arXiv Daily Digest

214

Papers

Adjacency-Based Spectral Proxy Control of Mobile Communication Agents

基于邻接矩阵的移动通信代理谱代理控制

del Castillo, Mariana, Larroca, Federico

Abstract

We consider a heterogeneous mobile-agent network composed of uncontrolled task agents and controllable communication agents. The objective is to reposition communication agents online as task agents move. Since throughput-based objectives are generally unsuitable for real-time control, spectral graph metrics such as algebraic connectivity are commonly adopted as surrogate objectives. However, controlling algebraic connectivity relies on the eigenvector corresponding to the second-smallest eigenvalue of a graph's Laplacian matrix (i.e., the Fiedler vector), whose distributed estimation requires an unbounded number of communication rounds to converge. In this work, we identify a structural decomposition of this Fiedler-gradient controller into a local interaction rule and a graph embedding component, suggesting the use of alternative embeddings that are easier to estimate distributively than the Fiedler vector. As a particular instance, we propose A-Fiedler, which replaces the Fiedler embedding with the dominant eigenvector of the adjacency matrix, commonly used as a graph embedding of nodes into a latent geometry. This representation is more naturally suited for distributed implementation under local communication constraints. We evaluate A-Fiedler against the classical Fiedler-gradient controller. Results show comparable network performance in the absence of communication constraints and improved robustness under distributed estimation. For instance, under the same number of communication rounds, the Fielder-gradient may even converge to disconnected configurations whereas our proposition maintains performance. We believe our contribution provides a simpler path toward distributed network control.

Chinese Translation

我们考虑一个由不受控任务代理和可控通信代理组成的异构移动代理网络。我们的目标是在任务代理移动时在线重新定位通信代理。由于基于吞吐量的目标通常不适合实时控制，因此谱图度量（如代数连通性）通常被采用作为替代目标。然而，控制代数连通性依赖于图的拉普拉斯矩阵的第二小特征值对应的特征向量（即 Fiedler 向量），其分布式估计需要无限数量的通信轮次才能收敛。在本研究中，我们识别出这个 Fiedler 梯度控制器的结构分解为局部交互规则和图嵌入组件，建议使用比 Fiedler 向量更容易进行分布式估计的替代嵌入。作为一个具体实例，我们提出了 A-Fiedler，它用邻接矩阵的主特征向量替代了 Fiedler 嵌入，后者通常被用作节点嵌入到潜在几何中的图嵌入。这种表示在局部通信约束下更自然地适合于分布式实现。我们将 A-Fiedler 与经典的 Fiedler 梯度控制器进行了评估。结果显示，在没有通信约束的情况下，网络性能相当，而在分布式估计下则表现出更好的鲁棒性。例如，在相同的通信轮次下，Fiedler 梯度可能会收敛到不连通的配置，而我们的提议则保持了性能。我们相信我们的贡献为分布式网络控制提供了一条更简单的路径。

View on arXiv Download PDF AI Translation

cs.RO / 2 / 2608.13678

hint$^2$: Hierarchical World Models for Inference-Time Temporal Logic Guidance

hint$^2$: 分层世界模型用于推理时的时序逻辑指导

Zoellner, Moritz, Manganaris, Anastasios, Qureshi, Ahmed H., Paleja, Rohan

Abstract

A central goal of robot learning is to enable robots to execute rich instructions specified at runtime. Large-scale language-conditioned policies have made substantial progress toward this goal, yet still struggle with temporal structure and safety constraints. Linear Temporal Logic (LTL) provides a powerful language to express complex, non-Markovian instructions. However, guiding learned manipulation policies toward LTL satisfaction remains challenging because modern policies generate short-horizon action chunks and replan in closed loop, while almost all LTL specifications are evaluated over long-horizon trajectories. In this paper, we introduce hint$^2$, a method for guiding short-horizon policies toward satisfying complex LTL specifications at inference time using hierarchical world models. Our key idea is to derive two separate guidance objectives using each world model's abstraction level. A high-level model predicts future action-induced transitions in task-relevant atomic propositions to guide progress through the LTL automaton, while a low-level dynamics model predicts immediate state evolution for accurate local safety guidance. Our results show that hint$^2$ overcomes the limitations of current LTL-guided diffusion methods, outperforms existing inference-time steering methods in CALVIN, and successfully completes instructions with complex liveness and safety constraints more elegantly than language-conditioned alternatives. Finally, we demonstrate that hint$^2$ can handle complex instructions on a real UR5e manipulator.

Chinese Translation

机器人学习的一个核心目标是使机器人能够在运行时执行丰富的指令。大规模语言条件策略在实现这一目标方面取得了显著进展，但在时序结构和安全约束方面仍然面临挑战。线性时序逻辑（LTL）提供了一种强大的语言来表达复杂的非马尔可夫指令。然而，引导学习到的操作策略满足LTL要求仍然具有挑战性，因为现代策略生成短期动作片段并在闭环中重新规划，而几乎所有LTL规范都是在长期轨迹上进行评估的。本文介绍了hint$^2$，一种利用分层世界模型在推理时引导短期策略满足复杂LTL规范的方法。我们的关键思想是利用每个世界模型的抽象层次推导出两个独立的指导目标。高层模型预测任务相关原子命题中的未来动作引起的转变，以引导LTL自动机的进展，而低层动态模型则预测即时状态演变，以实现准确的局部安全指导。我们的结果表明，hint$^2$克服了当前LTL引导扩散方法的局限性，在CALVIN中优于现有的推理时引导方法，并比语言条件替代方案更优雅地完成具有复杂活性和安全约束的指令。最后，我们证明了hint$^2$能够在真实的UR5e操纵器上处理复杂指令。

View on arXiv Download PDF AI Translation

cs.RO / 3 / 2608.13723

Graph-MambaNav: Spatial-Temporal Graph Mamba Leveraging Object-Relation Knowledge for Object-Goal Navigation

Graph-MambaNav：利用对象关系知识的时空图Mamba进行目标导向导航

Sun, Leyuan, Chen, Genxin, Ye, Linwei, Zhang, Yan, Kan, Xi, Sun, Yanfei

Abstract

Object-goal navigation requires an agent to reason over object relationships and prioritize target-relevant objects for efficient decision making in unseen environments. While existing graph-based methods incorporate target-awareness at the feature or attention level, they remain permutation-invariant and lack an explicit mechanism to control information propagation order, limiting their ability to model target-dependent importance and long-range dependencies. In contrast, Graph-Mamba highlights that node prioritization through sequence ordering is critical for effective global reasoning. In this work, we investigate the node prioritization mechanism in Graph-Mamba and study its role in object navigation. We propose Graph-MambaNav, a target-aware spatial-temporal graph encoding framework that introduces a heuristic ordering over objects based on their relevance to the target, allowing more informative objects to be processed later to aggregate richer context. Both node ordering and edge weights are initialized from LLM-derived commonsense object relationships, providing a unified prior for structured reasoning. A spatial module integrates local message passing with global GraphMamba-based selective scanning, while a temporal module applies Mamba-based sequence modeling over object-wise temporal orders, allowing selective aggregation of historical context for long-range temporal reasoning. Experiments on AI2-THOR and RoboTHOR demonstrate improved navigation performance with generalization, and additional real-world robot deployment further validates the effectiveness of our proposed approach.

Chinese Translation

目标导向导航要求智能体推理对象之间的关系，并优先考虑与目标相关的对象，以便在未知环境中进行高效决策。尽管现有的基于图的方法在特征或注意力层面上融入了目标意识，但它们仍然是排列不变的，并且缺乏控制信息传播顺序的显式机制，限制了它们建模目标依赖重要性和长距离依赖的能力。相比之下，Graph-Mamba强调通过序列排序进行节点优先级设置对于有效的全局推理至关重要。在本研究中，我们探讨了Graph-Mamba中的节点优先级机制，并研究其在对象导航中的作用。我们提出了Graph-MambaNav，这是一种目标意识的时空图编码框架，基于对象与目标的相关性引入了一种启发式排序，使得更具信息量的对象可以在后期处理，从而聚合更丰富的上下文。节点排序和边权重均基于大规模语言模型（LLM）推导的常识对象关系进行初始化，为结构化推理提供了统一的先验。空间模块将局部消息传递与基于GraphMamba的全局选择性扫描相结合，而时间模块则在对象时间顺序上应用基于Mamba的序列建模，允许对历史上下文进行选择性聚合，以实现长距离时间推理。在AI2-THOR和RoboTHOR上的实验表明，导航性能得到了改善并具有良好的泛化能力，额外的现实世界机器人部署进一步验证了我们提出的方法的有效性。

View on arXiv Download PDF AI Translation

cs.RO / 4 / 2608.13878

Knowledge-Data-Dual-Driven Reinforcement Learning for Autonomous Vehicle Control in Mixed Traffic

知识-数据双驱动强化学习在混合交通中的自主车辆控制

Fang, Jie, Zheng, Wei, Xu, Mengyun, Kim, Eui-Jin

Abstract

In mixed traffic, decision-making for autonomous vehicles (AVs) confronts three interrelated challenges. First, physics-based priors incorporated into reinforcement learning (RL) models fail to capture latent interactive vehicle intentions and diverse driver behaviors, limiting the proactive reasoning capabilities. Second, abrupt maneuvers by surrounding vehicles cause non-stationarity, leaving long-tail safety events under-explored. Third, hybrid action spaces destabilize unified RL training due to the different temporal scales of continuous car-following and discrete lane-changing maneuvers. To address these issues, we propose Knowledge-Data Dual-driven Reinforcement Learning (KDDRL). First, a conditional deep generative model synthesizes intention-aware future trajectories, converting passive perception into proactive predictive states. Second, a knowledge-data dual-driven paradigm operates on these predictive states, fusing probabilistic data-driven insights with physical constraints to guide safe exploration through safety-critical scenarios. Third, a coupling module compresses both intention-aware trajectories and physical constraints into compact shared embeddings. This unified representation enables asynchronous multi-timescale optimization of continuous car-following and discrete lane-changing while preserving mutual information. Evaluations on dataset-calibrated simulations demonstrate that KDDRL effectively handles intention uncertainty, accelerates training convergence, and outperforms conventional baseline methods in terms of safety, efficiency, and comfort.

Chinese Translation

在混合交通中，自主车辆（AVs）的决策面临三个相互关联的挑战。首先，基于物理的先验知识融入强化学习（RL）模型中，无法捕捉潜在的互动车辆意图和多样化的驾驶行为，从而限制了主动推理能力。其次，周围车辆的突然操控导致非平稳性，使得长尾安全事件未得到充分探索。第三，混合动作空间由于连续跟车和离散变道操控的不同时间尺度而使统一的RL训练不稳定。为了解决这些问题，我们提出了知识-数据双驱动强化学习（KDDRL）。首先，一个条件深度生成模型合成意图感知的未来轨迹，将被动感知转化为主动预测状态。其次，知识-数据双驱动范式在这些预测状态上运行，将基于概率的数据驱动洞察与物理约束相融合，以指导在安全关键场景中的安全探索。第三，一个耦合模块将意图感知的轨迹和物理约束压缩为紧凑的共享嵌入。这种统一表示使得连续跟车和离散变道的异步多时间尺度优化成为可能，同时保持互信息。对数据集校准的模拟评估表明，KDDRL有效地处理了意图不确定性，加速了训练收敛，并在安全性、效率和舒适性方面超越了传统基线方法。

View on arXiv Download PDF AI Translation

cs.RO / 5 / 2608.13901

Ontology-Grounded World Models for Failure Diagnosis and Closed-Loop Repair in Physical AI Systems

基于本体的世界模型用于物理人工智能系统中的故障诊断与闭环修复

Wang, Kailin, Jie, Haoxiang, Yan, Yaoyuan, Zhou, Jiacheng, Heng, Zhiyou

Abstract

EV-WM represents candidate quality with feature and event scores, but these scores do not explicitly record an unmet task predicate, a route label for an available correction mechanism, or a post-correction acceptance result. We present Onto-EV-WM, an ontology-grounded diagnosis and verification-gated correction interface layered above EV-WM rather than a replacement world-model architecture. The implemented task-local TBox defines entity types, predicate signatures, and constraints; source-specific grounding maps predicted or simulator-observed states to task ABoxes; and deterministic rules retain each missing predicate and its arguments when assigning a route label. Learned or heuristic proposers remain separate from this symbolic interface; native task predicates determine acceptance, and the bounded protocol determines whether a failed verification is retried. In the aligned PointMaze evaluation, EV-WM and Onto-EV-WM both report 94% success, with mean final-state distances of 0.90573 and 0.61177, respectively; the separately budgeted search reaches 100% success. On LIBERO-Goal, the ontology represents failed task conditions as typed records, retains their predicate arguments, and associates them with the declared source/joint correction route and predicate-gated acceptance; the complete configuration reports 93.8% corrected-window success on seed 0 and 94.05 +- 0.30% across four evaluation-sampling seeds. On the fixed 10,030-task LIBERO-Plus registry, Onto-EV-WM succeeds on 8,526 tasks (85.00%), with suite-level success rates of 65.98% for LIBERO-10, 91.39% for LIBERO-Goal, and 91.38% for both LIBERO-Object and LIBERO-Spatial. These numbers report the performance of the complete ontology-grounded configurations under the tested simulator protocols; an ontology-only causal share is not measured separately, and real-robot recovery is not evaluated.

Chinese Translation

EV-WM通过特征和事件评分来表示候选质量，但这些评分并未明确记录未满足的任务谓词、可用修正机制的路径标签或修正后的接受结果。我们提出了Onto-EV-WM，这是一种基于本体的诊断和验证门控修正接口，层叠在EV-WM之上，而不是替代的世界模型架构。实现的任务局部TBox定义了实体类型、谓词签名和约束；源特定的基础映射将预测或模拟观察到的状态映射到任务ABoxes；确定性规则在分配路径标签时保留每个缺失的谓词及其参数。学习的或启发式的提议者与该符号接口保持分离；本地任务谓词决定接受，有限协议决定是否重试失败的验证。在对齐的PointMaze评估中，EV-WM和Onto-EV-WM均报告94%的成功率，最终状态的平均距离分别为0.90573和0.61177；单独预算的搜索达到了100%的成功率。在LIBERO-Goal上，本体将失败的任务条件表示为类型记录，保留其谓词参数，并将其与声明的源/联合修正路径和谓词门控接受关联；完整配置在种子0上报告93.8%的修正窗口成功率，在四个评估采样种子上报告94.05 ± 0.30%。在固定的10,030任务LIBERO-Plus注册表中，Onto-EV-WM在8,526个任务上成功（85.00%），LIBERO-10的套件级成功率为65.98%，LIBERO-Goal为91.39%，而LIBERO-Object和LIBERO-Spatial均为91.38%。这些数字报告了在测试的模拟器协议下完整的基于本体的配置的性能；仅本体的因果份额未单独测量，真实机器人恢复未进行评估。

View on arXiv Download PDF AI Translation

cs.RO / 6 / 2608.13904

Communication in modular robotic motor control: Bilateral controllers under realistic constraints

模块化机器人运动控制中的通信：在现实约束下的双向控制器

Li, Jingwen, Kuhlmann, Levin, Friedman, Jason, Kowadlo, Gideon

Abstract

Robotic motor control in musculoskeletal systems requires fast, accurate movement and robust postural stabilization under signal-dependent noise (where motor command variance scales with command magnitude) and energetic cost. Modular controllers can distribute these competing demands across interacting submodules, but it remains unclear whether they outperform monolithic architectures under realistic constraints, and how inter-module communication shapes the resulting strategy. Inspired by the bilateral hemispheric organization of the brain, we introduce a recurrent controller of two GRU-based modules connected by a learnable, delayed inter-hemispheric channel, trained end-to-end in a differentiable two-arm musculoskeletal simulator. Across reaching and holding tasks, the modular architecture substantially outperforms a capacity-matched monolithic baseline. Compared to a matched modular controller without communication, learned inter-hemispheric communication reshapes the solution: improved endpoint precision, lower energetic cost in non-zero-delay regimes, and reduced muscle co-contraction. Our findings show that for robotics, biologically inspired modular controllers offer a practical route to robust movement under noise and energetic constraints, with inter-module communication providing a mechanism to tune trade-offs between precision, stability, and actuation cost.

Chinese Translation

在肌肉骨骼系统中，机器人运动控制需要快速、准确的运动和在信号依赖噪声（即运动指令方差与指令幅度成比例）和能量成本下的稳健姿态稳定。模块化控制器可以将这些相互竞争的需求分配到相互作用的子模块中，但在现实约束下，它们是否优于单体架构仍不清楚，以及模块间通信如何塑造最终策略。受到大脑双侧半球组织的启发，我们引入了一种递归控制器，由两个基于门控循环单元（GRU）的模块组成，通过一个可学习的延迟跨半球通道连接，在一个可微分的双臂肌肉骨骼模拟器中进行端到端训练。在抓取和保持任务中，模块化架构显著优于容量匹配的单体基线。与一个没有通信的匹配模块控制器相比，学习到的跨半球通信重塑了解决方案：提高了终点精度，在非零延迟条件下降低了能量成本，并减少了肌肉共收缩。我们的研究结果表明，对于机器人技术而言，生物启发的模块化控制器提供了一条在噪声和能量约束下实现稳健运动的实用途径，而模块间通信则提供了一种调节精度、稳定性和驱动成本之间权衡的机制。

View on arXiv Download PDF AI Translation

cs.RO / 7 / 2608.13924

BICPO-VLA: Behavior-Identified Continuation Preference Optimization for Smooth Asynchronous Vision-Language-Action Control

BICPO-VLA：基于行为识别的连续偏好优化用于平滑异步视觉-语言-动作控制

Shang, Ming, Huang, Yuchen, Chen, Jiaoyang, Hu, Haoyuan, Yu, Han, Song, Liping, Feng, Luyun, Bao, Shuo, Dong, Wei, Wang, Xinzhou, Sun, Fuchun

Abstract

The request-to-handoff gap has three coupled sources: ambiguity about the behavior intended at request time, physical-state drift accumulated during action generation, and residual incompatibility when the new action finally assumes control. BICPO-VLA addresses them in sequence. First, an instruction-aware causal history encoder identifies the behavior supported by the command and current task progress. Second, sequential Haar subspace generation decomposes each action chunk into complementary pairwise scaffold and residual coefficients, enabling two specialized generation stages followed by exact reconstruction. By reducing iterative refinement in the original action space, it shortens the interval over which the robot continues moving before the new chunk becomes available. Finally, BICPO rolls the known outgoing actions to the actual handoff state and applies reference-relative Flow-DPO among behaviorally matched candidates, adapting the generated chunk to the remaining request-to-handoff mismatch without changing its intended behavior.

Chinese Translation

请求与交接之间的间隙有三个相互关联的来源：请求时意图行为的模糊性、在动作生成过程中累积的物理状态漂移，以及当新动作最终接管控制时的残余不兼容性。BICPO-VLA 依次解决这些问题。首先，基于指令的因果历史编码器识别由命令和当前任务进展所支持的行为。其次，序列 Haar 子空间生成将每个动作块分解为互补的成对支架和残余系数，从而实现两个专门的生成阶段，随后进行精确重构。通过减少在原始动作空间中的迭代精炼，它缩短了机器人在新块可用之前继续移动的时间间隔。最后，BICPO 将已知的外发动作滚动到实际的交接状态，并在行为匹配的候选者中应用相对参考的 Flow-DPO，调整生成的块以适应剩余的请求与交接不匹配，而不改变其预期行为。

View on arXiv Download PDF AI Translation

cs.RO / 8 / 2608.14028

AdvDex: Learning Dexterous Manipulation from Human Demonstrations via Joint-Aligned Actions and Adversarial Learning

AdvDex：通过关节对齐动作和对抗学习从人类示范中学习灵巧操作

Zhao, Zhiyue, Wu, Jingyi, Liu, Hairuo, Liu, Mingyu, Li, Liyang, Zhang, Hengdi, He, Tong, Cheng, Zhengxue

Abstract

Dexterous manipulation is a fundamental capability for embodied intelligence, but scaling it remains difficult because robot demonstrations are expensive to collect and action spaces vary across embodiments. Policies trained on heterogeneous data can also entangle task-relevant visual cues with embodiment-specific appearance, limiting cross-embodiment generalization. We present AdvDex, a unified Vision-Language-Action framework for learning dexterous manipulation from human and robot demonstrations. First, we introduce OmniShare, a large-scale multimodal dataset of human manipulation demonstrations that provides high-quality kinematic supervision and tactile measurements while reducing reliance on robot teleoperation. Second, we propose the Joint-Aligned Action Space (JAAS), a canonical action representation comprising an $\mathrm{SE}(3)$ wrist pose and 15 finger joints, thereby functionally aligning human hands, dexterous robot hands, and parallel grippers. Finally, we use domain-adversarial learning to reduce embodiment-specific information in the learned visual representation. Experiments on hand-action prediction and real-world dexterous manipulation show consistent improvements over baselines, effective zero-shot human-to-robot skill transfer, generalization to unseen objects and environments, and data-efficient few-shot adaptation.

Chinese Translation

灵巧操作是具身智能的基本能力，但由于机器人示范的收集成本高昂以及不同具身体之间的动作空间差异，扩展这一能力仍然困难。基于异构数据训练的策略可能会将与任务相关的视觉线索与具身特定的外观混合，从而限制跨具身体的泛化能力。我们提出了AdvDex，一个统一的视觉-语言-动作框架，用于从人类和机器人示范中学习灵巧操作。首先，我们引入了OmniShare，一个大规模的多模态数据集，包含人类操作示范，提供高质量的运动学监督和触觉测量，同时减少对机器人遥操作的依赖。其次，我们提出了关节对齐动作空间（Joint-Aligned Action Space, JAAS），这是一种标准化的动作表示，包含一个$ ext{SE}(3)$腕部姿态和15个手指关节，从而在功能上对齐人类手、灵巧机器人手和并行夹具。最后，我们使用领域对抗学习来减少学习到的视觉表示中的具身特定信息。在手部动作预测和现实世界灵巧操作的实验中，我们在基线之上显示出一致的改进，成功实现零样本的人类到机器人技能转移，泛化到未见过的物体和环境，以及数据高效的少样本适应。

View on arXiv Download PDF AI Translation

cs.RO / 9 / 2608.14031

Demonstration of Space Robot Teleoperation over a Lossy and Delayed Network using ATMOS

基于ATMOS的空间机器人在有损失和延迟网络下的远程操作演示

Jang, Inkyu, Marchesini, Gregorio, De Carli, Nicola, Kim, Byeongjun, Hwang, Sunwoo, Kim, Dabin, Krantz, Elias, Kong, Youngkyoung, Jiang, Frank J., Wong, Annika, Roque, Pedro, Sanjaya, Prasetyo W. L., Bastianello, Nicola, Dhullipalla, Mani H., Johansson, Karl H., Shim, Hyungbo, Dimarogonas, Dimos V., Kim, H. Jin

Abstract

We present a demonstration showcasing the Autonomy Testbed for Multi-purpose Orbiting Systems (ATMOS), a planar spacecraft-analog robot designed for hardware-in-the-loop evaluation of guidance and control strategies in microgravity-like conditions. Using ATMOS as the physical test platform, we investigate the design, analysis, and performance evaluation of control architectures for remotely operated spacecraft under round-trip communication delays. In this work, we develop and experimentally validate a control strategy that combines state prediction and trajectory tracking control to perform a docking maneuver, accounting for time-varying random communication latency between ground operators and the ATMOS system. The demonstration includes a long-distance remote control experiment between Seoul and Stockholm, introducing realistic intercontinental delays and variability. The results highlight the capability of ATMOS to support rapid, reliable, and cost-effective testing of spacecraft teleoperation concepts, establishing a first step toward robust validation of on-orbit operations in microgravity-like environments.

Chinese Translation

我们展示了多用途轨道系统自主测试平台（Autonomy Testbed for Multi-purpose Orbiting Systems，ATMOS）的演示，该平台是一种平面航天器类机器人，旨在微重力条件下对引导和控制策略进行硬件在环评估。利用ATMOS作为物理测试平台，我们研究了在往返通信延迟下，远程操作航天器控制架构的设计、分析和性能评估。在本研究中，我们开发并实验验证了一种控制策略，该策略结合了状态预测和轨迹跟踪控制，以执行对接机动，同时考虑了地面操作员与ATMOS系统之间时间变化的随机通信延迟。演示包括在首尔与斯德哥尔摩之间的远程控制实验，模拟了现实的洲际延迟和变异性。结果突显了ATMOS支持快速、可靠且具有成本效益的航天器远程操作概念测试的能力，为在微重力环境中稳健验证在轨操作奠定了第一步基础。

View on arXiv Download PDF AI Translation

cs.RO / 10 / 2608.14047

Evolve Vision-Language-Action Model into an Agent with On-the-fly Tool-use

将视觉-语言-动作模型演化为具备即时工具使用能力的智能体

Ding, Yi, Yu, Yanzhao, Dai, Xili, Qi, Xianbiao, Sun, Peiwen, Wang, Xueqian, Yue, Xiangyu, Wang, Jianan

Abstract

This paper integrates end-to-end Visual-Language-Action (VLA) models with agentic tool-use to propose Agentic Robot with Tool-use (ART). ART is a tool-injection framework that tunes any VLA model to leverage off-the-shelf tool modules for low-level vision, high-level affordance, and embodiment enhancement. Compared to vanilla VLA models with a whole continuous action solution space, ART reduces the complexity of the action solution space through tool-use, which not only improves generalizability across different tasks but also reduces data dependency. To demonstrate the advantages (high generalizability and low data dependency) of this framework, we first built a dataset of 30K tool-use trajectories and action demonstrations, which is much smaller than those used by baseline methods. We then designed a training regimen for long-trajectory tool-use reasoning in challenging environments. Experiments show that ART achieves a 20% higher success rate than mainstream baselines on simulation and real-world tasks, such as pick-and-place in the dark at novel viewpoints. Empirical results highlight the benefits of an agent-based approach: modular tool utilization enables more efficient training, lightweight deployment, and scalable integration of new tools. This design fosters robustness, adaptability, and extensibility, paving the way for the practical deployment of VLA systems in complex real-world scenarios.

Chinese Translation

本文将端到端的视觉-语言-动作（VLA）模型与智能工具使用相结合，提出了具备工具使用能力的智能机器人（ART）。ART是一个工具注入框架，旨在调整任何VLA模型，以利用现成的工具模块来增强低级视觉、高级可供性和具身性。与传统的VLA模型相比，ART通过工具使用降低了动作解决空间的复杂性，这不仅提高了在不同任务中的泛化能力，还减少了对数据的依赖。为了展示该框架的优势（高泛化能力和低数据依赖），我们首先构建了一个包含3万条工具使用轨迹和动作示范的数据集，远小于基线方法所使用的数据集。随后，我们设计了一种训练方案，以应对复杂环境中的长轨迹工具使用推理。实验结果表明，ART在模拟和现实任务（如在新视角下的黑暗环境中的拾取与放置）中，成功率比主流基线高出20%。实证结果突显了基于智能体的方法的优势：模块化工具的利用使得训练更加高效、部署更加轻量，并且新工具的集成更加可扩展。这一设计促进了系统的鲁棒性、适应性和可扩展性，为VLA系统在复杂现实场景中的实际部署铺平了道路。

View on arXiv Download PDF AI Translation

cs.RO / 11 / 2608.14049

FlatLab: A Unified Methodology Framework and Simulation-Based Benchmark for Robotic Manipulation of Flat Objects

FlatLab：一种统一的方法论框架和基于仿真的平面物体机器人操控基准

Zhu, Xingyu, Han, Wenshuo, Wang, Zhouyu, Wang, Yuran, Wu, Ruihai, Dong, Hao, Tang, Fan, Chen, Hechang, Chang, Hyung Jin, Gao, Yixing

Abstract

Robotic manipulation of flat objects is challenging due to the ungraspable configurations and strong variations in object geometry and material. Existing methods rely on heuristic pre-manipulation and are often evaluated in closed settings with limited generalization. We propose a unified framework that decouples the manipulation into a strategy generator and an action execution module. The strategy generator predicts appropriate manipulation strategies from object point clouds by learning strategy-centric, object-invariant representations via simulated data transformation and contrastive learning. Conditioned on the predicted strategy, the execution module decomposes long-horizon manipulation into reusable action primitives and dynamically composes them to generate stable trajectories. To enable systematic evaluation, we introduce FlatLab, a comprehensive simulation benchmark for robotic flat object manipulation. FlatLab provides high-fidelity physical simulation of diverse rigid and deformable flat objects, automated multi-modal data collection, and standardized task definitions and evaluation protocols. Experiments conducted in FlatLab demonstrate that our approach generalizes effectively to unseen objects and categories, outperforming existing baselines. The project page and the code are provided at https://flatlab-web.github.io/.

Chinese Translation

平面物体的机器人操控具有挑战性，因为其存在不可抓取的配置以及物体几何形状和材料的强烈变化。现有方法依赖于启发式的预操控，并且通常在封闭环境中进行评估，缺乏广泛的泛化能力。我们提出了一种统一的框架，将操控过程解耦为策略生成器和动作执行模块。策略生成器通过学习以策略为中心、与物体无关的表示，从物体点云中预测适当的操控策略，利用仿真数据转换和对比学习进行训练。在预测策略的条件下，执行模块将长时间跨度的操控分解为可重用的动作原语，并动态组合这些原语以生成稳定的轨迹。为了实现系统化评估，我们引入了FlatLab，这是一个全面的平面物体机器人操控仿真基准。FlatLab提供了多样化的刚性和可变形平面物体的高保真物理仿真、自动化的多模态数据收集以及标准化的任务定义和评估协议。在FlatLab中进行的实验表明，我们的方法能够有效地泛化到未见过的物体和类别，超越了现有的基准。项目页面和代码可在 https://flatlab-web.github.io/ 获取。

View on arXiv Download PDF AI Translation

cs.RO / 12 / 2608.14082

PILOT: Privileged Imitation Learning for End-to-End Motion Planning of Autonomous UAVs under Partial Observability

PILOT：针对部分可观测环境下自主无人机端到端运动规划的特权模仿学习

Zhang, Qingrui, Xue, Feng, Zhou, Xiang, Yu, Chenghao

Abstract

Autonomous navigation in cluttered environments is hampered by partial observability and dynamic constraints. This paper presents PILOT, a constraint-aware privileged imitation learning framework for vision-based end-to-end UAV motion planning under partial observability. The framework distills planning strategies from a computationally intensive optimal control expert into a student policy regularized toward safety and dynamic requirements via a dual-objective loss function. To mitigate partial observability, a spatiotemporal perception fusion module using a Temporal Convolutional Network (TCN) is developed to integrate historical depth images and odometry. This module infers task-relevant latent context from historical observations, enhancing spatial awareness beyond the instantaneous FOV without maintaining persistent map memory. A trajectory parameterization layer mapping network outputs to a structured trajectory, while enabling explicit continuity, dynamic-consistency, and obstacle soft penalties during training, encouraging constraint satisfaction for unseen observations without formal guarantees. Simulations on quadrotor and fixed-wing aircraft demonstrate that PILOT achieves performance comparable to the privileged expert while reducing computational overhead by over 80\%. Successful indoor and outdoor zero-shot deployment confirms the practical feasibility and cross-domain generalization of the planner.

Chinese Translation

在复杂环境中，自主导航受到部分可观测性和动态约束的制约。本文提出了PILOT，一种基于视觉的端到端无人机运动规划的约束感知特权模仿学习框架，适用于部分可观测性。该框架将来自计算密集型最优控制专家的规划策略提炼为一个学生策略，并通过双目标损失函数对安全性和动态要求进行正则化。为了缓解部分可观测性，开发了一个使用时间卷积网络（Temporal Convolutional Network, TCN）的时空感知融合模块，以整合历史深度图像和里程计数据。该模块从历史观测中推断出与任务相关的潜在上下文，增强了超越瞬时视场（FOV）的空间感知，而无需保持持久的地图记忆。轨迹参数化层将网络输出映射到结构化轨迹，同时在训练过程中显式地实现连续性、动态一致性和障碍物软惩罚，鼓励在没有正式保证的情况下对未见观测的约束满足。对四旋翼和固定翼飞机的仿真表明，PILOT的性能与特权专家相当，同时计算开销降低超过80%。成功的室内和室外零样本部署验证了该规划器的实际可行性和跨领域泛化能力。

View on arXiv Download PDF AI Translation

cs.RO / 13 / 2608.14135

AgilePE: Autonomous UAV Pursuit-Evasion via Self-Play Reinforcement Learning

AgilePE：通过自我对弈强化学习实现自主无人机追逐-逃避

Tang, Wenhao, Chen, Tianyang, Cui, Zhejun, An, Boyuan, Chen, Jiayu, Zhang, Ruize, Liu, Huidong, Wu, Tianyue, Liao, Qingmin, Gao, Fei, Wang, Yu, Yu, Chao

Abstract

Autonomous pursuit-evasion is a fundamental challenge for Unmanned Aerial Vehicles (UAVs), requiring rapid decision-making under tightly coupled dynamics and continuously changing opponent behaviors. Traditional rule-based or differential-game approaches often struggle with high-dimensional aerial interactions and agile maneuvering. We present AgilePE, a complete system for autonomous UAV pursuit-evasion via self-play reinforcement learning. AgilePE integrates agile low-level control, competitive policy optimization, and sim-to-real deployment in a unified framework. The policy directly maps onboard state observations to Collective Thrust and Body Rates (CTBR) commands, enabling end-to-end agile maneuvering without intermediate trajectory planners or waypoint controllers. For training, we use competitive self-play with Prioritized Fictitious Self-Play (PFSP) and a diversified opponent pool, enabling agents to improve against historical policies while stabilizing optimization and reducing policy oscillation. This process leads to the emergence of sophisticated pursuit and evasion strategies. For real-world deployment, we develop a hardware-aligned simulation pipeline that models actuator-response dynamics, communication latency, and domain randomization. The learned policies transfer zero-shot to real quadrotors without task-specific tuning. Real-world experiments reproduce pursuit-evasion tactics observed in simulation, including rapid dodging and flanking, and demonstrate interactive two-agent zero-shot deployment.

Chinese Translation

自主追逐-逃避是无人机（UAV）面临的一项基本挑战，要求在紧密耦合的动态环境和不断变化的对手行为下进行快速决策。传统的基于规则或微分博弈的方法在高维空中交互和灵活机动方面常常面临困难。我们提出了AgilePE，这是一个通过自我对弈强化学习实现自主无人机追逐-逃避的完整系统。AgilePE将灵活的低级控制、竞争性策略优化和从仿真到现实的部署整合在一个统一的框架中。该策略直接将机载状态观测映射为集体推力和机体速率（Collective Thrust and Body Rates, CTBR）指令，使得在没有中间轨迹规划器或航点控制器的情况下实现端到端的灵活机动。为了进行训练，我们使用了优先虚构自我对弈（Prioritized Fictitious Self-Play, PFSP）和多样化的对手池进行竞争性自我对弈，使得智能体能够在稳定优化和减少策略振荡的同时，针对历史策略进行改进。这个过程导致了复杂的追逐和逃避策略的出现。为了实现现实世界的部署，我们开发了一个与硬件对齐的仿真管道，模拟执行器响应动态、通信延迟和领域随机化。学习到的策略可以零-shot迁移到真实的四旋翼无人机上，而无需特定任务的调优。现实世界的实验重现了在仿真中观察到的追逐-逃避战术，包括快速躲避和侧翼攻击，并展示了交互式双智能体的零-shot部署。

View on arXiv Download PDF AI Translation

cs.RO / 14 / 2608.14160

OccPlanner: Goal-Aware Occupancy-Conditioned Diffusion Planner for Pixel-Goal Navigation

OccPlanner：面向目标的占用条件扩散规划器用于像素目标导航

Huang, Binling, Ye, Nianjin, Yang, Xi, Hu, Liang, Huang, Zhou, Wei, Shuang, Yang, Longrui, Chen, Yanchi, Jia, Lanpeng

Abstract

Pixel-goal navigation specifies targets directly in the agent's camera view, but a target pixel provides neither metric depth nor traversability, making 3D goal grounding and collision-free continuous planning challenging. We present OccPlanner, a goal-aware occupancy-conditioned diffusion planner that grounds pixel goals in egocentric metric space and sequentially conditions the goal representation on temporal visual context and learned local 3D occupancy features. To provide occupancy supervision at scale, we introduce L3ROcc, which converts monocular RGB navigation videos into robot-centric local 3D occupancy annotations through geometric reconstruction and ray-based visibility reasoning. We train OccPlanner on InternData-N1 and evaluate it in closed-loop simulation across four unseen scene categories from InternScenes and two goal-distance ranges. In the 5-8 m setting, OccPlanner increases the average success rate (SR) over NavDP from 20.81% to 71.55% across the four categories, reaching 86.20% and 84.92% in cluttered-easy and cluttered-hard scenes, respectively. Real-world open-loop experiments on a Unitree Go2 further provide initial evidence of sim-to-real transfer and adaptation with L3ROcc-generated supervision.

Chinese Translation

像素目标导航直接在智能体的摄像头视野中指定目标，但目标像素既不提供度量深度也不提供可通行性，这使得3D目标定位和无碰撞的连续规划变得具有挑战性。我们提出了OccPlanner，一种面向目标的占用条件扩散规划器，它在自我中心的度量空间中定位像素目标，并基于时间视觉上下文和学习到的局部3D占用特征顺序地对目标表示进行条件化。为了大规模提供占用监督，我们引入了L3ROcc，它通过几何重建和基于光线的可见性推理将单目RGB导航视频转换为机器人中心的局部3D占用注释。我们在InternData-N1上训练OccPlanner，并在来自InternScenes的四个未见场景类别和两个目标距离范围内进行闭环仿真评估。在5-8米的设置中，OccPlanner在四个类别中将平均成功率（SR）从NavDP的20.81%提高到71.55%，在杂乱-简单和杂乱-困难场景中分别达到86.20%和84.92%。在Unitree Go2上的真实世界开环实验进一步提供了L3ROcc生成的监督在模拟到现实转移和适应方面的初步证据。

View on arXiv Download PDF AI Translation

cs.RO / 15 / 2608.14207

MMUSV-Sim: A Perception-Oriented Simulation and Data-Generation Platform for Multi-USV Cooperative Perception

MMUSV-Sim：面向感知的多无人水面艇协同感知仿真与数据生成平台

Li, Ziao, Ye, Jianxiong, Tang, Biao, Zhang, Leping, Zuo, Kun, Huang, Siyu, Gao, Chenqiang

Abstract

Cooperative perception among multiple unmanned surface vehicles (USVs) combines complementary observations to extend maritime target sensing beyond the view range and field of a single platform. Developing such systems at scale calls for a unified workflow for configurable multi-USV scenarios, multimodal acquisition, and shared annotations. We present MMUSV-Sim, a perception-oriented maritime simulation and data-generation platform built on Unreal Engine 5 and Project AirSim. It provides island, open-sea, and port environments; configurable weather, time of day, and wave conditions; a diverse vessel asset library; and spline-based multi-vessel motion. MMUSV-Sim acquires RGB, depth, semantic, LiDAR, and radar observations across multiple USVs and captures a common world state for per-agent annotation export. Experiments verify that the configured wave settings produce the intended changes in vessel heave, roll, and pitch, and evaluate the geometric consistency between projected annotations and semantic renderings. In LiDAR-based cooperative BEV vessel detection experiments on the generated multi-USV dataset, Early Fusion achieves an [email protected] of 72.74, compared with 45.54 using a single USV.

Chinese Translation

多无人水面艇（USV）之间的协同感知通过结合互补观察，扩展了海洋目标感知的范围，超出了单个平台的视野和领域。大规模开发此类系统需要一个统一的工作流程，以支持可配置的多USV场景、多模态采集和共享注释。我们提出了MMUSV-Sim，这是一个基于虚幻引擎5（Unreal Engine 5）和Project AirSim构建的面向感知的海洋仿真与数据生成平台。它提供了岛屿、开阔海域和港口环境；可配置的天气、时间和波浪条件；多样的船舶资产库；以及基于样条曲线的多船舶运动。MMUSV-Sim在多个USV上获取RGB、深度、语义、激光雷达（LiDAR）和雷达观察，并捕获每个代理的共同世界状态以便进行注释导出。实验验证了配置的波浪设置在船舶的纵摇、横摇和俯仰方面产生了预期的变化，并评估了投影注释与语义渲染之间的几何一致性。在基于LiDAR的协同鸟瞰视角（BEV）船舶检测实验中，使用生成的多USV数据集，早期融合（Early Fusion）方法在[email protected]上达到了72.74，而单个USV的结果为45.54。

View on arXiv Download PDF AI Translation

cs.RO / 16 / 2608.14244

Vibration Suppression in Collaborative Flexible Payload Manipulation Using Passive Force Control

使用被动力控制的协作柔性载荷操作中的振动抑制

Abderrahim, Alaa, Rosales, Antonio, Milella, Ferdinando, Suomalainen, Markku, Li, Shuai

Abstract

In large and heavy structures, vibrations arise during motion, posing significant challenges for precise manipulation. To accomplish the desired motion, control algorithms must effectively suppress these structural vibrations. In cutting edge projects, such as remote maintenance of future fusion energy reactors (tokamaks), the manipulation of this type of structure is defined as a crucial task. This paper presents a control strategy to suppress transverse vibrations in flexible payloads during motion using a collaborative payload manipulation approach. Two different industrial robot arms are arranged in a leader follower configuration for the manipulation strategy. The leader robot guides the motion with shaped velocity commands, while the follower robot ensures compliance with the estimated external forces applied by the leader on the payload through an admittance controller. Unlike existing methods, the proposed approach enables collaborative manipulation of heavier and larger flexible objects, addressing additional challenges such as vibration suppression and heterogeneous robot specifications. The dynamics of the leader follower payload system are modeled using an equivalent mass spring damper model, and it is shown that, with appropriate admittance parameters, the total energy of the system is passively dissipated. A stability proof is also provided. Numerical simulations validate the proposed method, and experimental results demonstrate its effectiveness.

Chinese Translation

在大型和重型结构中，运动过程中会产生振动，这对精确操作构成了重大挑战。为了实现所需的运动，控制算法必须有效抑制这些结构振动。在前沿项目中，例如未来聚变能反应堆（托卡马克）的远程维护，这种结构的操作被定义为一项关键任务。本文提出了一种控制策略，通过协作载荷操作方法抑制柔性载荷在运动过程中的横向振动。两种不同的工业机器人臂以领导-跟随配置排列以实现操作策略。领导机器人通过设定的速度指令引导运动，而跟随机器人则通过导纳控制器确保对领导机器人施加在载荷上的估计外力的顺应性。与现有方法不同，所提出的方法能够协作操作更重和更大的柔性物体，解决了振动抑制和异构机器人规格等额外挑战。领导-跟随载荷系统的动力学使用等效质量弹簧阻尼器模型进行建模，并且证明在适当的导纳参数下，系统的总能量被被动耗散。还提供了稳定性证明。数值仿真验证了所提出的方法，实验结果展示了其有效性。

View on arXiv Download PDF AI Translation

cs.RO / 17 / 2608.14266

Accelerating Large-scale Bundle Adjustment for LiDAR Mapping via Parallel Computing

通过并行计算加速大规模LiDAR映射的束调整

Cai, Yixi, Li, Rundong, Xie, Yuhan, Zhang, Qingwen, Jensfelt, Patric, Zhang, Fu

Abstract

LiDAR bundle adjustment is widely utilized in mapping to construct globally consistent point cloud maps. In this paper, we propose the first fully parallel computing framework to accelerate LiDAR bundle adjustment for large-scale mapping, incorporating three key techniques. First, we design an adaptive, asynchronous data loading strategy to efficiently process large-scale point cloud datasets on memory-constrained GPUs. Secondly, we present a novel bottom-up voxelization method for extracting planar features, enabling fully parallelized pre-processing. Thirdly, we build upon a majorization-minimization formulation to accelerate compute-intensive tasks in the optimization via parallel computation, including the computation of residuals, Jacobian and Hessian matrices, and a parallel increment solver. To support our design, we provide both theoretical and experimental analysis of the time complexity of our approach. Extensive benchmarking on large-scale public datasets across various computational platforms validates the robustness and adaptability of our approach, achieving up to a tenfold improvement in computational efficiency while preserving mapping accuracy comparable to state-of-the-art methods. To benefit future research, the implementation code is available on GitHub.

Chinese Translation

LiDAR束调整广泛应用于映射，以构建全球一致的点云地图。在本文中，我们提出了第一个完全并行计算框架，以加速大规模映射中的LiDAR束调整，结合了三项关键技术。首先，我们设计了一种自适应的异步数据加载策略，以高效处理在内存受限的GPU上大规模点云数据集。其次，我们提出了一种新颖的自下而上的体素化方法，用于提取平面特征，实现完全并行的预处理。第三，我们基于极小化主要化公式，通过并行计算加速优化中的计算密集型任务，包括残差、雅可比矩阵和海森矩阵的计算，以及并行增量求解器。为了支持我们的设计，我们提供了理论和实验分析，评估我们方法的时间复杂度。在各种计算平台上对大规模公共数据集进行的广泛基准测试验证了我们方法的鲁棒性和适应性，在保持与最先进方法相当的映射精度的同时，实现了计算效率最高可达十倍的提升。为了促进未来的研究，实施代码已在GitHub上提供。

View on arXiv Download PDF AI Translation

cs.RO / 18 / 2608.14284

PRM-as-a-Judge 1.5: A Toolkit for Robot Process Assessment

PRM-as-a-Judge 1.5：机器人过程评估工具包

Liu, Yuyang, Shen, Yanqing, Chen, Ruike, Zhao, Jifan, Tian, Yuxuan, Zhang, Yichi, Long, Tianfeng, Yin, Zixuan, Wang, Yipu, Qin, Ziheng, Tan, Wenxing, Shi, Yang, Cao, Mingyu, Xiao, Runze, Wang, Ziqi, Yin, Zhixin, Chu, Shiwei, Zhang, Yi-Fan, Mu, Yao, Ji, Yuheng, Wang, Yihao, Yan, Jun, Wang, Zhongyuan, Wang, Pengwei, Zheng, Xiaolong

Abstract

Fine-grained robotic evaluation matters for understanding embodied models, going beyond binary success rates and rule-based process scores. We present PRM-as-a-Judge 1.5, a toolkit for robot process assessment that turns rollout videos into dense progress curves and derives multiple fine metrics. PRM-as-a-Judge 1.5 introduces three metrics, building on version 1.0, that characterize failure-side progress, post-drawdown recovery, and success-side execution quality, helping users understand embodied model capability. Based on the rollout videos from benchmarks, we perform a comprehensive assessment of the embodied models, providing some fine-grained metric results and key findings. We further introduce RoboPulse++ to evaluate the reliability of process reward models (PRM), providing evaluators with a more accurate testing platform. Moreover, we release a user-friendly assessment suite, including the benchmark, metric implementation, and visualization tools, to support reproducible manipulation process evaluation. We call on the community to rethink how robots are evaluated and establish transparent, procedural, and reproducible assessment as a foundation for the next generation of embodied intelligence.

Chinese Translation

细粒度的机器人评估对于理解具身模型至关重要，超越了二元成功率和基于规则的过程评分。我们介绍了PRM-as-a-Judge 1.5，这是一个用于机器人过程评估的工具包，它将回放视频转化为密集的进展曲线，并衍生出多个细致的指标。PRM-as-a-Judge 1.5在1.0版本的基础上引入了三个指标，分别表征失败侧进展、回落后的恢复以及成功侧执行质量，帮助用户理解具身模型的能力。基于基准测试的回放视频，我们对具身模型进行了全面评估，提供了一些细粒度的指标结果和关键发现。我们进一步引入了RoboPulse++来评估过程奖励模型（PRM）的可靠性，为评估者提供了一个更准确的测试平台。此外，我们发布了一个用户友好的评估套件，包括基准测试、指标实现和可视化工具，以支持可重复的操作过程评估。我们呼吁社区重新思考机器人评估的方式，并建立透明、程序化和可重复的评估作为下一代具身智能的基础。

View on arXiv Download PDF AI Translation

cs.RO / 19 / 2608.14332

CORAL: Curriculum-Optimized Reward Adaptation for LiDAR-Based Goal-Directed Urban Driving

CORAL：基于激光雷达的目标导向城市驾驶的课程优化奖励适应

Saleem, Anisa, Kim, Duksu

Abstract

Reinforcement learning is promising for autonomous urban driving, but long-horizon goal-directed navigation asks a policy to acquire several competing behaviors at once--reaching a distant goal, tracking a route, avoiding obstacles, obeying signals--and a fixed objective gives no order in which to learn them. This paper presents CORAL, which advances two schedules together: a five-stage curriculum that progressively lengthens routes and tightens behavioral constraints, and a stage-aware reward whose component weights shift emphasis from mission progress toward route following, safety, smoothness, and rule compliance as the task hardens. The policy is a multi-stream actor-critic network trained with Proximal Policy Optimization (PPO) in CARLA on a compact 99-dimensional state pairing a polar LiDAR histogram with vehicle telemetry, ego-frame route geometry, and traffic-rule indicators--no point-cloud encoder, no bird's-eye-view rasterization. Against two PPO baselines under an identical protocol, CORAL reaches the goal in all twenty evaluation episodes on the longest routes under the full set of behavioral constraints, where the baselines reach 5% and 10%; a factorial ablation shows that neither schedule alone matches their combination: removing either lowers both success and route completion, and disabling both drops success to 55%. Trained in one town, the policy transfers zero-shot to seven unseen towns, succeeding in 68-98% of episodes on routes of the same 100-150 m length, with mean lateral deviation below 0.35 m.

Chinese Translation

强化学习在自主城市驾驶中展现出良好的前景，但长时间目标导向导航要求策略同时获得多种竞争行为——到达远处目标、跟踪路线、避免障碍、遵守信号——而固定目标并未提供学习这些行为的顺序。本文提出了CORAL，它同时推进两种调度：一个五阶段的课程，逐步延长路线并收紧行为约束，以及一个阶段感知的奖励，其组成权重随着任务难度的增加，从任务进展转向路线跟随、安全性、平稳性和规则遵守。该策略是一个多流的演员-评论家网络，在CARLA中使用近端策略优化（PPO）进行训练，状态为紧凑的99维，结合了极坐标激光雷达直方图、车辆遥测、自我框架路线几何和交通规则指示器——没有点云编码器，没有鸟瞰图栅格化。在相同协议下的两个PPO基线中，CORAL在最长路线的所有二十个评估回合中都成功到达目标，而基线仅达到5%和10%；一个因子消融实验表明，单独的调度无法匹配它们的组合：去掉任一调度都会降低成功率和路线完成率，而禁用两个调度则使成功率降至55%。在一个城镇训练后，该策略在七个未见过的城镇中零次转移，在相同长度为100-150米的路线中成功率为68%-98%，平均横向偏差低于0.35米。

View on arXiv Download PDF AI Translation

cs.RO / 20 / 2608.14379

Reflex: Enabling Fast and Predictive Vision-Language-Action Models for Reaction-Critical Manipulation

Reflex：为反应关键的操作启用快速和预测性的视觉-语言-动作模型

Chen, Yuxuan, Zhang, Wanruo, Li, Xiao

Abstract

Vision-Language-Action (VLA) models have recently achieved promising performance in robotic manipulation. However, existing benchmarks mainly evaluate generalization on static manipulation tasks and largely overlook dynamic interaction scenarios. To address this gap, we present ReflexBench, a benchmark for reaction-critical manipulation. ReflexBench contains six dynamic tasks and introduces an evaluation framework that decouples simulator stepping from robot control while supporting configurable latency under synchronous and asynchronous inference. Building upon ReflexBench, we propose ReflexVLA, an efficient VLA model designed for reaction-critical manipulation without large-scale robot-data pretraining. ReflexVLA enhances temporal reasoning through latent future prediction and multi-frame temporal fusion within the vision backbone, while reducing deployment latency through batched visual encoding and CUDA Graph replay. Experiments show that ReflexVLA consistently improves dynamic manipulation performance while maintaining competitive accuracy on standard static manipulation benchmarks, and real-world experiments further demonstrate its effectiveness under practical deployment conditions. Project website: https://reflexvla.github.io

Chinese Translation

视觉-语言-动作（VLA）模型最近在机器人操作中取得了令人鼓舞的表现。然而，现有基准主要评估静态操作任务的泛化能力，且在很大程度上忽视了动态交互场景。为了解决这一问题，我们提出了ReflexBench，这是一个针对反应关键操作的基准。ReflexBench包含六个动态任务，并引入了一个评估框架，该框架将模拟器步进与机器人控制解耦，同时支持在同步和异步推理下的可配置延迟。在ReflexBench的基础上，我们提出了ReflexVLA，这是一个高效的VLA模型，旨在实现反应关键的操作，而无需大规模的机器人数据预训练。ReflexVLA通过潜在未来预测和视觉主干中的多帧时间融合增强了时间推理，同时通过批量视觉编码和CUDA图重放减少了部署延迟。实验表明，ReflexVLA在动态操作性能上始终有所提升，同时在标准静态操作基准上保持了竞争力的准确性，实际部署条件下的实验证明了其有效性。项目网站：https://reflexvla.github.io

View on arXiv Download PDF AI Translation

cs.RO / 21 / 2608.14406

Effect of Twisted-Yarn Architecture on Pressure and Proximity Sensing Characteristics of Textile Capacitive Sensors for Robotic Skin

扭绳结构对机器人皮肤纺织电容传感器压力和接近感应特性的影响

Zahir, Ishtia, Saleh, Eslam, Rezayati, Maryam, Grabher, Güunter, Hossain, Gaffar

Abstract

Textile-integrated capacitive sensors offer flexible and conformable tactile sensing for wearable electronics and human-robot interaction; however, the influence of yarn-level architecture on capacitive transduction characteristics remains insufficiently quantified. This work presents a textile capacitive sensing platform based on silver-coated yarns coated with polydimethylsiloxane and assembled into one-, two-, and four-layer twisted configurations. The influence of effective electrode overlap area and inter-fiber separation on the capacitive response is systematically investigated, enabling architecture-dependent tuning of pressure and proximity sensing characteristics. Pressure was calculated using the localized single-fiber contact area, corresponding to stresses of 0.4-3.9 MPa. Increasing the layer number improved mechanical strength and sensing performance: elongation at break increased from 37.5% to 62.5% and 85.0%, while the maximum load increased from 23.3 to 42.7 and 89.7 N. Sensitivity increased with layer number and frequency, reaching 0.1331 MPa$^{-1}$ for the four-layer sensor at 100 kHz. The four-layer configuration also exhibited low hysteresis, minimal thermal drift from 25 to 90 $^\circ$C, and stable operation over 15,000 cycles. Proximity detection ranges of 60, 50, and 40 mm were obtained for the one-, two-, and four-layer sensors, respectively, revealing an architecture-dependent sensitivity-range trade-off. A 4$\times$4 textile sensing array enabled spatial contact mapping, while robotic-arm integration demonstrated real-time touch and proximity detection with an end-to-end robotic system latency (from detection to robot reaction) of 403 ms. The results establish yarn architecture as a tunable design parameter governing the measurement characteristics of textile-integrated capacitive sensing systems.

Chinese Translation

集成纺织品的电容传感器为可穿戴电子设备和人机交互提供了灵活且适应性强的触觉感知；然而，纱线级结构对电容传感特性的影响尚未得到充分量化。本研究提出了一种基于银涂层纱线的纺织电容传感平台，这些纱线涂覆有聚二甲基硅氧烷，并组装成单层、双层和四层扭绳配置。系统地研究了有效电极重叠面积和纤维间距对电容响应的影响，从而实现了基于结构的压力和接近感应特性的调节。压力通过局部单纤维接触面积计算，对应的应力范围为0.4-3.9 MPa。增加层数提高了机械强度和感应性能：断裂伸长率从37.5%增加到62.5%和85.0%，而最大负载从23.3 N增加到42.7 N和89.7 N。灵敏度随着层数和频率的增加而提高，在100 kHz时四层传感器的灵敏度达到0.1331 MPa$^{-1}$。四层配置还表现出低滞后性，从25到90 $^ ext{°C}$的热漂移最小，并在15,000个循环中保持稳定运行。单层、双层和四层传感器的接近检测范围分别为60 mm、50 mm和40 mm，揭示了结构依赖的灵敏度与范围的权衡。4$ imes$4的纺织传感阵列实现了空间接触映射，而机器人臂的集成展示了实时触摸和接近检测，机器人系统的端到端延迟（从检测到机器人反应）为403 ms。结果表明，纱线结构是决定纺织集成电容传感系统测量特性的可调设计参数。

View on arXiv Download PDF AI Translation

cs.RO / 22 / 2608.14448

Control-Informed Constraint Adaptation in Minimum-Time Trajectory Planning for Autonomous Racing

基于控制信息的约束自适应在自主赛车最短时间轨迹规划中的应用

Schwehn, Ann-Kathrin, Langmann, Alexander, Piccinini, Mattia, Betz, Johannes

Abstract

Autonomous racecars operate at the limits of vehicle dynamics, where small control errors translate into safety-critical behavior and lost performance. Trajectory planners assume perfect tracking and remain blind to execution errors. To guarantee safety, trajectory planners therefore restrict themselves to conservative spatial margins, leaving usable track space untapped. To overcome these issues, we introduce a control-informed online trajectory planning framework that learns from its own execution errors. By measuring systematic tracking deviations during runtime, we dynamically adapt spatial track constraints and iteratively expand the free-space planning area. The planner remains time-optimal while compensating for accumulated execution errors. This method was analyzed in a high-fidelity closed-loop simulation environment with autonomous racecars. The results demonstrate that our approach reduces lap time by 1.8\,s without increasing computational burden, maintaining a median runtime of 25 ms. Our finding indicates that feeding control-induced deviations back into the planning layer unlocks performance previously inaccessible to modular architectures and enables autonomous vehicles to exploit track limits systematically.

Chinese Translation

自主赛车在车辆动力学的极限状态下运行，微小的控制误差会导致安全关键行为和性能损失。轨迹规划器假设完美跟踪，并对执行错误视而不见。为了保证安全，轨迹规划器因此限制自己在保守的空间边界内，导致可用的赛道空间未被充分利用。为了解决这些问题，我们提出了一种基于控制信息的在线轨迹规划框架，该框架能够从自身的执行错误中学习。通过在运行时测量系统性的跟踪偏差，我们动态调整空间轨道约束，并迭代扩展自由空间规划区域。该规划器在补偿累积的执行错误的同时，仍然保持时间最优性。该方法在一个高保真闭环仿真环境中进行了分析，使用自主赛车进行测试。结果表明，我们的方法在不增加计算负担的情况下，将圈速减少了1.8秒，保持了25毫秒的中位运行时间。我们的研究发现，将控制引起的偏差反馈到规划层中，可以释放出以前对模块化架构不可及的性能，使自主车辆能够系统地利用赛道极限。

View on arXiv Download PDF AI Translation

cs.RO / 23 / 2608.14462

THRIVE: Therapeutic Humanoid Robot In Virtual Environment

THRIVE：虚拟环境中的治疗类人机器人

Xu, Jin, Chen, Yu-Ping, Howard, Ayanna

Abstract

This paper presents THRIVE (Therapeutic Humanoid Robot In Virtual Environment), an at-home rehabilitation platform that integrates a suite of virtual-reality upper-body rehabilitation games, a real-time camera-based motion-tracking system, and a socially interactive robot therapist. The system is designed for therapy and intervention in children with upper-limb motor impairments, which can be improved through consistent, task-specific practice. THRIVE features a set of newly designed, engaging games that target functional reaching, grasping, and object-manipulation movements through customizable popping, hitting, catching, and grabbing tasks, while the camera-based tracking system captures the child's kinematic performance during play. A robot therapist - deployable either as a physical robotic coach or as a remote-presence virtual agent - delivers adaptive, dynamic feedback to motivate the child and guide their movements toward therapeutic goals. THRIVE decouples the therapeutic games from the robot embodiment, extending the platform to support various embodiments and different robots within one modular system. This robot-agnostic design makes THRIVE affordable, scalable, and readily adaptable for sustained use in the home, offering a practical pathway to more consistent and engaging upper-limb therapy for children with motor function impairments.

Chinese Translation

本文介绍了THRIVE（虚拟环境中的治疗类人机器人），这是一个家庭康复平台，集成了一系列虚拟现实上肢康复游戏、实时基于摄像头的运动追踪系统和社交互动机器人治疗师。该系统旨在为上肢运动障碍的儿童提供治疗和干预，这些障碍可以通过持续的、特定任务的练习得到改善。THRIVE具有一套新设计的、引人入胜的游戏，针对功能性伸手、抓取和物体操作动作，通过可定制的弹出、击打、接球和抓取任务进行训练，同时基于摄像头的追踪系统在游戏过程中捕捉儿童的运动表现。机器人治疗师可以作为物理机器人教练或远程虚拟代理部署，提供适应性和动态反馈，以激励儿童并指导他们的动作朝向治疗目标。THRIVE将治疗游戏与机器人实体解耦，扩展了该平台以支持多种实体和不同的机器人在一个模块化系统中。这种与机器人无关的设计使THRIVE具有经济性、可扩展性，并且易于适应家庭中的持续使用，为上肢功能障碍儿童提供了更一致和更具吸引力的治疗途径。

View on arXiv Download PDF AI Translation

cs.RO / 24 / 2608.14466

Expected Free Energy-based Informative Path Planning for Robotic Mars Exploration

基于期望自由能的机器人火星探索信息路径规划

Meera, Ajith Anil, Lanillos, Pablo, Kouw, Wouter

Abstract

An autonomous robot efficiently exploring an unknown environment, such as looking for water sources on Mars, faces two simultaneous demands: building an accurate information map while quickly finding the regions of greatest value, and paying for every meter of travel and the cost of every measurement it takes. Classical information-seeking and reward-seeking criteria address only one of these objectives at a time. Here, we propose Expected Free Energy (EFE), the principled action-selection objective from active inference, as a unifying criterion for budgeted robotic informative path planning. Maintaining a Gaussian-process belief over the information field, our agent plans continuous trajectories that minimize expected free energy under hard path-length constraints. The results from multiple realizations show that EFE-based planning yields accurate posterior maps and locates the highest-value regions simultaneously, outperforming information-theoretic baselines under the same settings. In robotic exploration, these unified, easy-to-tune principled information-gathering strategies facilitate autonomous deployment while enforcing efficiency and resource constraints.

Chinese Translation

自主机器人在未知环境中高效探索，例如寻找火星上的水源，面临两个同时的需求：一方面需要构建准确的信息地图，同时快速找到最有价值的区域；另一方面则需要考虑每米行进的成本以及每次测量的费用。经典的信息寻求和奖励寻求标准只能单独解决其中一个目标。在此，我们提出期望自由能（Expected Free Energy, EFE），作为主动推理中的原则性行动选择目标，作为预算机器人信息路径规划的统一标准。通过对信息场保持高斯过程信念，我们的智能体规划连续轨迹，以在严格的路径长度约束下最小化期望自由能。多次实现的结果表明，基于EFE的规划能够同时产生准确的后验地图并定位最高价值区域，在相同设置下优于信息理论基准。在机器人探索中，这些统一且易于调整的原则性信息收集策略促进了自主部署，同时确保了效率和资源约束。

View on arXiv Download PDF AI Translation

cs.RO / 25 / 2608.14481

Ensuring Safe Physical AI in Urban Mobility via Hazard-Informed Synthesized Envelopes

通过基于危险的合成包确保城市移动中的安全物理人工智能

Odinokov, Alexei, Yavorskiy, Rostislav

Abstract

As heterogeneous robotic systems deploy across diverse urban zones, maintaining safety amid complex human-robot interactions remains a critical challenge. We present a unified framework that bridges systematic hazard analysis and runtime enforcement using hazard-informed safety envelopes. Rather than treating safety as a static constraint isolated within individual software modules, we introduce a cross-layer safety transformation process spanning symbolic, spatial, and dynamic world models. We show how this representation naturally interfaces with physical AI runtime harnesses to guarantee safe urban mobility.

Chinese Translation

随着异构机器人系统在多样化的城市区域部署，如何在复杂的人机交互中保持安全仍然是一个关键挑战。我们提出了一个统一框架，结合了系统性的危险分析和运行时强制执行，使用基于危险的信息安全包。我们并不将安全视为孤立于各个软件模块的静态约束，而是引入了跨层安全转换过程，涵盖符号、空间和动态世界模型。我们展示了这种表示如何自然地与物理人工智能运行时工具相接口，以保证安全的城市移动。

View on arXiv Download PDF AI Translation

cs.RO / 26 / 2608.14531

Spatiotemporal Tube-Based Safety-Certificate for Autonomous Navigation of Articulated Vehicles

基于时空管道的自主导航关节车辆安全证书

Faruqui, Mohd. Faizuddin, Das, Ratnangshu, L, Ravi Kumar, Jagtap, Pushpak

Abstract

Articulated vehicles are the workhorses of freight transportation, and their autonomous navigation is challenging. Their physical characteristics and motion constraints pose significant challenges in manoeuvring these vehicles on narrow routes. This paper presents a spatiotemporal tube-based approach to plan autonomous navigation of vehicles like tractor semi-trailers, truck/ tractor trailers, towing Automated Guided Vehicles (AGVs), and road trains. This planning approach provides a certified path plan for the truck or tractor, ensuring that the towed series of trailers always remains within the road corridor, limited by permissible corrections. The planning leverages the kinematics of the linked elements along with sway constraints to arrive at a safe tube for the actuated prime mover. We modify the spatiotemporal tube using permissible corrections to provide a route safety certificate to the vehicle for the given route. The proposed planning method is verified on a truck-trailer navigation simulation for a complex route.

Chinese Translation

关节车辆是货物运输的主力军，其自主导航面临诸多挑战。它们的物理特性和运动约束在狭窄路线上的操控中带来了显著的困难。本文提出了一种基于时空管道的方法，用于规划像拖拉机半挂车、卡车/拖拉机拖车、拖曳自动导引车（AGVs）和公路列车等车辆的自主导航。该规划方法为卡车或拖拉机提供了经过认证的路径规划，确保被拖动的系列拖车始终保持在道路走廊内，并受到允许的修正限制。该规划利用了连接元素的运动学以及摆动约束，以确定驱动主机的安全管道。我们通过允许的修正对时空管道进行修改，为车辆在给定路线上的行驶提供安全证书。所提出的规划方法在复杂路线的卡车-拖车导航仿真中得到了验证。

View on arXiv Download PDF AI Translation

计算机视觉 (Computer Vision)

cs.CV / 1 / 2608.13660

What to Preserve, Where to Adapt: A Depth-Wise Analysis of Forgetting in Continual Gynecological Image Segmentation

保留什么，适应哪里：持续妇科图像分割中遗忘的深度分析

Saqib, Amal, Saleem, Tausifa Jan, Saeed, Numan, Yaqub, Mohammad

Abstract

Medical image segmentation models are typically trained under the assumption that all data are available simultaneously. However, in clinical practice, datasets often arrive sequentially, requiring models to adapt continuously to evolving data distributions. We study this problem in gynecological image segmentation, where substantial heterogeneity across imaging modalities, anatomical structures, and annotation protocols creates a particularly challenging continual learning setting. Under these large distribution shifts, existing continual learning methods struggle to preserve previously learned knowledge, leading to catastrophic forgetting. To better understand forgetting in this setting, we investigate how different encoder--decoder regions influence segmentation performance and forgetting during continual gynecological segmentation. Through block-wise ablation analysis, we observe that ablating early encoder and late decoder regions results in the largest performance degradation, indicating that segmentation performance depends unevenly across the network hierarchy. Using controlled adaptation experiments, we further show that forgetting remains limited when updates are restricted to bottleneck-adjacent regions, but increases sharply once shallower encoders and decoders become trainable, even when only a small subset of parameters is updated. These findings suggest that forgetting in the encoder-decoder architecture is strongly influenced by where updates occur across network depth during continual learning. Full code and analysis pipelines will be made publicly available upon acceptance.

Chinese Translation

医学图像分割模型通常是在假设所有数据同时可用的情况下进行训练的。然而，在临床实践中，数据集往往是顺序到达的，这要求模型不断适应不断变化的数据分布。我们研究了这一问题在妇科图像分割中的表现，其中成像模态、解剖结构和标注协议之间的显著异质性使得持续学习的环境特别具有挑战性。在这些大规模分布转变下，现有的持续学习方法难以保留先前学习的知识，导致灾难性遗忘。为了更好地理解在这一环境中的遗忘现象，我们调查了不同编码器-解码器区域如何影响分割性能和持续妇科分割过程中的遗忘。通过逐块消融分析，我们观察到消融早期编码器和晚期解码器区域会导致性能显著下降，表明分割性能在网络层次结构中分布不均。通过控制适应实验，我们进一步表明，当更新限制在瓶颈邻近区域时，遗忘保持在有限范围内，但一旦较浅的编码器和解码器变得可训练，遗忘现象会急剧增加，即使仅更新少量参数。这些发现表明，在持续学习过程中，编码器-解码器架构中的遗忘现象受到网络深度中更新位置的强烈影响。完整的代码和分析流程将在接受后公开。

View on arXiv Download PDF AI Translation

cs.CV / 2 / 2608.13669

Multiphase-Diff: Diffusion-Based Generative Modeling for High-Contrast Multiphase Physical Systems with Sharp Interfaces

多相扩散：基于扩散的生成建模用于具有锐利界面的高对比度多相物理系统

Huang, Yining, Liang, Zhenyu

Abstract

Physics-constrained diffusion for high-contrast, sharp-interface multiphase fields faces three coupled difficulties. At coefficient jumps, expanded pointwise strong-form PDE residuals contain singular gradient terms that can penalize physical interfaces. Under extreme contrast, low-magnitude phases may fall below the diffusion noise floor and be erased, misscaled, or generated with negative coefficients, while a global likelihood scale allows high-magnitude phases to dominate supervision. We therefore propose Multiphase-Diff, which makes three corresponding contributions: (i) a conservative flux residual that avoids differentiating discontinuous coefficients and enforces discrete conservation; (ii) an analytic bijective representation that maps low-amplitude signals to order-one latent scales and guarantees coefficient positivity through exponential decoding; and (iii) a Jacobi-preconditioned likelihood that normalizes local residual scales for balanced supervision. Experiments on three complementary multiphase benchmarks demonstrate the superiority of Multiphase-Diff over seven baselines in both physical and distributional fidelity and its robustness across phase contrasts and compositions, establishing its effectiveness for scientific sample generation in this challenging regime.

