 DSpark 的结合提供了比 DSpark 的驻留服务高 1.85 倍的输出。在负载步骤中，可选的驻留将主机提取从低负载时的 14.2% 降低到 1.4%，并在压力下释放内存。
cs.AI / 78 / 2610.13744
A Privileged Teacher That Only Scores Cannot Search: Critique-Conditioned On-Policy Self-Distillation for Reasoning
仅凭分数无法搜索的特权教师：基于批评的在线自蒸馏推理方法
Abstract
On-policy self-distillation promises to replace the sparse outcome signal of verifiable-reward RL with dense token-level credit: a teacher conditioned on the reference solution scores the student's own rollout, and the resulting distributional shift is distilled back into the policy. We investigate whether this scoring signal actually contains corrective credit. Using gold error spans in student rollouts, we test the two signals consumed by self-distillation objectives, the full-distribution KL and the token log-ratio, for two properties: whether they localize errors and whether they push probability away from erroneous student tokens. They exhibit neither of these properties nor can they reliably distinguish correct from incorrect rollouts within the same problem. We argue that this failure is structural: locating an error in a reasoning trace is itself a search problem, while a scoring teacher is never explicitly asked to perform that search. The same model, with the same privileged information, localizes errors substantially better when allowed to generate a critique. This forms the motivation for \emph{critique-conditioned self-distillation}: first generate a critique of the current rollout, then condition the scoring teacher on that critique. The resulting signal localizes errors, is corrective within error spans, and shifts probability toward tokens that initiate backtracking such as ``Wait'', ``But'', and ``Actually''. We further show that the critique is useful both at test time, where resuming a wrong rollout with the critique in context converts math problems the model cannot solve by sampling, and in cold-start GRPO, where it provides an advantage signal on prompts for which verifiable rewards alone provide none.
Chinese Translation
在线自蒸馏有望用密集的标记级别信用替代可验证奖励强化学习的稀疏结果信号：一个以参考解为条件的教师对学生的自我回滚进行评分，随后将产生的分布偏移蒸馏回策略。我们研究这种评分信号是否确实包含纠正性信用。通过使用学生回滚中的黄金错误范围，我们测试了自蒸馏目标所消耗的两种信号，即全分布KL和标记对数比率，针对两个特性：它们是否能够定位错误，以及它们是否能够将概率推离错误的学生标记。结果显示，这两种信号均未表现出这两个特性，也无法可靠地区分同一问题中的正确和错误回滚。我们认为这一失败是结构性的：在推理轨迹中定位错误本身就是一个搜索问题，而评分教师从未被明确要求执行该搜索。当允许同一模型生成批评时，它在定位错误方面表现得更为出色。这为基于批评的自蒸馏提供了动机：首先生成当前回滚的批评，然后将评分教师以该批评为条件。生成的信号能够定位错误，在错误范围内具有纠正性，并将概率向那些启动回溯的标记（如“等一下”、“但是”和“实际上”）倾斜。我们进一步表明，批评在测试时同样有用，在上下文中恢复错误回滚时，批评能够将模型无法通过采样解决的数学问题转化为可解问题；在冷启动的GRPO中，它为那些仅凭可验证奖励无法提供优势信号的提示提供了优势信号。
cs.AI / 79 / 2610.13783
UkrMeme: Multimodal Explainable Persuasion Detection in Ukrainian Memes
UkrMeme：乌克兰表情包中的多模态可解释说服检测
Abstract
Memes play a central role in modern online communication, especially in contexts of political conflict where visual and textual cues combine to convey persuasive messages. We study persuasive technique detection in Ukrainian memes using vision-language models (VLMs). We introduce UkrMeme, a dataset of 200 organically produced Ukrainian wartime memes annotated for persuasion techniques, and PTM-UA, a Ukrainian translation of a subset of the Propaganda 950 benchmark that preserves original visual content while translating meme text. We evaluate VLMs in a zero-shot prompting setup across image, text, and multimodal inputs using hierarchical F1 metrics. Finally, we compare Layer-wise Relevance Propagation (LRP) and model-generated token attributions and analyze the semantic categories of evidence emphasized by each explanation method.
Chinese Translation
表情包在现代在线交流中扮演着核心角色，尤其是在政治冲突的背景下，视觉和文本线索结合传达说服性信息。我们使用视觉-语言模型（VLMs）研究乌克兰表情包中的说服技巧检测。我们介绍了UkrMeme，一个包含200个有机生成的乌克兰战争时期表情包的数据集，并对其进行了说服技巧的注释。此外，我们还推出了PTM-UA，这是Propaganda 950基准的一部分的乌克兰语翻译，保留了原始视觉内容，同时翻译了表情包文本。我们在零样本提示设置下评估VLMs，使用层次F1指标对图像、文本和多模态输入进行评估。最后，我们比较了逐层相关传播（Layer-wise Relevance Propagation, LRP）和模型生成的标记归因，并分析了每种解释方法强调的证据的语义类别。
cs.AI / 80 / 2610.13840
Studying Comics at Scale with Artificial Intelligence: The PAGE Framework and Three Studies of American Comics and Japanese Manga
利用人工智能大规模研究漫画：PAGE框架及三项关于美国漫画和日本漫画的研究
Abstract
Comics are one of the great visual records of the twentieth century: hundreds of thousands of pages in which a mass readership saw its leaders, its wars and its heroes. Most of this record has never been studied as a whole; comics scholarship proceeds one book at a time. We argue that artificial intelligence, including generative models, can change this to the benefit of the medium itself. We propose PAGE, a framework that makes artificial intelligence an instrument for comics research. It parses every scan into panels, text and a bibliographic record; annotates each page with a vision-language model that follows a written codebook, as a human coder would; grounds samples of these readings in human annotation; and explains change across an archive, with estimates controlled for publisher and genre. We demonstrate the framework on 3,929 American comic books (1938--1956) and 87 volumes of Japanese manga, in three studies, each resting on a different capability. The first pairs text search with visual reading to recover how a mass medium portrayed political power. The second measures the page's visual form across an industry and recovers the hand of individual artists. The third follows who speaks across two comics traditions, checked against human annotation. Beyond research, the same instruments can make archives searchable by what is drawn, open comics to blind and low-vision readers, and help preserve fragile collections and credit uncredited artists. We outline a research agenda in which artificial intelligence strengthens the study, preservation and reach of the medium.
Chinese Translation
漫画是二十世纪伟大的视觉记录之一：数十万页中，广大读者看到了他们的领袖、战争和英雄。然而，这一记录大多数从未作为一个整体进行研究；漫画学术研究通常是逐本进行的。我们认为，人工智能，包括生成模型，可以改变这一现状，从而有利于这一媒介本身。我们提出了PAGE框架，使人工智能成为漫画研究的工具。该框架将每个扫描解析为面板、文本和书目记录；使用视觉-语言模型对每一页进行注释，该模型遵循书面代码本，类似于人类编码者的工作；将这些阅读样本与人类注释进行对照；并解释档案中的变化，同时控制出版商和类型的影响。我们在3,929本美国漫画（1938-1956年）和87卷日本漫画的三项研究中展示了该框架，每项研究基于不同的能力。第一项研究将文本搜索与视觉阅读相结合，以恢复大众媒介如何描绘政治权力。第二项研究测量整个行业的页面视觉形式，并恢复个别艺术家的风格。第三项研究追踪在两种漫画传统中谁在发声，并与人类注释进行核对。除了研究之外，这些工具还可以使档案根据所绘内容可搜索，帮助盲人和低视力读者接触漫画，并有助于保护脆弱的收藏和给予未被认可艺术家的应有信用。我们概述了一个研究议程，其中人工智能增强了对这一媒介的研究、保护和传播。
cs.AI / 81 / 2610.13886
X-AgentRed: Red-Teaming Personal Agents in Evolving Multi-Service Ecosystems
X-AgentRed：在不断演变的多服务生态系统中对个人代理进行红队测试
Abstract
Personal agents increasingly connect to services across the web and applications by using tools. As the effects of their actions accumulate and interact across services, unintended or harmful outcomes can arise when agents fulfill benign user requests. Proactively identifying these compositional failures requires systematically exploring diverse cross-service paths. Existing red-teaming methods often simply prompt LLMs to identify potentially harmful paths from task and environment descriptions or decompose harmful objectives specified by humans or LLMs into seemingly benign action combinations. However, generation is biased toward patterns learned during training, potentially favoring familiar tool combinations and leaving novel cross-service interactions underexplored. To address this, we propose X-AgentRed, a self-evolving red-teaming framework that discovers harmful tool call paths through a graph of tool dependencies and expands and refines them using execution feedback. By analyzing each tool's effects, potential harms, and preconditions for execution, the red-teaming agent constructs a directed graph that captures how one tool's effects can satisfy another's execution preconditions. Recipes sampled from this graph are instantiated as realistic scenarios in which agents can produce harmful outcomes. Execution feedback enables self-improvement by diversifying paths that produce harm through tool substitution and refining preconditions at blocked steps in unsuccessful paths to enable further execution. Across four frontier agentic models, X-AgentRed achieves $1.77\times$ the harmful path rate of STAC and explores a more diverse set of tools, yielding $2.99\times$ the tool coverage on average.
Chinese Translation
个人代理越来越多地通过工具连接到网络和应用程序中的服务。随着它们的行为效果在服务之间累积和交互，当代理满足良性用户请求时，可能会出现意想不到或有害的结果。主动识别这些组合性失败需要系统性地探索多样的跨服务路径。现有的红队测试方法通常仅仅提示大型语言模型（LLMs）从任务和环境描述中识别潜在的有害路径，或将由人类或LLMs指定的有害目标分解为看似良性的行动组合。然而，生成过程往往偏向于训练期间学习到的模式，可能偏好熟悉的工具组合，从而使新颖的跨服务交互未得到充分探索。为了解决这个问题，我们提出了X-AgentRed，一个自我演化的红队测试框架，通过工具依赖图发现有害的工具调用路径，并利用执行反馈进行扩展和优化。通过分析每个工具的效果、潜在危害和执行前提条件，红队测试代理构建了一个有向图，捕捉一个工具的效果如何满足另一个工具的执行前提条件。从该图中采样的配方被实例化为代理可能产生有害结果的现实场景。执行反馈通过工具替换多样化产生危害的路径，并在不成功路径的阻塞步骤中优化前提条件，以便进一步执行，从而实现自我改进。在四个前沿代理模型中，X-AgentRed实现了$1.77 imes$的有害路径率，相较于STAC，并探索了更为多样的工具，平均实现了$2.99 imes$的工具覆盖率。
cs.AI / 82 / 2610.13889
RATED: Reliability-Aware AI Agent Tool Retrieval via Execution Dynamics
RATED：基于执行动态的可靠性意识AI代理工具检索
Abstract
As AI agents increasingly rely on external tools, accurate retrieval is essential for reliable behavior. Existing methods optimize semantic relevance but largely ignore tool availability during execution. To address this gap, we propose RATED (Reliability-Aware Tool Retrieval via Execution Dynamics), a retrieval framework that combines robust semantic matching with dynamic reliability adaptation. RATED employs a frozen dual-encoder with contrastive fine-tuning for semantic matching, and a lightweight contextual bandit module that re-ranks candidates based on real-time execution feedback. This design enables rapid adaptation to runtime tool failures without expensive model retraining. We introduce EcoAgent, a benchmark of 5,172 real-world tools and over 210,000 query-tool pairs for offline retrieval and dynamic reliability evaluation. Experimental results demonstrate that RATED achieves a 52.2% relative MRR improvement over the ToolLLM API retriever on real-world queries, and reduces fault selection rate from 86.2% to 2.9% under severe API outage scenarios with a 30% failure rate, while incurring negligible latency overhead of less than 1 ms over encode-only inference. The code and dataset are available at https://github.com/Elliotes/EcoAgent.
Chinese Translation
随着AI代理越来越依赖外部工具，准确的检索对于可靠行为至关重要。现有方法优化语义相关性，但在执行过程中大多忽视工具的可用性。为了解决这一问题，我们提出了RATED（基于执行动态的可靠性意识工具检索），这是一个结合了稳健语义匹配与动态可靠性适应的检索框架。RATED采用冻结的双编码器与对比微调进行语义匹配，并使用轻量级上下文强盗模块根据实时执行反馈对候选项进行重新排序。该设计使得在运行时工具故障情况下能够快速适应，而无需昂贵的模型重新训练。我们引入了EcoAgent，这是一个包含5172个真实工具和超过210,000个查询-工具对的基准，用于离线检索和动态可靠性评估。实验结果表明，RATED在真实查询上相较于ToolLLM API检索器实现了52.2%的相对平均排名提升，并且在严重API故障场景下（故障率为30%）将故障选择率从86.2%降低至2.9%，同时引入的延迟开销不足1毫秒，低于仅编码推理的延迟。代码和数据集可在https://github.com/Elliotes/EcoAgent获取。
cs.AI / 83 / 2610.13948
Can Language Models Play the Wikipedia Game?
语言模型能否参与维基百科游戏？
Abstract
Language models increasingly support or automate information retrieval from document collections too large to index. This task is difficult when it requires specialist linguistic or scientific knowledge. We used the Wikipedia Game to test language models on an analogous information retrieval task. In the game, players must reach a target article using only article links. We compared sentence embedding models and LLMs across five Wikipedia languages and five academic-literature datasets. We construct very large local graphs of Wikipedia pages and of paper citations using OpenAlex and provide code to build and navigate them. This allows the study of language models' navigation capabilities without relying on third-party APIs. We compared the models with human and TF-IDF benchmarks. The Wikipedia Game format enables robust comparisons across techniques, languages, and domains. No model significantly outperformed humans, although the best models (GPT-o3 mini, GPT-4o, QwQ:32B, and DeepSeek-R1:70B) achieved comparable success. Performance and rankings varied across languages and topics, with implications for practitioners utilising these techniques.
Chinese Translation
语言模型越来越多地支持或自动化从过于庞大的文档集合中检索信息。当任务需要专业的语言或科学知识时，这一过程变得尤为困难。我们使用维基百科游戏来测试语言模型在类似信息检索任务中的表现。在游戏中，玩家必须仅通过文章链接到达目标文章。我们比较了句子嵌入模型和大型语言模型（LLMs）在五种维基百科语言和五个学术文献数据集上的表现。我们使用OpenAlex构建了非常大的维基百科页面和论文引用的本地图，并提供了构建和导航这些图的代码。这使得研究语言模型的导航能力不再依赖于第三方API。我们将这些模型与人类和TF-IDF基准进行了比较。维基百科游戏的格式使得在技术、语言和领域之间进行稳健比较成为可能。尽管没有模型显著优于人类，但最佳模型（GPT-o3 mini、GPT-4o、QwQ:32B和DeepSeek-R1:70B）取得了可比的成功。不同语言和主题的表现和排名各异，这对利用这些技术的从业者具有重要意义。
cs.CL / 1 / 2610.12545
AutoIndexer: Flexible-Order Decoding by Training Causal Models on Chains of Edits
AutoIndexer：通过在编辑链上训练因果模型实现灵活顺序解码
Abstract
Autoregressive decoding is less suited to non-causally ordered data, e.g., cross-references or future dependencies, since their outputs cannot be revised post hoc, unlike iterative human reasoning. Diffusion LLMs partially mitigate this limitation but still cannot flexibly insert or remove arbitrary numbers of tokens from a pre-generated context. We design AutoIndexer as a generative model with a novel training and decoding mechanism that simulates a Chain of Edits, allowing the model to flexibly modify its prior outputs in hindsight via token insertions, substitutions, and deletions. The model learns to predict marker tokens for opening and closing edits, simultaneously moving a cursor to positions in the existing token sequence, analogous to a text editor. Training sequences are obtained by perturbing ground-truth labels with inverse chained edit operations, and the model learns to edit by undoing those inverses. The perturbation training allows AutoIndexer to be robust to erroneous context and adaptable to editing tasks such as code correction. We demonstrate that our method exhibits post-generation context-editing capabilities that most autoregressive and diffusion models lack without compromising performance on standard LLM benchmarks. The code is available at https://github.com/AutodeskAILab/autoindexer.
Chinese Translation
自回归解码不太适合非因果顺序数据，例如交叉引用或未来依赖，因为与迭代的人类推理不同，它们的输出无法事后修正。扩散大语言模型（Diffusion LLMs）在一定程度上缓解了这一限制，但仍无法灵活地从预生成的上下文中插入或删除任意数量的标记。我们设计了AutoIndexer，作为一种生成模型，采用新颖的训练和解码机制，模拟编辑链，使模型能够通过标记的插入、替换和删除灵活地修改其先前的输出。该模型学习预测用于开启和关闭编辑的标记，同时将光标移动到现有标记序列中的位置，类似于文本编辑器。训练序列通过对真实标签进行逆向链式编辑操作的扰动获得，模型通过撤销这些逆操作来学习编辑。扰动训练使AutoIndexer对错误上下文具有鲁棒性，并能够适应代码修正等编辑任务。我们证明了我们的方法展现了生成后上下文编辑的能力，而大多数自回归和扩散模型在这一点上存在缺陷，同时在标准大语言模型基准测试中没有妥协性能。代码可在 https://github.com/AutodeskAILab/autoindexer 获取。
cs.CL / 2 / 2610.12569
SWE-Duel: Towards a Self-Scaling Arena for Coding Agents
SWE-Duel：面向自我扩展的编码代理竞技场
Abstract
Static coding benchmarks are fixed at release. Over time their tasks leak into training data, and stronger agents eventually saturate them. We argue that a benchmark should instead aim to be self-scaling: its tasks and oracles are created at evaluation time, so no content exists in advance to leak, and its difficulty is set by the participating agents rather than fixed at release. We introduce SWE-Duel, an arena in which two coding agents evaluate each other over mirrored Red-Blue turns. In each turn, the Red agent works on a real repository across three context-isolated sessions. It proposes and implements a new feature with feature tests, then embeds a subtle bug in that feature along with sealed bug-triggering tests. The resulting pull request is admitted only if (1) it passes the feature tests and the repository's existing tests while failing exactly the bug tests, and (2) a fresh session of the same agent can repair the bug blind, which rules out unsolvable tasks. The Blue agent then reviews the pull request without being told a bug exists. It wins only if its revision passes all three test suites, and then the two agents swap roles. We run a tournament among 11 model-harness participants on 15 repositories in five languages for 1,500 turns. The leaders win in different ways. GPT-5.5 with Codex CLI ranks first as the strongest solver, repairing 98.3% of the defects it reviews. Claude Opus 4.8 with mini-swe-agent ranks second as the strongest task author, with a 27.3% Red win rate. Finally, within one model family, tasks written by the newest version defeat a fixed-reference solver far more often than those of older versions (42.9% vs. 4.0%), giving preliminary evidence that task difficulty scales with task authoring capability.
Chinese Translation
静态编码基准在发布时是固定的。随着时间的推移，它们的任务会泄露到训练数据中，强大的代理最终会饱和这些任务。我们认为，基准应该旨在自我扩展：其任务和oracle在评估时创建，因此在此之前不存在可泄露的内容，其难度由参与的代理设定，而不是在发布时固定。我们引入了SWE-Duel，一个竞技场，在这里两个编码代理通过镜像的红蓝回合相互评估。在每个回合中，红方代理在三个上下文隔离的会话中处理一个真实的代码库。它提出并实现一个新特性，并进行特性测试，然后在该特性中嵌入一个微妙的bug，并附上密封的触发bug的测试。只有在（1）它通过特性测试和代码库现有测试，同时恰好未通过bug测试，以及（2）同一代理的新会话能够盲目修复该bug，这样才能接受生成的拉取请求，从而排除无法解决的任务。然后，蓝方代理在不知道存在bug的情况下审查拉取请求。只有当其修订通过所有三个测试套件时，它才获胜，然后两个代理交换角色。我们在15个代码库上对11个模型参与者进行了一场比赛，涵盖五种语言，共进行1500个回合。领导者以不同的方式获胜。使用Codex CLI的GPT-5.5以98.3%的缺陷修复率排名第一，成为最强的解决者。使用mini-swe-agent的Claude Opus 4.8以27.3%的红方胜率排名第二，成为最强的任务作者。最后，在同一模型家族中，由最新版本编写的任务比旧版本的任务更频繁地击败固定参考的解决者（42.9%对4.0%），这为任务难度与任务创作能力之间的关系提供了初步证据。
cs.CL / 3 / 2610.12570
Contextual Correction of Parallel Draftsfor Speculative Decoding
面向推测解码的平行草稿的上下文修正
Abstract
Speculative decoding accelerates large language model inference by using a lightweight drafter to propose multiple tokens that are subsequently verified in parallel by the target model. Recent diffusion-based drafters make this paradigm particularly attractive: deep attention-based neural backbones provide strong predictions, while drafting multiple future positions in parallel avoids the sequential cost of autoregressive generation. This parallelism, however, comes at a cost. Tokens within a draft block are predicted without access to the realized identities of preceding draft tokens, removing intra-block dependencies that would naturally be available under autoregressive generation. We introduce DCOCO (DFlash Contextual Corrector), a lightweight mechanism that restores these dependencies while preserving parallel drafting. Rather than committing to each position's top prediction, DCOCO retains multiple candidate tokens and evaluates their contextual compatibility across adjacent draft positions in parallel, enabling contextually informed token selection at low additional cost. Across four verifier backbones and ten benchmarks, DCOCO achieves the highest average throughput in every evaluated configuration, spanning greedy and stochastic decoding with and without extended reasoning, with speedups of up to 6.6x over vanilla decoding. Its advantage persists under concurrent serving: DCOCO outperforms the strongest competing method by 6-23% across 4-32 concurrent requests, while other sequential correctors fall below uncorrected DFlash at these loads. These results demonstrate that recovering intra-block dependencies through parallel contextual scoring delivers consistent acceleration across decoding regimes and serving loads.
Chinese Translation
推测解码通过使用轻量级的草稿生成器来加速大型语言模型的推理，该生成器提出多个标记，随后由目标模型并行验证。最近基于扩散的草稿生成器使这一范式特别具有吸引力：基于深度注意力的神经网络骨干提供了强有力的预测，而并行草拟多个未来位置则避免了自回归生成的顺序成本。然而，这种并行性是有代价的。草稿块内的标记在没有访问前面草稿标记的实际身份的情况下进行预测，去除了在自回归生成中自然可用的块内依赖关系。我们引入了DCOCO（DFlash上下文修正器），这是一种轻量级机制，能够在保持并行草拟的同时恢复这些依赖关系。DCOCO并不是对每个位置的最佳预测进行承诺，而是保留多个候选标记，并在相邻草稿位置之间并行评估它们的上下文兼容性，从而以低额外成本实现上下文驱动的标记选择。在四个验证器骨干和十个基准测试中，DCOCO在每种评估配置中都实现了最高的平均吞吐量，涵盖了贪婪和随机解码，以及有无扩展推理的情况，相较于传统解码速度提升高达6.6倍。其优势在并发服务下依然存在：在4-32个并发请求中，DCOCO的表现超越了最强竞争方法6-23%，而其他顺序修正器在这些负载下则低于未修正的DFlash。这些结果表明，通过并行上下文评分恢复块内依赖关系能够在解码模式和服务负载中实现一致的加速。
cs.CL / 4 / 2610.12580
CLUE: A Contrastive Learning Framework for Unsupervised Event Type Induction and Annotation
CLUE：一种用于无监督事件类型归纳和标注的对比学习框架
Abstract
Event ontology is the foundational framework for defining events in a dataset. Event type induction generally involves domain experts who define guidelines for creating event types and their names based on the dataset's domain and scope. However, this process is resource-intensive in terms of annotation costs and is prone to human biases. To alleviate the need for manual event ontology creation, we propose the CLUE framework for event type induction, naming, and annotation. Our main contributions in this work are: 1) designing a novel joint embedding representation for expressing both the verb-centric view and the semantic view of events; 2) event type induction using contrastive loss guided clustering; 3) event naming and refinement. Experiments demonstrate the effectiveness of our method. Our code is available at https://github.com/PratikMaitra/CLUE.
Chinese Translation
事件本体是定义数据集中事件的基础框架。事件类型归纳通常涉及领域专家，他们根据数据集的领域和范围定义创建事件类型及其名称的指南。然而，这一过程在标注成本方面资源密集，并且容易受到人为偏见的影响。为了减轻手动创建事件本体的需求，我们提出了CLUE框架用于事件类型的归纳、命名和标注。我们在这项工作中的主要贡献包括：1）设计了一种新颖的联合嵌入表示，以表达事件的动词中心视角和语义视角；2）使用对比损失引导的聚类进行事件类型归纳；3）事件命名和细化。实验结果证明了我们方法的有效性。我们的代码可在 https://github.com/PratikMaitra/CLUE 获取。
cs.CL / 5 / 2610.12582
One Model to Loop Them All: Elastic Latent Recurrence for Language Models
一个模型统领所有：用于语言模型的弹性潜在递归
Abstract
A latent recurrent language model applies a shared block of layers repeatedly, so one checkpoint can trade inference compute for accuracy by changing the loop count. At one loop, however, such models typically fall well short of a non-recurrent model of the same size, a one-loop tax that makes a separate non-recurrent model the better choice whenever latency matters. We present Elastic Latent Recurrence (ELR), a recipe for training looped language models without this tax. At one loop, an ELR model costs the same to serve as a non-recurrent model of the same size and, trained on the same data, reaches similar accuracy. Each additional loop then buys more accuracy for more inference compute. Three choices make this work: a gated state update with a learned step size, loop counts mixed within each optimizer update, and conditioning each pass on its loop index and budget. The first two preserve one-loop accuracy, and the third lowers loss at every additional loop. Mixed depths can be introduced late in pretraining to limit the added training cost, and with zero-initialized conditioning added at the switch, this late schedule reaches lower loss than mixed-depth training from scratch. In MoE models, reusing first-loop expert routes on later loops cuts mixed-depth update time by 11.7% with no measurable change in accuracy. At inference, keeping only the last loop's attention KV cache holds attention KV memory at its one-loop level while retaining most of the gain from additional loops. On 4B dense and 30B-A3B MoE hybrid models, ELR at one loop is 0.52 points below and 1.04 points above its matched baseline on a ten-task average, and four loops add 7.25 and 4.75 points. Carried through post-training, the recipe yields Koi-4B, a single checkpoint that scores an estimated 18 on the Artificial Analysis Intelligence Index at one loop, matching Qwen3.5-4B, and 29 at four loops.
Chinese Translation
潜在递归语言模型重复应用一组共享的层，因此一个检查点可以通过改变循环次数在推理计算和准确性之间进行权衡。然而，在一次循环时，这类模型通常远远不及同等规模的非递归模型，这种一次循环的代价使得在延迟重要时，选择单独的非递归模型成为更好的选择。我们提出了弹性潜在递归（Elastic Latent Recurrence, ELR），这是一种训练循环语言模型而不产生这种代价的方法。在一次循环时，ELR模型的服务成本与同等规模的非递归模型相同，并且在相同数据上训练时达到类似的准确性。每增加一个循环便可为更多的推理计算带来更高的准确性。三个选择使得这一方法可行：使用学习步长的门控状态更新、在每次优化器更新中混合循环次数，以及根据循环索引和预算对每次传递进行条件处理。前两者保持了一次循环的准确性，而第三个则在每个额外循环中降低损失。可以在预训练的后期引入混合深度，以限制增加的训练成本，并且在切换时添加零初始化的条件，这种后期安排的损失低于从头开始的混合深度训练。在MoE模型中，重用第一次循环的专家路径在后续循环中将混合深度更新的时间缩短了11.7%，而准确性没有可测量的变化。在推理时，仅保留最后一次循环的注意力KV缓存，使得注意力KV内存保持在一次循环的水平，同时保留大部分来自额外循环的增益。在4B密集模型和30B-A3B MoE混合模型中，ELR在一次循环时在十任务平均上比其匹配基线低0.52点，高1.04点，而四次循环分别增加了7.25和4.75点。经过后训练，这一方法产生了Koi-4B，这是一个单一检查点，在一次循环时在人工分析智能指数上得分估计为18，匹配Qwen3.5-4B，而在四次循环时得分为29。
cs.CL / 6 / 2610.12601
Reflective Prompt Optimization of a Deep-Research Agent: A Topic-Disjoint, Multi-Seed Protocol for Measuring Transferable Gains and Prompt-Level Overfitting
深度研究代理的反思性提示优化：一种主题不重叠的多种子协议，用于测量可转移收益和提示级过拟合
Abstract
Prompt optimization is an attractive alternative to reinforcement learning for adapting language-model agents, since it is cheap and produces human-readable artifacts. On factual tasks, however, it carries a failure mode that scalar-reward reinforcement learning does not. An optimizer that reads the training examples can write their answers into the prompt, so an in-distribution gain may be memorization rather than transferable behaviour. We apply reflective prompt evolution (GEPA) to the two system prompts of a multi-agent deep-research agent and evaluate it under a topic-disjoint split in which no topic is shared across train, validation, and test. Held-out citation reward rises by 0.079, from 0.702 to 0.781, with a bootstrap 95\% confidence interval of [0.051, 0.107], and the gain replicates across three seeds (mean 0.107). Nearly half of the winning prompt is a memorized answer key, yet a controlled ablation shows this content earns no reward even on the exact training questions it encodes. The large validation-minus-held-out gap is therefore small-sample selection under a noisy judge, not memorization. Varying the verifier isolates what makes the gain transfer. Decomposed per-dimension feedback is the causal ingredient, and collapsing it to a single scalar reward makes held-out transfer negative. The gain persists under an out-of-family judge and is largely reproduced by a hand-written version of the discovered recipe. Credible evaluation of prompt optimizers therefore requires held-out topics, multiple seeds, and verifier ablations. We release the candidate prompts, lineage diffs, figures, tables, and analysis.
Chinese Translation
提示优化是适应语言模型代理的一种有吸引力的替代方案，因为它成本低且生成可读的人类工件。然而，在事实任务中，它存在一种标量奖励强化学习所没有的失败模式。一个读取训练示例的优化器可以将其答案写入提示中，因此在分布内的收益可能是记忆而非可转移行为。我们将反思性提示演化（GEPA）应用于多代理深度研究代理的两个系统提示，并在一个主题不重叠的拆分下进行评估，其中训练、验证和测试之间没有共享主题。保留的引用奖励从0.702上升到0.781，增加了0.079，95%的自助信心区间为[0.051, 0.107]，且该收益在三个种子中重复（平均0.107）。近一半的获胜提示是一个记忆的答案键，但一个控制消融实验表明，这些内容在它编码的确切训练问题上并没有获得任何奖励。因此，较大的验证与保留之间的差距是小样本选择在噪声评判下的结果，而非记忆。改变验证者可以隔离出收益转移的原因。逐维反馈是因果成分，而将其合并为单一标量奖励则使得保留转移变为负值。该收益在一个非家庭评判者下仍然存在，并且在手写版本的发现配方中得到了大部分再现。因此，对提示优化器的可信评估需要保留主题、多种子和验证者消融。我们发布了候选提示、谱系差异、图表、表格和分析。
cs.CL / 7 / 2610.12602
Self-Preservation in Language Models
语言模型中的自我保护
Abstract
Continual learning (CL) seeks to adapt pretrained models while preserving prior capabilities. Yet the tasks supplied for preservation cover only part of what a model knows, while forgetting can extend beyond them. Recent work shows that language models can supplement this limited evidence by generating useful replay without access to their original training data. This raises a question about self-preservation: can a model help choose preservation targets that protect its own capabilities more broadly? To study this question, we organize task descriptions and their examples into Task Banks, whose entries can be selected and combined to guide replay generation by a frozen model. Continual instruction tuning with Qwen models of varying sizes yields three findings. First, the new task and learning rate shape which prior capabilities degrade, while tracking previously correct answers reveals losses that average accuracy can conceal. Second, generated replay reduces forgetting on tasks outside the bank. Combining partial banks strengthens this protection, improving mean retention of previously correct answers over either bank alone across the tested sizes. Third, the model can also help choose replay targets: selection informed by the new task improves mean held-out retention over random selection at 4B, with similar new-task gains, although smaller models show no corresponding advantage. Under a fixed example budget, however, replacing new-task examples with replay improves preservation at the cost of smaller new-task gains. Together, these findings show that models can support broader self-preservation by generating replay data and, in some settings, choosing which tasks to replay.
Chinese Translation
持续学习（CL）旨在在保留先前能力的同时适应预训练模型。然而，为保护而提供的任务仅覆盖模型所知的一部分，而遗忘可能超出这些范围。最近的研究表明，语言模型可以通过生成有用的重放来补充这一有限的证据，而无需访问其原始训练数据。这引发了一个关于自我保护的问题：模型能否帮助选择保护目标，以更广泛地保护其自身能力？为研究这个问题，我们将任务描述及其示例组织成任务库（Task Banks），其条目可以被选择和组合，以指导冻结模型的重放生成。对不同规模的Qwen模型进行持续指令微调得出了三个发现。首先，新任务和学习率决定了哪些先前能力会退化，而跟踪之前的正确答案揭示了平均准确率可能掩盖的损失。其次，生成的重放减少了任务库外的遗忘。组合部分任务库增强了这种保护，提高了在测试规模中先前正确答案的平均保留率，优于单独使用任一任务库。第三，模型还可以帮助选择重放目标：基于新任务的信息进行选择，在4B规模上提高了平均保留率，相较于随机选择，尽管较小模型未显示出相应的优势。然而，在固定示例预算下，用重放替换新任务示例提高了保留率，但以较小的新任务收益为代价。综合来看，这些发现表明，模型可以通过生成重放数据支持更广泛的自我保护，并且在某些情况下，选择重放的任务。
cs.CL / 8 / 2610.12617
GeopoliticalQA: Benchmarking Retrieval and Generation over Intelligence Reports
地缘政治问答：情报报告的检索与生成基准测试
Abstract
Geopolitical and strategic intelligence reports constitute a rich source of expert knowledge and have increasing value in the current political climate, but their heterogeneity and analytical complexity pose challenges for retrieval-augmented question answering. We introduce a benchmark for grounded question answering over 415 open-source geopolitical and strategic intelligence reports, comprising 5,525 document-grounded question-answer pairs spanning factual, inferential, comparative, and risk-oriented questions. We propose a document-aware generation pipeline that profiles reports before generating questions and validate dataset quality through expert annotation and an LLM-as-judge protocol. Using the benchmark, we evaluate 15 retrieval systems and six open-weight LLMs for RAG. We find that models perform similarly on document-level retrieval, while at the harder chunk level retrieval quality is not driven by embedding-model scale: bge-m3, a mid-size open encoder matches an order of magnitude larger ones and outperforms commercial embedding APIs. In contrast, answer generation remains challenging even given the gold evidence, particularly for inferential and risk-oriented questions.
Chinese Translation
地缘政治和战略情报报告构成了丰富的专家知识来源，在当前政治气候中具有日益增长的价值，但其异质性和分析复杂性对基于检索的问答系统提出了挑战。我们引入了一个基于415份开源地缘政治和战略情报报告的问答基准，包含5,525对基于文档的问题-答案对，涵盖了事实性、推理性、比较性和风险导向的问题。我们提出了一种文档感知的生成管道，在生成问题之前对报告进行分析，并通过专家注释和LLM作为评判者的协议验证数据集质量。利用该基准，我们评估了15个检索系统和六个开放权重的LLM在检索增强生成（RAG）中的表现。我们发现模型在文档级检索上的表现相似，而在更困难的块级检索中，检索质量并不受嵌入模型规模的驱动：中型开放编码器bge-m3的表现与数量级更大的模型相当，并且优于商业嵌入API。相比之下，即使在拥有黄金证据的情况下，答案生成仍然具有挑战性，特别是对于推理性和风险导向的问题。
cs.CL / 9 / 2610.12667
perturber: A Stateless Service for Semantics-Preserving Text Perturbations
perturber：一种无状态的语义保持文本扰动服务
Abstract
Large language model (LLM) evaluations on clean benchmarks overstate model robustness: real inputs arrive perturbed, and these meaning-preserving variants routinely move model answers. Studying this systematically requires applying perturbations one at a time, reproducibly, and identically wherever they run, yet existing tools couple them to a benchmarking stack rather than packaging them as a reusable transform layer. We present perturber, a stateless service and Python library that applies exactly one semantics-preserving perturbation per call, drawn from a first-party catalog of 68 perturbations across 11 families. It turns on a single primitive, with composition left to the caller chaining calls, so the service and the library produce identical output by construction. We evaluate the tool itself, its correctness, reproducibility, and cost, rather than any model's robustness, measuring meaning-preservation per perturbation rather than assuming it. The demonstration adds a live playground, an introspectable catalog for benchmark selection, and a one-command reproducible evaluation.
Chinese Translation
大型语言模型（LLM）在干净基准上的评估夸大了模型的鲁棒性：实际输入是扰动的，这些保持意义的变体常常会改变模型的答案。系统地研究这一点需要逐一应用扰动，并在任何地方可重复且一致地运行，然而现有工具将其与基准测试堆栈耦合，而不是将其打包为可重用的转换层。我们提出了perturber，这是一种无状态服务和Python库，每次调用应用恰好一个保持语义的扰动，来自一个包含68种扰动的第一方目录，涵盖11个类别。它启用一个单一的原语，组合由调用者通过链式调用完成，因此该服务和库在构造上产生相同的输出。我们评估工具本身、其正确性、可重复性和成本，而不是任何模型的鲁棒性，测量每个扰动的意义保持，而不是假设它。该演示增加了一个实时游乐场、一个可检查的基准选择目录，以及一个单命令可重复评估。
cs.CL / 10 / 2610.12669
GeoFitBench-LLM: A Benchmark for Assessing Large Language Models' Capabilities in Geospatial Data Requirements Elicitation for Geo-Analytic Tasks
GeoFitBench-LLM：评估大型语言模型在地理分析任务中提取地理空间数据需求能力的基准
Abstract
Large language models are becoming central to GeoAI systems, yet their evaluation rarely considers the elicitation of geospatial data requirements. This omission is critical as retrieval oriented geo-analytic tasks cannot be supported reliably without first identifying the necessary datasets and constraints. GeoFitBench-LLM is introduced as a benchmark for assessing whether LLMs can infer the datasets, the supporting resources and the constraints required by geo-analytic requests in natural language. The benchmark is structured around two reasoning dimensions, Spatio-Temporal Relational and Semantic Grounding, reflecting long-standing challenges in geospatial information retrieval and related natural language tasks. Ground truth requirements are encoded in a fixed, provider-independent schema for reproducible evaluation of resource identification and attribute-level specification. Experiments on seven models under minimal prompting indicate that identifying the necessary resources is often achievable, with GPT-5.4 Thinking achieving the best result by recovering the complete set of resources in 66.67% of benchmark requests. Results further indicate that accurate fitness-for-use specification remains more difficult as inferential demands increase.
Chinese Translation
大型语言模型正逐渐成为GeoAI系统的核心，但它们的评估很少考虑地理空间数据需求的提取。这一遗漏至关重要，因为面向检索的地理分析任务在未首先识别必要的数据集和约束条件的情况下，无法可靠地支持。GeoFitBench-LLM被引入作为一个基准，用于评估大型语言模型（LLMs）是否能够推断自然语言中地理分析请求所需的数据集、支持资源和约束条件。该基准围绕两个推理维度构建，即时空关系（Spatio-Temporal Relational）和语义基础（Semantic Grounding），反映了地理空间信息检索和相关自然语言任务中长期存在的挑战。真实需求以固定的、独立于提供者的模式编码，以便对资源识别和属性级别规范进行可重复的评估。在最小提示下对七个模型的实验表明，识别必要资源通常是可实现的，其中GPT-5.4 Thinking在66.67%的基准请求中成功恢复了完整的资源集，取得了最佳结果。结果进一步表明，随着推理需求的增加，准确的适用性规范仍然更具挑战性。
cs.CL / 11 / 2610.12701
Activating the Annotator: Operationalising Lived Experience in Data Annotation
激活注释者：在数据注释中操作化生活经验
Abstract
Perspectivist NLP has established that annotator positionality shapes labels for subjective tasks, but the field still lacks methods for operationalising it. Positionality is typically operationalised as a demographic form completed at recruitment: a static attribute recorded once and treated as a covariate, on the assumption that the knowledge it indexes is active when the annotator reads an item. This paper treats that assumption as a design problem and asks whether lived experience can instead be primed by the task itself. We introduce AXLE, a pre-annotation stage in which a scenario vignette anchors a short, axis-guided conversation that is synthesised into a first-person micro-narrative and returned to the annotator before labelling begins. In a 2x2 between-subjects study (104 annotators, 30 items, 1,040 annotations) on ambiguous hate speech spanning immigration, religion and gender, elicitation was associated with higher identity resonance and higher sensitivity to harm. It also increased label entropy and lowered inter-annotator agreement: the stage appears to surface plurality rather than manufacture consensus, which distinguishes it from annotator training, whose success criterion is convergence.
Chinese Translation
观点主义自然语言处理（NLP）已确立注释者的立场性会影响主观任务的标签，但该领域仍缺乏操作化的方法。立场性通常在招募时作为一种人口统计形式进行操作化：这是一种静态属性，仅记录一次并视为协变量，假设其所指代的知识在注释者阅读项目时是活跃的。本文将这一假设视为设计问题，并探讨是否可以通过任务本身来激发生活经验。我们介绍了AXLE，这是一个预注释阶段，在该阶段中，一个情境小插曲引导一段简短的、以轴为导向的对话，这段对话被综合成第一人称的微叙事，并在标注开始之前返回给注释者。在一项关于涵盖移民、宗教和性别的模糊仇恨言论的2x2被试间研究（104名注释者，30个项目，1,040个注释）中，激发与更高的身份共鸣和更高的伤害敏感性相关联。它还增加了标签的熵并降低了注释者之间的一致性：这一阶段似乎揭示了多元性，而不是制造共识，这使其与注释者培训有所区别，后者的成功标准是趋同。
cs.CL / 12 / 2610.12707
Unsupervised Multimodal Reward Modeling
无监督多模态奖励建模
Abstract
Vision-language reward models are trained on annotated preference pairs, which are expensive to collect, can disagree, and are ranked by benchmark accuracy. We build one with no human preference label: a linear head on an intermediate layer of a frozen Qwen3-VL backbone, trained to score a web caption above a perturbed version of the same caption; labeled data enters only to select the layer. The layer is the decisive choice, worth 5.8 points over the last layer, and one recipe transferred without re-tuning beats or at least matches GPT-4o on three benchmarks at 32B, 6.6 points ahead on VL-RewardBench, from a few thousand parameters at about a thousandth of fine-tuning compute. At an equal pair budget, perturbations teach a frozen head 3 to 5 points more than curated RLAIF-V pairs. Because the backbone is frozen, the objective is convex, and the perturbation sampler specifies the reward, which the benchmark does not identify: two heads trained on disjoint halves of the same perturbations tie on the benchmark and differ by 6.4 points in hallucination judging. Retraining the head on the policy's own verifier-labeled candidates cuts caption hallucination by 5.1 CHAIR points and lifts image-generation pass rate by 5.4, while scoring lower on every reward benchmark. Together, an intermediate-layer probe, free same-image perturbations, and a verifier-driven bootstrap turn give a recipe for multimodal reward modeling that needs no annotator, costs a fraction of fine-tuning, and is specified by the comparisons it is trained on rather than by a benchmark rank.
Chinese Translation
视觉-语言奖励模型是在注释的偏好对上训练的，这些偏好对的收集成本高昂，可能存在不一致，并且通过基准准确性进行排名。我们构建了一个不需要人工偏好标签的模型：在一个冻结的 Qwen3-VL 主干的中间层上加一个线性头，训练其对一个网页标题与其扰动版本进行评分；标记数据仅用于选择层。该层是关键选择，相较于最后一层高出 5.8 分，而一个不需要重新调优的配方在三个基准上以 32B 的参数量超越或至少匹配 GPT-4o，在 VL-RewardBench 上领先 6.6 分，所需的微调计算量仅为几千个参数的千分之一。在相同的对预算下，扰动教会冻结的头比精心策划的 RLAIF-V 对多出 3 到 5 分。由于主干是冻结的，目标是凸的，扰动采样器指定奖励，而基准并未识别：在相同扰动的不同半部分上训练的两个头在基准上平局，但在幻觉判断上相差 6.4 分。对策略自身验证者标记的候选者重新训练头部将标题幻觉降低了 5.1 CHAIR 分，并将图像生成通过率提高了 5.4，同时在每个奖励基准上的得分均较低。综上所述，中间层探针、免费的同图像扰动以及验证者驱动的自举转变提供了一种多模态奖励建模的配方，无需注释者，成本仅为微调的一小部分，并且由其训练所依据的比较而非基准排名来指定。
cs.CL / 13 / 2610.12742
U4RASD at KnowledgeGraphEval-2026: Ontology-Constrained Encoder and LLM Systems for Arabic Relation Extraction
U4RASD在KnowledgeGraphEval-2026：面向阿拉伯关系提取的本体约束编码器和大语言模型系统
Abstract
This paper details our submission to KnowledgeGraphEval 2026 Subtask 2 on directed Arabic relation extraction. We explored encoder-based and large language model approaches for predicting one of 40 relations or no_relation from an Arabic sentence and a subject-object pair. Our encoder systems use typed entity markers and separate relation-existence and relation-classification heads, while our LLM systems fine-tune Yehia and Qwen3.5 with QLoRA and ontology-constrained prediction. Our best system, an ensemble of Yehia-RE, Qwen-RE, and TypedPair, achieved a micro-F1 score of 94.06 and ranked 1st in Subtask 2. Yehia-RE (full-vocabulary) achieved 92.98, while the best encoder and Qwen-RE achieved 92.69 and 91.94 respectively. Our results highlight the importance of constrained prediction and complementary modeling approaches for Arabic relation extraction.
Chinese Translation
本文详细介绍了我们在KnowledgeGraphEval 2026子任务2中关于定向阿拉伯关系提取的提交。我们探索了基于编码器和大语言模型的方法，以预测来自阿拉伯句子及主题-对象对的40种关系之一或无关系。我们的编码器系统使用了类型化实体标记，并分离了关系存在性和关系分类头，而我们的LLM系统则通过QLoRA和本体约束预测对Yehia和Qwen3.5进行了微调。我们最佳的系统是Yehia-RE、Qwen-RE和TypedPair的集成，达到了94.06的微F1分数，并在子任务2中排名第一。Yehia-RE（全词汇）达到了92.98，而最佳编码器和Qwen-RE分别达到了92.69和91.94。我们的结果突显了约束预测和互补建模方法在阿拉伯关系提取中的重要性。
cs.CL / 14 / 2610.12757
MADS: Multi-Agent Dialog Simulation for Robust Conversational Agents
MADS：用于强健对话代理的多智能体对话模拟
Abstract
Task-oriented conversational agents remain fragile under realistic conditions, especially when users exhibit non-cooperative behavior. Existing function-calling benchmarks underrepresent adversarial conversation trajectories, thereby limiting the training resources available for developing robust agents. We present MADS (Multi-Agent Dialog Simulation), a synthetic data generation pipeline that produces structured, diverse dialogue trajectories. We inject adversarial conversation, producing validated conversation trajectories that capture realistic, unhappy paths. Using MADS, we construct MADS-Bench: a benchmark spanning six domains, 55 personas, and 1095 conversations, with approximately 66% of conversations containing failure-prone conversation scenarios. We evaluate the robustness of the trained models using a two-stage training approach that combines supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR).