Chinese Translation

受物理约束的扩散在高对比度、锐利界面的多相场中面临三种耦合困难。在系数跳跃处，扩展的逐点强形式偏微分方程残差包含奇异梯度项，这可能会惩罚物理界面。在极端对比下，低幅度相可能低于扩散噪声底线而被抹去、缩放错误或以负系数生成，而全局似然尺度则使高幅度相主导监督。因此，我们提出了Multiphase-Diff，做出了三项相应的贡献：（i）一种保守的通量残差，避免对不连续系数进行微分并强制执行离散守恒；（ii）一种解析双射表示，将低幅度信号映射到一阶潜在尺度，并通过指数解码保证系数的正性；（iii）一种雅可比预处理的似然，规范化局部残差尺度以实现平衡监督。在三个互补的多相基准测试上的实验表明，Multiphase-Diff在物理和分布保真度方面优于七个基线，并且在相对比和组成方面表现出鲁棒性，确立了其在这一挑战性领域进行科学样本生成的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 3 / 2608.13671

PROVE: Training-Free Prompt Recovery using Verifiable Evidence

PROVE：基于可验证证据的无训练提示恢复

Mallick, Rupayan, Khoshnoodi, Mahsa, Bargal, Sarah Adel

Abstract

Modern text-to-image models can generate highly realistic images from natural-language prompts, while recent advances in prompt inversion have made it increasingly feasible to recover those prompts from generated outputs, raising new concerns for copyright protection and content ownership. As prompt marketplaces emerge, recovered prompts can enable both the unauthorized reproduction and redistribution of copyrighted creative works, and the exposure of the prompts that encode an artist's creative recipe in AI-generated content. Existing prompt inversion methods rely on gradient-based optimization, autoregressive captioning, or reinforcement learning. However, optimization-based methods often produce unreadable prompts, captioning methods hallucinate unverified details, and RL-based approaches frequently overfit to specific generators while introducing evaluation circularity. We introduce PROVE (Prompt Recovery with Verified Evidence), a training-free, black-box prompt inversion attack that reconstructs prompts by composing verifiable scene descriptions rather than optimizing token sequences, targeting both original copyrighted works and AI-generated content. The resulting prompts are fully auditable, with every recovered claim grounded in explicit image evidence, and are formalized through a precision-constrained recall maximization objective. Across MS-COCO, Flickr30K, and Lexica, using state-of-the-art text-to-image generators, PROVE consistently outperforms optimization, captioning, and RL-based baselines on image similarity (DINO, LPIPS) and text-image alignment (CLIP), without any training, generator access, or fine-tuning, demonstrating a stronger and more practical prompt inversion attack.

Chinese Translation

现代文本到图像模型能够根据自然语言提示生成高度逼真的图像，而最近在提示反演方面的进展使得从生成的输出中恢复这些提示变得越来越可行，这引发了对版权保护和内容所有权的新关注。随着提示市场的出现，恢复的提示可能会导致未经授权的复制和再分发受版权保护的创作作品，以及暴露编码艺术家创作配方的提示在AI生成内容中。现有的提示反演方法依赖于基于梯度的优化、自回归字幕生成或强化学习。然而，基于优化的方法往往产生难以阅读的提示，字幕生成方法则会幻觉出未经验证的细节，而基于强化学习的方法常常对特定生成器过拟合，同时引入评估的循环性。我们提出了PROVE（基于验证证据的提示恢复），这是一种无训练的黑箱提示反演攻击，通过组合可验证的场景描述而不是优化令牌序列来重建提示，针对原始受版权保护的作品和AI生成的内容。生成的提示是完全可审计的，每个恢复的声明都基于明确的图像证据，并通过精度约束的召回最大化目标进行形式化。在MS-COCO、Flickr30K和Lexica上，使用最先进的文本到图像生成器，PROVE在图像相似性（DINO，LPIPS）和文本-图像对齐（CLIP）方面始终优于基于优化、字幕生成和强化学习的基线，且无需任何训练、生成器访问或微调，展示了更强大和更实用的提示反演攻击。

View on arXiv Download PDF AI Translation

cs.CV / 4 / 2608.13690

MedPlex: Deep Vision-Language Co-Adaptation for Clinically Grounded Medical Segmentation

MedPlex：用于临床基础医学分割的深度视觉-语言共适应

Sultan, Rafi Ibn, Zhu, Hui, Li, Chengyin, Zhu, Dongxiao

Abstract

Medical image segmentation is still largely treated as a vision-only problem, although clinical interpretation often relies on textual knowledge of anatomy, location, appearance, and surrounding context. Existing text-guided segmentation methods within the Vision-Language Model (VLM) paradigm often use language only as a late conditioning signal, limiting its influence on visual representation learning. We introduce MedPlex (Medical Plexus of Vision and Language), an end-to-end VLM framework that makes text guidance a continuous, clinically grounded component of segmentation learning. Through Bi-Fusion (Bidirectional Fusion), visual and textual representations evolve jointly across the encoding hierarchy. MedPlex further introduces class-level and region-level concept alignment to organize the shared representation at complementary granularities. Class-level alignment anchors each anatomical target to an aggregated clinical concept profile, while region-level alignment preserves individual concepts, such as shape, location, appearance, and texture, through class-specific visual evidence. In this way, language provides structured supervision throughout the encoder rather than serving only as a late-stage cue. MedPlex achieves state-of-the-art performance across CT and MR benchmarks for multi-organ, cardiac substructure, and tumor segmentation, including settings with real free-text clinical supervision. Code: https://github.com/rafiibnsultan/MedPlex.

Chinese Translation

医学图像分割仍然主要被视为一个仅涉及视觉的问题，尽管临床解读往往依赖于解剖学、位置、外观和周围环境的文本知识。在视觉-语言模型（Vision-Language Model, VLM）范式下，现有的文本引导分割方法通常仅将语言作为后期条件信号，从而限制了其对视觉表征学习的影响。我们提出了MedPlex（医学视觉与语言的联结），这是一个端到端的VLM框架，使文本引导成为分割学习的一个持续的、基于临床的组成部分。通过双向融合（Bi-Fusion），视觉和文本表征在编码层次中共同演变。MedPlex进一步引入了类级和区域级概念对齐，以在互补的粒度上组织共享表征。类级对齐将每个解剖目标锚定到一个聚合的临床概念轮廓，而区域级对齐则通过类特定的视觉证据保留个体概念，如形状、位置、外观和纹理。通过这种方式，语言在整个编码器中提供结构化的监督，而不仅仅是作为后期提示。MedPlex在多脏器、心脏亚结构和肿瘤分割的CT和MR基准测试中实现了最先进的性能，包括具有真实自由文本临床监督的设置。代码链接： https://github.com/rafiibnsultan/MedPlex。

View on arXiv Download PDF AI Translation

cs.CV / 5 / 2608.13729

Limitations of Synthetic Data Generation in Specialized Data-Scarce Domains

合成数据生成在专业数据稀缺领域的局限性

Zhang, Edward, Hussing, Marcel, Tandon, Tanay, Kannapiran, Shenbagaraj, Hughes, Jason, Wang, Youkang, Caswell, Joshua, Kratimenos, Agelos, Li, Yi Fan, Manoj, Milan, Sanchez, Ethan, Shrote, Sumukh, Taylor, Camillo Jose, Hashimoto, Daniel A., Eaton, Eric

Abstract

Advances in diffusion-based generative models have motivated the use of synthetic image generation to alleviate data scarcity in vision tasks. While this strategy has shown promise in natural image benchmarks such as ImageNet, its effectiveness in sparse, high-variance real-world domains remains unclear. In this work, we focus on domains where images differ substantially from common image datasets and additional data are expensive to obtain. Against non-generative data augmentation baselines, we evaluate the downstream classifier performance improvements yielded by two schools of generative sparse data extension: distribution modeling and sample perturbation. Across five trauma classification tasks using subject-wise train--validation splits, no generative approach consistently outperforms a strong non-generative baseline. Feature-space analysis reveals recurring failure modes: memorization or collapse, distributional drift, and generation of visually plausible but simplified canonical instances that are easier to classify than real data.

Chinese Translation

基于扩散的生成模型的进展促使了合成图像生成的使用，以缓解视觉任务中的数据稀缺问题。尽管这一策略在自然图像基准（如ImageNet）中显示出前景，但其在稀疏、高方差的真实世界领域中的有效性仍不明确。在本研究中，我们关注于图像与常见图像数据集显著不同且额外数据获取成本高昂的领域。我们评估了两种生成稀疏数据扩展方法（分布建模和样本扰动）在与非生成数据增强基线的对比中，对下游分类器性能提升的影响。在五个创伤分类任务中，使用受试者划分的训练-验证集，没有任何生成方法在性能上持续优于强大的非生成基线。特征空间分析揭示了反复出现的失败模式：记忆或崩溃、分布漂移，以及生成视觉上似是而非但简化的典型实例，这些实例比真实数据更容易分类。

View on arXiv Download PDF AI Translation

cs.CV / 6 / 2608.13751

CAST: Closed-form Analytic Semantic Transfer for Zero-Shot Classifier Extension

CAST：用于零样本分类器扩展的闭式解析语义转移

Heyden, William, Ullah, Habib, Siddiqui, Muhammad Salman, Machot, Fadi Al

Abstract

Large pre-trained models have become foundational components of modern machine learning systems. Yet adapting these models to novel categories typically requires examples from the target distribution. In many domains, however, such data are unavailable. Zero-shot learning (ZSL) permits recognition under these limitations through relying on auxiliary semantic information such as textual descriptions. We introduce CAST (Closed-form Analytic Semantic Transfer), a training-free, image-free framework for extending a pre-trained classifier to previously unseen classes through weight injection. We provide a theoretical foundation for CAST and derive a finite-sample error decomposition that identifies the \emph{semantic extrapolation residual} $\rho_u$. The residual is a computable, model-agnostic measure and provides a principled criterion for dataset curation and benchmark design. Experiments on standard zero-shot learning benchmarks demonstrate that CAST matches or exceeds existing image-free approaches and approaches the performance of few-shot adaptation methods, while requiring neither iterative optimization nor examples from the target distribution.

Chinese Translation

大型预训练模型已成为现代机器学习系统的基础组成部分。然而，将这些模型适应于新类别通常需要来自目标分布的示例。然而，在许多领域，这类数据是不可用的。零样本学习（ZSL）允许在这些限制下进行识别，依赖于辅助语义信息，如文本描述。我们提出了CAST（闭式解析语义转移），这是一个无训练、无图像的框架，通过权重注入将预训练分类器扩展到以前未见过的类别。我们为CAST提供了理论基础，并推导出一个有限样本误差分解，识别出 extit{语义外推残差} $ ho_u$。该残差是一个可计算的、模型无关的度量，为数据集策划和基准设计提供了原则性标准。在标准零样本学习基准上的实验表明，CAST的性能与现有的无图像方法相当或更优，并接近少样本适应方法的性能，同时不需要迭代优化或来自目标分布的示例。

View on arXiv Download PDF AI Translation

cs.CV / 7 / 2608.13766

ChartProbe: A Diagnostic Study on Visual Reasoning through Perception, Grounding, and Simple Reasoning

ChartProbe：通过感知、基础和简单推理进行视觉推理的诊断研究

Khoshnoodi, Mahsa, Bargal, Sarah Adel

Abstract

Vision-language models (VLMs) remain unreliable on chart questions that require reasoning over visual quantities, and this weakness is usually attributed to a reasoning deficit and addressed with more reasoning supervision. We ask whether the difficulty lies in reasoning itself, or in the simpler skills that reasoning operates on: reading the plotted elements (\emph{perception}), locating them and binding them to their labels (\emph{grounding}), and performing single-step computations such as ranking, totals, and differences (\emph{simple reasoning}). We introduce \textbf{ChartProbe}, a diagnostic framework whose probes are generated directly from the code that renders each chart, so every gold answer is exact by construction, needs no human annotation, and attributes each failure to a single skill. ChartProbe enables an intervention prior work does not attempt: instead of synthesizing complex-reasoning data, we withhold complex questions and reasoning traces entirely, fine-tune on one simple skill at a time, and measure transfer to held-out complex-reasoning questions. Across three open-weight VLMs, supervising the simpler skills alone produces large gains on complex-reasoning questions the model never trained on: where these skills are weak and the model can be taught to read the image, training them recovers much of complex reasoning at no reasoning-data cost. The gains hold across three out-of-distribution settings: an unseen chart type (pie charts), a human-written benchmark disjoint from our images and templates (ChartQA), and a non-chart visual domain (CLEVR). Complex visual reasoning can therefore improve without complex-reasoning supervision.

Chinese Translation

视觉语言模型（VLMs）在需要对视觉量进行推理的图表问题上仍然不可靠，这一弱点通常归因于推理能力不足，并通过更多的推理监督来解决。我们提出一个问题：困难是出在推理本身，还是在推理所依赖的更简单技能上：读取绘制元素（ extit{感知}）、定位这些元素并将其与标签绑定（ extit{基础}），以及执行单步计算，如排名、总数和差异（ extit{简单推理}）。我们引入了 extbf{ChartProbe}，这是一个诊断框架，其探针直接从渲染每个图表的代码生成，因此每个金标准答案都是通过构造精确生成的，无需人工注释，并且将每个失败归因于单一技能。ChartProbe 使得一种干预成为可能，而以往的工作并未尝试：我们不合成复杂推理数据，而是完全保留复杂问题和推理轨迹，逐一对单一简单技能进行微调，并测量其对保留的复杂推理问题的迁移。在三个开放权重的 VLMs 中，仅对简单技能进行监督就能在模型从未训练过的复杂推理问题上产生显著提升：当这些技能较弱且模型能够学习读取图像时，训练它们可以在没有推理数据成本的情况下恢复大部分复杂推理。这些提升在三个分布外的设置中保持有效：一个未见过的图表类型（饼图）、一个与我们的图像和模板不相交的人类编写基准（ChartQA），以及一个非图表视觉领域（CLEVR）。因此，复杂视觉推理可以在没有复杂推理监督的情况下得到改善。

View on arXiv Download PDF AI Translation

cs.CV / 8 / 2608.13769

Kolmogorov-Arnold Networks for Spatially Independent Multispectral Land Classification

用于空间独立多光谱土地分类的科尔莫戈罗夫-阿诺德网络

Bauer, Katherine L., Harkonen, Teemu, Sarkka, Simo, Sanchez-Azofeifa, Arturo

Abstract

Land classification from satellite imagery is important for land management, environmental monitoring, and urban planning. Machine learning methods such as random forests and multilayer perceptrons have shown strong performance on multispectral data, while the Kolmogorov-Arnold network has emerged as an alternative architecture with compact model structures. This study evaluates the Kolmogorov-Arnold network for land classification using Landsat 8 imagery and compares it with random forest and multilayer perceptron models. The models were trained and tested on data from Edmonton, Alberta and evaluated on an independent dataset from Calgary, Alberta across five land classes: agriculture, urban, water, forest, and bare ground. For the Calgary dataset, the Kolmogorov-Arnold network matched the accuracy of the random forest and outperformed the multilayer perceptron, while requiring substantially fewer trainable parameters and providing greater interpretability.

Chinese Translation

从卫星影像进行土地分类对土地管理、环境监测和城市规划至关重要。机器学习方法如随机森林和多层感知器在多光谱数据上表现出色，而科尔莫戈罗夫-阿诺德网络则作为一种具有紧凑模型结构的替代架构出现。本研究评估了科尔莫戈罗夫-阿诺德网络在使用Landsat 8影像进行土地分类的效果，并将其与随机森林和多层感知器模型进行了比较。模型在阿尔伯塔省埃德蒙顿的数据上进行了训练和测试，并在来自阿尔伯塔省卡尔加里的独立数据集上进行了评估，涵盖五个土地类别：农业、城市、水体、森林和裸地。在卡尔加里数据集中，科尔莫戈罗夫-阿诺德网络的准确性与随机森林相当，并且优于多层感知器，同时所需的可训练参数显著更少，并提供了更好的可解释性。

View on arXiv Download PDF AI Translation

cs.CV / 9 / 2608.13783

Doomed to Re-Annotate, Forever: The ImageNet Story

注定要永远重新标注：ImageNet的故事

Volkov, Illia, Kisel, Nikita, Mishkina, Tetiana, Janouskova, Klara, Matas, Jiri

Abstract

Top-1 accuracy on ImageNet-1k remains the most commonly reported metric in visual recognition. Quality issues with the dataset have been repeatedly reported, yet the original 2012 noisy labels are still predominantly used. The paper presents a comprehensive effort, which goes well beyond prior correction attempts, towards obtaining accurate and complete ImageNet-1k validation set annotations. The result, ReImageNet, includes multilabel correction, object localization, revised class definitions, and semantic attributes (text-recognition, rendition, reflection, crowd, dominant). The reannotation reveals that approximately 12% of the original ImageNet-1k labels are incorrect, 33.3% of images are multilabel and 3.8% contain no object from an ImageNet-1k class. With the new labels, top-1 accuracy increases by up to 1.2% for supervised models and by 5-6% for MLLMs. We argue that annotation at ImageNet scale cannot realistically be completed in one pass, as errors and definitional issues are discovered only through annotating, and we build our pipeline around repeated refinement and error checking. We observed that human and LLM collaboration with appropriate tooling represents the current quality ceiling for annotation at this scale. ImageNet-1k issues propagate into its derivative test sets, indicating that the problem is structural rather than specific to any single benchmark. All annotations, class definitions, guidelines, and analysis code have been publicly released. Project page: https://vrg.fel.cvut.cz/reimagenet Annotations: https://huggingface.co/datasets/vrg-prague/ReImageNet Code: https://github.com/klarajanouskova/ImageNet

Chinese Translation

ImageNet-1k上的Top-1准确率仍然是视觉识别中最常报告的指标。关于数据集的质量问题已经多次被报道，但2012年原始的嘈杂标签仍然被广泛使用。本文提出了一项全面的努力，远超以往的修正尝试，旨在获得准确且完整的ImageNet-1k验证集标注。结果，ReImageNet，包含多标签修正、物体定位、修订的类别定义和语义属性（文本识别、表现、反射、拥挤、主导）。重新标注揭示大约12%的原始ImageNet-1k标签是错误的，33.3%的图像是多标签的，3.8%不包含任何来自ImageNet-1k类别的物体。使用新标签后，监督模型的Top-1准确率提高了最多1.2%，而MLLMs则提高了5-6%。我们认为，在ImageNet规模下，标注不可能在一次性完成，因为错误和定义问题只能通过标注过程中发现，我们围绕重复的精细化和错误检查构建了我们的流程。我们观察到，适当工具下的人类与LLM的协作代表了这一规模标注的当前质量上限。ImageNet-1k的问题会传播到其衍生的测试集，表明该问题是结构性的，而非特定于任何单一基准。所有标注、类别定义、指南和分析代码均已公开发布。项目页面：https://vrg.fel.cvut.cz/reimagenet 标注：https://huggingface.co/datasets/vrg-prague/ReImageNet 代码：https://github.com/klarajanouskova/ImageNet

View on arXiv Download PDF AI Translation

cs.CV / 10 / 2608.13858

Face Re-morphing: Differential Morphing Attack Detection via Feature-Space Similarity Changes

面部再变形：通过特征空间相似性变化进行差异性变形攻击检测

Jin, Jie, Nishigaki, Masakatsu, Ohki, Tetsushi

Abstract

Face morphing attacks pose a serious threat to face recognition systems because a single morphed document image can be matched to multiple contributors. Differential morphing attack detection (D-MAD) addresses this threat by comparing a document image with a trusted live image, but existing methods often rely on static feature differences, constituent-face reconstruction, or multi-cue fusion. This paper proposes Face Re-morphing, a D-MAD method that uses the feature-space response to an additional morphing operation as a detection cue. Given a document image and a trusted live image, the proposed method generates a re-morphed image and uses the change between the document--live and live--re-morphed cosine similarities as the detection score. Experiments on FRLL-Morphs and FEI Morph show that the proposed cue is effective across different morphing conditions, re-morphing methods, and face recognition models. Comparisons with existing methods show favorable results on AMSL and indicate that the proposed method performs well under the Criminal condition on FEI Morph Version~1, particularly when using MorDIFF. These results indicate that re-morphing-induced similarity change provides a complementary cue for D-MAD.

Chinese Translation

面部变形攻击对面部识别系统构成严重威胁，因为单个变形文档图像可以与多个贡献者匹配。差异性变形攻击检测（D-MAD）通过将文档图像与可信的实时图像进行比较来应对这一威胁，但现有方法往往依赖于静态特征差异、组成面重建或多线索融合。本文提出了面部再变形（Face Re-morphing），一种D-MAD方法，利用对额外变形操作的特征空间响应作为检测线索。给定文档图像和可信的实时图像，所提方法生成一个再变形图像，并使用文档-实时图像与实时-再变形图像之间的余弦相似性变化作为检测得分。在FRLL-Morphs和FEI Morph的实验表明，所提线索在不同的变形条件、再变形方法和面部识别模型下均有效。与现有方法的比较在AMSL上显示出良好的结果，并表明所提方法在FEI Morph Version~1的犯罪条件下表现良好，特别是在使用MorDIFF时。这些结果表明，再变形引起的相似性变化为D-MAD提供了一个补充线索。

View on arXiv Download PDF AI Translation

cs.CV / 11 / 2608.13861

XSA-MAD: Cross-modal Semantic Alignment for Morphing Attack Detection

XSA-MAD：用于变形攻击检测的跨模态语义对齐

Jin, Jie, Tokumasu, Mahiro, Makino, Yu, Nishigaki, Masakatsu, Ohki, Tetsushi

Abstract

Morphing attacks pose a serious threat to face recognition systems. However, existing image-based morphing attack detection (MAD) methods often generalize poorly to unseen generation techniques because they rely solely on visual cues. We propose XSA-MAD, a CLIP-based multimodal framework that explicitly models semantic inconsistencies between bona-fide and morphed faces. Morphing concepts are decomposed into four interpretable attributes, including identity, facial geometry, texture, and consistency, and are encoded as structured and attribute-aware textual representations. The image encoder is progressively aligned with this discriminative textual space, resulting in a unified semantic representation that captures generation-invariant and concept-level discrepancies between bona-fide and morph images. Experiments on MAD22 and MorDIFF, following training on SMDD, demonstrate strong generalization across diverse morphing principles. In particular, XSA-MAD achieves an equal error rate of 2.92% on GAN-based morphs and consistently outperforms existing methods under high-fidelity generative attacks.

Chinese Translation

变形攻击对人脸识别系统构成了严重威胁。然而，现有的基于图像的变形攻击检测（MAD）方法往往对未见过的生成技术泛化能力较差，因为它们仅依赖于视觉线索。我们提出了XSA-MAD，一个基于CLIP的多模态框架，明确建模真实人脸与变形人脸之间的语义不一致性。变形概念被分解为四个可解释的属性，包括身份、面部几何、纹理和一致性，并被编码为结构化和属性感知的文本表示。图像编码器逐步与这一判别性文本空间对齐，形成统一的语义表示，捕捉真实人脸与变形图像之间的生成不变性和概念层面的差异。在SMDD上训练后，在MAD22和MorDIFF上的实验表明，XSA-MAD在多样化的变形原则下具有强大的泛化能力。特别是，XSA-MAD在基于GAN的变形上实现了2.92%的等错误率，并在高保真生成攻击下始终优于现有方法。

View on arXiv Download PDF AI Translation

cs.CV / 12 / 2608.13865

Attention Capture Is Not Detection: A Two-Stage Account of How Humans Miss Localized AI Image Edits

注意捕捉并非检测：人类如何错过局部AI图像编辑的双阶段解释

Deng, Chiao-Chieh

Abstract

As AI-generated image edits proliferate, the platforms meant to curb the resulting disinformation treat detectability as a single, undifferentiated property: an edit either gets a warning or it does not. We show this is the wrong model. Across a controlled eye-tracking study ($N=59$, Latin-square design, four conditions crossing edit area and semantic plausibility), a mixed-effects analysis reveals that whether an edit is noticed and whether it is correctly judged as fake are dissociable stages, governed by different factors: edit area drives attention capture ($p<0.001$) while semantic plausibility drives judgment accuracy and look-but-fail-to-see (LBFS) error rates ($p<0.001$). This dissociation survives correction for multiple comparisons; a secondary interaction between the two factors does not. This two-stage account extends a long-standing distinction in visual attention research (between pre-attentive capture and effortful recognition) into the new domain of AI-edit detectability. We then test whether a generative eye-movement model can computationally operationalize the attention-capture stage: a Transformer trained to generate scanpaths tracks per-image attention with strong discriminative power (Pearson $r=0.77$--$0.82$ across held-out stimuli) and, on the harder task of predicting LBFS incidence, modestly outperforms a two-parameter linear baseline even without access to the plausibility label ($r=0.52$ vs. $r=0.48$). We report this comparison, our ablations, and our method's limitations (a single fixed train/validation split, not leave-one-subject-out) without inflation, consistent with responsibly communicating what a machine learning system can and cannot do to help curb AI-driven disinformation.

Chinese Translation

随着AI生成的图像编辑日益增多，旨在遏制由此产生的虚假信息的平台将可检测性视为一个单一、未加区分的属性：编辑要么得到警告，要么不被警告。我们表明这一模型是错误的。在一项受控的眼动追踪研究中（$N=59$，拉丁方设计，四个条件交叉编辑区域和语义合理性），混合效应分析揭示了编辑是否被注意到以及是否被正确判断为虚假是可分离的阶段，由不同因素主导：编辑区域驱动注意捕捉（$p<0.001$），而语义合理性驱动判断准确性和看但未见（LBFS）错误率（$p<0.001$）。这种分离在多重比较校正后依然成立；而两个因素之间的次要交互作用则不成立。这一双阶段解释将视觉注意研究中长期存在的区分（在预注意捕捉和费力识别之间）扩展到AI编辑可检测性的新领域。随后，我们测试了生成的眼动模型是否可以计算上操作化注意捕捉阶段：一个经过训练的Transformer生成扫描路径，能够以强大的判别能力追踪每幅图像的注意力（在保留的刺激上，Pearson $r=0.77$--$0.82$），并且在更难的预测LBFS发生率的任务上，即使在没有访问合理性标签的情况下，仍然适度优于一个两参数线性基线（$r=0.52$ vs. $r=0.48$）。我们报告了这一比较、我们的消融实验以及我们方法的局限性（单一固定的训练/验证划分，而非逐个被试剔除），以负责任的方式传达机器学习系统能够和不能做的事情，以帮助遏制AI驱动的虚假信息。

View on arXiv Download PDF AI Translation

cs.CV / 13 / 2608.13889

Consensus-gated Multi-Agent Neural Architecture Search for Seismic Fault Segmentation

共识引导的多智能体神经架构搜索用于地震断层分割

Baig, Shehram, Mustafa, Ahmad

Abstract

Neural networks for seismic fault segmentation are often borrowed from computer vision and medical imaging domains where they train under relatively much larger labeled data resources. Optimizing their architecture under tight labeled data budgets as are common in geophysical applications is not a trivial problem. Manually designing data-optimal architectures is time-consuming while classical neural architecture search (NAS) is restricted to hand-crafted search spaces and large compute budgets. We present an agentic NAS system in which a panel of three large language models (Claude, GPT-5.1, and Gemini~2.5~Pro) debates each candidate architecture to unanimous consensus, authors the complete PyTorch implementation, cross-reviews it, and submits it to an automated validate-train-score loop with a hard 450K parameter budget, keep-or-revert lineage, and a memory of failed mechanisms. Operating on source code rather than a predefined operation menu, the search ran on a single consumer GPU and trained only eight candidates. It discovered \ours{}: a 425K-parameter encoder-decoder with a strip-pooling bottleneck, squeeze-and-excitation gating, an asymmetric one-conv decoder, and a feature-pyramid fusion neck. Trained under a protocol identical to all baselines on sections derived from the Thebe fault dataset, it attains the highest F1 (0.578) and IoU of all models tested while being the smallest, outperforming a published-capacity U-Net (31M parameters, F1 0.484), DeepLabV3-ResNet50 (39.6M, 0.516), an Attention U-Net(1.83M, 0.502). The search cost 101 LLM calls ($\sim$1.15M input / 0.39M output tokens) and roughly one GPU-day, making consensus-gated LLM panels a practical, low-cost route to domain-specific architecture discovery.

Chinese Translation

用于地震断层分割的神经网络通常借鉴自计算机视觉和医学成像领域，这些领域在相对较大的标注数据资源下进行训练。在地球物理应用中，优化其架构以适应有限的标注数据预算并非易事。手动设计数据最优架构耗时较长，而传统的神经架构搜索（NAS）受限于手工设计的搜索空间和较大的计算预算。我们提出了一种智能化的NAS系统，其中三个大型语言模型（Claude、GPT-5.1和Gemini 2.5 Pro）对每个候选架构进行辩论以达成一致共识，撰写完整的PyTorch实现，进行交叉审查，并将其提交给一个自动化的验证-训练-评分循环，该循环具有严格的45万参数预算、保留或恢复的血统以及失败机制的记忆。该搜索在源代码上运行，而不是预定义的操作菜单，仅在一台消费级GPU上训练了八个候选架构。它发现了 extit{ours}{}：一个具有425K参数的编码-解码器，采用条带池化瓶颈、挤压与激励门控、非对称的一次卷积解码器和特征金字塔融合颈部。在与所有基线相同的协议下，在来自Thebe断层数据集的部分上进行训练，获得了所有测试模型中最高的F1（0.578）和IoU，同时也是最小的，超越了已发布容量的U-Net（3100万参数，F1 0.484）、DeepLabV3-ResNet50（3960万，0.516）和Attention U-Net（183万，0.502）。该搜索耗费了101次大型语言模型调用（约115万输入/39万输出标记）和大约一天的GPU时间，使得共识引导的LLM面板成为领域特定架构发现的实用、低成本途径。

View on arXiv Download PDF AI Translation

cs.CV / 14 / 2608.13918

Beyond Control Points: Arcsecond Relative-Motion Estimation of Vision Measurement Platforms With Incomplete or Absent Control Fields

超越控制点：具有不完整或缺失控制场的视觉测量平台的弧秒级相对运动估计

Lian, Meng, Wang, Jian, Pan, Shuixin, Liu, Haibo, Zhang, Yueqiang, Guo, Yulan

Abstract

Long-range vision-based deformation monitoring is highly sensitive to motion of the camera platform. Absolute-pose differencing typically relies on dedicated control data and propagates two independent pose errors into the relative-motion estimate. We develop a control-adaptive differential framework that estimates inter-frame platform motion directly from image displacements and known 3D points. With no dedicated control point, the framework recovers platform rotation from measurement-point observations. One surveyed control point enables prior-constrained translation recovery, while two nonparallel control rays recover full 3D translation. The framework requires neither nonlinear optimization nor an initial pose estimate. Excluding control data from the rotation stage makes the rotation estimate exactly immune to contamination confined to the control field. The inherited differential formulation also cancels translational extrinsic errors exactly. We derive the rotation observability condition, a leakage bound for unmodeled translation and nonrigid point motion, and the single-point axial-prior bias law. Under 0.5-pixel image noise, attitude changes of up to 30~arcmin, and 3D point perturbations of up to 2~mm, the multi-camera estimator achieves a rotation RMSE of 2.97~arcsec and an average runtime of 0.46~ms. With one surveyed control point, its prior-constrained translation RMSE is 1.19~mm. In a bridge experiment without a stable control field, the median coordinate-wise displacement RMSE relative to total-station measurements is 0.85~mm. The estimator also maintains zero divergence under the tested 3D coordinate perturbations on public RGB-D and stereo sequences. These results establish state-of-the-art accuracy, calibration robustness, and computational efficiency among the evaluated methods.

Chinese Translation

基于长距离视觉的变形监测对相机平台的运动高度敏感。绝对姿态差分通常依赖于专用控制数据，并将两个独立的姿态误差传播到相对运动估计中。我们开发了一种控制自适应差分框架，该框架直接从图像位移和已知的三维点中估计帧间平台运动。在没有专用控制点的情况下，该框架通过测量点观测恢复平台旋转。一个被调查的控制点使得先验约束的平移恢复成为可能，而两个非平行的控制光线则恢复完整的三维平移。该框架不需要非线性优化或初始姿态估计。在旋转阶段排除控制数据使得旋转估计完全免受局限于控制场的污染。继承的差分公式也完全抵消了平移外部误差。我们推导了旋转可观测性条件、未建模平移和非刚性点运动的泄漏界限，以及单点轴向先验偏差法则。在0.5像素的图像噪声、最大30弧分的姿态变化和最大2毫米的三维点扰动下，多摄像头估计器实现了2.97弧秒的旋转均方根误差（RMSE）和0.46毫秒的平均运行时间。在一个被调查的控制点下，其先验约束的平移均方根误差为1.19毫米。在没有稳定控制场的桥梁实验中，相对于全站仪测量的中位坐标位移均方根误差为0.85毫米。该估计器在公共RGB-D和立体序列上也在测试的三维坐标扰动下保持零发散。这些结果在评估的方法中确立了最先进的精度、校准鲁棒性和计算效率。

View on arXiv Download PDF AI Translation

cs.CV / 15 / 2608.13923

OpenBelief-Nav: Evidence-Preserving Object Memory for Open-Vocabulary Language-Guided Navigation

OpenBelief-Nav：用于开放词汇语言引导导航的证据保留对象记忆

Nguyen, Dinh Tuan, Dao, Anh, Dang, Phuong Nam, Pham, Quan-Dung, Le, Tuyen P., Nguyen, Truong, Nguyen, Quan

Abstract

Open-vocabulary 3D scene graphs provide compact semantic memory for language-guided navigation, but mapped objects are often exposed through a single fused feature or committed semantic label. Such commitment can remove minority yet task-relevant hypotheses from the task-time interface. We present OpenBelief-Nav, an evidence-preserving object memory that retains observation-level phrases, reliability cues, and frame-mask provenance while maintaining separate aggregate geometric and visual representations. Semantically related phrases are consolidated into a vocabulary-independent object belief from which task-specific readouts perform fixed-vocabulary projection or free-form retrieval. On five ScanNet200 and eight Replica scenes, full-belief projection achieves mIoU scores of 0.2742 and 0.2912, compared with 0.2393 and 0.2701 for a matched early-commit readout. Across 78 HM3D-YCB navigation trials, consensus and early-commit retrieval each achieve 60/78 successes, compared with 58/78 for belief-weighted retrieval and 55/78 for DualMap. Across 20 Unitree G1 runs organized as 10 matched evaluation cases, a correction policy permitting at most two verified candidate attempts improves target-confirmation success from 6/10 to 8/10 relative to top-1-only execution. Code will be released upon acceptance at https://openbelief-nav.github.io/.

Chinese Translation

开放词汇的3D场景图为语言引导导航提供了紧凑的语义记忆，但映射的对象通常通过单一的融合特征或固定的语义标签进行表示。这种承诺可能会将少数但与任务相关的假设从任务时间界面中移除。我们提出了OpenBelief-Nav，一种证据保留对象记忆，能够保留观察级短语、可靠性线索和帧掩码来源，同时保持独立的聚合几何和视觉表示。语义相关的短语被整合为一个与词汇无关的对象信念，从中任务特定的读出可以执行固定词汇投影或自由形式检索。在五个ScanNet200和八个Replica场景中，完全信念投影的mIoU得分分别为0.2742和0.2912，而匹配的早期承诺读出的得分为0.2393和0.2701。在78个HM3D-YCB导航试验中，共识和早期承诺检索各自取得了60/78的成功率，而信念加权检索为58/78，DualMap为55/78。在20个Unitree G1的运行中，组织为10个匹配评估案例，允许最多两次验证候选尝试的修正策略将目标确认成功率从6/10提高到8/10，相较于仅执行前1个候选。代码将在接受后发布于https://openbelief-nav.github.io/。

View on arXiv Download PDF AI Translation

cs.CV / 16 / 2608.13929

RGBX-Next: Towards Realistic Generative Rendering from G-Buffers

RGBX-Next：朝着基于 G-Buffers 的真实感生成渲染迈进

Zeng, Zheng, Salvi, Marco, Wu, Lifan, Novák, Jan, Lin, Daqi, Hadadan, Saeed, Sheng, Yichen, Pottorff, Robert, Liu, Shiqiu, Ramamoorthi, Ravi, Yan, Ling-Qi, Hašan, Miloš

Abstract

Diffusion models have achieved impressive results in image, video, and streaming generation. However, compared to traditional 3D rendering, they still lack precise control over the generated output. We believe a viable path forward is to use generative models as learned renderers conditioned on traditionally rendered G-buffers. We introduce RGBX-Next, a unified generative framework for forward and inverse rendering, which allows estimating G-buffers from images, videos, and streams, and rendering realistic images, videos, and streams from G-buffers. Our key contribution is a general recipe for finetuning diffusion transformer (DiT) models into generative forward and inverse renderers. We show that the resulting models achieve high quality in both realistic generative rendering and intrinsic decomposition. We will make all our models publicly available. We believe that the design principles presented in this paper will benefit future research on controllable generative forward and inverse rendering.

Chinese Translation

扩散模型在图像、视频和流媒体生成方面取得了令人瞩目的成果。然而，与传统的 3D 渲染相比，它们在生成输出的精确控制方面仍然存在不足。我们认为，一条可行的前进路径是将生成模型作为基于传统渲染的 G-buffers 的学习渲染器。我们提出了 RGBX-Next，这是一个统一的生成框架，用于正向和逆向渲染，能够从图像、视频和流媒体中估计 G-buffers，并从 G-buffers 渲染出真实感的图像、视频和流媒体。我们的主要贡献是为微调扩散变换器（Diffusion Transformer, DiT）模型成为生成正向和逆向渲染器提供了一般性的方法。我们展示了所得到的模型在真实感生成渲染和内在分解方面都达到了高质量。我们将公开所有模型。我们相信，本文提出的设计原则将有助于未来在可控生成正向和逆向渲染方面的研究。

View on arXiv Download PDF AI Translation

cs.CV / 17 / 2608.13938

CoANeRV: Coordinate-Aware Token-Space Neural Video Representation

CoANeRV：坐标感知的令牌空间神经视频表示

Guo, Jialong, Liu, Ke, Li, Mengxuan, Bu, Jiajun, Wang, Haishuai

Abstract

Neural representations for videos (NeRV) have shown strong reconstruction fidelity by storing video-specific information in network weights. However, existing formulations typically require either costly per-video optimization or video-specific weight generation, making it difficult to scale to efficient amortized video representation. We propose CoANeRV, a coordinate-aware token-space framework that adapts the broader token-conditioned neural-field paradigm to amortized video representation. CoANeRV forms compact video tokens in one feed-forward pass and uses a shared coordinate-conditioned decoder to reconstruct continuous spatio-temporal queries, avoiding per-video decoder optimization or generation while retaining coordinate-level reconstruction flexibility. To make token-space reconstruction effective, CoANeRV introduces a coordinate-aware decoding architecture that aligns spatio-temporal queries with video tokens through axis-adaptive positional encoding and temperature-modulated cross-attention. Block-wise coordinate querying further reduces peak attention memory, making high-resolution reconstruction practical. Experiments on diverse video datasets show that CoANeRV consistently improves reconstruction quality over prior feed-forward NeRV and INR baselines, reduces peak memory compared with attention-based coordinate decoders, and provides efficient amortized encoding without per-video optimization. These results support the proposed video-specific combination of feed-forward token formation, spatio-temporal coordinate retrieval, and memory-bounded dense querying. The code is available at https://github.com/jialong2023/CoANeRV.

Chinese Translation

视频的神经表示（NeRV）通过在网络权重中存储视频特定信息展现了强大的重建保真度。然而，现有的公式通常需要昂贵的逐视频优化或视频特定的权重生成，这使得其在高效的摊销视频表示中难以扩展。我们提出了CoANeRV，一种坐标感知的令牌空间框架，它将更广泛的令牌条件神经场范式适配于摊销视频表示。CoANeRV在一次前向传播中形成紧凑的视频令牌，并使用共享的坐标条件解码器来重建连续的时空查询，避免了逐视频解码器的优化或生成，同时保留了坐标级别的重建灵活性。为了使令牌空间重建有效，CoANeRV引入了一种坐标感知的解码架构，通过轴自适应位置编码和温度调制的交叉注意力将时空查询与视频令牌对齐。块级坐标查询进一步减少了峰值注意力内存，使高分辨率重建成为可能。在多样化的视频数据集上的实验表明，CoANeRV在重建质量上始终优于先前的前向NeRV和INR基线，与基于注意力的坐标解码器相比减少了峰值内存，并提供了高效的摊销编码而无需逐视频优化。这些结果支持了所提出的基于视频的前向令牌形成、时空坐标检索和内存受限的密集查询的组合。代码可在 https://github.com/jialong2023/CoANeRV 获取。

View on arXiv Download PDF AI Translation

cs.CV / 18 / 2608.13939

CMCNet: Aligning Ultrasound Image Embeddings with Textual TI-RADS Representations for Fine-Grained Thyroid Classification

CMCNet：将超声图像嵌入与文本TI-RADS表示对齐以实现细粒度甲状腺分类

Yu, Bingxin, Wang, Xueli, Zhou, Jerry, Wang, Wenyan, Wen, Li, Huang, Lan, Feng, Xin, Zhou, Fengfeng, Li, Kewei

Abstract

Ultrasound is the primary imaging modality for assessing thyroid nodules, and the ACR TI-RADS framework standardizes diagnosis through five ultrasound feature categories that are aggregated into five risk levels (TR1-TR5). Although widely adopted in clinical practice, most deep learning approaches focus on binary malignancy classification, while multi-class prediction and explicit utilization of feature-level supervision remain underexplored, largely due to limited annotated data. In this study, we introduce the STN dataset of 600 thyroid nodules with paired transverse and longitudinal ultrasound images, bounding box annotations, and complete labels for all five TI-RADS feature categories. Following the clinical decision process, we investigate how structured feature information can guide representation learning during training while requiring only images at inference. We demonstrate that text embeddings derived from standardized feature descriptions form a stable surrogate representation for TI-RADS risk levels. Based on this observation, we propose CMCNet, which aligns image embeddings to fixed textual embeddings via a Center-Margin Contrastive Loss that simultaneously promotes intra-class compactness and inter-class separation. Experimental results show that this embedding alignment strategy is more data-efficient and robust than direct multitask learning, and consistently outperforms InfoNCE, center loss, a strong multitask baseline, and a VQA-style multimodal model, particularly in imbalanced settings. The dataset is freely available at doi: 10.5281/zenodo.19125693 and the source code is available at: https://www.healthinformaticslab.org/supp/.

Chinese Translation

超声是评估甲状腺结节的主要影像学手段，ACR TI-RADS框架通过五个超声特征类别标准化诊断，这些特征被汇总为五个风险等级（TR1-TR5）。尽管在临床实践中被广泛采用，大多数深度学习方法仍集中于二元恶性分类，而多类预测和特征级监督的明确利用仍然未得到充分探索，这在很大程度上是由于标注数据的有限性。在本研究中，我们引入了STN数据集，其中包含600个甲状腺结节的配对横向和纵向超声图像、边界框注释以及所有五个TI-RADS特征类别的完整标签。根据临床决策过程，我们研究了结构化特征信息如何在训练期间指导表示学习，同时在推理时仅需要图像。我们证明，从标准化特征描述中派生的文本嵌入形成了TI-RADS风险等级的稳定替代表示。基于这一观察，我们提出了CMCNet，它通过中心-边距对比损失将图像嵌入与固定文本嵌入对齐，同时促进类内紧凑性和类间分离。实验结果表明，这种嵌入对齐策略比直接多任务学习更具数据效率和鲁棒性，并在不平衡设置中始终优于InfoNCE、中心损失、一个强大的多任务基线和VQA风格的多模态模型。数据集可在doi: 10.5281/zenodo.19125693免费获取，源代码可在：https://www.healthinformaticslab.org/supp/获取。

View on arXiv Download PDF AI Translation

cs.CV / 19 / 2608.13949

Fast Implicit Neural Light Field Representation via Geometric Decomposition and Multi-Resolution Low-Rank Features

通过几何分解和多分辨率低秩特征实现快速隐式神经光场表示

Guo, Yao, Shi, Ligen, Sun, Shuchen, Qiu, Jun, Liu, Chang

Abstract

Implicit neural representations provide a compact and continuous way to reconstruct dense light fields from sampled ray coordinates. However, fast light field reconstruction remains challenging because a light field is a high-dimensional signal with strong spatial-angular redundancy and structured disparity variations. Directly fitting 4D ray coordinates with a neural network often requires considerable optimization time to recover both view appearance and cross-view consistency. To address this issue, this paper proposes a fast implicit light field representation based on geometric decomposition and multi-resolution low-rank features. The proposed method decomposes a 4D light field into a horizontal disparity plane, a spatial texture plane, and a vertical disparity plane. Each plane is represented by a low-rank structure that combines a low-resolution 2D grid with the element-wise product of two high-resolution 1D line features at multiple resolution levels. The fused features are decoded by a lightweight multilayer perceptron to predict RGB values. Experiments on public light field datasets show that the proposed method achieves competitive reconstruction quality while providing a better trade-off among model parameters, training time, and inference efficiency.

Chinese Translation

隐式神经表示提供了一种紧凑且连续的方式，从采样的光线坐标重建稠密光场。然而，快速光场重建仍然具有挑战性，因为光场是一个高维信号，具有强烈的空间-角度冗余和结构化的视差变化。直接用神经网络拟合4D光线坐标通常需要相当大的优化时间，以恢复视图外观和视图间一致性。为了解决这个问题，本文提出了一种基于几何分解和多分辨率低秩特征的快速隐式光场表示方法。该方法将4D光场分解为一个水平视差平面、一个空间纹理平面和一个垂直视差平面。每个平面由一个低秩结构表示，该结构结合了一个低分辨率的2D网格和在多个分辨率级别下两个高分辨率1D线特征的逐元素乘积。融合的特征通过一个轻量级的多层感知器解码，以预测RGB值。在公共光场数据集上的实验表明，所提方法在重建质量上具有竞争力，同时在模型参数、训练时间和推理效率之间提供了更好的权衡。

View on arXiv Download PDF AI Translation

cs.CV / 20 / 2608.13967

SAFE: Scene-Aware Feature Modulation for Color Constancy with Learned Color Space in Pure-Color Scenes

SAFE：基于场景感知的特征调制在纯色场景中的颜色恒常性与学习颜色空间

Lee, Yuan-Kang, Chen, Kuan-Lin, Chang, Chih-Heng, Ding, Jian-Jiun

Abstract

Color constancy on pure-color scenes is challenging: when most pixels share a narrow band of hues, every chromaticity-based cue collapses to a single point and standard estimators become ambiguous. We propose a compact framework that couples two innovations: (i) SAFE, a Scene-Aware FeaturE modulation network that organizes illumination cues into a structured four-token representation, which is then selectively reweighted based on scene complexity features; (ii) the Learned Color Space (LCS), a scene-dependent chromaticity normalization that directly addresses the chromaticity collapse problem for pure-color scenes. Experiment results show that SAFE consistently improves performance in pure-color scenes. Compared to the best-performing baseline in each metric, it reduces the mean angular error by 10%, the best-25% error by 20%, and the worst-25% error by 5.8%.

Chinese Translation

在纯色场景中实现颜色恒常性具有挑战性：当大多数像素共享狭窄的色调带时，基于色度的线索会崩溃为一个单一的点，标准估计器变得模糊。我们提出了一个紧凑的框架，结合了两个创新：(i) SAFE，一个场景感知特征调制网络，它将照明线索组织成结构化的四个标记表示，然后根据场景复杂性特征进行选择性重加权；(ii) 学习颜色空间（Learned Color Space, LCS），一种依赖场景的色度归一化，直接解决了纯色场景中的色度崩溃问题。实验结果表明，SAFE在纯色场景中持续提高了性能。与每个指标中表现最佳的基线相比，它将平均角度误差降低了10%，最佳25%误差降低了20%，最差25%误差降低了5.8%。

View on arXiv Download PDF AI Translation

cs.CV / 21 / 2608.13969

PPOM: Marginalizing Patch-Grid Phase for CLIP-Based Generalizable Vision-Language Prompt Tuning

PPOM：边际化补丁网格相位以实现基于CLIP的可泛化视觉-语言提示调优

Wang, Liang, Li, Haoyang, Wang, Chao, Long, Guodong, Jiang, Jing, Peng, Yan

Abstract

Prompt tuning adapts CLIP-based vision-language models with few trainable parameters, yet its predictions remain sensitive to the spatial sampling imposed by a frozen vision transformer. In particular, non-overlapping patch tokenization makes predictions depend on the alignment (phase) between image and the patch lattice. To reduce prediction sensitivity to patch-grid alignment, we introduce Patch-Phase Orbit Marginalization (PPOM), a training-free inference operator that treats phase shift as a nuisance variable. Given a patch stride, PPOM evaluates the identity view and reflection-padded translations, pairs opposite shifts into horizontal, vertical, and diagonal antithetic families, and assigns equal mass to these families and the identity prediction to avoid view-count bias during phase integration. In summary, PPOM provides a deterministic interface between prompt adaptation and patch-grid sensitivity. Across multiple prompt-learning hosts, PPOM improves host performance without re-training.

Chinese Translation

提示调优通过少量可训练参数适应基于CLIP的视觉-语言模型，但其预测仍然对冻结的视觉变换器施加的空间采样敏感。特别是，非重叠的补丁标记化使得预测依赖于图像与补丁格子之间的对齐（相位）。为了减少预测对补丁网格对齐的敏感性，我们引入了补丁相位轨道边际化（PPOM），这是一种无训练的推理操作符，将相位偏移视为干扰变量。在给定的补丁步幅下，PPOM 评估身份视图和反射填充的平移，将相对偏移配对成水平、垂直和对角线的对立家族，并在相位整合过程中将这些家族与身份预测赋予相等的权重，以避免视图计数偏差。总之，PPOM 提供了提示适应与补丁网格敏感性之间的确定性接口。在多个提示学习主机上，PPOM 在不重新训练的情况下提高了主机性能。

View on arXiv Download PDF AI Translation

cs.CV / 22 / 2608.13973

Rethinking Auxiliary Modalities in Multimodal Zero-shot Anomaly Detection: From Semantic Fusion to Conditional Modulation

重新思考多模态零样本异常检测中的辅助模态：从语义融合到条件调制

Wu, Peng, Ge, Xin, Sun, Yujia, Pang, Guansong

Abstract

Recent foundation model-based methods have endowed RGB images with strong zero-shot anomaly detection (ZSAD) through vision-language pretraining. However, RGB observations alone remain limited in perceiving anomalies dominated by geometric deformation, depth variation, or subtle surface changes. Auxiliary modalities can provide complementary structural information, but existing multimodal methods typically fuse them directly into a shared semantic space, which may disturb the text-aligned anomaly semantics established by RGB foundation models and often requires modality-specific architectures. To address this issue, we propose a plug-and-play auxiliary-conditioned enhancement framework for zero-shot anomaly detection. Instead of reconstructing a joint multimodal anomaly semantic space, our framework preserves the original RGB image-text anomaly matching pathway and uses auxiliary observations as conditional signals for RGB feature refinement, allowing auxiliary modalities to seamlessly enhance existing RGB-based zero-shot anomaly detectors. Specifically, a lightweight meta-learning module takes global RGB and auxiliary representations as input and generates sample-adaptive low-rank residual updates to determine how RGB features should be refined. We further construct uncertainty-aware spatial modulation from the initial RGB anomaly response and auxiliary reliability, which determines where local residual updates are strengthened or suppressed. This global-to-local conditional modulation enables selective multimodal enhancement while preserving the original RGB anomaly semantics. Extensive experiments on MVTec 3D-AD and Eyecandies demonstrate that our framework consistently improves multiple popular RGB-based zero-shot anomaly detectors, achieving state-of-the-art performance for multimodal zero-shot anomaly detection.

Chinese Translation

近期基于基础模型的方法通过视觉-语言预训练赋予RGB图像强大的零样本异常检测（ZSAD）能力。然而，仅依靠RGB观察在感知以几何变形、深度变化或微妙表面变化为主的异常时仍然存在局限性。辅助模态可以提供互补的结构信息，但现有的多模态方法通常将其直接融合到共享的语义空间中，这可能会干扰RGB基础模型所建立的文本对齐异常语义，并且通常需要特定于模态的架构。为了解决这个问题，我们提出了一种即插即用的辅助条件增强框架用于零样本异常检测。我们的框架不再重建联合多模态异常语义空间，而是保留原始RGB图像-文本异常匹配路径，并将辅助观察作为RGB特征精炼的条件信号，从而使辅助模态能够无缝增强现有的基于RGB的零样本异常检测器。具体而言，一个轻量级的元学习模块将全局RGB和辅助表示作为输入，生成样本自适应的低秩残差更新，以确定RGB特征应如何被精炼。我们进一步构建了基于初始RGB异常响应和辅助可靠性的感知不确定性空间调制，以确定局部残差更新的增强或抑制位置。这种从全局到局部的条件调制使得选择性多模态增强成为可能，同时保留了原始RGB异常语义。在MVTec 3D-AD和Eyecandies上的广泛实验表明，我们的框架持续提升了多种流行的基于RGB的零样本异常检测器，达到了多模态零样本异常检测的最新性能。

View on arXiv Download PDF AI Translation

cs.CV / 23 / 2608.13974

ProFocus: Interpreting Affective Experience in Artistic Images with Progressive Visual Focusing

ProFocus：通过渐进视觉聚焦解读艺术图像中的情感体验

Zhang, Zhiyan, Yan, Zicheng, Chen, Jianqi, Song, Peipei, Wang, Shanshan, Yang, Xun

Abstract

Interpreting the emotional responses triggered by images is central to achieving emotional intelligence. Compared with natural images, visual art is intentionally created to elicit emotional responses from its viewers through abstract concepts and visual metaphors, making affective interpretation particularly challenging. However, most existing methods rely on general-purpose visual embeddings (e.g., CLIP), failing to capture the nuanced cues underlying artistic emotion. To address this gap, we propose \textbf{ProFocus}, a novel framework that models affective experience in artistic images via progressive visual focusing. The key idea is to model visual representation learning inspired by a hierarchical cognitive theory of human aesthetic appreciation. Technically, ProFocus contains two core components: a Hierarchical Art Critic (HAC) and a Progressive Hint Fusion (PHF) module. HAC leverages multimodal large language models to generate structured linguistic priors at three cognitive levels--atmospheric style, narrative subjects, and concrete details--thereby translating artistic perception into coherent semantic guidance. Building upon these priors, PHF departs from conventional cross-modal fusion by sequentially injecting the hierarchical hints into visual features, enabling a progressive focusing process that mirrors human perception. This design allows the model to capture subtle affective cues and produce more faithful explanations. Extensive experiments on the ArtEmis v1.0 and v2.0 datasets demonstrate that ProFocus consistently outperforms state-of-the-art methods in both emotion recognition and affective explanation. Project page: https://github.com/Zhang-Zhiyan/ProFocus.

Chinese Translation

解读图像引发的情感反应是实现情感智能的核心。与自然图像相比，视觉艺术是有意创作的，旨在通过抽象概念和视觉隐喻引发观众的情感反应，这使得情感解读尤为具有挑战性。然而，现有大多数方法依赖于通用的视觉嵌入（例如，CLIP），未能捕捉艺术情感背后的细微线索。为了解决这一问题，我们提出了 extbf{ProFocus}，一个通过渐进视觉聚焦建模艺术图像中情感体验的新框架。其核心思想是基于人类审美欣赏的层次认知理论来建模视觉表征学习。在技术上，ProFocus包含两个核心组件：层次艺术评论家（HAC）和渐进提示融合（PHF）模块。HAC利用多模态大型语言模型在三个认知层次（氛围风格、叙事主题和具体细节）生成结构化语言先验，从而将艺术感知转化为连贯的语义指导。在这些先验的基础上，PHF通过将层次提示顺序注入视觉特征，突破传统的跨模态融合，启用一种渐进聚焦过程，模拟人类感知。这一设计使模型能够捕捉细微的情感线索并生成更真实的解释。在ArtEmis v1.0和v2.0数据集上的广泛实验表明，ProFocus在情感识别和情感解释方面始终优于最先进的方法。项目页面：https://github.com/Zhang-Zhiyan/ProFocus。

View on arXiv Download PDF AI Translation

cs.CV / 24 / 2608.13980

FIRM: Fine-Grained Intra-Token Representation of Masks for Remote Sensing Reasoning Segmentation

FIRM：用于遥感推理分割的细粒度内部标记表示

Tang, Weidong, Li, Kaiyu, Wang, Yikai, Wu, Yanan, Gan, Haotian, Wang, Shihong, Cao, Xiangyong

Abstract

Reasoning segmentation requires multimodal large language models (MLLMs) to translate implicit instructions into precise pixel-level masks. MLLMs encode an image as visual tokens, each of which merges a group of image patches. In remote sensing images, small targets, thin structures, and adjacent instances can occupy different parts of the same visual token. Assigning a single binary mask label to such a token loses its internal spatial structure, causing nearby targets to merge and object boundaries to become coarse. To bridge this representational gap, we introduce FIRM, a Fine-grained Intra-token Representation of Masks. For each visual token, FIRM predicts a mask code that specifies an $r\times r$ binary sub-cell pattern rather than a single foreground/background label. Given a target identified by the MLLM, the complete grid of mask codes is predicted in one mask pass. Fixed lookup converts the predicted codes into a discrete sub-cell mask, while marginalizing the code distribution yields a soft structural field. To further recover fine-grained boundaries within each sub-cell, we introduce a lightweight continuous renderer that refines this field using pre-merge visual features and image details. Across five reasoning and referring segmentation benchmarks on satellite and UAV images, FIRM achieves leading results, including $70.5/80.5$ gIoU/cIoU on LaSeRS and a $3.0$-point average gain on EarthReason. These results demonstrate the value of explicitly representing intra-token mask patterns for fine-grained MLLM segmentation.

Chinese Translation

推理分割需要多模态大型语言模型（MLLMs）将隐含指令转换为精确的像素级掩码。MLLMs将图像编码为视觉标记，每个标记合并了一组图像块。在遥感图像中，小目标、细结构和相邻实例可能占据同一视觉标记的不同部分。将单一的二进制掩码标签分配给这样的标记会丧失其内部空间结构，导致相邻目标合并，物体边界变得粗糙。为了解决这一表示差距，我们提出了FIRM，即细粒度内部标记表示。对于每个视觉标记，FIRM预测一个掩码代码，该代码指定一个$r imes r$的二进制子单元模式，而不是单一的前景/背景标签。给定由MLLM识别的目标，完整的掩码代码网格在一次掩码传递中被预测。固定查找将预测的代码转换为离散的子单元掩码，而对代码分布的边际化则产生一个软结构场。为了进一步恢复每个子单元内的细粒度边界，我们引入了一种轻量级连续渲染器，利用合并前的视觉特征和图像细节来细化该场。在五个卫星和无人机图像的推理和引用分割基准测试中，FIRM取得了领先的结果，包括在LaSeRS上达到$70.5/80.5$的gIoU/cIoU，以及在EarthReason上获得$3.0$的平均增益。这些结果证明了明确表示内部标记掩码模式在细粒度MLLM分割中的价值。

View on arXiv Download PDF AI Translation

cs.CV / 25 / 2608.14015

MedClaw: Heuristic Agent Harness for Long-Horizon Surgical Video Reasoning

MedClaw：用于长时间手术视频推理的启发式代理工具

Fan, Yingying, Du, Penghui, Zhu, Leyan, He, Runze, Wu, Zimeng, Zhang, Yuxuan, Chen, Liang, Xie, Jiahao, Wang, Jiangtang, Shao, Shuai, Yang, Anchao, Bai, Yutong, Wang, Yan

Abstract

Understanding tens-of-minutes surgical videos requires long-horizon temporal reasoning, answering what happens before, after, or across stages of a procedure by grounding the question in visual evidence spread across time. Existing approaches handle this poorly: a one-shot vision-language model (VLM) compresses the whole procedure to fit its context window and loses the detail a "before" or "after" question depends on, while video agents that train the model where to look are data-hungry and transfer poorly to out-of-domain surgery. We build an agent harness that separates reasoning from perception and improves by evolving context rather than optimizing weights. A text-only orchestrator plans which evidence to gather and issues an auditable sequence of tool calls, while frozen vision-language sub-agents execute each call over the pixels, viewing, cropping, inspecting frames, and retrieving external knowledge. We further propose a gradient-free, reward-gated Heuristic Skill Distillation loop that mines the agent's own low-scoring traces and keeps a candidate skill only when it raises a validation reward, yielding reusable retrieval skills, notably directed re-look. Growing an external skill library rather than tuning weights, the loop adapts from only about 100 labeled examples, far fewer than supervised or reinforcement fine-tuning requires. To evaluate this agent, we introduce MedClawBench, a de-leaked, doctor-grounded benchmark of 1,123 questions over self-built long neurosurgery recordings and a held-out public lecture-video test split. Across both datasets and all four evaluation dimensions, our agent consistently outperforms one-shot VLMs and general video-agent frameworks, with the largest gains on the long, out-of-domain neurosurgery videos. Project page: https://fyycs.github.io/medclaw/.

Chinese Translation

理解数十分钟的手术视频需要长时间的时间推理，通过将问题与跨时间的视觉证据相结合，回答在手术的不同阶段之前、之后或之间发生了什么。现有方法对此处理不佳：一种单次视觉语言模型（VLM）将整个过程压缩以适应其上下文窗口，导致失去“之前”或“之后”问题所依赖的细节，而训练模型观察位置的视频代理则需要大量数据，并且在非领域手术中迁移效果不佳。我们构建了一个代理工具，将推理与感知分离，并通过演变上下文而不是优化权重来改进。一个仅文本的协调器规划要收集的证据，并发出可审计的工具调用序列，而冻结的视觉语言子代理则在像素上执行每个调用，查看、裁剪、检查帧并检索外部知识。我们进一步提出了一种无梯度、奖励门控的启发式技能蒸馏循环，该循环挖掘代理自身的低得分轨迹，并仅在提高验证奖励时保留候选技能，从而产生可重用的检索技能，特别是定向重看。该循环通过建立外部技能库而不是调整权重，仅需大约100个标记示例，远少于监督或强化微调所需的数量。为了评估该代理，我们引入了MedClawBench，这是一个去泄露、以医生为基础的基准，包含1,123个问题，基于自建的长时间神经外科录音和一个保留的公共讲座视频测试集。在这两个数据集和所有四个评估维度上，我们的代理始终优于单次VLM和通用视频代理框架，在长时间的非领域神经外科视频上获得了最大的提升。项目页面：https://fyycs.github.io/medclaw/

View on arXiv Download PDF AI Translation

cs.CV / 26 / 2608.14016

Content Based Video Narration of Gameplay with Vision Language Models

基于内容的视频游戏解说与视觉语言模型

Varghese, Mathew

Abstract

Live game commentary is scarce: it exists for professional esports broadcasts and almost nowhere else. We present a content-based video narration system that produces spoken, esports-style commentary for arbitrary gameplay recordings using a general-purpose vision-language model (VLM) and a text-to-speech back end, with no game-specific instrumentation, no engine telemetry, and no task-specific training. Three mechanisms carry the system. Temporal mosaic packing arranges nine uniformly sampled frames into a single 3x3 image, letting an image-native VLM reason about motion while consuming one image payload per segment instead of nine. Context-conditioned prompting replays the K most recent narrations as assistant-role history, suppressing the repetition that dominates per-segment captioning of static scenes. Duration-conditioned generation and elastic alignment constrain narration length in the prompt, then time-scale or symmetrically pad the synthesized audio so each utterance fills its segment slot exactly, giving frame-accurate muxing without a forced aligner. The implementation supports either cloud TTS or a 6-bit quantized 4B-parameter on-device TTS model on Apple silicon, making the speech stage fully local. We report a qualitative case study on real-time strategy footage, a cost model showing the mosaic reduces per-minute image payloads by 9x, and a candid account of observed failure modes - hallucinated game state, resolution loss from mosaicking, and prosody artifacts from time-scaling. We release the system as a reproducible baseline, with an evaluation protocol for the quantitative study a full version will report.

Chinese Translation

实时游戏解说非常稀缺：它仅存在于职业电子竞技直播中，几乎没有其他地方。我们提出了一种基于内容的视频解说系统，该系统利用通用的视觉语言模型（VLM）和文本转语音（TTS）后端，为任意游戏录制生成口语化的电子竞技风格解说，无需游戏特定的仪器、引擎遥测或任务特定的训练。该系统由三种机制支撑。时间马赛克打包将九个均匀采样的帧排列成一个3x3的图像，使得图像本地的VLM能够在消耗每个片段一个图像负载的同时推理运动，而不是九个。上下文条件提示重放最近K条解说作为助手角色历史，抑制了在静态场景的每个片段字幕中占主导地位的重复现象。持续时间条件生成和弹性对齐在提示中约束解说长度，然后对合成音频进行时间缩放或对称填充，使每个发声恰好填充其片段位置，实现帧精确的复用，而无需强制对齐器。该实现支持云端TTS或在Apple硅芯片上的6位量化4B参数的本地TTS模型，使得语音阶段完全本地化。我们报告了一个关于实时策略视频的定性案例研究，一个成本模型显示马赛克将每分钟的图像负载减少了9倍，以及对观察到的失败模式的坦诚描述——虚构的游戏状态、马赛克造成的分辨率损失，以及时间缩放带来的韵律伪影。我们将该系统作为可重复的基线发布，并提供了一个评估协议，以便进行定量研究，完整版本将对此进行报告。

View on arXiv Download PDF AI Translation

cs.CV / 27 / 2608.14022

ForgeWM: Progressive Causal Training for Few-Step Action-Conditioned Video World Models

ForgeWM：面向少步动作条件视频世界模型的渐进式因果训练

Li, Xinye, Lin, Lingshuai, Wang, Lei, Zhang, Liuzhou, Cui, Jialin, Li, Qingshan, Wang, Guanchu, Liu, Qingbin, Chen, Xi, Bian, Jiang, Lam, Wai

Abstract

Action-conditioned video world models require low-latency causal generation and reliable responses to game-native controls. Although causal distillation enables one- or few-step video synthesis, extending it to interactive world models remains challenging, as discrete keyboard states and continuous mouse motion must remain aligned with temporally compressed latent chunks during causal training and autoregressive rollout. We introduce ForgeWM, a progressive framework that transforms a bidirectional action-conditioned video generator into efficient few-step world models through domain adaptation, teacher-forced causal training, causal consistency distillation, and on-policy distribution matching with a bidirectional teacher. The resulting budget-specialized students operate at steady-state denoising budgets of 1, 2, and 4 steps. ForgeWM further supports a dual-path deployment protocol combining latency-critical interaction with optional replay-time refinement, where the one-step student re-noises and refines its saved draft. On paired Minecraft trajectories, ForgeWM leads the evaluated systems in Imaging Quality, reference-aligned motion-profile agreement, action-sign accuracy, and mouse-control accuracy, while achieving the lowest reference LPIPS; the same four-stage recipe transfers to gamepad-controlled FPS gameplay. Replay-time refinement matches four-step reference quality while remaining roughly three times closer to the experienced trajectory than regeneration from noise. These results demonstrate ForgeWM's effectiveness for controllable few-step video generation.

Chinese Translation

动作条件视频世界模型需要低延迟的因果生成和对游戏原生控制的可靠响应。尽管因果蒸馏使得一或少步视频合成成为可能，但将其扩展到交互式世界模型仍然具有挑战性，因为离散的键盘状态和连续的鼠标运动必须在因果训练和自回归生成过程中与时间压缩的潜在块保持一致。我们提出了ForgeWM，一个渐进式框架，通过领域适应、教师强制因果训练、因果一致性蒸馏和与双向教师的策略匹配，将双向动作条件视频生成器转变为高效的少步世界模型。最终生成的预算专用学生在1、2和4步的稳态去噪预算下运行。ForgeWM还支持一种双路径部署协议，结合了对延迟敏感的交互和可选的回放时间精炼，其中一阶学生对其保存的草稿进行去噪和精炼。在配对的Minecraft轨迹上，ForgeWM在成像质量、参考对齐的运动轮廓一致性、动作信号准确性和鼠标控制准确性方面领先于评估系统，同时实现了最低的参考LPIPS；相同的四阶段方案可以转移到手柄控制的第一人称射击游戏中。回放时间精炼在质量上匹配四步参考，同时与经验轨迹的距离大约接近三倍于从噪声再生的结果。这些结果证明了ForgeWM在可控的少步视频生成中的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 28 / 2608.14024

SSP: An Event-Matched Syn2Sim2Phy Cross-Domain Evaluation Framework for Autonomous Driving VLA Models

SSP：一种事件匹配的Syn2Sim2Phy跨域评估框架用于自主驾驶VLA模型

Feng, Haojie, Zhang, Peizhi, Zhang, Xinrui, Li, Zhuoren, Huang, Junpeng, Wang, Xiurong, Yin, Dongxiao, Zhang, Yuxiang, Zhu, Junfan, Xiong, Lu

Abstract

Vision-language-action (VLA) models for autonomous driving jointly produce scene interpretation, language-based reasoning, and driving trajectories. Existing evaluations often use independently selected synthetic, simulated, and physical data, so measured performance gaps can be confounded by changes in scenario content rather than genuine domain sensitivity. We propose SSP (Synthetic-Simulation-Physical), an event-matched Syn2Sim2Phy evaluation framework that anchors cross-domain comparison to the same safety-critical interaction. Starting from a synthetic long-tail video, SSP builds a validated event specification that preserves road topology, participant roles, relative motion, conflict evolution, passing order, response constraints, and event phases. Platform-specific realizations are then constructed in CARLA and on a closed proving ground and are evaluated only after transfer audits confirm preservation of mandatory event properties. SSP maps heterogeneous outputs from OpenEMMA, LLaViDA, and Alpamayo-R1 into common semantic slots and a 1 s trajectory window to assess output validity, semantic accuracy, critical-interaction recognition, trajectory quality, and risk response. Across Cut-in and vulnerable-road-user crossing cases, the macro-averaged Integrated VLA Capability Scores are 0.259, 0.291, and 0.325 in the Synthetic, Simulation, and Physical domains, respectively, while the best domain varies by scenario. Alpamayo-R1, OpenEMMA, and LLaViDA obtain scores of 0.405, 0.338, and 0.131. SSP provides a reproducible scene-transfer chain and an evidence-qualified evaluation of VLA behavior without assuming that the Physical domain is universally superior.

Chinese Translation

自主驾驶的视觉-语言-行动（VLA）模型共同生成场景解释、基于语言的推理和驾驶轨迹。现有评估通常使用独立选择的合成、模拟和物理数据，因此测量的性能差距可能受到场景内容变化的干扰，而非真实的领域敏感性。我们提出了SSP（合成-模拟-物理），一种事件匹配的Syn2Sim2Phy评估框架，将跨域比较锚定到相同的安全关键交互上。从合成的长尾视频开始，SSP构建了一个经过验证的事件规范，保留了道路拓扑、参与者角色、相对运动、冲突演变、通过顺序、响应约束和事件阶段。然后在CARLA平台和封闭试验场构建特定平台的实现，并仅在转移审核确认保留强制事件属性后进行评估。SSP将来自OpenEMMA、LLaViDA和Alpamayo-R1的异构输出映射到共同的语义槽和1秒的轨迹窗口，以评估输出有效性、语义准确性、关键交互识别、轨迹质量和风险响应。在切入和脆弱道路使用者穿越的案例中，合成、模拟和物理领域的宏观平均综合VLA能力分数分别为0.259、0.291和0.325，而最佳领域因场景而异。Alpamayo-R1、OpenEMMA和LLaViDA的得分分别为0.405、0.338和0.131。SSP提供了一个可重复的场景转移链和对VLA行为的证据合格评估，而不假设物理领域在普遍意义上优于其他领域。

View on arXiv Download PDF AI Translation

cs.CV / 29 / 2608.14027

E-S2Feat:Semantic-Guided Spiking Local Feature Detection and Description for Event Cameras

E-S2Feat：基于语义引导的脉冲局部特征检测与描述方法用于事件相机

Yi, Yang, Hua, Juntao, Zhang, Jinpu, Fan, Liangwei, Shen, Hui, Hu, Dewen

Abstract

Benefiting from high temporal resolution and dynamic range, event-based local feature methods have attracted increasing attention. However, event sparsity, noise, and limited texture still hinder robust local feature learning. Deploying such methods on resource-constrained platforms such as unmanned aerial vehicles also requires balancing accuracy and energy efficiency. To address these challenges, this paper proposes \textbf{E-S2Feat}, a spiking neural network framework for event-based local feature detection and description. The framework jointly optimizes local feature learning from the perspectives of feature representation and selection. First, a module-specific spiking activation mechanism preserves fine-grained structural cues and discriminative information under low-bit, energy-efficient inference, thereby improving overall representation fidelity. Furthermore, a semantic-guided feature modulation mechanism leverages semantic priors to refine keypoint response distributions and enhance local descriptor discriminability, thereby guiding the model to extract local features with greater geometric stability and stronger discriminative capability. Experiments on the ECD and EDS datasets show that the proposed method significantly outperforms baseline methods such as SuperEvent in pose estimation accuracy. It also achieves accuracy comparable to its artificial neural network counterpart while delivering an approximately 4.8-fold improvement in theoretical computational energy efficiency. Visual-inertial odometry experiments on the TUM-VIE dataset further verify the effectiveness and practical application potential of the proposed method in complete SLAM systems.

Chinese Translation

得益于高时间分辨率和动态范围，基于事件的局部特征方法受到越来越多的关注。然而，事件稀疏性、噪声和有限的纹理仍然阻碍了稳健的局部特征学习。在资源受限的平台（如无人机）上部署此类方法还需要在准确性和能效之间取得平衡。为了解决这些挑战，本文提出了 extbf{E-S2Feat}，一种用于基于事件的局部特征检测与描述的脉冲神经网络框架。该框架从特征表示和选择的角度共同优化局部特征学习。首先，模块特定的脉冲激活机制在低位、节能推理下保留细粒度的结构线索和判别信息，从而提高整体表示的保真度。此外，基于语义引导的特征调制机制利用语义先验来细化关键点响应分布，增强局部描述子的判别能力，从而引导模型提取具有更大几何稳定性和更强判别能力的局部特征。在ECD和EDS数据集上的实验表明，所提出的方法在姿态估计准确性上显著优于基线方法，如SuperEvent。同时，它在准确性上与其人工神经网络对应物相当，同时在理论计算能效上实现了约4.8倍的提升。在TUM-VIE数据集上的视觉惯性里程计实验进一步验证了所提方法在完整SLAM系统中的有效性和实际应用潜力。

View on arXiv Download PDF AI Translation

cs.CV / 30 / 2608.14043

Beyond Text Conditioning: A Systematic Study of MLLM-DiT Fusion for Video Generation

超越文本条件：关于MLLM-DiT融合在视频生成中的系统研究

Ding, Yanbo, Fan, Yijia, Shan, Caihua, Yang, Yifan, Shen, Yifei, Wang, Weijie, Hu, Xirui, Li, Dongsheng, Qiu, Lili, Yang, Yuqing, Wang, Yali

Abstract

Diffusion Transformers (DiTs) have become the dominant paradigm for high-fidelity video generation, yet their ability to perform high-level semantic planning remains limited. While hybrid architectures integrating MLLMs with diffusion backbones have shown strong advantages in image synthesis, such designs remain underexplored in video generation, where existing approaches often treat MLLMs primarily as frozen feature encoders rather than semantic generators. To fill this gap, we systematically study how an MLLM should be integrated with a DiT for video generation by answering three questions: what intermediate representation should bridge the MLLM and DiT, how the MLLM should generate it, and how the DiT should incorporate it during diffusion rendering. Our analysis reveals three key findings: (1) discrete semantic visual tokens produced by an EMA-based tokenizer provide a stable and expressive interface, (2) autoregressive causal modeling is effective for generating these tokens, and (3) explicit visual-token conditioning is more effective than prompt refinement or latent bridging. Based on these findings, we propose BiVidGen, a hybrid framework where an MLLM first generates semantic visual tokens and a DiT renders videos conditioned on both text and these tokens via multi-layer cross-attention. Extensive experiments show that BiVidGen improves semantic alignment and temporal coherence over a fine-tuned DiT baseline, achieving stronger performance on VBench-Long. These results demonstrate that explicit MLLM-based visual planning provides an effective intermediate interface for text-to-video generation beyond text-only conditioning.

Chinese Translation

扩散变换器（Diffusion Transformers, DiTs）已成为高保真视频生成的主导范式，但其在高层语义规划方面的能力仍然有限。尽管将多模态大语言模型（MLLMs）与扩散骨干网络结合的混合架构在图像合成中显示出强大的优势，但在视频生成中，这类设计仍然未得到充分探索，现有方法往往将MLLMs主要视为冻结的特征编码器，而非语义生成器。为填补这一空白，我们系统地研究了如何将MLLM与DiT结合用于视频生成，回答了三个问题：什么样的中间表示应当连接MLLM和DiT，MLLM应如何生成该表示，以及DiT在扩散渲染过程中应如何整合该表示。我们的分析揭示了三个关键发现：（1）基于EMA的标记器生成的离散语义视觉标记提供了一个稳定且富有表现力的接口；（2）自回归因果建模在生成这些标记方面是有效的；（3）显式视觉标记条件化比提示优化或潜在桥接更为有效。基于这些发现，我们提出了BiVidGen，一个混合框架，其中MLLM首先生成语义视觉标记，DiT则通过多层交叉注意力在文本和这些标记的条件下渲染视频。大量实验表明，BiVidGen在语义对齐和时间一致性方面优于经过微调的DiT基线，在VBench-Long上取得了更强的性能。这些结果表明，基于MLLM的显式视觉规划为超越仅基于文本条件的视频生成提供了有效的中间接口。

View on arXiv Download PDF AI Translation

cs.CV / 31 / 2608.14046

Source-Agnostic Image Translation Based on Latent Aware Adaptive Masking

基于潜在感知自适应掩膜的源无关图像转换

Dobrički, Tomislav, Hong, Byung-Woo

Abstract

In this work, we propose a source-agnostic framework that dynamically refines a binary mask throughout the reverse diffusion process by computing the discrepancies of a pretrained diffusion model's prediction for each latent time step. Rather than relying on a fixed threshold, our method introduces a time-dependent statistical thresholding scheme derived from the empirical mean and standard deviation of prediction discrepancies across the latent noisy images from the target distribution. This allows the mask to adapt to the model's varying predictive confidence at different noise levels, effectively isolating domain-specific regions while preserving global structural coherence. Experimental results on the AFHQ and Celeba-HQ datasets demonstrate that our approach outperforms state-of-the-art unsupervised Image-to-Image methods in both realism (FID, KID) and faithfulness (SSIM, LPIPS). By requiring only a pretrained model of the target domain, our approach enables precise, automated localization and seamless translation across diverse source distributions without any specialized training. The project source code is available at: https://github.com/dtoma95/PM-Edit

Chinese Translation

在本研究中，我们提出了一种源无关框架，该框架通过计算预训练扩散模型在每个潜在时间步的预测差异，动态地在反向扩散过程中细化二进制掩膜。我们的方法并不依赖于固定阈值，而是引入了一种基于目标分布中潜在噪声图像的预测差异的经验均值和标准差推导出的时间依赖统计阈值方案。这使得掩膜能够适应模型在不同噪声水平下的变化预测信心，有效地隔离特定领域的区域，同时保持全局结构的一致性。在AFHQ和Celeba-HQ数据集上的实验结果表明，我们的方法在现实性（FID, KID）和忠实性（SSIM, LPIPS）方面均优于最先进的无监督图像到图像方法。通过仅需目标领域的预训练模型，我们的方法实现了对多样源分布的精确自动定位和无缝转换，而无需任何专门训练。项目源代码可在以下网址获取：https://github.com/dtoma95/PM-Edit

View on arXiv Download PDF AI Translation

cs.CV / 32 / 2608.14051

Discovery and Spatial Characterisation of Multiple Shortcut Groups for Auditing Vision Model Bias

审计视觉模型偏差的多重快捷组的发现与空间特征化

Achara, Akshit, Manickam, Vishnunarayan, Day, Thomas, Anton, Esther Puyol, Hammers, Alexander, King, Andrew P.

Abstract

Deep learning models trained on datasets with spurious correlations can achieve high average accuracy whilst relying on shortcut features that do not generalise out of distribution. Whilst out-of-distribution testing highlights subgroup performance disparities arising from shortcut learning, it does not localise the regions within images that are associated with it. Existing research mostly uses attribution maps from interpretability methods to understand the spatial nature of spurious correlations. For example, conditional alignment methods separate task-relevant evidence from evidence tied to spurious correlations by comparing attribution maps from a task model, a sensitive attribute model, and a bias-reduced reference model. This yields shortcut-aligned and task-aligned contribution maps for each image. However, existing methods aggregate these maps across the dataset, potentially masking recurring spatial shortcut patterns that occur only in subsets of images. We address this limitation by grouping per-image shortcut and task contribution maps into recurring spatial patterns using K-means and non-negative matrix factorisation, and visualising the resulting shortcut groups through contribution maps and representative examples. Across CelebA, CheXpert, Waterbirds, Camelyon17, and ISIC2019, and across ResNet and ViT models, the discovered shortcut groups reveal both shared and distinct spatial patterns of shortcut and task contribution, with varying subgroup composition and error rates, enabling targeted inspection of image subsets with higher error rates. We perform input occlusion and internal test-time interventions to show that masking or suppressing task contribution regions substantially degrades the model classification performance and propose a combined shortcut suppression and task amplification feature intervention approach which generally reduces performance disparities.

Chinese Translation

在具有虚假相关性的数据库上训练的深度学习模型可以实现高平均准确率，但依赖于不具备泛化能力的快捷特征。尽管分布外测试突显了由于快捷学习而产生的子组性能差异，但并未定位与之相关的图像区域。现有研究主要使用可解释性方法的归因图来理解虚假相关性的空间特性。例如，条件对齐方法通过比较任务模型、敏感属性模型和偏差减少参考模型的归因图，将与任务相关的证据与与虚假相关性相关的证据分开。这为每个图像生成了快捷对齐和任务对齐的贡献图。然而，现有方法在数据集上聚合这些图，可能掩盖了仅在图像子集中出现的重复空间快捷模式。我们通过使用 K-means 和非负矩阵分解将每个图像的快捷和任务贡献图分组为重复的空间模式，来解决这一局限性，并通过贡献图和代表性示例可视化所得到的快捷组。在 CelebA、CheXpert、Waterbirds、Camelyon17 和 ISIC2019 数据集，以及 ResNet 和 ViT 模型中，发现的快捷组揭示了快捷和任务贡献的共享和独特的空间模式，具有不同的子组组成和错误率，从而使得能够针对性地检查具有较高错误率的图像子集。我们进行输入遮挡和内部测试时干预，表明遮挡或抑制任务贡献区域会显著降低模型分类性能，并提出了一种结合快捷抑制和任务增强的特征干预方法，通常减少性能差异。

View on arXiv Download PDF AI Translation

cs.CV / 33 / 2608.14058

Voxel-based 3D Facies Segmentation from Seismic Data: A Comparative Study

基于体素的三维地层分割：来自地震数据的比较研究

Pham, Duc-Thanh, Pham, Minh-Tan, Nguyen, Anh, Nguyen, Van

Abstract

Seismic facies segmentation has emerged as a significant challenge in geophysics, requiring robust methods and systems to effectively identify geologically analogous facies with limited labeled data. Although existing studies have shown promising results in 2D facies segmentation, they often preprocess the original 3D seismic volumes into sets of 2D slices, typically the inline and crossline directions, and treat this problem as a purely 2D segmentation task. This simplification introduces discontinuities across slices and fails to preserve the spatial and structural continuity in 3D seismic data, thus limiting the model's ability to learn coherent geological patterns. In this work, we present a comparative and reproducible benchmark for voxel-based 3D seismic facies segmentation, built upon publicly available seismic volumes including the Netherlands F3 and the Parihaka datasets, with standardized data splits and evaluation metrics. By evaluating the three representative families of modern 3D segmentation architectures, we establish strong baseline results that highlight the potential and remaining challenges for future research in this domain.

Chinese Translation

地震地层分割已成为地球物理学中的一个重大挑战，要求采用稳健的方法和系统来有效识别具有地质相似性的地层，尤其是在标注数据有限的情况下。尽管现有研究在二维地层分割中显示出良好的结果，但它们通常将原始的三维地震体积预处理为一组二维切片，通常是沿着内线和交叉线方向，并将此问题视为纯粹的二维分割任务。这种简化导致切片之间出现不连续性，未能保持三维地震数据中的空间和结构连续性，从而限制了模型学习一致的地质模式的能力。在本研究中，我们提出了一个基于体素的三维地震地层分割的比较和可重复基准，基于包括荷兰F3和Parihaka数据集在内的公开可用地震体积，采用标准化的数据划分和评估指标。通过评估三种现代三维分割架构的代表性家族，我们建立了强有力的基线结果，突显了该领域未来研究的潜力和面临的挑战。

View on arXiv Download PDF AI Translation

cs.CV / 34 / 2608.14070

InstructVVT: Instruction-Driven Video Virtual Try-On without Auxiliary Spatial Priors

InstructVVT：无辅助空间先验的指令驱动视频虚拟试穿

Shao, Dingbao, Wu, Song, Chen, Xinyu, Wang, Qian, Li, Jiahang, Jiang, Kuai, Lin, Jiang, Liu, Yuhang, Chen, Ziyu, Li, Duo, Hu, Jiaxin, Gu, Shengrong, Tang, Ziheng, Liu, Rongrong, Peng, Yanlun, Li, Liang, Feng, Junlan, Jin, Lujia, Zhang, Ting, Yang, Jian, Yi, Zili

Abstract

Video virtual try-on is a highly constrained editing task requiring the precise replacement of a target person's clothing while strictly preserving the original video's spatial structure and temporal dynamics. Existing methods heavily rely on auxiliary handcrafted spatial priors (e.g., masks, poses) for editing control. However, these priors are prone to failure in unconstrained real-world videos and often compress rich visual context into incomplete structural signals. Furthermore, standard reconstruction objectives fail to fully capture try-on-specific human preferences. To address these challenges, we propose InstructVVT, an instruction-driven and reference-guided video virtual try-on framework based on a Diffusion Transformer (DiT) that operates without inference-time spatial priors. Our core insight is to recover fine-grained control directly from the input triplet (source video, reference garment, and instruction) via a dual-level reference conditioning scheme. Specifically, an MLLM infers semantic edit tokens for target disambiguation and structural preservation, while a lightweight conditioning pathway explicitly injects fine-grained visual garment details. Finally, we design a try-on-specific reward and utilize the DiffusionNFT algorithm to align the model with human preferences. Extensive experiments on ViViD-S and TripVVT-Bench demonstrate that InstructVVT outperforms state-of-the-art open-source methods in garment fidelity, structural preservation, and temporal consistency, despite requiring fewer inference-time controls.

Chinese Translation

视频虚拟试穿是一项高度受限的编辑任务，要求精确替换目标人物的服装，同时严格保持原视频的空间结构和时间动态。现有方法在编辑控制上严重依赖于辅助手工制作的空间先验（例如，掩膜、姿势）。然而，这些先验在不受约束的现实世界视频中容易失效，且常常将丰富的视觉上下文压缩为不完整的结构信号。此外，标准重建目标未能充分捕捉试穿特定的人类偏好。为了解决这些挑战，我们提出了InstructVVT，这是一种基于扩散变换器（Diffusion Transformer，DiT）的指令驱动和参考引导的视频虚拟试穿框架，能够在推理时不依赖空间先验。我们的核心见解是通过双层参考条件方案直接从输入三元组（源视频、参考服装和指令）恢复细粒度控制。具体而言，MLLM推断出用于目标消歧和结构保留的语义编辑标记，而轻量级条件路径则明确注入细粒度的视觉服装细节。最后，我们设计了一种特定于试穿的奖励，并利用DiffusionNFT算法使模型与人类偏好对齐。在ViViD-S和TripVVT-Bench上的大量实验表明，尽管需要更少的推理时间控制，InstructVVT在服装保真度、结构保留和时间一致性方面优于最先进的开源方法。

View on arXiv Download PDF AI Translation

cs.CV / 35 / 2608.14078

Owner3D: Ownership-Guided Style Writing for Training-Free Localized 3D Stylization

Owner3D：基于所有权引导的风格写作用于无训练的局部3D风格化

Tao, Suchang, Shi, Kaifeng, Liu, Zhiyan, Jiang, Zhuoyuan, Ouyang, Yuqi

Abstract

Localized 3D stylization aims to modify the appearance of a specified object part while preserving the remaining surfaces. In large reconstruction models (LRMs), this task is challenging because style is injected into intermediate appearance representations before rendering, while compact triplane features are shared across target and non-target surfaces, causing style leakage and boundary ambiguity. We propose Owner3D, a training-free framework for localized 3D stylization that integrates localized appearance control directly into the LRM reconstruction process. Specifically, Owner3D introduces ownership-guided style writing to restrict reference-style injection to target regions, producing a single localized stylized triplane without additional training while avoiding separate global style and appearance representations. To resolve appearance ambiguity near semantic boundaries, we further introduce boundary dual slots that maintain separate local feature sources for target and non-target regions. Finally, a surface-first texture readout hierarchically combines surface, 3D, and triplane ownership evidence to robustly recover appearance under incomplete visibility. Experiments on a benchmark constructed from Google Scanned Objects and PartNet demonstrate that Owner3D consistently outperforms existing 3D stylization methods in target-region style fidelity and non-target appearance preservation, reducing appearance leakage by 86.4% and 89.9% compared with StyleSplat and LAENeRF, respectively.

Chinese Translation

局部3D风格化旨在修改指定对象部分的外观，同时保持其余表面的完整性。在大型重建模型（LRMs）中，这一任务具有挑战性，因为风格在渲染之前被注入到中间外观表示中，而紧凑的三平面特征在目标和非目标表面之间共享，导致风格泄漏和边界模糊。我们提出了Owner3D，这是一种无训练的局部3D风格化框架，它将局部外观控制直接集成到LRM重建过程中。具体而言，Owner3D引入了基于所有权引导的风格写作，以限制参考风格的注入仅限于目标区域，从而在不需要额外训练的情况下生成单一的局部风格化三平面，同时避免单独的全局风格和外观表示。为了消除语义边界附近的外观模糊，我们进一步引入了边界双槽，以保持目标和非目标区域的独立局部特征源。最后，表面优先的纹理读取通过分层结合表面、3D和三平面所有权证据，稳健地恢复在不完整可见性下的外观。在基于Google扫描对象和PartNet构建的基准测试中的实验表明，Owner3D在目标区域风格保真度和非目标外观保持方面始终优于现有的3D风格化方法，与StyleSplat和LAENeRF相比，外观泄漏分别减少了86.4%和89.9%。

View on arXiv Download PDF AI Translation

cs.CV / 36 / 2608.14085

CoDS: Robust Collaborative Perception via Expert-driven Detection and BEV Segmentation

CoDS：通过专家驱动的检测和鸟瞰视图分割实现稳健的协同感知

Wang, Jinlong, Jia, Yuang, Lin, Junhong, Li, Nannan, Gao, Wei

Abstract

Collaborative perception breaks through single-view limitations via multi-agent information exchange. However, multi-source noise such as pose errors and communication delays degrades fusion feature quality, constraining perception performance. Joint training of detection and BEV segmentation provides a natural remedy, where segmented road regions help constrain target distributions and detection bounding boxes help recover ambiguous segmentation boundaries. To this end, we propose a robust Collaborative perception framework with expert-driven Detection and bev Segmentation (CoDS). To address spatial inconsistency in fusion quality, we first introduce the Collaborative Reliability Map (CoRM) to explicitly quantify feature quality distribution. Based on CoRM, we design the Semantic Mixture-of-Experts (S-MoE) module to extract differentiated features for inconsistent feature demands. Finally, to further mitigate feature noise degradation, the Bidirectional Task Complementary Interaction (BTCI) refines task-aware features through bidirectional injection. Extensive experiments on OPV2V and V2V4Real datasets show that our CoDS surpasses existing baselines on both tasks and maintains stable robustness under multi-source noise. Code: https://github.com/JinlongW128/CoDS and https://openi.pcl.ac.cn/OpenAIDriving/CoDS.

Chinese Translation

协同感知通过多智能体信息交换突破了单视角的局限。然而，姿态误差和通信延迟等多源噪声降低了融合特征的质量，从而限制了感知性能。检测与鸟瞰视图（BEV）分割的联合训练提供了一种自然的解决方案，其中分割的道路区域有助于约束目标分布，而检测边界框则有助于恢复模糊的分割边界。为此，我们提出了一种稳健的协同感知框架，称为专家驱动的检测与鸟瞰视图分割（CoDS）。为了解决融合质量中的空间不一致性，我们首先引入了协同可靠性图（CoRM），以明确量化特征质量分布。基于CoRM，我们设计了语义专家混合（S-MoE）模块，以提取针对不一致特征需求的差异化特征。最后，为了进一步减轻特征噪声的降解，双向任务互补交互（BTCI）通过双向注入来精炼任务感知特征。在OPV2V和V2V4Real数据集上的大量实验表明，我们的CoDS在这两项任务上超越了现有基线，并在多源噪声下保持了稳定的鲁棒性。代码链接： https://github.com/JinlongW128/CoDS 和 https://openi.pcl.ac.cn/OpenAIDriving/CoDS。

View on arXiv Download PDF AI Translation

cs.CV / 37 / 2608.14112

Fixed-Budget Gaussian Volume Encoding with Structure-Aware Allocation

固定预算高斯体积编码与结构感知分配

Martin, Michael R., Insley, Joseph, Mateevitsi, Victor A., Rizzi, Silvio, Ma, Kwan-Liu

Abstract

Scientific simulations often produce scalar volumes faster than they can be stored, transferred, and loaded, while in situ reduction must use only a limited share of simulation resources. This work encodes scalar fields as anisotropic Gaussian primitives under a fixed budget. The complete primitive set is allocated analytically from local field structure, including position, orientation, and shape, then refined directly against the scalar field without densification, pruning, or count changes. The selected budget determines encoded storage before refinement and, together with the iteration schedule, provides a controllable refinement-time budget. In a controlled benchmark, truncation-aware field evaluation reduces encoding time by up to 51x; 1.4 million Gaussians encode a billion-voxel volume in at most four minutes on one desktop GPU, with reduced-iteration refinement completing in under one minute. Across five datasets spanning 2.1 million to 1.1 billion evaluated voxels, compression-useful configurations achieve 15.0-38.7 dB PSNR at compression ratios from 2.2x to over 40,000x. Pre-encoding structure statistics characterize fields for which one-shot allocation yields limited gains from additional capacity. Because primitives retain scalar attributes rather than baked appearance, a single compact model serves every subsequent visualization state - supporting post-hoc transfer-function, colormap, lighting, and viewpoint changes without re-encoding.

Chinese Translation

科学模拟通常以比存储、传输和加载更快的速度生成标量体积，而原位降维必须仅使用有限的模拟资源。本文在固定预算下将标量场编码为各向异性高斯原语。完整的原语集根据局部场结构（包括位置、方向和形状）进行解析分配，然后直接针对标量场进行精炼，而无需进行密集化、修剪或计数变化。所选预算决定了精炼前的编码存储，并与迭代计划一起提供可控的精炼时间预算。在一个受控基准测试中，考虑截断的场评估将编码时间减少了多达51倍；140万个高斯原语在最多四分钟内编码十亿体素体积，减少迭代的精炼在一分钟内完成。在五个数据集中，评估体素数量从210万到11亿，压缩有用的配置在压缩比从2.2倍到超过40,000倍的情况下实现了15.0-38.7 dB的峰值信噪比（PSNR）。预编码结构统计特征化了那些一次性分配对额外容量收益有限的场。由于原语保留了标量属性而不是烘焙外观，因此单个紧凑模型可以服务于每个后续可视化状态——支持后期转移函数、色图、照明和视点变化，而无需重新编码。

View on arXiv Download PDF AI Translation

cs.CV / 38 / 2608.14136

HiCo-GS: Hierarchical Context Aggregation and Geometric Consistency for Octree Gaussian Splatting

HiCo-GS：用于八叉树高斯溅射的层次上下文聚合与几何一致性

Zhang, Wei, Yu, Shengkai, Gong, Shiqiang, Zhang, Qi, Li, Qiang, Wang, Qi

Abstract

Octree-based anchor Gaussian Splatting has emerged as a scalable representation for city-scale novel view synthesis, where multi-level anchors adaptively capture scene content from coarse building structures to fine architectural details. However, we identify a fundamental limitation in existing methods: cross-level feature isolation, where each level's anchor features are optimized independently with no inter-level communication, causing color drift on building facades and over-smoothing in textured regions. We present HiCo-GS, a high-fidelity reconstruction framework with two complementary modules. Cross-Level Context Aggregation (CLCA) enables bidirectional hierarchical prior injection by leveraging the octree's spatial containment structure to aggregate per-level context vectors into parent-self-child triplets, fused via a lightweight MLP with residual connection. Coarse-level structural priors flow down to inform fine-level anchors, while fine-level detail statistics feed back to prevent over-smoothing, at negligible computational overhead. Depth-Normal Geometric Consistency (DNGC) regularization enforces agreement between rendered normals and depth-derived normals through an alpha-weighted consistency loss, complemented by edge-aware smoothness losses with progressive warmup that exploit the strong planar priors ubiquitous in urban geometry to suppress floating artifacts. We further introduce the China-Pagoda dataset comprising 8 ancient Chinese pagodas with over 1,200 images each, featuring dense ornamental carvings, curved multi-layer eaves, and repetitive fine-grained textures. Extensive experiments on Mill19, UrbanScene3D, MatrixCity, and China-Pagoda demonstrate that HiCo-GS achieves state-of-the-art rendering quality and substantially cleaner geometry across real-world and synthetic urban benchmarks.Code: https://github.com/WZ-CS/HiCo-GS.

Chinese Translation

基于八叉树的锚点高斯溅射已成为城市规模新视角合成的可扩展表示，其中多层锚点自适应地捕捉从粗略建筑结构到精细建筑细节的场景内容。然而，我们发现现有方法存在一个基本限制：跨层特征隔离，即每一层的锚点特征独立优化，缺乏层间通信，导致建筑立面的颜色漂移和纹理区域的过度平滑。我们提出了HiCo-GS，一个高保真重建框架，包含两个互补模块。跨层上下文聚合（Cross-Level Context Aggregation, CLCA）通过利用八叉树的空间包含结构，聚合每层上下文向量为父-自-子三元组，借助轻量级的残差连接多层感知器（MLP）进行融合，实现双向层次先验注入。粗层结构先验向下流动以通知细层锚点，而细层细节统计则反馈以防止过度平滑，计算开销微乎其微。深度-法线几何一致性（Depth-Normal Geometric Consistency, DNGC）正则化通过加权一致性损失强制渲染法线与深度导出的法线之间的一致性，辅以边缘感知平滑损失和逐步预热，利用城市几何中普遍存在的强平面先验来抑制浮动伪影。我们进一步引入了中国宝塔数据集，包括8座古代中国宝塔，每座宝塔拥有超过1200张图像，特征包括密集的装饰雕刻、弯曲的多层屋檐和重复的细粒度纹理。在Mill19、UrbanScene3D、MatrixCity和中国宝塔的广泛实验中，HiCo-GS实现了最先进的渲染质量，并在真实世界和合成城市基准测试中显著提高了几何清晰度。代码：https://github.com/WZ-CS/HiCo-GS。

View on arXiv Download PDF AI Translation

cs.CV / 39 / 2608.14138

SPARGen: Unifying Spatial Perception and Reasoning through Native Multimodal Generation

SPARGen：通过原生多模态生成统一空间感知与推理

Quan, Jinsheng, Li, Jianhua, Xie, Siyi, Shi, Xuanke, Deng, Kewang, Chen, Zukai, Shao, Feifei, Yang, Lei, Wang, Quan, Luo, Yawei

Abstract

Spatial perception and reasoning from visual observations require recovering geometric structure, establishing correspondences, and understanding spatial relations. Existing approaches typically address these capabilities separately using task-specific architectures or external geometric modules, limiting knowledge transfer among complementary representations of the same physical scene. We introduce SPARGen, a unified multimodal framework that casts 3D reconstruction, dense correspondence, and spatial reasoning as instruction-conditioned generation tasks. SPARGen serializes compact structured and linguistic outputs as token sequences while generating dense geometric fields in image-aligned forms, enabling spatial supervision to jointly shape shared representations within a native multimodal generative model. Experiments across benchmarks for 3D reconstruction, correspondence, and spatial reasoning show that SPARGen achieves competitive performance across heterogeneous spatial tasks within a single native multimodal generative framework.

Chinese Translation

从视觉观察中进行空间感知和推理需要恢复几何结构、建立对应关系以及理解空间关系。现有方法通常使用特定任务的架构或外部几何模块分别解决这些能力，这限制了同一物理场景的互补表示之间的知识转移。我们提出了SPARGen，一个统一的多模态框架，将3D重建、密集对应和空间推理视为指令条件生成任务。SPARGen将紧凑的结构化和语言输出序列化为标记序列，同时以图像对齐的形式生成密集几何场，使空间监督能够共同塑造原生多模态生成模型中的共享表示。在3D重建、对应和空间推理的基准测试中进行的实验表明，SPARGen在单一原生多模态生成框架内，在异构空间任务中实现了具有竞争力的性能。

View on arXiv Download PDF AI Translation

cs.CV / 40 / 2608.14142

PISA: A Pseudo-Individual Source-Domain Feature Adaptation Framework for Test-Time Open-Vocabulary Object Detection

PISA：一种伪个体源域特征适应框架用于测试时开放词汇物体检测

He, Ziyan, Yang, Xiongtai, Wang, Tao

Abstract

Open-vocabulary object detection test-time adaptation (OVOD-TTA) aims to address the performance degradation that pre-trained base models suffer when encountering image-domain shifts. Existing source-free OVOD-TTA methods rely either on refined test-time information for re-scoring or on pseudo-labels for self-training, leading to significant accuracy degradation when initial predictions are poor. Meanwhile, most conventional source-domain estimation methods recover abstract, sparse representations suitable for the classification task, but fail to capture the dense, concrete features required for detection. To address these issues, we propose PISA, a novel source-free OVOD-TTA method that can be seamlessly integrated into open-vocabulary visual backbones. The core components of our method are the Corruption-Invariant Feature Extractor (CIFE), the Feature Alignment Module (FAM), and a multi-scale alignment framework (BAA). To capture detection-suitable features, we develop CIFE to exploit the invariance of CLIP's visual features across corrupted images, ensuring robustness against various corruptions. We further develop FAM and BAA for the pre-training and adaptation to transform the corruption-invariant features into pseudo-individual source-domain features that are close to the original source-domain features. In this way, dense and concrete pseudo-individual source-domain features are used for supervision instead of unreliable pseudo-label signals. Experiments on the corrupted VOC-C, COCO-C, and LVIS-C benchmarks across three base models demonstrate that PISA substantially improves both the localization precision and the category recognition accuracy of the original models. Notably, PISA achieves state-of-the-art performance without requiring access to source-domain data, surpassing existing methods by 3.92% in AP@50% on COCO-C.

Chinese Translation

开放词汇物体检测测试时适应（OVOD-TTA）旨在解决预训练基础模型在遇到图像域偏移时性能下降的问题。现有的无源OVOD-TTA方法要么依赖于精细的测试时信息进行重新评分，要么依赖于伪标签进行自我训练，这在初始预测不佳时导致显著的准确性下降。同时，大多数传统的源域估计方法恢复了适合分类任务的抽象、稀疏表示，但未能捕捉到检测所需的密集、具体特征。为了解决这些问题，我们提出了PISA，这是一种新颖的无源OVOD-TTA方法，可以无缝集成到开放词汇视觉骨干网络中。我们方法的核心组件包括抗干扰特征提取器（CIFE）、特征对齐模块（FAM）和多尺度对齐框架（BAA）。为了捕捉适合检测的特征，我们开发了CIFE，以利用CLIP的视觉特征在受损图像中的不变性，确保对各种干扰的鲁棒性。我们进一步开发了FAM和BAA用于预训练和适应，将抗干扰特征转化为接近原始源域特征的伪个体源域特征。通过这种方式，密集且具体的伪个体源域特征被用于监督，而不是不可靠的伪标签信号。在对受损的VOC-C、COCO-C和LVIS-C基准测试的三种基础模型进行的实验中，PISA显著提高了原始模型的定位精度和类别识别准确性。值得注意的是，PISA在不需要访问源域数据的情况下实现了最先进的性能，在COCO-C上的AP@50%超越了现有方法3.92%。

View on arXiv Download PDF AI Translation

cs.CV / 41 / 2608.14144

Self-Supervised Visual On-Policy Distillation

自监督视觉在线蒸馏

Li, Yijiang, Liang, Yijun, Tian, Yunjie, Wang, Bingyang, Zhang, Ke, Yin, Zhenfei, Fu, Di, Torr, Philip, Vasconcelos, Nuno

Abstract

Visual on-policy distillation relies heavily on an informative teacher-student asymmetry, through either a larger, stronger teacher or privileged supervision, such as reference answers or ground-truth regions of interest. This raises a fundamental question: where can informative asymmetry come from when nothing privileged is available? We answer this by inverting where the asymmetry comes from. Rather than adding privileged information to the teacher, we subtract information from the student. This asymmetry creates the same effective learning signal for free as a teacher with access to information unavailable to the student, without ground-truth annotations, rewards, or a separate stronger teacher model. Building on this principle, we introduce Self-Supervised Visual On-Policy Distillation (S$^2$VOPD), a simple yet effective method that constructs on-policy learning signals from asymmetric augmented views. S$^2$VOPD distills the teacher's distribution conditioned on the original image on-policy into the student distribution conditioned on a strongly augmented view of the same image. We systematically explore a broad design space of visual augmentations and uncover that (1) asymmetry matters: all four augmentation families improve performance, while symmetric self-distillation degrades it; (2) strength matters: performance peaks at a moderate strength; and (3) the gap must remain task-consistent: augmentations that completely remove the question-relevant evidence can induce large but uninformative discrepancies. Across six fine-grained perception benchmarks, S$^2$VOPD improves Qwen3.5-4B from 70.7% to 77.4%, above all open-source models compared, up to Qwen3-VL at 235B, and surpasses GPT-5.4. While holding training data the same, it recovers 96% of the improvement achieved by methods with privileged information. Website is at https://williamium3000.github.io/s2vopd

Chinese Translation

视觉在线蒸馏在很大程度上依赖于信息丰富的教师-学生不对称性，这可以通过更大、更强的教师或特权监督（例如参考答案或真实感兴趣区域）来实现。这引发了一个根本性的问题：当没有特权信息可用时，信息丰富的不对称性来自何处？我们通过逆转不对称性来源来回答这个问题。我们不是向教师添加特权信息，而是从学生中减去信息。这种不对称性创造了与获取学生无法获得的信息的教师相同的有效学习信号，而无需真实标注、奖励或单独的更强教师模型。在此原则基础上，我们提出了自监督视觉在线蒸馏（Self-Supervised Visual On-Policy Distillation，S$^2$VOPD），这是一种简单而有效的方法，通过不对称增强视图构建在线学习信号。S$^2$VOPD将基于原始图像的教师分布在线蒸馏到基于同一图像的强增强视图的学生分布。我们系统地探索了广泛的视觉增强设计空间，并发现（1）不对称性很重要：所有四种增强类别都提高了性能，而对称自蒸馏则降低了性能；（2）强度很重要：性能在适度强度时达到峰值；（3）差距必须保持任务一致性：完全去除与问题相关证据的增强可能会引发大的但无信息的差异。在六个细粒度感知基准测试中，S$^2$VOPD将Qwen3.5-4B的性能从70.7%提升至77.4%，超过所有比较的开源模型，甚至超过Qwen3-VL（235B），并超越GPT-5.4。在保持训练数据不变的情况下，它恢复了通过具有特权信息的方法所获得的96%的提升。网站地址为 https://williamium3000.github.io/s2vopd

View on arXiv Download PDF AI Translation

cs.CV / 42 / 2608.14146

CSG-Mamba: A Convolutional Scoring Gating Vision State Space Network for Endoscopic Polyp Segmentation

CSG-Mamba：一种用于内窥镜息肉分割的卷积评分门控视觉状态空间网络

Wang, Yuliang, Wu, Jiaqi, Song, Jiaye, Ren, Shuxia

Abstract

Accurate polyp segmentation is critical for computer-aided colonoscopy, yet endoscopic images often contain low-contrast boundaries, mucosal texture interference, specular highlights, and device-dependent appearance shifts. Vision State Space Models (SSMs) provide efficient long-range modeling with linear complexity, but existing Vision Mamba segmentation models typically convert 2D features into 1D scanning sequences, which may weaken local geometric continuity and over-smooth irregular contours. We propose CSG-Mamba, a convolutional scoring gating Vision State Space network for endoscopic polyp segmentation. Built on a VM-UNet-style asymmetric U-shaped encoder-decoder, CSG-Mamba inserts a Convolutional Scoring Gating (CSG) module at the semantically rich bottleneck. CSG generates a local spatial score map through pointwise and large-kernel depthwise convolutions and recalibrates state-space features by multiplicative gating. Experiments with three random seeds show that CSG-Mamba achieves 0.9220 Dice and 15.87 HD95 on Kvasir-SEG, and 0.7418 Dice and 0.6570 mIoU on CVC-ColonDB, outperforming the baselines on most overlap and recall metrics while maintaining competitive boundary accuracy.

Chinese Translation

准确的息肉分割对于计算机辅助结肠镜检查至关重要，但内窥镜图像通常包含低对比度边界、黏膜纹理干扰、镜面高光以及设备依赖的外观变化。视觉状态空间模型（SSMs）提供了线性复杂度的高效长程建模，但现有的视觉Mamba分割模型通常将2D特征转换为1D扫描序列，这可能削弱局部几何连续性并过度平滑不规则轮廓。我们提出了CSG-Mamba，一种用于内窥镜息肉分割的卷积评分门控视觉状态空间网络。CSG-Mamba基于VM-UNet风格的不对称U型编码器-解码器，在语义丰富的瓶颈处插入了卷积评分门控（CSG）模块。CSG通过逐点卷积和大核深度卷积生成局部空间评分图，并通过乘法门控重新校准状态空间特征。使用三个随机种子的实验表明，CSG-Mamba在Kvasir-SEG上达到了0.9220的Dice系数和15.87的HD95，在CVC-ColonDB上达到了0.7418的Dice系数和0.6570的mIoU，在大多数重叠和召回指标上优于基线，同时保持了竞争性的边界准确性。

View on arXiv Download PDF AI Translation

cs.CV / 43 / 2608.14148

SCVIB: Editable State-Conditioned Visual Instance Binding forMulti-Turn Personalized Localization

SCVIB：可编辑的状态条件视觉实例绑定用于多轮个性化定位

Yang, Xiongtai, He, Ziyan, Wang, Tao

Abstract

We introduce editable state-conditioned visual instance binding, a multi-turn localization setting in which several support-defined instances are introduced across turns and protocol-defined state events determine the final target. We instantiate this setting as SCVIB, comprising 1,050 manually verified support--query base pairs and 1,500 episodes spanning five visual domains, three difficulty levels, and four target-state dependency groups. Direct Seq-free inference reaches only 60.13\% [email protected], indicating that resolving the final reference does not ensure effective use of the corresponding visual evidence for query-side localization. We address this gap with TT-VG (Transition-Tree Visual Grounding), which combines a Target-State Transition Tree (TSTT) with Visual Evidence Grounding Adaptation (VEGA). TSTT compiles the visible interaction into protocol-defined events, executes them over versioned target states, and resolves the final-query reference to the corresponding support evidence. Adapted on trajectory-derived same-instance pairs, VEGA performs support-conditioned grounding of the resolved instance using a Visual Evidence Package. TT-VG reaches 70.27\% [email protected]; under matched target resolution, VEGA exceeds the strongest comparison method by 16.20 points. Gains over direct inference are largest on Counter-Recency and Rollback, which require routing to non-latest or restored support evidence. Together, these results establish SCVIB as a controlled testbed and highlight the effective use of resolved support evidence for query-side same-instance localization as a central challenge in multi-turn personalized localization.

Chinese Translation

我们提出了可编辑的状态条件视觉实例绑定，这是一种多轮定位设置，其中在多个回合中引入了若干支持定义的实例，而协议定义的状态事件决定最终目标。我们将这一设置实例化为SCVIB，包括1,050对手动验证的支持-查询基础对和1,500个跨越五个视觉领域、三个难度级别和四个目标状态依赖组的情节。直接的无序列推理仅达到60.13%的[email protected]，表明解决最终参考并不能确保有效利用相应的视觉证据进行查询侧定位。我们通过TT-VG（过渡树视觉定位）来解决这一问题，该方法结合了目标状态过渡树（TSTT）和视觉证据定位适应（VEGA）。TSTT将可见交互编译为协议定义的事件，在版本化的目标状态上执行这些事件，并将最终查询参考解析为相应的支持证据。VEGA在基于轨迹的同实例对上进行适应，使用视觉证据包对解析出的实例进行支持条件定位。TT-VG达到了70.27%的[email protected]；在匹配目标解析下，VEGA超过了最强比较方法16.20分。相较于直接推理，在反向最近性和回滚上获得的提升最大，这些情况需要路由到非最新或恢复的支持证据。总的来说，这些结果确立了SCVIB作为一个受控测试平台，并强调了解析的支持证据在查询侧同实例定位中的有效利用是多轮个性化定位中的一个核心挑战。

View on arXiv Download PDF AI Translation

cs.CV / 44 / 2608.14172

Concept Guidance: Precise, Training-Free Latent Control for Text-to-Image Generation

概念引导：精确的无训练潜在控制用于文本到图像生成

Röhrich, Nikolai, Hans, Isabell, Krause, Felix, Ommer, Björn

Abstract

Text-to-image diffusion models have two major drawbacks that severely limit their practical utility: (1) standard models lack an intrinsic mechanism for continuous, concept-specific guidance (e.g., for precisely controlling how aesthetically pleasing an image looks), and (2) they lack reliability for tasks requiring high local coherence (e.g., generating text or human hands). To tackle these issues, we introduce a novel notion of concept-wise mutual information and find large, concept-dependent differences between individual layers, demonstrating that the generation of specific structures is localized in distinct parts of the network. We exploit this insight by reinforcing the impact of concept-relevant layers in Concept Guidance (CoG), a precise, target-specific guidance method that works for models out-of-the-box without additional training, external models, gradients, or prompt engineering. CoG first quantifies each layer's concept-specific impact and then guides denoising using a weighted combination of predictions generated with concept-relevant layers skipped. We demonstrate performance increases across various targets and popular models like PixArt-alpha, SD3, SD3.5, and FLUX.1-dev. Code is available at https://github.com/CompVis/concept_guidance

Chinese Translation

文本到图像的扩散模型存在两个主要缺陷，严重限制了其实际应用： (1) 标准模型缺乏连续的、特定概念的内在引导机制（例如，精确控制图像的美观程度），以及 (2) 在需要高局部一致性的任务中缺乏可靠性（例如，生成文本或人手）。为了解决这些问题，我们引入了概念互信息的新概念，并发现个别层之间存在显著的、依赖于概念的差异，证明特定结构的生成是局部化在网络的不同部分。我们利用这一洞察，通过在概念引导（Concept Guidance, CoG）中强化与概念相关的层的影响，这是一种精确的、目标特定的引导方法，能够在不需要额外训练、外部模型、梯度或提示工程的情况下，直接应用于模型。CoG首先量化每一层的概念特定影响，然后使用跳过与概念相关层生成的预测的加权组合来引导去噪。我们在多个目标和流行模型（如PixArt-alpha、SD3、SD3.5和FLUX.1-dev）上展示了性能的提升。代码可在 https://github.com/CompVis/concept_guidance 获取。

View on arXiv Download PDF AI Translation

cs.CV / 45 / 2608.14178

LightTeaNet: A Weakly Supervised Lightweight CNN for Multi-Label Tea Leaf Disease Detection and Localization

LightTeaNet：一种弱监督轻量级卷积神经网络用于多标签茶叶病害检测与定位

Chowdhury, Naif Haider, Rahim, Md, Hasan, Syed Farhan, Hasan, Murad, Bhattacharjee, Prithwiraj

Abstract

Tea is known as an important crop in many parts of South and Southeast Asia, yet the production of tea is still hampered by the multiple diseases that decrease the quantity and quality. Traditional methods of inspection, which are manual, are not consistent, labor-intensive, and depend on extensive monitoring. This paper introduces a lightweight convolutional neural network (CNN) designed for weakly supervised multi-label classification and disease localization in tea leaves called LightTeaNet. LightTeaNet learns directly from image-level labels and employs Class Activation Mapping (CAM) to localize disease-affected regions automatically, unlike conventional object detection models such as YOLO, which require extensive bounding box annotations. For Parameter efficiency, the network integrates Depthwise Separable Convolutions, and for enhanced feature discrimination, it integrates Channel Attention. LightTeaNet has achieved a Precision of 0.9615, a Recall of 0.8772, and an F1-score of 0.9179, while it shows [email protected]=0.1810 without any manual annotations, which delivers a competitive localization performance in the experimental results. These results validate the model as an interpretable as well as a resource-efficient framework for intelligent disease monitoring in agriculture.

Chinese Translation

茶叶被认为是南亚和东南亚许多地区的重要作物，但茶叶的生产仍然受到多种疾病的影响，从而降低了产量和质量。传统的人工检查方法不一致、劳动密集，并且依赖于广泛的监测。本文介绍了一种名为LightTeaNet的轻量级卷积神经网络（CNN），旨在进行弱监督的多标签分类和茶叶病害定位。LightTeaNet直接从图像级标签中学习，并采用类激活映射（Class Activation Mapping, CAM）自动定位受病害影响的区域，这与传统的物体检测模型（如YOLO）不同，后者需要大量的边界框注释。为了提高参数效率，该网络集成了深度可分离卷积（Depthwise Separable Convolutions），并为增强特征区分能力集成了通道注意力（Channel Attention）。LightTeaNet在没有任何手动注释的情况下，达到了0.9615的精确率、0.8772的召回率和0.9179的F1分数，同时在实验结果中显示[email protected]=0.1810，展现出具有竞争力的定位性能。这些结果验证了该模型作为一种可解释且资源高效的农业智能病害监测框架的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 46 / 2608.14226

RankT2I: A Submodular Framework for Discovering Interpretable and Diverse Semantics in Text-to-Image Models

RankT2I：一个用于发现文本到图像模型中可解释和多样化语义的子模框架

Allada, Ritika, Yanardag, Pinar

Abstract

Recent advances in text-to-image (T2I) models have revolutionized the field of image generation and editing. However, identifying semantics that a T2I model can successfully edit in an image continues to be a challenging task. Most existing approaches require users to manually specify semantics to modify a particular image, a time-consuming process that often involves extensive trial and error. In this paper, we present RankT2I, a novel, training-free, and model-agnostic framework that automates the discovery of editable semantics in diffusion and FLUX-based models. Given a visual domain, we first utilize a multimodal vision-language model to gather a broad set of candidate semantics. We then frame semantic discovery as a set selection problem and use a submodular objective to identify semantics that are relevant, editable, and diverse. Our method helps users efficiently identify a wide range of semantics for text-to-image editing models across several domains while outperforming existing methods.

Chinese Translation

最近，文本到图像（T2I）模型的进展彻底改变了图像生成和编辑领域。然而，识别T2I模型在图像中能够成功编辑的语义仍然是一项具有挑战性的任务。现有的大多数方法要求用户手动指定要修改特定图像的语义，这一过程耗时且通常涉及大量的反复试验。在本文中，我们提出了RankT2I，一个新颖的、无训练的、模型无关的框架，自动发现扩散和基于FLUX的模型中的可编辑语义。给定一个视觉领域，我们首先利用多模态视觉-语言模型收集广泛的候选语义集。然后，我们将语义发现框架化为一个集合选择问题，并使用子模目标来识别相关的、可编辑的和多样化的语义。我们的方法帮助用户高效识别多个领域中用于文本到图像编辑模型的广泛语义，同时优于现有方法。

View on arXiv Download PDF AI Translation

cs.CV / 47 / 2608.14235

AppleScab-LT: A Longitudinal Real-Field Apple Scab Dataset for Temporal Disease Progression Analysis

AppleScab-LT：一个用于时间性病害进展分析的纵向真实场地苹果黑腐病数据集

Hilal, Aamir, Sofi, Shabir Ahmad, Goel, Neeraj

Abstract

The development of reliable plant disease monitoring systems is constrained by limited longitudinal datasets capturing disease progression under natural field conditions. Although existing plant disease datasets have advanced image-based recognition, most consist of static images acquired at a single time point, limiting analysis of temporal disease evolution and severity progression. To address this gap, this study presents AppleScab-LT, a longitudinal real-field dataset developed to monitor apple scab progression through repeated observations of individually tracked infected leaves. Guided by a research-question-driven framework, the dataset was systematically developed, validated, and characterized for reliable longitudinal disease analysis. AppleScab-LT was constructed through systematic orchard monitoring under natural environmental conditions, incorporating longitudinal leaf tracking, expert-guided disease verification, polygon-based annotation, leaf isolation, disease severity quantification, and temporal sequence construction. A comprehensive quality assurance framework, including standardized annotation protocols, expert validation, automated integrity checks, sequence-level verification, and temporal consistency analysis, was applied throughout curation. The dataset contains 21 longitudinal leaf sequences, 2,101 high-resolution images, and 264 progressive temporal samples from repeated monitoring of same infected leaves. It captures variability in severity accumulation, progression rates, monitoring duration, and inter-leaf progression. Quantitative disease descriptors based on pixel severity, color-intensity severity, and normalized relative severity provide standardized measurements for temporal disease analysis. AppleScab-LT provides a reliable resource for temporal disease intelligence, disease progression modelling, precision agriculture, and future crop health monitoring

Chinese Translation

可靠的植物病害监测系统的发展受到自然田间条件下捕捉病害进展的纵向数据集有限的制约。尽管现有的植物病害数据集在基于图像的识别方面取得了进展，但大多数数据集仅由在单一时间点获取的静态图像组成，这限制了对病害时间演变和严重性进展的分析。为了解决这一问题，本研究提出了AppleScab-LT，这是一个旨在通过对单独追踪的感染叶片进行重复观察来监测苹果黑腐病进展的纵向真实场地数据集。在研究问题驱动的框架指导下，该数据集经过系统开发、验证和特征化，以便进行可靠的纵向病害分析。AppleScab-LT是在自然环境条件下通过系统的果园监测构建的，涵盖了纵向叶片追踪、专家指导的病害验证、多边形标注、叶片隔离、病害严重性量化和时间序列构建。整个数据集的整理过程中应用了全面的质量保证框架，包括标准化的标注协议、专家验证、自动完整性检查、序列级验证和时间一致性分析。该数据集包含21个纵向叶片序列、2101张高分辨率图像和264个来自对同一感染叶片的重复监测的渐进时间样本。它捕捉了严重性积累、进展速度、监测持续时间和叶片间进展的变异性。基于像素严重性、颜色强度严重性和标准化相对严重性的定量病害描述符为时间性病害分析提供了标准化的测量。AppleScab-LT为时间性病害智能、病害进展建模、精准农业和未来作物健康监测提供了可靠的资源。

View on arXiv Download PDF AI Translation

cs.CV / 48 / 2608.14243

Zero-Shot Skeleton-Based Action Anticipation

零样本基于骨架的动作预测

Wang, Hongsong, Yan, Pengbo, Zhang, Yang, Lai, Qiuxia

Abstract

Action anticipation (AA) aims to recognize ongoing human or humanoids actions from partial observations, enabling robots to predict intentions before the actions are completed. Although skeleton-based AA offers efficiency advantages, existing approaches assume that all action classes are seen during training, which limits their deployment in real-world scenarios where novel actions inevitably arise. To address this gap, we study the new task of Zero-Shot Skeleton-Based Action Anticipation (ZS-SkAA). This task requires recognizing unseen action classes using only limited early-stage skeleton sequences, combining the challenges of partial observations, temporal dynamics, and zero-shot generalization. To establish foundational research for ZS-SkAA, we introduce:(1) A baseline model comprising a spatio-temporal feature extractor and a mutual information estimation and maximization module. This baseline model explicitly aligns partial visual features with semantic class embeddings across modalities by estimating and maximizing their mutual information, enhancing generalization to unseen classes.(2) A benchmark protocol using the NTU RGB+D dataset, which is adapted for rigorous ZS-SkAA evaluation. Experiments demonstrate the effectiveness of our model as a strong baseline for ZS-SkAA, achieving high zero-shot accuracy on NTU RGB+D. This work establishes ZS-SkAA as a vital research direction for real-world systems requiring generalization to novel actions.

Chinese Translation

动作预测（AA）旨在从部分观察中识别正在进行的人类或类人动作，使机器人能够在动作完成之前预测意图。尽管基于骨架的AA具有效率优势，但现有方法假设在训练过程中所有动作类别均已见过，这限制了它们在现实场景中的应用，因为新颖动作不可避免地会出现。为了解决这一问题，我们研究了零样本基于骨架的动作预测（ZS-SkAA）这一新任务。该任务要求仅使用有限的早期骨架序列来识别未见过的动作类别，结合了部分观察、时间动态和零样本泛化的挑战。为了为ZS-SkAA奠定基础研究，我们引入了：(1) 一个基线模型，包括一个时空特征提取器和一个互信息估计与最大化模块。该基线模型通过估计和最大化部分视觉特征与语义类别嵌入在不同模态之间的互信息，显式地对齐这些特征，从而增强对未见类别的泛化能力。(2) 一个基准协议，使用NTU RGB+D数据集，经过调整以进行严格的ZS-SkAA评估。实验表明，我们的模型作为ZS-SkAA的强基线表现出色，在NTU RGB+D上实现了高零样本准确率。这项工作确立了ZS-SkAA作为需要对新颖动作进行泛化的现实系统的重要研究方向。

View on arXiv Download PDF AI Translation

cs.CV / 49 / 2608.14262

On the Robustness of Temporal Vision-Language Models for Surgical Endoscopy Videos

外科内窥镜视频的时间视觉-语言模型的鲁棒性研究

Rashid, Darakshan, Imam, Raza, Khan, Ufaq, Bilal, Muhammad, Ashraf, Shazad, Mahapatra, Dwarikanath, Yaqub, Mohammad, Khan, Muhammad Haris, Razzak, Imran, Lall, Brejesh, Maier-Hein, Lena, Xie, Yutong

Abstract

Temporal vision-language models (TVLMs) offer a reusable, prompt-based interface for surgical video understanding, yet, their robustness under clinically realistic acquisition artifacts in endoscopy remains insufficiently characterized. In practice, degradations such as defocus, haze, motion blur, noise, cautery smoke, and packet loss introduce structured distribution shifts which may compromise video-text alignment. We study the robustness of temporal VLMs under such shifts caused by corruptions in clip frames. We introduce Endo-C6, a compact corruption benchmark of six endoscopy-realistic perturbations evaluated at a fixed high severity, and apply it to public Gastrointestinal (GI) endoscopy and laparoscopic cholecystectomy videos. Under a standardized prompt protocol, we benchmark 3 recent surgical TVLM baselines and analyze robustness in both mean and worst-case settings, spanning 294 dataset-level evaluations. Finally, we present RobustEndoCLIP, obtained by few-shot parameter-efficient tuning with VeRA, outperforming existing TVLM baselines. Our findings show that off-the-shelf TVLMs can exhibit severe worst-case collapse under endoscopy-specific corruptions, whereas lightweight few-shot adaptation can substantially improve corrupted performance and robustness without changing the prompt-based interface. We expect Endo-C6 to support standardized robustness reporting and promote more reliable clinical vision-language systems.

Chinese Translation

时间视觉-语言模型（TVLMs）为外科视频理解提供了一种可重用的基于提示的接口，但它们在临床实际获取伪影下的鲁棒性仍然不足以充分表征。实际上，失焦、雾霾、运动模糊、噪声、烧灼烟雾和数据包丢失等降质现象引入了结构化的分布偏移，这可能会影响视频与文本的对齐。我们研究了在由剪辑帧中的腐败引起的这种偏移下，时间视觉语言模型的鲁棒性。我们引入了Endo-C6，这是一个紧凑的腐败基准，包含六种与内窥镜相关的扰动，并在固定的高严重性下进行评估，并将其应用于公共的胃肠道（GI）内窥镜和腹腔镜胆囊切除视频。在标准化的提示协议下，我们对三个最近的外科TVLM基线进行了基准测试，并分析了在平均和最坏情况下的鲁棒性，涵盖了294个数据集级别的评估。最后，我们提出了RobustEndoCLIP，通过与VeRA的少量样本参数高效调优获得，超越了现有的TVLM基线。我们的研究结果表明，现成的TVLM在特定于内窥镜的腐败下可能表现出严重的最坏情况崩溃，而轻量级的少量样本适应可以在不改变基于提示的接口的情况下显著改善腐败性能和鲁棒性。我们期望Endo-C6能够支持标准化的鲁棒性报告，并促进更可靠的临床视觉-语言系统。

View on arXiv Download PDF AI Translation

cs.CV / 50 / 2608.14281

Learning to Forecast Crop Growth from Earth Observation Data

从地球观测数据中学习预测作物生长

Senti, Dominik, Turkoglu, Mehmet Ozgur, Volpi, Michele, Aasen, Helge

Abstract

Forecasting crop growth across agricultural landscapes is important for improving the productivity, resilience, and operational management of farming systems. In this work, we investigate whether Earth observation time series and meteorological drivers can be used to predict future canopy development at country scale. We focus on winter wheat and formulate crop growth prediction as forecasting future leaf area index (LAI) trajectories beyond the last available Sentinel-2 observation. We evaluate this task on a multi-year dataset which spans the entire country of Switzerland, containing over 20 million pixel-level Sentinel-2-derived LAI time series paired with meteorological variables. Because cloud cover and revisit gaps leave LAI supervision sparse, models fit the few valid (cloud-free) LAI observations yet oscillate implausibly between them, producing trajectories no real canopy could follow. We introduce a lightweight unimodal shape regulariser which improves trajectory plausibility with negligible loss in accuracy. We compare deep learning sequence-to-sequence (Seq2Seq) models with classic machine learning baselines and show that Seq2Seq models generalise well across years, achieving $\mathrm{R}^2$ above 0.8 and consistently outperforming conventional approaches. Together, these results demonstrate that remote sensing and weather-driven sequence modelling can learn crop growth dynamics at landscape scale. S

Chinese Translation

在农业景观中预测作物生长对于提高农业系统的生产力、韧性和运营管理至关重要。在本研究中，我们探讨了是否可以利用地球观测时间序列和气象驱动因素来预测国家尺度上的未来冠层发展。我们重点关注冬小麦，并将作物生长预测表述为预测超出最后可用的 Sentinel-2 观测的未来叶面积指数（LAI）轨迹。我们在一个跨越整个瑞士的多年度数据集上评估这一任务，该数据集包含超过 2000 万个像素级的 Sentinel-2 派生的 LAI 时间序列，并与气象变量配对。由于云层覆盖和重访间隙导致 LAI 监督稀疏，模型只能拟合少量有效（无云）LAI 观测，但在这些观测之间不切实际地振荡，产生的轨迹无法被真实的冠层所遵循。我们引入了一种轻量级的单模态形状正则化器，能够在几乎不损失准确性的情况下改善轨迹的合理性。我们将深度学习序列到序列（Seq2Seq）模型与经典机器学习基线进行了比较，结果表明 Seq2Seq 模型在不同年份之间具有良好的泛化能力，$ ext{R}^2$ 值超过 0.8，并且始终优于传统方法。综合来看，这些结果表明，遥感和气象驱动的序列建模能够在景观尺度上学习作物生长动态。

View on arXiv Download PDF AI Translation

cs.CV / 51 / 2608.14282

MAGneT-3D: Monocular and Domain-Generalizable Temporal 3D Detection

MAGneT-3D：单目和领域通用的时序三维检测

Kotb, Mohamed, Meier, Johannes, Reich, Christoph, Dhaouadi, Oussema, Denninger, Luis, Cremers, Daniel

Abstract

Monocular temporal 3D detection aims to detect objects in 3D, given a monocular video. Query-based 3D detectors unify detection and cross-view association, but their learnable queries fit the spatial distribution of the training data (e.g., field-of-view). We show that this issue is especially severe when these models are applied to monocular video, hindering generalization to unseen datasets and environments. To address this limitation, we introduce MAGneT-3D, the first method for domain-generalized monocular temporal 3D object detection. Instead of relying on static learnable queries, we propose a Domain-Robust Anchor Generator (DRAG) approach that adaptively derives 3D proposals during inference. To further enable domain generalization, we propose a Temporal Refinement and Identity Merging (TRIM) strategy, reducing dependence on specific 3D proposals. To enable comprehensive domain-generalization evaluation, we establish a cross-dataset benchmark spanning nuScenes, Waymo, Lyft, and ONCE. Under zero-shot domain shifts, MAGneT-3D outperforms all baselines, improving NDS from 12.1% to 18.6% while also increasing in-domain accuracy.

Chinese Translation

单目时序三维检测旨在根据单目视频检测三维物体。基于查询的三维检测器统一了检测和跨视图关联，但其可学习的查询适应于训练数据的空间分布（例如，视场）。我们表明，当这些模型应用于单目视频时，这一问题尤其严重，阻碍了对未见数据集和环境的泛化。为了解决这一限制，我们提出了MAGneT-3D，这是首个用于领域通用的单目时序三维物体检测的方法。我们不再依赖静态的可学习查询，而是提出了一种领域鲁棒锚生成器（Domain-Robust Anchor Generator, DRAG）方法，该方法在推理过程中自适应地产生三维提议。为了进一步实现领域泛化，我们提出了一种时序精炼与身份合并（Temporal Refinement and Identity Merging, TRIM）策略，减少对特定三维提议的依赖。为了实现全面的领域泛化评估，我们建立了一个跨数据集基准，涵盖nuScenes、Waymo、Lyft和ONCE。在零样本领域转移下，MAGneT-3D超越了所有基线，将NDS从12.1%提升至18.6%，同时也提高了领域内的准确性。

View on arXiv Download PDF AI Translation

cs.CV / 52 / 2608.14286

Seeing Red, Thinking Bad: Color Bias in Vision Language Models

看见红色，思考不良：视觉语言模型中的颜色偏见

Ide, Kohsuke, Yamada, Ryousuke, Fukuhara, Yoshihiro, Kataoka, Hirokatsu, Satoh, Yutaka

Abstract

Vision language models (VLMs) are increasingly used in industrial decision-making systems, such as recruitment support and recommendation. This motivates careful analysis of how VLMs process visual and textual information. In this work, we study how VLMs interpret text rendered as an image, and investigate the influence of visual styling biases. To this end, we introduce Stealth Visual Prompts, which subtly change visual styling of text, such as color and contrast, while preserving semantic content. Using these prompts, we systematically control the visual styling of words in text and measure their impact on the analysis performed by VLMs. We further analyze how such visual perturbations affect the latent representations of the vision encoder. From our experiments, we observed that coloring positive words in green consistently shifts sentiment predictions toward a positive direction. As a result, VLMs often fail to properly account for negative words present in the text. Our analysis suggests that this behavior is correlated with changes in the latent representations of the vision encoder induced by color variations. In addition, we show that reducing text--background contrast increases reliance on visually salient cues and leads to more incorrect Visual Question Answering (VQA) outputs. These results suggest that the visual styling of rendered text can guide VLMs' interpretation in ways that diverge from human semantic understanding. Project page: https://github.com/KohsukeIde/color-bias-vlm

Chinese Translation

视觉语言模型（VLMs）在工业决策系统中越来越多地被应用，例如招聘支持和推荐。这促使我们对VLMs如何处理视觉和文本信息进行仔细分析。在本研究中，我们探讨了VLMs如何解读作为图像呈现的文本，并研究视觉风格偏见的影响。为此，我们引入了隐形视觉提示（Stealth Visual Prompts），这些提示微妙地改变文本的视觉风格，如颜色和对比度，同时保持语义内容。通过这些提示，我们系统地控制文本中单词的视觉风格，并测量其对VLMs分析的影响。我们进一步分析了这些视觉扰动如何影响视觉编码器的潜在表示。从我们的实验中，我们观察到将积极词汇用绿色着色会持续地将情感预测向积极方向偏移。因此，VLMs往往未能正确考虑文本中存在的消极词汇。我们的分析表明，这种行为与颜色变化引起的视觉编码器潜在表示的变化相关。此外，我们还展示了降低文本与背景的对比度会增加对视觉显著线索的依赖，从而导致更多错误的视觉问答（Visual Question Answering, VQA）输出。这些结果表明，渲染文本的视觉风格可以以偏离人类语义理解的方式引导VLMs的解读。项目页面：https://github.com/KohsukeIde/color-bias-vlm

View on arXiv Download PDF AI Translation

cs.CV / 53 / 2608.14293

Conditional Neural Optimal Transport for Predicting Cellular Phenotypes from Molecular Structure

用于从分子结构预测细胞表型的条件神经最优传输

Avité, Gauthier, Sanchez-Renauld, Maxime, Bourriez, Nicolas, Genovesio, Auguste

Abstract

High-content microscopy enables systematic profiling of cellular responses to chemical perturbations, but the scale of the chemical space makes exhaustive phenotypic characterization experimentally infeasible. This motivates computational models that can predict image-derived phenotypes without acquiring the corresponding treated cells. We formulate molecule-induced phenotype prediction as an inductive conditional transport problem in image representation space. Given a negative-control phenotype and the structure of a molecule, we aim to predict the phenotype induced by the corresponding molecule. We first evaluate classical optimal transport baselines and show that static couplings do not yield useful predictions on large-scale phenotypic image datasets. We then introduce a molecule-conditioned Neural Optimal Transport (NOT) model with a Monge-Gap regularization training objective that learns to transport negative-control unperturbed phenotypes toward perturbed phenotypes using molecular structure as conditioning information. NOT recovers molecule-specific phenotypic effects while reducing microscopy-associated technical variation, thereby facilitating comparisons across experimental batches. On unseen active molecules, the model outperforms baseline approaches, demonstrating that chemically conditioned transport can generalize beyond the molecules observed during training. We identified the molecular encoder as the main limitation to this generalization, while transport in a compressed representation space improves performance and scalability. These results establish NOT as a promising framework for predicting cellular phenotypes from molecular structure and negative-control phenotypes, while highlighting the development of more informative molecular representations as a key direction for improving out-of-distribution performance.

Chinese Translation

高通量显微镜技术使得系统性地分析细胞对化学扰动的反应成为可能，但化学空间的规模使得全面的表型特征在实验上不可行。这促使了能够在不获取相应处理细胞的情况下预测图像派生表型的计算模型的开发。我们将分子诱导的表型预测形式化为图像表示空间中的归纳条件传输问题。给定一个负对照表型和一个分子的结构，我们旨在预测由相应分子诱导的表型。我们首先评估经典的最优传输基线，并表明静态耦合在大规模表型图像数据集上并未产生有用的预测。随后，我们引入了一种分子条件的神经最优传输（Neural Optimal Transport, NOT）模型，该模型具有Monge-Gap正则化训练目标，学习使用分子结构作为条件信息将负对照未扰动表型传输到扰动表型。NOT能够恢复特定分子的表型效应，同时减少与显微镜相关的技术变异，从而促进不同实验批次之间的比较。在未见过的活性分子上，该模型的表现优于基线方法，证明了化学条件传输能够超越训练期间观察到的分子进行泛化。我们确定分子编码器是这一泛化的主要限制，而在压缩表示空间中的传输则提高了性能和可扩展性。这些结果确立了NOT作为一个有前景的框架，用于从分子结构和负对照表型预测细胞表型，同时强调开发更具信息性的分子表示作为提高分布外性能的关键方向。

View on arXiv Download PDF AI Translation

cs.CV / 54 / 2608.14309

Spatial Message Passing in Language Space for Pathology Image Interpretation

语言空间中的空间消息传递用于病理图像解读

Yang, Jing-Cheng, Wang, Hao-Jung, Du, Jinhao, Hu, Yang, Tsai, Ming-shan, Rittscher, Jens, Li, Bin

Abstract

Multimodal Large Language Models (MLLMs) can generate pathological descriptions from histological images, but gigapixel Whole Slide Images (WSIs) exceed their visual context limits. The standard tiling workaround makes WSIs tractable yet severs the tissue neighborhoods that define tumor-stroma interfaces and morphology. We introduce Spatial Language Message Passing (SLMP), a framework that performs spatial reasoning entirely in language space, human-readable by construction. SLMP represents a WSI region as a spatial text graph: tiles are nodes initialized with MLLM descriptions, and edges encode spatial adjacency. For each tile, an LLM refines its description by integrating language messages from adjacent tiles under a shared aggregation policy that, on the tile grid, acts as an adaptive local kernel operating on text rather than learned embeddings. This policy is an inspectable prompt that can be refined from model-observed tissue phenotypes via textual gradients, enabling automatic semantic optimization from local cellular context to broader tissue morphology without fine-tuning MLLM weights. On representative HER2 and CAMELYON16 regions, SLMP improves tile-level tumor description accuracy in settings spanning general-purpose and pathology-specialized backbones, with gains of +3.3 to +19.6 percentage points. Random-neighbor ablations confirm that these gains stem from spatial context rather than additional text alone, and inspecting the optimized policies reveals interpretable, tissue-specific decision rules. Besides, without any weight updates or fine-tuning the backbone MLLM, SLMP substantially improves general-purpose MLLMs and narrows its gap to pathology-specialized counterparts, offering a transparent and flexible mechanism for incorporating spatial reasoning into MLLM-based pathology analysis.