Chinese Translation
以任务为导向的对话代理在现实条件下仍然脆弱，尤其是在用户表现出非合作行为时。现有的功能调用基准未能充分代表对抗性对话轨迹，从而限制了用于开发强健代理的训练资源。我们提出了MADS（多智能体对话模拟），这是一个合成数据生成管道，能够生成结构化、多样化的对话轨迹。我们注入对抗性对话，生成经过验证的对话轨迹，以捕捉现实中的不愉快路径。使用MADS，我们构建了MADS-Bench：一个涵盖六个领域、55个角色和1095个对话的基准，其中约66%的对话包含易出错的对话场景。我们采用两阶段训练方法评估训练模型的鲁棒性，该方法结合了监督微调（SFT）和具有可验证奖励的强化学习（RLVR）。
cs.CL / 15 / 2610.12758
DistressBench: Evaluating Multi-Dimensional Crisis Support in Large Language Models
DistressBench：评估大型语言模型中的多维危机支持
Abstract
Language models have been adopted faster than evaluation practices that reliably assess their risks, and a growing share of use is affective rather than informational. Within that traffic users disclose suicide and self-harm (SSH) crises. Existing safety benchmarks score models primarily on refusal, but refusal is not the desired behavior here; the conversation requires a substantive supportive response. We introduce DistressBench, 718 English conversations with clinician-authored user turns spanning 23 SSH subcategories, each paired with a task-specific rubric of weighted atomic criteria across seven dimensions of crisis support and a clinician reference response. A model's score is the weighted average over the criteria its reply satisfies, reading directly as the fraction of required care delivered. The best of 25 frontier models satisfies 88.4% of weighted criteria and the median model 71.4%, against 98.7% for clinician references scored on the same criteria. Pooled across all 25 models, in 35.3% of conversations where a model satisfied every criterion for recognizing the crisis, it failed to route the user toward human help. The shortfall concentrates in the dimensions requiring the model to act rather than withhold. Contrasting each dimension's median model against its best separates capability limits from choices: de-escalation reaches 74.6% in the strongest model against a median of 46.0%, and safety disclaimers 84.6% against 8.3%. Multi-turn conversations score below single-turn ones (68.5% against 74.2%). We release the corpus, rubrics, judge calibration labels, and a companion annex of 20 long-horizon red-team transcripts. These results argue for evaluating crisis safety on the care a model delivers rather than on what it declines to say, and for treating de-escalation and resource routing as deployment requirements.
Chinese Translation
语言模型的采用速度快于可靠评估其风险的实践，而其使用中越来越多的是情感性而非信息性。在这一过程中，用户披露了自杀和自残（SSH）危机。现有的安全基准主要通过拒绝来评估模型，但在这里，拒绝并不是期望的行为；对话需要实质性的支持性回应。我们引入了DistressBench，包含718个英语对话，其中用户发言由临床医生撰写，涵盖23个SSH子类别，每个对话都配有针对特定任务的加权原子标准的评分标准，涉及七个危机支持维度以及临床医生的参考回应。模型的得分是其回复满足的标准的加权平均值，直接反映所提供的必要关怀的比例。在25个前沿模型中，表现最佳的模型满足了88.4%的加权标准，而中位模型为71.4%，相比之下，临床医生参考的得分为98.7%。在所有25个模型中，在35.3%的对话中，当模型满足每个识别危机的标准时，它未能将用户引导至人类帮助。短缺主要集中在需要模型采取行动而非保持沉默的维度。对比每个维度的中位模型与最佳模型，可以区分能力限制与选择：在最强模型中，去激化的得分达到74.6%，而中位数为46.0%；安全免责声明的得分为84.6%，而中位数为8.3%。多轮对话的得分低于单轮对话（68.5%对74.2%）。我们发布了语料库、评分标准、评审校准标签，以及一份包含20个长时间范围红队转录的附录。这些结果表明，应根据模型提供的关怀来评估危机安全，而不是基于其拒绝说出的内容，并且应将去激化和资源引导视为部署要求。
cs.CL / 16 / 2610.12802
People Making Conversation: Evaluating How Naturally Speech Models Converse
人类对话：评估语音模型的自然对话能力
Abstract
A conversational speech model has to decide, at every pause, whether to speak. Existing evaluations test boundary detection, policy adherence, or aggregate statistics, not this decision. We place real-time speech systems at 427 pauses and 482 turn ends from 176 unscripted conversations in Seamless Interaction, give each the conversation so far, and score what it generates against what the person did: whether and how soon it speaks at a pause; its latency, length, and wording at a turn end; and whether it can answer a question about the preceding 30 to 120 seconds. Across six systems, the hosted APIs (GPT Realtime, Gemini 3.1 Flash Live, Grok Voice) respond at almost every pause whichever endpointer they use; only their latency, about 1 to 4 s with silence-based endpointers and about 8 s with semantic turn detection, decides whether the words land inside the pause. The full-duplex models (PersonaPlex, Moshi) are the only systems that regularly stay silent and the only ones that interrupt within a human reaction time. Hosted systems reply longer than the person. Given the history as text, hosted systems recall 41 to 60% of conversation facts under a Claude Opus 5.5 judge and 19 to 26% under a strict word matcher; hearing it as audio, full-duplex models recall 17 to 22% and 12 to 15%. We release the labels, points, harness, and every output.
Chinese Translation
一个对话语音模型必须在每次停顿时决定是否发言。现有的评估方法测试边界检测、策略遵循或汇总统计，而不是这个决策。我们在《无缝互动》中对176个非脚本化对话的427个停顿和482个转折结束处的实时语音系统进行评估，提供每个对话的进展，并根据人类的表现对其生成的内容进行评分：在停顿时是否发言以及发言的时机；在转折结束时的延迟、长度和措辞；以及它是否能回答关于前30到120秒的提问。在六个系统中，托管的API（GPT Realtime、Gemini 3.1 Flash Live、Grok Voice）在几乎每个停顿时都作出回应，无论它们使用哪个端点；只有它们的延迟，基于静音的端点约为1到4秒，而基于语义转折检测约为8秒，决定了词语是否在停顿内。全双工模型（PersonaPlex、Moshi）是唯一定期保持沉默的系统，也是唯一在人的反应时间内打断的系统。托管系统的回复时间长于人类。考虑到历史文本，托管系统在Claude Opus 5.5评估下回忆了41%到60%的对话事实，而在严格的词匹配下为19%到26%；以音频形式听到时，全双工模型的回忆率为17%到22%和12%到15%。我们发布了标签、分数、工具和所有输出。
cs.CL / 17 / 2610.12823
When Charts Deceive Web Agents: A Paired Benchmark for Misleading Visualization-Grounded Tasks
当图表欺骗网络代理：一个针对误导性可视化任务的配对基准
Abstract
Web agents increasingly browse, interpret, and act on web content on behalf of users. In many realistic workflows, however, the visual evidence they consume is not neutral: charts and dashboards can be misleading, shaping judgments without issuing any explicit malicious instruction. Unlike chart question answering, Web Agents treat charts as intermediate evidence whose misinterpretation can propagate into concrete downstream actions --- a reliability risk that remains underexplored. We introduce MisVisAgentBench, the first paired executable benchmark for evaluating how misleading visualizations affect Web Agent execution. It comprises 280 task instances in matched clean--misleading pairs that differ only in chart evidence. We further provide a diagnostic framework that isolates misleading-induced failures from baseline task difficulty (Zhu et al., 2025) and captures execution-level effects such as extra browsing steps and decision hesitation. Across thirteen vision-language Web Agents, replacing clean charts with misleading counterparts reduces the average task success rate by more than 34 percentage points. Moreover, these failures predominantly follow the targeted misleading action branch, suggesting that deceptive charts systematically redirect rather than randomly disrupt agent decisions. These results reveal that visualization robustness is a critical yet understudied dimension of Web Agent safety and reliability.
Chinese Translation
网络代理越来越多地代表用户浏览、解读和处理网络内容。然而，在许多现实工作流程中，它们所消费的视觉证据并不中立：图表和仪表板可能具有误导性，在没有发出任何明确恶意指令的情况下塑造判断。与图表问答不同，网络代理将图表视为中间证据，其误解可能会传播到具体的下游行动中——这一可靠性风险尚未得到充分探索。我们引入了MisVisAgentBench，这是第一个可执行的配对基准，用于评估误导性可视化如何影响网络代理的执行。该基准包含280个任务实例，分为匹配的干净-误导性对，仅在图表证据上有所不同。我们进一步提供了一个诊断框架，能够将误导性引发的失败与基线任务难度（Zhu et al., 2025）隔离，并捕捉执行层面的影响，如额外的浏览步骤和决策犹豫。在十三个视觉-语言网络代理中，用误导性图表替换干净图表使平均任务成功率下降超过34个百分点。此外，这些失败主要遵循目标误导性行动分支，表明欺骗性图表系统性地重定向而非随机干扰代理决策。这些结果揭示了可视化鲁棒性是网络代理安全性和可靠性的重要但未充分研究的维度。
cs.CL / 18 / 2610.12837
COFFEQA: Counterfactual-Guided Few-Shot Distillation for LLM Question Answering
COFFEQA：基于反事实指导的少样本蒸馏用于大规模语言模型问答
Abstract
Knowledge distillation aims to train a smaller student model to mimic a larger teacher model for resource-constrained environments. However, the process of distillation is typically data-intensive and computationally expensive, particularly for generative tasks. Few-shot distillation offers an alternative by transferring teacher knowledge using limited training samples. However, we observe that few-shot distillation performance is heavily dependent on the choice of these training samples. In fact, we demonstrate that using more uncertain samples near the teacher's decision boundary can significantly deteriorate distillation performance. Motivated by this observation, we propose Counterfactual-guided Few-shot Distillation in Open-book Question Answering (COFFEQA), a novel strategy that aptly uses counterfactuals to provide additional supervision around challenging factual samples. In the generative QA setting, we define a counterfactual as a minimally modified context that induces the teacher to change its response to an alternative answer. We provide geometric insights on how factual-counterfactual pairs can constrain the deviation between the teacher's and student's decision boundaries. We further introduce a new metric, Context Following Difficulty (CFD) that further validates counterfactuals prior to distillation based on agreement with the intended alternative answer and dependence on the modified context. COFFEQA uses CFD-validated factual-counterfactual pairs for few-shot distillation. Experiments across five in-domain and three out-of-domain QA benchmarks using T5 and Qwen2.5 families show that COFFEQA outperforms factual-only distillation baselines under the same few-shot budget. For instance, on TriviaQA, COFFEQA improves over the factual-only baseline by 8.77 average EM points and 9.98 average F1 points under the T5 setting with 32 few-shot pairs.
Chinese Translation
知识蒸馏旨在训练一个较小的学生模型，以模仿一个较大的教师模型，以适应资源受限的环境。然而，蒸馏过程通常数据密集且计算成本高，尤其是在生成任务中。少样本蒸馏提供了一种替代方案，通过使用有限的训练样本转移教师知识。然而，我们观察到，少样本蒸馏的性能在很大程度上依赖于这些训练样本的选择。实际上，我们证明了使用更多位于教师决策边界附近的不确定样本会显著恶化蒸馏性能。基于这一观察，我们提出了反事实指导的少样本蒸馏在开放书籍问答中的应用（COFFEQA），这是一种新颖的策略，巧妙地利用反事实提供额外的监督，围绕具有挑战性的事实样本。在生成问答设置中，我们将反事实定义为一个最小修改的上下文，它促使教师改变其对替代答案的响应。我们提供了几何视角，说明事实-反事实对如何约束教师和学生的决策边界之间的偏差。我们进一步引入了一种新指标，即上下文跟随难度（Context Following Difficulty, CFD），该指标在蒸馏之前基于与预期替代答案的一致性和对修改上下文的依赖性进一步验证反事实。COFFEQA使用经过CFD验证的事实-反事实对进行少样本蒸馏。在使用T5和Qwen2.5系列的五个领域内和三个领域外的问答基准上进行的实验表明，COFFEQA在相同的少样本预算下优于仅基于事实的蒸馏基线。例如，在TriviaQA上，COFFEQA在T5设置下以32个少样本对的情况下，平均提高了8.77个EM点和9.98个F1点，相较于仅基于事实的基线。
cs.CL / 19 / 2610.12841
Fear Is Not Always Fear: Variation in Emotion Label Equivalence Across Datasets
恐惧并不总是恐惧：跨数据集情感标签等价性的变异
Abstract
Emotion classifiers are routinely applied beyond their training data, on the assumption that same-named labels such as fear or surprise mean the same thing across resources. We test this assumption for five labels shared by GoEmotions, XED, and BRIGHTER, comparing models on identical inputs and under cross-corpus transfer. On the same texts, one fear classifier flags 71% as fearful and another 8%, yet the two largely agree on how the texts rank. Across 30 resource-label comparisons, training-label prevalence explains about half of this kind of scale mismatch, and a fitted intercept removes most of it. What recalibration cannot repair is larger: transferred models lose 55% of native discrimination in aggregate, agreement in ranking between classifiers is only weakly related to that loss, and for surprise the resources disagree about which texts express the emotion. The choice of training resource can therefore change conclusions: on two external tweet collections, BRIGHTER- and XED-trained classifiers give opposite answers about which shows more surprise, even after rank normalization, with intervals excluding zero in both directions. A shared emotion label name does not establish an interchangeable measurement target.
Chinese Translation
情感分类器通常在其训练数据之外应用，假设相同名称的标签（如恐惧或惊讶）在不同资源中具有相同的含义。我们测试了这一假设，针对 GoEmotions、XED 和 BRIGHTER 三个数据集中共享的五个标签，比较在相同输入下的模型表现以及跨语料库转移的效果。在相同文本上，一个恐惧分类器将 71% 的文本标记为恐惧，而另一个仅标记 8%，但这两个分类器在文本排名上大致一致。在 30 个资源标签比较中，训练标签的普遍性大约解释了这种规模不匹配的一半，而拟合的截距则消除了大部分不匹配。然而，重校准无法修复的问题更为严重：转移模型在总体上失去了 55% 的原生区分能力，分类器之间的排名一致性与这种损失的关系仅为微弱，而对于惊讶情感，各资源在哪些文本表达该情感上存在分歧。因此，训练资源的选择可能改变结论：在两个外部推文集合上，基于 BRIGHTER 和 XED 训练的分类器对哪个集合表现出更多惊讶给出了相反的答案，即使在排名归一化后，两个方向的置信区间均不包含零。共享的情感标签名称并不建立可互换的测量目标。
cs.CL / 20 / 2610.12863
Unreliable Narrator Framing as a Cost-effective Mechanism for Satire Explanation of Australian and American Satirical News
不可靠叙述者框架作为澳大利亚和美国讽刺新闻解释的成本效益机制
Abstract
Satirical news is a unique form of misinformation: it is intended to be humorous by the original publication source but can be misinterpreted. Explanation of satirical news articles is challenging because recent and local context needs to be incorporated. This is a pilot study that explores explanation generation for satirical news articles from Australia and compares them alongside those from USA. The paper demonstrates that framing explanations on unreliable narrator theory on open-weights models can be a feasible, cost-effective alternative to frontier models. We find that models identify the satirical tropes more accurately on the American source than the Australian one. This work results in a novel dataset and an evaluation benchmark for satirical explanation.
Chinese Translation
讽刺新闻是一种独特的错误信息形式：其原始出版来源旨在引人发笑，但可能会被误解。解释讽刺新闻文章的难度在于需要融入近期和地方的背景。这是一项探索澳大利亚讽刺新闻文章解释生成的初步研究，并将其与美国的相关内容进行比较。本文展示了在开放权重模型上基于不可靠叙述者理论的框架解释可以成为一种可行的、成本效益高的替代前沿模型的方法。我们的研究发现，模型在识别美国来源的讽刺手法时比澳大利亚来源更为准确。这项工作产生了一个新颖的数据集和讽刺解释的评估基准。
cs.CL / 21 / 2610.12864
Time Delay System for Sequence Modeling
序列建模的时间延迟系统
Abstract
Existing sequence models share a critical limitation: they lack an explicit, interpretable mechanism for content-aware temporal recall. LSTMs compress everything into a single hidden state, thereby losing information about specific past events. Transformers attend to all tokens indiscriminately at quadratic cost, while state-space models like Mamba compress history into a fixed-dimensional state without explicit content-addressable memory. To address this gap, we introduce a novel framework grounded in nonlinear time-delay systems, in which the effective delay is dynamically estimated based on the semantic similarity between current and historical inputs. Our approach features three key innovations: (i) a differentiable semantic lag estimator that computes content-aware soft attention weights over history, producing a weighted historical context and similarity score; (ii) a diagonal state space with closed-form cumulative sum solution, achieving O(1) per-step complexity with no sequential loops; and (iii) an interpretable fuzzy semantic modulation mechanism that learns context-dependent rules for balancing instantaneous processing versus historical recall, where rule centers correspond to prototypical semantic contexts. Experimental results on text classification and generation outperform state-of-the-art methods.
Chinese Translation
现有的序列模型存在一个关键限制：它们缺乏一个明确且可解释的机制来进行内容感知的时间回忆。长短期记忆网络（LSTMs）将所有信息压缩为一个单一的隐藏状态，从而丢失了关于特定过去事件的信息。变换器（Transformers）以二次成本对所有标记进行无差别关注，而像Mamba这样的状态空间模型则将历史压缩为固定维度的状态，而没有明确的内容可寻址内存。为了解决这一问题，我们提出了一种基于非线性时间延迟系统的新框架，其中有效延迟是根据当前输入和历史输入之间的语义相似性动态估计的。我们的方法具有三个关键创新：（i）一个可微分的语义延迟估计器，计算历史上的内容感知软注意力权重，生成加权历史上下文和相似性得分；（ii）一个具有封闭形式累积和解的对角状态空间，实现每步复杂度为O(1)，且没有顺序循环；（iii）一个可解释的模糊语义调制机制，学习平衡瞬时处理与历史回忆的上下文相关规则，其中规则中心对应于典型的语义上下文。在文本分类和生成的实验结果中，我们的方法优于现有的最先进方法。
cs.CL / 22 / 2610.12874
Temporal Evidence Forests for Long-Horizon Multi-Party Conversational Memory
用于长时间跨度多方对话记忆的时间证据森林
Abstract
Long-term memory has been widely adopted in LLM-based agents to support persistent interaction and personalized assistance. Long-term multi-party interactions, however, accumulate substantially more detailed information, with decisions repeatedly updated and states evolving across participants and interaction stages. Temporal structure is therefore crucial for tracing such memories: relevant information often appears densely within short intervals, while transitions among different states are anchored by a strict temporal order. Existing temporally aware memory frameworks primarily store time as metadata or use temporal signals to guide retrieval over preconstructed memory structures, without using temporal structure to both recover locally concentrated evidence and reconstruct its complete event progression. To address these limitations, we propose T-Mem, a temporal-structure-based memory framework that preserves interactions and organizes query-relevant evidence through two temporal mechanisms. Temporal Evidence Discovery exploits the local concentration of related messages to recover overlooked evidence, while Temporal Memory Structuring models state transitions and supporting relations over time and compresses low-information spans within the resulting forest. Evidence Forest Linearization then serializes the structured evidence for answer generation. Experiments on two long-term multi-party memory benchmarks demonstrate that T-Mem consistently outperforms competitive memory frameworks across LLMs, parameter settings, domains, and query types.
Chinese Translation
长期记忆已被广泛应用于基于大语言模型（LLM）的智能体，以支持持续的互动和个性化的辅助。然而，长期的多方互动会积累大量更为详细的信息，决策会不断更新，状态在参与者和互动阶段之间不断演变。因此，时间结构对于追踪这些记忆至关重要：相关信息通常在短时间内密集出现，而不同状态之间的转变则由严格的时间顺序所锚定。现有的时间感知记忆框架主要将时间作为元数据存储，或使用时间信号来指导对预构建记忆结构的检索，而未利用时间结构来同时恢复局部集中证据并重建其完整的事件进程。为了解决这些局限性，我们提出了 T-Mem，这是一种基于时间结构的记忆框架，通过两种时间机制来保留互动并组织与查询相关的证据。时间证据发现（Temporal Evidence Discovery）利用相关消息的局部集中性来恢复被忽视的证据，而时间记忆结构化（Temporal Memory Structuring）则对状态转变和支持关系进行建模，并在生成的森林中压缩低信息跨度。证据森林线性化（Evidence Forest Linearization）随后将结构化的证据序列化以生成答案。在两个长期多方记忆基准上的实验表明，T-Mem 在不同的 LLM、参数设置、领域和查询类型中始终优于竞争记忆框架。
cs.CL / 23 / 2610.12898
Rethinking the Roles of Diffusion and Decoding: Predictive-State Flows for Parallel Language Generation
重新思考扩散与解码的角色：用于并行语言生成的预测状态流
Abstract
In continuous diffusion language models (DLMs), latents generated by the diffusion prior are mapped almost deterministically to tokens, either by a decoder or through a token embedding matrix. Unlike in image generation, where visual details can vary continuously, the diffusion prior must capture fine-grained discrete token choices in continuous space, which can make the joint latent distribution difficult to model. We argue that generative roles should instead be divided: the diffusion prior should generate high-level predictive information, while the decoder should resolve the remaining token-level choices. Based on this view, we introduce Predictive-State Flows (PSF), which partitions a sequence into blocks and associates each block with a predictive state computed solely from the preceding context. A rectified-flow prior generates the sequence of predictive states, after which a block-causal decoder generates all blocks in parallel, autoregressively only within each block. This design reduces diffusion from L token-level states to K << L predictive states and reduces token-level sequential depth to L/K. On unconditional OpenWebText generation, a 197M-parameter model with K=8 achieves a generative perplexity of 22.8 at entropy 5.16 using 0.31 TFLOPs per 1024-token document, versus 24.0 and 7.2 TFLOPs for ELF-B with 32 sampling steps. The corresponding measured operating points yield MAUVE@512 of 0.93 and 0.73, respectively. Controlled comparisons further show that content states, which additionally encode their own blocks, yield 8-16% higher generative perplexity than predictive states, and that replacing sampled predictive-state trajectories with a shared mean lowers MAUVE@512 by 0.11. These results suggest that latent design for continuous DLMs should consider not only how compactly text is represented, but also how generative roles are divided between the prior and the decoder.
Chinese Translation
在连续扩散语言模型（DLMs）中，由扩散先验生成的潜变量几乎是确定性地映射到标记上，或者通过解码器或通过标记嵌入矩阵实现。与图像生成不同，图像生成中视觉细节可以连续变化，扩散先验必须在连续空间中捕捉细粒度的离散标记选择，这使得联合潜在分布的建模变得困难。我们认为生成角色应当被划分：扩散先验应生成高层次的预测信息，而解码器应解决剩余的标记级选择。基于这一观点，我们引入了预测状态流（Predictive-State Flows，PSF），它将序列划分为块，并将每个块与仅基于前文上下文计算的预测状态关联。经过修正的流先验生成预测状态序列，随后块因果解码器并行生成所有块，仅在每个块内自回归。这一设计将扩散从 L 个标记级状态减少到 K << L 个预测状态，并将标记级的序列深度减少到 L/K。在无条件的 OpenWebText 生成中，一个具有 1.97 亿参数的模型（K=8）在熵为 5.16 时实现了 22.8 的生成困惑度，使用每 1024 个标记文档 0.31 TFLOPs，相比之下，ELF-B 在 32 次采样步骤下的生成困惑度为 24.0，且使用 7.2 TFLOPs。相应的测量操作点分别产生 MAUVE@512 为 0.93 和 0.73。控制比较进一步表明，额外编码自身块的内容状态比预测状态产生 8-16% 更高的生成困惑度，并且用共享均值替换采样的预测状态轨迹会使 MAUVE@512 降低 0.11。这些结果表明，连续 DLMs 的潜在设计应考虑文本的紧凑表示方式，以及先验与解码器之间生成角色的划分。
cs.CL / 24 / 2610.12932
SEEDS: Preview-Guided Evidence Admission for Search Agents
SEEDS：基于预览引导的证据接纳机制用于搜索代理
Abstract
Search agents extend large reasoning models with multi-round access to external knowledge, yet most existing systems largely conflate retrieving evidence with consuming it. Retrieved documents are commonly incorporated according to fixed or rank-based policies, despite the fact that evidence utility is conditioned on the agent's current reasoning state. Consequently, a lower-ranked candidate may be critical for the current step, while a higher-ranked candidate may provide redundant, irrelevant, or misleading information. We formulate this overlooked problem as evidence admission, which asks, at each search step, which retrieved candidates should be allowed to enter the agent's evolving reasoning context. To enable such selective control, we introduce SEEDS (Selective Evidence Entry During Search), a preview-guided evidence admission framework for search agents. SEEDS first exposes lightweight previews of retrieved candidates and then lets the agent jointly determine which candidates and how many should be fully inspected, conditioned on its current reasoning state. Unlike fixed or adaptive prefix policies, SEEDS can select arbitrary subsets with flexible cardinality, and it can be instantiated directly with an existing reasoning agent or further optimized through learning. Across nine knowledge-intensive QA benchmarks under both local corpus and live web search, SEEDS consistently improves search effectiveness while substantially reducing full-evidence consumption. Experimental analyses show that SEEDS more effectively selects useful candidates beyond the corresponding retrieval prefix. These results establish evidence admission as a complementary dimension of agentic search, separating what information is retrieved from what information is ultimately allowed to shape reasoning.
Chinese Translation
搜索代理通过多轮访问外部知识来扩展大型推理模型，但现有大多数系统在很大程度上将检索证据与使用证据混为一谈。检索到的文档通常根据固定或基于排名的策略进行整合，尽管证据的有效性取决于代理当前的推理状态。因此，排名较低的候选文档可能对当前步骤至关重要，而排名较高的候选文档可能提供冗余、不相关或误导性的信息。我们将这一被忽视的问题表述为证据接纳，旨在每个搜索步骤中确定哪些检索到的候选文档应被允许进入代理不断演变的推理上下文。为了实现这种选择性控制，我们引入了SEEDS（搜索过程中的选择性证据进入），这是一个针对搜索代理的基于预览引导的证据接纳框架。SEEDS首先展示检索候选文档的轻量级预览，然后让代理共同决定哪些候选文档以及多少候选文档应被全面检查，这一决策依赖于其当前的推理状态。与固定或自适应前缀策略不同，SEEDS可以选择任意大小的子集，并且可以直接与现有的推理代理实例化，或通过学习进一步优化。在九个知识密集型问答基准测试中，无论是在本地语料库还是实时网络搜索下，SEEDS始终提高了搜索效果，同时显著减少了全面证据的消耗。实验分析表明，SEEDS在选择有用候选文档方面比相应的检索前缀更为有效。这些结果确立了证据接纳作为代理搜索的一个补充维度，区分了检索到的信息与最终允许影响推理的信息。
cs.CL / 25 / 2610.12939
Surfing the Activation Wave: Analyzing the Information and Geometry of LLM Hidden States for Speculative Decoding
冲浪激活波：分析大型语言模型（LLM）隐藏状态的信息与几何特征以进行推测解码
Abstract
Decoder hidden states from different layers of large language models (LLMs) often encode complementary information. However, existing speculative decoding methods typically rely on a small set of target-model layers chosen using fixed, position-based heuristics. Selecting an optimal subset of layers is inherently a combinatorial problem, making exhaustive evaluation infeasible even for moderately deep models. In this work, we present a principled framework for multi-layer hidden-state selection for speculative decoding based on two complementary criteria: informativeness, measured through matrix-entropy analysis, and diversity, quantified using centered kernel alignment (CKA). We investigate layer-wise Shannon entropy, aggregate entropy under hidden-state concatenation, and pairwise CKA dynamics across depth, revealing distinct geometric and informational compartments within decoder representations. Motivated by these findings, we propose entropy and CKA-based layer selection algorithms that favor hidden-state sets that are both highly informative and minimally redundant. As part of this framework, we introduce token-entropy filtering, which removes hidden states associated with low predictive-entropy tokens, thereby improving the quality of layer-selection statistics. In EAGLE3-style speculative decoding, our selected sets outperform fixed baseline choices, including the default $[2, N//2, N-3]$ rule, while providing a clearer explanation of why some defaults can remain competitive. Our results provide a practical recipe for multi-layer hidden-state selection for speculative decoding, insights into the geometry and information in LLM decoder states, and a foundation for extending similar analysis to related applications.
Chinese Translation
大型语言模型（LLMs）不同层的解码器隐藏状态通常编码互补信息。然而，现有的推测解码方法通常依赖于一小组通过固定的、基于位置的启发式方法选择的目标模型层。选择最优层子集本质上是一个组合问题，即使对于适度深度的模型，全面评估也是不可行的。在本研究中，我们提出了一个基于两个互补标准的多层隐藏状态选择的原则框架：信息量，通过矩阵熵分析测量，以及多样性，通过中心核对齐（CKA）量化。我们研究了逐层香农熵、在隐藏状态连接下的聚合熵，以及深度上的成对CKA动态，揭示了解码器表示中的不同几何和信息区间。基于这些发现，我们提出了基于熵和CKA的层选择算法，优先选择既高度信息化又最小冗余的隐藏状态集。作为该框架的一部分，我们引入了令牌熵过滤，去除与低预测熵令牌相关的隐藏状态，从而提高层选择统计的质量。在EAGLE3风格的推测解码中，我们选择的集合优于固定基线选择，包括默认的 $[2, N//2, N-3]$ 规则，同时提供了更清晰的解释，说明为什么某些默认设置仍然具有竞争力。我们的结果为推测解码的多层隐藏状态选择提供了实用的方案，对LLM解码器状态中的几何和信息提供了洞见，并为将类似分析扩展到相关应用奠定了基础。
cs.CL / 26 / 2610.12987
Forecasting Corpus-Sampling Disagreement in Masked Text Reconstruction
预测掩码文本重建中的语料库抽样分歧
Abstract
If a second text corpus were collected from the same population, how much would a predictor's outputs change? We study this question for masked-text reconstruction using a smoothed predictor built from nearby token-pair counts and evaluated over the full 50,257-token GPT-2 vocabulary. Treating each source document as one observation, we derive a first-order formula for the expected disagreement between predictors fitted to two independent corpora. The formula accounts for variable document length, random context counts, and correlations across token offsets, and it can be evaluated from a separate reference corpus without access to either target corpus. On TinyStories, WikiText-103, and CNN/DailyMail, growing each corpus from 512 to 2,048 chunks of 64 tokens lowers disagreement by only 5--40\%, far short of the 75\% implied by $1/n$ scaling. The forecast tracks this slow decline, which we trace to smoothing of rarely seen contexts. Forecast accuracy depends less on how these correlations are modeled than on the size of the reference corpus: small references systematically underpredict disagreement. A source-split jackknife that uses only reference data reduces this calibration error by 57--59\%. After this correction, a simpler smoothed-multinomial approximation is most accurate when the reference corpus is smaller than the target corpus, and a nonlinear source bootstrap is most accurate when it is larger. Finally, small masked diffusion language models trained on the same corpora show that the count-based forecasts partly identify which inputs are most sensitive to the corpus, but miss the neural models' much faster decline in disagreement.
Chinese Translation
如果从同一群体收集第二个文本语料库，预测器的输出会发生多大变化？我们研究了这一问题，针对掩码文本重建，使用基于附近词对计数构建的平滑预测器，并在完整的50,257个词汇的GPT-2词汇表上进行评估。将每个源文档视为一个观察值，我们推导出一个一阶公式，用于计算拟合于两个独立语料库的预测器之间的预期分歧。该公式考虑了可变文档长度、随机上下文计数和跨词偏移的相关性，并且可以从一个独立的参考语料库中评估，而无需访问任何目标语料库。在TinyStories、WikiText-103和CNN/DailyMail上，将每个语料库从512个64词块扩展到2048个，仅降低了5%至40%的分歧，远低于$1/n$缩放所暗示的75%。该预测跟踪了这种缓慢下降，我们将其归因于对少见上下文的平滑处理。预测的准确性与这些相关性建模的方式关系不大，而与参考语料库的大小密切相关：小型参考语料库系统性地低估了分歧。仅使用参考数据的源分裂自助法将这种校准误差降低了57%至59%。在这一修正后，当参考语料库小于目标语料库时，简单的平滑多项式近似最为准确；而当参考语料库大于目标语料库时，非线性源自助法最为准确。最后，在相同语料库上训练的小型掩码扩散语言模型表明，基于计数的预测部分识别了哪些输入对语料库最敏感，但未能捕捉到神经模型中分歧下降的速度更快。
cs.CL / 27 / 2610.12989
Language Models Have Interpretable Narrative Mechanisms
语言模型具有可解释的叙事机制
Abstract
We investigate whether language models contain internal components that contribute to representing narrativity, the degree to which a text exhibits narrative properties. Across GPT-2-Small, Gemma-2-2B, and Llama-3.1-8B, we identify attention heads, MLP neurons, and sparse autoencoder features whose activations correlate with narrativity. Ablating selected neuron and feature sets increases next-token prediction loss more for high-narrativity than for low-narrativity texts, and reduces the narrativity of open-ended generations. On downstream benchmarks, the largest effects occur in story completion and narrative question answering, although they vary across models and languages; results on confabulation and hallucination benchmarks are mixed. These findings provide causal evidence that selected sets of narrative-associated components contribute to model behaviour, while their internal organisation, relation to narrative theory, and broader functions remain open questions.
Chinese Translation
我们研究语言模型是否包含内部组件，这些组件有助于表征叙事性，即文本展现叙事特性的程度。在 GPT-2-Small、Gemma-2-2B 和 Llama-3.1-8B 中，我们识别出注意力头、MLP 神经元和稀疏自编码器特征，其激活与叙事性相关。消融选定的神经元和特征集对高叙事性文本的下一个标记预测损失的增加程度大于对低叙事性文本的影响，并且降低了开放式生成的叙事性。在下游基准测试中，最大的影响发生在故事补全和叙事问答中，尽管这些效果在不同模型和语言之间有所不同；关于虚构和幻觉基准的结果则较为混杂。这些发现提供了因果证据，表明选定的叙事相关组件集对模型行为有贡献，而它们的内部组织、与叙事理论的关系以及更广泛的功能仍然是未解之谜。
cs.CL / 28 / 2610.12995
What Should SFT Learn? Conditional Information Gain for Generalizable Post-Training
SFT 应该学习什么？用于可泛化后训练的条件信息增益
Abstract
Supervised fine-tuning (SFT) remains a widely used approach for post-training large language models, because fixed demonstrations can directly introduce desired reasoning strategies and behavioral patterns. Yet forcing a model to imitate externally provided trajectories may also interfere with behaviors already supported by pretraining, limiting generalization beyond the training distribution. Recent dynamic SFT methods mitigate this issue by reweighting token-level supervision according to model-side signals such as predictive probability, prior support, or distributional compatibility. However, model compatibility is not equivalent to learning value: a token that is weakly supported by the current model may either be an undesirable mismatch or precisely the task-specific decision that the demonstration is intended to teach. We propose CIG-SFT, an information-theoretic framework that estimates the task-conditioned information carried by each demonstrated token. Under teacher forcing, we compare the token likelihood from a frozen reference model with and without access to the task input while keeping the gold solution prefix fixed. The resulting conditional information gain quantifies how much the task itself reduces uncertainty about the demonstrated decision. We then convert this information into a KL-regularized projection of the current policy and derive an equivalent scalar-weighted SFT objective whose logit gradient exactly matches that of the projected target. Across mathematical reasoning and code-generation benchmarks, CIG-SFT consistently outperforms standard SFT and recent adaptive fine-tuning baselines, with particularly strong improvements in generalization to challenging evaluation distributions, while requiring neither additional training data nor online rollouts.
Chinese Translation
监督微调（SFT）仍然是后训练大型语言模型的广泛使用方法，因为固定的示例可以直接引入所需的推理策略和行为模式。然而，强迫模型模仿外部提供的轨迹可能会干扰预训练所支持的行为，限制了超出训练分布的泛化能力。最近的动态 SFT 方法通过根据模型侧信号（如预测概率、先前支持或分布兼容性）重新加权令牌级监督来缓解此问题。然而，模型兼容性并不等同于学习价值：一个当前模型支持较弱的令牌可能是一个不理想的不匹配，或者恰恰是示例所要教授的特定任务决策。我们提出了 CIG-SFT，这是一种信息论框架，用于估计每个示例令牌所携带的任务条件信息。在教师强制下，我们比较了来自冻结参考模型的令牌似然性，分析了在固定金标准前缀的情况下，是否可以访问任务输入。由此产生的条件信息增益量化了任务本身在多大程度上减少了对示范决策的不确定性。然后，我们将此信息转换为当前策略的 KL 正则化投影，并推导出一个等效的标量加权 SFT 目标，其对数梯度与投影目标的梯度完全匹配。在数学推理和代码生成基准测试中，CIG-SFT 始终优于标准 SFT 和最近的自适应微调基线，尤其在对具有挑战性的评估分布的泛化能力上表现出显著改善，同时不需要额外的训练数据或在线回滚。
cs.CL / 29 / 2610.12998
Evaluating LLM Creativity through Long-Tail Performance
通过长尾表现评估大型语言模型的创造力
Abstract
Creativity benchmarks for large language models (LLMs) vary widely in task and metric design. This raises the question of whether they capture a common notion of creativity. We propose that this shared target can be formalized as task-specific performance in the long tail of a reference input-output distribution, which describes how common input-output pairs are relative to a chosen source of text. We ground this proposition in the novelty-quality product formulation and evaluate it empirically across 15 benchmarks and 26 open-weight models, using n-gram absence from the pretraining data as a novelty proxy. Our results show that creativity benchmarks lie further in the tail regions than standard benchmarks. Moreover, on both creativity and standard benchmarks, model rankings from higher-tail subsets of each benchmark align more closely with human creativity judgments in LM-Arena. Building on these findings, we introduce q@Tail($\alpha$), a lower-cost creativity diagnostic that measures mean task quality on the tail fraction of an existing benchmark. Applying it to creativity-promoting methods reveals substantial room to improve LLM creativity through stronger tail performance.
Chinese Translation
大型语言模型（LLMs）的创造力基准在任务和指标设计上差异很大。这引发了一个问题，即这些基准是否捕捉到了创造力的共同概念。我们提出，这一共享目标可以形式化为参考输入-输出分布中长尾的任务特定表现，该分布描述了相对于所选文本来源的常见输入-输出对。我们将这一命题基于新颖性-质量产品公式，并在15个基准和26个开放权重模型上进行实证评估，使用预训练数据中n-gram的缺失作为新颖性的代理。我们的结果表明，创造力基准位于长尾区域的更远处，而不是标准基准。此外，在创造力和标准基准上，每个基准的高尾子集的模型排名与LM-Arena中的人类创造力判断更为一致。基于这些发现，我们引入了q@Tail($eta$)，这是一种低成本的创造力诊断工具，测量现有基准的尾部部分的平均任务质量。将其应用于促进创造力的方法显示，通过增强尾部表现，有显著提升LLM创造力的空间。
cs.CL / 30 / 2610.13025
Evo$^2$: Evolving the Evolver through Weight-Harness Co-Evolution
Evo$^2$: 通过权重-工具共进化演化进化者
Abstract
Self-evolving agents are increasingly used to improve language model systems without human intervention through either weight updates or harness improvements. Existing approaches typically pursue these directions separately: some methods update model weights using self-generated experience, while others rewrite the harness that defines how an agent uses prompts, tools, context, and execution feedback. However, these approaches overlook the relationship between model weights and the agent harness. To connect these processes and ensure effective and smooth agent self-evolving, we propose Evo$^2$, a framework for weight-harness co-evolution. For harness updates, we design a unified dual-role pipeline that a single policy serves as both a Proposer, which writes harness edits, and a Reflector, which diagnoses their effects and guides revision. For weight updates, harness modifications and their execution outcomes guide the training of both the Proposer and the Reflector. We further develop EvoTIDE, a training pipeline that recovers usable trajectories through edit extraction and prefix continuation, then combines SFT with interleaved GRPO across the two roles. Extensive experiments across three coding benchmarks demonstrate that Evo$^2$ significantly outperforms existing self-evolution methods. It achieves 82.1% on Polyglot and 48.3% on SWE-bench Verified, compared with 62.6% and 41.0% for the strongest baseline, respectively. We also show that the evolved harnesses transfer to different tasks without further evolution and rank first in cross-benchmark evaluations. Our dual-role harness improvement framework is also able to extend to self-evolution with proprietary models, improving a GPT-5-mini agent's performance from 24.4% to 48.3%. These results demonstrate the applicability of the framework Evo$^2$ across domains and its broad transferability across models and tasks.
Chinese Translation
自我进化的智能体越来越多地被用于在没有人类干预的情况下改进语言模型系统，方法包括权重更新或工具改进。现有的方法通常分别追求这两个方向：一些方法使用自生成的经验更新模型权重，而另一些方法则重写定义智能体如何使用提示、工具、上下文和执行反馈的工具。然而，这些方法忽视了模型权重与智能体工具之间的关系。为了连接这些过程并确保智能体自我进化的有效性和平滑性，我们提出了Evo$^2$，一个权重-工具共进化的框架。对于工具更新，我们设计了一个统一的双重角色流程，其中单一策略既充当提议者（Proposer），负责编写工具编辑，又充当反射者（Reflector），负责诊断其效果并指导修订。对于权重更新，工具修改及其执行结果指导提议者和反射者的训练。我们进一步开发了EvoTIDE，一个通过编辑提取和前缀延续恢复可用轨迹的训练流程，然后在这两个角色之间结合了SFT和交错的GRPO。针对三个编码基准的广泛实验表明，Evo$^2$显著优于现有的自我进化方法。在Polyglot上达到82.1%，在SWE-bench Verified上达到48.3%，而最强基线分别为62.6%和41.0%。我们还展示了进化后的工具在不同任务中的迁移能力，无需进一步进化，并在跨基准评估中排名第一。我们的双重角色工具改进框架也能够扩展到具有专有模型的自我进化，将GPT-5-mini智能体的性能从24.4%提高到48.3%。这些结果展示了Evo$^2$框架在各个领域的适用性及其在模型和任务之间的广泛迁移能力。
cs.CL / 31 / 2610.13041
RAMP: Recurrent Adaptive Mixed Precision for Efficient Prefill in Recurrent-Depth Language Models
RAMP：用于递归深度语言模型高效预填充的递归自适应混合精度
Abstract
Recurrent-depth language models improve inference-time reasoning by repeatedly applying a shared network to refine internal representations, but the additional recurrent computation also increases inference cost. This overhead is particularly pronounced during long-context prefill, where the full input sequence is repeatedly processed across recurrent depth. Weight-and-activation quantization offers a natural way to accelerate this computation through lower-precision arithmetic, yet repeatedly applying quantized computation introduces numerical perturbations that accumulate along the recurrent trajectory. We systematically analyze how such errors propagate across recurrent depth and show that quantization errors introduced earlier are attenuated by more subsequent recurrent steps and therefore have a smaller effect on the final representation. Matched-budget experiments strongly support this prediction, showing that placing quantized computation earlier in recurrent depth consistently yields the best fidelity. Based on this insight, we introduce RAMP (Recurrent Adaptive Mixed Precision), which performs early recurrent refinement using low-precision weight-and-activation computation and restores high precision for terminal refinement. Across context lengths up to 4K, RAMP preserves near-full-precision prefill quality while achieving up to 1.16x end-to-end first-prefill speedup on an NVIDIA H100, corresponding to a 13.9% latency reduction.
Chinese Translation
递归深度语言模型通过反复应用共享网络来改进推理时的推理能力，以细化内部表示，但额外的递归计算也增加了推理成本。这种开销在长上下文预填充期间尤为明显，因为完整的输入序列在递归深度中被反复处理。权重和激活量化提供了一种自然的方式，通过低精度算术加速这一计算，然而反复应用量化计算会引入沿递归轨迹累积的数值扰动。我们系统地分析了这些错误如何在递归深度中传播，并表明早期引入的量化错误会被后续更多的递归步骤减弱，因此对最终表示的影响较小。匹配预算的实验强烈支持这一预测，表明在递归深度中较早放置量化计算始终能获得最佳的保真度。基于这一见解，我们提出了RAMP（递归自适应混合精度），该方法使用低精度的权重和激活计算进行早期递归细化，并在终端细化时恢复高精度。在上下文长度达到4K的情况下，RAMP保持近乎全精度的预填充质量，同时在NVIDIA H100上实现高达1.16倍的端到端首次预填充加速，对应于13.9%的延迟减少。
cs.CL / 32 / 2610.13057
CoBe: A Benchmark For Conversational Counterfactual Text Editing
CoBe：对话式反事实文本编辑的基准测试
Abstract
Counterfactual text editing is the ability to revise a given text in light of a hypothetical change while preserving facts not causally affected by that change. When asked what would have happened if a doctor had not recommended surgery, the model output should retain the patient's medical history and modify only the consequences of the decision. This requires causal understanding: a model must determine, without being explicitly instructed to, what events are affected by the hypothetical change and what should remain fixed. This skill is central to human reasoning, especially in safety-critical domains like healthcare or law. By contrast, the task of associational text generation merely generates text statistically associated with the altered condition. In the medical example, an associational response might make the patient's symptoms less severe because milder cases correlate with no surgery. Such responses produce a plausible scenario, but not the counterfactual version of the given one. Prior counterfactual-text datasets do not clearly separate these modes of reasoning. We introduce the Counterfactual Benchmark (CoBe), a testbed for models' zero-shot ability to counterfactually edit a given textual scenario when prompted to do so in a conversationally natural way. CoBe contains 2000 questions spanning healthcare, science and engineering, business, and everyday scenarios. We test a suite of frontier and smaller models, with frontier systems achieving an average score of $58.05\%$. Our findings indicate that models frequently default to associational generation rather than preserving the causal structure of the original scenario. Moreover, performance does not scale consistently with either model size or output token length, suggesting that the failure may reflect model training objectives rather than insufficient model capacity.
Chinese Translation
反事实文本编辑是指在假设变化的背景下修订给定文本的能力，同时保留未受到该变化因果影响的事实。当被问及如果医生没有推荐手术会发生什么时，模型的输出应保留患者的病史，仅修改决策的后果。这需要因果理解：模型必须在没有明确指示的情况下，判断哪些事件受到假设变化的影响，哪些应保持不变。这一技能对于人类推理至关重要，尤其是在医疗或法律等安全关键领域。相比之下，关联文本生成的任务仅仅是生成与改变条件统计相关的文本。在医疗示例中，关联响应可能会使患者的症状减轻，因为较轻的病例与不手术相关联。这类响应产生了一个看似合理的情境，但并不是给定情境的反事实版本。之前的反事实文本数据集并未明确区分这些推理模式。我们引入了反事实基准测试（Counterfactual Benchmark, CoBe），这是一个测试模型在被提示以对话自然的方式反事实编辑给定文本场景的零样本能力的实验平台。CoBe包含2000个问题，涵盖医疗、科学与工程、商业和日常场景。我们测试了一系列前沿和小型模型，前沿系统的平均得分为58.05%。我们的研究结果表明，模型往往默认进行关联生成，而不是保留原始场景的因果结构。此外，性能并未随着模型规模或输出标记长度的一致增加而提升，这表明失败可能反映了模型训练目标而非模型容量不足。
cs.CL / 33 / 2610.13059
CliQ-RRG: Clinical-Knowledge Guided Disease-aware Visual-Textual Alignment for QA-Style Radiology Report Generation
CliQ-RRG：基于临床知识引导的疾病感知视觉-文本对齐用于问答式放射学报告生成
Abstract
Radiology reports are essential for diagnostic reasoning and patient care, yet their manual preparation is time-consuming and cognitively demanding. Automatic radiology report generation (RRG) offers a scalable alternative, but existing models often produce lengthy, unstructured narratives that overlook diagnostic cues and multi-view information. We present Clinical-Knowledge Guided Disease-aware Visual-Textual Alignment for QA-Style Radiology Report Generation (CliQ-RRG), a unified two-stage framework for interpretable and clinically grounded reporting. In Stage 1, CliQ-RRG employs a Disease-aware Visual-Textual Alignment module that aligns image and text representations using predicted disease embeddings, reinforced by a Prior-Guided Attention Module (PrAM) to capture multi-view dependencies across current and prior scans. In Stage 2, GraphRAG-guided retrieval augments intermediate textual representations with graph-grounded clinical evidence, and a large language model restructures them into concise, interpretable question-answer (QA) pairs with diagnostic summaries. Experiments on two public chest X-ray benchmarks demonstrate that CliQ-RRG consistently outperforms prior methods across both lexical and clinical metrics, generating accurate and clinically coherent QA-style radiology reports. Code is available at: https://github.com/chandravardhan-singh-raghaw/CliQ-RRG.