Chinese Translation

多模态大型语言模型（MLLMs）能够从组织学图像生成病理描述，但千兆像素的全切片图像（WSIs）超出了它们的视觉上下文限制。标准的切片解决方案使得WSIs变得可处理，但却切断了定义肿瘤- stroma 接口和形态的组织邻域。我们提出了空间语言消息传递（SLMP），这是一个完全在语言空间中进行空间推理的框架，构建上可被人类理解。SLMP将WSI区域表示为一个空间文本图：切片作为节点，初始化为MLLM描述，边缘编码空间邻接关系。对于每个切片，LLM通过整合来自相邻切片的语言消息来优化其描述，采用共享的聚合策略，该策略在切片网格上作为一个自适应局部核在文本上操作，而不是在学习的嵌入上。这一策略是一个可检查的提示，可以通过文本梯度从模型观察到的组织表型进行优化，从而实现从局部细胞上下文到更广泛的组织形态的自动语义优化，而无需微调MLLM权重。在代表性的HER2和CAMELYON16区域中，SLMP在涵盖通用和病理专用骨干网络的设置中提高了切片级肿瘤描述的准确性，增幅为+3.3到+19.6个百分点。随机邻居消融实验确认这些增益源于空间上下文，而不仅仅是额外的文本，并且检查优化后的策略揭示了可解释的、特定于组织的决策规则。此外，在没有任何权重更新或微调骨干MLLM的情况下，SLMP显著改善了通用MLLM，并缩小了与病理专用模型之间的差距，提供了一种透明且灵活的机制，将空间推理纳入基于MLLM的病理分析中。

View on arXiv Download PDF AI Translation

cs.CV / 55 / 2608.14317

Intelligent Detection of Mechanical, Electrical, and Plumbing (MEP) Metrics Based on 2D Floor Plans

基于二维平面图的机械、电气和管道（MEP）指标智能检测

Mandhiratta, Tarandeep Singh, Zaman, ANK, Mawlood-Yunis, Abdul-Rahman

Abstract

This research developed a neural network-based model to extract various information from 2D floor plans. We detect lighting symbols, identify the appropriate type of light, and extract the associated texts with lights. The study aims to enable efficient floor designing and determining the number and type of lights needed per floor, i.e., allow efficient design and estimate the power requirement of the floor plan. The model was developed using Mask RCNN as the base. The images were annotated and converted into a Coco data format for training the model. The model achieved bbox\_mAP and segm\_mAP values of 0.7596 and 0.7111, respectively. It also performed well at different IoU thresholds, i.e., with bbox\_mAP 50 and segm\_mAP 75 values of 0.9850 and 0.9219, respectively. The developed model will help various industries, such as architecture and construction, to improve design time and create efficient workflows by automatically detecting Mechanical, Electrical, and Plumbing (MEP) objects from floor plans, and it is the first step towards building tools that will help energy-efficient building design.

Chinese Translation

本研究开发了一种基于神经网络的模型，用于从二维平面图中提取各种信息。我们检测照明符号，识别适当类型的灯具，并提取与灯具相关的文本。该研究旨在实现高效的楼层设计，并确定每层所需的灯具数量和类型，即实现高效设计并估算平面图的电力需求。该模型以 Mask RCNN 为基础进行开发。图像经过标注并转换为 Coco 数据格式以用于模型训练。模型在 bbox_mAP 和 segm_mAP 指标上分别达到了 0.7596 和 0.7111 的值。在不同的 IoU 阈值下，模型表现良好，bbox_mAP 50 和 segm_mAP 75 的值分别为 0.9850 和 0.9219。所开发的模型将帮助建筑和施工等多个行业提高设计效率，并通过自动检测平面图中的机械、电气和管道（MEP）对象来创建高效的工作流程，这是构建有助于节能建筑设计工具的第一步。

View on arXiv Download PDF AI Translation

cs.CV / 56 / 2608.14321

TRIAGE: Risk-Controlled Pseudo-Label Admission for Annotation-Efficient Semi-Supervised Retinal OCT Classification

TRIAGE：风险控制的伪标签录取用于注释高效的半监督视网膜OCT分类

Akash, Md Ashraful Hossen, Afroge, Shyla, Mamun, Abdullah Al, Morol, Md. Kishor, Liew, Tze Hui

Abstract

The advanced retinal disease diagnosing imaging modality, optical coherence tomography (OCT), encounters a lack of automation because of the high expenses for annotations performed by specialists. The use of SSL solves the problem of insufficient annotations using unlabeled B-scans; however, most of the current techniques for generating pseudo-labels are based on prediction confidence without considering the asymmetry between different types of errors. This paper proposes TRIAGE, a risk-controlled semi-supervised framework for OCT scans classification, which uses the concept of a patient-level conformal risk controller with an asymmetric cost matrix. TRIAGE unites three crucial modules: a hierarchical classifier that is capable of working with partially abnormal supervision of the disease subtypes, a patient-grouped conformal risk controller with primal-dual coverage control, and a context-aware Transformer teacher for cross-slice verification. On the dataset from Noor Eye Hospital (16,822 B-scans, 161 patients, and 554 volumes) with a test set of unseen patients, TRIAGE demonstrates 89.66% scan-level accuracy, 0.8805 macro-F1, 0.9641 macro-AUC, and an 8.34% under-grading rate when using only 20% of the labeled data. With only 5% of the labeled data, TRIAGE keeps 76.88% accuracy and a 0.1656 under-grading rate. Compared with the other six state-of-the-art semi-supervised methods, TRIAGE significantly outperforms them with ablation study demonstrating the contribution of each module in the overall framework performance (by 42.7% in terms of under-grading rate comparing to fixed threshold methods). TRIAGE demonstrates 98.00% accuracy for 3-class classification with 1% labeled data and 95.94% accuracy for 8-class classification with 10% labeled data on the OCT-C8 dataset.

Chinese Translation

先进的视网膜疾病诊断成像方式——光学相干断层扫描（OCT），由于专家进行注释的高昂费用，面临缺乏自动化的问题。使用半监督学习（SSL）可以通过未标记的B扫描解决注释不足的问题；然而，目前大多数生成伪标签的技术基于预测置信度，而未考虑不同类型错误之间的不对称性。本文提出了TRIAGE，一种用于OCT扫描分类的风险控制半监督框架，该框架利用患者级别的符合风险控制器和不对称成本矩阵的概念。TRIAGE结合了三个关键模块：一个能够处理疾病亚型部分异常监督的层次分类器，一个具有原始-对偶覆盖控制的患者分组符合风险控制器，以及一个用于跨切片验证的上下文感知Transformer教师。在来自诺尔眼科医院的数据集（16,822个B扫描，161名患者，554个体积）上，使用未见患者的测试集，TRIAGE在仅使用20%的标记数据时表现出89.66%的扫描级准确率，0.8805的宏F1值，0.9641的宏AUC值，以及8.34%的低估率。在仅使用5%的标记数据时，TRIAGE保持了76.88%的准确率和0.1656的低估率。与其他六种最先进的半监督方法相比，TRIAGE显著优于它们，消融研究表明了每个模块对整体框架性能的贡献（与固定阈值方法相比，低估率提高了42.7%）。在OCT-C8数据集上，TRIAGE在使用1%标记数据时实现了98.00%的三类分类准确率，在使用10%标记数据时实现了95.94%的八类分类准确率。

View on arXiv Download PDF AI Translation

cs.CV / 57 / 2608.14366

Weakly Supervised Polar Low Segmentation in Sentinel-1 SAR Imagery

哨兵-1 SAR 图像中的弱监督极地低压分割

Federici, Andrea, Grahn, Jakob, Boracchi, Giacomo, Bianchi, Filippo Maria

Abstract

Polar lows are intense maritime cyclones that form rapidly at high latitudes. Deep learning can detect them in Synthetic Aperture Radar (SAR) imagery, but pixel-level segmentation remains an open challenge. No pixel-level masks are available for training, and a polar low's extent is inherently subjective, with diffuse boundaries that even experts delineate inconsistently. We propose Constrained Region Erasing with Soft Targets (CREST), a Weakly Supervised Semantic Segmentation (WSSS) framework that generates masks solely from image-level labels. Our approach builds on Adversarial Erasing (AER), which iteratively mines discriminative regions, erases them, and retrains a classifier to reveal complementary cues that become pseudo-labels for segmentation. However, standard AER also collects irrelevant background features, degrading pseudo-label quality. CREST addresses this with (i) a Constrained Ordinal Region Expansion (CORE) module that encodes the spatial-connectedness prior of polar lows, constraining region expansion from a high-confidence seed, and (ii) a Dynamic Bootstrapping (DB) loss that treats the mining order as a proxy for label reliability, attenuating supervision from noisier, later-mined regions. On Sentinel-1 SAR data, CREST follows the cyclone structure more closely than standard AER, and returns a multi-class rather than binary mask whose classes indicate the reliability assigned to each region. We further evaluate on BUS-UCLM breast ultrasound and PASCAL VOC person data, whose targets satisfy the same connectedness prior but come with the dense masks the SAR data lacks. On both datasets, CREST performs better than the equivalent AER pipeline under identical settings.

Chinese Translation

极地低压是快速形成于高纬度地区的强烈海洋气旋。深度学习可以在合成孔径雷达（SAR）图像中检测到它们，但像素级分割仍然是一个开放的挑战。由于缺乏用于训练的像素级掩膜，极地低压的范围本质上是主观的，其模糊的边界甚至使得专家的划分也不一致。我们提出了带有软目标的约束区域擦除（CREST）框架，这是一种弱监督语义分割（WSSS）方法，仅通过图像级标签生成掩膜。我们的方法基于对抗性擦除（AER），该方法迭代挖掘判别区域，擦除这些区域，并重新训练分类器以揭示补充线索，这些线索成为分割的伪标签。然而，标准的 AER 还会收集无关的背景特征，从而降低伪标签的质量。CREST 通过 (i) 约束序数区域扩展（CORE）模块来解决这个问题，该模块编码了极地低压的空间连通性先验，从高置信度种子约束区域扩展，以及 (ii) 动态自助抽样（DB）损失，将挖掘顺序视为标签可靠性的代理，减弱来自噪声较大、后期挖掘区域的监督。在哨兵-1 SAR 数据上，CREST 更加紧密地遵循气旋结构，并返回一个多类而非二元掩膜，其类别指示分配给每个区域的可靠性。我们还在 BUS-UCLM 乳腺超声和 PASCAL VOC 人物数据集上进行了进一步评估，这些目标满足相同的连通性先验，但具有 SAR 数据所缺乏的密集掩膜。在这两个数据集上，CREST 在相同设置下的表现优于等效的 AER 流水线。

View on arXiv Download PDF AI Translation

cs.CV / 58 / 2608.14389

GBU-Palm: A Multimodal Video Dataset and Benchmark for Palm Presentation Attack Detection

GBU-Palm：一种用于掌纹呈现攻击检测的多模态视频数据集和基准

Ma, Yingjie, Yu, Zitong, Jia, Wei, Kumar, Ajay, Shen, Linlin

Abstract

Existing palm presentation attack detection (PAD) datasets are often limited by static imagery, restricted acquisition conditions, or insufficient multimodal video data, hindering systematic evaluation across environments, modalities, and attack types. We present GBU-Palm, a large-scale multimodal video dataset and benchmark containing 21,326 videos from 105 subjects and 210 palms across six acquisition environments, including bona fide, Print, and Replay presentations, with 6,310 synchronized RGB-NIR samples. We construct leakage-controlled protocols that separate palm identity and attack lineage and benchmark four representative video architectures under environment-matched and held-out-environment settings. Results reveal substantial architecture-dependent degradation under environmental shift and show that RGB-NIR fusion does not consistently outperform RGB-only input. We further analyze model behavior through true accept (TA), true reject (TR), false accept (FA), and false reject (FR) decomposition, spectral masking, temporal-order intervention, and frozen-backbone NIR probing, revealing distinct failure patterns and evidence utilization across architectures. GBU-Palm provides a unified and challenging benchmark for developing and evaluating robust multimodal palm PAD methods under cross-environment conditions.

Chinese Translation

现有的掌纹呈现攻击检测（PAD）数据集往往受到静态图像、受限的采集条件或不足的多模态视频数据的限制，阻碍了在不同环境、模态和攻击类型下的系统评估。我们提出了GBU-Palm，一个大规模的多模态视频数据集和基准，包含来自105个受试者和210个掌纹的21,326个视频，涵盖六种采集环境，包括真实（bona fide）、打印（Print）和重放（Replay）呈现，配有6,310个同步的RGB-NIR样本。我们构建了泄漏控制协议，以分离掌纹身份和攻击谱系，并在环境匹配和保留环境设置下对四种代表性的视频架构进行基准测试。结果显示，在环境变化下，架构依赖性降级显著，并且RGB-NIR融合并不总是优于仅使用RGB输入。我们进一步通过真实接受（TA）、真实拒绝（TR）、错误接受（FA）和错误拒绝（FR）分解、光谱掩蔽、时间顺序干预和冻结主干NIR探测分析模型行为，揭示了不同架构的独特失败模式和证据利用。GBU-Palm为在跨环境条件下开发和评估鲁棒的多模态掌纹PAD方法提供了一个统一且具有挑战性的基准。

View on arXiv Download PDF AI Translation

cs.CV / 59 / 2608.14391

Can We Defend Against AI-Generated Video Attacks on Real-World Crisis Events? A Systematic Evaluation of Detectors, Generators and Social Dissemination

我们能否防御针对现实危机事件的AI生成视频攻击？对检测器、生成器和社会传播的系统评估

Liang, Shuo, Ma, Yixing, Zhou, Pengfei, Chen, Xingyan, Mei, Zihan, Li, Manting, Chen, Feihan, Wang, Zhiwen, Xu, Bin, Zhang, Haotian, Song, Jiajun, Su, Shiya, Liu, Run, Ni, Zhenghang, Yu, Yifa, Hong, Jintao, Feng, Bolong, Liu, Yifei, Zhang, Zirui, Zhang, Jingxuan, Zhao, Songlin, Bai, Yifan, Tan, Kang, Liu, Yizhe, Du, Junhao, Ge, Yongtao, Xv, Zhaopan, Zhang, Xinyuan, Ma, Mengru, Shen, Chunhua, Wang, Wei, You, Yang, Zhu, Zheng, Zhang, Kaipeng, Zhao, Wangbo

Abstract

Recent video generators can fabricate realistic depictions of wars, disasters, public emergencies, and other real-world crises, creating substantial risks of misinformation. Existing benchmarks, however, provide limited evidence on detector and generator behavior in such settings, including how detectability varies with generation conditions, how people perceive generated videos, and whether detectors remain reliable during social dissemination. To address this gap, we introduce RA-Bench, a benchmark for AI-generated video detection that uses Real videos as Anchors. RA-Bench contains 17,886 videos, comprising 1,830 real-video anchors across 10 social-risk categories and 16,056 generated clips from four open-source and five closed-source generators. Based on RA-Bench, we organize our evaluation along three dimensions. We first assess detector generalization across seven traditional detectors, ten zero-shot multimodal models under three review settings, and two MLLMs specifically fine-tuned on AI-generated video detection. Across these methods, none of the three detector families generalizes consistently across RA-Bench instances. We then examine how detectability varies with generation quality, conditioning information, and sampling seeds. These analyses show that generation properties affect detector families differently, while source-level detection patterns remain stable across seeds. Finally, we study human authenticity judgments and detector reliability during social dissemination. We find that videos that mislead people are also difficult for current detectors, and that social dissemination makes detection harder. Together, these findings show that current methods struggle to detect realistic AI-generated videos, highlighting the need for detectors robust to evolving video generators.

Chinese Translation

近期的视频生成器能够制造出战争、灾难、公共紧急事件及其他现实危机的逼真描绘，造成了重大虚假信息风险。然而，现有基准对检测器和生成器在此类环境中的表现提供的证据有限，包括可检测性如何随生成条件变化、公众如何感知生成的视频，以及在社会传播过程中检测器的可靠性是否保持。为了解决这一问题，我们引入了RA-Bench，这是一个用于AI生成视频检测的基准，使用真实视频作为锚点。RA-Bench包含17,886个视频，其中包括来自10个社会风险类别的1,830个真实视频锚点和来自四个开源及五个闭源生成器的16,056个生成片段。基于RA-Bench，我们沿三个维度组织我们的评估。首先，我们评估七个传统检测器、十个在三种审查设置下的零样本多模态模型，以及两个专门针对AI生成视频检测进行微调的MLLM的检测器泛化能力。在这些方法中，三类检测器在RA-Bench实例中并未表现出一致的泛化能力。然后，我们研究可检测性如何随生成质量、条件信息和采样种子变化。这些分析表明，生成属性对检测器家族的影响各不相同，而源级检测模式在不同种子间保持稳定。最后，我们研究在社会传播过程中人类的真实性判断和检测器的可靠性。我们发现，误导人们的视频同样难以被当前的检测器识别，而社会传播使检测变得更加困难。综合来看，这些发现表明当前方法在检测逼真的AI生成视频方面存在困难，突显了对能够适应不断演变的视频生成器的检测器的需求。

View on arXiv Download PDF AI Translation

cs.CV / 60 / 2608.14394

IRGNN: Efficient Invariant Radar Graph Neural Network for Radar Point Cloud Object Detection

IRGNN：高效的不变雷达图神经网络用于雷达点云目标检测

Guo, Xiao, Xia, Wanke, Yang, Lili, Wu, Caicong

Abstract

Perception is a fundamental component of autonomous driving systems. While LiDAR-based methods have achieved remarkable progress in object detection, their reliability can degrade under adverse weather conditions. Radar point clouds provide a robust alternative due to their resilience to bad weather and low-illumination scenarios. However, radar point clouds are typically sparse, unordered, and less informative than LiDAR data, making it challenging to directly apply existing LiDAR-based perception methods. To address these challenges, we propose IRGNN, an Invariant Radar Graph Neural Network for radar point cloud object detection. IRGNN first reconstructs radar point clouds into graph representations using translation- and rotation-invariant feature designs, enabling robust modeling of sparse radar measurements. It then employs an improved message passing neural network (MPNN) with residual connections and a virtual node layer to enhance local feature propagation and global context modeling. Finally, task-specific heads are applied to the learned graph representations for object classification and bounding box prediction. Experimental results on the RadarScenes dataset show that IRGNN outperforms existing radar-based object detection methods and achieves competitive performance. In addition, IRGNN significantly reduces computational cost and memory usage during inference, demonstrating its effectiveness and practical potential for efficient radar-based perception in autonomous driving.

Chinese Translation

感知是自动驾驶系统的一个基本组成部分。虽然基于激光雷达（LiDAR）的方法在目标检测方面取得了显著进展，但在恶劣天气条件下，它们的可靠性可能会下降。雷达点云由于其对恶劣天气和低光照场景的抗干扰能力，提供了一种稳健的替代方案。然而，雷达点云通常是稀疏的、无序的，并且信息量较激光雷达数据少，这使得直接应用现有的基于激光雷达的感知方法变得具有挑战性。为了解决这些问题，我们提出了IRGNN，一种用于雷达点云目标检测的不变雷达图神经网络。IRGNN首先通过平移和旋转不变特征设计将雷达点云重构为图表示，从而实现对稀疏雷达测量的稳健建模。然后，它采用改进的消息传递神经网络（MPNN），结合残差连接和虚拟节点层，以增强局部特征传播和全局上下文建模。最后，将特定任务的头部应用于学习到的图表示，以进行目标分类和边界框预测。在RadarScenes数据集上的实验结果表明，IRGNN的性能优于现有的基于雷达的目标检测方法，并且达到了竞争性的性能。此外，IRGNN在推理过程中显著降低了计算成本和内存使用，证明了其在自动驾驶中高效雷达感知的有效性和实际潜力。

View on arXiv Download PDF AI Translation

cs.CV / 61 / 2608.14403

CRAFT: Constrained Reward via Attention Fine-Tuning for Subject Personalization without Composed Targets

CRAFT：通过注意力微调实现的受限奖励，用于无合成目标的主体个性化

Park, Jihun, Lee, Kyoungmin, Gim, Jongmin, Jo, Hyeonseo, Kim, Jaeyeul, Zou, Han, Zhan, Zhenpeng, Zhang, Yan, Im, Sunghoon

Abstract

Subject-driven image personalization---generating new images that preserve the identity of one or several reference subjects in novel scenes---is a foundational capability for modern visual content creation. It is currently dominated by generalized methods that fine-tune a pretrained multimodal diffusion transformer (MMDiT) on hundreds of thousands to millions of paired \emph{(reference, composed-target)} examples, where each composed target is a synthesized image of the subject in a novel scene. Producing such targets demands a costly multi-stage curation pipeline---LLM-based prompt generation, T2I-based composed-target synthesis, reference-subject extraction, VLM-based quality filtering, and correspondence labeling---and tightly couples each method to a particular target synthesizer and curation choice. We introduce \emph{CRAFT} (Constrained Reward via Attention Fine-Tuning), a single-step ReFL framework that fine-tunes a pre-trained \emph{reference-aware} MMDiT via LoRA adapters using a compact reference-only data construction---$10$K reference images and subject masks, with no composed-target supervision. CRAFT realizes a \emph{Where to look} principle: attention-level rewards align noise- and phrase-token attention with the correct reference subject, and the resulting per-subject attention masks gate a pixel-level identity reward to keep image-space supervision consistent with the learned attention routing. Applied to FLUX.2-klein-9B, CRAFT achieves state-of-the-art performance on XVerseBench \rev{while using no composed-target supervision---only $10$K reference-only samples, whereas prior generalized methods require $150$K to over $2$M composed-target pairs}. The same recipe transfers to other reference-aware backbones, consistently improving performance. Project page: https://jihun999.github.io/projects/CRAFT/.

Chinese Translation

以主体为驱动的图像个性化——生成在新场景中保留一个或多个参考主体身份的新图像——是现代视觉内容创作的基础能力。目前，这一领域主要由通用方法主导，这些方法在数十万到数百万对 extit{(参考, 合成目标)}示例上微调预训练的多模态扩散变换器（MMDiT），其中每个合成目标都是在新场景中合成的主体图像。生成此类目标需要一个昂贵的多阶段策划流程——基于大型语言模型（LLM）的提示生成、基于文本到图像（T2I）的合成目标合成、参考主体提取、基于视觉语言模型（VLM）的质量过滤和对应标注——并将每种方法紧密耦合到特定的目标合成器和策划选择。我们提出了 extit{CRAFT}（通过注意力微调实现的受限奖励），这是一个单步的ReFL框架，通过LoRA适配器微调预训练的 extit{参考感知} MMDiT，使用紧凑的仅参考数据构建——$10$K参考图像和主体掩膜，无需合成目标监督。CRAFT实现了 extit{Where to look}原则：注意力级别的奖励将噪声和短语标记的注意力与正确的参考主体对齐，生成的每个主体注意力掩膜为像素级身份奖励提供了门控，以保持图像空间监督与学习到的注意力路由一致。应用于FLUX.2-klein-9B，CRAFT在XVerseBench上实现了最先进的性能——而不使用合成目标监督，仅使用$10$K仅参考样本，而之前的通用方法需要$150$K到超过$2$M的合成目标对。相同的方法可以转移到其他参考感知的骨干网络，持续提高性能。项目页面：https://jihun999.github.io/projects/CRAFT/

View on arXiv Download PDF AI Translation

cs.CV / 62 / 2608.14428

GhostPoint: Self-Supervised Representation Learning by Hallucinating Occluded LiDAR Structure

GhostPoint：通过幻觉化遮挡的LiDAR结构进行自监督表示学习

Abdelsamad, Mohamed, Yang, Bin, Ulrich, Michael, Zhang, Miao, Miron, Yakov, Condurache, Alexandru Paul, Valada, Abhinav

Abstract

3D object detection from LiDAR point clouds is a core problem in autonomous driving. Recent advances in self-supervised learning (SSL) enable scalable pretraining and transfers well to per-point tasks such as semantic and panoptic segmentation, but transfer to 3D detection remains weaker. We analyze recent SSL methods and find that most objectives are defined only on measured LiDAR returns from visible surfaces, leaving occluded and unobserved regions unconstrained. This visible-surface bias can be sufficient for point-wise prediction, but 3D detection requires robustness to missing structure. To address this gap, we propose GhostPoint, an SSL framework that hallucinates latent features in local neighborhoods around discovered instances, generated via a novel instance voxel dilation. In GhostPoint, an encoder processes observed returns, and an additional predictor infers neighborhood representations from observed context. In addition to standard encoder-level supervision, we introduce a predictor-level supervision scheme on sampled voxels from generated neighborhoods. Specifically, observed (visible/masked) voxels match teacher-encoder targets, while unobserved voxels match teacher-predictor hallucinations. This design encourages the learned representation to explicitly model structure beyond observed returns. Extensive evaluations on nuScenes and Waymo demonstrate that our method achieves state-of-the-art performance, consistently improving downstream 3D detection, especially under sparse scans and limited labels.

Chinese Translation

从LiDAR点云中进行3D物体检测是自动驾驶中的核心问题。最近自监督学习（SSL）的进展使得可扩展的预训练成为可能，并且在语义和全景分割等逐点任务中表现良好，但在3D检测中的迁移效果仍然较弱。我们分析了近期的SSL方法，发现大多数目标仅在来自可见表面的测量LiDAR返回上定义，导致遮挡和未观察区域没有约束。这种可见表面偏差对于逐点预测可能是足够的，但3D检测需要对缺失结构的鲁棒性。为了解决这一问题，我们提出了GhostPoint，一个SSL框架，通过一种新颖的实例体素扩张技术，在发现的实例周围的局部邻域中幻觉化潜在特征。在GhostPoint中，编码器处理观察到的返回，而额外的预测器则从观察到的上下文中推断邻域表示。除了标准的编码器级监督外，我们还在生成邻域中采样的体素上引入了预测器级监督方案。具体而言，观察到的（可见/遮挡）体素与教师编码器目标匹配，而未观察到的体素与教师预测器的幻觉匹配。这种设计鼓励学习到的表示明确建模超出观察返回的结构。在nuScenes和Waymo上的广泛评估表明，我们的方法实现了最先进的性能，持续改善下游3D检测，尤其是在稀疏扫描和有限标签的情况下。

View on arXiv Download PDF AI Translation

cs.CV / 63 / 2608.14435

Style or Signature? Artist-Disjoint Evaluation of Style Classification in Frozen Vision Embeddings

风格还是签名？艺术家分离评估在冻结视觉嵌入中的风格分类

Ashton, Rory

Abstract

Frozen image embeddings from models such as CLIP are increasingly used to classify paintings by art-historical style, with high reported accuracy. We ask whether this accuracy reflects an understanding of style or the recognition of individual artists. Standard evaluation uses random splits in which works by the same artist appear on both sides, so a classifier can succeed by recognising the painter rather than the movement. We re-evaluate style classification under an artist-disjoint protocol, holding out every artist in turn so that no work is ever classified using other works by its own painter. On a balanced dataset of 320 paintings across four twentieth-century movements, 5-NN style accuracy falls from 0.87 to 0.77 under this protocol, and the drop is sharply uneven. Impressionism and Cubism barely move, while Surrealism falls twenty points. The pattern holds across four image encoders, including a vision-only self-supervised model, which places the effect in visual structure rather than language. Where an encoder captures genuine shared form, individual artists are barely recognisable yet style is robust, while Surrealism shows the opposite. We argue that artist-disjoint evaluation is necessary to measure stylistic understanding in frozen embeddings.

Chinese Translation

来自如 CLIP 等模型的冻结图像嵌入正被越来越多地用于根据艺术历史风格对绘画进行分类，并报告了较高的准确性。我们探讨这种准确性是否反映了对风格的理解，还是对个别艺术家的识别。标准评估使用随机拆分，其中同一艺术家的作品出现在两侧，因此分类器可以通过识别画家而不是艺术流派来取得成功。我们在艺术家分离协议下重新评估风格分类，逐一排除每位艺术家，以确保任何作品都不会使用其自身画家的其他作品进行分类。在一个包含四个二十世纪艺术流派的320幅绘画的平衡数据集中，5-NN 风格准确率在该协议下从0.87降至0.77，且下降幅度明显不均匀。印象派和立体主义几乎没有变化，而超现实主义则下降了二十个百分点。这个模式在四种图像编码器中都存在，包括一个仅限视觉的自监督模型，这将效果归因于视觉结构而非语言。当编码器捕捉到真正共享的形式时，个别艺术家几乎不可识别，但风格却保持稳健，而超现实主义则显示出相反的情况。我们认为，艺术家分离评估对于测量冻结嵌入中的风格理解是必要的。

View on arXiv Download PDF AI Translation

cs.CV / 64 / 2608.14530

Marionette: Predicting World States, Rendering Geometry, Painting Appearance

木偶：预测世界状态、渲染几何、绘制外观

Meng, Zian, Li, Zhen, Li, Chuanhao, Li, Qiang, Zhang, Kaipeng

Abstract

Interactive game world models typically autoregress visual observations directly in pixel or latent space, forcing structured properties such as pose, geometry, and occlusion to be implicitly maintained by the same generative sequence. Over long horizons, errors in these latent world properties accumulate, making consistency and controllability fragile. We explicitly model the evolving world state, delegate exact geometric computation to a fixed, zero-parameter renderer, and leave the neural model to synthesize appearance. We instantiate this idea as Marionette, a world model for interactive games with articulated characters. First, a two-stage autoregressive dynamics model predicts an explicit and interpretable 276-dimensional 3D world state comprising multi-entity articulated skeletons, metric root trajectories, and rotations. Second, a zero-parameter graphics bridge converts the predicted state into pose-control videos, computing world-space geometry and occlusion in closed form. Third, a control-conditioned video-diffusion observation model synthesizes photorealistic RGB observations from the resulting structured controls. Our experiments establish two properties of Marionette. First, the predicted world state is directly controllable. Forcing a mismatched action stream changes root-aligned joint error by 31% across 48 held-out segments. Second, long-horizon behaviour is determined in the state, and can be repaired there. Left free, the two generated characters drift to 21.2 m apart (recorded sessions stay near 5 m) and a third of frames show ground penetration. Two rules imposed on the explicit state, a terrain collider and a separation cap, cut penetration by 66% and keep the pair engaged, with no change to the observation model. Routing appearance through the predicted state costs no fidelity we can detect, at an FVD of 831 against 799 for recorded pose.

Chinese Translation

交互式游戏世界模型通常直接在像素或潜在空间中自回归视觉观察，迫使姿态、几何和遮挡等结构化属性通过相同的生成序列隐式维持。在长时间范围内，这些潜在世界属性的错误会累积，使得一致性和可控性变得脆弱。我们明确建模不断变化的世界状态，将精确的几何计算委托给一个固定的零参数渲染器，并让神经模型合成外观。我们将这一理念具体化为木偶（Marionette），这是一个用于具有关节角色的交互式游戏的世界模型。首先，一个两阶段自回归动态模型预测一个明确且可解释的276维3D世界状态，包括多实体关节骨架、度量根轨迹和旋转。其次，一个零参数图形桥将预测的状态转换为姿态控制视频，以封闭形式计算世界空间几何和遮挡。第三，一个控制条件的视频扩散观察模型从生成的结构化控制中合成逼真的RGB观察。我们的实验确立了木偶的两个特性。首先，预测的世界状态是直接可控的。强迫不匹配的动作流使得根对齐的关节误差在48个保留段中变化了31%。其次，长时间范围内的行为在状态中被决定，并可以在其中修复。自由状态下，两个生成的角色漂移至21.2米远（记录的会话保持在5米附近），三分之一的帧显示地面穿透。对显式状态施加的两个规则，一个地形碰撞器和一个分离上限，将穿透减少了66%，并保持这对角色的互动，观察模型没有变化。通过预测状态路由外观的成本在我们可以检测到的保真度上没有损失，FVD为831，而记录姿态的FVD为799。

View on arXiv Download PDF AI Translation

cs.CV / 65 / 2608.14539

Decoding the Past: An Uncertainty-Aware Deep Learning Framework for Sex Attribution in Prehistoric Hand Stencils

解码过去：一种基于不确定性的深度学习框架用于史前手印的性别归属

Becerra, Karel, Mederos, Boris, Snow, Dean, Mollineda, Ramón A.

Abstract

Determining the biological sex of the individuals who created Upper Paleolithic hand stencils remains a challenging problem due to the absence of ground truth, population differences between contemporary and prehistoric groups, and the uncertainty introduced by image degradation. Traditional morphometric methods suffer from high structural overlap across sexes, poor cross-population generalizability, and subjective feature engineering. This study presents an uncertainty-aware deep learning framework for sex attribution in prehistoric hand stencils that explicitly models, propagates, and aggregates uncertainty throughout the analytical pipeline. The methodology combines dual image processing, dual contour extraction, structured silhouette augmentation, model architectural diversity, and ensemble-based decision aggregation. The pipeline generates twelve plausible silhouette realizations per stencil to capture boundary uncertainties, which are processed by two ensembles of ten deep neural networks each (EfficientNet-B3 and MobileViT-S) trained on 14,036 contemporary hand samples. Furthermore, a triangulated validation scheme integrates ensemble predictions with unsupervised 2D latent-space manifold mapping (UMAP + k-NN) and explainable AI spatial attributions (LayerCAM) to ensure anatomical consistency. On contemporary data, ensemble models achieve strong classification performance, with accuracies exceeding 88% in older age groups. When applied to prehistoric stencils, the framework produces both sex predictions and confidence measures of internal agreement, enabling the distinction between morphologically stable and ambiguous cases. Convergence across ensemble predictions, latent-space structure, and interpretability analyses shows that uncertainty can become a measurable component of archaeological inference, enabling robust and reproducible decoding of ancient rock art.

Chinese Translation

由于缺乏真实数据、当代与史前群体之间的人口差异以及图像退化带来的不确定性，确定创造上旧石器时代手印个体的生物性别仍然是一个具有挑战性的问题。传统的形态测量方法受到性别间结构重叠高、跨人群泛化能力差以及主观特征工程的限制。本研究提出了一种基于不确定性的深度学习框架，用于史前手印的性别归属，该框架在整个分析流程中明确建模、传播和聚合不确定性。该方法结合了双重图像处理、双重轮廓提取、结构化轮廓增强、模型架构多样性和基于集成的决策聚合。该流程为每个手印生成十二个合理的轮廓实现，以捕捉边界不确定性，这些轮廓由两个各包含十个深度神经网络的集成处理（EfficientNet-B3 和 MobileViT-S），这些网络在14,036个当代手样本上进行训练。此外，一个三角验证方案将集成预测与无监督的二维潜在空间流形映射（UMAP + k-NN）和可解释的人工智能空间归属（LayerCAM）相结合，以确保解剖学的一致性。在当代数据上，集成模型实现了强大的分类性能，老年组的准确率超过88%。当应用于史前手印时，该框架不仅生成性别预测，还提供内部一致性的置信度度量，从而能够区分形态学稳定和模糊的案例。集成预测、潜在空间结构和可解释性分析之间的收敛表明，不确定性可以成为考古推理的可测量组成部分，从而实现对古代岩画的稳健和可重复解码。

View on arXiv Download PDF AI Translation

cs.CV / 66 / 2608.14543

MagnifiQ: Patch-aware Text Guided Progressive Upscaling for High-Resolution Image Restoration

MagnifiQ：基于补丁感知的文本引导渐进式高分辨率图像恢复

Reddy, Mahesh, Savani, Yashesh, Mercier, Antoine, Cai, Hong, Porikli, Fatih, Berger, Guillaume

Abstract

High-resolution image restoration from degraded inputs is challenging because it must preserve global structural consistency while recovering fine-grained local details, especially at 4K resolution where direct diffusion-based restoration is computationally expensive and prone to repeated or inconsistent textures. In this work, we introduce MagnifiQ, an image restoration framework that progressively upscales and restores images across resolutions, e.g., from 1024x1024 to 4096x4096. Our approach leverages a pre-trained text-to-image diffusion model such as SDXL and adapts it for more scalable high-resolution inference by replacing its original self-attention layers with convolutional operations whose computational cost grows linearly with image resolution. We further propose a progressive upscaling strategy that iteratively restores images over multiple resolution stages, refining each intermediate output rather than directly hallucinating the final 4K image, thereby improving global coherence and reducing high-resolution artifacts. To enhance local details while controlling content drift, MagnifiQ uses patch-specific text prompts that provide spatially localized semantic guidance during restoration. Extensive experiments on synthetic and real-world degraded images show that MagnifiQ outperforms prior diffusion-based restoration methods in perceptual quality and human preference, producing sharper textures and more coherent 4K results while offering practical speed--quality trade-offs through its scalable backbone and progressive design.

Chinese Translation

从退化输入中恢复高分辨率图像是一项具有挑战性的任务，因为它必须在恢复细粒度局部细节的同时保持全局结构的一致性，尤其是在4K分辨率下，直接基于扩散的恢复计算成本高且容易出现重复或不一致的纹理。在本研究中，我们提出了MagnifiQ，一个图像恢复框架，它逐步放大并恢复不同分辨率的图像，例如，从1024x1024到4096x4096。我们的方法利用了预训练的文本到图像扩散模型（如SDXL），并通过用卷积操作替换其原始自注意力层来适应更可扩展的高分辨率推理，这样计算成本随着图像分辨率线性增长。我们进一步提出了一种渐进式放大策略，该策略在多个分辨率阶段迭代恢复图像，精细化每个中间输出，而不是直接幻觉最终的4K图像，从而提高全局一致性并减少高分辨率伪影。为了增强局部细节并控制内容漂移，MagnifiQ使用特定于补丁的文本提示，在恢复过程中提供空间局部的语义指导。在对合成和真实世界退化图像进行的广泛实验中，MagnifiQ在感知质量和人类偏好方面优于先前的基于扩散的恢复方法，产生了更清晰的纹理和更连贯的4K结果，同时通过其可扩展的骨干网络和渐进式设计提供了实用的速度与质量权衡。

View on arXiv Download PDF AI Translation

cs.CV / 67 / 2608.14546

CPI-Bench: A Comprehensive,Practical and Intelligent Benchmark for Real-World Image Editing

CPI-Bench：一个全面、实用且智能的真实世界图像编辑基准

Zhou, Qinye, Zheng, Jun, Du, Yongchao, Wang, Yuan, Chen, Zhengrui, Gao, Zuan, Hu, Taihang, Lin, Chao, Shen, Yefeng, Wang, Xingjian, Wang, Zhao, Wu, Zhengtao, Xu, Xiaoli, Xu, Zhengze, Yan, Hao, Yang, Denghui, Yu, Yuhang, Zhang, Huayu, Zhang, Mingzhou, Chen, Mengting

Abstract

With the rapid advancement of image editing models and their widespread application across various domains, there is an increasingly urgent need to deploy these model capabilities directly into real-world scenarios. However, existing benchmarks remain confined to simple single-image tasks, suffering from limited coverage dimensions and an inability to effectively differentiate performance among diverse models. Consequently, they fail to reliably evaluate model performance in complex multi-image editing, highly demanding reasoning instructions, and practical deployment settings. To address these limitations, we propose CPI-Bench, a Comprehensive, Practical andIntelligent benchmark for real-world image editing. CPI-Bench comprises three core subsets: CPI-General-Bench, which comprehensively covers diverse editing tasks and pioneers the inclusion of multi-image editing evaluation; CPI-Practical-Bench, which focuses on high-frequency real-user application scenarios; and CPI-Intelligent-Bench, which is dedicated to evaluating capabilities in highly demanding reasoning-based editing. Evaluation results of mainstream image editing models based on CPI-Bench demonstrate that CPI-Bench enhances performance differentiation among models. It provides a comprehensive and reliable quantification of gaps in general editing capabilities, practical deployment efficacy, and advanced reasoning-based editing, offering invaluable guidance for the future optimization of image editing models. Crucially, our ranking analysis reveals that CPI-Bench achieves the highest alignment with the Arena Image Edit Leaderboard, indicating it faithfully captures the preferences and perceptual judgments of human evaluators, serving as a robust proxy for real-world user experience.

Chinese Translation

随着图像编辑模型的快速发展及其在各个领域的广泛应用，直接将这些模型能力部署到真实场景中的需求日益迫切。然而，现有基准仍局限于简单的单图像任务，覆盖维度有限，且无法有效区分不同模型的性能。因此，它们无法可靠地评估模型在复杂的多图像编辑、高要求推理指令和实际部署环境中的性能。为了解决这些局限性，我们提出了CPI-Bench，一个全面、实用且智能的真实世界图像编辑基准。CPI-Bench包含三个核心子集：CPI-General-Bench，全面覆盖多样的编辑任务，并首创多图像编辑评估的纳入；CPI-Practical-Bench，专注于高频真实用户应用场景；以及CPI-Intelligent-Bench，致力于评估在高要求推理基础编辑中的能力。基于CPI-Bench的主流图像编辑模型评估结果表明，CPI-Bench增强了模型之间的性能区分。它提供了对一般编辑能力、实际部署有效性和高级推理基础编辑差距的全面可靠量化，为未来图像编辑模型的优化提供了宝贵指导。重要的是，我们的排名分析显示，CPI-Bench与Arena Image Edit Leaderboard的对齐度最高，表明它真实地捕捉了人类评估者的偏好和感知判断，成为真实用户体验的有力代理。

View on arXiv Download PDF AI Translation

人工智能 (Artificial Intelligence)

cs.AI / 1 / 2608.13564

Inducing Reward-Free Judging Rubrics that Reduce Over-Crediting in Agent Evaluation

诱导无奖励评估标准以减少代理评估中的过度评分

Quinn, Darragh, Dylan, David, Healy, Roisin, Carroll, Fionn, Donnelly, Maeve, Sheehan, Cormac

Abstract

Evaluating language-model agents at scale increasingly relies on a second language model as an automatic judge, because the gold signal, an executable environment reward, is expensive, slow, or unavailable at deployment time. Such a judge is a reward-free proxy whose value depends on whether it can be trusted, yet existing judges either hand-write the scoring rubric, as in G-Eval, or fine-tune the judge's weights, and both tend to credit fluent but unsuccessful trajectories as successes. We instead induce the text of an agent-judging rubric from a small set of ground-truth-labeled trajectories, grounding it in true outcomes. We present RubricForge, which evolves a judge rubric by reflective evolution against labeled trajectories to maximize agreement with the environment reward, freezes it, and applies it to held-out trajectories in one model call with no environment access. The optimized artifact is human-readable text, so every verdict is attributable to named criteria. Using one frozen 7B model as both agent and judge, on tau-bench (173 labeled trajectories drawn from 220 rollouts) and WebShop (160), the principal gain is faithfulness rather than raw agreement. The edge over a generic G-Eval judge is not statistically significant (McNemar p = 0.248), and absolute-score calibration marginally favors the generic judge (|err| difference -0.048, p = 2x10^-4). Yet RubricForge over-credits failed trajectories roughly half as often (0.115 vs. 0.173 false-pass rate on tau-bench, with three over-credit catches and zero reversals) and ranks graded WebShop outcomes more faithfully (Spearman 0.410 vs. 0.370). For a reward-free evaluator the false-pass rate, not aggregate agreement, is the deployment-relevant quantity, since a false pass ships a broken agent whereas a false fail merely costs a retry.

Chinese Translation

在大规模评估语言模型代理时，越来越依赖于第二个语言模型作为自动评审者，因为金标准信号（可执行环境奖励）在部署时成本高、速度慢或不可用。这种评审者是一个无奖励的代理，其价值取决于其可信度，而现有的评审者要么手动编写评分标准（如 G-Eval），要么微调评审者的权重，二者都倾向于将流畅但不成功的轨迹视为成功。我们则从一小组真实标注的轨迹中诱导出代理评审标准的文本，以真实结果为基础。我们提出了 RubricForge，它通过反思进化在标注轨迹上演化评审标准，以最大化与环境奖励的一致性，随后将其固定，并在一次模型调用中应用于保留轨迹，无需访问环境。优化后的产物是人类可读的文本，因此每个裁决都可以归因于特定标准。在 tau-bench（从 220 次回合中抽取的 173 条标注轨迹）和 WebShop（160 条）上，使用一个冻结的 7B 模型作为代理和评审者，主要的收益是忠实性而非原始一致性。与通用 G-Eval 评审者相比，优势在统计上并不显著（McNemar p = 0.248），绝对分数校准略微偏向通用评审者（|err| 差异 -0.048, p = 2x10^-4）。然而，RubricForge 对失败轨迹的过度评分大约只有一半的频率（在 tau-bench 上为 0.115 对比 0.173 的误判通过率，且有三次过度评分捕获和零次反转），并且对 WebShop 结果的排名更为忠实（Spearman 0.410 对比 0.370）。对于无奖励评估者而言，误判通过率而非总体一致性是与部署相关的量，因为误判通过会导致一个损坏的代理被交付，而误判失败仅仅会导致重试。

View on arXiv Download PDF AI Translation

cs.AI / 2 / 2608.13565

Depth-Aware Sensitivity Analysis of Mixture-of-Experts Models via Magnitude-Based Expert Masking

基于幅度的专家掩码的混合专家模型深度感知灵敏度分析

Sharma, Pradeep Kumar, Godbole, Shantanu, Shrivastava, Hritvik

Abstract

Mixture-of-Experts (MoE) architectures scale large language models (LLMs) while preserving computational efficiency through sparse activation. Despite their widespread adoption, the relative importance of individual MoE layers remains insufficiently characterized, particularly for model compression. This paper presents a systematic layer-wise sensitivity analysis of the Qwen3.6-35B-A3B model (40 MoE layers, 256 experts per layer, top-8 routing) using magnitude-based expert masking on the XLCoST cross-lingual code translation benchmark. We conduct a multi-phase study spanning 100, 300, and 500 prompt evaluation scales across three H100 GPU servers. Our central finding is that layer sensitivity is strongly depth-dependent: early layers (0-9) and middle layers (10-29) are highly fragile to expert masking, while late layers (30-39), and especially very-late layers (35-39), tolerate aggressive masking of low-magnitude experts. Flat all-layer masking at 30% retains only 150/300 Good+Similar outputs at 300-prompt scale, whereas late-focused policies retain 249-255/300 while masking 640-1,145 experts. On a later 500-prompt held-out validation slice, the narrow very-late policy (layers 35-39 @ 50%) achieves the strongest quality/masked-expert tradeoff among tested candidates, retaining 419/500 Good+Similar outputs while masking only 640 of 10,240 total experts. We additionally characterize top-k routing width reduction from 8 to 6 active experts per token, which shows a large observed wall-clock reduction on a 100-prompt probe with no Good+Similar loss, though it does not yet compose cleanly with aggressive expert masking. These findings provide an empirical foundation for depth-aware MoE expert masking and establish a practical path toward physical weight surgery, activation-based expert scoring, and training-based recovery.

Chinese Translation

混合专家（Mixture-of-Experts, MoE）架构在保持计算效率的同时扩展大型语言模型（Large Language Models, LLMs），通过稀疏激活实现。尽管其广泛应用，但个别 MoE 层的相对重要性仍未得到充分表征，尤其是在模型压缩方面。本文对 Qwen3.6-35B-A3B 模型（40 个 MoE 层，每层 256 个专家，前 8 路由）进行了系统的层级灵敏度分析，采用基于幅度的专家掩码，基于 XLCoST 跨语言代码翻译基准进行研究。我们在三台 H100 GPU 服务器上进行了跨 100、300 和 500 个提示评估规模的多阶段研究。我们的主要发现是，层灵敏度与深度密切相关：早期层（0-9）和中间层（10-29）对专家掩码高度脆弱，而后期层（30-39），尤其是非常后期层（35-39），能够容忍低幅度专家的激进掩码。在 300 个提示规模下，30% 的全层掩码仅保留了 150/300 个 Good+Similar 输出，而以后期为重点的策略在掩码 640-1,145 个专家的情况下保留了 249-255/300 个输出。在后来的 500 个提示保留验证切片中，狭窄的非常后期策略（层 35-39 @ 50%）在测试候选中实现了最佳的质量/掩码专家权衡，保留了 419/500 个 Good+Similar 输出，同时仅掩码了 640 个总共 10,240 个专家。我们还表征了从每个 token 的 8 个活跃专家减少到 6 个活跃专家的 top-k 路由宽度，这在 100 个提示探测中显示出显著的墙钟时间减少，且没有 Good+Similar 损失，尽管它尚未与激进的专家掩码良好组合。这些发现为深度感知的 MoE 专家掩码提供了实证基础，并为物理权重手术、基于激活的专家评分和基于训练的恢复建立了实用路径。

View on arXiv Download PDF AI Translation

cs.AI / 3 / 2608.13567

Modular Cognitive Architecture Emerges in Large Language Models

大型语言模型中的模块化认知架构的出现

Han, Pengrui, Andreas, Jacob, Fedorenko, Evelina, de Varda, Andrea Gregor

Abstract

The human brain exhibits a striking degree of functional specialization, with distinct networks supporting language, formal reasoning, reasoning about other minds, and reasoning about the physical world. Is this modular organization a fundamental principle of how intelligent systems must be built, or an evolutionary accident specific to biological brains? Here, we test whether a similar organization emerges in Large Language Models--another class of intelligent systems created through a very different optimization process. Using circuit analyses across N=46 tasks spanning four cognitive domains (language, formal reasoning, social reasoning, physical reasoning), we find that LLMs develop a modular architecture that mirrors the human brain: tasks drawing on the same network in humans recruit overlapping neurons in LLMs, whereas tasks drawing on different networks recruit distinct neurons. The convergent emergence of modularity in brains and neural networks suggests that it may be a fundamental property of intelligent systems.

Chinese Translation

人脑表现出显著的功能专业化程度，具有支持语言、形式推理、他人思维推理和物理世界推理的不同网络。这种模块化组织是智能系统构建的基本原则，还是生物大脑特有的进化偶然？在此，我们测试了大型语言模型（Large Language Models，LLMs）中是否出现类似的组织结构——这是一类通过非常不同的优化过程创建的智能系统。通过对跨越四个认知领域（语言、形式推理、社会推理、物理推理）的46个任务进行电路分析，我们发现LLMs发展出一种与人脑相似的模块化架构：在任务中，利用人类相同网络的任务在LLMs中招募重叠的神经元，而利用不同网络的任务则招募不同的神经元。大脑和神经网络中模块化的趋同出现表明，这可能是智能系统的一个基本属性。

View on arXiv Download PDF AI Translation

cs.AI / 4 / 2608.13573

A Year in LLM Serving: Workload Evolution, Caching and Load-Balancing

大型语言模型服务的一年：工作负载演变、缓存与负载均衡

Nixon, William, Durbin, Jon, Standhartinger, Florian, Gunawi, Haryadi S., Yang, Juncheng

Abstract

Large Language Model (LLM) serving has become a critical cloud workload, and realistic traces are essential for motivating and benchmarking serving systems. However, existing LLM serving workload studies remain limited in scale and scope. They often observe short time periods and provide limited visibility into how users interact with models in production. As a result, they do not fully capture how LLM serving workloads evolve over time or how user-model interactions shape production traffic. In this work, we further the understanding of real-world LLM serving workloads through both a global characterization and a longitudinal study of a one-year production trace from Chutes. Unlike prior studies, our trace captures full production behavior across many models and users, including both popular and long-tail models. We analyze the workload from aggregate, temporal, model-level, and user-level perspectives, revealing workload evolution and user-model structure that are typically hidden behind aggregate views. To support future research, we will release the full one-year trace with the paper, enabling downstream studies of production behavior without relying on sampled or synthetically generated workloads.

Chinese Translation

大型语言模型（LLM）服务已成为关键的云工作负载，真实的工作负载记录对于激励和基准测试服务系统至关重要。然而，现有的LLM服务工作负载研究在规模和范围上仍然有限。它们通常观察短时间段，并且对用户如何与生产中的模型互动提供了有限的视角。因此，它们未能充分捕捉LLM服务工作负载随时间演变的过程，或用户与模型的互动如何影响生产流量。在本研究中，我们通过对Chutes一年生产记录的全球特征化和纵向研究，进一步理解现实世界中的LLM服务工作负载。与之前的研究不同，我们的记录捕捉了多个模型和用户的完整生产行为，包括热门模型和长尾模型。我们从整体、时间、模型级别和用户级别的视角分析了工作负载，揭示了通常隐藏在整体视图背后的工作负载演变和用户-模型结构。为了支持未来的研究，我们将在论文中发布完整的一年记录，使下游研究能够在不依赖于采样或合成生成的工作负载的情况下研究生产行为。

View on arXiv Download PDF AI Translation

cs.AI / 5 / 2608.13574

Agentao: A Governed Local-First Runtime for Tool-Using LLM Agents

Agentao：一种用于工具使用的 LLM 代理的受控本地优先运行时

Jin, Bo, Jiao, Qiang, Tong, Xin

Abstract

LLM agents increasingly operate as execution systems that invoke tools, modify local state, use persistent memory, and interact with external protocols. These capabilities make agents useful, but they also introduce risks related to over-privileged actions, weak auditability, prompt injection, tool poisoning, and uncontrolled side effects. This paper presents Agentao, a governed local-first runtime for tool-using LLM agents. Agentao separates model-generated action proposals from host-authorized execution through a layered architecture consisting of host-facing surfaces, a host contract, a runtime core, a permission-mediated tool system, and supporting subsystems for memory, replay, plugins, skills, sub-agents, and protocol integration. We describe the motivation, threat model, design goals, governance model, execution pipeline, and structured event interface of the system. Agentao does not provide formal safety guarantees; rather, it demonstrates how permissions, state, protocol boundaries, and execution traces can be made explicit runtime abstractions for building agents that are more governable, inspectable, and suitable for host-controlled local environments. The code is publicly available at https://github.com/jin-bo/agentao.

Chinese Translation

LLM 代理越来越多地作为执行系统运行，调用工具、修改本地状态、使用持久内存并与外部协议交互。这些能力使代理变得有用，但也引入了与过度特权操作、审计能力弱、提示注入、工具中毒和不可控副作用相关的风险。本文介绍了 Agentao，一种用于工具使用的 LLM 代理的受控本地优先运行时。Agentao 通过一个分层架构将模型生成的动作提议与主机授权的执行分离，该架构包括面向主机的表面、主机合同、运行时核心、权限中介工具系统以及用于内存、重放、插件、技能、子代理和协议集成的支持子系统。我们描述了该系统的动机、威胁模型、设计目标、治理模型、执行管道和结构化事件接口。Agentao 并不提供正式的安全保证；相反，它展示了如何将权限、状态、协议边界和执行痕迹明确化为构建更易于治理、可检查和适合主机控制的本地环境的代理的运行时抽象。代码可在 https://github.com/jin-bo/agentao 上公开获取。

View on arXiv Download PDF AI Translation

cs.AI / 6 / 2608.13577

AI Evaluation Should Work With Humans

人工智能评估应与人类协作

Kulveit, Jan, Leech, Gavin, Gavenčiak, Tomáš, Douglas, Raymond

Abstract

This position paper argues that the dominant paradigm of AI evaluation (which focuses on superhuman autonomous performance and so implicitly targets the goal of replacing humans) is guiding AI development in the wrong direction. Instead, the AI community should pivot to evaluating the performance of human--AI teams. We argue that this collaborative shift will foster AI systems that act as true complements to human capabilities and therefore lead to far better societal outcomes than will the current process.

Chinese Translation

本文立场论文认为，当前主导的人工智能评估范式（侧重于超人类的自主性能，因此隐含地以取代人类为目标）正在引导人工智能的发展走向错误的方向。相反，人工智能社区应转向评估人类与人工智能团队的表现。我们认为，这一协作转变将促进人工智能系统作为人类能力的真正补充，从而带来比当前过程更好的社会结果。

View on arXiv Download PDF AI Translation

cs.AI / 7 / 2608.13591

Stable Miscalibration in Large Language Models: A Practical View of High-Confidence Errors

大型语言模型中的稳定误校准：高置信度错误的实用视角

Okutomi, Akira

Abstract

High-confidence errors in large language models are often treated as evidence of fragile internal inference. We study a different possibility: stable miscalibration, where a confident wrong answer remains locally stable under small perturbations. We combine two diagnostics: a label-aware output-level audit score that ranks domains by confidence variation and overconfident mistakes under a forced-answer baseline, and an internal sensitivity probe that measures hidden-state movement. On a multi-domain binary factual audit set, this audit score tracks where abstention-aware self-critique reduces decision loss, although direct labeled baselines rank the same gain more strongly. Internally, self-critical prompting consistently reduces hidden-state sensitivity across layers in three open-weight models. This supports prompt-induced local stabilization rather than a purely output-level abstention pattern, but it does not imply calibration: audit-defined overconfident errors are not clearly more locally sensitive than confidently correct answers, so some high-confidence errors may be stable and miscalibrated rather than simply fragile.

Chinese Translation

大型语言模型中的高置信度错误常被视为脆弱内部推理的证据。我们研究了另一种可能性：稳定误校准，其中一个自信的错误答案在小扰动下保持局部稳定。我们结合了两种诊断工具：一种基于标签的输出级审计分数，它通过置信度变化和在强制回答基线下的过度自信错误对领域进行排名，以及一种内部敏感性探测器，用于测量隐状态的移动。在一个多领域的二元事实审计集上，这个审计分数追踪了自我批评意识的弃权如何减少决策损失，尽管直接标记的基线对同样的增益给予了更强的排名。在内部，自我批评提示在三个开放权重模型的各层中一致地降低了隐状态敏感性。这支持了由提示引发的局部稳定，而不是单纯的输出级弃权模式，但这并不意味着校准：审计定义的过度自信错误并没有明显比自信正确答案更具局部敏感性，因此一些高置信度错误可能是稳定且误校准的，而不仅仅是脆弱的。

View on arXiv Download PDF AI Translation

cs.AI / 8 / 2608.13598

Measuring Cross-Task Behavioral Consistency in Language Model Agents

测量语言模型代理的跨任务行为一致性

Banerjee, Amritesh, Raichura, Pranil

Abstract

Agent evaluation relies almost entirely on outcome metrics such as success rate, which capture whether an agent succeeds but not how consistently it behaves. We argue that behavioral consistency across tasks is a distinct and measurable property, and we introduce the Behavioral Consistency Metric (BCM) to quantify it. BCM trains a model to predict task success from behavioral features of agent execution traces, derives a per-trajectory feature-attribution vector, and measures the mean pairwise similarity of these vectors within an agent system. Across roughly 9,000 trajectories from six language model agents on software engineering tasks, our central finding is that cross-task and within-task consistency are distinct axes that can diverge: some systems are locally reproducible, behaving similarly on repeated attempts at one task, yet globally fragmented, with no stable strategy across different tasks, while others are consistent at both scales. Prior work measures only same-task reproducibility and so cannot observe this separation. We further find that consistency is not reducible to success rate, since systems with comparable success can differ sharply in consistency, and that the frontier-versus-open-source consistency gap persists under a within-task control that holds task difficulty constant. We position BCM as a process-level reliability signal that complements outcome metrics, and we are explicit about the conditions under which it is meaningful.

Chinese Translation

代理评估几乎完全依赖于成功率等结果指标，这些指标能够捕捉代理是否成功，但无法反映其行为的一致性。我们认为跨任务的行为一致性是一个独特且可测量的属性，并引入了行为一致性指标（Behavioral Consistency Metric, BCM）来量化这一属性。BCM训练一个模型，通过代理执行轨迹的行为特征预测任务成功，推导出每个轨迹的特征归因向量，并测量这些向量在代理系统内的平均成对相似性。在大约9000条来自六个语言模型代理在软件工程任务上的轨迹中，我们的核心发现是跨任务和同任务一致性是两个不同的维度，可能会出现分歧：一些系统在单一任务的重复尝试中表现出局部可重复性，但在不同任务之间却没有稳定的策略，显示出全局的碎片化，而另一些系统在这两个维度上都表现出一致性。之前的研究仅测量同任务的可重复性，因此无法观察到这种分离。我们进一步发现，一致性无法简化为成功率，因为在成功率相当的情况下，系统的一致性可能有显著差异，并且在保持任务难度不变的同任务控制下，前沿与开源的一致性差距仍然存在。我们将BCM定位为一个过程级的可靠性信号，补充结果指标，并明确指出其有意义的条件。

View on arXiv Download PDF AI Translation

cs.AI / 9 / 2608.13604

Cross-Disciplinary Taxonomy and Modeling of Misunderstanding Generation, Amplification, and Detection, from Pragmatics to AI Agents

跨学科误解生成、放大与检测的分类与建模：从语用学到人工智能代理

Abbaschian, Babak

Abstract

Detection of misunderstanding is an urgent problem to solve because communication has moved away from real-time, in-person interaction and is increasingly handled by AI-mediated channels. This shift cuts communicators off from the resources repair depends on faster than new means of detection are being built. In this paper we analyse misunderstanding as a layered process in which a divergence is generated, may then be amplified, and is either detected and repaired or left to persist unnoticed. Consolidating accounts from nine fields of research that do not ordinarily cite one another, we identify eleven exact failure modes and show that each operates at a specific point in a communicative process rather than anywhere within it. Those points give eight analytical layers, derived from the literature rather than adopted from an existing model. Eight of the mechanisms primarily generate a divergence, two primarily amplify one already present, and one governs whether a divergence is detected and repaired. We model the eight layers formally, extending information and communication theory from the transmission of signals to the reconstruction of meaning, and we supply a source-by-source evidence matrix that makes every rating auditable, a coding manual, and nine analysed dialogue cases. No prior classification of misunderstanding both locates mechanisms at points in the process and types them by function.

Chinese Translation

误解的检测是一个亟待解决的问题，因为沟通已不再是实时的面对面互动，而越来越多地通过人工智能介导的渠道进行。这一转变使得沟通者失去了修复所依赖的资源，速度快于新检测手段的建立。在本文中，我们将误解分析为一个分层过程，其中产生了偏差，可能会被放大，并且要么被检测和修复，要么被忽视而持续存在。整合来自九个通常不相互引用的研究领域的观点，我们识别出十一种具体的失败模式，并展示每种模式在交流过程中运作的特定点，而不是在过程的任何位置。这些点提供了八个分析层次，源自文献而非现有模型。八种机制主要生成偏差，两种机制主要放大已存在的偏差，一种机制决定偏差是否被检测和修复。我们正式建模这八个层次，将信息与通信理论从信号传输扩展到意义重构，并提供了一个逐源证据矩阵，使每个评分可审计，以及一个编码手册和九个分析对话案例。此前对误解的分类既定位了机制在过程中的具体点，又按功能对其进行了分类。

View on arXiv Download PDF AI Translation

cs.AI / 10 / 2608.13605

Active Perception for Embodied Disambiguation

用于具身消歧的主动感知

Liu, Yiwei, Yang, Luwei

Abstract

Natural language provides robots with a flexible task interface, but target ambiguity in embodied environments arises not only from user intent; it can also result from missing taskrelevant physical evidence in the current observation. Existing interactive disambiguation methods primarily obtain additional information by asking the user, whereas occlusion, restricted viewpoints, unreadable text, and unobserved targets require the robot to actively change its observation. We propose an active-perception framework for embodied target disambiguation that uses active observation as the backbone for information acquisition and uses a vision-language model to decide, on the basis of accumulated visual evidence and interaction information, whether to continue observing, request clarification, or complete target selection. Active observation can both directly recover missing discriminative evidence and reveal object names, labels, and semantic attributes, thereby improving user clarification when it remains necessary. Real-robot experiments show that the framework combines physical information acquisition and userintent clarification within a unified embodied disambiguation process.

Chinese Translation

自然语言为机器人提供了灵活的任务接口，但在具身环境中，目标模糊性不仅源于用户意图，还可能由于当前观察中缺失与任务相关的物理证据而产生。现有的交互式消歧方法主要通过询问用户来获取额外信息，而遮挡、受限视角、不可读文本和未观察到的目标则要求机器人主动改变其观察方式。我们提出了一种用于具身目标消歧的主动感知框架，该框架以主动观察为信息获取的基础，并利用视觉-语言模型根据累积的视觉证据和交互信息来决定是继续观察、请求澄清，还是完成目标选择。主动观察可以直接恢复缺失的区分性证据，并揭示物体名称、标签和语义属性，从而在必要时改善用户的澄清。真实机器人实验表明，该框架在统一的具身消歧过程中结合了物理信息获取和用户意图澄清。

View on arXiv Download PDF AI Translation

cs.AI / 11 / 2608.13606

MobileMem: Learning from a Year of Mobile Experiences

MobileMem：从一年的移动体验中学习

Deng, Xinle, Xue, Yida, Ru, Xiangyuan, Xu, Haoming, Qiao, Shuofei, Wang, Mengru, Chen, Yijun, Xu, Buqiang, Jiang, Chen, Jiang, Yuchen Eleanor, Wang, Lizhong, Wang, Jianfeng, Zeng, Li, Wang, Haofen, Qi, Guilin, Chen, Huajun, Zhang, Ningyu

Abstract

The next generation of AI agents is increasingly moving beyond systems that answer isolated questions toward persistent personal assistants that can understand, remember, and continuously learn from users' experiences. Such assistants require long-term memory to accumulate and leverage user-specific experiences over time, yet existing benchmarks remain inadequate for realistic mobile settings, where experiences are heterogeneous, multimodal, evolving, and deeply personal. We introduce MobileMem, a benchmark and framework for studying on-device long-term memory, grounded in a year-scale collection of mobile experiences. MobileMem employs a knowledge-grounded synthesis pipeline to construct coherent and temporally consistent long-horizon trajectories from user-app sessions. It provides complementary text and multimodal settings covering multi-hop and temporal reasoning, knowledge updating, and implicit preference inference. Specifically, MobileMem enables agents to remember the past, understand the present, and adapt to the future. By modeling experiences rather than isolated facts, MobileMem moves memory beyond information retrieval toward experiential intelligence for continuous personal learning.

Chinese Translation

下一代人工智能代理正逐渐超越仅回答孤立问题的系统，向能够理解、记忆并持续从用户体验中学习的持久个人助手发展。这类助手需要长期记忆，以便随着时间的推移积累和利用用户特定的经验，但现有基准在现实移动环境中仍显不足，因为这些经验是异质的、多模态的、不断发展的，并且深具个人性。我们提出了MobileMem，一个用于研究设备端长期记忆的基准和框架，基于一年规模的移动体验数据集。MobileMem采用知识驱动的合成管道，从用户-应用会话中构建连贯且时间一致的长时间轨迹。它提供了互补的文本和多模态设置，涵盖多跳和时间推理、知识更新以及隐含偏好推断。具体而言，MobileMem使代理能够记住过去、理解现在并适应未来。通过对经验而非孤立事实的建模，MobileMem将记忆从信息检索推进到持续个人学习的经验智能。

View on arXiv Download PDF AI Translation

cs.AI / 12 / 2608.13607

No Universal Signal Predicts Sample-Level LLM Regression under Version Updates

没有普遍信号可以预测版本更新下的样本级 LLM 回归

Sheng, Jia, Lu, Yiwei

Abstract

Frontier LLMs are updated frequently and typically outperform their predecessors in aggregate. But aggregate gains say little about individual samples: an update can still cause sample-level regression, where a response correct under the old model becomes incorrect under the new one. This paper studies how to predict such regressions from signals available at inference time. We compare single-model signals (confidence, logit margin, attention entropy) against cross-version signals (output KL divergence, likelihood drift, token-level KL, representation drift) under a unified added-value test that isolates each signal's gain over a confidence baseline. Across six benchmarks in three task families (multiple-choice question answering, or MCQ; math reasoning; code generation) and six model update pairs, we find that (1) signal effectiveness is task-dependent: confidence is strongest on MCQ and simpler math, while likelihood/KL signals give the most frequent gains on harder math and code; (2) no signal is universally best across model updates either; and (3) some cross-version signals stay informative even when confidence fails, including without labels, which supports a proof-of-concept selective fallback that routes high-risk samples back to the old model. Practitioners can use these task-level patterns to choose which regression signal to trust for a given update. Code is available at https://github.com/jiashengsally/llm-regression-signals.

Chinese Translation

前沿的 LLM（大型语言模型）更新频繁，通常在整体上优于其前身。然而，整体增益对个别样本的意义不大：一次更新仍可能导致样本级回归，即在旧模型下正确的响应在新模型下变得不正确。本文研究如何从推理时可用的信号中预测这种回归。我们在一个统一的增值测试下比较了单模型信号（置信度、对数边际、注意力熵）与跨版本信号（输出 KL 散度、似然漂移、令牌级 KL、表示漂移），该测试能够隔离每个信号相对于置信度基线的增益。在三个任务家族（多项选择问答（MCQ）、数学推理、代码生成）和六对模型更新的六个基准测试中，我们发现：(1) 信号的有效性依赖于任务：在 MCQ 和简单数学中，置信度最强，而在更难的数学和代码中，似然/KL 信号提供了最频繁的增益；(2) 没有任何信号在模型更新中普遍最佳；(3) 一些跨版本信号在置信度失效时仍然保持信息量，包括在没有标签的情况下，这支持了一种概念验证的选择性回退机制，将高风险样本重新路由回旧模型。实践者可以利用这些任务级模式来选择在给定更新中值得信赖的回归信号。代码可在 https://github.com/jiashengsally/llm-regression-signals 获取。

View on arXiv Download PDF AI Translation

cs.AI / 13 / 2608.13608

Evaluating Agentic Learning Harness Capabilities Without Labels via the Scaling Hypothesis

通过缩放假设评估无标签的自主学习工具的能力

Luthra, Aryan, Jain, Kshitij, Arya, Siddharth, Filar, Bobby, Bertiger, Anna

Abstract

Agentic "Continual Learning Harnesses", systems that pair an LLM with retrieval or memory to improve from feedback without retraining, have shown growing value in cybersecurity. But their value is conventionally measured by gains against labeled benchmarks, an approach that often fails in operational security settings. Benchmark labels are scarce, stale, and unrepresentative, so a practitioner often cannot tell whether a given harness helps at all or which of two is better for their task. Traditional LLM-as-a-judge offers little signal because it is no stronger than the agent it evaluates, and distillation is unreliable on scarce, sporadic, and biased labels. We propose a framework for evaluating learning harnesses end-to-end without a labeled benchmark, grounded in the scaling hypothesis. A stronger teacher model provides sparsely sampled corrections to a smaller student with a continual learning harness. We score a harness by how much its student converges toward the teacher over time. Across security tasks, model families, and harness designs, we show that improvement relative to the teacher correlates with improvement relative to a held-out gold standard, validating teacher-relative lift as a proxy for true harness uplift when labels are absent. We further show that LLM-as-a-judge between similarly powered models yields no usable signal. These results suggest that a teacher-sized model can be improved through the same harness when humans provide the same kind of sparse, high-precision corrections.