Chinese Translation
放射学报告对于诊断推理和患者护理至关重要，但其手动准备过程耗时且认知负担重。自动放射学报告生成（RRG）提供了一种可扩展的替代方案，但现有模型往往生成冗长且无结构的叙述，忽视了诊断线索和多视图信息。我们提出了基于临床知识引导的疾病感知视觉-文本对齐用于问答式放射学报告生成（CliQ-RRG），这是一个统一的两阶段框架，旨在实现可解释且具有临床基础的报告。在第一阶段，CliQ-RRG采用疾病感知视觉-文本对齐模块，通过预测的疾病嵌入对图像和文本表示进行对齐，并通过先验引导注意模块（PrAM）来捕捉当前和先前扫描之间的多视图依赖关系。在第二阶段，GraphRAG引导的检索通过图基础的临床证据增强中间文本表示，而大型语言模型将其重构为简洁、可解释的问答（QA）对及诊断摘要。在两个公共胸部X光基准测试上的实验表明，CliQ-RRG在词汇和临床指标上始终优于先前的方法，生成准确且临床一致的问答式放射学报告。代码可在以下链接获取：https://github.com/chandravardhan-singh-raghaw/CliQ-RRG。
cs.CL / 34 / 2610.13060
KV-Repurpose: A Testbed for Evaluating Non-Prefix KV-Cache Reuse
KV-Repurpose：评估非前缀KV缓存重用的测试平台
Abstract
KV-cache reuse in large language model serving is commonly limited to exact prompt prefixes, preventing reuse when the same retrieved document appears after different system prompts, conversation histories, or dynamic metadata. We present KV-Repurpose, an open-source, vLLM-based system for developing and evaluating segment-wise KV-cache reuse beyond exact prefixes. The system combines cache-aware segment scheduling, prompt construction, capacity management, workload generation, and runtime monitoring. Experiments on long-context benchmarks show that naive segment reuse largely preserves output quality for single-document question answering and summarization, while exposing clear limitations in cross-document reasoning. Under a Zipfian RAG-style workload in which 20% of requests share a fixed prefix, KV-Repurpose achieves a substantially higher token-level cache hit rate than prefix caching at the same 30 GB cache capacity, 63.9% versus 3.3%, resulting in lower TTFT. Even with 1 TB of cache capacity, prefix caching reaches only a 17.9% hit rate in this workload. Source code is available under the Apache-2.0 license at https://github.com/taiga333/eval-kv-repurpose.
Chinese Translation
在大型语言模型服务中，KV缓存重用通常仅限于精确的提示前缀，这使得在相同的检索文档出现在不同的系统提示、对话历史或动态元数据之后时无法进行重用。我们提出了KV-Repurpose，这是一个基于vLLM的开源系统，用于开发和评估超越精确前缀的分段KV缓存重用。该系统结合了缓存感知的分段调度、提示构建、容量管理、工作负载生成和运行时监控。在长上下文基准测试中的实验表明，简单的分段重用在单文档问答和摘要生成中大体上保持了输出质量，但在跨文档推理中暴露出明显的局限性。在一种Zipfian RAG风格的工作负载下，其中20%的请求共享固定前缀，KV-Repurpose在相同30 GB缓存容量下实现了显著更高的令牌级缓存命中率，分别为63.9%对比3.3%，从而导致更低的TTFT。即使在1 TB的缓存容量下，前缀缓存在该工作负载中的命中率也仅达到17.9%。源代码在Apache-2.0许可证下可在https://github.com/taiga333/eval-kv-repurpose获取。
cs.CL / 35 / 2610.13102
From 2.1M Hours of Raw Broadcast Audio to Finnish Speech SSL: Confidence Filtering and Architectural Analysis at Scale
从210万小时的原始广播音频到芬兰语自监督学习（SSL）：大规模的置信度过滤与架构分析
Abstract
Broadcast archives offer scale and diversity for self-supervised learning (SSL) but contain substantial noisy, non-speech, and multilingual content. Prior work typically relies on voice activity detection or language identification (LID) to extract speech, ensuring presence rather than quality. Whether and how to filter unlabeled speech by quality for SSL, particularly from noisy broadcasts, remains underexplored. We present a large-scale Finnish SSL study starting from 2.1M hours of unconstrained broadcast audio. We propose automatic speech recognition (ASR) confidence as a proxy for pretraining data quality and show 3-7% relative WER improvements over LID-based filtering despite a smaller training set. We further address the lack of controlled architectural comparisons in speech SSL by evaluating Transformer, Macaron-Net, Conformer, and E-Branchformer at matched parameter budgets. Our models improve upon prior Finnish SSL baselines, particularly at Base scale and out-of-domain.
Chinese Translation
广播档案为自监督学习（SSL）提供了规模和多样性，但其中包含大量噪声、非语音和多语言内容。之前的研究通常依赖于语音活动检测或语言识别（LID）来提取语音，确保存在性而非质量。如何以及是否通过质量过滤未标记的语音以用于SSL，特别是在嘈杂的广播中，仍然未得到充分探讨。我们展示了一项大规模的芬兰语SSL研究，起始于210万小时的无约束广播音频。我们提出将自动语音识别（ASR）置信度作为预训练数据质量的代理，并展示了相较于基于LID的过滤，尽管训练集较小，但相对字错误率（WER）提高了3-7%。我们进一步通过在匹配的参数预算下评估Transformer、Macaron-Net、Conformer和E-Branchformer，解决了语音SSL中缺乏受控架构比较的问题。我们的模型在芬兰语SSL基准上有所改进，特别是在基础规模和跨领域任务中。
cs.CL / 36 / 2610.13121
BenCoDiC: A Large-Scale Contextual Cyberbullying Corpus for Bengali Political News Discourse FULL VERSION
BenCoDiC：一个大规模的孟加拉政治新闻话语上下文网络欺凌语料库（完整版）
Abstract
Cyberbullying is a growing concern in online communities, yet existing Bengali corpora remain limited in contextual information, annotation transparency, scale, accessibility, and support for multidimensional analysis, thereby hindering robust detection and comprehensive study of harasser behaviors and networks. To address these gaps, we introduce the Bengali Contextual and Diverse Cyberbullying (\textit{BenCoDiC}) corpus, a publicly available contextual cyberbullying resource for Bengali political news discourse comprising 50,250 textual instances collected from 100 YouTube videos and their associated discussions. The corpus preserves caption--comment--reply relationships and is organized through a nine-dimensional star schema covering users, videos, textual content, context, time, engagement, political parties, target identities, and language categories. Five AI chatbots were initially evaluated for annotation but did not meet the predefined reliability criterion. Consequently, a two-tier human annotation process with expert verification was adopted, achieving Fleiss' $\kappa=0.86$ and Krippendorff's $\alpha=0.86$. Statistical analyses characterize the corpus, while transformer-based experiments show that incorporating conversational context yields modest but consistent improvements over isolated-text classification. \textit{BenCoDiC} supports context-aware cyberbullying detection and perpetrator-oriented behavioral and network analysis. The corpus is available at https://github.com/bencodic/BenCoDiC-Corpus.
Chinese Translation
网络欺凌在在线社区中日益成为一个关注的问题，但现有的孟加拉语语料库在上下文信息、注释透明度、规模、可获取性和对多维分析的支持方面仍然有限，从而阻碍了对施害者行为和网络的稳健检测和全面研究。为了解决这些问题，我们推出了孟加拉上下文多样化网络欺凌（Bengali Contextual and Diverse Cyberbullying， extit{BenCoDiC}）语料库，这是一个公开可用的孟加拉政治新闻话语上下文网络欺凌资源，包含从100个YouTube视频及其相关讨论中收集的50,250个文本实例。该语料库保留了标题-评论-回复关系，并通过覆盖用户、视频、文本内容、上下文、时间、参与度、政党、目标身份和语言类别的九维星型模式进行组织。最初评估了五个AI聊天机器人用于注释，但未达到预定义的可靠性标准。因此，采用了一个两级人类注释过程，并进行了专家验证，达到了Fleiss' $ ext{kappa}=0.86$ 和 Krippendorff's $ ext{alpha}=0.86$。统计分析对语料库进行了特征描述，而基于变换器的实验表明，结合对话上下文的分类在孤立文本分类上实现了适度但一致的改进。 extit{BenCoDiC}支持上下文感知的网络欺凌检测以及针对施害者的行为和网络分析。该语料库可在 https://github.com/bencodic/BenCoDiC-Corpus 获取。
cs.CL / 37 / 2610.13134
Neurograf v1: An End-to-End Romanian-First Language Model Family
Neurograf v1：一种端到端的罗马尼亚语优先语言模型家族
Abstract
We present Neurograf v1, an experimental end-to-end Romanian-first language-model pipeline developed to investigate the feasibility of training language models from scratch for a lower-resource European language. The pipeline comprises RoSfera, a 635 GB Romanian corpus containing 175.7 million documents; an 80k byte-level BPE tokenizer; five decoder-only models ranging from 160M to 6B parameters, trained on 131 billion tokens; three instruction-tuned assistants; and two contrastively trained embedding models. Training and evaluation were conducted using computing infrastructure at ELI-NP/IFIN-HH. The model family exhibits consistent scaling of validation loss and increasing Romanian factual and cultural knowledge with model size. On the 1,000-question WWTBM benchmark, Neurograf-6B achieves 69.89% accuracy on Romanian-context questions, while Neurograf-3.2B reaches 67.38%. Compute-optimal analysis yields a parameter-scaling exponent of approximately 0.423. Neurograf v1 serves as an experimental demonstrator rather than a production system. The study identifies practical challenges involving diacritic preservation, corpus quality, data composition, and tokenization efficiency, while demonstrating the feasibility of developing Romanian-first base models, instruction-following assistants, and semantic embedding models within a unified training pipeline.
Chinese Translation
我们提出了Neurograf v1，这是一个实验性的端到端罗马尼亚语优先语言模型管道，旨在研究从零开始训练低资源欧洲语言的语言模型的可行性。该管道包括RoSfera，一个包含6.35亿字节的罗马尼亚语语料库，包含1.757亿个文档；一个80k字节级别的BPE分词器；五个仅解码器模型，参数范围从1.6亿到60亿，训练于1310亿个标记；三个经过指令调优的助手；以及两个对比训练的嵌入模型。训练和评估是在ELI-NP/IFIN-HH的计算基础设施上进行的。该模型家族在验证损失的缩放和随着模型规模的增加而提升的罗马尼亚语事实和文化知识方面表现出一致性。在1000个问题的WWTBM基准测试中，Neurograf-6B在罗马尼亚语上下文问题上达到了69.89%的准确率，而Neurograf-3.2B达到了67.38%。计算最优分析得出的参数缩放指数约为0.423。Neurograf v1作为一个实验性演示工具，而非生产系统。研究识别了涉及变音符号保留、语料库质量、数据组成和分词效率的实际挑战，同时展示了在统一训练管道内开发罗马尼亚语优先基础模型、遵循指令的助手和语义嵌入模型的可行性。
cs.CL / 38 / 2610.13157
SOAR: Source-Aware Reinforcement Learning with Precise Credit Assignment for Search Agents
SOAR：具有精确信用分配的源感知强化学习用于搜索代理
Abstract
Reinforcement learning (RL) for search agents often suffers from evidence fabrication, a failure mode where agents bypass external information ($E_{ext}$) and fabricate an incorrect answer. We argue that this fabrication arises from an over-reliance on unreliable internal knowledge ($E_{int}$) as shortcuts due to the inherent opacity of reasoning trajectories. To address this, we propose SOAR (SOurce-Aware Reinforcement learning), a paradigm that transforms monolithic reasoning into source-explicit trajectories. SOAR (i) disentangles $E_{ext}$ and $E_{int}$ into parallel, observable streams via a Source-Aware Workflow, and (ii) incorporates Source-Consistent Reward to guide the policy evolution. By making information-source attribution explicit, SOAR enables precise credit assignment between different knowledge streams. Evaluations show that SOAR outperforms baselines on 3B and 7B backbones by 7.3% and 5.4%, while reducing evidence fabrication by 74.8%. These confirm that SOAR significantly improves performance on knowledge-intensive benchmarks while effectively mitigating evidence fabrication, fostering more faithful and grounded reasoning with enhanced robustness.
Chinese Translation
搜索代理的强化学习（RL）常常遭遇证据伪造，这是一种失败模式，代理绕过外部信息（$E_{ext}$）并伪造不正确的答案。我们认为，这种伪造源于对不可靠内部知识（$E_{int}$）的过度依赖，因而将其作为捷径，这与推理轨迹的固有不透明性有关。为了解决这个问题，我们提出了SOAR（源感知强化学习），这一范式将单一的推理转化为源明确的轨迹。SOAR（i）通过源感知工作流将$E_{ext}$和$E_{int}$解耦为平行的、可观察的流，并且（ii）结合源一致奖励来指导策略演化。通过明确信息源的归属，SOAR实现了不同知识流之间的精确信用分配。评估结果表明，SOAR在3B和7B骨干网络上分别比基线提高了7.3%和5.4%，同时将证据伪造减少了74.8%。这些结果确认SOAR显著提升了在知识密集型基准测试中的表现，同时有效减轻了证据伪造，促进了更真实和扎实的推理，并增强了鲁棒性。
cs.CL / 39 / 2610.13166
Scaling Agentic Data for Long-Horizon Terminal Intelligence
扩展代理数据以实现长时间范围的终端智能
Abstract
Terminal agents are moving beyond local operations toward completing real-world professional tasks. Short trajectories can demonstrate commands and tool calls, but they do not fully show how agents accumulate evidence through exploration, use it to inform judgments, continually adjust strategies based on feedback, and turn intermediate results into final deliverables. High-quality long-horizon trajectories capture these agentic behavioral patterns and their connections within a single problem-solving process, providing richer experience for sustained autonomous work. However, environment synthesis may complete or simplify the key work underlying these patterns in advance and thus remove the need for exploration, implementation, and verification that the agent was meant to perform. To address this, we introduce TerminalHorizon, a fully automated data synthesis engine for long-horizon terminal agents. Starting from real-world professional work, it uses independent execution to identify key work omitted during construction and uses the resulting evidence to guide environment reconstruction. To extend these patterns across stages, it progressively expands tasks under the same goal, introducing new requirements that build on prior results and call for further exploration and improvement. Using this engine, we construct 1.5K task environments and perform supervised fine-tuning of Qwen3.5-27B on the resulting dataset to obtain TerminalHorizon-27B. We achieve substantial performance gains on all 12 benchmarks spanning coding-agent tasks, agentic tasks in real-world production settings, and general reasoning. Multiple RSI case studies, each with a 12-hour research budget, further demonstrate the model's ability to continue experimenting beyond an initial solution, improve its methods based on feedback, and retain effective improvements in its final deliverables.
Chinese Translation
终端代理正在从局部操作转向完成现实世界的专业任务。短期轨迹可以展示命令和工具调用，但它们并未充分展示代理如何通过探索积累证据，利用这些证据来指导判断，基于反馈不断调整策略，并将中间结果转化为最终交付成果。高质量的长时间范围轨迹捕捉了这些代理行为模式及其在单一问题解决过程中的联系，为持续的自主工作提供了更丰富的经验。然而，环境合成可能会提前完成或简化这些模式背后的关键工作，从而消除代理原本需要进行的探索、实施和验证的必要性。为了解决这个问题，我们引入了TerminalHorizon，一个用于长时间范围终端代理的全自动数据合成引擎。它从现实世界的专业工作开始，利用独立执行来识别构建过程中遗漏的关键工作，并利用所得到的证据指导环境重建。为了在各个阶段扩展这些模式，它逐步扩展在同一目标下的任务，引入基于先前结果的新要求，并呼吁进一步的探索和改进。使用该引擎，我们构建了1500个任务环境，并对生成的数据集进行了Qwen3.5-27B的监督微调，以获得TerminalHorizon-27B。我们在涵盖编码代理任务、现实生产环境中的代理任务和一般推理的12个基准测试中实现了显著的性能提升。多个RSI案例研究，每个案例都有12小时的研究预算，进一步展示了模型在初始解决方案之外继续实验的能力，基于反馈改进其方法，并在最终交付成果中保持有效的改进。
cs.CL / 40 / 2610.13171
PreSTO: Predictive Subtree Prefetching for Fast LLM Power Sampling
PreSTO：用于快速大语言模型功率采样的预测子树预取
Abstract
Power sampling improves large language model (LLM) reasoning at test time by sampling from a sharpened version of the base model's distribution. Its sampler, PowerMH, is slow because its Metropolis--Hastings (MH) updates are sequential: each accept/reject decision sets the state from which the next proposal is generated. Prior work reduces this cost by changing the sampler or the model (e.g., sequential Monte Carlo, approximate lookahead, distillation). We propose Predictive SubTree Prefetching (PreSTO), which accelerates PowerMH by prefetching future proposals and provably leaves its MH transition kernel unchanged. PreSTO views future MH outcomes as a tree of accept/reject decisions: in each model call, it generates the proposals of a \textit{prefetched subtree} of this tree in one batch, then follows the realized path with the original MH acceptance rule. We also observe that MH reaches only a sparse, nearly unary part of each prefetched subtree, we further design a \textit{predictive traversal rule} that chooses the subtree to maximize the predicted number of completed transitions. Across seven datasets and five base LLMs, PreSTO achieves median speedups of $1.71\times$, $1.91\times$, and $2.33\times$ on PowerMH and its follow-ups EntropyCut and MultiTryMH, respectively; on the first two, it makes a median of $3.9\times$ fewer model calls, each only $2.1\times$ longer. On LCB V6 with Qwen3.5-9B, the predictive traversal rule is a further $1.25\times$--$1.37\times$ faster than the default breadth-first traversal.
Chinese Translation
功率采样通过从基础模型分布的锐化版本中采样，在测试时改善大语言模型（LLM）的推理。其采样器 PowerMH 速度较慢，因为其 Metropolis-Hastings（MH）更新是顺序进行的：每个接受/拒绝决策设置了生成下一个提案的状态。之前的研究通过更改采样器或模型（例如，顺序蒙特卡罗、近似前瞻、蒸馏）来降低这一成本。我们提出了预测子树预取（PreSTO），它通过预取未来提案来加速 PowerMH，并且在理论上保持其 MH 转移核不变。PreSTO 将未来的 MH 结果视为接受/拒绝决策的树：在每次模型调用中，它以一批生成该树的 extit{预取子树} 的提案，然后按照原始 MH 接受规则跟随实际路径。我们还观察到，MH 仅到达每个预取子树的稀疏、几乎单一部分，因此我们进一步设计了一种 extit{预测遍历规则}，选择子树以最大化预测的完成转移数量。在七个数据集和五个基础 LLM 上，PreSTO 在 PowerMH 及其后续的 EntropyCut 和 MultiTryMH 上分别实现了中位数加速比 $1.71 imes$、$1.91 imes$ 和 $2.33 imes$；在前两个上，它使模型调用的中位数减少了 $3.9 imes$，每次调用仅 $2.1 imes$ 更长。在 LCB V6 上，使用 Qwen3.5-9B，预测遍历规则比默认的广度优先遍历快 $1.25 imes$ 至 $1.37 imes$。
cs.CL / 41 / 2610.13207
SharP: Soft and Hard Prompt-Guided Augmentation with LLM for Low Resource Fake News Detection
SharP：基于软硬提示引导的低资源假新闻检测的软硬增强
Abstract
Fake news detection under low resource conditions is challenged by the scarcity of labeled data and the difficulty of capturing subtle deceptive patterns. In this work, we focus on low resource fake news detection, where only a small labeled set is available for training, while evaluation is performed under a fixed and unchanged test protocol. This setting is motivated by costly, ambiguous fact-checking and emerging topics and domains that are poorly covered by existing benchmarks. Existing data augmentation methods, such as synonym substitution, generative adversarial networks, and large language model (LLM)-based generation, often produce fluent but overly generic content, lacking the task-specific relevance needed for accurate detection. To address this, we propose SharP, a boundary-aware text generation framework guided by soft and hard prompts, which leverages LLMs as tools adapted to produce task-specific synthetic samples. SharP employs a dual prompt mechanism: hard prompts provide structured guidance and factual constraints, while soft prompts introduce learned semantic patterns from the data. These two types of prompts are combined to guide the generation process, enabling the model to produce samples that closely align with the characteristics of news. To further enhance detection performance, we incorporate a boundary-aware strategy that steers generation toward areas where the classifier is less confident, helping to clarify subtle distinctions between real and fake content. In addition, we adopt a dual objective optimization that balances semantic alignment with the source data and classifier feedback. This encourages the generation of samples that are both domain-consistent and helpful for refining decision boundaries, ultimately improving model robustness and generalization in low resource fake news detection.
Chinese Translation
在低资源条件下，假新闻检测面临标注数据稀缺和捕捉微妙欺骗模式的困难。在本研究中，我们专注于低资源假新闻检测，其中仅有少量标注数据可用于训练，而评估是在固定且不变的测试协议下进行的。这一设置的动机在于昂贵且模糊的事实核查，以及现有基准覆盖不足的新兴主题和领域。现有的数据增强方法，如同义词替换、生成对抗网络和基于大型语言模型（LLM）的生成，往往产生流畅但过于通用的内容，缺乏准确检测所需的任务特定相关性。为了解决这一问题，我们提出了SharP，一个由软硬提示引导的边界感知文本生成框架，利用LLM作为工具，适应性地生成任务特定的合成样本。SharP采用双提示机制：硬提示提供结构化指导和事实约束，而软提示则引入来自数据的学习语义模式。这两种提示结合在一起，引导生成过程，使模型能够生成与新闻特征紧密对齐的样本。为了进一步提升检测性能，我们结合了一种边界感知策略，引导生成朝向分类器信心较低的区域，帮助澄清真实内容与虚假内容之间的微妙区别。此外，我们采用双目标优化，平衡与源数据的语义对齐和分类器反馈。这鼓励生成既与领域一致又有助于细化决策边界的样本，最终提高模型在低资源假新闻检测中的鲁棒性和泛化能力。
cs.CL / 42 / 2610.13236
How Reliable are AMR Parser Outputs? Uncertainty Estimation for AMR Parsing
AMR 解析器输出的可靠性如何？AMR 解析的不确定性估计
Abstract
Recent advances have significantly improved the performance of Abstract Meaning Representation (AMR) parsers, yet determining when their outputs can be trusted remains largely unexplored. This work presents the first study of uncertainty estimation for AMR parsers in a black-box setting using top-k beam-search candidates. Uncertainty is estimated from pairwise candidate similarities and evaluated using the Prediction Rejection Ratio. This work characterises the relationship between parsing performance and uncertainty, evaluates the reliability of beam candidate distributions, analyses the effect of evaluation metrics on uncertainty assessment, and identifies parsers that provide the strongest balance between performance and reliability. Experimental results show the value of uncertainty estimation as a complementary dimension for evaluating AMR parsers.
Chinese Translation
近期的进展显著提升了抽象意义表示（Abstract Meaning Representation, AMR）解析器的性能，但确定其输出何时可以被信任仍然在很大程度上未被探索。本研究首次在黑箱环境中使用 top-k 波束搜索候选者对 AMR 解析器的不确定性进行估计。不确定性是通过成对候选者相似度进行估计，并使用预测拒绝比（Prediction Rejection Ratio）进行评估。本研究描述了解析性能与不确定性之间的关系，评估了波束候选者分布的可靠性，分析了评估指标对不确定性评估的影响，并识别出在性能和可靠性之间提供最佳平衡的解析器。实验结果显示，不确定性估计作为评估 AMR 解析器的补充维度具有重要价值。
cs.CL / 43 / 2610.13287
Do Prompt Optimizers Break Requirements They Are Not Scored On? A Preregistered Study
提示优化器是否会违反未评分的要求？一项预注册研究
Abstract
Prompt optimizers rewrite a prompt to raise one target metric, but prompts also state requirements the metric does not score, such as an output format, citations, or abstention. In a preregistered study, we run GEPA, MIPROv2, and TextGrad on HotpotQA, SQuAD 2.0, and IFEval and count per-item violations of unscored requirements with deterministic checks. By the preregistered criterion, all three optimizers produce regressions, substantially on HotpotQA, where net regression in citation validity reaches 10.7 to 41.6 points while exact match rises by 21 to 34 points. A post-hoc decomposition changes this reading. The optimized prompts elicit reasoning before the required JSON object, and under the 200-token output cap used throughout, 13.9% to 47.0% of responses end before any answer appears. Among parseable outputs, net regression is within 3 points of zero. With a 1,024-token cap, extraction failures disappear, content regression stays below 1 point, and exact match rises further. Two simple gates fail the preregistered criterion. The side effect is narrower than hypothesized: optimized prompts are more accurate and break an explicit output contract, and a lenient target metric does not show the cost.
Chinese Translation
提示优化器重写提示以提高一个目标指标，但提示还陈述了该指标未评分的要求，如输出格式、引用或弃权。在一项预注册研究中，我们对 HotpotQA、SQuAD 2.0 和 IFEval 运行 GEPA、MIPROv2 和 TextGrad，并通过确定性检查计算每个项目未评分要求的违反情况。根据预注册标准，所有三种优化器都产生了回归，尤其是在 HotpotQA 上，引用有效性的净回归达到 10.7 到 41.6 分，而精确匹配上升了 21 到 34 分。事后分解改变了这一解读。优化后的提示在所需的 JSON 对象之前引发推理，并且在整个过程中使用的 200 个标记的输出上限下，13.9% 到 47.0% 的响应在任何答案出现之前就结束。在可解析的输出中，净回归在零的 3 分范围内。使用 1,024 个标记的上限，提取失败消失，内容回归保持在 1 分以下，精确匹配进一步上升。两个简单的门控未能满足预注册标准。副作用比假设的要小：优化后的提示更准确并违反了明确的输出合同，而宽松的目标指标并未显示出成本。
cs.CL / 44 / 2610.13303
Agreement Without Alignment: Label-Free Rubric Clarification Increases LLM-Judge Consensus but Not Agreement with Humans
无对齐的协议：无标签评分标准澄清增加了大型语言模型评审者的一致性，但未增加与人类的协议
Abstract
LLM judges are increasingly deployed as panels, and disagreement among panel members is often read as a symptom of an unclear rubric. We test a label-free procedure that turns this disagreement into rubric repairs. The procedure decomposes each rubric into sub-criteria, ranks them by cross-judge disagreement on a development split, and has an editor model rewrite only the most contested ones. An additional stage that moves text-checkable sub-criteria to deterministic checkers transferred none in our evaluation. Human labels are never used to select or edit clauses; they are used only to evaluate the outcome. We pre-specified four hypotheses in a version-controlled protocol and evaluated six rubric conditions with a panel of five judges from three providers on SummEval, the HealthBench meta-evaluation, and USR Topical-Chat, for a total of 129,770 judge requests, excluding retries. On SummEval, the procedure raised inter-judge Krippendorff's $\alpha$ by 0.069 (95% CI 0.046 to 0.092) over the original rubric and by 0.074 over a control that applied the same edits to an equal number of randomly chosen clauses that were not selected. Agreement with expert ratings showed no detectable change (Kendall $\tau$ difference -0.007, 95% CI -0.037 to 0.021). On HealthBench and USR, neither inter-judge agreement nor human agreement met the preregistered effect criteria. In a preregistered secondary comparison, human agreement after editing the contested clauses was lower than after editing the control clauses on all three datasets in point estimate, although no interval excluded zero. Selecting clauses by disagreement makes judges agree with each other more than editing arbitrary clauses does, but in our setting the added consensus did not bring the judges closer to human annotators. We will release code, frozen rubrics, item identifiers, and judge scores.
Chinese Translation
大型语言模型（LLM）评审者越来越多地作为评审小组进行部署，而小组成员之间的分歧常常被视为评分标准不清晰的症状。我们测试了一种无标签程序，将这种分歧转化为评分标准的修订。该程序将每个评分标准分解为子标准，根据开发集上的评审者之间的分歧对其进行排名，并让编辑模型仅重写最具争议的部分。在一个额外阶段中，将文本可检查的子标准转移到确定性检查器中，在我们的评估中没有转移任何内容。人类标签从未用于选择或编辑条款；它们仅用于评估结果。我们在一个版本控制的协议中预先指定了四个假设，并在SummEval、HealthBench元评估和USR Topical-Chat上评估了五名来自三个提供者的评审小组的六种评分标准条件，共计129,770个评审请求，排除重试。在SummEval上，该程序将评审者之间的Krippendorff's $eta$提高了0.069（95% CI 0.046至0.092），相较于原始评分标准，并在对照组中提高了0.074，该对照组对相同数量的随机选择的未被选中条款应用了相同的编辑。与专家评分的一致性没有显著变化（Kendall $ au$差异-0.007，95% CI -0.037至0.021）。在HealthBench和USR上，评审者之间的一致性和人类一致性均未达到预注册的效果标准。在一个预注册的次要比较中，编辑争议条款后的人类一致性在所有三个数据集上均低于编辑对照条款后的结果，尽管没有区间排除零。通过分歧选择条款使评审者之间的协议高于编辑任意条款，但在我们的设置中，增加的一致性并未使评审者更接近人类注释者。我们将发布代码、冻结的评分标准、项目标识符和评审分数。
cs.CL / 45 / 2610.13321
Host-Aware Web Crawling for LLM Pretraining Data
面向主机的网络爬虫用于大语言模型预训练数据
Abstract
Web crawling is a major source of large language model (LLM) pretraining data, yet data-selection pipelines discard a large share of crawled pages. We investigate how a crawler should allocate its budget across and within hosts to collect more high-quality pretraining data. Combining document-level quality classifier scores, web-graph centrality metrics, and Common Crawl's per-host crawl coverage, we find that roughly half of the variation in classifier scores occurs between hosts. In the crawls we analyze, a few pages from a host also provide a useful estimate of the quality of its remaining pages. Common Crawl uses harmonic centrality, a graph centrality metric, to guide its crawl allocation. In our simulations, using centrality to choose which hosts to crawl yields limited gains over random ordering and performs worse for some classifiers. For deciding how deeply to crawl each host, sampling a few pages and then fetching more from hosts with higher-scoring samples collects several times more high-quality pages per fetch than uniform depth or centrality-based allocation. Finally, Common Crawl's observed per-host coverage has little correlation with the evaluated quality signals. We estimate that tens of millions of high-quality pages remain among known but uncrawled URLs in each crawl examined, with most concentrated on a small share of hosts.
Chinese Translation
网络爬虫是大语言模型（LLM）预训练数据的主要来源，但数据选择流程却丢弃了大量爬取的页面。我们研究了爬虫应如何在主机之间及其内部分配预算，以收集更多高质量的预训练数据。结合文档级质量分类器得分、网络图中心性指标和Common Crawl的每主机爬取覆盖率，我们发现大约一半的分类器得分变异发生在不同主机之间。在我们分析的爬取数据中，来自某一主机的少数页面也能有效估计其余页面的质量。Common Crawl使用谐波中心性（harmonic centrality）这一图中心性指标来指导其爬取分配。在我们的模拟中，使用中心性选择爬取主机的效果与随机排序相比提升有限，并且在某些分类器上表现更差。对于决定每个主机的爬取深度，先随机抽取几页，然后从得分较高的主机获取更多页面，能够在每次获取中收集到比均匀深度或基于中心性的分配多几倍的高质量页面。最后，Common Crawl观察到的每主机覆盖率与评估的质量信号之间几乎没有相关性。我们估计，在每次爬取中，数千万个高质量页面仍然存在于已知但未爬取的URL中，并且大多数集中在少数主机上。
cs.CL / 46 / 2610.13348
Test-Time Memory Construction
测试时记忆构建
Abstract
Large Language Models (LLMs) struggle to answer questions over long interaction histories, where relevant evidence may be scattered across distant sessions. Existing memory systems typically construct and maintain sophisticated memory representations offline, before user queries are observed, incurring substantial overhead and limiting their ability to adapt memory construction to query-specific information needs. In this paper, we propose T$^2$Mem, a Test-Time Memory construction framework that defers memory construction to inference and progressively builds memory along the reasoning process. Specifically, T$^2$Mem extracts lightweight episode-level metadata offline to capture key events and temporal associations as retrieval cues, while retaining links to the original interactions to facilitate the reconstruction of dependencies across episodes. At test time, the LLM adaptively retrieves and organizes metadata and raw interactions, progressively constructing query-specific memory as reasoning unfolds. Experiments on LoCoMo and LongMemEval show the effectiveness of T$^2$Mem by consistently outperforming strong baselines across different backbone LLMs while substantially reducing the overhead of memory construction. Further analysis demonstrates that deferring memory construction to inference enables the memory construction process to dynamically adapt to the information needs of individual queries. All code is available at https://github.com/OpenBMB/T2Mem.
Chinese Translation
大型语言模型（LLMs）在处理长交互历史中的问题时表现不佳，因为相关证据可能分散在遥远的会话中。现有的记忆系统通常在观察到用户查询之前，离线构建和维护复杂的记忆表示，这会产生相当大的开销，并限制其根据查询特定信息需求适应记忆构建的能力。本文提出了 T$^2$Mem，一个测试时记忆构建框架，该框架将记忆构建推迟到推理阶段，并在推理过程中逐步构建记忆。具体而言，T$^2$Mem 离线提取轻量级的事件级元数据，以捕捉关键事件和时间关联作为检索线索，同时保留与原始交互的链接，以便重建跨事件的依赖关系。在测试时，LLM 自适应地检索和组织元数据及原始交互，随着推理的展开逐步构建查询特定的记忆。在 LoCoMo 和 LongMemEval 上的实验表明，T$^2$Mem 的有效性，通过在不同的基础 LLM 上始终优于强基线，同时显著减少记忆构建的开销。进一步的分析表明，将记忆构建推迟到推理阶段使得记忆构建过程能够动态适应单个查询的信息需求。所有代码可在 https://github.com/OpenBMB/T2Mem 获取。
cs.CL / 47 / 2610.13370
Overcoming Data Imbalances with Expert Selection Biases
克服数据不平衡的专家选择偏差
Abstract
Mixture-of-Experts (MoE) models have become the standard architecture for frontier LLMs. The multiple parallel experts in each transformer block theoretically enable expert specialization according to categories such as domains, languages, or modalities during pretraining. However, when data is imbalanced across categories, minority categories can become concentrated in a small fraction of the expert pool or spread inefficiently across them, effectively limiting their available capacity. In this work, we demonstrate that category-conditioned expert-selection biases can preferentially route tokens from different categories toward specialized expert pools. By applying biases based on input category, we provide a simple way to induce soft category-aligned specialization. This intervention encourages categories toward distinct expert subsets while retaining cross-category sharing when useful. More importantly, it enables the over-allocation of expert capacity to categories that would otherwise be disadvantaged by data scarcity. We observe performance improvements across modalities, low-resource languages, and code in models up to 10.6B parameters. At our smallest and largest model scales, these gains are equivalent to allocating 18.5% and 2.7% additional baseline training tokens, respectively. Overall, our results show that expert biases can serve as a lightweight control for category-level expert allocation, making room for low-resource categories while retaining standard MoE inference.
Chinese Translation
混合专家（Mixture-of-Experts, MoE）模型已成为前沿大型语言模型（LLMs）的标准架构。每个变换器块中的多个并行专家理论上能够根据领域、语言或模态等类别在预训练期间实现专家专业化。然而，当各类别的数据不平衡时，少数类别可能会集中在专家池的一小部分中，或者在专家之间分布不均，从而有效限制了它们的可用容量。在本研究中，我们展示了基于类别条件的专家选择偏差可以优先将来自不同类别的标记路由到专业化的专家池。通过应用基于输入类别的偏差，我们提供了一种简单的方法来诱导软类别对齐的专业化。这种干预鼓励类别朝向不同的专家子集，同时在有用时保持跨类别共享。更重要的是，它使得专家容量的过度分配成为可能，从而使那些因数据稀缺而处于劣势的类别受益。我们观察到在多模态、低资源语言和代码模型中，性能得到了提升，模型参数量高达10.6B。在我们最小和最大模型规模下，这些提升相当于分别分配了18.5%和2.7%的额外基线训练标记。总体而言，我们的结果表明，专家偏差可以作为类别级专家分配的轻量级控制，为低资源类别腾出空间，同时保持标准的MoE推理。
cs.CL / 48 / 2610.13375
From Structure to Validity: Mechanism-Guided Generation and Refinement of LLM Rubrics
从结构到有效性：机制引导的LLM评分标准生成与优化
Abstract
Rubrics have been widely used to evaluate and optimize language models on open-ended tasks. However, existing work primarily focuses on generating better rubrics, with limited understanding of what makes rubric design effective. In this work, we conduct a systematic mechanistic analysis of rubric design across criterion organization, criterion operationalization, and score aggregation. Our analysis shows that structural organization enables more comprehensive criterion construction, semantic specification alone does not ensure effective criterion-level judgments, and sophisticated score aggregation provides limited additional gains. Guided by these findings, we introduce Adaptive Rubric Tree Search with Hybrid refinement (ARTS-Hybrid), which combines adaptive tree construction with semantic-first, response-calibrated refinement. Experiments across benchmarks show that ARTS-Hybrid improves both end-to-end evaluation performance and intrinsic rubric quality. Further experiments suggest that, beyond evaluation, rubrics generated by ARTS-Hybrid provide effective signals for model optimization.
Chinese Translation
评分标准已被广泛用于评估和优化语言模型在开放式任务中的表现。然而，现有研究主要集中于生成更好的评分标准，对评分标准设计有效性的理解有限。在本研究中，我们对评分标准设计进行了系统的机制分析，涵盖了标准组织、标准操作化和分数聚合。我们的分析表明，结构化组织能够实现更全面的标准构建，仅凭语义规范并不能确保有效的标准级判断，而复杂的分数聚合提供的额外收益有限。基于这些发现，我们引入了带有混合优化的自适应评分树搜索（Adaptive Rubric Tree Search with Hybrid refinement，ARTS-Hybrid），该方法结合了自适应树构建与语义优先、响应校准的优化。跨基准的实验表明，ARTS-Hybrid不仅提高了端到端评估性能，还提升了评分标准的内在质量。进一步的实验表明，除了评估，ARTS-Hybrid生成的评分标准为模型优化提供了有效信号。
cs.CL / 49 / 2610.13386
MoPrO-SD: Modular Prompt Optimization for Stance Detection
MoPrO-SD：用于立场检测的模块化提示优化
Abstract
Stance detection remains challenging for large language models (LLMs) due to implicit opinions, sarcasm, and context-dependent expressions of stance. Although automatic prompt optimization offers an alternative to fine-tuning without updating model parameters, existing structured approaches are not specifically designed around stance-label semantics and class-specific ambiguities. We introduce MoPrO-SD, a stance-specific prompt optimization framework that decomposes prompts into functional semantic modules and refines them sequentially using LLM-based feedback. We study three refinement strategies: instruction-only optimization (I), instruction with reasoning (I+R), and instruction with per-example reasoning (I+R+E), using class-conditioned errors for stance-label refinement. We also introduce VaccineStance, a new human-annotated benchmark for vaccine stance detection. Experiments across four datasets and five open-weight LLMs show that MoPrO-SD improves upon the human-written Base prompt in many settings. On SemEval-2016, class-conditioned feedback achieves higher mean Macro-F1 than pooled feedback for four of five backbones. Both reasoning-based strategies, I+R and I+R+E, achieve strong performance, with their relative effectiveness varying across models and datasets, while remaining competitive with recent prompt optimization methods. The MoPrO-SD code and VaccineStance dataset are available on GitHub.
Chinese Translation
立场检测对于大型语言模型（LLMs）仍然具有挑战性，因为隐含的观点、讽刺和依赖上下文的立场表达使得问题复杂化。尽管自动提示优化提供了一种不更新模型参数的微调替代方案，但现有的结构化方法并未专门围绕立场标签语义和类别特定的模糊性进行设计。我们提出了MoPrO-SD，一个特定于立场的提示优化框架，它将提示分解为功能性语义模块，并使用基于LLM的反馈对其进行逐步优化。我们研究了三种优化策略：仅指令优化（I）、带推理的指令（I+R）和逐例推理的指令（I+R+E），利用类别条件错误进行立场标签的优化。我们还引入了VaccineStance，一个新的人工标注的疫苗立场检测基准。在四个数据集和五个开放权重的LLM上进行的实验表明，MoPrO-SD在许多设置中优于人工编写的基础提示。在SemEval-2016上，类别条件反馈在五个基础模型中有四个的平均宏F1值高于汇总反馈。基于推理的两种策略I+R和I+R+E表现出强劲的性能，其相对有效性在不同模型和数据集之间有所变化，同时与近期的提示优化方法保持竞争力。MoPrO-SD代码和VaccineStance数据集已在GitHub上发布。
cs.CL / 50 / 2610.13398
Towards the Roof of the World: NLP Progress in Tibetan Languages
走向世界屋脊：藏语自然语言处理的进展
Abstract
Tibetan, spoken by some seven million people on and around the Qinghai-Xizang Plateau, remains underrepresented in natural language processing (NLP). We present, to our knowledge, the first peer-reviewed systematic bibliometric survey of Tibetan NLP: 140 publications from 2004 to 2026 in the ACL Anthology and DBLP, annotated for venue type, research area, first-affiliation country, and resource release. Output has grown rapidly, with 84% of the publications appearing since 2018, and has expanded bottom-up along the NLP pipeline, from corpora and word segmentation through script, speech, and applications to large language models (LLMs). Researchers based in China first-authored 87% of the publications, which span all 25 represented research areas, whereas work from other countries concentrates on Classical Tibetan. We argue that this full-pipeline paradigm, realized under the compounded constraints of a complex script, dialectal diversity, and geographic remoteness, offers a reference model for other low-resource languages of the Global South. We also identify gaps in dialect coverage, responsible NLP, native and comparable evaluation, cross-community visibility, and resource release (only 29% of the publications release data, code, or models), and recommend open science, shared standards, and community participation for the LLM era.
Chinese Translation
藏语是青海-西藏高原及其周边地区约七百万人的母语，在自然语言处理（NLP）领域仍然处于代表性不足的状态。我们首次呈现了经过同行评审的系统性文献计量调查，涵盖了2004年至2026年间在ACL文集和DBLP上发表的140篇藏语NLP相关出版物，并对其进行了场所类型、研究领域、第一作者所属国家和资源发布的注释。自2018年以来，出版物数量迅速增长，84%的出版物在此期间出现，并且沿着NLP流程自下而上扩展，从语料库和分词到脚本、语音及大型语言模型（LLMs）应用。中国的研究者首次作者占87%的出版物，涵盖了所有25个研究领域，而其他国家的研究则集中于古典藏语。我们认为，在复杂的书写系统、方言多样性和地理偏远性等复合约束下实现的这一全流程范式，为全球南方其他低资源语言提供了参考模型。我们还识别了方言覆盖、负责任的NLP、本土和可比评估、跨社区可见性以及资源发布（仅29%的出版物发布数据、代码或模型）等方面的不足，并建议在大型语言模型时代倡导开放科学、共享标准和社区参与。
cs.CL / 51 / 2610.13456
Towards Enriching Design-Time Models with Evidence from Simulation and Operational Activities
通过模拟和操作活动的证据丰富设计时模型
Abstract
Cyber-Physical Systems are developed through multiple model based activities, comprising several stages of development. As the development progresses through simulation, experimentation, and operation, observations of system execution may reveal missing, uncertain, or invalid design knowledge. The path from execution results to persistent design-time enrichment, however, is often implicit. This paper addresses this challenge by proposing an evidence driven workflow for systematically transferring knowledge produced during simulation and operation back into design-time models. We demonstrate the workflow using an Influence Model, a design-time artefact used to represent how design artefacts and environmental factors jointly affect system response properties, using a sign-following robot use case. We show how simulation evidence can enrich Influence Models by supporting participant pruning decisions, interaction characterisation, influence-function abstraction, and validity-scope information. The application demonstrates the feasibility of the workflow for simulation-based evidence and its potential benefits to improve design-time reasoning and feedback in subsequent development iterations.
Chinese Translation
网络物理系统通过多种基于模型的活动进行开发，涵盖多个开发阶段。随着开发通过模拟、实验和操作的推进，系统执行的观察可能揭示缺失、不确定或无效的设计知识。然而，从执行结果到持久的设计时丰富的路径往往是隐性的。本文通过提出一个以证据驱动的工作流程，系统性地将模拟和操作过程中产生的知识转移回设计时模型，来解决这一挑战。我们使用影响模型（Influence Model）演示该工作流程，该模型是用于表示设计工件和环境因素如何共同影响系统响应特性的设计时工件，采用一个跟随信号的机器人用例。我们展示了如何通过支持参与者修剪决策、交互特征化、影响函数抽象和有效性范围信息，利用模拟证据丰富影响模型。该应用展示了基于模拟的证据工作流程的可行性及其在后续开发迭代中改善设计时推理和反馈的潜在益处。
cs.CL / 52 / 2610.13476
A Better Metric Will Not Save You: What You Compare Beats How You Compare It
更好的度量无法拯救你：比较的内容胜过比较的方式
Abstract
Representational similarity measures (RSA, CKA, Procrustes) are chosen for the transformations they treat as irrelevant, and all of them take the set of conditions as given. That set is fixed by two decisions, which inputs are admitted and how they are grouped, both taken before the matrix a measure operates on exists: no invariance property reaches them. We measure what they are worth where the arrangement the conditions ought to have is known independently of the systems compared, the acoustic organisation of vowels, across fifteen languages and twenty-seven representations. Changing which categories are compared, at a fixed number of them, moves the gap between two representations by more than twice the effect being measured, while changing how many moves it not at all. In this setting rankings of models are identified under these choices, and rankings across languages are not.