Chinese Translation

自主的“持续学习工具”是将大型语言模型（LLM）与检索或记忆相结合，以便在不重新训练的情况下从反馈中改进的系统，在网络安全领域显示出越来越大的价值。但其价值通常通过与标记基准的比较来衡量，这种方法在操作安全环境中往往失败。基准标签稀缺、过时且不具代表性，因此从业者常常无法判断某个工具是否有帮助，或者在两个工具中哪个更适合他们的任务。传统的LLM作为评判者提供的信号微弱，因为它的能力不强于被评估的代理，而在稀缺、偶发和偏见标签上进行蒸馏是不可靠的。我们提出了一个框架，用于在没有标记基准的情况下端到端评估学习工具，基于缩放假设。一个更强的教师模型为一个配备持续学习工具的小型学生提供稀疏采样的修正。我们通过学生在时间上向教师的收敛程度来评分一个工具。在安全任务、模型家族和工具设计中，我们展示了相对于教师的改进与相对于保留的金标准的改进相关联，验证了在缺乏标签时，教师相对提升作为真实工具提升的代理。我们进一步表明，在相似能力模型之间使用LLM作为评判者不会产生可用信号。这些结果表明，当人类提供相同类型的稀疏、高精度修正时，教师大小的模型可以通过相同的工具得到改进。

View on arXiv Download PDF AI Translation

cs.AI / 14 / 2608.13612

SemPlan: Benchmarking Structured Semantic Planning for LLM-Based Queries over Enterprise Data

SemPlan：基于大型语言模型的企业数据结构化语义规划基准测试

Teixeira, Bruno Santos

Abstract

Natural-language interfaces to enterprise data must translate underspecified requests into governed, executable behavior while controlling invalid queries, policy failures, cost, and nondeterminism. SemPlan Benchmark evaluates this architectural design space with a deterministic synthetic bilingual benchmark containing 1,800 cases in English and Brazilian Portuguese; 1,200 cases form the frozen scientific evaluation subset. Four architectures are compared under the same model configuration: direct SQL generation (A1), a bounded tool-agent baseline (A2), structured semantic-request generation followed by deterministic planning and execution (A3), and a clarification/stateful semantic-plan variant (A4). Across 4,800 primary records, answer correctness was low in absolute terms: 22.25% for A1, 22.58% for A2, 25.67% for A3, and 24.25% for A4. A3 had the highest observed correctness and significantly exceeded A1, A2, and A4 in the pre-specified paired correctness analysis, while A1 retained the highest policy-correct rate and the lowest unsafe-or-invalid rate. A4 had the lowest mean API cost and lowest false-refusal rate. On a preselected 150-case stability subset, answer-correct repeatability ranged from 92.00% to 98.67%. The results support a trade-off interpretation rather than a universal ranking: additional structural constraints changed failure modes and efficiency, but did not monotonically improve correctness or solve ambiguity and multi-turn state consistency.

Chinese Translation

企业数据的自然语言接口必须将不明确的请求转化为受控的、可执行的行为，同时控制无效查询、政策失败、成本和非确定性。SemPlan基准测试评估了这一架构设计空间，使用了一个包含1800个案例的确定性合成双语基准，涵盖英语和巴西葡萄牙语；其中1200个案例构成了冻结的科学评估子集。比较了四种架构在相同模型配置下的表现：直接SQL生成（A1）、有界工具代理基线（A2）、结构化语义请求生成后进行确定性规划和执行（A3），以及澄清/状态语义规划变体（A4）。在4800个主要记录中，答案的正确性在绝对值上较低：A1为22.25%，A2为22.58%，A3为25.67%，A4为24.25%。A3的观察到的正确性最高，并在预先指定的配对正确性分析中显著超过了A1、A2和A4，而A1保持了最高的政策正确率和最低的不安全或无效率。A4的平均API成本最低，虚假拒绝率也最低。在预选的150个案例稳定性子集中，答案正确的重复性范围为92.00%至98.67%。结果支持一种权衡解释，而非普遍排名：额外的结构约束改变了失败模式和效率，但并未单调改善正确性或解决模糊性和多轮状态一致性问题。

View on arXiv Download PDF AI Translation

cs.AI / 15 / 2608.13617

How Compliant is Sepsis Treatment? An Expert-Guided Neuro-symbolic Pipeline for Generating Clinical Compliance Insights

脓毒症治疗的合规性如何？一个专家指导的神经符号管道用于生成临床合规性洞察

Tripathi, Himanshu, Roy, Kaushik, Neupane, Subash, Rahimi, Shahram

Abstract

Verifying whether clinical care follows evidence-based protocols is a natural neuro-symbolic problem, yet the safety-critical setting defeats either paradigm alone. We present an expert-guided pipeline that constrains a large language model strictly to semantic normalization, mapping messy drug and microbiology strings onto a fixed clinical vocabulary, while a Sugeno fuzzy inference system reasons over the normalized events. The fuzzy layer encodes eight Surviving Sepsis Campaign bundle rules and replaces binary judgments with graded scores in [0,1]. Applied to 2,438 MIMIC-IV v3.1 sepsis episodes, it surfaces antibiotic timing as the most critical breakdown (mean 0.24, 13% within one hour), Hour-1 underperformance (mean 36.7%), a 51% elevated-lactate drop-off, and descriptive differences in ICU stay across compliance groups (3.8 versus 5.1 days).

Chinese Translation

验证临床护理是否遵循基于证据的协议是一个自然的神经符号问题，但安全关键环境使得单一范式难以应对。我们提出了一个专家指导的管道，该管道严格限制大型语言模型仅进行语义标准化，将混乱的药物和微生物字符串映射到固定的临床词汇上，同时使用Sugeno模糊推理系统对标准化事件进行推理。模糊层编码了八项生存脓毒症运动（Surviving Sepsis Campaign）捆绑规则，并用[0,1]范围内的分级评分替代了二元判断。应用于2438个MIMIC-IV v3.1脓毒症病例，结果显示抗生素使用时机是最关键的失误（平均0.24，13%在一小时内），第一小时表现不佳（平均36.7%），乳酸升高后51%的下降，以及不同合规组在ICU停留时间上的描述性差异（3.8天对比5.1天）。

View on arXiv Download PDF AI Translation

cs.AI / 16 / 2608.13618

Algorithm Design and Physician Liability

算法设计与医生责任

Luan, Shujie, Singh, Shubhranshu, Dai, Tinglong

Abstract

A single clinical algorithm can deliver unequal accuracy across patient groups, and concern about such disparity has grown as artificial intelligence (AI) spreads through clinical decision-making. In response, a liability rule introduced in the United States holds healthcare providers responsible when their reliance on disparate algorithms contributes to erroneous clinical decisions. We examine how such liability considerations reshape (i) an AI firm's algorithm design decisions that drive group-specific accuracy and (ii) a physician's decisions to use AI in healthcare delivery. The AI firm designs an algorithm for two patient groups, and improving accuracy for the disadvantaged group is more costly. The physician (who remains the accountable decision-maker) then decides whether to consult AI, weighing the reduction in clinical uncertainty against expected liability exposure when AI errors disproportionately affect the disadvantaged group. We find the liability rule can induce disparate use of AI: the physician may reduce AI use overall and, over an intermediate range of liability, rely on AI less for disadvantaged patients. The effect is non-monotone. As liability increases, the physician's use of AI for disadvantaged patients first declines, then rises as the firm reallocates investment toward reducing disparity or switches to an equal-accuracy design. Mandating equal algorithmic accuracy across patient groups can then inadvertently harm both groups, because a uniform accuracy requirement distorts the firm's investment incentives and the physician's equilibrium AI-use decisions.

Chinese Translation

单一的临床算法在不同患者群体中可能会产生不均等的准确性，随着人工智能（AI）在临床决策中的广泛应用，对这种差异的关注也在增加。对此，美国引入了一项责任规则，要求医疗提供者在其依赖不同算法导致错误临床决策时承担责任。我们研究了这种责任考量如何重塑（i）AI公司在推动特定群体准确性方面的算法设计决策，以及（ii）医生在医疗服务中使用AI的决策。AI公司为两个患者群体设计算法，而改善弱势群体的准确性成本更高。医生（作为负责的决策者）随后决定是否咨询AI，权衡临床不确定性的减少与当AI错误对弱势群体产生不成比例影响时的预期责任风险。我们发现责任规则可能导致AI的使用不均：医生可能整体上减少AI的使用，并且在责任中等范围内，对弱势患者的AI依赖减少。这个影响是非单调的。随着责任的增加，医生对弱势患者使用AI的频率先下降，随后随着公司将投资重新分配到减少差异或转向均等准确性设计而上升。强制要求不同患者群体之间的算法准确性相等可能会无意中对两个群体造成伤害，因为统一的准确性要求扭曲了公司的投资激励和医生的均衡AI使用决策。

View on arXiv Download PDF AI Translation

cs.AI / 17 / 2608.13621

Your Probabilistic JEPA Is Secretly a Hidden Markov Model: A State-Space Interpretation of Joint-Embedding Predictive Learning

你的概率 JEPA 实际上是一个隐马尔可夫模型：联合嵌入预测学习的状态空间解释

Huang, Yongchao

Abstract

A hidden Markov model (HMM) combines three roles: inference of a hidden-state belief from observations, propagation through a Markov transition, and emission back to observation space. We show that full, time-indexed Predictive Information Bottleneck VJEPA (PIB-VJEPA) exposes the same computational structure: a stochastic context encoder plays the role of an amortized filtering distribution, a probabilistic predictor defines latent-state dynamics, and a decoder, inverse target encoder, or induced implicit conditional supplies the emission direction. We distinguish 4 progressively stronger levels of correspondence and give sufficient conditions for exact sequence-level HMM equivalence. To make the connection concrete, we introduce Markov-Chain JEPA (MCJEPA), which replaces the latent predictor by a learned transition matrix; in the finite time-homogeneous case, matrix powers guarantee exact multi-horizon Chapman--Kolmogorov consistency. Conditioned discrete-state transitions, continuous-state Markov kernels, and continuous-time dynamics extend this construction, while deterministic temporal JEPA appears as a degenerate Dirac-kernel special case. We further interpret predictive information-bottleneck learning as seeking a compact predictive state: compression promotes minimality, while residual predictability tests sufficiency. Controlled experiments support transition composition, the filtering interpretation, predictive Markovization in a known synthetic process, and the distinction between JEPA latent prediction and HMM-style sequence learning. Together, these results give temporal JEPA a principled state-space interpretation.

Chinese Translation

隐马尔可夫模型（HMM）结合了三个角色：从观察中推断隐藏状态信念，通过马尔可夫转移传播，以及向观察空间发射。我们展示了完整的、时间索引的预测信息瓶颈 VJEPA（PIB-VJEPA）暴露了相同的计算结构：随机上下文编码器充当了摊销过滤分布，概率预测器定义了潜在状态动态，而解码器、逆目标编码器或诱导的隐式条件提供了发射方向。我们区分了四个逐渐增强的对应级别，并给出了精确序列级 HMM 等价的充分条件。为了使这一联系具体化，我们引入了马尔可夫链 JEPA（MCJEPA），它通过学习的转移矩阵替代了潜在预测器；在有限时间齐次情况下，矩阵幂保证了精确的多时间步 Chapman-Kolmogorov 一致性。条件离散状态转移、连续状态马尔可夫核和连续时间动态扩展了这一构造，而确定性时间 JEPA 作为一个退化的 Dirac 核特例出现。我们进一步将预测信息瓶颈学习解释为寻求一个紧凑的预测状态：压缩促进最小性，而残余可预测性测试充分性。受控实验支持转移组合、过滤解释、已知合成过程中的预测马尔可夫化，以及 JEPA 潜在预测与 HMM 风格序列学习之间的区别。这些结果共同为时间 JEPA 提供了一个有原则的状态空间解释。

View on arXiv Download PDF AI Translation

cs.AI / 18 / 2608.13622

ARC: Fair Relative Advantage Comparison in Open-Ended Real-World Interaction

ARC：开放式真实世界交互中的公平相对优势比较

Tong, Yongqi, Faith, Tan Li Hui, Marcus, Choy Zhen Wen, Jin, Zhou, Fu, Kewei, Yang, Jiang-Ming, Li, Jianshe, Zhang, Xin

Abstract

Open-ended real-world interaction admits multiple valid behaviors: an agent may answer directly, ask for clarification, provide progress updates, or confirm before acting. This flexibility breaks a core assumption behind group-based RL: rollouts compared within a group are no longer guaranteed to be behaviorally comparable. As a result, reward-model preferences over interaction style can distort relative advantages and steer optimization toward reward-preferred behaviors rather than context-appropriate ones. We formalize this as a \textit{reward fairness problem} and propose \textbf{ARC} (Advantage Regularization via Conditioning), a training recipe that restores fairer relative comparison through strategy-conditioned rollout grouping, together with hybrid rewards and entropy regularization. We study ARC in our proposed \inter, a novel paradigm for responsive, steerable, and execution-aware user-agent interaction that decouples user-visible communication from latent reasoning and tool use. \inter\ also provides the annotation and distillation pipeline for constructing \inter-86K, our strategy-annotated training corpus for supervised and RL training. Empirically, ARC substantially strengthens the core $\tau/\tau^2$ tool-use benchmarks, while \inter\ reduces time-to-first-token from 4.91s to 1.27s relative to a think-style baseline. Together, these results suggest that a central bottleneck in open-ended interactive learning is not only how agents are rewarded, but whether their behaviors are compared fairly in the first place. The ARC implementation and \inter-86K training data will be released.

Chinese Translation

开放式真实世界交互允许多种有效行为：代理可以直接回答、请求澄清、提供进展更新或在行动前确认。这种灵活性打破了基于群体的强化学习（RL）背后的核心假设：在一个群体内比较的回合不再保证在行为上是可比的。因此，基于奖励模型的交互风格偏好可能扭曲相对优势，并将优化引导向奖励偏好的行为，而非适合上下文的行为。我们将其形式化为 extit{奖励公平性问题}，并提出 extbf{ARC}（通过条件化的优势正则化），这是一种训练方案，通过策略条件的回合分组、混合奖励和熵正则化来恢复更公平的相对比较。我们在我们提出的 extit{inter}中研究ARC，这是一种新颖的响应式、可引导和执行感知的用户-代理交互范式，它将用户可见的沟通与潜在推理和工具使用解耦。 extit{inter}还提供了构建 extit{inter-86K}的注释和蒸馏管道，这是我们用于监督学习和强化学习训练的策略注释训练语料库。从经验上看，ARC显著增强了核心的$ au/ au^2$工具使用基准，而 extit{inter}将首次令牌的时间从4.91秒减少到1.27秒，相较于思考风格的基线。这些结果表明，开放式互动学习的一个核心瓶颈不仅在于代理如何获得奖励，还在于它们的行为是否首先得到了公平的比较。ARC的实现和 extit{inter-86K}训练数据将被发布。

View on arXiv Download PDF AI Translation

cs.AI / 19 / 2608.13625

Reward Machines for Signal Temporal Logic

信号时序逻辑的奖励机器

Bozkurt, Alper Kamil, Zhang, Shangtong, Motai, Yuichi

Abstract

Signal temporal logic (STL) provides a formal language for specifying real-time properties of real-valued observations, along with a quantitative robustness score for monitoring satisfaction. Control synthesis from STL specifications is of interest since manual controller design becomes infeasible as real-world systems grow in complexity. Moreover, many modern autonomous and AI-enabled systems lack accurate and complete system models, which makes optimization-based synthesis approaches unsuitable and motivates learning-based control. Prior work uses STL robustness scores as rewards in reinforcement learning (RL) to obtain control policies satisfying given specifications; however, robustness depends on execution history, leading to intractable state space expansion for general long-horizon specifications with arbitrarily nested temporal operators. This work introduces a novel automata-based approach that provides an efficient memory mechanism and associated Markovian rewards suitable for RL frameworks. Our approach constructs a timed alternating automaton from the given STL specifications, augments the state space with automaton locations and clock valuations, and derives rewards from the automaton acceptance condition. We empirically demonstrate that our approach learns policies that achieve higher robustness scores and satisfaction rates than those learned by existing approaches using robustness-based rewards.

Chinese Translation

信号时序逻辑（STL）为指定实值观测的实时属性提供了一种形式化语言，并提供了用于监测满足度的定量鲁棒性评分。从 STL 规范进行控制合成受到关注，因为随着现实世界系统复杂性的增长，手动控制器设计变得不可行。此外，许多现代自主和人工智能驱动的系统缺乏准确和完整的系统模型，这使得基于优化的合成方法不适用，并促使基于学习的控制方法的发展。之前的研究使用 STL 鲁棒性评分作为强化学习（RL）中的奖励，以获得满足给定规范的控制策略；然而，鲁棒性依赖于执行历史，导致对于具有任意嵌套时序运算符的一般长时间范围规范，状态空间扩展变得不可处理。本研究提出了一种新颖的基于自动机的方法，提供了一种高效的记忆机制和适合 RL 框架的马尔可夫奖励。我们的方法从给定的 STL 规范构建一个定时交替自动机，利用自动机位置和时钟值扩展状态空间，并从自动机接受条件中推导奖励。我们通过实证展示了我们的方法学习的策略在鲁棒性评分和满足率上优于现有基于鲁棒性奖励的方法。

View on arXiv Download PDF AI Translation

cs.AI / 20 / 2608.13626

A Calibrated Test of Internal Action Maps: State Signals Without Global Affine Closure

内部行动图的校准测试：没有全局仿射闭包的状态信号

Yang, Dekun

Abstract

A hidden state signal can be decodable or causally usable without supporting a reusable action map. We test whether action maps fitted without a source reach its natural post-action activation and compose. We organize the tests as an evidence lattice and validate the geometric branch on a known affine S_5 carrier: all held-source folds pass one-step, composition, inverse, decoding, and commutativity gates. Structured curvature and held-domain conjugacy raise error monotonically, but only 23/30 strongest cells flip a closure gate, bounding rather than universalizing calibration. In post-trained Qwen/Qwen3-4B, frozen final-token h28 affine maps have mean held-entity error .519, versus .398 for within-test-domain cross-fit. Seven randomized entity splits and map geometry do not support a purely entity-specific account. Earlier h4/h16 layers fit one-step transitions better, but h4 conflict-state decoding is weak and lexical controls remain unresolved. Three matched intervention datasets regenerated from one frozen checkpoint show causal effects only at h28/h36. Outcome-aware refitting improves h28 one-step error to .474 (.469 with weighting), yet no refit passes composition. Learned finite worlds likewise preserve relative algebraic signals or shared charts without held-source affine closure. Within the tested carriers, state availability, causal use, local geometry, and reusable closure are separable. The result is limited to one pretrained model, sampled final-token layers, two finite worlds, and the tested affine or diagnostic function classes.

Chinese Translation

一个隐藏状态信号可以在没有支持可重用行动图的情况下被解码或因果使用。我们测试在没有源的情况下拟合的行动图是否能够达到其自然的后行动激活并进行组合。我们将测试组织为证据格，并在已知的仿射 S_5 承载体上验证几何分支：所有持源折叠通过了一步、组合、逆、解码和交换门。结构曲率和持域共轭单调增加错误，但只有 23/30 个最强单元翻转了闭包门，从而限制而非普遍化校准。在后训练的 Qwen/Qwen3-4B 中，冻结的最终标记 h28 仿射图的平均持实体错误为 0.519，而在测试域内交叉拟合为 0.398。七个随机实体分割和图几何不支持纯粹的实体特定解释。较早的 h4/h16 层更好地拟合一步过渡，但 h4 冲突状态解码较弱，词汇控制仍未解决。从一个冻结检查点再生的三个匹配干预数据集仅在 h28/h36 处显示因果效应。结果感知的重新拟合将 h28 一步错误改善至 0.474（加权后为 0.469），但没有重新拟合通过组合。在测试的承载体中，状态可用性、因果使用、局部几何和可重用闭包是可分的。该结果仅限于一个预训练模型、采样的最终标记层、两个有限世界和测试的仿射或诊断函数类。

View on arXiv Download PDF AI Translation

cs.AI / 21 / 2608.13630

Exploring ESC Winners with Nested Diagrams

使用嵌套图探索ESC获奖者

Sharma, Anurag, Nöhre, Marcel, Stumme, Gerd

Abstract

We present ConceptFlow, a scikit-learn-compatible Python library for Formal Concept Analysis that constructs and renders nested line diagrams from many-valued formal contexts. Given a many-valued context and a partition of its attributes into conceptual scales, ConceptFlow performs conceptual scaling, computes the factor lattices, identifies filled nodes of the corresponding subdirect product, and produces an interactive visualization. We apply ConceptFlow to the winners of the Eurovision Song Contest from 1975 to 2025, exploring relationships between voting patterns and musical characteristics. Voting support is captured by an outer scale spanning regional, cultural, historical, and political dimensions, while an inner scale captures musical characteristics via tempo and key. The resulting nested line diagram reveals implications across both scales, exposing dependencies between how winning entries were voted for and the musical properties they share.

Chinese Translation

我们提出了ConceptFlow，这是一个与scikit-learn兼容的Python库，用于形式概念分析，能够从多值形式背景中构建和渲染嵌套线图。给定一个多值背景及其属性的概念尺度划分，ConceptFlow执行概念尺度化，计算因子格，识别相应子直接积的填充节点，并生成交互式可视化。我们将ConceptFlow应用于1975年至2025年间的欧洲歌唱大赛获奖者，探索投票模式与音乐特征之间的关系。投票支持通过一个外部尺度捕捉，涵盖区域、文化、历史和政治维度，而内部尺度则通过节奏和调性捕捉音乐特征。生成的嵌套线图揭示了两个尺度之间的影响，暴露了获奖作品的投票方式与其共享的音乐属性之间的依赖关系。

View on arXiv Download PDF AI Translation

cs.AI / 22 / 2608.13662

Ontology-Grounded Project Memory for Coding Agents

基于本体的编码代理项目记忆

Adam, James

Abstract

Coding agents have become the primary means of generating new code in many software projects, and the resulting velocity of changes makes keeping track of the reasons behind those changes challenging. This paper introduces MOOSEDev, a system designed to give coding agents structured, ontology-grounded project memory. The system captures architectural decisions, lessons, constraints, and rationales in a knowledge graph exposed to agents via a Model Context Protocol (MCP) interface. Records carry lifecycle status, provenance, and supersession links, queryable via MOOSE, a proprietary neurosymbolic engine that treats the symbolic layer as the primary reasoning substrate. We compared MOOSEDev against a production vector-memory tool on a neutral public corpus of 835 typed records. MOOSEDev returned the expected answer set essentially in full (0.98-1.00) on supersession, set-completeness, and negation questions, whereas the baseline's top-k retrieval surfaced between 6% and 27%. Conversely, relevance recall and token cost were largely equivalent between the two systems. We also describe a temporal commit-history bootstrap of our own codebase, a pre-registered live trial, and lessons learned.

Chinese Translation

编码代理已成为许多软件项目中生成新代码的主要手段，而随之而来的变更速度使得追踪这些变更背后的原因变得具有挑战性。本文介绍了MOOSEDev，一个旨在为编码代理提供结构化、基于本体的项目记忆的系统。该系统通过模型上下文协议（Model Context Protocol, MCP）接口捕捉架构决策、经验教训、约束和理由，并将其存储在知识图谱中，供代理访问。记录携带生命周期状态、来源和替代链接，可以通过MOOSE这一专有的神经符号引擎进行查询，该引擎将符号层视为主要推理基础。我们将MOOSEDev与一个生产向量记忆工具在一个中立的公共语料库（包含835条类型记录）上进行了比较。MOOSEDev在替代、集合完整性和否定问题上几乎完全返回了预期的答案集（0.98-1.00），而基线的前k个检索结果仅显示了6%到27%的相关性。相反，两个系统在相关性召回和标记成本方面大致相当。我们还描述了我们自己代码库的时间提交历史引导、一个预注册的实时试验以及所获得的经验教训。

View on arXiv Download PDF AI Translation

cs.AI / 23 / 2608.13667

Second Thought: Reasoning in Parallel as LLM Agents Act and Observe

第二思考：在LLM代理行动和观察时的并行推理

Sun, Zhensu, Yang, Chengran, Lyu, Yunbo, Shi, Jieke, Lo, David

Abstract

LLM agents in the ReAct paradigm alternate between reasoning, acting, and observing, but deliberate reasoning is confined to the Thought phase: while the agent serializes an action and waits for the environment, its reasoning is frozen. We identify this recurring interval for Action and Observation as a reasoning idle window and ask whether it can host additional reasoning in parallel that serves future turns. Therefore, we propose Second Thought, a training-free inference framework that forks four auxiliary branches the instant each Thought phase concludes, decodes them concurrently with the main loop, and merges the generated thoughts back when the environment observation arrives. In this way, Second Thought relocates the added reasoning off the main thread's sequential decoding path. Across three agentic benchmarks and three reasoning LLMs, Second Thought lowers the average turn count in all nine (model,benchmark) pairs and reduces main thread decoding in six of them by up to 43% (roughly 20% on average among those settings), while leaving it essentially unchanged in a seventh; Pass@1 shows no significant change in seven of nine pairs and the two significant differences are +12.4 and +10.2 points. Against a compute-matched control that forces an equivalent budget onto the main thread's own reasoning, it attains strictly higher Pass@1 with 1.3 to 3.2 less sequential decoding in all four settings where the control applies.

Chinese Translation

在ReAct范式下，LLM代理在推理、行动和观察之间交替进行，但深思熟虑的推理仅限于思考阶段：当代理序列化一个动作并等待环境时，其推理处于冻结状态。我们将这一反复出现的行动和观察间隔称为推理闲置窗口，并探讨是否可以在此期间并行进行额外的推理，以服务于未来的回合。因此，我们提出了第二思考（Second Thought），这是一种无训练推理框架，在每个思考阶段结束的瞬间分叉出四个辅助分支，与主循环并行解码，并在环境观察到达时将生成的思考合并回去。通过这种方式，第二思考将额外的推理从主线程的顺序解码路径中移出。在三个代理基准和三个推理LLM的实验中，第二思考在所有九个（模型，基准）对中降低了平均回合数，并在其中六个中将主线程解码减少了最多43%（在这些设置中平均约20%），而在第七个中基本保持不变；Pass@1在九对中的七对没有显著变化，两个显著差异分别为+12.4和+10.2点。与一个计算匹配的控制组相比，该控制组强制将等效预算施加于主线程自身的推理，第二思考在所有四个适用设置中实现了严格更高的Pass@1，同时减少了1.3到3.2的顺序解码。

View on arXiv Download PDF AI Translation

cs.AI / 24 / 2608.13684

Learning to Assemble Novel Structures with Unfamiliar Parts under Semantic Constraints

在语义约束下学习使用不熟悉部件组装新结构

Park, Jonghyuk, Lascarides, Alex, Ramamoorthy, Subramanian

Abstract

This paper describes a neurosymbolic architecture for learning to assemble novel structures using evidence from embodied conversations and task demonstrations. We focus on scenarios where an agent encounters, after deployment, semantic constraints on structures--in other words, constraints as to which part types and features make valid structures--that were not available during training, and where it is initially unaware of the relevant structure and component part concepts. The agent must acquire and exploit such knowledge through user interactions while attempting assembly. We study this setting in a simulated toy truck assembly domain, learning from symbolic evidence encoded in natural language and from dense visual observations. Our experiments show that communicating semantic constraints through natural language (e.g., "dump trucks have a dumper") yields more data-efficient online adaptation than relying only on task demonstrations and/or only naming the parts through natural language.

Chinese Translation

本文描述了一种神经符号架构，用于学习在具身对话和任务演示的证据基础上组装新结构。我们关注的场景是，代理在部署后遇到对结构的语义约束——换句话说，即哪些部件类型和特征构成有效结构的约束——这些约束在训练期间并不可用，并且代理最初对相关的结构和组件概念并不知情。代理必须通过用户交互在尝试组装的过程中获取并利用这些知识。我们在一个模拟的玩具卡车组装领域研究这一设置，从自然语言编码的符号证据和密集的视觉观察中学习。我们的实验表明，通过自然语言传达语义约束（例如，“自卸卡车有一个卸料斗”）比仅依赖任务演示和/或仅通过自然语言命名部件更能有效地进行在线适应。

View on arXiv Download PDF AI Translation

cs.AI / 25 / 2608.13719

Coverage Aware Active Evaluation for Failure Discovery with Paired Systems

基于覆盖意识的主动评估方法用于配对系统的故障发现

Parashar, Anjali, Luo, Rachel, Sharma, Apoorva, Veer, Sushant, Schmerling, Edward, Sobolewski, Carson, Yu, Mingxin, Fan, Chuchu, Pavone, Marco

Abstract

Autonomous systems can fail in rare and heterogeneous ways, making real-world failure discovery difficult under limited testing budgets. Although cheaper proxies such as simulators, lower-fidelity systems, or related policies can be sampled extensively to find failures, proxy failures often do not transfer to the real world due to sim-to-real and system-to-system gaps. The key challenge is therefore to effectively leverage proxy system information for accurate prediction of severe target system failures. We propose an adaptive failure discovery method that combines proxy evaluations with limited target system results to guide scenario selection for target system testing. Our method learns a local predictor of target risk by correcting proxy failure signals using control-variate-inspired residual modeling. To find failures that are both likely and diverse, we combine this predictor with a support-aware mutual-information objective that favors realistic, well-supported regions while expanding coverage across failure modes. Across autonomous driving, manipulation, and quadruped velocity-tracking tasks, our method discovers up to 2$\times$ as many failures as random sampling and active-learning baselines, including severe and diverse failures missed by competing methods.

Chinese Translation

自主系统可能以稀有且异质的方式发生故障，这使得在有限测试预算下进行真实世界的故障发现变得困难。尽管可以广泛采样更便宜的代理，如模拟器、低保真度系统或相关策略来寻找故障，但代理故障往往由于模拟到真实（sim-to-real）和系统间（system-to-system）差距而无法转移到真实世界。因此，关键挑战在于有效利用代理系统信息来准确预测目标系统的严重故障。我们提出了一种自适应故障发现方法，该方法结合了代理评估和有限的目标系统结果，以指导目标系统测试的场景选择。我们的方法通过使用受控变量启发的残差建模来修正代理故障信号，从而学习目标风险的局部预测器。为了发现既可能又多样的故障，我们将该预测器与支持感知的互信息目标相结合，该目标偏向于现实的、得到良好支持的区域，同时在故障模式中扩展覆盖范围。在自主驾驶、操控和四足动物速度跟踪任务中，我们的方法发现的故障数量是随机采样和主动学习基线的最多2倍，包括被竞争方法遗漏的严重且多样的故障。

View on arXiv Download PDF AI Translation

cs.AI / 26 / 2608.13754

Explanation Multiplicity: Circuit-Level Interpretability Evidence Does Not Survive Defensible Analytic Variation

解释的多样性：电路级可解释性证据在可辩护的分析变异下不再成立

Mahale, Ajay Pravin

Abstract

The EU AI Act requires providers of high-risk systems to file technical documentation describing how the system reaches its decisions. Mechanistic interpretability is the obvious source of such evidence, and circuit discovery is its most developed instrument. We ask whether that evidence survives the condition under which it would be relied upon: two competent analysts, the same system, the same tool, different defensible settings. We pre-registered a crossed grid of seven analytic axes, every level taken from a published implementation, and mapped each discovered circuit through a deterministic claim map to a structured Annex IV statement. Across 15,840 pre-registered specifications on GPT-2 small and the indirect object identification task, of which 7,561 produced a claim, the derived statement flips across 73.2% of specification pairs (95% CI 0.725 to 0.738) and the modal claim commands 41.1% of the space. The evidence fails a filability criterion at every tolerance a conformity assessment body would plausibly accept. Standardising the single most influential choice, the evaluation metric, leaves the flip rate at 59.4%. Removing circuit size from the claim entirely and holding it fixed leaves 27.1% (95% CI 0.255 to 0.286), still above the pre-registered threshold. The circuits underlying these claims are structurally near-disjoint, median pairwise Jaccard overlap 4%, and functionally uncorrelated at Cohen's kappa 0.015, so the instability is not one mechanism described in different words. We give the filability criterion as a standalone protocol, and we report that one of the seven documented discovery objectives does not execute at all on the library's own canonical task. The study covers one model and one task, and whether the conclusion holds at scale is untested.

Chinese Translation

欧盟人工智能法案要求高风险系统的提供者提交技术文档，描述系统如何做出决策。机械可解释性显然是此类证据的来源，而电路发现是其最成熟的工具。我们探讨该证据在何种条件下能够被依赖：两个合格的分析师、相同的系统、相同的工具、不同的可辩护设置。我们预注册了七个分析轴的交叉网格，每个级别均取自已发布的实现，并通过确定性声明映射将每个发现的电路映射到结构化的附录 IV 声明中。在 15,840 个预注册的 GPT-2 small 规格和间接对象识别任务中，其中 7,561 个产生了声明，得出的声明在 73.2% 的规格对之间发生翻转（95% 置信区间 0.725 至 0.738），而模态声明占据了 41.1% 的空间。证据在每个符合性评估机构可能接受的容忍度下都未能满足可归档标准。标准化单一最具影响力的选择，即评估指标，翻转率仍为 59.4%。完全移除声明中的电路大小并保持其固定，翻转率为 27.1%（95% 置信区间 0.255 至 0.286），仍高于预注册阈值。这些声明所依据的电路在结构上几乎不相交，中位数成对 Jaccard 重叠为 4%，在 Cohen's kappa 下功能上不相关为 0.015，因此不稳定性并不是用不同措辞描述的同一机制。我们将可归档标准作为独立协议，并报告七个文档化发现目标中的一个在库的标准任务上根本无法执行。本研究涵盖了一个模型和一个任务，结论在大规模下是否成立尚未测试。

View on arXiv Download PDF AI Translation

cs.AI / 27 / 2608.13767

Simulation-Aware In-Context Policy Improvement for LLM-Aided Analog Layout Refinement

面向仿真的上下文策略改进用于大语言模型辅助的模拟布局优化

Liu, Bingyang, Wei, Ziming, Gao, Xiaohan, Pan, David Z.

Abstract

Analog IC layout design remains a labor-intensive iterative process dominated by simulation-driven refinement. Although end-to-end layout generators accelerate initial placement and routing, they still require experts to manually tune layout optimization parameters with repeated post-layout simulations for stringent design specifications. While Bayesian Optimization (BO) is widely adopted for parameter tuning in analog IC design, at the layout level it typically requires hundreds to thousands of evaluations, each involving costly parasitic extraction and post-layout simulation, which makes it impractical. Recently, Large Language Models (LLMs) have demonstrated potential in improving the sample efficiency of such simulation-driven tuning. However, their restricted access to geometric layout context and design-specific heuristics limits their ability to manipulate the layout optimization process. In this paper, we propose a simulation-aware LLM multi-agent framework that performs in-context policy improvement (ICPI) by iteratively updating layout optimization parameters exposed by an analog layout generator through an act-observe-reflect loop on compact structured layout representations. Experiments on real-world analog circuits show that, with only tens of post-layout simulations, our approach improves post-layout performance over the generator's built-in heuristics and BO-based tuning method.

Chinese Translation

模拟集成电路布局设计仍然是一个劳动密集型的迭代过程，主要依赖于仿真驱动的优化。尽管端到端布局生成器加速了初始放置和布线，但仍然需要专家通过反复的后布局仿真手动调整布局优化参数，以满足严格的设计规范。虽然贝叶斯优化（Bayesian Optimization, BO）在模拟集成电路设计中的参数调优中被广泛采用，但在布局层面上通常需要数百到数千次评估，每次评估都涉及昂贵的寄生提取和后布局仿真，这使得其在实际应用中不够可行。最近，大语言模型（Large Language Models, LLMs）在提高此类仿真驱动调优的样本效率方面显示出了潜力。然而，它们对几何布局上下文和设计特定启发式的有限访问限制了它们操控布局优化过程的能力。在本文中，我们提出了一种面向仿真的LLM多智能体框架，通过在紧凑结构化布局表示上进行行为-观察-反思循环，迭代更新由模拟布局生成器暴露的布局优化参数，从而实现上下文策略改进（In-Context Policy Improvement, ICPI）。在真实世界的模拟电路实验中，我们的方法在仅进行几十次后布局仿真的情况下，显著提升了后布局性能，相较于生成器内置的启发式和基于BO的调优方法。

View on arXiv Download PDF AI Translation

cs.AI / 28 / 2608.13774

FLARE MCMC: Fidelity-based Layer-Adaptive REcursive proposals for MCMC

FLARE MCMC：基于保真度的层自适应递归提议的马尔可夫链蒙特卡洛方法

Venkatesan, Harini, Shelton, Christian, Ho, Ming-Feng, Bird, Simeon, Wu, Mengxuan

Abstract

Markov chain Monte Carlo (MCMC) requires only the ability to evaluate the likelihood, making it a common technique for inference in complex models. However, it can have a slow mixing rate, requiring the generation of many samples to obtain good estimates and an overall high computational cost. FLARE MCMC is a multi-fidelity layered MCMC method that exploits lower-fidelity approximations of the true likelihood calculation to improve mixing and leads to overall faster performance. Such lower-fidelity likelihoods are commonly available in scientific and engineering applications where the model involves a simulation whose resolution or accuracy can be tuned. Our technique uses recursive, layered chains with simple layer tuning; it does not require the likelihood to take any form or have any particular internal mathematical structure. We demonstrate experimentally that FLARE MCMC achieves larger effective sample sizes for the same computational time across different scientific domains including hydrology and cosmology.

Chinese Translation

马尔可夫链蒙特卡洛（MCMC）只需能够评估似然性，这使其成为复杂模型推断中的一种常用技术。然而，MCMC的混合速率可能较慢，通常需要生成大量样本以获得良好的估计，从而导致整体计算成本较高。FLARE MCMC是一种多保真度分层MCMC方法，它利用真实似然计算的低保真度近似来改善混合性，从而实现整体更快的性能。这种低保真度的似然在科学和工程应用中通常是可用的，尤其是在模型涉及可调分辨率或精度的模拟时。我们的方法使用递归的分层链和简单的层调节；它不要求似然具有任何特定形式或内部数学结构。我们通过实验表明，FLARE MCMC在不同科学领域（包括水文学和宇宙学）中，在相同计算时间内实现了更大的有效样本量。

View on arXiv Download PDF AI Translation

cs.AI / 29 / 2608.13787

From Passive Delegates to Strategic Negotiators: Reinforcing Social Reasoning in Small Language Models with SocialRL

从被动代理到战略谈判者：通过SocialRL增强小型语言模型的社会推理能力

Hua, Wenyue, Huang, Zachary, Payne, Tyler, Yousefi, Safoora, Amershi, Saleema, Celikyilmaz, Asli

Abstract

AI agents increasingly act on their users' behalf, handling tasks such as scheduling meetings, comparing offers, and haggling over prices. These principal-driven tasks routinely place the agent across from a counterpart (another user's agent, a seller, a recruiter) whose goals may conflict with its principal's. Yet the dispositions that make an assistant pleasant can make it a poor delegate: a friendly, helpful frontier model may disclose its principal's private information unprompted and concede at the first sign of resistance. We present SocialRL, a general recipe that trains social reasoning directly, and apply it to a 4B model across six domains: Deal-or-No-Deal, CaSiNo, Craigslist, Job Interview, Calendar, and Marketplace. Every domain is trained in-domain under the same recipe, and every policy is evaluated on all six. We find that (1) in-domain training reaches the frontier: on held-out scenarios the 4B matches or exceeds the GPT-5 family per domain, closing 73-122% of the baseline-to-frontier gap on the negotiation games, with 78% of buyer openings anchoring below target versus 3% untrained; (2) cross-domain transfer follows game structure: structurally paired games lift each other, a broad multi-issue donor lifts nearly all domains, and structurally isolated games transfer nothing; (3) guided by this transfer structure, two strategies, cascade RL and multi-teacher on-policy distillation (OPD), consolidate the per-domain specialists into a single unified 4B that reaches 0.627 average utility across all six environments, matching or exceeding GPT-4.1 (0.625), GPT-5.1 (0.619), and GPT-5.2 (0.613); (4) an explicit theory-of-mind scaffold helps only through training: distilling the ToM trace, rather than actions alone, lifts utility on every environment and generalizes better across them, and of the two ToM skills, only next-action prediction predicts negotiation outcomes.

Chinese Translation

人工智能代理越来越多地代表用户行事，处理诸如安排会议、比较报价和讨价还价等任务。这些以委托人为驱动的任务通常使代理与一个目标可能与其委托人相冲突的对手（另一个用户的代理、卖家、招聘人员）对峙。然而，使助手愉快的性格特征可能使其成为一个糟糕的代理：一个友好、乐于助人的前沿模型可能会在没有提示的情况下泄露其委托人的私人信息，并在遇到抵抗的第一迹象时就妥协。我们提出了SocialRL，这是一种直接训练社会推理的通用方法，并将其应用于一个4B模型，涵盖六个领域：Deal-or-No-Deal、CaSiNo、Craigslist、Job Interview、Calendar和Marketplace。每个领域都在相同的框架下进行领域内训练，并且每个策略在所有六个领域上进行评估。我们发现：(1) 领域内训练达到了前沿：在保留的场景中，4B在每个领域的表现与GPT-5系列相匹配或超越，谈判游戏的基线到前沿的差距缩小了73-122%，其中78%的买方开盘低于目标，而未训练的仅为3%；(2) 跨领域迁移遵循游戏结构：结构上配对的游戏相互提升，一个广泛的多议题捐赠者几乎提升了所有领域，而结构上孤立的游戏则没有任何迁移；(3) 在这种迁移结构的指导下，两种策略，级联强化学习（cascade RL）和多教师在线策略蒸馏（multi-teacher on-policy distillation, OPD），将每个领域的专家整合为一个统一的4B，在所有六个环境中达到0.627的平均效用，匹配或超越GPT-4.1（0.625）、GPT-5.1（0.619）和GPT-5.2（0.613）；(4) 明确的心智理论框架仅通过训练起作用：蒸馏心智理论痕迹，而不仅仅是动作，提升了每个环境的效用，并在这些环境中更好地泛化，而在两种心智理论技能中，只有下一步行动预测能够预测谈判结果。

View on arXiv Download PDF AI Translation

cs.AI / 30 / 2608.13820

SDO: Subspace Deconflicting Operator for Multi-Adapter Composition

SDO：用于多适配器组合的子空间去冲突操作符

Wang, Zhongsheng, Lin, Zhedong, Liu, Qian, Zhang, Xinyu, Liu, Jiamou

Abstract

Composing independently trained adapters within a shared diffusion backbone provides a modular approach to multi-character generation, but naive joint deployment often causes identity mixing, cross-character attribute leakage, and unstable scene composition. We study this interference from a parameter-space perspective and hypothesize that it arises partly from conflicts between overlapping dominant subspaces in shared layers. To address this issue, we propose \textbf{SDO}, a \textbf{S}ubspace \textbf{D}econflicting \textbf{O}perator for multi-adapter composition. SDO reconstructs layer-wise low-rank updates from the selected adapters, extracts compact subspace signatures, measures pairwise conflict through output-subspace overlap, and applies a permutation-equivariant transformation that suppresses harmful shared directions while retaining identity-specific characteristics. The resulting representations are mapped back to standard adapter updates and can be directly incorporated into existing diffusion inference pipelines. Experiments demonstrate that SDO consistently improves identity fidelity and compositional stability, with particularly clear gains as the number of jointly composed adapters increases.

Chinese Translation

在共享扩散主干中组合独立训练的适配器提供了一种模块化的多角色生成方法，但简单的联合部署往往会导致身份混合、跨角色属性泄漏和不稳定的场景组合。我们从参数空间的角度研究这种干扰，并假设它部分源于共享层中重叠主子空间之间的冲突。为了解决这个问题，我们提出了 extbf{SDO}，一种用于多适配器组合的 extbf{S}ubspace extbf{D}econflicting extbf{O}perator。SDO从选定的适配器中重建层级低秩更新，提取紧凑的子空间特征，通过输出子空间重叠测量成对冲突，并应用一种置换等变变换，抑制有害的共享方向，同时保留特定身份的特征。最终生成的表示被映射回标准适配器更新，并可以直接纳入现有的扩散推理管道。实验表明，SDO在身份保真度和组合稳定性方面始终有所提升，特别是在联合组合的适配器数量增加时，效果尤为明显。

View on arXiv Download PDF AI Translation

cs.AI / 31 / 2608.13863

Joint Optimization of Memory and Computing Frequency for Energy-Efficient DNN Inference

面向能效的深度神经网络推理的内存与计算频率联合优化

Han, Yunchu, Nan, Zhaojun, Zhou, Sheng, Niu, Zhisheng

Abstract

Deep neural network (DNN) inference on mobile devices often incurs high latency and energy consumption due to limited computing and memory resources. To enable energy-efficient DNN inference, most existing studies focus on dynamic voltage and frequency scaling (DVFS) for adjusting the computing frequency, while the impact of memory frequency on the inference performance has been greatly overlooked. In this paper, we consider the impact of memory frequency and computing frequency on DNN inference time, and jointly optimize these two frequencies together with communication resources for energy-efficient DNN inference. Based on a realistic inference time model, we formulate an optimization problem to minimize the energy consumption of all mobile devices under the deadline constraint. For local inference, we derive a near-optimal closed-form solution via convex optimization, while an optimal closed-form solution for transmission power is obtained for edge inference with the given bandwidth. Furthermore, we propose a low-complexity heuristic algorithm to effectively solve the overall problem with polynomial time complexity. Simulation results based on measured data show that the proposed near-optimal solution for local inference can achieve optimal performance under strict deadline constraints, with a performance gap of up to 2.5% compared with the optimal solution. Meanwhile, our proposed algorithm significantly reduces the energy consumption of devices by up to 10.4% compared to other methods.

Chinese Translation

移动设备上的深度神经网络（DNN）推理由于计算和内存资源的限制，通常会导致高延迟和能耗。为了实现能效的DNN推理，大多数现有研究集中在动态电压与频率调整（DVFS）上，以调整计算频率，而内存频率对推理性能的影响则被大大忽视。本文考虑了内存频率和计算频率对DNN推理时间的影响，并联合优化这两种频率及通信资源，以实现能效的DNN推理。基于一个现实的推理时间模型，我们制定了一个优化问题，以在截止时间约束下最小化所有移动设备的能耗。对于本地推理，我们通过凸优化推导出一个近似最优的闭式解，而对于边缘推理，在给定带宽的情况下获得了传输功率的最优闭式解。此外，我们提出了一种低复杂度的启发式算法，以有效解决整体问题，其多项式时间复杂度。基于测量数据的仿真结果表明，所提出的本地推理近似最优解在严格的截止时间约束下能够实现最佳性能，与最优解相比，性能差距可达2.5%。同时，我们提出的算法相比其他方法显著降低了设备的能耗，最高可达10.4%。

View on arXiv Download PDF AI Translation

cs.AI / 32 / 2608.13883

MemoryLake on MemoryArena: A Matched Study of Agent Memory Backends

MemoryLake 在 MemoryArena 上：代理记忆后端的匹配研究

Zhan, Chaoqun, Zhou, Qiang, Li, Guannan, Huang, Zhenqiang, Wang, Qianjin

Abstract

Most agent-memory benchmarks test post-hoc recall, whereas MemoryArena evaluates whether memory supports interdependent, multi-session task completion. We compare MemoryLake, a structured multi-track memory backend, with Mem0, text-embedding-3-small vector RAG, and a long-context control across all five MemoryArena domains. The systems share the same agent framework, requested gpt-5-mini model alias, task samples, and scoring code; the memory integration is the intentionally changed component. Because each backend bundles write, retrieval, consolidation, budgeting, and prompt-assembly choices, the study is a matched system-level comparison, not a representation-only ablation or a cost-matched experiment. On the shared evaluation sets, MemoryLake has the highest observed success rate (SR) in mathematics (9/40), physics (12/20), and progressive retrieval (4/20). Every system has zero SR in travel planning, and web shopping yields a single bundle-level success (long context, 1/150); MemoryLake ranks third on both the travel soft process score and shopping step match. Following MemoryArena's suite-level convention, a post-hoc equal-weight average over the five SRs is 20.5% for MemoryLake versus 13.6% for the best comparator. These are point estimates: sample sizes are modest, confidence intervals overlap, and we do not report paired significance tests. A separate MemoryLake-only run over all 221 progressive queries yields a failure-counted SR of 26.7% (59/221) and is not a baseline comparison. The results support a workload-dependent view of memory backends and an observed lead among the four evaluated systems on the shared sets; they do not establish benchmark-wide state of the art or a causal advantage of representation structure.

Chinese Translation

大多数代理记忆基准测试侧重于事后回忆，而 MemoryArena 评估记忆是否支持相互依赖的多会话任务完成。我们将 MemoryLake（一个结构化的多轨记忆后端）与 Mem0（文本嵌入-3-小型向量 RAG）和一个长上下文控制进行比较，涵盖所有五个 MemoryArena 领域。这些系统共享相同的代理框架、请求的 gpt-5-mini 模型别名、任务样本和评分代码；记忆集成是故意更改的组件。由于每个后端都捆绑了写入、检索、整合、预算和提示组装选项，因此该研究是一个匹配的系统级比较，而不是仅仅代表性的消融或成本匹配实验。在共享评估集上，MemoryLake 在数学（9/40）、物理（12/20）和渐进检索（4/20）中观察到的成功率（SR）最高。每个系统在旅行规划中均无成功率，而网络购物仅产生一个捆绑级成功（长上下文，1/150）；MemoryLake 在旅行软过程得分和购物步骤匹配中均排名第三。根据 MemoryArena 的套件级惯例，五个成功率的事后等权重平均值为 MemoryLake 的 20.5% 对比最佳比较者的 13.6%。这些都是点估计：样本量适中，置信区间重叠，我们不报告配对显著性测试。对所有 221 个渐进查询进行的单独 MemoryLake 运行显示失败计数的成功率为 26.7%（59/221），并不是基线比较。结果支持记忆后端的工作负载依赖视角，并在共享集上观察到四个评估系统之间的领先地位；它们并未确立基准范围内的最新技术或表示结构的因果优势。

View on arXiv Download PDF AI Translation

cs.AI / 33 / 2608.13921

When Personal Memory Has No Single Answer: Evaluating LLM Agents under Irreducible Conflict

当个人记忆没有单一答案时：在不可简化冲突下评估大型语言模型代理

Yang, Lu, Xu, Shusheng, Li, Zhuoran, Yang, Tongkai, Huang, Longbo

Abstract

LLM agents increasingly maintain personal memory across sessions, but it can conflict. Preferences depend on context, behavior evolves, and sources can conflict. When a query lacks context, time, or source authority to interpret conflict, treating one memory as definitive converts unresolved conflict into an unjustified, overconfident action. Existing benchmarks recover one answer from conflicting evidence, overlooking whether agents recognize underdetermination, preserve alternatives, seek missing information, and choose appropriate actions. We introduce \underline{T}esting \underline{A}gents' \underline{N}avigation of \underline{G}enuine, \underline{L}atent, and \underline{E}ntangled Memory Conflicts (\textsc{TANGLE}), a benchmark for genuinely unresolvable memory conflicts. It comprises 541 instances across 40 personas and three types: Context-Partitioned Conflict (CPC), Behavior-Oscillation Conflict (BOC), and Source-Contradiction Conflict (SCC). We evaluate two tracks---an oracle track with curated memory and a pipeline track that extracts memory from multi-session dialogues---on five dimensions: conflict perception, causal reasoning, confidence calibration, clarification seeking, and memory faithfulness. Experiments reveal pipeline challenges. With curated memory, models recognize conflicts more reliably than they calibrate actions or seek targeted clarification. With end-to-end pipeline memory, extraction fails to preserve conflict-bearing relations needed for downstream reasoning. Policy comparisons show fixed rules are insufficient when actions must reflect conflict. These findings motivate Conflict-Aware Action Policy (CAAP), which adapts actions to each conflict using available evidence. \textsc{TANGLE} frames conflict handling as recognizing underdetermination, retaining conflicting evidence, and acting without forcing a definitive answer.

Chinese Translation

大型语言模型（LLM）代理越来越多地在会话中维持个人记忆，但这些记忆可能会发生冲突。偏好依赖于上下文，行为会演变，信息来源可能存在矛盾。当查询缺乏上下文、时间或来源权威来解释冲突时，将某一记忆视为决定性答案会将未解决的冲突转化为不合理的过度自信行为。现有基准测试从冲突证据中恢复单一答案，却忽视了代理是否识别不确定性、保留替代方案、寻求缺失信息以及选择适当的行动。我们引入了 extsc{TANGLE}（测试代理在真实、潜在和纠缠记忆冲突中的导航），这是一个针对真正无法解决的记忆冲突的基准测试。它包含541个实例，涵盖40个角色和三种类型：上下文分割冲突（CPC）、行为振荡冲突（BOC）和来源矛盾冲突（SCC）。我们在五个维度上评估两个轨道——一个具有策划记忆的神谕轨道和一个从多会话对话中提取记忆的管道轨道：冲突感知、因果推理、自信度校准、澄清寻求和记忆忠实度。实验揭示了管道的挑战。在策划记忆的情况下，模型比校准行动或寻求针对性澄清更可靠地识别冲突。在端到端管道记忆中，提取未能保留下游推理所需的冲突关系。政策比较表明，当行动必须反映冲突时，固定规则是不够的。这些发现促使了冲突意识行动政策（CAAP）的提出，该政策根据可用证据调整每个冲突的行动。 extsc{TANGLE}将冲突处理框架化为识别不确定性、保留冲突证据以及在不强迫给出决定性答案的情况下采取行动。

View on arXiv Download PDF AI Translation

cs.AI / 34 / 2608.13926

Never the Number: Structural Abstention for AI Systems Whose Answers Are Consumed as Fact

从不只是数字：针对将答案视为事实的人工智能系统的结构性弃权

Zhelun, Wu

Abstract

Large language models have made natural language interfaces to databases (NLIDB) newly credible, but LLM text-to-SQL systems fail in a way that matters for deployment: a hallucinated column or a mis-aggregated total yields a fluent wrong answer, indistinguishable at the point of use from a right one. Where the consumer cannot inspect the generated query, as in enterprise AI deployments and operational dashboards, and increasingly where the consumer is a tool-using agent rather than a person, accuracy alone is insufficient: nothing marks which answers to distrust. This is a reliability problem before it is an accuracy problem. We propose an architectural pattern for such systems, a trusted kernel with a generative shell, resting on one invariant: a component that can fabricate may influence which question the system answers, never which value it returns. A generative shell interprets underspecified input and phrases replies; a deterministic kernel matches fully specified questions against a bounded set of answerable question shapes and compiles them to queries by deterministic execution. The two meet at a confirmation the user reads before any value is computed, and requests the kernel cannot express are declined rather than approximated. We call this structural abstention, and distinguish it from the statistical abstention of selective prediction and calibrated confidence: refusal here needs no confidence estimate, because unanswerable requests are unrepresentable. We specify the pattern implementation-independently, give a five-decision recipe and work it across three domains, extend the invariant from returned values to the actions of agentic systems, and report a two-year production case study alongside two generative alternatives, a fine-tuned parser and a tool-retrieval agent. We close against enterprise and reliability benchmarks published since.

Chinese Translation

大型语言模型使得数据库的自然语言接口（NLIDB）重新获得可信度，但LLM文本到SQL系统在部署时存在一个重要的失败：虚构的列或错误聚合的总数会产生流畅的错误答案，在使用时与正确答案难以区分。当消费者无法检查生成的查询时，例如在企业AI部署和操作仪表板中，且越来越多的情况下，消费者是使用工具的代理而非个人，仅仅依赖准确性是不够的：没有任何标记可以指示哪些答案不可信。这是一个可靠性问题，而不是准确性问题。我们为此类系统提出了一种架构模式，即一个具有生成外壳的可信内核，基于一个不变性原则：一个能够生成内容的组件可能会影响系统回答的问题，而绝不会影响返回的值。生成外壳解释不明确的输入并表述回复；确定性内核将完全指定的问题与有限的可回答问题形式进行匹配，并通过确定性执行将其编译为查询。这两者在用户在计算任何值之前阅读的确认处相遇，而内核无法表达的请求则被拒绝而不是近似。我们称之为结构性弃权，并将其与选择性预测和校准置信度的统计性弃权区分开来：此处的拒绝不需要置信度估计，因为无法回答的请求是不可表示的。我们独立于实现规范该模式，提供了一个五个决策的配方，并在三个领域中应用，扩展了从返回值到代理系统行为的不变性，并报告了一个为期两年的生产案例研究，以及两个生成替代方案，一个经过微调的解析器和一个工具检索代理。我们在企业和可靠性基准测试中进行了总结。

View on arXiv Download PDF AI Translation

cs.AI / 35 / 2608.13940

AI Research Preference Models

人工智能研究偏好模型

Foster, Thomas Simon, Omari, Bassel Al, Fu, Tingchen, Mann, Thomas, Domond, Carl, Cipolina-Kun, Lucia, Gauri, Bhavul, Aghamelu, Muna, Goldie, Alexander D., Helenowski, Eryk, Gagnon-Audet, Jean-Christophe, Pepe, Alberto, Nazir, Saba, Izcovich, Daniel, Levi, Noam, Hazra, Rishi, Hambardzumyan, Karen, Baldwin, Nicolas, Li, Xian, Josifoski, Martin, Giampouras, Paris, Sabet, Masoud Jalili, Sims, Anya, Momand, Hela, Shavrina, Tatiana, Magka, Despoina, Weston, Jason, Wang, Yulin, Goyal, Anirudh, Henriques, João, Bachrach, Yoram, McMilin, Emily, Foerster, Jakob Nicolaus

Abstract

AI research agents (AIRA) can now propose, implement, and evaluate their own machine learning experiments, but progress on frontier tasks is throttled by cost: a candidate solution can be written in minutes, whereas evaluating it can take hours to days of GPU time. An agent can therefore propose far more candidates than it can afford to run, and its progress depends on its research preference: how it allocates a fixed execution budget across many candidates. We introduce AI Research Preference Models (RPMs) that predict which of multiple candidate solutions are most worth executing, without paying the cost of executing them all. We build RPMs from frozen pretrained language models (with no task-specific training), in two forms: an inference-only model that reasons over candidate plans, code, and prior executed solutions, and an agentic model that additionally runs small-scale pilot experiments before deciding. We integrate both into the AIRA-dojo search agent and evaluate on AIRS-Bench, a recent benchmark of machine learning research tasks for AI research agents. The two variants raise the average normalized score from 0.684 to 0.711 and 0.729 respectively, and reach the unguided agent's 24-hour performance in roughly 15 hours, using less than two-thirds of its execution budget. Our best RPMs also yield new state-of-the-art results on two AIRS-Bench tasks.

Chinese Translation

人工智能研究代理（AIRA）现在可以提出、实施和评估自己的机器学习实验，但在前沿任务上的进展受到成本的限制：一个候选解决方案可以在几分钟内编写，而评估它可能需要数小时到数天的GPU时间。因此，一个代理可以提出远多于其能够运行的候选方案，其进展取决于其研究偏好：如何在多个候选方案之间分配固定的执行预算。我们引入了人工智能研究偏好模型（RPMs），它可以预测多个候选解决方案中哪些最值得执行，而无需支付执行所有方案的成本。我们从冻结的预训练语言模型（没有特定任务的训练）构建RPMs，分为两种形式：一种是仅推理模型，能够对候选计划、代码和先前执行的解决方案进行推理；另一种是代理模型，此外在决定之前还会运行小规模的试点实验。我们将两者整合到AIRA-dojo搜索代理中，并在AIRS-Bench上进行评估，这是一个针对人工智能研究代理的机器学习研究任务的最新基准。这两种变体将平均归一化得分从0.684提高到0.711和0.729，并在大约15小时内达到无指导代理的24小时性能，使用的执行预算不到其三分之二。我们的最佳RPMs在两个AIRS-Bench任务上也取得了新的最先进结果。

View on arXiv Download PDF AI Translation

cs.AI / 36 / 2608.13951

HELIX: Model-Harness Co-evolution for Recursive Self-Improvement

HELIX：模型-工具共同进化以实现递归自我改进

Fan, Tianyu, Huang, Chao

Abstract

Scaling agent capability has largely focused on improving the model, yet an interactive agent acts through a runtime harness that mediates context, tools, control flow, and stopping. The harness shapes both what a model can accomplish and the trajectories from which it learns. This coupling motivates model-harness co-evolution for recursive self-improvement: build harnesses for a fixed model, update the model from verified sibling trajectories, and rebuild the harnesses as model capabilities change. Realizing this loop requires a controlled way to evolve harnesses while preserving intervention identity and effect. We present HELIX, a source-traceable substrate for harness evolution. HELIX decomposes agent systems into typed ports, reusable atoms, recipes, product shells, and runtime policies. It makes interventions explicit and auditable while retaining trajectories, test outcomes, and provenance. Harness evolution thus serves two linked roles: improving fixed-model execution and producing matched successes, regressions, near misses, and alternative solutions as data for subsequent model improvement. We evaluate HELIX in one evolution round on code repair. A 65-candidate portfolio discovers a fixed harness that improves task coverage by 4.0% over Pi, while the full portfolio exposes up to 58.0% more verified coverage through complementary sibling behavior. Selected candidates are assessed with repeated runs and the SWE-bench evaluator. A 200-slot sibling slice yields 438 verified SFT, critic, filter, and preference records. These results show how harness, model, and data form a feedback system: harness evolution expands current capability and creates learning signal for the next model; model updates motivate the next round of harness evolution. HELIX provides an auditable interface for studying this recursive process. Code is available at https://github.com/HKUDS/HELIX.

Chinese Translation

提升智能体能力的研究主要集中在改进模型上，然而交互智能体通过运行时工具进行操作，该工具调节上下文、工具、控制流和停止。工具不仅影响模型能够完成的任务，还影响模型学习的轨迹。这种耦合促使模型-工具共同进化以实现递归自我改进：为固定模型构建工具，从经过验证的同类轨迹中更新模型，并在模型能力变化时重建工具。实现这一循环需要一种受控的方法来进化工具，同时保持干预的身份和效果。我们提出了HELIX，一个可追溯源的工具进化基础。HELIX将智能体系统分解为类型化端口、可重用原子、配方、产品外壳和运行时策略。它使干预变得明确且可审计，同时保留轨迹、测试结果和来源。因此，工具进化承担了两个相互关联的角色：改善固定模型的执行，并生成匹配的成功、回归、接近失败和替代解决方案的数据，以便后续模型改进。我们在代码修复的一个进化轮次中评估了HELIX。一个包含65个候选者的组合发现了一个固定工具，该工具在任务覆盖率上比Pi提高了4.0%，而完整组合通过互补的同类行为揭示了多达58.0%的额外验证覆盖率。选定的候选者通过重复运行和SWE-bench评估器进行评估。一个包含200个插槽的同类切片产生了438个经过验证的SFT、评论、过滤器和偏好记录。这些结果展示了工具、模型和数据如何形成一个反馈系统：工具进化扩展当前能力，并为下一个模型创造学习信号；模型更新则激励下一轮工具进化。HELIX提供了一个可审计的接口，用于研究这一递归过程。代码可在 https://github.com/HKUDS/HELIX 获取。

View on arXiv Download PDF AI Translation

cs.AI / 37 / 2608.13958

Implementing Computational Law in Wolfram Language for the Governance of Artificial Intelligence

在沃尔夫拉姆语言中实施计算法以治理人工智能

Wiles, James K.

Abstract

How do we govern AI systems whose reasoning we cannot fully inspect? Governance does not require understanding a system's reasoning. It requires stating what the system is obliged, permitted, and forbidden to do, and checking whether it complied. I present an implementation of Reified Input/Output Logic, the formalism behind the DAPRECO knowledge base, in Wolfram Language: the core I/O axioms, obligations, permissions, constitutive norms, reified eventualities, and temporal operators. I then test whether GPT-4 can translate English legal statements into the formalism, and report the failures: hallucinated functions, omitted temporal scope, deviation from the formalism, and (in the worst cases) code that runs, reads plausibly, but silently encodes the wrong norm. A case study, an AI guard dog operating under a computational contract, shows how formalized rules can extend from a contract directly into the operational code of an embodied agent, producing symbolic, auditable justifications for its behaviour. I argue that computational law can be used as a governance tool and that a desirable goal would be to formalize the law that can and ought to be programmatically executable.

Chinese Translation

我们如何治理那些我们无法完全检查其推理的人工智能系统？治理并不要求理解系统的推理。它要求明确系统的义务、允许和禁止的行为，并检查其是否遵守。我在沃尔夫拉姆语言中展示了具象输入/输出逻辑（Reified Input/Output Logic）的实现，这是DAPRECO知识库背后的形式化工具：核心的输入/输出公理、义务、权限、构成规范、具象事件和时间运算符。然后，我测试了GPT-4是否能够将英文法律陈述翻译成该形式化工具，并报告了失败的情况：虚构的函数、遗漏的时间范围、偏离形式化工具，以及（在最坏的情况下）运行正常、看似合理但默默编码错误规范的代码。一个案例研究，即在计算合同下运作的人工智能看门狗，展示了如何将形式化规则直接扩展到具身代理的操作代码中，从而为其行为生成符号化的、可审计的理由。我认为计算法可以作为治理工具，并且一个理想的目标是形式化那些可以且应该以程序化方式执行的法律。

View on arXiv Download PDF AI Translation

cs.AI / 38 / 2608.13987

Nanbeige4.2-3B on Apple Silicon: Fixing Deployment Bugs and Decreasing Looped Transformer Memory Overhead

Nanbeige4.2-3B在苹果硅上的应用：修复部署错误并减少循环变换器的内存开销

Halloran, John T.

Abstract

Nanbeige4.2-3B is a 3B-parameter agentic model built around a Looped Transformer (LT) that reuses one stack of layers for a second forward pass, adding effective depth without additional parameters. Evaluated on Apple Silicon (MPS), we identify five independent bugs which prevent the released checkpoint from running via Hugging Face transformers out of the box (including a silently-zeroed RoPE buffer and calls to removed transformers cache APIs). Furthermore, we show that fixing these bugs is still not sufficient for agentic tasks, due to the LT's layer-reuse strategy (which effectively doubles peak attention memory) used to achieve parameter efficiency. We thus introduce a chunked-prefill strategy which alleviates the incurred memory-capacity penalty, extending allowable context width by $2.7 \times$ on 32~GiB shared memory. However, even with the reduced memory overhead, we show that patches are required to render Nanbeige4.2-3B usable; resolving both system prompt and MPS-native memory bugs finally allows reliable evaluation on standard MCP and tool-calling benchmarks. On a subset of MCPMark, the debugged model completes up to 30\% of real agentic tasks (up from the original's 0\%), while, on BFCL, it is near-perfect at single tool calls (yet fails the majority of multi-tool tests). We release the patched checkpoint, system prompt optimizer, and evaluation harnesses at https://github.com/johnhalloran321/Nanbeige4.2-3B-mps-fix.

Chinese Translation

Nanbeige4.2-3B是一个拥有30亿参数的智能模型，基于循环变换器（Looped Transformer, LT）构建，该模型在第二次前向传播中重用一组层，从而在不增加额外参数的情况下增加有效深度。在苹果硅（MPS）上进行评估时，我们发现了五个独立的错误，这些错误阻止了发布的检查点通过Hugging Face变换器直接运行（包括一个静默归零的RoPE缓冲区和对已移除的变换器缓存API的调用）。此外，我们展示了修复这些错误仍然不足以满足智能任务的需求，因为LT的层重用策略（有效地使峰值注意力内存翻倍）用于实现参数效率。因此，我们引入了一种分块预填充策略，缓解了所产生的内存容量惩罚，在32 GiB共享内存上将允许的上下文宽度扩展了2.7倍。然而，即使在减少内存开销的情况下，我们仍然表明需要补丁才能使Nanbeige4.2-3B可用；解决系统提示和MPS原生内存错误最终使得在标准MCP和工具调用基准上的可靠评估成为可能。在MCPMark的一个子集中，调试后的模型完成了高达30%的真实智能任务（相比于原始模型的0%），而在BFCL上，它在单个工具调用时几乎完美，但在大多数多工具测试中失败。我们在https://github.com/johnhalloran321/Nanbeige4.2-3B-mps-fix发布了修补后的检查点、系统提示优化器和评估工具。

View on arXiv Download PDF AI Translation

cs.AI / 39 / 2608.13990

Content Depth Matters in Short-Video Recommendation: Rethinking the Attention Economy

短视频推荐中的内容深度至关重要：重新思考注意力经济

Deng, Liwei, Jiang, Jing, Li, Zhiwei, Wang, Yang, Long, Guodong

Abstract

Driven by the attention economy, short-video Recommender Systems (RSs) are primarily optimized to maximize user engagement by promoting videos that capture attention within seconds. These systems inherently favor shallow-content videos that are effective at attracting immediate attention. However, growing evidence suggests that prolonged exposure to such content may negatively affect users' cognitive engagement and mental well-being, raising concerns about the long-term societal impact of the short-video platform. To tackle this challenge, this paper introduces a new metric, the \textbf{Content Depth Score (CDS)}, to quantify the content depth of short videos. CDS measures the extent to which a video is expected to stimulate higher-order cognitive processes, using a seven-level scale grounded in established theories of cognitive psychology and learning. As an initial step toward this vision, we present \textbf{SCOPE-Bench}, the first benchmark for content-depth evaluation in short-video recommendation. Built upon a large-scale open-source short-video dataset, SCOPE-Bench provides CDS annotations for 150K videos, enabling systematic evaluation of RSs from a cognitive-content perspective. Leveraging SCOPE-Bench, we evaluate 13 representative RSs and reveal a consistent preference for shallow-content videos. Moreover, we find that these algorithms recommending cognitively deep content are only marginally better than random selection, highlighting a previously overlooked limitation of existing recommendation objectives. Our code and datasets are available at https://liweidengdavid.github.io/SCOPE-Bench/.

Chinese Translation

在注意力经济的驱动下，短视频推荐系统（Recommender Systems, RSs）主要优化以最大化用户参与度，通过推广能够在几秒钟内吸引注意力的视频。这些系统本质上偏向于能够有效吸引即时注意力的浅层内容视频。然而，越来越多的证据表明，长时间接触此类内容可能会对用户的认知参与度和心理健康产生负面影响，引发了对短视频平台长期社会影响的担忧。为了解决这一挑战，本文引入了一种新的指标——内容深度评分（Content Depth Score, CDS），用于量化短视频的内容深度。CDS衡量视频刺激高阶认知过程的预期程度，采用基于认知心理学和学习理论的七级量表。作为实现这一愿景的初步步骤，我们提出了SCOPE-Bench，这是短视频推荐中内容深度评估的首个基准。SCOPE-Bench基于一个大规模的开源短视频数据集，为150K个视频提供CDS注释，使得从认知内容的角度对RS进行系统评估成为可能。利用SCOPE-Bench，我们评估了13个具有代表性的RS，并揭示出它们对浅层内容视频的一致偏好。此外，我们发现这些推荐认知深度内容的算法仅比随机选择稍微好一些，突显了现有推荐目标中一个被忽视的局限性。我们的代码和数据集可在 https://liweidengdavid.github.io/SCOPE-Bench/ 获取。

View on arXiv Download PDF AI Translation

cs.AI / 40 / 2608.13993

Simulation-Driven Vehicular Traffic Data Augmentation: Extending Sensor Coverage Through Virtual Sensing

基于仿真的车辆交通数据增强：通过虚拟传感器扩展传感器覆盖范围

Guastella, Davide Andrea, Porras, Eladio Montero, Pournaras, Evangelos, Bontempi, Gianluca

Abstract

Urban traffic management relies on sensor networks whose spatial coverage is limited by deployment costs and privacy regulations. Machine learning models trained on such sparse data cannot generalize to unmonitored locations and must be retrained whenever the sensor infrastructure changes. We propose a simulation-based methodology that addresses this problem by generating augmented traffic count datasets in which each physical sensor is replaced by a virtual sensor placed at a surrogate location in the road network. Virtual sensors are selected by a graph-search heuristic that jointly maximises vehicle-flow continuity and traffic-metric similarity between the original and surrogate locations, while enforcing a minimum spatial displacement to ensure diversity of observed traffic conditions. We validate the method on two Belgian cities: Brussels, using a calibrated model, and Namur, using synthetic models. The augmented datasets preserve the bimodal daily demand profile and the dynamics of traffic at the observed locations.

Chinese Translation

城市交通管理依赖于传感器网络，但其空间覆盖受到部署成本和隐私法规的限制。基于稀疏数据训练的机器学习模型无法推广到未监测的位置，并且每当传感器基础设施发生变化时必须重新训练。我们提出了一种基于仿真的方法，通过生成增强的交通计数数据集来解决这一问题，其中每个物理传感器被替换为在道路网络中放置的虚拟传感器。虚拟传感器的选择采用图搜索启发式算法，旨在最大化车辆流动的连续性和原始位置与替代位置之间的交通指标相似性，同时强制执行最小空间位移以确保观察到的交通条件的多样性。我们在比利时的两个城市进行验证：布鲁塞尔（Brussels），使用经过校准的模型，以及纳慕尔（Namur），使用合成模型。增强的数据集保留了双峰的日需求曲线和观察位置的交通动态。

View on arXiv Download PDF AI Translation

cs.AI / 41 / 2608.14014

Buy the Rumor, Sell the News: When Is News Priced In?

买谣言，卖新闻：新闻何时被定价？

Kargarzadeh, Alireza, Khaledian, Nariman, Parvini, Navid, Ghatak, Sid, Khaledian, Arman

Abstract

Two old market sayings hold that news is already priced in by the time it is published, and that the rumor is bought while the news is sold. Both place the price move associated with a piece of news before and at publication rather than after it. Whether the claims hold, for which kinds of news, and by how much are basic questions about how fast markets absorb public information. We test them on 4.57 million financial news articles covering roughly 3,000 US stocks (2023-2026). A large language model teacher, distilled into a compact classifier through active learning, assigns each article one of 17 event tags and five attributes; articles are clustered into stories to separate first reports from follow-up coverage; and beta-adjusted abnormal returns are measured around the resulting 1.68 million stock-day events, with 364,405 neutral-sentiment events as a placebo group. Three results follow. First, the price move associated with news concentrates before and at publication: pooled across all signed events, the cumulative move in the news direction by the close of publication day is 2.8 times its value 20 days later, and for rumor-flagged events the rumor day captures the entire move while the subsequent confirmation contributes nothing. Second, measured against the placebo of comparable stocks, markets underreact to numbers and overreact to stories: quantified fundamental news (earnings, dividends, guidance, analyst actions) keeps drifting in the direction of the news for weeks, while soft story-driven news (launches, macro commentary, leadership) gives back its move. Third, news carries width as well as direction: publicity raises volatility before the publication day, and volatility declines once the news is out, because publication resolves uncertainty. The study also produces a table of measured drift for each event tag, usable as a prior in news-conditioned forecasting models.

Chinese Translation

两句古老的市场谚语认为，新闻在发布时已经被定价，而谣言则是在新闻发布前被买入。这两者都将与新闻相关的价格变动置于发布之前和发布时，而非之后。关于这些说法是否成立、适用于何种类型的新闻以及其影响程度是市场吸收公共信息速度的基本问题。我们对覆盖大约3000只美国股票（2023-2026年）的457万篇金融新闻文章进行了测试。一种大型语言模型教师通过主动学习被提炼为一个紧凑的分类器，为每篇文章分配17个事件标签和五个属性；文章被聚类为故事，以区分首次报道与后续报道；并测量围绕结果产生的168万股日事件的贝塔调整异常收益，364,405个中性情绪事件作为对照组。研究得出了三个结果。首先，与新闻相关的价格变动集中在发布前和发布时：在所有已签名事件中，发布日收盘时的累积变动是20天后其值的2.8倍，而对于标记为谣言的事件，谣言日捕捉了整个变动，而随后的确认没有贡献任何东西。其次，与可比股票的对照组相比，市场对数字反应不足，对故事反应过度：量化的基本新闻（收益、股息、指导、分析师行动）在数周内持续朝着新闻方向漂移，而软性故事驱动的新闻（发布、宏观评论、领导层变动）则回吐其变动。第三，新闻不仅具有方向性，还具有宽度：在发布日前，宣传提高了波动性，而一旦新闻发布，波动性便下降，因为发布消除了不确定性。该研究还生成了每个事件标签的漂移测量表，可用作新闻条件预测模型中的先验。

View on arXiv Download PDF AI Translation

cs.AI / 42 / 2608.14021

Residual Dominance as a Structural Account of Last-Item Reliance in Causal Self-Attention Recommenders

残差主导性作为因果自注意推荐系统中最后一项依赖的结构性解释

Kozaki, Keito, Sakurai, Keigo, Togo, Ren, Ogawa, Takahiro, Haseyama, Miki

Abstract

Transformer-based sequential recommenders with causal self-attention often rely heavily on the most recent interaction at inference time, but how this behavior is structurally expressed in the representation used for prediction remains unclear. We combine prediction-time diagnostics with norm-based analysis of the full attention block. First, we show that SASRec-style models exhibit highly localized last-item reliance. We then find that, although self-attention aggregates contextual information, residual addition sharply shifts the full-block representation toward same-position contributions, which we term residual dominance. To probe this interpretation, we use inference-time residual scaling as a controlled diagnostic intervention. Changing the residual strength induces a monotonic trade-off between structural mixing and last-item reliance, while reducing residual strength recovers a subset of final-position misses for which representations at non-final positions already rank the ground-truth item correctly. Our results provide a structural account linking extreme last-item reliance to residual dominance at inference time. The code is publicly available.

Chinese Translation

基于Transformer的因果自注意序列推荐系统在推理时通常高度依赖最近的交互，但这种行为在用于预测的表示中是如何结构性表达的仍不清楚。我们结合了预测时的诊断与基于范数的全注意力块分析。首先，我们展示了SASRec风格模型表现出高度局部化的最后一项依赖。然后，我们发现，尽管自注意力聚合了上下文信息，但残差加法显著将全块表示向同位置贡献倾斜，我们称之为残差主导性。为了探讨这一解释，我们使用推理时的残差缩放作为一种受控的诊断干预。改变残差强度会在结构混合与最后一项依赖之间引发单调的权衡，而降低残差强度则恢复了一部分最终位置的遗漏，这些位置在非最终位置的表示中已经正确地对真实项进行了排名。我们的结果提供了一种结构性解释，将极端的最后一项依赖与推理时的残差主导性联系起来。代码已公开发布。

View on arXiv Download PDF AI Translation

cs.AI / 43 / 2608.14035

Agent-Orchestration in Autonomous Chip Design

自主芯片设计中的代理协调

Li, Linyang

Abstract

Recent developments in large language models (LLMs) and tool-using agents encourage people to explore the potential of using agents in chip design. The core question is what kind of AI we really need in such a sophisticated industry. To this end, we bring the idea of modeling a chip-design superintelligence as an enormous \textit{AI-organization}.

Chinese Translation

近期大型语言模型（LLMs）和工具使用代理的发展促使人们探索在芯片设计中使用代理的潜力。核心问题是，在如此复杂的行业中，我们真正需要什么样的人工智能。为此，我们提出将芯片设计的超级智能建模为一个庞大的 extit{AI-organization}（人工智能组织）的概念。

View on arXiv Download PDF AI Translation

cs.AI / 44 / 2608.14036

Demystifying Agent Skills: Why They Work-Until They Don't

揭示智能体技能的奥秘：为何有效——直到无效

Jiang, Zhiyuan, Huang, Fangrui, Xing, Hanwen, Wu, Xander, Gao, Yipeng, Cao, Rui, Wang, Mengdi, Liu, Shilong, Li, Yijiang

Abstract

Skills have emerged as a practical and effective approach for enhancing LLM agents at inference time through structured packages of knowledge. However, existing evaluations largely measure whether skills improve aggregated task success, leaving a more fundamental question underexplored: \emph{\textbf{When do skills help, why do they work, and where do they fail?}} Through controlled experiments across various benchmarks, agent harnesses and LLMs, we isolate the effects of representation, outcome annotation, retrieval difficulty, and cross-framework robustness of skills. To further answer this question, we design a contrastive study that combines controlled quantitative experiments with paired trajectory analysis. We normalize 8,135 trial records from controlled experiments and retain 238 valid unique labels from 240 open-coded records. We consolidate these observations into a taxonomy of three high-level categories and twelve skill-use modes: skills work when noisy trajectories become procedural anchors that stabilize execution. Skills improve over Workflow Memory by 6.06 points in matched comparisons. Procedural anchoring accounts for 65.7\% of skill cases, versus 4.5\% for explicit knowledge injection, showing that skills stabilize action rather than inject missing facts. Retrieval is a separate bottleneck: as pools grow from 5 to 100, actual-use precision falls from 29.6\% to 3.3\%. Confusable distractors impair offline identification, yet downstream success remains stable; exact ground-truth invocation is neither sufficient nor necessary. Skills fail under brittle assumptions, incompatible contexts, or insufficient adaptation. These findings move evaluation beyond aggregate success rates and guide reliable self-evolving agents.

Chinese Translation

技能已成为在推理时通过结构化知识包增强大型语言模型（LLM）智能体的一种实用有效的方法。然而，现有评估主要衡量技能是否提高了任务的整体成功率，留下了一个更为根本的问题未被深入探讨： extbf{技能在何时有效，为什么有效，以及在何处失效？} 通过在各种基准、智能体利用和LLM上的受控实验，我们隔离了表现、结果注释、检索难度和技能的跨框架稳健性等因素的影响。为了进一步回答这个问题，我们设计了一项对比研究，结合了受控定量实验和配对轨迹分析。我们对受控实验中的8,135条试验记录进行了标准化，并从240条开放编码记录中保留了238个有效的独特标签。我们将这些观察结果整合为三个高层次类别和十二种技能使用模式：当噪声轨迹成为稳定执行的程序锚时，技能便会有效。在匹配比较中，技能相较于工作流记忆提高了6.06分。程序锚定占技能案例的65.7\%，而显式知识注入仅占4.5\,这表明技能是通过稳定行动而非注入缺失事实来发挥作用。检索是一个独立的瓶颈：当池从5个增长到100个时，实际使用的精度从29.6\%下降到3.3\%。混淆干扰项妨碍了离线识别，但下游成功率保持稳定；确切的真实调用既不是充分条件也不是必要条件。技能在脆弱假设、不兼容的上下文或适应不足的情况下失效。这些发现将评估推向超越整体成功率，并为可靠的自我进化智能体提供指导。

View on arXiv Download PDF AI Translation

cs.AI / 45 / 2608.14063

Benchmarking data-driven material models on the classic Treloar dataset

基于经典Treloar数据集的数据驱动材料模型基准测试

Holthusen, Hagen, Flaschel, Moritz, Martonová, Denisa, Kuhl, Ellen

Abstract

Machine learning is rapidly reshaping constitutive modeling, offers new ways to learn material behavior directly from experimental data, and challenges long-established modeling paradigms. But with a growing number of machine-learning-based approaches available, how do they compare in practice? In this paper, we use the classic experimental data of Treloar to benchmark popular frameworks for hyperelasticity: (Generalized-Invariant) Constitutive Artificial Neural Networks, Physics-Augmented Neural Networks, (Adaptive) Material Fingerprinting, and Efficient Unsupervised Constitutive Law Identification & Discovery. We compare their fitting performance, computational cost, hyperparameter sensitivity, and ease of implementation. Furthermore, we discuss the trade-offs between predictive accuracy and model complexity. The latter is assessed by quantifying both the number of material parameters in the discovered models and the computational time required to evaluate the constitutive model and its derivatives. The results show that all methods can reproduce the benchmark data remarkably well. Rather than identifying a single winner, we highlight the strengths and limitations of each approach and provide practical guidance for their use. The source code for all six methods, including the training and comparison scripts, as well as all results and data used in this study, is publicly available via https://doi.org/10.5281/zenodo.21915635.

Chinese Translation

机器学习正在迅速重塑本构建模，提供了直接从实验数据中学习材料行为的新方法，并挑战了长期以来确立的建模范式。然而，随着越来越多基于机器学习的方法的出现，它们在实践中的比较如何？在本文中，我们使用经典的Treloar实验数据对流行的超弹性框架进行基准测试：广义不变本构人工神经网络（Generalized-Invariant Constitutive Artificial Neural Networks）、物理增强神经网络（Physics-Augmented Neural Networks）、自适应材料指纹（Adaptive Material Fingerprinting）以及高效无监督本构法则识别与发现（Efficient Unsupervised Constitutive Law Identification & Discovery）。我们比较了它们的拟合性能、计算成本、超参数敏感性和实现难易程度。此外，我们讨论了预测准确性与模型复杂性之间的权衡。后者通过量化发现模型中的材料参数数量以及评估本构模型及其导数所需的计算时间来进行评估。结果表明，所有方法都能出色地重现基准数据。我们并未确定单一的赢家，而是突出了每种方法的优缺点，并提供了实际使用的指导。所有六种方法的源代码，包括训练和比较脚本，以及本研究中使用的所有结果和数据，均可通过https://doi.org/10.5281/zenodo.21915635公开获取。

View on arXiv Download PDF AI Translation

cs.AI / 46 / 2608.14071

Scaling Domain Data Repetition in LLM Pretraining

在大型语言模型预训练中扩展领域数据重复

Li, Jingwei, Gu, Xinran, Dai, Rui, Hao, Xintong, Xu, Chengyin, Wu, Yan, Zheng, Shuran, Zhang, Jingzhao

Abstract

As large language models scale, their training-token budgets must also increase to maintain an appropriate tokens-per-parameter ratio ($\mathrm{TPP}$). However, high-quality domain data is much harder to scale than general web data. As model size and the training-token budget increase, its fraction in the training mixture tends to decrease. Repeating the available high-quality data provides an effective way to counteract this dilution, but excessive repetition may lead to overfitting. We study this trade-off under practical LLM scaling, where the training-token budget grows proportionally with model size. For a fixed domain, we first find that, surprisingly at a fixed $\mathrm{TPP}$, the optimal repetition count mildly increases with model size. Across different domains, we find that the optimal repetition count is strongly negatively correlated with the final validation loss of a domain: domains with lower loss can generally benefit from more repetitions. In contrast, the amount of unique domain data is only weakly related to the optimal repetition count. These findings suggest that repetition counts tuned on smaller proxy models with the same $\mathrm{TPP}$ can provide a practical estimate for larger models.

Chinese Translation

随着大型语言模型的规模扩大，其训练令牌预算也必须增加，以维持适当的每参数令牌比率（TPP）。然而，高质量领域数据的扩展远比一般网络数据困难。随着模型规模和训练令牌预算的增加，其在训练混合中的比例往往会下降。重复可用的高质量数据提供了一种有效的方式来抵消这种稀释，但过度重复可能导致过拟合。我们研究了这一权衡，特别是在实际的LLM扩展中，其中训练令牌预算与模型规模成比例增长。对于固定的领域，我们首先发现，令人惊讶的是，在固定的TPP下，最佳重复次数随着模型规模的增加而略有增加。在不同领域之间，我们发现最佳重复次数与领域的最终验证损失呈强负相关：损失较低的领域通常可以从更多的重复中受益。相比之下，独特领域数据的数量与最佳重复次数的关系较弱。这些发现表明，在相同的TPP下，针对较小代理模型调优的重复次数可以为更大模型提供实用的估计。

View on arXiv Download PDF AI Translation

cs.AI / 47 / 2608.14074

Mandato: Protocol-Level Enforcement of Digitally Signed Mandates on AI Agent Actions with Cryptographically Chained Audit Trails

Mandato：在协议层面强制执行数字签名授权的人工智能代理行为及其密码链审计轨迹

Racioppi, Giovanni

Abstract

AI agents increasingly act on external systems through standardized tool-calling protocols such as the Model Context Protocol (MCP), yet no infrastructure layer constrains their actions to what a principal has verifiably authorized: authorization logic lives in application code, is neither signed nor independently auditable, and the resulting logs lack evidentiary value. We present Mandato, a governance proxy that enforces digitally signed mandates on agent actions at the protocol level. A mandate is a machine-readable, cryptographically signed authorization artifact specifying which tools an agent may invoke, under which parameter constraints and contextual conditions, for how long, and on whose behalf; the proxy evaluates every tool call against the applicable mandate chain, blocks non-conforming calls in line, and records every decision -- permit, deny, and the evidence for each -- in an append-only, hash-chained audit log designed for evidentiary use and periodically anchored via qualified timestamps. The mandate is deliberately modeled on the civil-law institution of delegation of authority, making the artifact legible to lawyers and auditors, not only to engineers. We give the mandate model and its decision semantics, the reference architecture as an MCP-transparent proxy with separated decision and enforcement points, and a mapping of the mechanism onto EU AI Act Articles 12 and 14, GDPR accountability, NIS2, and eIDAS 2, including a roadmap to qualified attestation through Qualified Trust Service Providers (QTSPs). We describe the implementation status of the reference system and a quantitative evaluation plan covering enforcement overhead, audit completeness, and tamper-evidence verification cost.

Chinese Translation

人工智能代理越来越多地通过标准化的工具调用协议（如模型上下文协议 Model Context Protocol, MCP）在外部系统上执行操作，但没有基础设施层能够将其行为限制在可验证授权的范围内：授权逻辑存在于应用代码中，既未签名也无法独立审计，结果日志缺乏证据价值。我们提出了Mandato，这是一种治理代理，在协议层面上对代理行为强制执行数字签名的授权。授权是一个机器可读的、经过密码学签名的授权文档，指定代理可以调用哪些工具、在何种参数约束和上下文条件下、持续多长时间以及代表谁；该代理根据适用的授权链评估每个工具调用，阻止不符合的调用，并在一个仅追加、哈希链的审计日志中记录每个决策——允许、拒绝及其证据——该日志旨在用于证据用途，并通过合格时间戳定期锚定。该授权故意以民法中授权委托的制度为模型，使得该文档不仅对工程师可读，也对律师和审计人员可读。我们给出了授权模型及其决策语义，作为一个与MCP透明的代理的参考架构，分离决策和执行点，并将该机制映射到欧盟人工智能法第12和第14条、GDPR问责制、NIS2和eIDAS 2，包括通过合格信任服务提供商（Qualified Trust Service Providers, QTSPs）实现合格证明的路线图。我们描述了参考系统的实施状态和一个定量评估计划，涵盖执行开销、审计完整性和篡改证据验证成本。

View on arXiv Download PDF AI Translation

cs.AI / 48 / 2608.14075

A Pathway to General-Purpose Scientific AI: Multimodal Comprehension of Scientific Images

通向通用科学人工智能的途径：科学图像的多模态理解

D'Souza, Jennifer, Ahmed, Fahad, Andrade, Cecilia Andrea Bustamante, Frolova, Lina, Gnanasambandan, Poorani, Hussain, Dilshad, Khan, Muhammad Uzair, Nkengfoa, Nkembeng Kevin, J., Paul Praveen, Priante, Fabio, van der Werf, Sjoerd Franciscus, van Roeden, Thomas Frederik Jan

Abstract

Scientific figures and tables encode essential experimental evidence, yet remain difficult for digital libraries and multimodal AI systems to retrieve and interpret. The ALD/E-ImageMiner benchmark and ICDAR 2026 Competition on Information Extraction from Atomic Layer Deposition/Etching Scientific Figures provide 1,951 figures from 205 publications, expert-annotated for classification, data table extraction, summarization, and visual question answering. In these companion proceedings, we present a forward-looking perspective on how the benchmark can guide future scientific-image challenges. We examine how its tasks probe capabilities from visual and quantitative reading to domain-grounded reasoning and evidential justification, and how Bloom-informed question design can support deeper scientific understanding. We propose "scientific conceptual understanding from images" as a long-term benchmark objective, with future directions including broader domains and figure types, contextual and cross-document synthesis, hypothesis evaluation, provenance, uncertainty, counterfactual grounding, and open-ended multimodal research. This perspective connects the ICDAR 2026 challenge to a broader agenda for machine-actionable scientific visual knowledge and verifiable multimodal scientific AI.

Chinese Translation

科学图形和表格编码了重要的实验证据，但对于数字图书馆和多模态人工智能系统来说，检索和解释这些信息仍然困难。ALD/E-ImageMiner基准和ICDAR 2026原子层沉积/蚀刻科学图形信息提取竞赛提供了来自205篇出版物的1,951个图形，经过专家注释以进行分类、数据表提取、摘要和视觉问答。在这些伴随论文中，我们提出了一个前瞻性的视角，探讨该基准如何指导未来的科学图像挑战。我们考察了其任务如何探测从视觉和定量阅读到领域基础推理和证据证明的能力，以及如何通过Bloom启发的问题设计支持更深层次的科学理解。我们提出“从图像中获得科学概念理解”作为一个长期基准目标，未来方向包括更广泛的领域和图形类型、上下文和跨文档综合、假设评估、来源、确定性、反事实基础和开放式多模态研究。该视角将ICDAR 2026挑战与机器可操作的科学视觉知识和可验证的多模态科学人工智能的更广泛议程联系起来。

View on arXiv Download PDF AI Translation

cs.AI / 49 / 2608.14089

Regime-Conditional Verification: Correctness Estimation for Adapting and Monitoring Safety Classifiers

基于状态的验证：适应和监控安全分类器的正确性估计

Sandoval, Thiago, Topcu, Ufuk

Abstract

Safety classifiers deployed with large language models often fail for two reasons: their decisions reflect the policy learned during training rather than the deployer's desired policy, and their performance degrades as deployment traffic evolves. We present Regime-Conditional Verification (RCV), a lightweight wrapper that adapts an off-the-shelf safety classifier without retraining it. RCV estimates, from the classifier's internal representations, the probability that each prediction disagrees with the deployer's policy, and selectively corrects predictions likely to be wrong. The same correctness estimates also provide a label-free signal for detecting distribution shift, enabling a maintenance loop that updates the correctness estimation layer and resorts to classifier fine-tuning only when necessary. Across three off-the-shelf safety classifiers and two benchmark datasets, RCV improves adherence to the deployer's policy in every classifier-dataset combination, catching up to 0.81 of previously missed unsafe content without modifying the underlying classifier. In a deployment study with ten attack campaigns, each a harm category held out of RCV's training, RCV detects every campaign in a dedicated injection panel; in the maintenance census most drift episodes are repaired without updating the classifier, and the fine-tune is reserved for the residual episodes that repair does not restore.

Chinese Translation

部署在大型语言模型上的安全分类器通常因两个原因而失败：其决策反映了训练期间学习到的策略，而非部署者期望的策略；以及随着部署流量的演变，其性能下降。我们提出了基于状态的验证（Regime-Conditional Verification, RCV），这是一种轻量级的封装器，可以在不重新训练的情况下适应现成的安全分类器。RCV 从分类器的内部表示中估计每个预测与部署者策略不一致的概率，并选择性地纠正可能错误的预测。这些正确性估计还提供了一种无标签的信号，用于检测分布变化，从而启用一个维护循环，该循环更新正确性估计层，并仅在必要时进行分类器微调。在三个现成的安全分类器和两个基准数据集上，RCV 在每个分类器-数据集组合中提高了对部署者策略的遵循，捕获了高达 0.81 的先前遗漏的不安全内容，而无需修改基础分类器。在一项包含十个攻击活动的部署研究中，每个攻击类别均未纳入 RCV 的训练，RCV 在专用注入面板中检测到每个活动；在维护普查中，大多数漂移事件在不更新分类器的情况下得以修复，而微调则保留用于修复未恢复的剩余事件。

View on arXiv Download PDF AI Translation

cs.AI / 50 / 2608.14107

Retrieval Grounding Latent Reasoning for Dense Retrieval

检索基础的潜在推理用于密集检索

Zhou, Gang, Yu, Xiongxi, Tian, Hu, Wei, Yang, Pan, Lu, Zeng, Ke, Xu, Shibiao, Zheng, Xiaolong

Abstract

Reasoning-intensive retrieval requires text representations to capture not only semantic similarity, but also the reasoning needed to determine relevance under a given retrieval instruction. Existing reasoning-enhanced embedding models improve retrieval by incorporating reasoning information into dense representations, yet their supervision is typically dominated by the final retrieval objective. As a result, latent reasoning trajectories may learn shortcut reasoning patterns that preserve retrieval performance without producing meaningful incremental retrieval gains. We propose Retrieval Grounding Latent Reasoning (RGLT), a latent reasoning framework for dense retrieval that explicitly connects intermediate latent transitions with retrieval improvements. RGLT performs non-autoregressive reasoning in hidden space through an instruction-conditioned latent reasoning trajectory constructed from silent tokens. It combines process-supervised explicit-to-implicit distillation with retrieval-grounded supervision, using stage-wise CoT reconstruction to shape intermediate latent states and retrieval-effect credit to optimize incremental retrieval gains across the latent reasoning trajectories. Experiments on reasoning-intensive retrieval benchmarks show that RGLT consistently outperforms strong baselines while preserving efficient embedding inference.

Chinese Translation

推理密集型检索要求文本表示不仅捕捉语义相似性，还要捕捉在给定检索指令下确定相关性所需的推理。现有的增强推理的嵌入模型通过将推理信息纳入密集表示来改善检索，但其监督通常以最终检索目标为主导。因此，潜在推理轨迹可能学习到保留检索性能的捷径推理模式，而未能产生有意义的增量检索收益。我们提出了检索基础的潜在推理（Retrieval Grounding Latent Reasoning, RGLT），这是一个用于密集检索的潜在推理框架，明确将中间潜在过渡与检索改进联系起来。RGLT通过从静默标记构建的指令条件潜在推理轨迹，在隐藏空间中执行非自回归推理。它结合了过程监督的显式到隐式蒸馏与检索基础的监督，使用阶段性链式思维（CoT）重构来塑造中间潜在状态，并利用检索效果信用优化潜在推理轨迹中的增量检索收益。在推理密集型检索基准上的实验表明，RGLT在保持高效嵌入推理的同时，始终优于强基线。

View on arXiv Download PDF AI Translation

cs.AI / 51 / 2608.14109

A Graph-Based Reinforcement Learning Framework for Structured Drift Diagnosis and Recovery in Autonomous LLM Agents

基于图的强化学习框架用于自主LLM代理的结构化漂移诊断与恢复

Hamraoui, Ismail El, Jose, Sagar, Bureau, Nicolas, Plana, Robert

Abstract

Autonomous LLM agents are increasingly deployed in complex real-world workflows, yet they remain vulnerable to runtime behavioral drift, a silent deviation from the original task that can lead to irreversible side effects on external systems. Existing approaches address drift at the prompt level but lack structured mechanisms for step-level detection, risk assessment, and recovery decision. Because the main task-executing agent is often a large and expensive model that cannot be re-trained on every deployment, this work targets a plug-and-play recovery module instead. It introduces a graph-based framework in which a single small language model is trained via reinforcement learning to specialize at each node of a recovery graph, external to the main agent. Each node has a precise role\,: drift classification, operation detection, risk evaluation, or final decision and the model learns to produce structured XML-formatted reasoning adapted to that role. Training combines rule-based structural rewards with an LLM-as-judge semantic-quality signal, so that the model is graded both on how it answers (schema and length) and on what it says. Experiments on the public AppWorld benchmark show that the method generally exploits information about the suspected drift onset to issue correct recovery decisions using a small language model. In addition, the trained small language model reliably respects the prescribed output schema and produces semantically appropriate content in each field according to its assigned node role.

Chinese Translation

自主LLM代理越来越多地被部署在复杂的现实工作流程中，但它们仍然容易受到运行时行为漂移的影响，这是一种从原始任务的无声偏离，可能导致对外部系统的不可逆影响。现有方法在提示级别上解决漂移问题，但缺乏逐步检测、风险评估和恢复决策的结构化机制。由于主要任务执行代理通常是一个庞大且昂贵的模型，无法在每次部署时重新训练，因此本研究针对的是一个即插即用的恢复模块。它引入了一个基于图的框架，其中一个小型语言模型通过强化学习在恢复图的每个节点上进行专门训练，位于主代理之外。每个节点都有明确的角色：漂移分类、操作检测、风险评估或最终决策，模型学习生成适应该角色的结构化XML格式推理。训练结合了基于规则的结构奖励与LLM作为评判者的语义质量信号，使得模型在回答的方式（结构和长度）以及内容的准确性上都得到评分。对公共AppWorld基准的实验表明，该方法通常利用关于怀疑漂移开始的信息，使用小型语言模型发出正确的恢复决策。此外，训练后的小型语言模型可靠地遵循规定的输出结构，并根据其分配的节点角色在每个字段中生成语义上适当的内容。

View on arXiv Download PDF AI Translation

cs.AI / 52 / 2608.14122

Reinforcement Learning-Based Production Scheduling in an Industry-Based Coating Scenario Using the Digital Model Playground

基于强化学习的工业涂层场景中的生产调度：数字模型游乐场的应用

Kröger, Arne, Buschermöhle, Ralf, Hasselbring, Wilhelm, Wilbers, Henrik

Abstract

Production scheduling in complex manufacturing environments is challenging when sequence-dependent setup times, stochastic disturbances, and due-date constraints must be addressed simultaneously. While reinforcement learning (RL) methods have shown promising results in research, most studies rely on simplified benchmark processes, limiting their industrial relevance. This paper demonstrates the applicability of RL-based scheduling in an industry-inspired coating process that reflects practical complexities such as sequence-dependent setup times, machine breakdowns, and variable utilization. The open-source Digital Model Playground (DMPG), a discrete event simulation framework, is used to model the scenario and to train RL agents. Two standard algorithms, Deep Q-Networks and Proximal Policy Optimization, are benchmarked against conventional dispatching rules to illustrate feasibility and to provide a transparent testbed for further research. Results indicate that RL-based scheduling achieves balanced improvements across key performance indicators, with PPO delivering the most robust performance. The main contribution of this work is to bridge the gap between academic research and industrial practice by validating RL-based scheduling in a realistic, shareable scenario and by providing a reusable open-source framework for future studies.

Chinese Translation

在复杂制造环境中，生产调度面临着同时处理依赖于顺序的准备时间、随机干扰和交付期限约束的挑战。尽管强化学习（RL）方法在研究中显示出良好的前景，但大多数研究依赖于简化的基准过程，限制了其在工业中的相关性。本文展示了基于RL的调度在一个受工业启发的涂层过程中应用的可行性，该过程反映了实际复杂性，如依赖于顺序的准备时间、机器故障和可变利用率。我们使用开源的数字模型游乐场（Digital Model Playground, DMPG）作为离散事件仿真框架来建模该场景并训练RL代理。我们对两种标准算法，深度Q网络（Deep Q-Networks）和近端策略优化（Proximal Policy Optimization），与传统调度规则进行了基准测试，以展示其可行性并提供一个透明的测试平台以供进一步研究。结果表明，基于RL的调度在关键绩效指标上实现了均衡的改进，其中PPO表现出最强的性能。本工作的主要贡献在于通过在一个现实且可共享的场景中验证基于RL的调度，弥合学术研究与工业实践之间的差距，并为未来的研究提供一个可重用的开源框架。

View on arXiv Download PDF AI Translation

cs.AI / 53 / 2608.14125

Traj-LeWM: Path-Aware World-Model Planning via Latent Trajectory Cost

Traj-LeWM：通过潜在轨迹成本进行路径感知的世界模型规划

Huang, Xiaodi, Ding, Ziyi, Wan, Jingtian, Liu, Yuchen, Zhang, Yuan, Zhang, Xiao-Ping, Chen, Jiayu, Zhang, Zhang, Huang, Tao

Abstract

LeWM is a lightweight visual world model that learns latent dynamics end-to-end from pixels and ranks candidate action sequences by the distance between their predicted endpoints and the goal. However, LeWM has two limitations. First, during training, it learns local next-step transitions without evaluating complete trajectories relative to the task goal. Second, during planning, it ranks candidates solely by predicted endpoint distance. Because model predictions may differ from actual execution outcomes, the candidate whose predicted endpoint is closest to the goal may not perform best when executed in the environment. The evolution of the complete predicted trajectory can therefore provide complementary information beyond endpoint distance. To address these limitations, we propose Traj-LeWM, which retains LeWM's local-dynamics objective and endpoint score while introducing a goal-conditioned latent trajectory cost (LTC) that aggregates trajectory-level information as a complementary signal. During training, LTC-based trajectory-preference supervision complements next-step prediction in shaping the shared representation. During planning, LTC is combined with endpoint distance to incorporate intermediate-path information into candidate ranking. With joint endpoint-plus-LTC scoring, Traj-LeWM outperforms LeWM on Push-T, OGBench-Cube, Reacher, and Two-Room by $3$, $14$, $7$, and $7$ percentage points, respectively. Controlled experiments and ablations further verify the complementary roles of trajectory-level representation shaping and path-aware candidate ranking.

Chinese Translation

LeWM 是一种轻量级视觉世界模型，它从像素中端到端地学习潜在动态，并通过预测端点与目标之间的距离对候选动作序列进行排序。然而，LeWM 存在两个局限性。首先，在训练过程中，它学习局部的下一步转移，而没有评估相对于任务目标的完整轨迹。其次，在规划过程中，它仅通过预测端点距离对候选者进行排序。由于模型预测可能与实际执行结果不同，预测端点最接近目标的候选者在环境中执行时可能表现并不好。因此，完整预测轨迹的演变可以提供超越端点距离的补充信息。为了解决这些局限性，我们提出了 Traj-LeWM，它保留了 LeWM 的局部动态目标和端点评分，同时引入了一个目标条件的潜在轨迹成本（LTC），该成本聚合轨迹级信息作为补充信号。在训练过程中，基于 LTC 的轨迹偏好监督补充了下一步预测，以塑造共享表示。在规划过程中，LTC 与端点距离结合，将中间路径信息纳入候选排名。通过联合端点加 LTC 评分，Traj-LeWM 在 Push-T、OGBench-Cube、Reacher 和 Two-Room 上分别比 LeWM 提高了 $3$、$14$、$7$ 和 $7$ 个百分点。控制实验和消融实验进一步验证了轨迹级表示塑造和路径感知候选排名的互补作用。

View on arXiv Download PDF AI Translation

cs.AI / 54 / 2608.14149

QuaSAR: Quantization Compensation via Stable Activation-Aware Rank Truncation

QuaSAR：通过稳定的激活感知秩截断进行量化补偿

Lee, Lin-Fa, Chang, Yi-Yu, Yeh, Kuo-Hei

Abstract

Recent training-free post-training quantization methods restore model accuracy through closed-form residual compensation. To constrain additional model storage overhead, several existing methods gate layer selection by goodness-of-fit, retaining only those layers whose compensation yields a positive residual fit score and discarding the rest. In this paper, we show that, under the low-bit W4A4 setting, this gating mechanism fails to distinguish poorly predictable quantization error from numerical solver failure. Rank-deficient input activations yield severely ill-conditioned or numerically singular Gram matrices, causing the closed-form solver to become unstable and produce spuriously negative fit scores. Consequently, existing goodness-of-fit gates misclassify affected layers as uncompensable and discard them. Many of these discarded layers can nevertheless provide substantial error recovery when their compensation is computed using a numerically stable solver. To address this problem, we propose a parameter-free truncated pseudoinverse solver which removes collapsed directions prior to inversion. On ViT-B with the W4A4 setting, our training-free method achieves 81.42\% top-1 accuracy, outperforming prior post-training methods and fine-tuning-based baselines. Combined with joint low-rank and quantization compression, the proposed method reaches a deployable operating point of 80.26\% accuracy at 54.7 MB, providing a well-balanced trade-off between model size and accuracy.

Chinese Translation

近期的无训练后训练量化方法通过封闭形式的残差补偿恢复模型准确性。为了限制额外的模型存储开销，现有的几种方法通过拟合优度对层进行选择，仅保留那些补偿能产生正残差拟合分数的层，而丢弃其余层。在本文中，我们展示了在低比特 W4A4 设置下，这种门控机制无法区分可预测性差的量化误差与数值求解器失败。秩缺失的输入激活导致严重病态或数值奇异的 Gram 矩阵，使得封闭形式的求解器变得不稳定并产生虚假的负拟合分数。因此，现有的拟合优度门控错误地将受影响的层分类为不可补偿并将其丢弃。然而，这些被丢弃的层在使用数值稳定的求解器计算补偿时，仍然可以提供显著的误差恢复。为了解决这个问题，我们提出了一种无参数的截断伪逆求解器，该求解器在反演之前去除崩溃的方向。在 W4A4 设置下的 ViT-B 上，我们的无训练方法达到了 81.42\% 的 top-1 准确率，超越了先前的后训练方法和基于微调的基线。结合联合低秩和量化压缩，所提方法在 54.7 MB 的模型大小下达到了 80.26\% 的可部署操作点，提供了模型大小与准确性之间的良好平衡。

View on arXiv Download PDF AI Translation

cs.AI / 55 / 2608.14152

Towards Efficient Multimodal and Multilingual Opinion Extraction for STI: A QLoRA-Based Fine-Tuning Approach

面向高效的多模态和多语言观点提取用于科技情报：基于QLoRA的微调方法

Hong, Sheng, Wang, Xuanqi, Wang, Jiacheng, Wang, Yuwei

Abstract

Recent advances in large language models (LLMs) have reshaped semantic analysis. Opinion Extraction (OE) for Science and Technology Intelligence (STI) requires concise core opinions from large information streams. Off-the-shelf models struggle to filter noise from these streams and show limited structured-output reliability in zero-shot multilingual and multi-modal settings. To address information overload and extraction defocus, this study proposes a multimodal core-opinion extraction framework in which visual evidence serves as a contextual anchor for textual judgment. Using VideoLLaMA2 (VL2) and VideoLLaMA2.1 (VL2.1) as the base models, we apply Quantized Low-Rank Adaptation (QLoRA) fine-tuning on a curated dataset of 2,194 multilingual and multimodal samples. Under the selected Image-Augmented setting, fine-tuned VL2.1 generates structured JSON core-opinion outputs, achieving 64.98% Precision, 42.15% Recall, 51.14% F1-score, and 74.00% sample-level accuracy. Relative to the zero-shot VL2.1 setting, it raises the F1-scores of Spanish and Russian from 4.83% and 0.45% to 46.05% and 51.93%, respectively. The framework further incorporates a Fuzzy Cumulative Prospect Theory-based post-extraction triage module for case-level value assessment, providing a case-level value signal for downstream STI screening.

Chinese Translation

近期大型语言模型（LLMs）的进展重塑了语义分析。科技情报（STI）的观点提取（OE）需要从大量信息流中提炼出简洁的核心观点。现成模型在过滤这些信息流中的噪声方面表现不佳，并且在零-shot多语言和多模态环境下的结构化输出可靠性有限。为了解决信息过载和提取失焦的问题，本研究提出了一种多模态核心观点提取框架，其中视觉证据作为文本判断的上下文锚点。以VideoLLaMA2（VL2）和VideoLLaMA2.1（VL2.1）为基础模型，我们在一个经过精心挑选的包含2,194个多语言和多模态样本的数据集上应用了量化低秩适应（QLoRA）微调。在选定的图像增强设置下，微调后的VL2.1生成结构化的JSON核心观点输出，达到了64.98%的精确率、42.15%的召回率、51.14%的F1分数和74.00%的样本级准确率。相较于零-shot的VL2.1设置，西班牙语和俄语的F1分数分别从4.83%和0.45%提升至46.05%和51.93%。该框架进一步结合了基于模糊累积前景理论的后提取分流模块，用于案例级价值评估，为下游STI筛选提供案例级价值信号。

View on arXiv Download PDF AI Translation

cs.AI / 56 / 2608.14157

Removing Temporal Note Redundancy Improves Multimodal Reinforcement Learning for Medicine

去除时间性备注冗余提高医学多模态强化学习效果

Weng, Chenran, Lee, Joo Seung, Mahendra, Malini, Aswani, Anil

Abstract

Mechanical ventilation is a critical life-support intervention, requiring dynamic adjustments to ventilator settings as a patient's condition evolves. While reinforcement learning (RL) offers a promising framework for optimizing these sequential decisions, standard approaches rely primarily on structured electronic health record (EHR) data, missing crucial clinical context recorded in free-text notes. Integrating longitudinal clinical notes into RL state spaces is challenging because notes are heavily inflated by temporal redundancy, such as copy-forward text, templating, and repetitive documentation, which dilutes time-local updates and degrades state representation quality. To address this, we propose a redundancy-aware multimodal state representation framework that explicitly removes duplicated note text over time before policy learning. We evaluate two computationally efficient temporal decomposition strategies for removing duplicated note text: (1) an embedding-space decomposition using singular value decomposition on local history subspaces, and (2) an interpretable sentence-level diff operation that filters out previously documented sentences before text encoding. Using real-world ICU data, we demonstrate that state representations constructed by stripping temporal note redundancy significantly outperform both structured-only and raw-note baselines across multiple off-policy evaluation methods (Model-Based Rollouts, Fitted Q-Evaluation, Weighted Importance Sampling, and Weighted Doubly Robust Evaluation). Our findings show that explicitly isolating new clinical information from repeated note text yields higher-quality state representations and directly improves RL performance for clinical decision support.

Chinese Translation

机械通气是一项关键的生命支持干预措施，要求根据患者状况的变化动态调整通气机设置。尽管强化学习（RL）为优化这些连续决策提供了一个有前景的框架，但标准方法主要依赖于结构化的电子健康记录（EHR）数据，缺乏在自由文本备注中记录的重要临床背景。将纵向临床备注整合到RL状态空间中具有挑战性，因为备注中存在大量的时间性冗余，例如复制文本、模板化和重复文档，这稀释了时间局部更新并降低了状态表示的质量。为了解决这个问题，我们提出了一种冗余感知的多模态状态表示框架，该框架在策略学习之前显式去除重复的备注文本。我们评估了两种计算效率高的时间分解策略以去除重复的备注文本：（1）使用奇异值分解对局部历史子空间进行的嵌入空间分解，以及（2）一种可解释的句子级差异操作，在文本编码之前过滤掉先前记录的句子。利用真实世界的重症监护室（ICU）数据，我们证明通过去除时间性备注冗余构建的状态表示在多个离线策略评估方法（基于模型的回滚、拟合Q评估、加权重要性采样和加权双重稳健评估）中显著优于仅结构化和原始备注的基线。我们的研究结果表明，明确将新的临床信息与重复的备注文本隔离能够产生更高质量的状态表示，并直接改善临床决策支持的RL性能。

View on arXiv Download PDF AI Translation

cs.AI / 57 / 2608.14161

BiasTrace: Linking Reasoning Behaviours to Biased Outputs in LLMs

BiasTrace：将推理行为与大型语言模型中的偏见输出联系起来

Ramineni, Varsha, Rahmani, Hossein A., Ramos, Jerome, Sevegnani, Karin, Yilmaz, Emine

Abstract

LLMs exhibit social biases that can produce inaccurate and discriminatory inferences, posing risks in high-stakes applications. While prior work has made progress in measuring and mitigating bias, it largely focuses on final outputs of models, with limited understanding of the mechanisms that produce biased outcomes. Recent advances in LLM reasoning offers a new lens for investigating bias, yet the link between reasoning and bias remains poorly understood. Existing approaches focus primarily on final answer correctness or explicitly biased language, overlooking different behaviours in reasoning that can drive biased outcomes. We introduce BiasTrace, an annotation scheme for labelling reasoning behaviours in model-generated traces and linking them to biased outcomes. BiasTrace captures bias-specific behaviours (e.g., unsupported demographic assumptions) as well as general reasoning patterns that may implicitly contribute to bias (e.g. overthinking). We apply BiasTrace to reasoning traces in bias-sensitive contexts, scaled using validated LLM-as-a-judge methods, producing a large annotated dataset. Our analysis shows that biased outputs often stem from subtle reasoning behaviours rather than explicitly biased language, and that reasoning-level annotations improve bias detection. We further show that BiasTrace behaviours can be exploited for inference-time mitigation. These findings underscore the importance of examining a broader range of reasoning patterns to better understand bias in LLMs.

Chinese Translation

大型语言模型（LLMs）表现出社会偏见，这可能导致不准确和歧视性的推断，给高风险应用带来风险。尽管之前的研究在测量和减轻偏见方面取得了一定进展，但主要集中在模型的最终输出上，对产生偏见结果的机制了解有限。最近在LLM推理方面的进展为研究偏见提供了新的视角，但推理与偏见之间的联系仍然不够清晰。现有的方法主要关注最终答案的正确性或明显的偏见语言，忽视了可能导致偏见结果的不同推理行为。我们提出了BiasTrace，一种用于标注模型生成的推理轨迹中的推理行为并将其与偏见结果联系起来的注释方案。BiasTrace捕捉特定于偏见的行为（例如，未支持的人口假设）以及可能隐含地促成偏见的一般推理模式（例如，过度思考）。我们将BiasTrace应用于偏见敏感上下文中的推理轨迹，使用经过验证的LLM作为裁判的方法进行扩展，生成了一个大型注释数据集。我们的分析表明，偏见输出往往源于微妙的推理行为，而不是明显的偏见语言，并且推理级别的注释可以改善偏见检测。我们进一步表明，BiasTrace行为可以用于推断时的减轻。这些发现强调了审视更广泛的推理模式以更好地理解LLMs中的偏见的重要性。

View on arXiv Download PDF AI Translation

cs.AI / 58 / 2608.14179

Can Language Models Understand mmWave Data? Benchmarking Large Language Models for mmWave Radar-Based Human Understanding

语言模型能理解毫米波数据吗？基于毫米波雷达的人类理解的大型语言模型基准测试

Shin, Jeongwan, Kim, Jaehyeon, Ko, Donguk, Choi, Jaeho

Abstract

Large language models (LLMs) have shown remarkable reasoning and generative capabilities, motivating their use as universal reasoning engines for perception. While modern approaches such as vision-language models (VLMs) have attempted to incorporate reasoning capabilities into visual sensing, the integration of LLMs with the millimeter-wave (mmWave) modality-despite its unique advantages under low light and occlusion-remains largely unexplored. The principal bottlenecks stem from the scarcity of radar language pairs, severe cross-dataset heterogeneity, and the absence of a foundational mmWave encoder. We address this gap through a minimal textualization interface that serializes each mmWave point cloud into concise natural language, allowing off-the-shelf LLMs to operate in a question answering (QA) setting. Building on this, we present mmWave-QA, the first benchmark for language-conditioned mmWave human perception. mmWave-QA aggregates heterogeneous public mmWave datasets and harmonizes them via calibration-aware preprocessing and global taxonomy alignment, while providing natural language QA. Spanning six scenarios and five QA tasks, the benchmark enables standardized evaluation across diverse mmWave hardware and experimental conditions, establishing a foundation for scalable research on mmWave-LLM integration. We further evaluate and analyze LLMs on our mmWave-QA, highlighting their zero-shot reasoning potential for radar perception, as well as their robustness under visual degradation.

Chinese Translation

大型语言模型（LLMs）展现了卓越的推理和生成能力，这激励了它们作为感知的通用推理引擎的应用。尽管现代方法如视觉-语言模型（VLMs）已尝试将推理能力融入视觉感知，但将LLMs与毫米波（mmWave）模态结合的研究仍然相对较少，尽管其在低光照和遮挡条件下具有独特优势。主要瓶颈在于雷达语言对的稀缺、跨数据集的严重异质性以及缺乏基础的mmWave编码器。我们通过一个最小文本化接口来解决这一问题，该接口将每个mmWave点云序列化为简洁的自然语言，从而使现成的LLMs能够在问答（QA）环境中运行。在此基础上，我们提出了mmWave-QA，这是第一个针对语言条件下mmWave人类感知的基准测试。mmWave-QA汇集了异质的公共mmWave数据集，并通过校准感知的预处理和全球分类法对齐进行统一，同时提供自然语言问答。该基准涵盖六种场景和五个QA任务，能够在不同的mmWave硬件和实验条件下进行标准化评估，为mmWave-LLM集成的可扩展研究奠定基础。我们进一步在mmWave-QA上评估和分析LLMs，突出了它们在雷达感知中的零-shot推理潜力，以及在视觉退化下的鲁棒性。

View on arXiv Download PDF AI Translation

cs.AI / 59 / 2608.14205

FreeBalance: Pre-Routing Online Moe Load Balancing via Residual Workload Prediction

FreeBalance：通过剩余工作负载预测进行的预路由在线混合专家负载均衡

Chen, Pengfei, Wu, Yize, Kuang, Shouxu, Gao, Ke, Li, Ling

Abstract

Load imbalance poses a major bottleneck to the efficiency of expert parallelism in distributed inference of Mixture-of-Experts (MoE) models. The most heavily loaded rank stalls global execution due to skewed routing distributions, directly increasing latency. While offline expert placement can alleviate persistent imbalance, practical multi-task serving workloads exhibit layer- and batch-dependent routing dynamics, making online load balancing indispensable. Existing approaches rely on routing statistics collected after each MoE router, requiring expert weight load or migration to begin only after routing decisions are available, consequently placing migration overhead on the inference critical path. In this work, we observe that online balancing can instead be largely overlapped with computation before target routing (e.g., attention), if routing distributions can be predicted accurately in advance. Therefore, we propose FreeBalance, a lossless online load-balancing framework that overlaps expert migration with preceding computation stages via residual workload prediction. FreeBalance leverages cross-layer similarities in hidden representations within the residual network to build a lightweight workload predictor. This enables proactive expert migration planning before routing decisions are available, creating substantial overlap between weight transfer and computation-heavy pre-routing stages. Furthermore, a cost model constrains the number of swaps to fully hide the synchronization overhead within the available window. Experiments across models and datasets show that FreeBalance reduces the max-to-mean rank load ratio by 32.8% and end-to-end prefill latency by 13.1%. Specifically, our method hides balancing overhead of an average of 5.1 experts per layer, which would otherwise account for about 8.5% of the critical-path latency.

Chinese Translation

负载不均衡是混合专家（Mixture-of-Experts, MoE）模型在分布式推理中专家并行性效率的主要瓶颈。由于路由分布的不均匀，负载最重的节点会导致全局执行停滞，直接增加延迟。虽然离线专家放置可以缓解持续的负载不均衡，但实际的多任务服务工作负载表现出层和批次依赖的路由动态，因此在线负载均衡是不可或缺的。现有方法依赖于在每个 MoE 路由器后收集的路由统计数据，这要求专家权重负载或迁移仅在路由决策可用后才开始，从而将迁移开销置于推理的关键路径上。在本研究中，我们观察到，如果能够准确预测路由分布，在线负载均衡可以与目标路由（例如，注意力机制）之前的计算阶段大部分重叠。因此，我们提出了 FreeBalance，一个无损的在线负载均衡框架，通过剩余工作负载预测将专家迁移与前面的计算阶段重叠。FreeBalance 利用残差网络中隐藏表示的跨层相似性构建轻量级工作负载预测器。这使得在路由决策可用之前能够主动规划专家迁移，从而在权重转移和计算密集型的预路由阶段之间创造了显著的重叠。此外，一个成本模型限制了交换的数量，以完全隐藏可用窗口内的同步开销。跨模型和数据集的实验表明，FreeBalance 将最大与平均节点负载比降低了 32.8%，并将端到端预填充延迟降低了 13.1%。具体而言，我们的方法隐藏了每层平均 5.1 个专家的负载均衡开销，否则这将占据约 8.5% 的关键路径延迟。

View on arXiv Download PDF AI Translation

cs.AI / 60 / 2608.14212

APTER: Adaptive Post-Training with Expert-Grounded Rubrics

APTER：基于专家导向评分标准的自适应后训练

Wang, Xukai, Li, Liangqi, Xu, Zhiyue, Zhou, Jingang, Shi, Xiaoyu, Cai, Jiansheng, Zhang, Bo, Li, Zhe, Zhang, Xu-Yao

Abstract

As large language models enter professional domains, they must satisfy domain constraints, include critical evidence, and provide complete reasoning rather than merely produce fluent responses. Existing post-training methods often rely on holistic preferences or outcome-level verification, while recent rubric-based methods usually generate rubrics independently for each query. In specialized domains, such unconstrained rubrics may omit critical requirements and vary across samples, hindering the diagnosis and targeted repair of persistent capability deficiencies. We propose APTER (Adaptive Post-Training with Expert-Grounded Rubrics), a framework that integrates structured domain knowledge into fine-grained evaluation, optimization, and diagnosis for specialized complex reasoning. First, expert-grounded rubric construction starts from an expert criteria framework built by domain experts, where each criterion represents a stable professional capability. For each query, APTER selects relevant criteria and instantiates them into query-level rubrics linked to their source criteria, turning reusable expert criteria into executable query-level supervision without reference answers. Second, adaptive post-training uses rubric verdicts as both optimization and criterion-level diagnostic signals. Aggregating low-scoring verdicts by criterion ID reveals persistent deficiencies and triggers targeted supervised fine-tuning updates during reinforcement learning. Experiments on mathematical reasoning and medical question answering show consistent gains across both domains. Across three model generations, APTER improves the mathematics and medical averages over the corresponding base models by up to 15.86 and 8.04 points, respectively. Code and rubric datasets are available at https://github.com/AntDT-APTER/APTER.

Chinese Translation

随着大型语言模型进入专业领域，它们必须满足领域约束，包含关键证据，并提供完整的推理，而不仅仅是生成流畅的响应。现有的后训练方法通常依赖于整体偏好或结果级验证，而最近的基于评分标准的方法通常为每个查询独立生成评分标准。在专业领域，这种不受限制的评分标准可能会遗漏关键要求，并在样本之间存在差异，从而妨碍对持续能力缺陷的诊断和有针对性的修复。我们提出了APTER（基于专家导向评分标准的自适应后训练），这是一个将结构化领域知识整合到细粒度评估、优化和诊断中的框架，旨在处理专业复杂推理。首先，专家导向的评分标准构建始于由领域专家建立的专家标准框架，其中每个标准代表一种稳定的专业能力。对于每个查询，APTER选择相关标准，并将其实例化为与其源标准关联的查询级评分标准，从而将可重用的专家标准转化为无需参考答案的可执行查询级监督。其次，自适应后训练使用评分标准的裁决作为优化和标准级诊断信号。通过标准ID聚合低评分裁决，可以揭示持续的缺陷，并在强化学习过程中触发有针对性的监督微调更新。在数学推理和医学问答的实验中，两个领域均显示出一致的提升。在三代模型中，APTER在数学和医学平均分上分别比相应的基础模型提高了最多15.86和8.04分。代码和评分标准数据集可在 https://github.com/AntDT-APTER/APTER 获取。

View on arXiv Download PDF AI Translation

cs.AI / 61 / 2608.14220

A Generalized Parallelogram Rule for Proportional Analogies on Riemannian Manifolds

在黎曼流形上的比例类比的一般化平行四边形法则

Murena, Pierre-Alexandre, Hartmann, Marcelo

Abstract

Analogies are quaternary relations of the form "a is to b as c is to d", usually denoted a : b :: c : d. This notion is formalized in particular with the notion of proportional analogy, which imposes some constraints on the valid analogies. Whereas proportional analogies have been studied mostly in symbolic domains and in vector spaces, their use is limited in non-Euclidean spaces. In this paper, we introduce a proportional analogy relation in Riemannian domains, extending the parallelogram rule used for arithmetic analogies in Euclidean spaces. We illustrate the introduced analogy on various manifolds, such as the sphere, shape spaces and manifolds of probability distributions.

Chinese Translation

类比是形式为“a 对 b 如同 c 对 d”的四元关系，通常表示为 a : b :: c : d。这个概念特别通过比例类比的概念进行形式化，该概念对有效的类比施加了一些约束。尽管比例类比主要在符号领域和向量空间中得到了研究，但在非欧几里得空间中的应用却有限。在本文中，我们在黎曼领域中引入了一种比例类比关系，扩展了用于欧几里得空间算术类比的平行四边形法则。我们在各种流形上展示了所引入的类比，例如球面、形状空间和概率分布流形。

View on arXiv Download PDF AI Translation

cs.AI / 62 / 2608.14221

MathForm: Scaling Mathematical Autoformalization with Knowledge Retrieval and Verification-Guided Refinement

MathForm：通过知识检索和验证引导的精炼扩展数学自动形式化

Pu, Lushi, Zhang, Weiming, Xie, Xinheng, Fu, Zixuan, He, Bingxiang, Zhao, Hengyu, Lyu, Hongya, Li, Xin, Zhou, Jie, Wang, Yudong

Abstract

Autoformalization is commonly framed as translating natural-language mathematical statements into machine-verifiable formal languages such as Lean 4. However, faithful formalization requires more than translation. Models must map mathematical concepts to the complex hierarchy of types and definitions in formal libraries such as Mathlib, while ensuring that generated statements preserve the meaning of the source propositions. Existing approaches struggle because they rely heavily on the model's parametric memory for library-specific knowledge, while common data construction pipelines often resort to filtering single-pass outputs and lack mechanisms for feedback-driven revision. To address these challenges, we introduce MathForm, an autoformalization framework for constructing verified training data through Mathlib knowledge retrieval and verification-guided iterative refinement. Before generation, a retrieval planner gathers relevant definitions and existing formalizations from Mathlib to guide the formalization generator. Generated statements are then revised using compiler diagnostics and semantic-consistency feedback. Using this framework, we construct FormalVerse, a Lean 4 dataset containing approximately 367K verified examples across diverse mathematical domains and sources. We then train MathForm-8B through supervised fine-tuning followed by reinforcement learning. Across six benchmarks, MathForm-8B achieves average Pass@8 rates of 88.06% under Syntax Check (SC) and 72.37% under Consistency Check (CC), outperforming multiple specialized 32B autoformalizers. On the challenging FATE-H and FATE-X subsets, it attains CC pass rates of 63% and 37%, exceeding the strongest specialized baselines in both cases.

Chinese Translation

自动形式化通常被视为将自然语言数学陈述翻译为机器可验证的形式语言，如 Lean 4。然而，忠实的形式化不仅仅需要翻译。模型必须将数学概念映射到形式库（如 Mathlib）中复杂的类型和定义层次结构，同时确保生成的陈述保留源命题的含义。现有的方法面临困难，因为它们过于依赖模型的参数记忆来获取库特定知识，而常见的数据构建管道往往依赖于过滤单次输出，缺乏反馈驱动的修订机制。为了解决这些挑战，我们提出了 MathForm，一个通过 Mathlib 知识检索和验证引导的迭代精炼构建验证训练数据的自动形式化框架。在生成之前，检索规划器从 Mathlib 收集相关定义和现有形式化内容，以指导形式化生成器。生成的陈述随后使用编译器诊断和语义一致性反馈进行修订。利用该框架，我们构建了 FormalVerse，一个包含约 367K 个跨多种数学领域和来源的验证示例的 Lean 4 数据集。然后，我们通过监督微调和强化学习训练 MathForm-8B。在六个基准测试中，MathForm-8B 在语法检查（SC）下实现了 88.06% 的平均通过率，在一致性检查（CC）下实现了 72.37% 的平均通过率，优于多个专门的 32B 自动形式化工具。在具有挑战性的 FATE-H 和 FATE-X 子集上，它分别达到了 63% 和 37% 的 CC 通过率，超越了在这两种情况下最强的专门基线。

View on arXiv Download PDF AI Translation

cs.AI / 63 / 2608.14227

Attributing Preprocessing Invariance in Spectral Foundation Models

归因于光谱基础模型中的预处理不变性

Wei, Dongjun, Wu, Hongyi, Zou, Yinuo

Abstract

Preprocessing invariance is an appealing goal for spectral foundation models: a frozen model should remain useful when laboratories preprocess spectra differently. It is usually measured by training a classifier under one preprocessing pipeline and testing it under another, with preserved accuracy read as evidence of learning. We revisit that reading, using a Raman foundation model as a case study. Such models normalize their inputs before any learned parameter is applied. If that normalization maps two differently preprocessed spectra to the same vector, the encoder receives identical inputs, so the invariance cannot be attributed to learning. For a normalization that uses each spectrum's own statistics, this happens exactly when one spectrum is a positive multiple of the other plus a constant. Several standard preprocessing operations take that form. The encoder should therefore be measured against the normalization alone, which has no learned parameters. On six Raman evaluation datasets, the model does not measurably outperform its own normalization. It improves on raw spectra, but so does the normalization alone. Training does improve the encoder over random initialization, and a controlled experiment shows that it learns to ignore a transformation only when that transformation reaches it. A numerical test settles which transformations a given normalization removes. Across released systems in five modalities, most normalizations already remove transformations of that form, and several of those systems claim that invariance as learned. Replicating the comparison on two of them shows no gain either.

Chinese Translation

预处理不变性是光谱基础模型的一个吸引人的目标：一个冻结的模型在实验室以不同方式预处理光谱时仍应保持有效。通常通过在一个预处理管道下训练分类器，并在另一个预处理管道下进行测试来衡量，保留的准确性被视为学习的证据。我们重新审视了这一解读，以拉曼基础模型作为案例研究。这类模型在应用任何学习参数之前会对其输入进行归一化。如果该归一化将两个不同预处理的光谱映射到相同的向量，则编码器接收到相同的输入，因此不变性不能归因于学习。对于使用每个光谱自身统计量的归一化，当一个光谱是另一个光谱的正倍数加上一个常数时，这种情况恰好会发生。若干标准预处理操作正是这种形式。因此，编码器的性能应仅与归一化进行比较，而归一化本身没有学习参数。在六个拉曼评估数据集中，该模型在可测量上并未优于其自身的归一化。它在原始光谱上有所改善，但归一化本身也有改善。训练确实使编码器优于随机初始化，并且一个受控实验表明，只有当变换到达编码器时，它才学会忽略该变换。一个数值测试确定了给定归一化去除哪些变换。在五种模式下发布的系统中，大多数归一化已经去除了这种形式的变换，而这些系统中的多个系统声称这种不变性是学习得来的。在其中两个系统上复制比较显示没有收益。

View on arXiv Download PDF AI Translation

cs.AI / 64 / 2608.14246

Polaris : Multi Agentic System for Conversational Enterprise Analytics

Polaris：用于对话式企业分析的多智能体系统

K, Varuni H, Sarkar, Soham, Kumar, Jay, Krishnan, Goutham, Johari, Tanvi, Bharadwaj, Avinash, Hegde, Santosh

Abstract

In today's fast-paced environment, the ability to swiftly access, understand, and act on data is no longer optional; it is essential. Yet most organizations remain data-rich but insight-poor, constrained by the complexity of querying, interpreting, and explaining enterprise-scale information. We present Polaris, a supervisor-led multi-agent framework for conversational enterprise analytics that bridges this gap. Polaris introduces Dynamic Task Coordination (DTC), a decision-theoretic orchestration layer that models agent-task assignment as adaptive bipartite matching, enabling real-time coordination, recovery, and optimization across specialized agents for querying, visualization, and reasoning. By coupling DTC with reason-first, ReAct-style agents, Polaris transforms natural-language queries into coherent analytical workflows that not only retrieve and visualize data but also explain the underlying "why." Evaluation on structured enterprise datasets demonstrates high semantic fidelity and answer relevancy, underscoring the potential of multi-agent orchestration to deliver trustworthy, end-to-end business intelligence at scale.

Chinese Translation

在当今快速变化的环境中，迅速访问、理解和处理数据的能力已不再是可选的，而是必不可少的。然而，大多数组织虽然数据丰富，却缺乏洞察力，受限于查询、解释和说明企业级信息的复杂性。我们提出了Polaris，一个由监督者主导的多智能体框架，用于对话式企业分析，旨在弥补这一差距。Polaris引入了动态任务协调（Dynamic Task Coordination, DTC），这是一个决策理论的编排层，将智能体与任务的分配建模为自适应的二分匹配，能够实现实时协调、恢复和优化，适用于专门的智能体进行查询、可视化和推理。通过将DTC与以推理为先的ReAct风格智能体相结合，Polaris将自然语言查询转化为连贯的分析工作流，不仅检索和可视化数据，还解释其背后的“原因”。在结构化企业数据集上的评估显示出高语义保真度和答案相关性，突显了多智能体编排在大规模提供可靠的端到端商业智能方面的潜力。

View on arXiv Download PDF AI Translation

cs.AI / 65 / 2608.14252

Grounding Without Corrective Control: Truth-Tracking Profiles for Large Language Models

无需纠正控制的基础：大型语言模型的真相追踪特征

Reynolds, Brett

Abstract

Recent work suggests that some large language model representations have content or reference. Grounding can secure either without supplying live routes for correction. This paper asks what follows from that gap. An output is answerable when discrepancies can affect what a target- and task-specific arrangement produces, accepts, or withdraws. The arrangement has corrective control only when live, sufficiently independent routes can detect and repair fresh discrepancies. A route profile records which routes constrain the arrangement and how they are related. Those profiles support analysis of truth-tracking: patterned support for representational success. Language models are the pressure case; text-only arrangements provide a task-relative limiting case. Text-trained models inherit patterns of testimony, coherence, and prior correction. Where target-sensitive correction survives training, these can supply derivative answerability (inherited constraint); live answerability is the relation supplied by a current route for fresh discrepancies. Fluent failures should follow when a task requires independently informative access to the facts. Self-consistency, retrieval, tools, code execution, multimodal input, and feedback should help selectively. Route-by-task interactions test the distinctions. The decomposition's empirical burden is to predict held-out route--task combinations or improve intervention choice without conceptual refitting. Surface improvement and truth-tracking improvement can come apart.

Chinese Translation

近期研究表明，一些大型语言模型的表征具有内容或参考性。基础可以确保这两者，而无需提供实时的纠正路径。本文探讨了这一空白所带来的后果。当差异能够影响目标和任务特定的安排所产生、接受或撤回的内容时，输出是可回答的。仅当实时且足够独立的路径能够检测并修复新的差异时，安排才具有纠正控制。路径特征记录了哪些路径限制了安排以及它们之间的关系。这些特征支持真相追踪的分析：对表征成功的模式化支持。语言模型是压力案例；仅文本的安排提供了任务相关的极限案例。经过文本训练的模型继承了证言、一致性和先前纠正的模式。在目标敏感的纠正存活于训练的情况下，这些可以提供派生的可回答性（继承约束）；实时可回答性是当前路径对新差异所提供的关系。当任务需要独立的信息访问事实时，流畅的失败应该随之而来。自我一致性、检索、工具、代码执行、多模态输入和反馈应有选择性地提供帮助。路径与任务的交互测试了这些区别。分解的实证负担是预测保留的路径-任务组合或在不进行概念重构的情况下改善干预选择。表面改进和真相追踪改进可能会分离。

View on arXiv Download PDF AI Translation

cs.AI / 66 / 2608.14270

TimeSage-EV: A Live Benchmark for Agentic Time Series Analysis in Evolving Environments

TimeSage-EV：一个用于动态环境中智能时间序列分析的实时基准

Yao, Qingren, Kong, Yaxuan, Nie, Yuqi, Li, Yichen, Zohren, Stefan, Vettoruzzo, Anna, Wen, Qingsong, Jin, Ming, Vanschoren, Joaquin

Abstract

Time series analysis in high-stakes domains relies on recurring data releases, where new observations can alter the evidence base and the validity of later conclusions. Existing time series QA benchmarks mostly rely on fixed snapshots, leaving temporal validity and cutoff-aware evidence use unevaluated. We introduce TimeSage-EV, a live benchmark for agentic time series analysis in evolving environments. It tracks 60 real institutional scenarios across 6 domains, comprising 1,485 scenario-period QA pairs from Feb 2023 to May 2026 and spanning monthly, weekly, daily, and irregular release cadences. At each period, large language model (LLM) agents receive time series data and source reports, while the withheld target release provides ground truth. TimeSage-EV evaluates state identification, data summarization, and outlook reasoning. Experiments with frontier LLM agents and TimeSage-1.0, a novel self-evolving agent with a reusable analytical skill library, reveal significant performance gaps across model tiers and recurring failures in temporal validity, exogenous context use, and adaptation. We release TimeSage-EV as a research resource with monthly updates, code, a leaderboard, and failure-mode analyses.

Chinese Translation

高风险领域的时间序列分析依赖于定期的数据发布，其中新的观察结果可能会改变证据基础以及后续结论的有效性。现有的时间序列质量评估基准大多依赖于固定快照，未能评估时间有效性和截止日期意识下的证据使用。我们推出了TimeSage-EV，这是一个用于动态环境中智能时间序列分析的实时基准。它跟踪6个领域中的60个真实机构场景，包括从2023年2月到2026年5月的1,485个场景-时间段问答对，涵盖了每月、每周、每日和不规则发布频率。在每个时间段，大型语言模型（LLM）代理接收时间序列数据和源报告，而被保留的目标发布提供了真实答案。TimeSage-EV评估状态识别、数据总结和前景推理。与前沿LLM代理和TimeSage-1.0（一个具有可重用分析技能库的新型自我进化代理）的实验显示，不同模型层次之间存在显著的性能差距，以及在时间有效性、外部背景使用和适应性方面的反复失败。我们将TimeSage-EV作为研究资源发布，并提供每月更新、代码、排行榜和失败模式分析。

View on arXiv Download PDF AI Translation

cs.AI / 67 / 2608.14290

Intern-S2-Mobius: Foundation Model with Decoupled Knowledge and Reasoning

Intern-S2-Mobius：具有解耦知识与推理的基础模型

Chen, Kai, Ding, Jifeng, Ding, Ning, Ge, Jiaye, Gu, Lixin, Gu, Yicheng, Guo, Qipeng, Hua, Ermo, Huang, Haian, Hou, Haozheng, Hou, Jie, Hong, Xiangyu, Jiang, Che, Jin, Minxi, Liang, Cheng, Lin, Dahua, Liu, Dawei, Liu, Kuikun, Lv, Chengqi, Lv, Haijun, Lv, Han, Ma, Ningsheng, Qi, Biqing, Qian, Jianmin, Su, Shiya, Sun, Youbang, Tang, Huanze, Tian, Zhongbo, Wang, Hanjing, Wang, Rui, Wang, Ting, Wang, Yi, Wu, Baiting, Xu, Jun, Yang, Bowen, Wang, Hui, Wang, Weida, Ye, Haochen, Yu, Jiashuo, Yu, Shan, Yu, Xiaoyi, Zeng, Qirui, Zhang, Qi, Zhang, Ming, Zhang, Wenwei, Zhou, Bowen, Zhou, Xinyu

Abstract

We introduce Mobius-v0, an architecture that comprises a globally shared Memory (FFN) that stores knowledge vectors and multiple Reasoners (Self-Attn) that iteratively achieve compositional reasoning. Using hidden states as cache and carrier, reasoners repeatedly query memory for required knowledge-vectors, while the knowledge is transmitted back to reasoning operators. Through this knowledge-reasoning-separation architecture, Mobius achieves better knowledge compression and reasoning efficiency. Built upon Mobius-v0 architecture: 1) Our 7B model trained-from-scratch achieves similar downstream score as a 7B Transformer baseline with 62.6% of baseline's training data. 2) Our Intern-S2-Mobius, continually-pretrained from Qwen3.5-35B, achieves similar downstream score while delivering nearly 4x end-to-end inference speedup.