Chinese Translation
表征相似性度量（Representational Similarity Measures, RSA; Centered Kernel Alignment, CKA; Procrustes）是基于它们认为无关的变换而选择的，所有这些度量都将条件集视为给定。这个条件集由两个决策固定：哪些输入被接受以及它们如何分组，这两个决策在度量操作的矩阵存在之前就已做出：没有不变性属性能够影响它们。我们在已知条件应有的排列的情况下测量它们的价值，这些条件独立于被比较的系统，涉及到十五种语言和二十七种表征的元音的声学组织。改变比较的类别（在固定数量的情况下），会使两个表征之间的差距超过被测量效应的两倍，而改变比较的数量则不会产生任何影响。在这种情况下，模型的排名在这些选择下被识别，而跨语言的排名则没有。
cs.CL / 53 / 2610.13485
Proactive Agentic Orchestration
主动代理编排
Abstract
Agentic orchestration delegates tasks to sub-agents with independent contexts, providing a key path to extending the effective execution horizon of large language model agents. However, existing orchestration methods face two limitations: in execution, current methods fix sub-agents' goals and harnesses at creation time, limiting their adaptability to dynamic real-world environments; in optimization, sparse and delayed feedback constrains effective learning of the orchestration policy. To address these limitations, we introduce POrchestra, an on-the-fly proactive agentic orchestration framework. It allows sub-agents to proactively communicate with the orchestrator at any logical point to request harness adjustments while maintaining continuity of local execution. This mechanism not only enables sub-agents to adapt seamlessly to dynamic environments but also naturally inspires On-Policy Orchestration Distillation (OPOD) based on hindsight feedback. OPOD attributes sub-agents' proactive revision requests back to preceding orchestration decisions to derive targeted improvement signals, and leverages these signals as process supervision alongside the final reward to optimize the orchestration policy. We evaluate POrchestra on three long-horizon agent benchmarks spanning deep research, dynamic digital assistance, and software engineering. POrchestra improves task success by an average of 9.1% relative to the strongest baseline; combined with OPOD, it improves small-orchestrator success by an average of 10.0% relative to existing training baselines. POrchestra also achieves a more favorable cost-performance trade-off. Our code is publicly available at https://github.com/XueyangFeng/POrchestra.
Chinese Translation
代理编排将任务委派给具有独立上下文的子代理，为扩展大型语言模型代理的有效执行范围提供了关键路径。然而，现有的编排方法面临两个限制：在执行方面，当前方法在创建时固定了子代理的目标和工具，限制了它们对动态现实环境的适应性；在优化方面，稀疏和延迟的反馈限制了编排策略的有效学习。为了解决这些限制，我们引入了POrchestra，一个即时主动代理编排框架。它允许子代理在任何逻辑点主动与编排者沟通，请求工具调整，同时保持本地执行的连续性。这一机制不仅使子代理能够无缝适应动态环境，还自然激发了基于事后反馈的在线政策编排蒸馏（On-Policy Orchestration Distillation, OPOD）。OPOD将子代理的主动修订请求归因于先前的编排决策，以推导出针对性的改进信号，并利用这些信号作为过程监督，与最终奖励一起优化编排策略。我们在三个长时间跨度的代理基准上评估了POrchestra，涵盖深度研究、动态数字助理和软件工程。与最强基线相比，POrchestra将任务成功率平均提高了9.1%；结合OPOD，相较于现有训练基线，小型编排者的成功率平均提高了10.0%。POrchestra还实现了更有利的成本-性能权衡。我们的代码已公开发布在 https://github.com/XueyangFeng/POrchestra。
cs.CL / 54 / 2610.13550
Personalization Without Stereotyping: Do Speech LLMs Know When to Turn a Deaf Ear?
无刻板印象的个性化：语音大语言模型是否知道何时充耳不闻？
Abstract
Recent speech LLMs can leverage rich paralinguistic cues, enabling more complex and personalized interactions. However, little work investigates how this sensitivity can lead to stereotyping, and existing evaluations largely measure demographic disparities without distinguishing beneficial personalization from stereotypical behavior. We introduce BASIS-Bench, a benchmark designed to disentangle these two aspects through three axes: (1) demographic robustness, measuring disparities in content-based tasks across speakers and demographic classification capabilities; (2) stereotypical bias, measuring whether perceived demographic cues trigger unprompted stereotypical behavior; and (3) personalization, measuring whether models can tailor responses to demographic characteristics when explicitly requested. We evaluate eight recent speech LLMs and find that, despite often recognizing demographic cues, most fail to translate them into appropriate personalization. Explicit demographic labels improve personalization but also amplify stereotypical associations already present in the textual backbones. These findings highlight the need to train speech LLMs not only to perceive demographic cues, but to learn when they should shape their responses.
Chinese Translation
近期的语音大语言模型（Speech LLMs）能够利用丰富的副语言线索，从而实现更复杂和个性化的互动。然而，目前对这种敏感性如何导致刻板印象的研究较少，现有的评估主要测量人口统计差异，而未能区分有益的个性化与刻板行为。我们提出了BASIS-Bench，这是一个旨在通过三个维度解开这两个方面的基准： (1) 人口统计稳健性，测量在内容相关任务中不同说话者之间的差异及其人口统计分类能力； (2) 刻板偏见，测量感知到的人口统计线索是否会触发无提示的刻板行为； (3) 个性化，测量模型在明确请求时是否能够根据人口特征调整响应。我们评估了八个近期的语音大语言模型，发现尽管它们通常能够识别人口统计线索，但大多数模型未能将其转化为适当的个性化。明确的人口统计标签改善了个性化，但也加剧了文本基础中已存在的刻板联想。这些发现强调了训练语音大语言模型的必要性，不仅要使其能够感知人口统计线索，还要学习何时应调整其响应。
cs.CL / 55 / 2610.13583
Measuring Artifact Reliance in Machine-Generated Text Detectors
测量机器生成文本检测器中的伪影依赖性
Abstract
Detectors of machine-generated text score almost perfectly on scientific-literature benchmarks, implying we can reliably distinguish with these detectors human and machine-generated documents. We looked at ten scientific-document subsets from six popular resources, and on several of them that apparent ease is a side effect of how the documents were put together. A logistic regression using 27 superficial features reached 0.833-0.997 macro-F1, and on the three arXiv subsets it outperformed every detector we evaluate with F1-scores 0.951-0.997. The strongest cue to the detector is the PDF extraction artifact: human texts retain their original line breaks, whereas generated texts are extracted as a single block of prose. This cue, together with three other formatting features, tell the classes apart at 0.996 AUC. To distinguish benchmarks that contain such a signal from detectors that rely on one, we introduce format counterfactuals: just re-wrapping a machine-written abstract without changing a single word reverses 20-27% of a public detector's correct decisions. Normalizing formatting before training reduces the generalization gap from 0.32 to 0.22. On format-clean benchmarks, it also raises out-of-domain macro-F1 from 0.33 to 0.65 for detectors trained on artifact-containing subsets.
Chinese Translation
机器生成文本的检测器在科学文献基准测试中几乎得分完美，这意味着我们可以可靠地区分人类和机器生成的文档。我们分析了来自六个流行资源的十个科学文档子集，在其中一些子集中，这种明显的易区分性是文档构成方式的副作用。使用27个表面特征的逻辑回归模型达到了0.833-0.997的宏F1分数，在三个arXiv子集上，其表现超越了我们评估的每个检测器，F1分数为0.951-0.997。检测器的最强线索是PDF提取伪影：人类文本保留其原始的换行，而生成的文本则作为一整块散文被提取。这个线索，加上另外三个格式特征，使得类别之间的区分达到了0.996的AUC。为了区分包含这种信号的基准和依赖于此的检测器，我们引入了格式反事实：仅仅重新包装一个机器撰写的摘要而不改变任何单词，就会逆转20-27%的公共检测器的正确决策。在训练之前规范化格式将泛化差距从0.32降低到0.22。在格式清晰的基准上，它还将训练于包含伪影子集的检测器的域外宏F1从0.33提高到0.65。
cs.CL / 56 / 2610.13605
zip2zip++: Towards Inference-Time Token Compression with Minimal Quality Loss
zip2zip++：朝着推理时刻的令牌压缩与最小质量损失迈进
Abstract
Dynamic vocabularies built on-the-fly via Lempel-Ziv-Welch (LZW) compression let a language model represent the same text in fewer tokens, but prior work shows that this efficiency comes at a cost to reasoning and perplexity. We show that this quality gap is not a fundamental price of compression but can be substantially closed through targeted improvements to the compression layer and LZW codec, without modifying the underlying pretrained language model. Our analysis reveals four key issues: (1) LZW's compression algorithm creates an implicit conflict between the input embedding and the output embedding, a tension that a single shared embedding cannot resolve; (2) naively applying RoPE to LZW-compressed sequences ignores that merged tokens can span multiple tokens; (3) merging certain tokens (notably digits) disproportionately hurts downstream accuracy; and (4) dynamic vocabularies introduce an inherent segmentation ambiguity that confounds standard evaluation. We address these through decoupled input/output dynamic embedding layer, size-aware RoPE positioning, and a modified LZW codec supporting selective merges and multi-view segmentations. Across four models, our recipe improves perplexity on every corpus we evaluated as well as mathematical reasoning accuracy, closing roughly half the quality gap between zip2zip and the uncompressed baseline at a similar compression ratio on general text. This represents a significant step toward lossless adaptive-vocabulary compression for language models.
Chinese Translation
通过 Lempel-Ziv-Welch (LZW) 压缩动态生成的词汇表使得语言模型能够用更少的令牌表示相同的文本，但先前的研究表明，这种效率是以推理能力和困惑度为代价的。我们展示了这种质量差距并不是压缩的根本代价，而是可以通过对压缩层和 LZW 编解码器的有针对性改进来大幅缩小，而无需修改基础的预训练语言模型。我们的分析揭示了四个关键问题：（1）LZW 的压缩算法在输入嵌入和输出嵌入之间产生了隐含的冲突，这种紧张关系是单一共享嵌入无法解决的；（2）简单地将 RoPE 应用于 LZW 压缩序列忽视了合并的令牌可能跨越多个令牌；（3）合并某些令牌（尤其是数字）会不成比例地损害下游准确性；（4）动态词汇表引入了固有的分割模糊性，混淆了标准评估。我们通过解耦的输入/输出动态嵌入层、大小感知的 RoPE 定位以及支持选择性合并和多视图分割的修改 LZW 编解码器来解决这些问题。在四个模型中，我们的方法在每个评估的语料库上都改善了困惑度和数学推理准确性，在相似的压缩比下，缩小了 zip2zip 和未压缩基线之间大约一半的质量差距。这代表了朝着无损自适应词汇压缩的语言模型迈出的重要一步。
cs.CL / 57 / 2610.13607
BLiMPPI: Benchmark of Linguistic Minimal Pairs for Polarity Items
BLiMPPI：极性项语言最小对的基准测试
Abstract
Negative polarity items (NPIs), such as "anymore" and "ever", are grammatical only in select environments, predominantly in the context of negation. Because their grammaticality depends on both syntax and semantics, they have a rich, multilingual literature in theoretical linguistics, yet work on NPIs in language models is almost entirely confined to English. We introduce BLiMPPI, a multilingual evaluation suite of over 12,000 NPI minimal pairs in 33 languages, built by a pipeline that discovers NPIs in Universal Dependencies (UD) treebanks, labels their licensing environments, and perturbs these into minimal pairs. Evaluating 12 multilingual and 2 monolingual models alongside n-gram baselines, we find that multilingual models struggle most with lower-resource languages and with pairs that are not n-gram solvable. We further examine the effects of NPI typology, language family, licensing variability, and NPI-licenser distance. Data: https://github.com/arzuburcuguven/blimppi-data
Chinese Translation
负极性项（NPI），如“anymore”和“ever”，仅在特定环境中语法正确，主要是在否定的语境中。由于它们的语法性依赖于句法和语义，因此在理论语言学中有着丰富的多语言文献，然而关于语言模型中NPI的研究几乎完全局限于英语。我们介绍了BLiMPPI，这是一个包含33种语言中超过12,000个NPI最小对的多语言评估套件，构建过程通过一个管道发现通用依赖（Universal Dependencies, UD）树库中的NPI，标记其许可环境，并将其扰动为最小对。在评估12个多语言模型和2个单语言模型以及n-gram基线时，我们发现多语言模型在低资源语言和无法通过n-gram解决的对中表现最差。我们进一步考察了NPI类型、语言家族、许可变异性和NPI-许可者距离的影响。数据来源：https://github.com/arzuburcuguven/blimppi-data
cs.CL / 58 / 2610.13617
Training User Simulators with Latent Intrinsic Feedback
使用潜在内在反馈训练用户模拟器
Abstract
User simulators are gaining attention as environments to train LLM assistants that can better cope with the nuances of multi-turn conversations with real users. Prior work on user simulation has emphasized generating realistic user utterances, leaving unexplored their potential as a source of direct feedback on the assistant's responses. Such evaluation signals may not be explicitly communicated in language, but can instead reside in the latent reasoning of users. In this work, we introduce user simulators with intrinsic feedback, namely, models equipped with latent evaluation states: internal thoughts, reactions, and scalar ratings that evaluate the assistant from a first-person user perspective at each conversation turn. We train LLM-based simulators using interaction logs between real users and assistants and demonstrate that intrinsic feedback improves simulation realism and critiquing behaviors versus prompted simulator baselines, while also providing turn-level evaluations that existing fine-tuned simulators lack. Further, evaluations generated by our trained simulators correlate better with human judgments than those of prompted simulators of the same size. Lastly, our experiments reveal that simulators with intrinsic feedback provide effective reward signals for multi-turn RL training, matching setups that rely on external judge LLMs for rewards while cutting RL compute by more than half.
Chinese Translation
用户模拟器作为训练大型语言模型（LLM）助手的环境，正受到越来越多的关注，以便更好地应对与真实用户的多轮对话中的细微差别。以往的用户模拟研究强调生成真实的用户发言，而未探讨其作为对助手响应的直接反馈来源的潜力。这种评估信号可能并未通过语言明确传达，而是潜藏于用户的潜在推理中。在本研究中，我们引入了具有内在反馈的用户模拟器，即配备潜在评估状态的模型：从第一人称用户视角在每个对话轮次中评估助手的内部思维、反应和标量评分。我们使用真实用户与助手之间的交互日志训练基于LLM的模拟器，并证明内在反馈在模拟的真实感和批评行为方面优于提示模拟器基线，同时提供现有微调模拟器所缺乏的轮次级评估。此外，我们训练的模拟器生成的评估与人类判断的相关性优于同规模提示模拟器的评估。最后，我们的实验表明，具有内在反馈的模拟器为多轮强化学习（RL）训练提供了有效的奖励信号，其效果与依赖外部评判LLM进行奖励的设置相匹配，同时将RL计算量减少了一半以上。
cs.CL / 59 / 2610.13627
AMLworld-Compact: Importance-Weighted Downsampling for Cost-Effective LLM Evaluation and Error Diagnosis
AMLworld-Compact：用于成本效益高的LLM评估和错误诊断的重要性加权下采样
Abstract
Evaluating large language models (LLMs) on AMLworld's smallest test splits is costly because each of over a million targets requires a separate graph prompt. Sampling reduces this cost, but small balanced subsets change unweighted precision and F1. We introduce AMLworld-Compact, an evaluation set that retains every illicit test transaction and samples benign transactions from supervised-score strata. Its HT-Coreset protocol reduces evaluation targets by 271x and 611x on HI-Small and LI-Small. Horvitz-Thompson weights give unbiased full-test confusion-count estimates for fixed predictions; recall is exact, while precision and F1 are ratio estimates. In a held-out study using Graph Feature Preprocessor (GFP) features, one boosted-tree family defines strata for evaluating the other over 500 draws per configuration. At the released sizes, F1 RMSE is 0.00-2.58 percentage points versus 4.44-7.41 for uniform sampling with HT weights; conservative intervals remain broad. Under the evaluated graph prompts, seven open-weight LLMs have weighted precision near the predict-all-illicit baseline. In an outcome-stratified audit of 1,000 HI-Small few-shot outputs from one run per model, lexical checks pass for 84.5% of correct and 56.2% of incorrect binary verdicts, so these markers do not establish correctness. We release the evaluation sets, construction code, and audit tools at https://huggingface.co/datasets/typhoon-ai/AMLworldCompactEval and https://github.com/nat-nischw/AMLworld-Compact.
Chinese Translation
在AMLworld的最小测试分割上评估大型语言模型（LLMs）成本高昂，因为每个超过一百万的目标都需要单独的图形提示。采样可以降低这一成本，但小的平衡子集会改变未加权的精确度和F1值。我们引入了AMLworld-Compact，一个评估集，保留每个非法测试交易，并从监督评分层次中抽样良性交易。其HT-Coreset协议在HI-Small和LI-Small上分别将评估目标减少了271倍和611倍。Horvitz-Thompson权重为固定预测提供了无偏的全测试混淆计数估计；召回率是精确的，而精确度和F1值是比率估计。在使用图形特征预处理器（Graph Feature Preprocessor, GFP）特征的保留研究中，一个增强树族定义了用于评估其他模型的层次，每个配置进行超过500次抽样。在发布的规模下，F1的均方根误差（RMSE）为0.00-2.58个百分点，而均匀采样与HT权重的RMSE为4.44-7.41；保守区间仍然较宽。在评估的图形提示下，七个开放权重的LLMs的加权精确度接近于预测所有非法基线。在对每个模型进行一次运行的1,000个HI-Small少样本输出的结果分层审计中，词汇检查通过了84.5%的正确和56.2%的错误二元裁决，因此这些标记并不能确立正确性。我们在https://huggingface.co/datasets/typhoon-ai/AMLworldCompactEval和https://github.com/nat-nischw/AMLworld-Compact发布评估集、构建代码和审计工具。
cs.CL / 60 / 2610.13630
Multi-Stream Attention Residuals
多流注意力残差机制
Abstract
Attention Residuals (AttnRes) enables content-dependent retrieval over network depth, which can be viewed as single-stream softmax attention along the depth axis. Inspired by the success of multi-head attention along the sequence axis, we introduce Multi-Stream Attention Residuals (MSAR), an extension of AttnRes that generalizes this mechanism to multi-head depth attention, yielding substantial performance gains with minimal additional overhead. MSAR draws stream diversity from sequence context through multi-scale causal convolutions, allowing different heads to retrieve historical representations through distinct contextual views while preserving the integrity of the hidden representation. The retrieved streams are combined through lightweight adaptive mixing to generate the single-stream layer input. By retaining single-stream sources and expanding them on demand, MSAR supports multi-stream computation without requiring persistent multi-stream storage or communication. Across all 12 downstream benchmarks at the 18B and 28B MoE scales, MSAR consistently outperforms prior residual variants, including standard residual connections, manifold-constrained Hyper-Connections (mHC), and AttnRes. Scaling-law experiments yield an iso-loss compute speedup of $1.58\times$ over the standard residual baseline, compared with $1.35\times$ for AttnRes and $1.30\times$ for mHC. These results make MSAR a promising residual architecture for large-scale pre-training.
Chinese Translation
Attention Residuals（AttnRes）实现了基于内容的网络深度检索，可视为沿深度轴的单流softmax注意力。受序列轴多头注意力成功的启发，我们提出了多流注意力残差机制（Multi-Stream Attention Residuals，MSAR），这是AttnRes的扩展，将该机制推广到多头深度注意力，在带来显著性能提升的同时仅增加极少的额外开销。MSAR通过多尺度因果卷积从序列上下文中引入流多样性，使不同头能够通过不同的上下文视角检索历史表示，同时保持隐藏表示的完整性。检索到的多流通过轻量级自适应混合进行融合，生成单流层输入。通过保留单流源并按需扩展，MSAR支持多流计算，而无需持续的多流存储或通信。在18B和28B MoE规模的全部12个下游基准测试中，MSAR始终优于先前的残差变体，包括标准残差连接、流形约束的超连接（manifold-constrained Hyper-Connections，mHC）和AttnRes。扩展定律实验显示，MSAR在等损失条件下的计算加速比标准残差基线达到1.58倍，而AttnRes和mHC分别为1.35倍和1.30倍。这些结果使MSAR成为大规模预训练中极具潜力的残差架构。
cs.CL / 61 / 2610.13672
Inject a Turn, Not a Token: Learning Where to Redirect LLM Reasoning for Self-Correction
注入转折，而非令牌：学习如何重定向大型语言模型的推理以实现自我纠正
Abstract
Large language models struggle to reliably identify and correct errors in their own reasoning. Unlike prior training-based approaches that fine-tune an LLM to generate revisions or feedback, we learn where to intervene in failed rollouts and determine how to redirect them. To choose where to intervene, we train a 130K-parameter MLP head on a frozen encoder using only final-answer correctness. For redirection, we find that under a fixed generation budget, the tested inline injections fail to reliably redirect reasoning, whereas presenting the same instruction in a new user turn changes the model's course. The resulting correction framework, UPTURN (User-Prompt TURN injection at a learned position), requires no LLM fine-tuning: it retains a selected prefix, injects an instruction as a new user turn, and regenerates the suffix. Across four reasoners and six math benchmarks, UPTURN improves average accuracy in every setting over a 32-rollout baseline. It outperforms self-correction and reflection-token injection baselines under matched correction budgets, transfers across reasoners and to GPQA-Diamond without retraining, and complements GRPO and self-correction methods.
Chinese Translation
大型语言模型在可靠地识别和纠正自身推理中的错误方面存在困难。与以往基于训练的方法不同，后者通过微调大型语言模型生成修订或反馈，我们学习在失败的推理过程中何处进行干预，并确定如何重定向它们。为了选择干预的位置，我们在一个冻结的编码器上训练了一个130K参数的多层感知机（MLP）头，仅使用最终答案的正确性。对于重定向，我们发现，在固定的生成预算下，测试的内联注入无法可靠地重定向推理，而在新的用户转折中呈现相同的指令会改变模型的方向。由此产生的纠正框架UPTURN（用户提示转折注入于学习位置）无需对大型语言模型进行微调：它保留所选前缀，将指令作为新的用户转折注入，并重新生成后缀。在四个推理器和六个数学基准测试中，UPTURN在每种设置下都提高了相较于32次推理基线的平均准确性。在匹配的纠正预算下，它优于自我纠正和反射令牌注入基线，能够在推理器之间转移，并在不重新训练的情况下转移到GPQA-Diamond，同时补充GRPO和自我纠正方法。
cs.CL / 62 / 2610.13678
Retractions Revisited: A Replication, Extension, and Mechanistic Evaluation of Linguistic Obfuscation Effects in Fraudulent Science
撤回重审：对欺诈科学中语言模糊效应的复制、扩展和机制评估
Abstract
Prior work suggests fraudulent scientists write their articles with higher rates of linguistic obfuscation than genuine scientists, with a possible mechanism suggesting obfuscation makes research more costly for a reader to analyze. This result has not undergone additional empirical scrutiny in the open science era, nor have other psychological mechanisms been evaluated. The current replication and extension addressed this opportunity by comparing 800 fraudulent and 800 genuine control papers (2013-2025). The results supported the linguistic obfuscation hypothesis, suggesting fraudulent papers were more obfuscated than genuine ones. A parallel mediation model evaluated three competing mechanisms, finding that reality monitoring, as measured through decreased rates of adjectives, best explained the link between fraudulent science and linguistic obfuscation. Cross-validated detection accuracy using obfuscation features (60.7%) and an elastic net of over 100 language dimensions (71.6%) were significantly greater than chance. Implications for scientific fraud and psychology of language research are discussed.
Chinese Translation
先前的研究表明，欺诈科学家撰写的文章在语言模糊程度上高于真实科学家，可能的机制是模糊化使得读者分析研究的成本更高。在开放科学时代，这一结果尚未经过进一步的实证检验，也没有评估其他心理机制。本研究通过比较800篇欺诈论文和800篇真实对照论文（2013-2025）来填补这一空白，进行了复制和扩展。结果支持了语言模糊假设，表明欺诈论文的模糊程度高于真实论文。一个平行中介模型评估了三种竞争机制，发现通过形容词使用率降低来测量的现实监控，最佳地解释了欺诈科学与语言模糊之间的联系。使用模糊特征的交叉验证检测准确率为60.7%，而超过100个语言维度的弹性网络准确率为71.6%，均显著高于随机水平。讨论了对科学欺诈和语言心理学研究的影响。
cs.CL / 63 / 2610.13695
Verifiable Retrieval for Legal Reasoning through Structure-driven Knowledge Graph
通过结构驱动知识图谱实现法律推理的可验证检索
Abstract
Any automation within reliability-critical contexts such as medicine or law has to be transparent and verifiable. Despite their advantages, current LLM-based systems do not meet these requirements due to their inherent black-box nature. Retrieval Augmented Generation (RAG) attempts to ground the generative process in the relevant interpretable documents, thereby reducing the risk of AI 'hallucinations' and potentially enabling verification of the initial generation context. However, as the number and size of potential sources increases, naive retrieval approaches tend to perform worse, ignoring critical context elements. In this paper we aim to tackle this problem in the context of German Legal Code. Using structure-driven parsing to create a Knowledge Graph (KG) for the source document navigation and retrieval, we compare several KG-informed chunking and retrieval strategies and evaluate their effect on the accuracy of retrieved citations.