Chinese Translation

我们介绍了Mobius-v0，这是一种架构，包含一个全球共享的内存（FFN），用于存储知识向量，以及多个推理器（Self-Attn），它们通过迭代实现组合推理。推理器使用隐藏状态作为缓存和载体，反复查询内存以获取所需的知识向量，同时知识被传回推理操作。通过这种知识与推理分离的架构，Mobius实现了更好的知识压缩和推理效率。在Mobius-v0架构的基础上：1）我们的7B模型从零开始训练，达到了与7B Transformer基线相似的下游得分，使用了基线62.6%的训练数据。2）我们的Intern-S2-Mobius，基于Qwen3.5-35B进行持续预训练，达到了相似的下游得分，同时实现了近4倍的端到端推理速度提升。

View on arXiv Download PDF AI Translation

cs.AI / 68 / 2608.14306

Sensor-Driven Mission Synthesis for UAV/UGV Swarms: A TB-CSPN Coordination Architecture with Hardware-Enforced Safety

基于传感器驱动的无人机/无人地面车辆群体任务合成：一种具有硬件强制安全性的主题基础通信空间Petri网（TB-CSPN）协调架构

Borghoff, Uwe M., Bottoni, Paolo, Pareschi, Remo

Abstract

This paper presents a coordination architecture for heterogeneous UAV/UGV swarms that synthesises mission actions from uncertain, multi-modal sensor evidence while preserving hardware-enforced safety at the actuation boundary. The approach combines radar, RF, acoustic, and visual observations with Topic-Based Communication Space Petri Net (TB-CSPN) orchestration to support incremental mission formation under partial and evolving information. Consultant agents transform sensor outputs into temporally bounded semantic tokens, while supervisor agents provide authorisation and policy-governed release of mission transitions. This separation between interpretation, coordination, and execution yields auditable decision paths, constrains non-determinism within the coordination layer through guards and synchronisation, and enables bounded-time integration of heterogeneous evidence. To improve resilience in contested environments, including cyber compromise, spoofing, jamming, and communication loss, the digital coordination layer is complemented by independent analogue safety envelopes that clamp or veto unsafe actuator commands issued to individual vehicles. A coastal-surveillance case study illustrates how the proposed architecture enables dependable, governed, and physically safe swarm coordination under operational uncertainty.

Chinese Translation

本文提出了一种异构无人机/无人地面车辆群体的协调架构，该架构能够从不确定的多模态传感器证据中合成任务行动，同时在执行边界保持硬件强制安全性。该方法结合了雷达、射频、声学和视觉观测，并通过主题基础通信空间Petri网（TB-CSPN）进行编排，以支持在部分和不断变化的信息下逐步形成任务。顾问代理将传感器输出转化为时间限制的语义标记，而监督代理则提供授权和政策驱动的任务转移释放。这种解释、协调和执行之间的分离产生了可审计的决策路径，通过保护和同步限制协调层中的非确定性，并实现异构证据的有界时间集成。为了提高在竞争环境中的韧性，包括网络攻击、欺骗、干扰和通信丢失，数字协调层得到了独立的模拟安全信封的补充，这些信封可以限制或否决发给单个车辆的不安全执行命令。一个海岸监视案例研究展示了所提出的架构如何在操作不确定性下实现可靠、受管控和物理安全的群体协调。

View on arXiv Download PDF AI Translation

cs.AI / 69 / 2608.14320

AnchorBench: A Multi-Pathway Benchmark for the Anchoring Effect in LLMs

AnchorBench：用于大语言模型中锚定效应的多路径基准

Borjigin, Yiderigun, Hermann, Alexander, Cyron, Christian, Aydin, Roland

Abstract

The anchoring effect is a cognitive bias in which an initial reference value shifts a later judgment toward itself. This effect is well established in human judgment and decision-making, and recent work suggests that large language models (LLMs) exhibit similar behavior. However, existing work on anchoring in LLMs typically evaluates only a narrow set of anchor pathways and rarely distinguishes irrelevant from plausible anchors. We introduce AnchorBench, a benchmark for the anchoring effect in LLMs that evaluates multiple anchor pathways under an explicit anchor relevance axis. Across fourteen models, including ten open-weight models and four frontier API models, and a large set of controlled prompts, we find that (1) anchoring is strongly pathway-dependent, (2) plausible anchors usually induce larger shifts than irrelevant ones when introduced through stronger pathways, (3) anchor influence generally weakens as the anchor moves farther from the evidence-supported answer, most clearly on External and RAG, and (4) high task accuracy on the anchor-free control condition (Acc$_{10}$: answers within 10 points of gold) does not guarantee robustness: even frontier API models above 95% control accuracy remain susceptible to plausible anchors.

Chinese Translation

锚定效应是一种认知偏差，其中初始参考值会将后续判断向其自身偏移。该效应在人类判断和决策中已得到充分证实，最近的研究表明，大语言模型（LLMs）也表现出类似的行为。然而，现有关于LLMs中锚定的研究通常仅评估有限的锚定路径，并且很少区分无关锚和合理锚。我们提出了AnchorBench，这是一个用于评估LLMs中锚定效应的基准，能够在明确的锚定相关性轴下评估多条锚定路径。在包括十个开放权重模型和四个前沿API模型的十四个模型以及一组大量控制提示的实验中，我们发现：(1) 锚定效应强烈依赖于路径；(2) 合理锚通常在通过更强的路径引入时会引起比无关锚更大的偏移；(3) 随着锚离证据支持的答案越来越远，锚的影响通常减弱，尤其在External和RAG模型中表现得最为明显；(4) 在无锚定控制条件下的高任务准确率（Acc$_{10}$：答案在金标准的10分以内）并不保证鲁棒性：即使是控制准确率超过95%的前沿API模型也仍然容易受到合理锚的影响。

View on arXiv Download PDF AI Translation

cs.AI / 70 / 2608.14330

Program-space Diffusion for Morphology-to-Transcriptomics Prediction

形态学到转录组学预测的程序空间扩散

Swann, Ruyter, Reuben, Dorent, Daniel, Racoceanu

Abstract

Spatial transcriptomics (ST) enables genome-wide gene expression profiling while preserving tissue architecture, but its cost and limited scalability remain major bottlenecks. This has motivated models that predict spatial expression directly from routine histology. Despite promising results, most existing approaches operate at the gene level without leveraging established transcriptomic modeling practices and rely on heterogeneous gene selection strategies, which complicates fair comparison across methods. We propose to reformulate morphology-to-transcriptomics prediction as conditional generation in transcriptional program space, thereby exploiting coordinated transcriptional variation instead of predicting genes independently. Using consensus non-negative matrix factorization (cNMF), we extract a low-dimensional set of transcriptional programs capturing coordinated expression variation in the training data, and train a conditional diffusion model to generate program activations from histology. This formulation exploits coordinated transcriptional variation and substantially lowers the dimensionality of the conditional generative task.

Chinese Translation

空间转录组学（ST）能够在保留组织结构的同时进行全基因组基因表达谱分析，但其高成本和有限的可扩展性仍然是主要瓶颈。这促使了直接从常规组织学预测空间表达的模型的出现。尽管已有的研究结果颇具前景，但大多数现有方法在基因层面上操作，未能利用成熟的转录组建模实践，并依赖于异质的基因选择策略，这使得不同方法之间的公平比较变得复杂。我们建议将形态学到转录组学的预测重新表述为转录程序空间中的条件生成，从而利用协调的转录变异，而不是独立预测基因。通过共识非负矩阵分解（cNMF），我们提取了一组低维的转录程序，以捕捉训练数据中的协调表达变异，并训练一个条件扩散模型，从组织学中生成程序激活。这种表述利用了协调的转录变异，并显著降低了条件生成任务的维度。

View on arXiv Download PDF AI Translation

cs.AI / 71 / 2608.14339

Clearing the Fog: Towards Installing and Refining Proactive Exploration Capabilities in LLM Agents

消除迷雾：建立和完善大型语言模型代理的主动探索能力

Guan, Zhizhao, Huang, Chen, Liu, Ziming, Liang, Hongru, Lei, Wenqiang, Ng, See-Kiong, Chua, Tat-Seng, Cohn, Anthony G

Abstract

We study proactive exploration in LLM agents, i.e., the ability to explore an environment to acquire information that improves future decision-making. In this regard, we first identify two fundamental bottlenecks that hinder this capability and then propose \ours, a novel method designed to instill and refine proactive exploration. Specifically, \ours\ consists of two components: (1) Exploratory Data Construction, which synthesizes exploration-rich trajectories to mitigate the hindsight bias of standard demonstrations; and (2) RL Optimization with Contrastive Signal Guidance, which leverages contrastive trajectory pairs to distinguish productive exploration from redundant wandering. Extensive experiments demonstrate the effectiveness of \ours\ and provide insights into the characteristics of proactive exploration. Our code is available at: https://github.com/GuanZhizhao/SAFARI.

Chinese Translation

我们研究了大型语言模型（LLM）代理中的主动探索，即探索环境以获取信息以改善未来决策的能力。在这方面，我们首先识别出阻碍这一能力的两个基本瓶颈，然后提出了 extit{ours}，一种旨在培养和完善主动探索的新方法。具体而言， extit{ours}由两个组成部分构成：（1）探索数据构建，合成丰富的探索轨迹以减轻标准演示的事后偏见；（2）带有对比信号指导的强化学习优化，利用对比轨迹对来区分有效探索与冗余游荡。大量实验表明 extit{ours}的有效性，并提供了对主动探索特征的深入见解。我们的代码可在以下链接获取：https://github.com/GuanZhizhao/SAFARI。

View on arXiv Download PDF AI Translation

cs.AI / 72 / 2608.14354

ScienceFlow: A long-horizon agent for ML research, scientific discovery and beyond

ScienceFlow：用于机器学习研究、科学发现及其他领域的长时间跨度智能体

Zhao, Mingming, Dong, Jiqian, Xu, Kangping, Hasan, Zadid, Fan, Chengrui, Jiang, Shan, Mao, Shuai, Lingya, Ting, Zou, Linyi, Zhou, Tailin, Chan, Yun Hin, Zhang, Wenkai, Zhou, Zhanhong, Huang, Guowei, Li, Hongliang, Cun, Wenjing, Chen, Zhitang, Yuan, Mingxuan, Geng, Yanhui

Abstract

Enabling LLM agents to sustain productive, stable, and goal-aligned research over extended horizons is a central challenge for autonomous machine learning and scientific discovery, as progress hinges on continuously managing evolving state, exploration decisions, and computational resources. Pioneering autoresearch agents, despite great success, still lack mechanisms for continuity, recovery from dead ends, and value-driven compute allocation, which inherently undermines overall search efficiency, wastes computational resources, and lowers the chance of ultimate success. To bridge this gap, we introduce ScienceFlow, an end-to-end autoresearch agent framework that organizes long-horizon research work into research segments grounded in executable workspaces. It represents research progress as recoverable executable states, enabling efficient exploration, revision, and execution. Transitions between research segments are governed by Executable-State Transition through Re-Anchoring (ESTRA), which selects either the live state or an archived state as the next anchor and determines whether to continue or redirect the research trajectory. An evidence-aware execution controller allocates resources to physical jobs based on resource availability, remaining budget, and validated progress. We evaluate ScienceFlow on tasks spanning machine learning, scientific modeling, and mathematical optimization. Results on diverse long-horizon benchmarks demonstrate its ability to sustain effective research processes, highlighted by a SOTA 70.22 percent Any-Medal score on the full MLE-bench within a 24-hour budget, outperforming prior reported results by 4.92 percentage points. The efficacy of ScienceFlow further demonstrates that efficient state management, adaptive exploration, and objective-aligned execution are critical for scaling autonomous research beyond short-horizon interactions.

Chinese Translation

使大规模语言模型（LLM）智能体能够在较长时间跨度内维持高效、稳定且目标一致的研究是自主机器学习和科学发现的一个核心挑战，因为进展依赖于持续管理不断变化的状态、探索决策和计算资源。尽管开创性的自我研究智能体取得了巨大成功，但仍缺乏连续性、从死胡同中恢复和基于价值的计算资源分配机制，这在本质上削弱了整体搜索效率，浪费了计算资源，并降低了最终成功的机会。为了解决这一问题，我们提出了ScienceFlow，一个端到端的自我研究智能体框架，将长时间跨度的研究工作组织为基于可执行工作空间的研究段。它将研究进展表示为可恢复的可执行状态，从而实现高效的探索、修订和执行。研究段之间的过渡由可执行状态过渡与重新锚定（Executable-State Transition through Re-Anchoring, ESTRA）控制，该机制选择当前状态或归档状态作为下一个锚点，并决定是继续还是重新引导研究轨迹。一个基于证据的执行控制器根据资源可用性、剩余预算和验证进展为物理任务分配资源。我们在涵盖机器学习、科学建模和数学优化的任务上评估了ScienceFlow。在多样化的长时间跨度基准测试中的结果表明，其能够维持有效的研究过程，特别是在24小时预算内，在完整的MLE-bench上获得了70.22%的任何奖牌（Any-Medal）得分，超越了之前报告的结果4.92个百分点。ScienceFlow的有效性进一步表明，高效的状态管理、自适应探索和目标一致的执行对于将自主研究扩展到超越短时间跨度的交互至关重要。

View on arXiv Download PDF AI Translation

cs.AI / 73 / 2608.14355

Disentangled Shared Representations Improve Morpho-Transcriptomic Integration

解耦共享表示改善形态-转录组整合

Ostermaier, Julian, Ruyter, Swann, Dorent, Reuben, Racoceanu, Daniel

Abstract

Spatial transcriptomics (ST) enables the simultaneous profiling of gene expression and tissue morphology, creating an opportunity to learn multimodal representations capturing shared morpho-transcriptomic structure. However, standard multimodal models often compress modalities into a common latent space without explicitly separating shared and modality-specific sources of variation, which may limit downstream utility. We investigate whether explicit disentanglement of shared and private latent components improves multimodal representation learning for paired Hematoxylin \& Eosin (H\&E) and ST data. We compare VAE-based and contrastive approaches, each in standard and disentangled variants, across two cancer cohorts under matched experimental conditions. Representations are evaluated using cross-modal reconstruction, downstream probing and cross-modal probe transfer. The experiments suggest two main trends. First, contrastive objectives yield higher downstream probing performance than VAE-based models. Second, disentangled variants improve the selected reconstruction and probing metrics, although the gains depend on the model family, task, direction, and disentanglement strength. Overall, our results suggest that explicitly factorizing shared and modality-specific information can improve multimodal representation learning for spatial transcriptomics and provides a useful evaluation framework for future foundation models.

Chinese Translation

空间转录组学（ST）使得基因表达和组织形态的同时分析成为可能，为学习捕捉共享形态-转录组结构的多模态表示提供了机会。然而，标准的多模态模型通常将模态压缩到一个共同的潜在空间，而未明确区分共享和模态特异的变异来源，这可能限制下游应用的效用。我们研究了显式解耦共享和私有潜在成分是否能改善配对的苏木精-伊红（H&E）和ST数据的多模态表示学习。我们在两个癌症队列中比较了基于变分自编码器（VAE）和对比方法的标准和解耦变体，实验条件相匹配。通过跨模态重构、下游探测和跨模态探针转移来评估表示。实验结果表明了两个主要趋势。首先，对比目标的下游探测性能优于基于VAE的模型。其次，解耦变体改善了所选的重构和探测指标，尽管增益依赖于模型家族、任务、方向和解耦强度。总体而言，我们的结果表明，显式分解共享和模态特异信息可以改善空间转录组学的多模态表示学习，并为未来的基础模型提供了有用的评估框架。

View on arXiv Download PDF AI Translation

cs.AI / 74 / 2608.14359

Designing Sustainable Federated Learning as a Service using Neural Architecture Search

利用神经架构搜索设计可持续的联邦学习即服务

Patel, Keya, Mistry, Sajib, Fattah, Sheik, Kanneganti, Deepak, Krishna, Aneesh, Mahmud, Mufti, Bhuyan, Monowar

Abstract

The sustainability constraints of FLaaS consumers pose significant challenges to maintaining carbon-feasible federated training in FLaaS environments. These constraints often lead to infeasible consumer participation and unstable federated training under hard carbon constraints. We propose a Sustainable Federated Learning as a Service (SFLaaS), a carbon- constrained Neural Architecture Search (NAS) framework for heteroge- neous sustainable constraints. We introduce a requirement-driven search space that transforms consumer sustainability profiles into a feasible architecture region before federated execution. We develop a consumer-level carbon feasibility estimation mechanism to evaluate candidate architectures under dynamic carbon conditions. We propose a sustainable con- sumer scheduling strategy that adaptively selects feasible consumers and allocates local workloads to preserve consumer participation and statistical data coverage. An evolutionary search strategy jointly optimised for predictive performance, consumer feasibility, and participation coverage under hard carbon constraints. Experiments on real-world datasets and a simulated environment demonstrate the effectiveness of the proposed approach.

Chinese Translation

联邦学习即服务（FLaaS）消费者的可持续性约束对在FLaaS环境中维持碳可行的联邦训练提出了重大挑战。这些约束往往导致消费者参与不可行以及在严格的碳约束下联邦训练不稳定。我们提出了一种可持续的联邦学习即服务（SFLaaS），这是一个针对异构可持续约束的碳约束神经架构搜索（NAS）框架。我们引入了一种需求驱动的搜索空间，将消费者的可持续性特征转化为联邦执行前的可行架构区域。我们开发了一种消费者级碳可行性评估机制，以在动态碳条件下评估候选架构。我们提出了一种可持续的消费者调度策略，能够自适应选择可行的消费者并分配本地工作负载，以保持消费者参与和统计数据覆盖。在严格的碳约束下，采用一种进化搜索策略共同优化预测性能、消费者可行性和参与覆盖。对真实世界数据集和模拟环境的实验表明了所提方法的有效性。

View on arXiv Download PDF AI Translation

cs.AI / 75 / 2608.14375

Wrong but Useful: Trajectory Value Beyond Answer Correctness in Multi-Agent Messages

错误但有用：多智能体消息中超越答案正确性的轨迹价值

Yang, Chih-Hsuan, Chowdhury, Anjir Ahmed, Yang, Cheng-Hau, Zheng, Weijian, Llorente, Fernando, Ma, Xiaolong, Li, Xinyang, Huerta, Eliu A., Foster, Ian T., Thakur, Rajeev

Abstract

Multi-agent reasoning systems often use agreement, confidence, or automated scores to decide which messages should shape a final answer. Such filtering assumes that a message likely to be correct is also worth keeping. Yet a wrong answer can contain a useful decomposition, constraint, or scientific principle. We test this distinction with Diverse Hypothesis Deliberation (DHD), a controlled measurement protocol that caches five independently generated messages and replays the same downstream solver, called the integrator, with each message available or hidden. The replay comparison measures a message's trajectory value: whether making the message available helps or harms subsequent reasoning. Across five mathematics and science benchmarks and two openly available model families, gpt-oss-120b and gemma-4-31B-it, wrong-helpful messages appear in every benchmark-model combination. Among wrong-answer messages that change final correctness, more than four in ten changes are helpful in each model. Controlled repeats show that the number of repeatable message effects is unlikely to arise from replay variation alone (p=0.0002). A focused intervention on repeatable wrong-helpful messages finds that the complete message works best, while retaining its reasoning preserves more success than retaining only its answer; the source of the complete-message advantage remains open. Within the same problem, repeated trajectory-value evidence also identifies a better keep-or-remove choice than answer correctness alone. Answer correctness is therefore informative but does not determine trajectory value. DHD measures this missing property and produces reusable labels for learning when agents should listen.

Chinese Translation

多智能体推理系统通常使用一致性、信心或自动评分来决定哪些消息应影响最终答案。这种筛选假设一个可能正确的消息也值得保留。然而，一个错误的答案可能包含有用的分解、约束或科学原理。我们通过多样化假设审议（Diverse Hypothesis Deliberation, DHD）测试这一区别，这是一种控制测量协议，缓存五个独立生成的消息，并在每个消息可用或隐藏的情况下重放相同的下游求解器，称为整合器（integrator）。重放比较测量消息的轨迹价值：使消息可用是否有助于或损害后续推理。在五个数学和科学基准以及两个公开可用的模型系列（gpt-oss-120b 和 gemma-4-31B-it）中，错误有用的消息出现在每个基准-模型组合中。在改变最终正确性的错误答案消息中，每个模型中超过四成的变化是有帮助的。控制重复显示，可重复消息效果的数量不太可能仅由重放变异引起（p=0.0002）。对可重复的错误有用消息进行的集中干预发现，完整消息效果最佳，而保留其推理比仅保留其答案能保持更多成功；完整消息优势的来源仍然未解。在同一问题中，重复的轨迹价值证据也识别出比仅依赖答案正确性更好的保留或移除选择。因此，答案正确性是有信息量的，但并不决定轨迹价值。DHD测量这一缺失属性，并生成可重用标签，以便学习何时代理应倾听。

View on arXiv Download PDF AI Translation

cs.AI / 76 / 2608.14380

AgentRewind: Recoverable Execution for Long-Horizon LLM Agents

AgentRewind：长时间跨度 LLM 代理的可恢复执行

Zhuang, Yu, Chen, Kefei, Duan, Yitong, Zheng, Shuxin, Li, Jian, Zhang, Xu-Yao

Abstract

Many real-world tasks require LLM agents to interact with their environments over long execution horizons. Errors that occur early in execution may propagate through both the agent context and environment state, and their effects may be difficult to reverse through subsequent actions. Existing methods mainly seek to reduce such errors through plan refinement and safety checks but provide little support after errors occur. To enable recovery during long-horizon execution, we present AgentRewind, a runtime recovery framework that records aligned checkpoints of the agent context and controlled environment, allowing agents to return to an earlier state and resume execution with information from previous attempts. We also construct MettleBench, a benchmark for evaluating task completion and partial progress on long-horizon engineering assignments containing a series of related requirements. Experiments across tasks, multiple models, execution strategies, and agent harnesses show that AgentRewind improves task success rate and average checklist progress over the compared baselines.

Chinese Translation

许多现实世界任务要求 LLM 代理在长时间执行过程中与其环境进行交互。在执行早期发生的错误可能会在代理上下文和环境状态中传播，其影响可能难以通过后续动作逆转。现有方法主要通过计划优化和安全检查来减少此类错误，但在错误发生后提供的支持有限。为了在长时间执行过程中实现恢复，我们提出了 AgentRewind，这是一种运行时恢复框架，记录代理上下文和受控环境的对齐检查点，使代理能够返回到早期状态，并利用先前尝试的信息继续执行。我们还构建了 MettleBench，这是一个用于评估长时间跨度工程任务中任务完成和部分进展的基准，任务包含一系列相关要求。在多个任务、模型、执行策略和代理工具的实验中，结果表明 AgentRewind 提高了任务成功率和平均检查列表进展，相较于对比基线表现更佳。

View on arXiv Download PDF AI Translation

cs.AI / 77 / 2608.14392

Tripwire: Triggering Aligned Refusal via Statistically Certified Safety Neurons

Tripwire：通过统计认证的安全神经元触发对齐拒绝

Zhao, Wei, Li, Zhe, Zhang, Peixin, Sun, Jun

Abstract

Neuron- and path-level interventions offer the finest-grained route to defending large language models (LLMs) against jailbreak attacks, yet existing methods fall short of this promise, i.e., they often compromise model utility significantly. Specifically, one line of work suppresses toxic neurons to erase harmful semantics, but since such semantics are distributed across the network, blocking every pathway forces a large intervention footprint. An alternative line of research focus on identify safety neurons using external classifiers. While promising, the existing approaches suffer from compromising neurons that are important for the model utility as well. Moreover, both approaches remain always on and thus perturb every benign request even when no attack is present. To address these limitations, we present \ours{}, a training-free defense that first identifies safety-specific neurons through per-neuron hypothesis tests under false-discovery-rate control together with a utility-specificity filter. Based on this identification, a trigger-style clamp holds the selected neurons at their harmful-conditional mean activations, injecting an internal harmful-input signal that triggers the refusal behavior learned during alignment. The clamp is then realized by two provably equivalent deployment modes, namely a detector-gated inference-time intervention and an offline bias-patch weight edit. Extensive experiments across four safety-aligned LLMs and four representative attacks demonstrate that \ours{} reduces the average attack success rate to at most 2.0\% while incurring a utility drop of only 0.5\% to 5.3\% on MT-Bench, the smallest among all defenses. Code is available at https://anonymous.4open.science/r/Tripwire-65C4.

Chinese Translation

神经元和路径级干预提供了保护大型语言模型（LLMs）免受越狱攻击的最细粒度途径，但现有方法未能实现这一承诺，即它们往往显著损害模型的实用性。具体而言，一类研究通过抑制有害神经元来消除有害语义，但由于这些语义在网络中分布，阻断每条路径会导致较大的干预足迹。另一类研究则专注于使用外部分类器识别安全神经元。尽管前景可期，但现有方法也面临着损害对模型实用性重要的神经元的问题。此外，这两种方法始终处于开启状态，因此即使在没有攻击的情况下也会干扰每一个良性请求。为了解决这些局限性，我们提出了 extit{Tripwire}，这是一种无训练的防御方法，首先通过在假发现率控制下进行每个神经元的假设检验，并结合实用性特异性过滤器来识别安全特定神经元。基于这一识别，触发式夹具将所选神经元保持在其有害条件均值激活水平，注入一个内部有害输入信号，触发在对齐过程中学习到的拒绝行为。该夹具通过两种可证明等效的部署模式实现，即检测器门控推理时干预和离线偏置补丁权重编辑。在四个安全对齐的LLMs和四个代表性攻击的广泛实验中， extit{Tripwire}将平均攻击成功率降低至最多2.0 ext{%}，同时在MT-Bench上仅造成0.5 ext{%}至5.3 ext{%}的实用性下降，这是所有防御方法中最低的。代码可在 https://anonymous.4open.science/r/Tripwire-65C4 获取。

View on arXiv Download PDF AI Translation

cs.AI / 78 / 2608.14397

LLMs Don't Pay for the Jump

大型语言模型无法实现“跳跃”

Balani, Paras, Panda, Subhrakanta

Abstract

Zahavy [2026] argues that Large Language Models, despite their capabilities in induction and deduction, cannot perform the abductive "Jump" that produced Einstein's equivalence principle, and attributes this limitation to the absence of embodied simulation. Zheng-Xin [2026] and Farmer [2026] question whether embodiment is necessary for abduction, pointing to alternative routes to General Relativity and forms of abduction that require no sensorimotor grounding. Max Planck resolved the blackbody radiation problem in 1900. Planck's move to E = h{\nu} required no embodied simulation. It was motivated by a mathematical consequence of classical theory, an infinite predicted energy for a finite measured quantity, that could not be physically accepted. We show that neither induction nor deduction could have produced the postulate and argue that its adoption required a coupling between epistemic error and physical cost. We formalize this distinction through thermodynamic coupling and show that fixed-weight transformer inference lacks such coupling, regardless of model scale. This is consistent with empirical results showing that output entropy remains nearly unchanged across tasks with sharply increasing causal difficulty, even as accuracy falls from 100% to 17%. We therefore argue that the missing ingredient in machine abduction may lie deeper than embodiment: a system must have some physical mechanism through which epistemic error becomes costly enough to force revision.

Chinese Translation

Zahavy [2026] 认为，尽管大型语言模型在归纳和演绎方面具有能力，但它们无法执行产生爱因斯坦等效原理的推理“跳跃”，并将这一限制归因于缺乏具身模拟。Zheng-Xin [2026] 和 Farmer [2026] 质疑具身性是否对推理是必要的，指出通往广义相对论的替代路径以及不需要感知运动基础的推理形式。马克斯·普朗克在1900年解决了黑体辐射问题。普朗克的 E = h{ u} 的推导并不需要具身模拟，而是由经典理论的一个数学后果所驱动，即对有限测量量的无限预测能量，这是物理上无法接受的。我们表明，既没有归纳也没有演绎能够产生这一假设，并认为其采纳需要认知错误与物理成本之间的耦合。我们通过热力学耦合形式化了这一区别，并表明固定权重的变换器推理缺乏这种耦合，无论模型规模如何。这与实证结果一致，显示在因果难度急剧增加的任务中，输出熵几乎保持不变，即使准确率从100%降至17%。因此，我们认为机器推理中缺失的成分可能比具身性更深层次：一个系统必须具备某种物理机制，使得认知错误的成本足够高，以迫使其修正。

View on arXiv Download PDF AI Translation

cs.AI / 79 / 2608.14407

The Past and Future of AI Scientists

人工智能科学家的过去与未来

King, Ross D.

Abstract

We present a survey of the past and future of AI Scientists: machines capable of automating science. AI Scientists can originate hypotheses, deduce their consequences, design and execute experiments, interpret their results, and revise their beliefs. Such systems are integrated scientific agents, connected to the literature, formal knowledge, mathematical models, simulations, data-analysis systems and physical laboratories. Adam was the first machine to make novel scientific discoveries through cycles of hypothesis formation and physical experimentation. Eve established the architecture of the modern self-driving laboratory. Foundation models, autonomous agents and laboratory robotics now make it possible to build systems far more general than either Adam or Eve. The central problem is no longer whether individual components of science can be automated. They can. The problem is integration. AI Scientists must combine neural learning with logic, probability, mathematics, causal reasoning, simulation, experimental design, robotics and formal scientific records. AI Scientists have the potential to transform science: to make science faster, cheaper, more systematic and more reproducible. AI Scientists could investigate systems too complicated for unaided human science, and enable thousands of AI scientists to work together on single problems. The Nobel Turing Challenge sets the goal of developing by 2050 AI systems capable of automating Nobel-quality discoveries. Progress is ahead of schedule. When we succeed it will create a new form of science and transform the world.

Chinese Translation

我们对人工智能科学家的过去与未来进行了调查：这些机器能够自动化科学研究。人工智能科学家可以提出假设、推导其后果、设计并执行实验、解释结果并修正信念。这些系统是集成的科学代理，连接着文献、形式知识、数学模型、模拟、数据分析系统和物理实验室。亚当是第一台通过假设形成和物理实验循环进行新科学发现的机器。夏娃建立了现代自动化实验室的架构。基础模型、自治代理和实验室机器人现在使得构建比亚当或夏娃更为通用的系统成为可能。核心问题不再是科学的个别组成部分是否可以被自动化——它们可以。问题在于整合。人工智能科学家必须将神经学习与逻辑、概率、数学、因果推理、模拟、实验设计、机器人技术和正式科学记录相结合。人工智能科学家有潜力改变科学：使科学变得更快、更便宜、更系统和更可重复。人工智能科学家可以研究过于复杂而无法依靠人类科学独立解决的系统，并使成千上万的人工智能科学家能够共同研究单一问题。诺贝尔图灵挑战设定了到2050年开发能够自动化诺贝尔级发现的人工智能系统的目标。进展超出了预期。当我们成功时，它将创造一种新的科学形式并改变世界。

View on arXiv Download PDF AI Translation

cs.AI / 80 / 2608.14425

Knowing When to Stop: Bayesian Optimal Stopping for LLM Evaluations

知道何时停止：针对大型语言模型评估的贝叶斯最优停止方法

Pilditch, Toby D.

Abstract

LLM evaluations often use fixed sampling budgets, testing every item the same number of times even after estimates are precise. We introduce optstop, a precision-based adaptive stopping framework that treats evaluation as a sequential measurement problem: keep sampling where uncertainty remains high, and stop where estimates are precise or stable enough. The framework builds on hierarchical Bayesian inference, supports binary, ordinal, and continuous outcomes, and keeps every benchmark item eligible for sampling, without requiring a calibrated item bank. It runs live or retrospectively, and includes a safeguard that samples more cautiously as measured performance approaches zero, where rare successes matter most. In an illustrative 200-item, 10-epoch evaluation, it removes 57%-97% of planned trials across nine validation settings, with overall conclusions equivalent to the full run. These results show that LLM evaluation compute can be allocated by uncertainty rather than by fixed repetition counts, with the magnitude of savings depending on evaluation design.

Chinese Translation

大型语言模型（LLM）评估通常使用固定的采样预算，即使在估计值已经足够精确的情况下，也对每个项目进行相同次数的测试。我们提出了 optstop，这是一种基于精度的自适应停止框架，将评估视为一个序列测量问题：在不确定性仍然较高的地方继续采样，而在估计值足够精确或稳定的地方停止。该框架基于层次贝叶斯推断，支持二元、序数和连续结果，并保持每个基准项目都可以进行采样，而无需一个经过校准的项目库。它可以实时运行或回顾性分析，并包含一个保护机制，当测量性能接近零时更加谨慎地采样，因为在这种情况下，稀有的成功尤为重要。在一个示例性的200项、10轮评估中，它在九个验证设置中去除了57%-97%的计划试验，整体结论与完整运行相当。这些结果表明，LLM评估的计算资源可以根据不确定性进行分配，而不是固定的重复次数，节省的幅度取决于评估设计。

View on arXiv Download PDF AI Translation

cs.AI / 81 / 2608.14426

The Dynamics of Intelligence Explosions

智能爆炸的动态

Ord, Toby

Abstract

AI is increasingly being used to help with AI R&D. Under certain conditions this feedback loop might be able to produce an intelligence explosion, with rapidly escalating AI capabilities. I explore the mathematics of the most explosive possibilities, with an eye to understanding what drives the dynamics. I show that singular growth (towards a vertical asymptote) is harder to achieve than would be expected from recent economics-inspired modelling, and that there is an important but neglected class of growth rates that are faster than exponential but don't lead to a vertical asymptote. I draw out the generation time (the time to go around the feedback loop) as a neglected parameter that plays a pivotal role in determining the behaviour of any intelligence explosion --- one cannot have singular growth unless the generation time rapidly approaches zero.

Chinese Translation

人工智能（AI）正日益被用于支持人工智能的研发。在某些条件下，这种反馈循环可能会产生智能爆炸，导致人工智能能力的快速升级。我探讨了最具爆炸性可能性的数学，旨在理解驱动这些动态的因素。我展示了单一增长（朝向垂直渐近线）比近期经济学启发的模型所预期的更难以实现，并且存在一个重要但被忽视的增长率类别，其增长速度快于指数增长但不导致垂直渐近线。我指出了生成时间（反馈循环的时间）作为一个被忽视的参数，在决定任何智能爆炸行为中起着关键作用——除非生成时间迅速接近零，否则无法实现单一增长。

View on arXiv Download PDF AI Translation

cs.AI / 82 / 2608.14441

PACE-Bench: Benchmarking Physics Adaptation via Code Evolution in Dynamic Environments

PACE-Bench：在动态环境中通过代码演化进行物理适应性基准测试

Zhan, Yuhao, He, Bingxiang, Tang, Zecong, Xiao, Chaojun

Abstract

Self-evolving agents improve future behavior from interaction experience, yet existing evaluations typically optimize under fixed execution conditions and do not test recovery after those conditions change. To address this gap, we introduce PACE-Bench (Physics Adaptation via Code Evolution), a simulator-grounded benchmark of 144 source-to-target adaptation pairs across six physics domains. Each pair links a source environment to a mutated target environment with the same goal and interface. A code-driven design that succeeds in the source fails in the target, where agents must iteratively adapt it into a working target design using diagnostic sandbox feedback within a limited attempt budget. We compare ten self-evolving methods from four paradigms. The benchmark remains far from saturated: Reflexion + Qwen3-14B succeeds on only 35.9\% of full-benchmark pairs, while GPT-5.5 solves 66.7\% of the Statics subset under the full budget. Together, these results show that simulator-grounded reflection is more reliable than unverified self-revision, while memory anchors agents to early designs and broad tree search explores without converging. Even revealing exact physical changes does not raise the performance ceiling, pointing to mechanism redesign rather than parameter inference as the central bottleneck. Data and code are available at https://github.com/thunlp/PACE-Bench.

Chinese Translation

自我演化的智能体通过交互经验改善未来行为，但现有评估通常在固定执行条件下进行优化，并未测试在这些条件变化后能否恢复。为了解决这一问题，我们引入了PACE-Bench（通过代码演化进行物理适应性），这是一个基于模拟器的基准，包含144对源到目标的适应性实例，涵盖六个物理领域。每对实例将一个源环境与一个具有相同目标和接口的变异目标环境相连接。在源环境中成功的代码驱动设计在目标环境中失败，智能体必须在有限的尝试预算内，通过诊断沙盒反馈迭代地将其适应为有效的目标设计。我们比较了来自四个范式的十种自我演化方法。该基准测试仍远未饱和：Reflexion + Qwen3-14B仅在35.9%的完整基准对上成功，而GPT-5.5在完整预算下解决了66.7%的静力学子集。这些结果表明，基于模拟器的反思比未经验证的自我修正更可靠，而记忆锚定智能体于早期设计，广泛的树搜索则在不收敛的情况下进行探索。即使揭示确切的物理变化也未能提高性能上限，指出机制重设计而非参数推断是主要瓶颈。数据和代码可在 https://github.com/thunlp/PACE-Bench 获取。

View on arXiv Download PDF AI Translation

cs.AI / 83 / 2608.14446

Wyvern: An Agentic Framework for Generating Grounded Multimodal Reports

Wyvern：一种生成有依据的多模态报告的代理框架

Motetti, Beatrice Alessandra, Guandalino, Emilien, Pagliari, Daniele Jahier, Burrello, Alessio, Müller, Lorenz K., Berestizshevsky, Konstantin, Cavigelli, Lukas

Abstract

In the current artificial intelligence-driven innovation era, the pace of knowledge growth is accelerating, and is hard to keep up with. While generative models are increasingly used to synthesize content, they often lack in information grounding. To address these peculiarities of our time, we propose Wyvern, a multi-agent framework for the automated generation of grounded, multimodal technical reports. Wyvern allows for the generation of multimodal outputs, integrating images, tables, and text with supporting references in a unified report. Additionally, a particular focus is placed on the grounding of the content, with the implementation of a claims auto-revision stage. We conduct a human evaluation study to assess the quality of our proposed framework. The results show that the figures' informativeness is perceived as superior to that of a recent baseline in 87% of cases. Furthermore, Wyvern's reports are rated as more useful than those produced by three alternative methods in 63% to 100% of instances. We also carry out automatic evaluations showing that Wyvern gains up to 2.3$\times$ in citation recall and 1.6$\times$ in citation precision with respect to the baselines.

Chinese Translation

在当前人工智能驱动的创新时代，知识增长的速度正在加快，难以跟上。尽管生成模型越来越多地用于合成内容，但它们往往缺乏信息的基础性。为了解决我们时代的这些特殊问题，我们提出了Wyvern，一个用于自动生成有依据的多模态技术报告的多代理框架。Wyvern允许生成多模态输出，将图像、表格和文本与支持性参考整合到一个统一的报告中。此外，我们特别关注内容的基础性，实施了一个声明自动修订阶段。我们进行了一项人类评估研究，以评估我们提出的框架的质量。结果显示，在87%的情况下，图表的信息性被认为优于最近的基线。此外，Wyvern的报告在63%到100%的实例中被评估为比三种替代方法生成的报告更有用。我们还进行了自动评估，显示Wyvern在引用召回率上比基线提高了最多2.3倍，在引用精确度上提高了1.6倍。

View on arXiv Download PDF AI Translation

cs.AI / 84 / 2608.14452

SheetCompass: Hierarchical Relation Graphs for Agentic Spreadsheet Reasoning

SheetCompass：用于自主电子表格推理的层次关系图

He, Panjing, Cheng, Mingyue, Luo, Yucong, Li, Li, Zhang, Xiaohan

Abstract

Spreadsheets are widely used to organize, analyze, and manipulate semi-structured data, yet automated spreadsheet reasoning remains challenging for large language models (LLMs). Real-world workbooks often contain implicit cross-table associations, fine-grained column dependencies, and complex spatial layouts. Existing methods typically flatten these multidimensional structures into sequential strings, losing important intra-sheet boundaries and inter-sheet semantics. Consequently, LLMs cannot exploit the global spatial context that human experts naturally use when inspecting spreadsheets. We propose SheetCompass, a graph-guided and memory-driven agentic framework for spreadsheet reasoning and automation. SheetCompass explicitly models structural relationships within and across worksheets while maintaining task-relevant information in memory, enabling agents to reason more effectively over complex workbooks.

Chinese Translation

电子表格广泛用于组织、分析和处理半结构化数据，但对于大型语言模型（LLMs）而言，自动化电子表格推理仍然具有挑战性。现实世界中的工作簿通常包含隐含的跨表关联、细粒度的列依赖关系和复杂的空间布局。现有方法通常将这些多维结构扁平化为顺序字符串，从而丢失了重要的表内边界和表间语义。因此，LLMs无法利用人类专家在检查电子表格时自然使用的全局空间上下文。我们提出了SheetCompass，一种基于图引导和记忆驱动的自主框架，用于电子表格推理和自动化。SheetCompass明确建模了工作表内和工作表间的结构关系，同时在记忆中保持与任务相关的信息，使得代理能够更有效地对复杂工作簿进行推理。

View on arXiv Download PDF AI Translation

cs.AI / 85 / 2608.14456

Shift Aware Transfer Learning with Adaptive Dual-Encoder Fusion for PM Forecasting in Data-Limited Environments

具有自适应双编码器融合的转移学习在数据有限环境下的PM预测中的位移感知

Band, Shahab, Mohammadi, Hamed

Abstract

Short-horizon forecasting of fine particulate matter (PM2.5) remains difficult when observations from the target domain are limited and the statistical properties of the source and target domains differ. In these settings, models trained only on local data may not capture complex temporal dynamics, while direct transfer learning can result in negative transfer. This study develops a shift-aware dual-encoder transfer framework that combines source-domain knowledge with target-specific representation learning. The source encoder was pretrained using hourly observations from 10 U.S. monitoring locations. The framework was then adapted and evaluated using two years of hourly observations from 77 stations in Taiwan under a chronological train-validation-test protocol. Among the four principal baselines, the frozen-source dual-encoder model achieved the best performance, with MSE = 21.8960, MAE = 3.1597, and R^2 = 0.8725. This corresponds to an MSE reduction of approximately 7.1% relative to TL-v1 and 4.1% relative to TL-v2. The ablation analysis showed that removing the Taiwan-specific branch caused the largest decline in performance. Allowing the source encoder to adapt produced the best overall result, with MSE = 21.6575, MAE = 3.1383, and R^2 = 0.8739. SHAP analysis indicated that predictions were driven mainly by recent PM2.5 observations and meteorological variables related to pollutant transport and dispersion. These results suggest that source-domain knowledge is most effective when target-specific information is preserved and the transferred representation is allowed to adapt under target supervision.

Chinese Translation

在目标领域观测数据有限且源领域与目标领域的统计特性存在差异的情况下，细颗粒物（PM2.5）的短期预测仍然困难。在这些环境中，仅基于本地数据训练的模型可能无法捕捉复杂的时间动态，而直接的转移学习可能导致负转移。本研究开发了一种位移感知的双编码器转移框架，该框架结合了源领域知识与目标特定的表示学习。源编码器使用来自美国10个监测地点的每小时观测数据进行预训练。该框架随后在台湾77个站点的两年每小时观测数据上进行了适应和评估，采用了时间顺序的训练-验证-测试协议。在四个主要基准中，冻结源双编码器模型表现最佳，均方误差（MSE）为21.8960，平均绝对误差（MAE）为3.1597，决定系数（R^2）为0.8725。这与TL-v1相比，MSE减少了约7.1%，与TL-v2相比减少了4.1%。消融分析表明，去除台湾特定分支导致性能下降最大。允许源编码器适应产生了最佳的整体结果，MSE为21.6575，MAE为3.1383，R^2为0.8739。SHAP分析表明，预测主要受到近期PM2.5观测值和与污染物运输及扩散相关的气象变量的驱动。这些结果表明，当保留目标特定信息并允许转移表示在目标监督下适应时，源领域知识最为有效。

View on arXiv Download PDF AI Translation

cs.AI / 86 / 2608.14490

Twin: Playing an Unknown Game with a Test-Time Digital Twin

双子：与测试时数字双胞胎进行未知游戏

Skoutnev, Alexy, Acharya, Kirill, Longhitano, Gaston, Udell, Madeleine, Ellis, Kevin, Drori, Iddo

Abstract

We present a Test-time World-model Inference (Twin) system, in which a frontier coding agent writes an executable world model for completing continual learning tasks, such as ARC-AGI-3 games. Traditional approaches hand-engineer such models, one custom design per task. Each game hides its rules and goal, and our system constructs them from simulation and interaction alone. Its inductive prior over grid games is strong enough to recover the true transitions of the game and the goal on nearly all levels. Replay validation happens in a twin world model. The harness enforces that an action is not made until the program reproduces every previous observed game transition. Each mismatch between a world model prediction and the actual action result becomes a counterexample that is used to repair the world model. Twin clears 179 out of 183 levels (97.8%), and does so more efficiently than humans in 158 out of 179 levels (88.3%). The system infers the goal before any reward on 156 of the levels it clears (87.2%), and in the remaining levels automatically discovers the goal by search. The benchmark scores completion and action efficiency, between 0 and 100, against humans playing each game for the first time. Played directly, the base model scores only 7.8%; an off-the-shelf harness increases it to 61.1%, whereas our twin world model increases the same base model to 93.3%, clearing 23 out of 25 games. Building a usable world model is simpler than anticipated, whereas the harder problem is inferring the right goal.

Chinese Translation

我们提出了一种测试时世界模型推理系统（Twin），其中一个前沿编码代理为完成持续学习任务（如ARC-AGI-3游戏）编写可执行的世界模型。传统方法为每个任务手动设计这样的模型。每个游戏隐藏其规则和目标，而我们的系统仅通过模拟和交互构建这些规则和目标。其在网格游戏上的归纳先验足够强大，能够在几乎所有关卡中恢复游戏的真实转变和目标。回放验证发生在一个双胞胎世界模型中。该系统确保在程序重现每个先前观察到的游戏转变之前，不会执行任何动作。世界模型预测与实际动作结果之间的每个不匹配都成为一个反例，用于修复世界模型。Twin在183个关卡中清除了179个（97.8%），并且在179个关卡中有158个（88.3%）的效率超过人类。该系统在清除的156个关卡中在任何奖励之前推断出目标（87.2%），而在剩余关卡中通过搜索自动发现目标。基准测试对完成度和动作效率进行评分，范围在0到100之间，与人类首次玩每个游戏的表现进行比较。直接使用时，基础模型仅得分7.8%；而现成的工具将其提高到61.1%，而我们的双胞胎世界模型将相同的基础模型提高到93.3%，清除了25个游戏中的23个。构建一个可用的世界模型比预期的更简单，而更难的问题是推断正确的目标。

View on arXiv Download PDF AI Translation

cs.AI / 87 / 2608.14509

Split the Labor: Separating Evidence Interpretation from Decision Aggregation

分工：将证据解释与决策聚合分开

Wu, Zhelun

Abstract

Systems that ask a language model to reach a conclusion from many sources usually concatenate them into one prompt. This conflates two operations with different requirements. Interpreting a source rewards capacity and context. Combining interpretations rewards fixed arithmetic, comparability across instances, and the option to return nothing. Once separated, the design problem becomes the interface between them. We propose a four-field evidence tuple (hypothesis, reliability bucket, rationale, provenance) and show that fixing it determines both halves. The separation also reveals a failure mode in how such systems combine, which we call count-scale drift. Thresholding a sum of unnormalized weights is exactly posterior thresholding, but at an operating point that slides with the number of sources consulted. The slide grows with reader reliability. When source reliabilities differ, the vote rule and the posterior order instances differently, and no threshold reconciles them. Pooling calibrated log-likelihood ratios addresses both problems. The fix is arithmetic rather than architectural, and applies to a class of rules beyond language models: score-summing triage engines, diagnostic panels scored by counting positives, and additive multi-signal detectors. We then instantiate the principle twice on one longitudinal corpus, once after outcomes resolve and once before. The same partition helps in both, at different granularities: over reading in the first, over learning capacity in the second. There, a small sequence encoder on an easy auxiliary objective plus a tree ensemble carrying the censored survival loss reaches 0.921 AUPRC against 0.805 for a hand-crafted baseline. We separate what transfers from what must be re-estimated per domain, and state five predictions that would falsify the framework, three negative results, and which comparisons remain confounded.

Chinese Translation

要求语言模型从多个来源得出结论的系统通常将这些来源连接成一个提示。这将两种具有不同要求的操作混为一谈。解释一个来源需要能力和上下文，而组合解释则需要固定的算术、实例间的可比性以及返回空值的选项。一旦将这两者分开，设计问题就变成了它们之间的接口。我们提出了一个四字段证据元组（假设、可靠性桶、理由、来源），并表明修正它可以确定这两个部分的内容。分离还揭示了此类系统组合时的一种失败模式，我们称之为计数-规模漂移（count-scale drift）。对未归一化权重的总和进行阈值处理恰好是后验阈值处理，但在一个随着咨询来源数量滑动的操作点上。随着读者可靠性的提高，这种滑动会加剧。当来源的可靠性不同时，投票规则和后验顺序对实例的处理方式不同，且没有一个阈值能够调和它们。汇聚经过校准的对数似然比可以解决这两个问题。这个修正是算术上的，而非架构上的，并适用于超出语言模型的规则类别：得分求和的分诊引擎、通过计数阳性结果进行评分的诊断小组，以及加性多信号探测器。然后，我们在一个纵向语料库上两次实例化这一原则，一次在结果确定后，一次在结果确定之前。同样的分区在两者中都发挥了作用，但粒度不同：在第一次中侧重于阅读，在第二次中侧重于学习能力。在那里，一个在简单辅助目标上的小序列编码器加上一个携带审查生存损失的树集成达到了0.921的AUPRC，相较于手工制作的基线0.805。我们分离了可转移的内容与每个领域必须重新估计的内容，并提出了五个可以证伪该框架的预测、三个负面结果，以及哪些比较仍然混淆。

View on arXiv Download PDF AI Translation

cs.AI / 88 / 2608.14522

Participatory Moral AI Is Not Neutral: The Invisible Hand of Developers

参与式道德人工智能并非中立：开发者的隐形之手

Kim, Taenyun, Bogucka, Edyta, Quercia, Daniele

Abstract

As AI systems make more morally loaded decisions across society, one response has been moral preference elicitation. In this approach, researchers poll participants on hypothetical dilemmas and use the aggregated votes to train a policy that an AI model then applies at scale. Before any vote is cast, developers make three key choices in the moral AI elicitation pipeline: feature scoping, voter sampling, and question framing. In other words, they decide which features go to a vote, which voters to include, and how to present the question. These choices are often opaque, undocumented, and treated as technical details rather than normative ones. We examine each of these choices within a common empirical study and show that each can shape the preferences produced by moral AI elicitation. Across two phases (N = 809) in three deployment contexts (i.e., AI kidney allocation, AI agents simulating absent workers, and generative AI depictions of the deceased), we examine the three main stages of the moral AI elicitation pipeline. First, morally relevant features shift across contexts. This suggests that feature schemas should not be assumed to transfer across deployment domains. Second, preferences differ by political ideology for roughly one-third of features, with some differences reversing direction. The ideological composition of the voter pool can therefore affect the resulting aggregated preference profile. Third, the wording of the elicitation question can narrow or widen ideological gaps by up to a full scale point. The framing conditions also change how moral foundations are associated with participants' judgments. Taken together, these findings suggest that voting-based alignment cannot deliver fair or transparent AI by aggregation alone; at minimum, each stage of the moral AI elicitation pipeline should be audited and disclosed.

Chinese Translation

随着人工智能系统在社会中做出越来越多道德负担的决策，一种应对方式是道德偏好引导。在这种方法中，研究人员对参与者进行假设困境的调查，并利用汇总的投票结果训练一个政策，随后该政策被人工智能模型大规模应用。在投票之前，开发者在道德人工智能引导流程中做出三个关键选择：特征范围、投票者抽样和问题框架。换句话说，他们决定哪些特征进行投票、包括哪些投票者以及如何呈现问题。这些选择通常是不透明的、没有文档记录的，并且被视为技术细节而非规范性问题。我们在一个共同的实证研究中考察了这些选择，并展示了每个选择如何塑造道德人工智能引导所产生的偏好。在三个部署背景下的两个阶段（N = 809）（即人工智能肾脏分配、模拟缺席工人的人工智能代理和生成性人工智能对已故者的描绘），我们考察了道德人工智能引导流程的三个主要阶段。首先，道德相关特征在不同背景下发生变化。这表明特征模式不应假定可以在不同部署领域之间转移。其次，约三分之一的特征在政治意识形态上存在差异，某些差异的方向甚至会反转。因此，投票者池的意识形态构成可能会影响最终的汇总偏好轮廓。第三，引导问题的措辞可以将意识形态差距缩小或扩大至整整一个量表点。框架条件也改变了道德基础与参与者判断之间的关联。综合来看，这些发现表明，仅依靠投票基础的对齐无法通过汇总实现公平或透明的人工智能；至少，道德人工智能引导流程的每个阶段都应进行审计和披露。

View on arXiv Download PDF AI Translation

cs.AI / 89 / 2608.14528

Handover of In-Context Learning State Across Session Boundaries

跨会话边界的上下文学习状态交接

Kato, Masahiro, Kato, Taka

Abstract

This study investigates the methodological and theoretical properties of session handover in applications that use large language models. A task may continue in a new session when the context reaches the model's input limit, when the application restarts, or when another agent is asked to finish the task. The application must then decide which information from the earlier session to pass on. We formulate handover as the transfer of a task-relative in-context learning (ICL) state and distinguish exact recovery of earlier material from preservation of the target distribution. Under an exogeneity condition, predictive equivalence characterizes the coarsest deterministic sufficient handover and gives a fixed-length bit requirement. The analysis isolates the effects of the memory constraint, the writer, and the continuation procedure, and quantifies the cost of writing before the realized downstream query is known. We propose a three-part record that stores decisions and constraints exactly, uses task-justified statistics for repeated evidence, and retains original observations whose effect is not preserved by those statistics. Gaussian linear regression gives an exact finite-dimensional handover and finite-bit perturbation bounds, while nonparametric regression gives upper and lower bounds that relate memory to squared prediction error. These results provide a theory and method for deciding what a handover must retain and how its memory requirement depends on the continuation task.

Chinese Translation

本研究探讨了在使用大型语言模型的应用中会话交接的方法论和理论特性。当上下文达到模型的输入限制、应用程序重启或请求其他代理完成任务时，任务可能会在新会话中继续。应用程序必须决定从早期会话中传递哪些信息。我们将交接形式化为任务相关的上下文学习（In-Context Learning, ICL）状态的转移，并区分早期材料的精确恢复与目标分布的保持。在外生性条件下，预测等价性表征了最粗糙的确定性充分交接，并给出了固定长度的比特要求。该分析隔离了记忆约束、写入者和继续程序的影响，并量化了在实现下游查询已知之前写入的成本。我们提出了一种三部分记录，准确存储决策和约束，使用任务合理的统计数据来重复证据，并保留那些统计数据未能保留的原始观察。高斯线性回归提供了精确的有限维交接和有限比特扰动界限，而非参数回归则给出了与记忆相关的平方预测误差的上下界。这些结果为决定交接必须保留的内容以及其记忆要求如何依赖于继续任务提供了理论和方法。

View on arXiv Download PDF AI Translation

计算语言学 (Computation and Language)

cs.CL / 1 / 2608.13568

Does a Language Server Save Tokens for Coding Agents? A Measurement Methodology and Preliminary Study

语言服务器是否为编码代理节省了令牌？一种测量方法学和初步研究

Xu, Pengcheng

Abstract

Coding agents spend most of their context budget on retrieval. Lexical retrieval (grep) is universal, instant, and zero-setup, but noisy: it cannot tell a definition from a call from a comment. Semantic retrieval via the Language Server Protocol (LSP) is precise and typed, but needs a running, indexed server and pays a per-symbol round-trip. The claim that semantic retrieval is more token-efficient is, we find, asserted almost everywhere and measured almost nowhere: no public source isolates the LSP-vs-lexical token delta for an agent at equal task-success. This paper formalizes the question with one metric (tokens-to-success), specifies a five-arm ablation isolating semantic retrieval from confounds, maps three pre-stated failure modes onto measurable variables, and reports a preliminary study (Python and TypeScript repos; Claude Opus 4.8, Sonnet 4.6, Haiku 4.5). The answer is conditional and usually negative. On symbol-named localization the LSP costs tokens (+6% to +118%) and the agent ignores it when free. On reference-completeness it buys precision but not token savings and cannot raise the recall ceiling set by agent thoroughness; it saves tokens only for the weakest model. Tool choice is task-dependent: models default to grep on localization (0-6% semantic use) but reach for the LSP about half the time on reference tasks, unprompted. On edits scored by real test execution the gap is starkest: grep solves multi-file renames perfectly, a location-only LSP fails three-quarters of them by missing a call site, and even a complete, index-warmed, text-enriched LSP (each reference's line inline, as production LSP-MCP servers do) recovers most of the gap but cannot close it, since a rename must touch comments and strings that semantic references exclude. The implication is not LSP-always but an adaptive router keyed on task class, model capability, and lexical noise.

Chinese Translation

编码代理在检索上花费了大部分上下文预算。词汇检索（grep）是通用的、即时的且无需设置，但噪声较大：它无法区分定义、调用和注释。通过语言服务器协议（LSP）的语义检索是精确且类型化的，但需要一个正在运行的、已建立索引的服务器，并且每个符号都需要往返传输。我们发现，关于语义检索更具令牌效率的说法几乎无处不在，但几乎没有地方进行测量：没有公共来源将LSP与词汇令牌差异在相同任务成功率下进行隔离。本文通过一个指标（成功所需令牌数）形式化了这一问题，指定了一个五臂消融实验以隔离语义检索与混淆因素，映射了三种预先声明的失败模式到可测量变量，并报告了一项初步研究（Python和TypeScript代码库；Claude Opus 4.8，Sonnet 4.6，Haiku 4.5）。答案是有条件的，通常是负面的。在符号命名的定位上，LSP的令牌成本增加了6%到118%，而代理在免费时会忽略它。在引用完整性方面，它提高了精确度但未能节省令牌，并且无法提高代理彻底性所设定的召回上限；它仅对最弱的模型节省令牌。工具选择依赖于任务：模型在定位任务上默认使用grep（语义使用为0-6%），但在引用任务上大约一半的时间会选择LSP，且没有提示。在通过实际测试执行评分的编辑中，差距最为明显：grep完美解决了多文件重命名，而仅定位的LSP由于错过调用位置，失败了四分之三的案例，即使是一个完整的、索引预热的、文本丰富的LSP（每个引用的行内联，如生产LSP-MCP服务器所做）也恢复了大部分差距，但无法弥补，因为重命名必须涉及注释和字符串，而语义引用则将其排除。其含义不是LSP总是适用，而是一个适应性路由器，基于任务类别、模型能力和词汇噪声进行调整。

View on arXiv Download PDF AI Translation

cs.CL / 2 / 2608.13570

Think in Latent, Explain in Language: Self-Explainable Latent Reasoning

在潜在空间中思考，用语言解释：自解释的潜在推理

Zhao, Dayuan, Cao, Shengcao, Wang, Yu-Xiong, Gui, Liang-Yan

Abstract

Latent reasoning has emerged as a powerful alternative to text-based Chain-of-Thought (CoT), offering significant gains in computational efficiency by compressing verbose reasoning into compact embeddings. However, compressing reasoning into the latent space renders the thinking opaque, hindering its interpretability. Current methods present a stark trade-off: they either function as unexplainable ''black boxes'' (e.g., Coconut), where the latent reasoning is not human-readable, or rely on separate post-hoc decoders for explainability (e.g., Heima), introducing architectural overhead and decoupling the explanation from the actual reasoning process. In this work, we present a unified framework for Self-Explainable Latent Reasoning (SELR) that trains a single model to perform efficient and inherently explainable latent reasoning. Our core contribution is a novel multi-task training objective that optimizes for two goals simultaneously: (1) an Answer Loss that optimizes the latent reasoning trajectory to produce accurate final answers, and (2) a CoT Loss that explicitly trains the same model to decode its own latent representations back into human-understandable reasoning steps. This design ensures that generated latent representations are both task-effective and semantically interpretable, eliminating the need for external decoders. We validate the effectiveness of SELR on both Large Language Models (LLMs) and Vision-Language Models (VLMs), demonstrating that SELR achieves superior token efficiency and accuracy compared to baselines, while uniquely providing self-contained explainability without auxiliary models. Project page is available at https://jasondayuan.github.io/SELR/.

Chinese Translation

潜在推理已成为文本基础的思维链（Chain-of-Thought, CoT）的强大替代方案，通过将冗长的推理压缩为紧凑的嵌入，显著提高了计算效率。然而，将推理压缩到潜在空间使得思维变得不透明，妨碍了其可解释性。目前的方法存在明显的权衡：它们要么作为不可解释的“黑箱”（例如，Coconut）运作，其中潜在推理不可被人类理解，要么依赖于单独的后期解码器来实现可解释性（例如，Heima），这引入了架构开销并使解释与实际推理过程脱钩。在本研究中，我们提出了一个统一的自解释潜在推理（Self-Explainable Latent Reasoning, SELR）框架，该框架训练一个单一模型以执行高效且内在可解释的潜在推理。我们的核心贡献是一个新颖的多任务训练目标，旨在同时优化两个目标：（1）优化潜在推理轨迹以产生准确最终答案的答案损失（Answer Loss），以及（2）明确训练同一模型将其潜在表示解码回人类可理解的推理步骤的思维链损失（CoT Loss）。这一设计确保生成的潜在表示既在任务上有效，又在语义上可解释，消除了对外部解码器的需求。我们在大型语言模型（Large Language Models, LLMs）和视觉语言模型（Vision-Language Models, VLMs）上验证了SELR的有效性，证明SELR在令牌效率和准确性方面优于基线，同时独特地提供了无需辅助模型的自包含可解释性。项目页面可访问 https://jasondayuan.github.io/SELR/。

View on arXiv Download PDF AI Translation

cs.CL / 3 / 2608.13571

Not All Tokens Are Equal: Inflation-Aware Routing for Agentic LLM Systems

并非所有标记都是平等的：面向通货膨胀的代理型大语言模型系统路由

Fu, Heming, Lin, Shan, Xie, Qianqian, Xiong, Guojun

Abstract

When a language model fails to answer a query on the first attempt, an agentic system retries, consuming additional tokens each time. This retry overhead creates a gap between what a model's per-token price implies and what a full workflow actually costs. We call this gap \emph{token inflation} and define it as the ratio of true workflow cost to single-call cost. Systems like FrugalGPT route based on the latter, which can underestimate real cost by more than $2\times$ on difficult tasks. We address this with InflationAgent, a four-stage router that (1) measures token inflation systematically across model tiers and task types, finding inflation as high as $4.25\times$ for a 7B model on multi-hop question answering; (2) introduces CoT Branching Entropy (CBE), a pre-execution difficulty signal computed entirely from local inference, which predicts high inflation with AUROC 0.887; and (3) selects models by maximizing a Semantic Exchange Rate (SER) that divides expected accuracy by predicted true cost, with a fresh-escalation policy that discards failed chains before routing to a stronger model. On GSM8K under a fixed budget, InflationAgent achieves 94.7\% accuracy versus 91.0\% for FrugalGPT while using 31\% fewer tokens, and we show that forwarding a failed reasoning chain to GPT-4o reduces its accuracy by up to 34.8 percentage points, validating the fresh-escalation design.

Chinese Translation

当语言模型在第一次尝试中未能回答查询时，代理系统会进行重试，每次消耗额外的标记。这种重试开销造成了模型每标记价格所暗示的成本与完整工作流实际成本之间的差距。我们将这一差距称为 extit{标记通货膨胀}，并将其定义为真实工作流成本与单次调用成本的比率。像FrugalGPT这样的系统基于后者进行路由，这在困难任务中可能低估真实成本超过$2 imes$。我们通过InflationAgent来解决这个问题，该系统是一个四阶段路由器：(1) 系统地测量不同模型层级和任务类型的标记通货膨胀，发现对于7B模型在多跳问答中通货膨胀高达$4.25 imes$；(2) 引入了完全基于局部推理计算的执行前难度信号——CoT Branching Entropy (CBE)，其预测高通货膨胀的AUROC为0.887；(3) 通过最大化语义交换率（Semantic Exchange Rate，SER）来选择模型，该比率将预期准确率与预测真实成本相除，并采用新鲜升级策略，在路由到更强模型之前丢弃失败的链条。在固定预算下，InflationAgent在GSM8K上实现了94.7\%的准确率，而FrugalGPT为91.0\%，同时使用的标记减少了31\%。我们还展示了将失败的推理链转发给GPT-4o会使其准确率降低多达34.8个百分点，从而验证了新鲜升级设计的有效性。

View on arXiv Download PDF AI Translation

cs.CL / 4 / 2608.13578

BCMT: Blockwise Causal Memory Transformer

BCMT：块状因果记忆变换器

Arezki, Rachid

Abstract

Transformer architectures rely on dense self-attention to model long-range dependencies, but this mechanism exhibits quadratic complexity with respect to sequence length. We introduce BCMT (Blockwise Causal Memory Transformer), an architecture for long-context language modeling that decouples local token interactions from global context propagation. Dense causal self-attention is applied independently within local blocks, while each block produces an adaptive summary aggregated through an exponential causal memory. This memory is subsequently injected back into the token representations, enabling efficient propagation of long-range contextual information without relying on explicit global attention. Unlike standard Transformers and recurrent memory architectures, BCMT maintains neither dense interactions between distant tokens nor learned memory states. Its memory mechanism is fully parallelizable and remains compatible with standard implementations of dense self-attention. Experiments on language modeling with context lengths of up to 1024 tokens show that BCMT achieves validation performance comparable to that of Dense Transformers while significantly improving training throughput and reducing memory consumption. An ablation study further confirms that these improvements arise from the proposed memory mechanism. These results demonstrate that an exponential causal memory constructed from block summaries provides an effective alternative to dense global attention mechanisms for long-context language modeling.

Chinese Translation

变换器架构依赖于密集自注意力机制来建模长距离依赖关系，但该机制在序列长度方面表现出平方复杂度。我们提出了BCMT（块状因果记忆变换器），这是一种用于长上下文语言建模的架构，它将局部标记交互与全局上下文传播解耦。密集因果自注意力在局部块内独立应用，而每个块生成一个通过指数因果记忆聚合的自适应摘要。该记忆随后被注入回标记表示中，使得在不依赖显式全局注意力的情况下有效传播长距离上下文信息。与标准变换器和递归记忆架构不同，BCMT既不保持远程标记之间的密集交互，也不学习记忆状态。其记忆机制是完全可并行化的，并且与密集自注意力的标准实现兼容。在上下文长度达到1024个标记的语言建模实验中，BCMT的验证性能与密集变换器相当，同时显著提高了训练吞吐量并减少了内存消耗。消融研究进一步确认这些改进源于所提出的记忆机制。这些结果表明，从块摘要构建的指数因果记忆为长上下文语言建模提供了一种有效的替代方案，取代了密集全局注意力机制。

View on arXiv Download PDF AI Translation

cs.CL / 5 / 2608.13580

Jais 2: A Family of Arabic-Centric Open Large Language Models

Jais 2：一个以阿拉伯语为中心的大型开放语言模型家族

Anwar, Mohamed, Freihat, Abed Alhakim, Ibrahim, George, Awad, Mostafa, Sadallah, Abdelrahman, Gosal, Gurpreet, Ramakrishnan, Gokulakrishnan, Chandran, Sarath, Mishra, Biswajit, Joshi, Rituraj, Frikha, Ahmed, Goffinet, Etienne, Maiti, Abhishek, Filali, Ali El, AlBarri, Sarah, Ghosh, Samujjwal, Pal, Rahul, Mullah, Parvez, Shukla, Awantika, siddiki, Sajid, Kamboj, Samta, Pandit, Onkar, Sahu, Sunil Kumar, Elbadawy, AbdelRahman, Mohamed, Amr, Chamma, Ahmad, Dufraisse, Evan, Bounhar, Abdelaziz, Bouch, Dani, Abdine, Hadi, Shang, Guokan, Koto, Fajri, Wang, Yuxia, Xie, Zhuohan, Mekky, Ali, Elbadry, Rania, Ahmad, Sarfraz, Ahsan, Momina, Herraoui, Omar El, Orel, Daniil, Iqbal, Hasan, Elzeky, Kareem, Abassy, Mervat, Elozeiri, Kareem, Eletter, Saadeldine, Atif, Farah, Mukhituly, Nurdaulet, Li, Haonan, Han, Xudong, Singh, Aaryamonvikram, Quraishi, Zainul Abedien Ahmed, Sengupta, Neha, Murray, Larry, Sheinin, Avraham, Hestness, Joel, Vassilieva, Natalia, Ren, Hector Xuguang, Liu, Zhengzhong, Vazirgiannis, Michalis, Nakov, Preslav

Abstract

Jais 2 is a family of Arabic-centric large language models developed jointly by MBZUAI, Cerebras, and Inception, designed to advance Arabic-centric language modeling, with strong performance across the Arabic and culturally grounded benchmarks evaluated in this report. The family includes, to our knowledge, the largest open Arabic-centric LLM trained from scratch at 70B parameters, and a competitive 8B-parameter variant among the evaluated open models. A custom Arabic-centric vocabulary enables efficient training and inference. In addition, an optimized architecture and training recipe yield highly compute-efficient training. With a substantially smaller token budget than comparable models, Jais 2 achieves strong Arabic performance on the benchmarks considered in this report and competitive English results. The models obtain leading results among the evaluated open models on OALL2 and AraGen. They also perform strongly on several culturally grounded Arabic benchmarks, including poetry, religion, cuisine, and dream interpretation, as well as in general tasks such as translation and summarization. We release the models in HuggingFace under a commercially permissive license. Jais 2 70B is also released as a chat app on the Web, iOS, and Android; it runs on Cerebras hardware, delivering up to 2,000 tokens per second, and enabling high-throughput Arabic-centric chat serving in our deployment setting. By uniting scale, linguistic diversity, cultural fidelity, openness, and speed, Jais 2 provides an open-weight foundation intended to support further research and development in Arabic-centric LLMs.

Chinese Translation

Jais 2 是由 MBZUAI、Cerebras 和 Inception 联合开发的以阿拉伯语为中心的大型语言模型家族，旨在推动以阿拉伯语为中心的语言建模，在本报告评估的阿拉伯语和文化基准上表现出色。该家族包括我们所知的从零开始训练的最大开放阿拉伯语中心 LLM，参数量达到 70B，以及在评估的开放模型中具有竞争力的 8B 参数变体。定制的以阿拉伯语为中心的词汇表使得训练和推理更加高效。此外，优化的架构和训练方案实现了高计算效率的训练。在与可比模型相比的情况下，Jais 2 在本报告考虑的基准上实现了强大的阿拉伯语表现，并在英语结果上也具有竞争力。这些模型在评估的开放模型中在 OALL2 和 AraGen 上取得了领先结果。在多个文化基准上表现出色，包括诗歌、宗教、美食和梦境解读，以及翻译和摘要等一般任务。我们将在 HuggingFace 上以商业许可发布这些模型。Jais 2 70B 也作为聊天应用程序在 Web、iOS 和 Android 上发布；它运行在 Cerebras 硬件上，每秒可处理高达 2000 个标记，能够在我们的部署环境中实现高吞吐量的以阿拉伯语为中心的聊天服务。通过结合规模、语言多样性、文化忠实性、开放性和速度，Jais 2 提供了一个开放权重的基础，旨在支持以阿拉伯语为中心的 LLM 的进一步研究和开发。

View on arXiv Download PDF AI Translation

cs.CL / 6 / 2608.13588

IterCOMP: Reasoning-aware Adaptive Prompt Compression for Multi-hop Question Answering

IterCOMP：面向推理的自适应提示压缩用于多跳问答

Yun, JungMin, Kim, YoungBin

Abstract

Multi-hop question answering requires complex reasoning across multiple evidence segments, which often overwhelms retrieval-augmented generation systems with lengthy and noisy contexts, thereby undermining both efficiency and accuracy. While existing prompt compression methods attempt to address this issue, they are typically designed for single-turn queries and fail to capture interdependent reasoning steps. We propose IterCOMP, a unified, training-free prompt compression framework that incorporates multi-hop reasoning within an iterative compression loop. IterCOMP decomposes documents into evidence segments, evaluates question answerability, and generates targeted follow-up questions to iteratively integrate essential evidence, producing a compact, reasoning-oriented prompt. Experiments on MusiQue, 2WikiMultiHopQA, and HotpotQA demonstrate that IterCOMP achieves substantial improvements in Exact Match and F1 scores while reducing the token budget, outperforming existing baselines and exhibiting robustness as reasoning complexity increases.