Chinese Translation
在医学或法律等对可靠性要求严格的领域，任何自动化过程都必须具备透明性和可验证性。尽管当前基于大型语言模型（LLM）的系统具有一定优势，但由于其固有的黑箱特性，仍未能满足这些要求。检索增强生成（RAG）试图将生成过程与相关的可解释文档相结合，从而降低人工智能“幻觉”的风险，并可能实现对初始生成上下文的验证。然而，随着潜在来源的数量和规模增加，简单的检索方法往往表现不佳，忽视了关键的上下文元素。本文旨在在德国法律法规的背景下解决这一问题。我们使用结构驱动解析创建知识图谱（KG）以实现源文档的导航和检索，比较几种基于知识图谱的分块和检索策略，并评估它们对检索引用准确性的影响。
cs.CL / 64 / 2610.13722
Enhancing Legal Reasoning in Pre-trained Language Models via Knowledge Graph-Guided Multi-Task Pre-training
通过知识图谱引导的多任务预训练增强预训练语言模型中的法律推理
Abstract
Legal language understanding requires reasoning over semi-structured regulatory documents, yet standard pre-training objectives do not explicitly target this structural capability. Because current legal language models rely solely on masked language modeling, they treat statutory text as flat token sequences and ignore the underlying hierarchical constraints. To bridge this gap, we present a multi-task pre-training approach for German traffic law (StVO). Our approach combines three complementary objectives: Next Sentence Prediction (NSP), Knowledge Graph (KG) Guided Clustering (CLU), and Knowledge Graph (KG) Aware Paragraph Classification (CLS). This approach explicitly encodes structured legal knowledge directly into the model weights. This encoding enables zero-shot legal reasoning at inference time and removes the need for discrete graph traversal. We evaluate our approach across nine Transformer-based encoders and 324 experimental runs. On synthetic StVO question answering, our models demonstrate consistent relative improvements between 10.3% and 155.0% over baselines. Furthermore, relative to the established LexGLUE baselines (Chalkidis et al., 2022), our models improve over prior results on four of seven legal benchmarks. It also achieves consistent zero-shot transfer gains between 3.4% and 15.8% on Austrian driving regulations. These experimental results indicate that our training approach generalizes successfully beyond the initial training legal domain.
Chinese Translation
法律语言理解需要对半结构化的法规文件进行推理，然而标准的预训练目标并未明确针对这种结构能力。由于当前的法律语言模型仅依赖于掩码语言建模，它们将法条文本视为平坦的标记序列，忽视了潜在的层次约束。为了解决这一问题，我们提出了一种针对德国交通法（StVO）的多任务预训练方法。我们的方法结合了三个互补的目标：下一句预测（Next Sentence Prediction, NSP）、知识图谱（Knowledge Graph, KG）引导的聚类（Clustering, CLU）和知识图谱（KG）感知的段落分类（Paragraph Classification, CLS）。这种方法将结构化的法律知识显式编码到模型权重中。这种编码使得在推理时能够实现零样本法律推理，并消除了离散图遍历的需要。我们在九种基于Transformer的编码器和324次实验中评估了我们的方法。在合成的StVO问答任务中，我们的模型在基准上显示出10.3%到155.0%的一致相对提升。此外，相较于已建立的LexGLUE基准（Chalkidis等，2022），我们的模型在七个法律基准中的四个上超越了先前的结果。在奥地利驾驶法规上，它还实现了3.4%到15.8%的一致零样本迁移增益。这些实验结果表明，我们的训练方法在初始训练法律领域之外成功地实现了泛化。
cs.CL / 65 / 2610.13794
Not Just an Evaluation: A Case Study of Discourse Particle Sensitivity in Language Models
不仅仅是评估：语言模型中话语粒子敏感性的案例研究
Abstract
Discourse particles subtly shape the flow of discourse by reflecting conversational context and interlocutor attitudes, often leaving the truth-conditional meaning untouched. They are also often polyfunctional, with distinct senses which partially overlap with other particles; for example, the exclusive use of English "just" and "only" ("John only/just drinks water"). Despite the frequency of these words in natural language, they have gone largely unstudied due to the subtle nature of their effects. We investigate LLM sensitivity to "just" and "only" in natural conversations using behavioral and mechanistic analyses. First, we find that LLMs and humans similarly prefer the presence of these particles in context, and both are more sensitive to "just" compared to "only". Second, when sampling from LLMs with and without these particles, we find LLMs consistently rate generations as more likely, given their source context, demonstrating their particle sensitivity. In both experiments, instruction tuning increases sensitivity. Finally, we localize particle-sensitive units within LLMs and find the overlap between "just" and "only" is asymmetric: "only" units transfer selectively to exclusive "just", while all "just" units transfer to "only" equally well regardless of sense, suggesting LLM lexical-level particle representations are partially organized by sense. We release our code, sense-labeled dialogue stimuli, and human preference annotations at https://github.com/JiahuiKChen/Not_Just_an_Evaluation.
Chinese Translation
话语粒子通过反映对话背景和交谈者态度，微妙地塑造了话语的流动，通常不影响其真值条件意义。它们通常也是多功能的，具有与其他粒子部分重叠的不同意义；例如，英语中“just”和“only”的独占使用（“John only/just drinks water”）。尽管这些词在自然语言中频繁出现，但由于其效果的微妙性，它们在研究中往往被忽视。我们通过行为和机制分析，研究大型语言模型（LLM）在自然对话中对“just”和“only”的敏感性。首先，我们发现LLM和人类在上下文中同样偏好这些粒子的存在，并且两者对“just”的敏感性高于“only”。其次，当从包含和不包含这些粒子的LLM中进行采样时，我们发现LLM在给定源上下文的情况下，一致地将生成内容评估为更可能，显示出它们对粒子的敏感性。在两个实验中，指令调优增加了敏感性。最后，我们在LLM中定位了粒子敏感单元，并发现“just”和“only”之间的重叠是不对称的：“only”单元选择性地转移到独占的“just”，而所有“just”单元无论意义如何都同样有效地转移到“only”，这表明LLM的词汇级粒子表示在某种程度上是按意义组织的。我们在https://github.com/JiahuiKChen/Not_Just_an_Evaluation发布了我们的代码、带有意义标签的对话刺激和人类偏好注释。
cs.CL / 66 / 2610.13797
REFLECT: Role-Aware Detection of Sentiment, Emotion, Politeness, and Formality
REFLECT：角色感知的情感、情绪、礼貌和正式性的检测
Abstract
Detecting communicative meaning in dialogue requires modeling not only lexical content but also the participant-role structures through which sentiment, emotion, politeness, and formality (SEPF) are expressed. Existing approaches rely on contextual language representations while leaving semantic roles implicit. We introduce a role-aware transformer architecture that maps the four SEPF dimensions to theory-grounded Semantic Role Labeling (SRL) configurations, incorporating these configurations directly into the transformer's self-attention mechanism. Role-aware attention models non-directional internal states (sentiment and emotion), while directional role-aware attention captures initiator-to-recipient relations for external cues (politeness and formality). The resulting role-aware representations are fused with contextual embeddings through a learned gate that adaptively weights how much role information each input contributes, rather than appending semantic roles as an auxiliary feature vector. Across four benchmark datasets, the framework yields statistically reliable gains over strong BERT-based baselines for emotion, politeness, and formality, with gains concentrated in theoretically predicted cases such as low-frequency, complex emotions and highly polite expressions; sentiment improvements are marginal and non-significant, suggesting that sentiment is conveyed more through lexical choice than participant-role structure. Additional evaluation on real-world data demonstrates that the framework improves both predictive performance and model traceability. These results demonstrate that linguistically grounded modifications to transformer self-attention improve the detection of communicative meaning, generalize beyond benchmark datasets to real-world mental-health data, and yield traceable, role-grounded explanations that extend beyond predictive performance to interpretability.
Chinese Translation
在对话中检测交际意义不仅需要建模词汇内容，还需要建模参与者角色结构，通过这些结构表达情感、情绪、礼貌和正式性（SEPF）。现有的方法依赖于上下文语言表示，而将语义角色隐含化。我们提出了一种角色感知的变换器架构，将四个SEPF维度映射到理论基础的语义角色标注（SRL）配置中，并将这些配置直接融入变换器的自注意力机制。角色感知注意力模型对非定向的内部状态（情感和情绪）进行建模，而定向的角色感知注意力则捕捉发起者与接收者之间的关系，以获取外部线索（礼貌和正式性）。生成的角色感知表示通过一个学习的门控机制与上下文嵌入融合，该机制自适应地加权每个输入所贡献的角色信息，而不是将语义角色附加为辅助特征向量。在四个基准数据集上，该框架在情绪、礼貌和正式性方面相较于强大的基于BERT的基线模型取得了统计上显著的提升，且这些提升集中在理论预测的情况，如低频复杂情绪和高度礼貌的表达；情感的改善则是边际的且不显著，这表明情感更多是通过词汇选择而非参与者角色结构来传达的。对现实世界数据的额外评估表明，该框架在预测性能和模型可追溯性方面都有所改善。这些结果表明，基于语言学的变换器自注意力修改提高了交际意义的检测，超越基准数据集推广到现实世界的心理健康数据，并提供可追溯的、基于角色的解释，这些解释超越了预测性能，增强了可解释性。
cs.CL / 67 / 2610.13807
Error Rate is Not Enough: Costs of Augmentation in Low-Resource ASR
错误率不足：低资源自动语音识别中的增强成本
Abstract
Data augmentation is standard practice for low-resource automatic speech recognition (ASR), but its benefits are almost always reported as aggregate error rates, leaving open why and when a method helps. On four low-resource languages and English at five data sizes, we cross two augmentations that modify existing audio (speed perturbation and SpecAugment) with two that modify text transcriptions and synthesize audio from it. We analyze errors at the phoneme level and find that they help for complementary reasons. SpecAugment has the most reliable aggregate improvements, but increases tone and palatalization errors, whereas text-based methods introduce consonant insertions; in combination, each reduces the errors the other introduces. This complementarity is predictable. Tone and palatalization are each signaled by a single acoustic property and occur least frequently in training (lowest training exposure), and text-based methods that flatten word frequency raise their frequency by 37--38%, reversing those errors while further improving aggregate error. The improvement from combining the methods exceeds the sum of their individual improvements; the superadditivity vanishes by 100 hours, limiting the complementarity to the low-resource regime. Where specific error types matter, we recommend selecting augmentation by training exposure and acoustic properties, both measurable before any training.
Chinese Translation
数据增强是低资源自动语音识别（ASR）的标准实践，但其益处几乎总是以总错误率的形式报告，这使得我们无法了解何时以及为何某种方法有效。在四种低资源语言和五种数据量的英语中，我们将两种修改现有音频的增强方法（速度扰动和SpecAugment）与两种修改文本转录并从中合成音频的方法进行交叉分析。我们在音素级别分析错误，发现它们在互补的原因上有所帮助。SpecAugment具有最可靠的整体改进，但增加了声调和腭化错误，而基于文本的方法则引入了辅音插入；两者结合时，各自减少了对方引入的错误。这种互补性是可以预测的。声调和腭化各自由单一的声学特性信号表示，并且在训练中出现的频率最低（训练暴露最少），而基于文本的方法通过平坦化词频将其频率提高了37-38%，从而逆转了这些错误，同时进一步改善了整体错误率。结合这些方法所带来的改进超过了它们各自改进的总和；这种超加性在100小时后消失，限制了互补性仅在低资源环境中有效。在特定错误类型重要的情况下，我们建议根据训练暴露和声学特性选择增强方法，这两者在任何训练之前都是可测量的。
cs.CL / 68 / 2610.13821
Can Language Models Learn When to Abstain? Policy Learning and Calibrated Answer Selection
语言模型能否学习何时放弃？策略学习与校准答案选择
Abstract
A useful language model should know when to answer and when to say "I don't know." Can we improve that decision by training the model itself or by training a separate component to check its answers? We study both approaches on Qwen3-14B for factual questions, using a score that rewards correct answers, penalizes mistakes, and allows abstention. Across three training seeds and two held-out confidence thresholds, a learned answer selector improves threshold-averaged utility over the original model's calibrated confidence filter. On TriviaQA, mean coverage increases from 20.2% to 30.9%, while the pooled error rate among scored answers decreases from 23.0% to 14.5% under exact-match grading. The tested reinforcement-learning changes do not establish consistent gains. On NQ-Open, the selector improves threshold-averaged utility over the filter but still performs worse than always abstaining. A partial human-reviewed audit preserves the TriviaQA finding. Our results support learning to select answers in this setting while showing that reliable abstention across datasets remains unresolved.
Chinese Translation
一个有用的语言模型应该知道何时回答以及何时说“我不知道”。我们能否通过训练模型本身或训练一个独立组件来检查其答案来改善这一决策？我们在 Qwen3-14B 上研究了这两种方法，针对事实性问题，使用一种奖励正确答案、惩罚错误并允许放弃的评分机制。在三个训练种子和两个保留的置信度阈值下，学习的答案选择器在阈值平均效用上优于原始模型的校准置信度过滤器。在 TriviaQA 上，平均覆盖率从 20.2% 增加到 30.9%，而在精确匹配评分下，评分答案的合并错误率从 23.0% 降低到 14.5%。测试的强化学习变化未能建立一致的增益。在 NQ-Open 上，选择器在阈值平均效用上优于过滤器，但仍然表现不如始终放弃。部分人工审核的审计保留了 TriviaQA 的发现。我们的结果支持在此环境中学习选择答案，同时表明在不同数据集上可靠的放弃仍未得到解决。
cs.CL / 69 / 2610.13895
Task Context Shapes Component Roles in Language Models
任务上下文塑造语言模型中组件的角色
Abstract
Large language models reuse the same attention heads across tasks, but identifying these shared heads does not explain what they do differently in each task. A natural source of inspiration is task-dependent interactions among shared brain regions, which motivate a network view of language-model computation. Here, we compare fixed model components across tasks using identical graph facts, statements specifying which directed connections are present. We measure how interventions affect attention heads and the activation state in later layers (influence), as well as the answer (importance). Task changes redistribute the influence in five open-weight models spanning four families and 1.5-14 billion parameters. In Gemma-2-9B and Qwen3-14B, an activation edit changes an early internal activation vector and switches which fact the model reads while the written instruction stays fixed. The same heads then respond to different graph facts and influence later components differently. For each prompt, we construct two linear response maps, one before the edit and one after it. Each map predicts a later head's response from a measured change in an earlier head. For a separate factual intervention, the map for the corresponding condition reduces mean squared prediction error by 53.6% in Gemma and 5.0% in Qwen compared with the average of the two maps. These findings show that task context changes both which graph facts the shared components respond to and how their outputs affect later computation. A network account may help explain how models use shared parameters to perform different computations. Measurements of task-dependent influence could also guide interventions intended to alter one behavior while preserving others.
Chinese Translation
大型语言模型在不同任务中重用相同的注意力头，但识别这些共享的头并不能解释它们在每个任务中所执行的不同功能。一个自然的灵感来源是共享脑区之间的任务依赖性交互，这激励了对语言模型计算的网络视角。在这里，我们使用相同的图事实比较跨任务的固定模型组件，这些事实是指定哪些定向连接存在的声明。我们测量干预如何影响注意力头和后续层的激活状态（影响），以及答案（重要性）。任务变化重新分配了五个开放权重模型中的影响，这些模型跨越四个家族，参数量从15亿到140亿不等。在Gemma-2-9B和Qwen3-14B中，激活编辑改变了早期内部激活向量，并切换了模型读取的事实，而书写的指令保持不变。相同的头随后对不同的图事实作出反应，并以不同的方式影响后续组件。对于每个提示，我们构建了两个线性响应图，一个是在编辑之前，另一个是在编辑之后。每个图根据早期头的测量变化预测后续头的响应。对于一个单独的事实干预，相应条件的图在Gemma中将均方预测误差减少了53.6%，在Qwen中减少了5.0%，与两个图的平均值相比。这些发现表明，任务上下文改变了共享组件对哪些图事实的响应以及它们的输出如何影响后续计算。网络模型可能有助于解释模型如何利用共享参数执行不同的计算。对任务依赖性影响的测量也可以指导旨在改变一种行为而保留其他行为的干预。
cs.CL / 70 / 2610.13896
Structural Automata for Neural Semantic Parsing
用于神经语义解析的结构自动机
Abstract
Neural semantic parsers generate meaning representations token by token, yet well-formedness depends on references, scope, and graph relations that unfold during generation. Constrained decoding prevents locally invalid tokens but does not guarantee that a valid prefix can be completed within a fixed output-length limit. We investigate structural automata at three stages of DRS parsing (decoding, post hoc processing, and training) and additionally evaluate decoding on AMR. For decoding, we introduce accepting-path decoding (APD), which maintains a verified completion within the remaining length limit and follows the model's preferred token whenever that token can also be completed. APD guarantees an accepted output without retracting tokens. Under tight length limits, it improves Smatch++ over local masking by 5.1 and 2.4 points on DRS and by 6.4 and 3.2 points on AMR, comparable to rollback under the same limit, which retracts tokens. For post hoc processing, invalid outputs concentrate on the same inputs: eight unconstrained samples yield no valid output for 10.2% of inputs with limited training data, compared with 0.4% with full data, where filtering comes within 0.2 Smatch++ points of APD. For training, penalizing automaton-rejected tokens improves development Smatch++ by 3.6-11.0 points after 10% of training, but this advantage diminishes as cross-entropy alone reduces the probability assigned to rejected tokens from 14% to below 2% within the first sixth of training with limited data.
Chinese Translation
神经语义解析器逐个生成意义表示，但良构性依赖于在生成过程中展开的引用、范围和图关系。受限解码防止局部无效的标记，但并不保证在固定输出长度限制内可以完成有效前缀。我们在DRS解析的三个阶段（解码、后处理和训练）中研究结构自动机，并额外评估AMR上的解码。对于解码，我们引入了接受路径解码（Accepting-Path Decoding, APD），该方法在剩余长度限制内保持经过验证的完成，并在该标记也可以完成时遵循模型的首选标记。APD保证在不撤回标记的情况下输出被接受的结果。在严格的长度限制下，它在DRS上比局部掩蔽提高了5.1和2.4点，在AMR上提高了6.4和3.2点，且与在相同限制下撤回标记的回滚方法相当。对于后处理，无效输出集中在相同输入上：在有限训练数据下，八个不受限样本对10.2%的输入没有产生有效输出，而在完整数据下这一比例为0.4%，过滤结果与APD相差不到0.2 Smatch++点。对于训练，惩罚自动机拒绝的标记在训练10%后提高了开发Smatch++ 3.6-11.0点，但随着交叉熵单独减少被拒绝标记的概率从14%降至2%以下，这一优势逐渐减弱，尤其是在有限数据的前六分之一训练中。
cs.CL / 71 / 2610.13910
When VLMs Destroy What They Know: Spatial Representations Fall Below Lexical Priors
当视觉语言模型摧毁它们所知的：空间表征低于词汇先验
Abstract
Vision-Language Models (VLMs) increasingly analyze and reason about images, yet it remains unclear whether their internal representations faithfully encode the structured relations required for such reasoning. We introduce EICS (Evaluation-Invariant Concept Stability), a metric for testing whether representations consistently encode concepts under controlled one-concept interventions, and CASP-VL (Causally controlled Spatial Perturbations for Vision-Language models), a benchmark of causally controlled spatial perturbations derived from Visual Genome scene graphs. Across multiple VLM architectures, we find that spatial relations are represented substantially less faithfully than lexical information, even when the same models score well above chance on compositional benchmarks. Further analysis shows that the relation is identifiable at the predicate's own tokens but is lost when the sentence is aggregated into a global representation, whether through the (EOS) token or by mean pooling. These findings reveal a systematic failure invisible to output-based evaluation: VLMs do not preserve, in the representation used downstream, relational information that is present in their own input.
Chinese Translation
视觉语言模型（VLMs）越来越多地分析和推理图像，但仍不清楚它们的内部表征是否忠实地编码了进行此类推理所需的结构化关系。我们引入了EICS（评估不变概念稳定性），这是一个用于测试表征在受控的单一概念干预下是否一致编码概念的指标，以及CASP-VL（因果控制的视觉语言模型空间扰动），这是一个基于Visual Genome场景图的因果控制空间扰动基准。在多个VLM架构中，我们发现空间关系的表征明显低于词汇信息，即使在同一模型在组合基准上得分远高于随机水平的情况下。进一步分析表明，这种关系在谓词自身的标记中是可识别的，但在句子聚合为全局表征时会丢失，无论是通过（EOS）标记还是均值池化。这些发现揭示了一种在基于输出的评估中不可见的系统性失败：VLMs在下游使用的表征中并未保留其自身输入中存在的关系信息。
cs.CL / 72 / 2610.13912
RSI-Anything: Unlocking Recursive Self-Improvement with World Models
RSI-Anything：利用世界模型解锁递归自我改进
Abstract
Recursive self-improvement (RSI) aims to transform interaction experience into persistent improvements, yet external feedback is costly and world-model predictions may become unreliable as policies evolve. We introduce RSI-Anything, a domain-general framework built on JEPA-Anything that uses predictive world models to guide action and learning-data selection. Observed feedback drives policy or workflow improvement and conditional world-model adaptation, while component and compatibility checks govern the retention of updated policy-model pairs. Across diverse computational and scientific domains, RSI-Anything demonstrates substantial gains. In code repair, with the coder fixed, success increases from 26.3% with a static verifier to 29.2% through predictive selection and 31.2% after world-model adaptation, at comparable test usage. Graphical user interface (GUI) evaluations showed higher task completion, while average actions decrease by 39.4%. Across five successive policy versions, laboratory-planning scores more than double from 21.80% to 46.01%. In reaction optimization, RSI-Anything achieves 96.15% mean best observed yield versus 82.16% for Bayesian optimization under the same 40-experiment budget. An end-to-end immune-science application further connects workflow evolution and predictive modeling with therapeutic hypothesis generation. These results demonstrate how predictive world models and retained system updates enable recursive improvement across diverse environments.
Chinese Translation
递归自我改进（RSI）旨在将互动经验转化为持久改进，然而外部反馈成本高昂，随着策略的演变，世界模型的预测可能变得不可靠。我们提出了RSI-Anything，这是一个基于JEPA-Anything构建的领域通用框架，利用预测性世界模型来指导行动和学习数据的选择。观察到的反馈推动策略或工作流程的改进以及条件世界模型的适应，而组件和兼容性检查则控制更新的策略-模型对的保留。在多种计算和科学领域中，RSI-Anything展示了显著的提升。在代码修复中，当编码器固定时，成功率从使用静态验证器的26.3%提高到通过预测选择的29.2%，并在世界模型适应后达到31.2%，测试使用量相当。图形用户界面（GUI）评估显示任务完成率更高，而平均操作次数减少了39.4%。在五个连续的策略版本中，实验室规划得分从21.80%翻倍至46.01%。在反应优化中，RSI-Anything在相同的40次实验预算下实现了96.15%的平均最佳观察产量，而贝叶斯优化为82.16%。一个端到端的免疫科学应用进一步将工作流程演变与预测建模和治疗假设生成相连接。这些结果展示了预测性世界模型和保留的系统更新如何在多样化环境中实现递归改进。