Chinese Translation

多跳问答需要在多个证据片段之间进行复杂推理，这常常使得增强检索的生成系统面临冗长且嘈杂的上下文，从而影响效率和准确性。虽然现有的提示压缩方法试图解决这一问题，但它们通常是为单轮查询设计的，无法捕捉相互依赖的推理步骤。我们提出了IterCOMP，一个统一的、无训练的提示压缩框架，结合了多跳推理并采用迭代压缩循环。IterCOMP将文档分解为证据片段，评估问题的可回答性，并生成针对性的后续问题，以迭代整合必要的证据，从而生成紧凑的、面向推理的提示。在MusiQue、2WikiMultiHopQA和HotpotQA上的实验表明，IterCOMP在精确匹配和F1分数上实现了显著提升，同时减少了标记预算，超越了现有基准，并在推理复杂性增加时表现出鲁棒性。

View on arXiv Download PDF AI Translation

cs.CL / 7 / 2608.13624

Measuring Fairness in Large Audio Language Models via Semantic-Aware Bias Estimation

通过语义感知偏差估计测量大型音频语言模型的公平性

Liu, Zhe

Abstract

Large Audio Language Models (LALMs) have seen increasing use for audio understanding tasks such as speech recognition and audio question answering, raising concerns about fairness across demographic subgroups. Fairness evaluation in spoken-input settings is challenging due to confounding factors, including semantic variation in spoken content and speaker-specific characteristics. Ignoring these factors can result in misleading conclusions about model bias. We propose a semantic-aware mixed-effects regression framework for fairness evaluation in LALMs that explicitly accounts for these confounders. Our approach incorporates sentence-level semantic embeddings of reference text as covariates and models speaker identity as a random effect. Notably, semantic representations are extracted from the same LALM under evaluation, enabling semantic control over variation as perceived by the model itself. Experiments on simulated data and real-world benchmarks demonstrate that the proposed approach substantially reduces spurious fairness findings and yields more robust and interpretable estimates of subgroup performance differences.

Chinese Translation

大型音频语言模型（LALMs）在语音识别和音频问答等音频理解任务中的应用日益增加，这引发了对不同人口子群体公平性的关注。在口语输入环境中进行公平性评估面临挑战，因为存在混杂因素，包括口语内容的语义变化和说话者特征。忽视这些因素可能导致对模型偏差的误导性结论。我们提出了一种语义感知的混合效应回归框架，用于在LALMs中进行公平性评估，明确考虑这些混杂因素。我们的方法将参考文本的句子级语义嵌入作为协变量，并将说话者身份建模为随机效应。值得注意的是，语义表示是从同一被评估的LALM中提取的，从而使模型本身感知的变化能够得到语义控制。在模拟数据和真实世界基准上的实验表明，所提出的方法显著减少了虚假的公平性发现，并提供了更稳健和可解释的子群体性能差异估计。

View on arXiv Download PDF AI Translation

cs.CL / 8 / 2608.13698

GRPO Beyond English: A Large-Scale Study of GRPO in Non-English and Multilingual Settings

超越英语的 GRPO：非英语和多语言环境下的 GRPO 大规模研究

Dobler, Konstantin, Scozzafava, Federico, Janke, Jonathan, Ali, Mohamed, Lehnerer, Simon

Abstract

Reinforcement Learning with Verifiable Rewards (RLVR), often optimized with Group Relative Policy Optimization (GRPO), has become a central recipe for improving the reasoning capabilities of pretrained language models but current studies remain heavily English-centric. We conduct a large-scale empirical study of multilingual and non-English GRPO across a wide range of base models, training languages, and different reasoning language rewards. We find that training to reason in the native language often leaves only a small gap to training for English reasoning. We further observe strong crosslingual transfer: training in one language often improves performance in many others. However, specific trends are highly model- and language-dependent. In some cases, training in a particular language induces severe regressions on out-of-domain capabilities in other languages. Our analysis shows that RLVR beyond English can provide broad crosslingual gains, but also requires broad evaluation to detect language-specific regressions.

Chinese Translation

可验证奖励的强化学习（Reinforcement Learning with Verifiable Rewards, RLVR），通常通过群体相对策略优化（Group Relative Policy Optimization, GRPO）进行优化，已成为提升预训练语言模型推理能力的核心方法，但目前的研究仍然以英语为中心。我们对多语言和非英语 GRPO 进行了大规模的实证研究，涵盖了广泛的基础模型、训练语言和不同的推理语言奖励。我们发现，在母语中进行推理的训练与英语推理的训练之间往往只有很小的差距。我们进一步观察到强烈的跨语言迁移：在一种语言中的训练通常会提高其他多种语言的表现。然而，特定的趋势高度依赖于模型和语言。在某些情况下，在特定语言中的训练会对其他语言的领域外能力造成严重的退化。我们的分析表明，超越英语的 RLVR 可以提供广泛的跨语言收益，但也需要广泛的评估以检测语言特定的退化。

View on arXiv Download PDF AI Translation

cs.CL / 9 / 2608.13706

CLAIR-Fin: An Adversarial Multi-Agent Framework for Claim-Level Verification and Adaptive Debate in Cross-Modal Financial QA

Faria, Fatema Tuj Johora, Moin, Mukaffi Bin, Mahmud, Jubayer Al, Mridha, M. F., Hossain, Md. Alam

Abstract

Existing defenses against hallucination in retrieval-augmented and multi-agent pipelines remain partial: evidence is trusted despite modality disagreement, debate verifies an aggregate report rather than individual claims, and such verification occurs only after drafting, leaving inter-agent errors undetected until the final text. To close this gap, we present CLAIR-Fin, a nine-agent framework that decomposes each question into atomic claims maintained in a typed Financial Claim Ledger. Each claim is resolved through Asymmetric Evidence Authority, which conditions evidence trust on claim type rather than treating all modalities as equally reliable; Chain-of-Custody Verification, which checks grounding at the hand-off between drafting and adversarial review rather than only at the pipeline's exit; an Adaptive Rebuttal Cycle, which routes contested claims through adversarial debate whose depth scales with what that debate finds; and a terminal entailment audit paired with a continuous Hallucination Risk Index that distinguishes claims that passed scrutiny from claims never contested. We evaluate CLAIR-Fin on BB-FinQA-X, a 500-question cross-modal financial evaluation set built from Bangladesh Bank Annual Report material, stratified by query type, format, and difficulty. Relative to a single-pass retrieval-augmented generation baseline, it raises faithfulness ($0.780 \rightarrow 0.889$) while abstaining on 5.4% of questions when evidence is insufficient rather than forcing an unsupported response, and it exceeds stronger retrieval-strategy baselines such as HyDE and Graph-RAG on faithfulness ($\leq 0.874$).

View on arXiv Download PDF AI Translation

cs.CL / 10 / 2608.13708

TeachMateGPT: A Multi-Agent Knowledge-Grounded Framework for Pedagogical Assessment Generation from Science Curriculum Materials

TeachMateGPT：一个基于多智能体的知识驱动框架，用于从科学课程材料生成教学评估

Faria, Fatema Tuj Johora, Moin, Mukaffi Bin, Mridha, M. F., Mahmud, Jubayer Al

Abstract

Automatically generating textbook-grounded assessment items can reduce science teachers' workload, but existing retrieval-augmented generation (RAG) systems rely on flat retrieval, support only single-question generation, lack safeguards against weak evidence, and are ill-suited to low-resource, board-exam-structured curricula. We address these limitations with TeachMateGPT, a multi-agent system contributing four advances to curriculum-grounded science-assessment authoring. (i) COPE, a hierarchical knowledge base replacing token-window chunking with a multi-resolution index that segments documents along syllabus structure and links them at three granularities via a traversable graph-based lineage, matching evidence to each topic's instructional level. (ii) A staged, fail-closed agent pipeline replacing one-shot retrieve-then-generate: routing gates search, retrieval fuses dense and lexical evidence under a coverage gate that withholds generation on insufficient evidence, and specialist agents draft objective and constructed-response items. (iii) SAVER, a source-attributed verification protocol scoring faithfulness, relevance, and hallucination risk against retrieved evidence, applying stricter grounding checks across each creative question's four sub-parts, paired with teacher-in-the-loop evaluation rather than automatic filtering. (iv) NCTB-SciGen8, a curriculum-grounded dataset of 198 items (143 multiple-choice, 55 creative questions) spanning all 14 chapters of the NCTB Class 8 science textbook, produced by the pipeline and rated by three practicing teachers. TeachMateGPT raises faithfulness (0.68 $\rightarrow$ 0.96) and answer relevancy (0.60 $\rightarrow$ 0.89) over a vanilla RAG baseline.

Chinese Translation

自动生成基于教材的评估题目可以减轻科学教师的工作负担，但现有的检索增强生成（RAG）系统依赖于平面检索，仅支持单题生成，缺乏对弱证据的保护措施，并且不适合低资源和结构化的考试课程。我们通过TeachMateGPT解决了这些局限性，这是一个多智能体系统，为基于课程的科学评估创作提供了四项进展。(i) COPE，一个层次化知识库，用多分辨率索引替代了基于标记窗口的分块，沿着教学大纲结构对文档进行分段，并通过可遍历的图状谱系在三个粒度上链接它们，将证据与每个主题的教学水平相匹配。(ii) 一个分阶段的、失败关闭的智能体管道，替代了一次性检索后生成的方式：路由门进行搜索，检索在覆盖门下融合密集和词汇证据，在证据不足时暂停生成，专业智能体起草客观和构建性回答题。(iii) SAVER，一个源归属的验证协议，对检索到的证据进行忠实性、相关性和幻觉风险的评分，在每个创意问题的四个子部分上应用更严格的基础检查，配合教师参与的评估，而非自动过滤。(iv) NCTB-SciGen8，一个基于课程的数据集，包含198个题目（143个选择题，55个创意题），涵盖NCTB八年级科学教材的14个章节，由该管道生成并由三位在职教师评分。TeachMateGPT在忠实性（0.68 $ ightarrow$ 0.96）和答案相关性（0.60 $ ightarrow$ 0.89）上超越了基础的RAG模型。

View on arXiv Download PDF AI Translation

cs.CL / 11 / 2608.13717

StreamHear: Domain-Adapted Pseudo-Labeling for Semi-Supervised Streaming Speech Recognition

StreamHear：用于半监督流式语音识别的领域适应伪标签生成

Liu, Zefang, Zhu, Chenyang, Cho, Sangwoo, Peng, Xujun, Zhang, Shi-Xiong, Sahu, Sambit

Abstract

Streaming automatic speech recognition (ASR) underperforms on domain-shifted target audio, where labeled in-domain data is costly to prepare while unlabeled audio is abundant. We present StreamHear, a semi-supervised pipeline that adapts a pretrained streaming student by fine-tuning an offline transducer teacher on the labeled training set, generating pseudo-labels on the unlabeled portion, and fine-tuning the student on the mixture. We further introduce a prior-regularized dynamic-programming realignment step that fixes chunk-level word placement using an ASR-hypothesis anchor. Across four datasets spanning financial calls, prepared read speech, and phone-quality dialogue, StreamHear consistently outperforms supervised student fine-tuning and narrows the gap to the offline teacher.

Chinese Translation

在领域转移的目标音频下，流式自动语音识别（ASR）的表现不佳，其中标注的领域内数据准备成本高昂，而未标注的音频却十分丰富。我们提出了StreamHear，一个半监督的流程，通过在标注训练集上微调预训练的流式学生模型，生成未标注部分的伪标签，并在混合数据上微调学生模型。我们进一步引入了一种先验正则化的动态规划重对齐步骤，该步骤利用ASR假设锚点修正了块级别的单词位置。在涵盖金融电话、准备好的朗读语音和电话质量对话的四个数据集上，StreamHear始终优于监督学生微调，并缩小了与离线教师之间的差距。

View on arXiv Download PDF AI Translation

cs.CL / 12 / 2608.13722

BM25-Augmented Many-Shot Translation for Low-Resource North-Eastern Indian Languages

基于BM25的多样本翻译在低资源东北印度语言中的应用

Dhawan, Aashish, Driggers-Ellis, Christopher, Kasinets, Dzmitry, Grant, Christan, Wang, Daisy Zhe

Abstract

This paper describes the University of Florida Gators submission to the WMT26 Low-Resource Indic Language Translation shared task. We adapt the retrieval-augmented many-shot translation pipeline from our AmericasNLP 2026 system to translate between English and eleven North-Eastern Indian languages in both directions. At inference time, BM25 retrieves the most similar parallel examples from a language-specific training bank, and Gemini 2.5 Flash translates the input conditioned on these examples. No model fine-tuning is involved. Training banks combine official WMT26 data with publicly available corpora such as Samanantar and prior WMT shared task releases. A grid search over retrieval count r and development exemplar count d across all 22 language-direction pairs selects the best configuration for each submission.

Chinese Translation

本文描述了佛罗里达大学Gators团队在WMT26低资源印度语言翻译共享任务中的提交。我们将从AmericasNLP 2026系统中适配的检索增强多样本翻译流程应用于英语与十一种东北印度语言之间的双向翻译。在推理阶段，BM25从特定语言的训练库中检索最相似的平行示例，而Gemini 2.5 Flash则基于这些示例对输入进行翻译。该过程不涉及模型微调。训练库结合了官方的WMT26数据和公开可用的语料库，如Samanantar以及之前的WMT共享任务发布的数据。通过对所有22个语言方向对的检索数量r和开发示例数量d进行网格搜索，选择每个提交的最佳配置。

View on arXiv Download PDF AI Translation

cs.CL / 13 / 2608.13741

GALA: Generation-Aware Cross-Modal Alignment for Text-to-Time-Series Synthesis

GALA：生成感知的跨模态对齐用于文本到时间序列的合成

Zhang, Haochen, Zhang, Gengwei, Yao, Laura, Knoz, Nicholas, Chen, Tianlong

Abstract

Synthesizing time series from natural language is emerging as the most expressive form of controllable time series generation. However, existing text-conditioned generators either take caption embeddings frozen from off-the-shelf text encoders, or adapt the encoder end-to-end, letting the denoising loss shape the embeddings only as a by-product. In either case, the conditioning representation is never deliberately matched to the signal modality, leaving it ill-suited to guide generation. We address this by introducing GALA: Generation-Aware cross-modaL Alignment for text conditional time series generation. GALA is a two-stage approach that first contrastively couples a pretrained text encoder with a time-series foundation model into a shared embedding space with both encoders adapted to generation by an auxiliary generative loss, and then freezes the resulting caption embedding to drive a flow-matching generator. On TSFragment-600K, spanning four domains and three fragment lengths, GALA sets a new state of the art, ranking first in 30 of 36 metric columns and reaching an average rank of 1.08/1.08/1.42 at lengths 24/48/96 against 1.92/2.00/1.75 for the strongest baseline. We further find that generator-internal text encoders force a trade-off between fidelity and caption adherence, whereas conditioning on the aligned embedding breaks it: FID, CTTP, and JFTSD all improve at once. Ablating the auxiliary loss degrades FID, CTTP and JFTSD together, it indicates the generative term is a necessary component of the alignment rather than an add-on.

Chinese Translation

从自然语言合成时间序列正在成为可控时间序列生成中最具表现力的形式。然而，现有的文本条件生成器要么使用来自现成文本编码器的固定标题嵌入，要么端到端地适应编码器，让去噪损失仅作为副产品来塑造嵌入。在这两种情况下，条件表示从未被故意匹配到信号模态，导致其不适合引导生成。我们通过引入GALA来解决这个问题：生成感知的跨模态对齐用于文本条件时间序列生成。GALA是一种两阶段的方法，首先对比性地将一个预训练的文本编码器与一个时间序列基础模型耦合到一个共享的嵌入空间中，两个编码器都通过辅助生成损失适应生成，然后冻结生成的标题嵌入以驱动流匹配生成器。在涵盖四个领域和三种片段长度的TSFragment-600K上，GALA设定了新的最先进水平，在36个指标列中排名前30，并在长度为24/48/96时分别达到平均排名1.08/1.08/1.42，而最强基线为1.92/2.00/1.75。我们进一步发现，生成器内部的文本编码器在保真度和标题遵循之间迫使权衡，而对齐嵌入的条件则打破了这种权衡：FID、CTTP和JFTSD同时改善。消融辅助损失会导致FID、CTTP和JFTSD的共同下降，这表明生成项是对齐的必要组成部分，而不是附加项。

View on arXiv Download PDF AI Translation

cs.CL / 14 / 2608.13760

Amplified Does Not Mean Predictive: Reasoning Behaviors in Thinking Models

放大并不意味着预测：思维模型中的推理行为

Nyandwi, Jean de Dieu, Mathur, Leena, Bisk, Yonatan, Hawkins, Robert, Neubig, Graham

Abstract

Which reasoning behaviors are associated with correct answers in reasoning models, and does reasoning-oriented training amplify those behaviors? This distinction is important because reasoning-oriented training can make traces look more deliberative without amplifying the behaviors most tied to model correctness. We quantify this mismatch with Behavioral Lift, a metric that measures how much correctness changes when a behavior is present versus absent in a model's reasoning trace. Across 15 models and 6 benchmarks spanning text-only and vision-language reasoning, we annotate 15,282 traces with a taxonomy whose core behaviors are defined for both LLM and VLM traces. We find evidence for an Amplification-Lift Gap, in which thinking models strongly amplify self-correction, hypothesis testing, and uncertainty acknowledgment, while the highest-lift behaviors are confidence calibration, knowledge alignment, and self-awareness. Confidence calibration is among the strongest positive signals of correctness in both modalities, yet is barely amplified; uncertainty acknowledgment is amplified by 3--7$\times$, yet is weakly or negatively associated with correctness. We find that reasoning-oriented training does not preferentially amplify the highest-Lift behaviors, motivating process-level objectives that reward calibrated and grounded reasoning rather than surface form alone.

Chinese Translation

哪些推理行为与推理模型中的正确答案相关，并且推理导向的训练是否会放大这些行为？这一区分非常重要，因为推理导向的训练可以使痕迹看起来更具深思熟虑性，而不一定放大与模型正确性最相关的行为。我们通过行为提升（Behavioral Lift）这一指标量化这种不匹配，该指标测量在模型的推理痕迹中，当某一行为存在与缺失时，正确性变化的程度。在涵盖文本和视觉语言推理的15个模型和6个基准测试中，我们对15,282个痕迹进行了注释，采用了一种分类法，其核心行为在LLM（大语言模型）和VLM（视觉语言模型）痕迹中均有定义。我们发现了放大-提升差距（Amplification-Lift Gap）的证据，其中思维模型强烈放大自我修正、假设检验和不确定性承认，而提升最高的行为是信心校准、知识对齐和自我意识。信心校准在两种模式中都是正确性的最强正信号之一，但几乎没有被放大；不确定性承认被放大了3到7倍，但与正确性之间的关联较弱或为负。我们发现，推理导向的训练并没有优先放大提升最高的行为，这激励了奖励经过校准和扎实推理的过程级目标，而不仅仅是表面形式。

View on arXiv Download PDF AI Translation

cs.CL / 15 / 2608.13835

When Lexical Change Misleads: Rethinking Dynamic Topic Model Evaluation with Traditional and LLM-Based Metrics

当词汇变化误导时：用传统和基于大语言模型的指标重新思考动态主题模型评估

James, Charu Karakkaparambil

Abstract

Dynamic topic models capture evolving word distributions, but traditional coherence metrics may fail when vocabulary changes while semantic meaning persists. We evaluate 120 topics from CoNTM and DLDA across NYT, DBLP, and arXiv, using three human annotators and Low, Medium, and High lexical-change categories. Traditional temporal coherence shows highly variable agreement with human judgments ($\rho$=-0.256 to 0.614). In contrast, LLM-based semantic similarity agrees strongly with human semantic judgments for CoNTM on NYT ($\rho$=0.609), DBLP ($\rho$=0.721), and arXiv ($\rho$=0.502), but is less consistent for DLDA. Lexical-change stratification reveals variation hidden by aggregate evaluation. We therefore advocate lexical-change-aware evaluation, jointly reporting traditional coherence and LLM-based semantic measures as complementary rather than interchangeable signals.

Chinese Translation

动态主题模型捕捉不断演变的词汇分布，但当词汇发生变化而语义意义保持不变时，传统的连贯性指标可能会失效。我们评估了来自 CoNTM 和 DLDA 的 120 个主题，涵盖 NYT、DBLP 和 arXiv，使用了三位人工注释者和低、中、高词汇变化类别。传统的时间连贯性与人工判断的高度变异性一致（$ ho$=-0.256 到 0.614）。相比之下，基于大语言模型的语义相似性在 NYT（$ ho$=0.609）、DBLP（$ ho$=0.721）和 arXiv（$ ho$=0.502）上与 CoNTM 的人工语义判断高度一致，但对于 DLDA 的一致性较差。词汇变化的分层揭示了聚合评估所隐藏的变异。因此，我们倡导关注词汇变化的评估，联合报告传统连贯性和基于大语言模型的语义度量，作为互补而非可互换的信号。

View on arXiv Download PDF AI Translation

cs.CL / 16 / 2608.13840

ASSERT: A Measurement Pipeline for GenAI Audits

ASSERT：生成性人工智能审计的测量管道

Fogliato, Riccardo, Palia, Abhinav, Wang, Xiawei, Sheng, Emily, Atalla, Chad, Garcia-Gathright, Jean, Pangakis, Nicholas, Tan, Sharman, Vann, Dan, Washington, Hannah, Dow, P. Alex, Elfardy, Heba, Wallach, Hanna, Atluri, Sandeep

Abstract

Audits of generative AI (GenAI) systems often summarize behavior as a reported rate: how often the audited system complies with policy. Researchers and stakeholders use that rate to compare systems, track regressions, and gate deployment. A reported rate reflects both the system under audit and the measurement choices behind it, so a change in the rate can leave it unclear whether the system or those choices moved. We introduce ASSERT, a specification-driven measurement pipeline for GenAI audits that ties each reported rate to a written specification of the measurement choices used to produce it. ASSERT helps draft a behavioral rubric and test cases, then runs the audit against a GenAI system and returns a reported rate. In a case study on conversational deception, we observe that the reported rate moves substantially with the dialogue setup, the simulated user, the judge, and the evidence bar for non-compliance. These measurement choices substantially change the reported rate and can reorder GenAI system rankings. Because each reported rate is tied to an explicit specification, differences across audits are easier to attribute and interpret.

Chinese Translation

生成性人工智能（GenAI）系统的审计通常将行为总结为一个报告率：被审计系统遵循政策的频率。研究人员和利益相关者使用该率来比较系统、跟踪回归和控制部署。报告率反映了被审计系统及其背后的测量选择，因此率的变化可能使得不清楚是系统还是这些选择发生了变化。我们引入了ASSERT，这是一种以规范驱动的GenAI审计测量管道，将每个报告率与用于生成该率的测量选择的书面规范联系起来。ASSERT帮助起草行为标准和测试用例，然后对GenAI系统进行审计并返回报告率。在对会话欺骗的案例研究中，我们观察到报告率随着对话设置、模拟用户、评审者和不合规证据标准的变化而显著波动。这些测量选择显著改变了报告率，并可能重新排序GenAI系统的排名。由于每个报告率都与明确的规范相关联，因此不同审计之间的差异更容易归因和解释。

View on arXiv Download PDF AI Translation

cs.CL / 17 / 2608.13854

Bootstrapping Niche Multilingual Code Translation via Reinforcement Learning with Execution-Based Verifiable Supervision

通过基于执行的可验证监督的强化学习引导小众多语言代码翻译

Yuki, Kouki, Zeng, Jie, Ogawa, Kyoko, Ikeda, Ryunosuke, Kobashi, Yohei, Kojima, Takeshi, Yamada, Ikuya, Iwasawa, Yusuke, Matsuo, Yutaka

Abstract

Code translation must preserve executable behavior across many programming languages, yet neural code translation has largely focused on a few popular languages such as C++, Java, and Python. This leaves a niche, many-to-many setting where parallel supervision is sparse, producing plausible but non-executable translations. We address this setting with preference-based reinforcement learning driven by execution-based supervision. Our pipeline firstly expands verifiable seed Python programs into a multilingual pool of execution-validated codes. Using the pool, a base LLM generates translation candidates across language pairs, which we label by their execution outcomes. The resulting preferences are used to train a reward model that scores cross-language translation quality. Finally, we optimize our base LLMs with GRPO over 600 directed language pairs (25 x 24) using the reward model as a signal. To evaluate the niche translation capability, we introduce HumanEval-X++, an execution-based benchmark that extends HumanEval-X to a broad many-to-many language space. We evaluate our approach using Qwen-3.5 4B and 9B models. On HumanEval-X++ and existing benchmarks, it yields consistent gains over the untrained baselines. In particular, the 4B model achieves an average improvement of 13% across all languages on HumanEval-X++, with a gain of 21% on mid-tier languages. Our study establishes a reliable approach of data generation, training, and benchmarking, paving the way toward further bootstrapping the quality of many-to-many translation for programming languages.

Chinese Translation

代码翻译必须在多种编程语言中保持可执行行为，然而神经代码翻译在很大程度上集中于少数流行语言，如 C++、Java 和 Python。这导致了一个小众的多对多设置，其中并行监督稀缺，产生了看似合理但不可执行的翻译。我们通过基于执行的监督驱动的偏好强化学习来解决这一设置。我们的流程首先将可验证的种子 Python 程序扩展为一个多语言的执行验证代码池。利用该池，基础大型语言模型（LLM）生成跨语言对的翻译候选，并根据其执行结果进行标注。由此产生的偏好用于训练一个奖励模型，以评分跨语言翻译质量。最后，我们使用奖励模型作为信号，通过 GRPO 优化我们的基础 LLM，涵盖 600 对定向语言对（25 x 24）。为了评估小众翻译能力，我们引入了 HumanEval-X++，这是一个基于执行的基准，扩展了 HumanEval-X 到广泛的多对多语言空间。我们使用 Qwen-3.5 4B 和 9B 模型评估我们的方法。在 HumanEval-X++ 和现有基准上，它相较于未训练的基线表现出一致的提升。特别是，4B 模型在 HumanEval-X++ 上实现了所有语言平均提高 13%，在中等语言上提高 21%。我们的研究建立了一种可靠的数据生成、训练和基准评估方法，为进一步引导编程语言多对多翻译质量的提升铺平了道路。

View on arXiv Download PDF AI Translation

cs.CL / 18 / 2608.13947

Scaling Creative Writing Beyond Story-Centric Data with Attribute-Guided Genre Expansion

超越以故事为中心的数据进行创意写作的属性引导体裁扩展

Chang, Hwan, Kim, Yongil, Yeen, Heuiyeen, Kim, Yireun, Lee, Jinsik, Lee, Hwanhee

Abstract

High-quality creative writing data for large language models (LLMs) remains dominated by story-centric data, limiting models' ability to follow the structural and functional conventions of diverse creative formats. We propose an attribute-guided genre expansion framework for scaling creative writing data beyond story generation. By separating thematic breadth from genre-form control, our framework leverages human-authored story prompts as diverse creative seeds, while utilizing manually curated genre attributes to enforce distinct structural, stylistic, and formatting conventions. We combine these to prompt strong LLMs for genre-faithful query-response pairs, which are then quality-filtered. Applying this framework, we construct the Multi-Genre Collection, a 50K-example corpus spanning 13 creative genres, including story, rap, lyrics, scripts, game design, character design, and other creative formats. Experiments across out-of-distribution writing benchmarks and held-out genre diagnostics demonstrate that models fine-tuned on our data consistently surpass not only base models and writing-specialized baselines, but also models trained on existing writing corpora. Genre-count ablations further indicate that controlled genre expansion, rather than story-centric scaling alone, is a key driver of robust creative writing capability.

Chinese Translation

高质量的创意写作数据在大型语言模型（LLMs）中仍然以故事为中心的数据为主，这限制了模型遵循多样化创意格式的结构和功能规范的能力。我们提出了一种属性引导的体裁扩展框架，以将创意写作数据扩展到故事生成之外。通过将主题广度与体裁形式控制分离，我们的框架利用人类创作的故事提示作为多样化的创意种子，同时利用手动策划的体裁属性来强制执行不同的结构、风格和格式规范。我们将这些结合起来，以提示强大的LLMs生成符合体裁的查询-响应对，然后进行质量过滤。应用该框架，我们构建了多体裁集合（Multi-Genre Collection），这是一个包含50,000个示例的语料库，涵盖13种创意体裁，包括故事、说唱、歌词、剧本、游戏设计、角色设计及其他创意格式。在分布外写作基准和保留体裁诊断的实验表明，基于我们的数据进行微调的模型不仅持续超越基础模型和写作专业基线，还超越了在现有写作语料库上训练的模型。体裁数量的消融实验进一步表明，受控的体裁扩展，而非单纯的以故事为中心的扩展，是增强创意写作能力的关键驱动因素。

View on arXiv Download PDF AI Translation

cs.CL / 19 / 2608.13959

Repair, Not Improvement: Decomposing Constrained Decoding in Tool-Call Abstention

修复，而非改进：工具调用弃权中的约束解码分解

Lee, Janghoon

Abstract

Function calling is what the recent accounting of constrained generation explicitly sets aside: it finds the decoder's contribution small for format constraints, then warns in its Section 7 against extrapolating where a constraint encodes a correctness requirement, and names function calling as one. Tool abstention is that case at its sharpest: an enum leaves the wording of an answer alone and narrows the set of answers there are, and declining to call anything is the first it drops. We measure the excluded case. Three conditions over one byte-identical prompt separate a grammar's two jobs: it fixes where generation stops as well as which tokens may be emitted. We evaluate open-weight models from 0.6B to 4B on matched English and Korean items, so the language comparison is made within item. Against an unconstrained decoder, prior work's contrast is negative on abstention in four of six cells with intervals excluding zero, worst -29.5 points, and positive with an interval excluding zero in none. The total is a sum with opposite signs: on the smallest model in Korean the stop token costs -20.0, the enum returns +19.5, and the two leave -0.5. What it recovers is form: of 698 abstentions repaired, 545 had no readable answer and 0 were judgements the scorer refused. On tool-needed items it is positive throughout; abstention leads because it is the preregistered measure, and the pooled number being kinder to the intervention makes moving to it worse rather than better. Both preregistered language claims fail.

Chinese Translation

函数调用是最近关于约束生成的报告明确排除的内容：它发现解码器对格式约束的贡献微小，然后在第七节警告不要推断约束编码了正确性要求的地方，并将函数调用列为其中之一。工具弃权是这一情况的最尖锐表现：一个枚举保持答案的措辞不变，并缩小了可能的答案集合，而拒绝调用任何内容是它首先放弃的。我们测量被排除的情况。三个条件在一个字节完全相同的提示上分离了语法的两个任务：它固定了生成停止的位置以及可能发出的标记。我们在匹配的英语和韩语项目上评估了从0.6B到4B的开放权重模型，因此语言比较是在项目内部进行的。与无约束解码器相比，先前工作的对比在六个单元中的四个上对弃权的影响为负，区间不包括零，最差为-29.5分，而在没有任何区间不包括零的情况下为正。总和是一个符号相反的和：在韩语中最小的模型中，停止标记的成本为-20.0，枚举返回+19.5，两者合计为-0.5。它恢复的是形式：在698个修复的弃权中，545个没有可读答案，0个是评分者拒绝的判断。在需要工具的项目中，结果始终为正；弃权是因为它是预注册的测量，而汇总的数字对干预更为宽容，使得转向它的效果变得更糟而非更好。两个预注册的语言声明都未能成立。

View on arXiv Download PDF AI Translation

cs.CL / 20 / 2608.14003

Batch-wise Adaptive Pruning: Periodic Neuron Activation-Aware Weight Pruning for Language Reasoning Model

批量自适应剪枝：面向语言推理模型的周期性神经元激活感知权重剪枝

Kim, Yongmin, Takashiro, Shota, Iwasawa, Yusuke, Kojima, Takeshi, Matsuo, Yutaka

Abstract

Large Reasoning Models (LRMs) achieve strong performance on complex tasks through extended chain-of-thought generation, but incur substantial computational costs during inference. In production settings, batched inference is essential for high throughput, yet the existing training-free adaptive pruning methods we evaluate severely degrade in this regime. Because a batch must share a single pruning mask, these methods aggregate activations across samples and then apply threshold-based selection; the threshold, calibrated offline on unaggregated activations, no longer matches the aggregated distribution, so the realized sparsity ratio drifts and accuracy on reasoning tasks collapses under batched inference. In this work, we propose a training-free adaptive pruning method designed specifically for batched inference in LRMs, built on two components. First, we replace threshold-based selection with periodic top-k selection over the aggregated importance scores, which is unaffected by the shift that aggregation induces in the activation distribution, and which runs selection once per update period rather than at every token, preserving the speedup. Second, based on the observation that important neurons re-fire periodically during long reasoning generation, we introduce an activation memory that accumulates importance across update phases so that recurring neurons are retained. Experiments on diverse reasoning benchmarks demonstrate that our method outperforms the previous state-of-the-art adaptive pruning method by 39.7 percentage points in average accuracy at batch size 4 with 50% target sparsity on DeepSeek-R1-Distill-Qwen-7B, and reaches 1.40x speedup over dense inference at 50% actual sparsity.

Chinese Translation

大型推理模型（LRMs）通过扩展的思维链生成在复杂任务上取得了强大的性能，但在推理过程中却产生了巨大的计算成本。在生产环境中，批量推理对于高吞吐量至关重要，然而我们评估的现有无训练自适应剪枝方法在这一环境中严重退化。由于一个批次必须共享一个剪枝掩码，这些方法会在样本之间聚合激活值，然后应用基于阈值的选择；而阈值是在未聚合激活值上离线校准的，因此不再与聚合后的分布匹配，从而导致实现的稀疏率漂移，并且在批量推理下推理任务的准确性崩溃。在本研究中，我们提出了一种专门为LRMs的批量推理设计的无训练自适应剪枝方法，该方法基于两个组件。首先，我们用周期性top-k选择替代基于阈值的选择，这种选择基于聚合的重要性评分，不受聚合对激活分布造成的偏移影响，并且每个更新周期仅运行一次选择，而不是在每个标记上进行，从而保持了加速效果。其次，基于重要神经元在长时间推理生成过程中周期性重新激活的观察，我们引入了一种激活记忆，在更新阶段累积重要性，以便保留重复出现的神经元。在多样的推理基准上的实验表明，我们的方法在DeepSeek-R1-Distill-Qwen-7B上以批量大小为4、目标稀疏率为50%时，平均准确率比之前的最先进自适应剪枝方法提高了39.7个百分点，并且在实际稀疏率为50%时，相较于密集推理实现了1.40倍的加速。

View on arXiv Download PDF AI Translation

cs.CL / 21 / 2608.14029

S2Dialog: Multimodal Dialogue Retrieval with Semantic and Acoustic-Style Modeling

S2Dialog：基于语义和声学风格建模的多模态对话检索

Wang, Xueqi, Wang, Zhigang, Zhang, Runqing, Jia, Zhenqi, Zhao, Junfeng

Abstract

Multimodal dialogue retrieval aims to retrieve dialogues from multimodal dialogue banks that are similar to a target dialogue in terms of both textual semantics and acoustic conversational styles. Such dialogue-level retrieval is crucial for many dialogue-related tasks, including Emotion Recognition in Conversation, Spoken Dialogue Systems, and Conversational Speech Synthesis, where external dialogue examples can provide valuable semantic and stylistic references. However, existing retrieval methods are still largely limited to utterance-level or unimodal matching, and often fail to capture the global semantic coherence and stylistic consistency of an entire dialogue. To address this gap, we propose S2Dialog, a unified framework for dialogue-level semantic-style retrieval from multimodal dialogue banks. Specifically, S2Dialog consists of a Dialogue-level Textual Retriever and a Dialogue-level Acoustic Retriever, which encode the textual and acoustic modalities of a dialogue into dialogue-level representations, respectively. To further enhance multimodal retrieval, we introduce Dialogue-level Textual-Acoustic Contrastive Learning, which aligns semantically and stylistically similar dialogues while distinguishing unrelated ones. Extensive experiments on the multimodal dialogue dataset DailyTalk demonstrate that S2Dialog achieves outstanding retrieval performance.

Chinese Translation

多模态对话检索旨在从多模态对话库中检索与目标对话在文本语义和声学对话风格上相似的对话。这种对话级检索对于许多与对话相关的任务至关重要，包括对话中的情感识别、口语对话系统和对话语音合成，其中外部对话示例可以提供有价值的语义和风格参考。然而，现有的检索方法仍然主要局限于发言级或单模态匹配，往往无法捕捉整个对话的全局语义连贯性和风格一致性。为了解决这一问题，我们提出了S2Dialog，一个用于从多模态对话库中进行对话级语义-风格检索的统一框架。具体而言，S2Dialog由对话级文本检索器和对话级声学检索器组成，分别将对话的文本和声学模态编码为对话级表示。为了进一步增强多模态检索，我们引入了对话级文本-声学对比学习，该方法在对语义和风格相似的对话进行对齐的同时，区分不相关的对话。在多模态对话数据集DailyTalk上的大量实验表明，S2Dialog实现了卓越的检索性能。

View on arXiv Download PDF AI Translation

cs.CL / 22 / 2608.14055

HERMES: a multi-agent framework for structured knowledge extraction from ultra-long documents in geoscience

HERMES：一个用于地球科学超长文档结构化知识提取的多智能体框架

Song, Ziqi, Xiang, Zongyuan, Ogg, James G., Lieberman, Bruce S., Ogg, Gabi, Carranza, Natalia López, Du, Wen, Ye, Yufei, Li, Shuan, Peng, Zhong, Yu, Shaoqi, Wei, Juye, Zhou, Ying, Ye, Jieping, Yang, Jiang

Abstract

Authoritative scientific knowledge in geoscience remains largely trapped in legacy monographs and historical literature, where unstructured text and complex layouts hinder computational access. We introduce HERMES, a scalable multi-agent framework that extracts structured data from ultra-long scientific documents. Using a coordinating large language model, HERMES integrates domain constraints, validation rules and evidence tracing within a unified document-level extraction process that incorporates parsed text, tables, figures and captions. Applied to the 55-volume Treatise on Invertebrate Paleontology, the system produced a structured database of 32,277 fossil taxonomic entities and 451,878 attributes, released online at https://treatise.geolex.org. Extraction performance remained stable across fossil groups (average F1 scores of approximately 0.90 for entities and 0.91 for attributes), improving per-volume efficiency approximately sixfold relative to the tested fully manual baseline. Evaluation in palaeomagnetism and geochemistry, conducted without additional model training, demonstrated transfer across distinct geoscience domains. This work provides a practical pathway to transform historical scientific literature into FAIR-oriented structured data, offering a sustainable infrastructure for data-intensive disciplines and large-scale knowledge integration.

Chinese Translation

地球科学中的权威科学知识在很大程度上仍然被困于传统专著和历史文献中，这些文献中的非结构化文本和复杂布局妨碍了计算访问。我们介绍了HERMES，一个可扩展的多智能体框架，用于从超长科学文档中提取结构化数据。HERMES利用协调的大型语言模型，将领域约束、验证规则和证据追踪整合到一个统一的文档级提取过程中，该过程结合了解析文本、表格、图形和说明。应用于《无脊椎动物古生物学专论》的55卷本，该系统生成了一个包含32,277个化石分类实体和451,878个属性的结构化数据库，并在线发布于https://treatise.geolex.org。提取性能在化石组之间保持稳定（实体的平均F1分数约为0.90，属性的平均F1分数约为0.91），相对于测试的完全手动基线，逐卷效率提高了大约六倍。在没有额外模型训练的情况下进行的古地磁学和地球化学评估，展示了在不同地球科学领域之间的迁移能力。这项工作提供了一条将历史科学文献转化为面向FAIR的结构化数据的实用路径，为数据密集型学科和大规模知识整合提供了可持续的基础设施。

View on arXiv Download PDF AI Translation

cs.CL / 23 / 2608.14079

The conditional superiority of fast silicon sampling

快速硅采样的条件优越性

Lam, Nickolas Hock Yuen, Voo, Ji Xuan, Ma, Xiangyu

Abstract

Silicon sampling can produce surprisingly good population estimates at times. Does doing it fast attenuate such fidelity? In this study, we extend and assess ongoing work in silicon sampling by comparing the algorithmic fidelity of "fast" and "slow" modes of silicon sampling among a nationally representative sample of Singaporean survey respondents. We find that silicon sampling with contemporary frontier models remains a method in early development to be used only with great caution. While silicon samples are able to produce moderately faithful estimates of population means, they continue to understate opinion variance and distort the latent contextual space behind human opinions. Conditional on such limitations, we find "fast" modes of silicon sampling to be relatively superior to traditional "slow" modes of silicon sampling. Fast silicon sampling is significantly more efficient in compute resources and run-time while being monotonically superior to slower modes of sampling in algorithmic fidelity.

Chinese Translation

硅采样在某些情况下能够产生令人惊讶的良好人口估计。快速进行采样是否会削弱这种准确性？在本研究中，我们扩展并评估了硅采样的持续研究，通过比较新加坡全国代表性样本调查受访者中“快速”和“慢速”模式的算法准确性。我们发现，使用当代前沿模型的硅采样仍然是一种处于早期发展的方法，仅应谨慎使用。尽管硅样本能够产生对人口均值的中等忠实估计，但它们仍然低估了意见方差，并扭曲了人类意见背后的潜在情境空间。在这些限制条件下，我们发现“快速”模式的硅采样相对于传统的“慢速”模式具有相对优越性。快速硅采样在计算资源和运行时间上显著更高效，同时在算法准确性上也单调优于较慢的采样模式。

View on arXiv Download PDF AI Translation

cs.CL / 24 / 2608.14150

Leading-Silence Augmentation and Multi-Stage Synthetic Supervision for the Second MLC-SLM Challenge

领先静音增强与多阶段合成监督用于第二届多语言对话语音语言模型挑战

Shi, Kexin, Sun, Renhe, Huang, Yuge, Wang, Ximeng, Zhou, Jiayi, Liu, Jian, Zhang, Malu

Abstract

The second Multilingual Conversational Speech Language Model (MLC-SLM) Challenge evaluates two tasks over complete, unsegmented multilingual conversations: speaker diarization and recognition (Task 1) and conversational speech understanding (Task 2). Neither task provides oracle utterance boundaries or speaker labels at evaluation, and Task 2 provides no question-answer training set. For Task 1, we fine-tune VibeVoice-ASR-7B with random leading-silence cropping, consistent timestamp correction, and an exponential moving average (EMA) training strategy. For Task 2, we construct synthetic question-answer pairs through multimodal candidate generation, silent-audio filtering, and distribution-matched augmentation, and fine-tune Qwen3-Omni-30B-A3B-Instruct for tagged direct answering. On the Task 1 evaluation set, cropping reduces tcpMER from 18.30% to 17.27%, and EMA further reduces it to 16.73%. On the Task 2 evaluation set, jointly applying distribution-matched augmentation and tagged direct answering raises accuracy from 83.0% to 86.0%.

Chinese Translation

第二届多语言对话语音语言模型（MLC-SLM）挑战评估两个任务，涉及完整的、未分段的多语言对话：说话人分离与识别（任务1）和对话语音理解（任务2）。在评估中，两个任务均不提供真实的发言边界或说话人标签，且任务2没有问题-答案训练集。对于任务1，我们通过随机的领先静音裁剪、一致的时间戳校正和指数移动平均（EMA）训练策略对VibeVoice-ASR-7B进行微调。对于任务2，我们通过多模态候选生成、静音音频过滤和分布匹配增强构建合成的问题-答案对，并对Qwen3-Omni-30B-A3B-Instruct进行标记直接回答的微调。在任务1评估集上，裁剪将tcpMER从18.30%降低到17.27%，而EMA进一步将其降低至16.73%。在任务2评估集上，联合应用分布匹配增强和标记直接回答将准确率从83.0%提高到86.0%。

View on arXiv Download PDF AI Translation

cs.CL / 25 / 2608.14210

How Much Do Legal RAG Systems Still Hallucinate?

法律检索增强生成系统的幻觉现象有多严重？

Das, Souvick, Abualhaija, Sallam, Bianculli, Domenico

Abstract

Hallucination is a major challenge for retrieval-augmented generation (RAG) systems in the legal domain, where ungrounded answers can lead to serious consequences. To better understand this problem, we conduct a fine-grained analysis of hallucination behavior in eight legal RAG systems across two legal corpora, the GDPR (in English) and a national civil law (in French). Using claim-level and answer-level evaluation, we report on hallucination density and severity, analyze performance across question categories and user personas, and validate our findings on an independent set of 142 legal-expert-authored questions. Our results show that hallucinations remain pervasive, ranging from less than 10% of responses for the best-performing systems to nearly half in the worst case. We further find that false-premise questions, containing incorrect assumptions that must be rejected, produce high hallucination rates on the manually-drafted questions.

Chinese Translation

幻觉现象是法律领域检索增强生成（RAG）系统面临的主要挑战之一，因为无根据的答案可能导致严重后果。为了更好地理解这一问题，我们对八个法律 RAG 系统在两个法律语料库中的幻觉行为进行了细致分析，分别为 GDPR（英文）和国家民法（法文）。通过对声明级和答案级的评估，我们报告了幻觉的密度和严重性，分析了不同问题类别和用户角色的表现，并在一个独立的142个法律专家撰写的问题集上验证了我们的发现。我们的结果显示，幻觉现象依然普遍，表现为最佳系统的响应中幻觉比例低于10%，而最差情况下接近一半。我们进一步发现，包含必须被拒绝的错误假设的错误前提问题，在手动起草的问题中产生了较高的幻觉率。

View on arXiv Download PDF AI Translation

cs.CL / 26 / 2608.14229

The More Popular, The Harder to Forget: Adaptive Popularity for LLM Unlearning

越受欢迎，越难以遗忘：针对大规模语言模型的自适应流行度遗忘方法

Borisiuk, Anna, Savchenko, Andrey, Panchenko, Alexander, Tutubalina, Elena

Abstract

Popular facts are memorised more deeply during pretraining and resist removal longer than rare ones, yet existing LLM unlearning methods apply uniform gradient pressure regardless of training-data frequency. We propose the AdaPop (Adaptive Popularity) method, which combines local token confidence with a per-fact popularity-dependent exponent derived from an external proxy (e.g., Wikidata sitelinks, LLM-as-Judge), and automates the forget-retain balance via a dual-ascent controller that adjusts the retain penalty each epoch. Across three model families and two benchmarks, AdaPop leaks ~5x less forgotten content than competing methods under paraphrased queries and ~1.6x less under adversarial reformulations. We support our analysis with internal metrics: under our method, forget-set hidden states move further from the pre-unlearning model's states than under other methods, while retain-set representations remain close.

Chinese Translation

在预训练过程中，流行事实被记忆得更深，并且比稀有事实更难以删除，然而现有的大规模语言模型（LLM）遗忘方法对训练数据频率施加统一的梯度压力。我们提出了AdaPop（自适应流行度）方法，该方法结合了局部标记置信度和基于外部代理（例如，Wikidata 站点链接、LLM-as-Judge）得出的每个事实的流行度相关指数，并通过双上升控制器自动化遗忘与保留的平衡，该控制器在每个训练周期调整保留惩罚。在三个模型系列和两个基准测试中，AdaPop在改写查询下泄漏的遗忘内容比竞争方法少约5倍，在对抗性重构下少约1.6倍。我们通过内部指标支持我们的分析：在我们的方法下，遗忘集的隐藏状态比其他方法下的预遗忘模型状态更远，而保留集的表示则保持接近。

View on arXiv Download PDF AI Translation

cs.CL / 27 / 2608.14277

SimpleOPD: Simple Tokenizer-Agnostic On-Policy Distillation for Long-Context Reasoning

SimpleOPD：简单的与标记器无关的在线蒸馏用于长上下文推理

He, Haonan, Lei, Haodi, Luo, Yun, Zhang, Haoran, Zhang, Shunkai, Li, Yizhuo, Tang, Shengji, Wang, Zhilin, Zhan, Runzhe, Bai, Lei, Cui, Ganqu, Yu, Fangchen, Li, Yafu, Ye, Peng, Ding, Ning, Cheng, Yu

Abstract

On-policy distillation (OPD) offers a promising way to transfer reasoning capabilities from stronger teacher models, but applying it to long-context reasoning teachers and short-context students introduces practical challenges, including tokenizer mismatch, teacher-student distribution mismatch, response length explosion, and training instability. In this work, we study this setting by transferring proof-reasoning capabilities from the long-context reasoning model SU-01 to short-context student models. To handle tokenizer differences, we perform OPD in a shared text space and align only tokens that occupy identical text spans under the student and teacher tokenizers. To mitigate the problem of excessive generation length and frequent truncation, we introduce a student reference KL loss and mask the advantages of special termination tokens such as and <|im_end|>. This strategy constrains the student from drifting excessively from its initial policy, thereby mitigating the teacher-student distribution mismatch problem and fostering steady length growth. Experiments on both same-family and different-family student models, including Qwen3, Qwen3.5, Intern-S2, GLM-4.7, Gemma-4, show consistent gains in mathematical reasoning, especially natural-language math proving. Notably, Intern-S2-Preview improves by 21.2 points on ProofBench, reaching 55.2 and surpassing Gemini-2.5-Pro. It also improves on science benchmarks such as HLE and HiPhO, suggesting that OPD transfers reasoning capabilities that generalize beyond the mathematical training domain.

Chinese Translation

在线蒸馏（On-policy distillation, OPD）为从更强大的教师模型转移推理能力提供了一种有前景的方法，但将其应用于长上下文推理教师和短上下文学生时引入了一些实际挑战，包括标记器不匹配、教师-学生分布不匹配、响应长度爆炸和训练不稳定。在本研究中，我们通过将证明推理能力从长上下文推理模型SU-01转移到短上下文学生模型来研究这一设置。为了处理标记器差异，我们在共享文本空间中执行OPD，并仅对在学生和教师标记器下占据相同文本跨度的标记进行对齐。为了缓解生成长度过长和频繁截断的问题，我们引入了学生参考KL损失，并屏蔽了特殊终止标记（如和<|im_end|>）的优势。这一策略限制了学生过度偏离其初始策略，从而缓解了教师-学生分布不匹配的问题，并促进了长度的稳定增长。在同一系列和不同系列的学生模型（包括Qwen3、Qwen3.5、Intern-S2、GLM-4.7、Gemma-4）上的实验显示，在数学推理，尤其是自然语言数学证明方面，均取得了一致的提升。值得注意的是，Intern-S2-Preview在ProofBench上提高了21.2分，达到了55.2，超越了Gemini-2.5-Pro。在科学基准（如HLE和HiPhO）上也有所提升，这表明OPD转移的推理能力超越了数学训练领域的泛化能力。

View on arXiv Download PDF AI Translation

cs.CL / 28 / 2608.14312

Envs-FORGE: Frontier-Optimized Reward-Grounded Environment Synthesis for Agent RL

Envs-FORGE：面向前沿优化的基于奖励的环境合成用于代理强化学习

Wu, Xiaojun, Yang, Cehao, Liu, Honghao, Lin, Xueyuan, Shi, Zhichao, Zhou, Hao, Jiang, Xuhui, Xu, Chengjin, Li, Jia, Guo, Jian

Abstract

Reinforcement learning (RL) for terminal agents needs executable training environments with reliable rewards and useful difficulty. Fixed recipes such as few-shot, Self-Instruct, and Evol-Instruct apply the same prompting policy to every seed, even when the current policy would benefit from a harder, easier, or simply different task. We present Envs-FORGE, a prompting policy that converts verifier rewards into per-seed environment-synthesis actions. Envs-FORGE estimates seed pass rates, scores six projection--direction actions around a target learning frontier, and solves a per-seed mixed-integer linear program (MILP) to choose the action that conditions generation. The selected action drives synchronized rewriting of the instruction, fixtures, oracle solution, tests, and Docker environment; only gold-verified bundles enter RL training. The indexed MILP form also supports optional soft skill coverage for portfolio planning. On Qwen 3.5 35B, Envs-FORGE improves Pass@1 over Base by 9.2 percentage points on tb-core (40.0% to 49.2%) and 6.4 points on tb-2.0 (23.0% to 29.4%), exceeding the strongest fixed-recipe baseline by 2.4 and 2.1 points. It reaches 77.1% on SWE-bench Verified versus 73.4% for Base, and improves tb-core by 6.8--9.2 points across the evaluated 4B--35B models. All synthesis methods export 100 verified environments and use 2.27M--2.88M synthesis tokens, placing the comparison at the same downstream training-set size and the same operational scale. The source code is available at https://github.com/DataArcTech/DataArc-SynData-Toolkit/.

Chinese Translation

针对终端代理的强化学习（RL）需要可执行的训练环境，这些环境具有可靠的奖励和有用的难度。固定的配方如少量样本（few-shot）、自我指导（Self-Instruct）和进化指导（Evol-Instruct）对每个种子应用相同的提示策略，即使当前策略可能从更难、更简单或完全不同的任务中受益。我们提出了Envs-FORGE，这是一种提示策略，将验证者奖励转换为每个种子的环境合成动作。Envs-FORGE估计种子的通过率，围绕目标学习前沿评分六个投影方向的动作，并解决每个种子的混合整数线性规划（MILP），以选择条件生成的动作。所选动作驱动指令、固定件、oracle解决方案、测试和Docker环境的同步重写；只有经过金标准验证的包才能进入RL训练。索引的MILP形式还支持投资组合规划的可选软技能覆盖。在Qwen 3.5 35B上，Envs-FORGE在tb-core上将Pass@1提高了9.2个百分点（从40.0%提高到49.2%），在tb-2.0上提高了6.4个百分点（从23.0%提高到29.4%），超过了最强的固定配方基线2.4和2.1个百分点。在SWE-bench Verified上达到77.1%，而Base为73.4%，并在评估的4B-35B模型中将tb-core提高了6.8-9.2个百分点。所有合成方法导出了100个经过验证的环境，并使用了2.27M-2.88M的合成令牌，使比较在相同的下游训练集规模和相同的操作规模下进行。源代码可在https://github.com/DataArcTech/DataArc-SynData-Toolkit/获取。

View on arXiv Download PDF AI Translation

cs.CL / 29 / 2608.14361

Local and Global Regimes of Geometric Complexity in Language Model Representations

语言模型表征中的局部与全局几何复杂性机制

Osman, Arwa, Baroni, Marco, Macocco, Iuri

Abstract

Intrinsic dimensionality (ID) is widely used to probe the representational complexity of language models, but it remains unclear whether ID differences reflect properties of language itself or artefacts of how the underlying dataset was constructed. In this paper, we focus specifically on how lexical diversity, the number of unique last-token items present in a dataset, affects ID estimates of that dataset. We find a scale-dependent transition between two regimes: at low lexical diversity, conditions with fewer unique final words produce higher ID, while at high lexical diversity, this ordering reverses, and conditions with more unique words produce higher ID. We derive an exact, parameter-free formula for the point at which this reversal occurs, which matches the observed transition point at every scale tested. On the one hand, our results highlight how care must be taken when interpreting the intrinsic dimensionality of a set of representations as a straightforward cue of their complexity. On the other hand, our discovery of the two ID regimes reveals a general principle of organisation of linguistic data in LLMs that sheds new light on their inner manifold structures.

Chinese Translation

内在维度（ID）被广泛用于探测语言模型的表征复杂性，但尚不清楚ID差异是否反映了语言本身的特性，还是构建基础数据集时的伪影。本文专注于词汇多样性，即数据集中存在的独特最后词项的数量，如何影响该数据集的ID估计。我们发现了两个机制之间的尺度依赖性转变：在低词汇多样性下，具有较少独特最终词的条件产生更高的ID，而在高词汇多样性下，这一排序则反转，具有更多独特词的条件产生更高的ID。我们推导出一个精确的、无参数的公式，用于确定这一反转发生的点，该公式与每个测试尺度下观察到的转变点相匹配。一方面，我们的结果强调在将一组表征的内在维度解读为其复杂性的直接线索时必须谨慎。另一方面，我们发现的两个ID机制揭示了LLMs中语言数据组织的一般原则，为其内在流形结构提供了新的视角。

View on arXiv Download PDF AI Translation

cs.CL / 30 / 2608.14377

A Survey of Large Models in Sports

体育领域大型模型的调查

Xu, Yichen, Ma, Jianzhe, Wang, Chuhan, Cao, Zhonghao, Chen, Liangyu, Wang, Wenxuan, Jin, Qin

Abstract

Sports have witnessed growing global enthusiasm in recent years, serving as a vital force for physical health, cultural exchange, social connection, and economic growth. The rapid advancement of large models, particularly (multimodal) large language models (M)LLMs, has demonstrated transformative potential to reshape sports understanding, analysis, and interaction across diverse domains. This paper presents a comprehensive survey of large models in sports, including (i) an overview of tasks and applications across different participant groups; (ii) a detailed analysis of sports-related datasets and benchmarks; and (iii) a critical discussion of current challenges and future directions. Our goal is to establish a foundation for advancing research and practical development of large-model-driven sports intelligence. An open-source GitHub repository is maintained at: https://github.com/Road2Redemption/Awesome_Large_Models_In_Sports1.

Chinese Translation

近年来，体育在全球范围内的热情日益高涨，成为促进身体健康、文化交流、社会联系和经济增长的重要力量。大型模型的快速发展，特别是（多模态）大型语言模型（M）LLMs，展现出重塑体育理解、分析和互动的变革潜力，涵盖多个领域。本文对体育领域的大型模型进行了全面调查，包括（i）不同参与者群体的任务和应用概述；（ii）与体育相关的数据集和基准的详细分析；以及（iii）对当前挑战和未来方向的批判性讨论。我们的目标是为推动大型模型驱动的体育智能的研究和实际发展奠定基础。一个开源的GitHub代码库可在以下链接访问：https://github.com/Road2Redemption/Awesome_Large_Models_In_Sports1。

View on arXiv Download PDF AI Translation

cs.CL / 31 / 2608.14457

Information Satisfaction: A Reader-Centered Axis for Summarization Evaluation

信息满意度：以读者为中心的摘要评估轴心

Cachola, Isabel, Walden, William, Kriz, Reno, Dredze, Mark

Abstract

The majority of work on summarization evaluation focuses on general summary quality (e.g., ROUGE, BERTScore) or specific desired properties (e.g., readability, factuality). However, these metrics fail to measure the utility of a summary to an individual user. For example, a biomedical researcher learning about the latest vaccine research will have different informational needs from a family doctor. Query-focused summarization captures part of this need, but in practice, users rarely state everything relevant in a query: a single short query is likely inadequate to distinguish the needs of a researcher from those of a physician. By contrast, a reader's background or persona (their role and expertise) is comparatively stable across queries and recovers much of this missing context, which makes it a practical signal for assessing whether a summary satisfies that reader's needs. In this work, we assess how sensitive popular summarization metrics are to both informational and persona differences, and find that many popular metrics, including strong LLM-as-judge metrics, fail basic perturbation tests of informational content. We additionally conduct an expert human evaluation, measuring summary preferences based on information satisfaction given a specific person's background and use case. We find that both traditional and LLM-based metrics are insufficient measures of information satisfaction and agree poorly with human judgment.

Chinese Translation

大多数关于摘要评估的研究集中在一般摘要质量（例如，ROUGE、BERTScore）或特定期望属性（例如，可读性、事实性）上。然而，这些指标未能衡量摘要对个体用户的实用性。例如，一位生物医学研究人员了解最新疫苗研究的信息需求与家庭医生的需求不同。基于查询的摘要捕捉了部分这种需求，但在实际中，用户很少在查询中陈述所有相关信息：一个简短的查询往往不足以区分研究人员与医生的需求。相比之下，读者的背景或角色（他们的角色和专业知识）在不同查询中相对稳定，并恢复了许多缺失的上下文，这使其成为评估摘要是否满足该读者需求的实用信号。在本研究中，我们评估了流行摘要指标对信息和角色差异的敏感性，发现许多流行指标，包括强大的LLM作为评判指标，未能通过信息内容的基本扰动测试。此外，我们还进行了专家人工评估，根据特定个体的背景和使用案例测量摘要偏好。我们发现，无论是传统指标还是基于LLM的指标，都不足以衡量信息满意度，并且与人类判断的吻合度较低。

View on arXiv Download PDF AI Translation

cs.CL / 32 / 2608.14465

You Only Pass Once: Answering and Abstaining Together in a Single Forward Pass of a Frozen Language Model

你只需通过一次：在冻结语言模型的单次前向传递中同时回答和放弃

Luo, Ziyang, Chu, Zhongyao, He, Xinjie, Wang, Youting, Qin, Xukui, Wu, Runxiong, Chen, Yan-Syuan

Abstract

A frozen language model on reasoning tasks has two coupled weaknesses: it under-uses evidence its own residual stream already encodes, and it fails to detect when the input is insufficient to answer, so it confabulates. This paper consolidates two research lines that address these on the same residual stream: a conditional steering probe writes the stream at mid-stack layers and recovers reasoning accuracy from a frozen backbone, and a zero-shot sufficiency direction reads the stream and abstains when information is insufficient. Deployed in one forward pass they interfere: the steering write shifts the state the direction reads, costing up to 8 AUROC points of cross-domain transfer on small models; a separate clean pass doubles inference cost. We keep the direction fixed and train a small network to reconstruct the pre-steering residual from the steered one -- mean-squared error on (steered, clean) pairs, no sufficiency labels -- and read the direction on the reconstruction. The resulting system, YOPO (You Only Pass Once), answers, steers, and abstains in one forward pass of a frozen Qwen2.5 backbone (1.5B/3B/7B). End to end, three-way accuracy more than doubles the frozen baseline (0.375->0.798 on 1.5B alphaNLI) and one pass beats the two-pass reference at every scale (0.798/0.830/0.893 vs 0.753/0.790/0.863) and on ten backbones across six model families. We chart the capacity-transfer frontier quantifying the principle that abstention should not be trained in; a source-side audit catches our own alphaNLI construction leaking a surface artifact, so architectural claims are anchored on native-label replications (SQuAD2, RepLiQA, MuSiQue); and on the standard four-domain suite we contribute, to our knowledge, the first answer-or-abstain benchmark, where our gate tops every in-domain dataset and the label-free direction is the only gate family to survive domain transfer.

Chinese Translation

在推理任务中，冻结语言模型存在两个相互关联的弱点：它未充分利用其自身残差流中已经编码的证据，并且未能检测输入是否足够以给出答案，因此出现了虚构现象。本文整合了两条研究线索，旨在在同一残差流上解决这些问题：一个条件引导探针在中间堆栈层写入流，并从冻结的主干网络中恢复推理准确性；一个零-shot充分性方向读取流，并在信息不足时选择放弃。在一次前向传递中部署时，它们相互干扰：引导写入改变了方向读取的状态，导致小模型在跨域转移时损失高达8个AUROC点；而单独的清晰传递则使推理成本翻倍。我们保持方向固定，并训练一个小网络从引导后的残差中重建引导前的残差——在（引导，清晰）对上的均方误差，没有充分性标签——并在重建中读取方向。最终系统YOPO（You Only Pass Once）在冻结的Qwen2.5主干（1.5B/3B/7B）中在一次前向传递中实现回答、引导和放弃。整体而言，三路准确率比冻结基线翻倍以上（0.375->0.798，1.5B alphaNLI），并且在每个规模上，一次传递的表现优于两次传递的参考（0.798/0.830/0.893对比0.753/0.790/0.863），在六个模型系列的十个主干上均表现良好。我们绘制了能力转移边界，量化了放弃不应被训练的原则；源侧审计捕捉到我们自己的alphaNLI构建泄露了表面伪影，因此架构声明基于原生标签复制（SQuAD2、RepLiQA、MuSiQue）；在标准的四域套件上，我们贡献了据我们所知的第一个回答或放弃基准，其中我们的门控在每个领域数据集中表现最佳，而无标签方向是唯一在领域转移中存活的门控家族。

View on arXiv Download PDF AI Translation

arXiv Papers

Adjacency-Based Spectral Proxy Control of Mobile Communication Agents

hint$^2$: Hierarchical World Models for Inference-Time Temporal Logic Guidance

Graph-MambaNav: Spatial-Temporal Graph Mamba Leveraging Object-Relation Knowledge for Object-Goal Navigation

Knowledge-Data-Dual-Driven Reinforcement Learning for Autonomous Vehicle Control in Mixed Traffic

Ontology-Grounded World Models for Failure Diagnosis and Closed-Loop Repair in Physical AI Systems

Communication in modular robotic motor control: Bilateral controllers under realistic constraints

BICPO-VLA: Behavior-Identified Continuation Preference Optimization for Smooth Asynchronous Vision-Language-Action Control

AdvDex: Learning Dexterous Manipulation from Human Demonstrations via Joint-Aligned Actions and Adversarial Learning

Demonstration of Space Robot Teleoperation over a Lossy and Delayed Network using ATMOS

Evolve Vision-Language-Action Model into an Agent with On-the-fly Tool-use

FlatLab: A Unified Methodology Framework and Simulation-Based Benchmark for Robotic Manipulation of Flat Objects

PILOT: Privileged Imitation Learning for End-to-End Motion Planning of Autonomous UAVs under Partial Observability

AgilePE: Autonomous UAV Pursuit-Evasion via Self-Play Reinforcement Learning

OccPlanner: Goal-Aware Occupancy-Conditioned Diffusion Planner for Pixel-Goal Navigation

MMUSV-Sim: A Perception-Oriented Simulation and Data-Generation Platform for Multi-USV Cooperative Perception

Vibration Suppression in Collaborative Flexible Payload Manipulation Using Passive Force Control

Accelerating Large-scale Bundle Adjustment for LiDAR Mapping via Parallel Computing

PRM-as-a-Judge 1.5: A Toolkit for Robot Process Assessment

CORAL: Curriculum-Optimized Reward Adaptation for LiDAR-Based Goal-Directed Urban Driving

Reflex: Enabling Fast and Predictive Vision-Language-Action Models for Reaction-Critical Manipulation

Effect of Twisted-Yarn Architecture on Pressure and Proximity Sensing Characteristics of Textile Capacitive Sensors for Robotic Skin

Control-Informed Constraint Adaptation in Minimum-Time Trajectory Planning for Autonomous Racing

THRIVE: Therapeutic Humanoid Robot In Virtual Environment

Expected Free Energy-based Informative Path Planning for Robotic Mars Exploration

Ensuring Safe Physical AI in Urban Mobility via Hazard-Informed Synthesized Envelopes

Spatiotemporal Tube-Based Safety-Certificate for Autonomous Navigation of Articulated Vehicles

What to Preserve, Where to Adapt: A Depth-Wise Analysis of Forgetting in Continual Gynecological Image Segmentation

Multiphase-Diff: Diffusion-Based Generative Modeling for High-Contrast Multiphase Physical Systems with Sharp Interfaces

PROVE: Training-Free Prompt Recovery using Verifiable Evidence

MedPlex: Deep Vision-Language Co-Adaptation for Clinically Grounded Medical Segmentation

Limitations of Synthetic Data Generation in Specialized Data-Scarce Domains

CAST: Closed-form Analytic Semantic Transfer for Zero-Shot Classifier Extension

ChartProbe: A Diagnostic Study on Visual Reasoning through Perception, Grounding, and Simple Reasoning

Kolmogorov-Arnold Networks for Spatially Independent Multispectral Land Classification

Doomed to Re-Annotate, Forever: The ImageNet Story

Face Re-morphing: Differential Morphing Attack Detection via Feature-Space Similarity Changes

XSA-MAD: Cross-modal Semantic Alignment for Morphing Attack Detection

Attention Capture Is Not Detection: A Two-Stage Account of How Humans Miss Localized AI Image Edits

Consensus-gated Multi-Agent Neural Architecture Search for Seismic Fault Segmentation

Beyond Control Points: Arcsecond Relative-Motion Estimation of Vision Measurement Platforms With Incomplete or Absent Control Fields

OpenBelief-Nav: Evidence-Preserving Object Memory for Open-Vocabulary Language-Guided Navigation

RGBX-Next: Towards Realistic Generative Rendering from G-Buffers

CoANeRV: Coordinate-Aware Token-Space Neural Video Representation

CMCNet: Aligning Ultrasound Image Embeddings with Textual TI-RADS Representations for Fine-Grained Thyroid Classification

Fast Implicit Neural Light Field Representation via Geometric Decomposition and Multi-Resolution Low-Rank Features

SAFE: Scene-Aware Feature Modulation for Color Constancy with Learned Color Space in Pure-Color Scenes

PPOM: Marginalizing Patch-Grid Phase for CLIP-Based Generalizable Vision-Language Prompt Tuning

Rethinking Auxiliary Modalities in Multimodal Zero-shot Anomaly Detection: From Semantic Fusion to Conditional Modulation

ProFocus: Interpreting Affective Experience in Artistic Images with Progressive Visual Focusing

FIRM: Fine-Grained Intra-Token Representation of Masks for Remote Sensing Reasoning Segmentation

MedClaw: Heuristic Agent Harness for Long-Horizon Surgical Video Reasoning

Content Based Video Narration of Gameplay with Vision Language Models

ForgeWM: Progressive Causal Training for Few-Step Action-Conditioned Video World Models

SSP: An Event-Matched Syn2Sim2Phy Cross-Domain Evaluation Framework for Autonomous Driving VLA Models

E-S2Feat:Semantic-Guided Spiking Local Feature Detection and Description for Event Cameras

Beyond Text Conditioning: A Systematic Study of MLLM-DiT Fusion for Video Generation

Source-Agnostic Image Translation Based on Latent Aware Adaptive Masking

Discovery and Spatial Characterisation of Multiple Shortcut Groups for Auditing Vision Model Bias

Voxel-based 3D Facies Segmentation from Seismic Data: A Comparative Study

InstructVVT: Instruction-Driven Video Virtual Try-On without Auxiliary Spatial Priors

Owner3D: Ownership-Guided Style Writing for Training-Free Localized 3D Stylization

CoDS: Robust Collaborative Perception via Expert-driven Detection and BEV Segmentation

Fixed-Budget Gaussian Volume Encoding with Structure-Aware Allocation

HiCo-GS: Hierarchical Context Aggregation and Geometric Consistency for Octree Gaussian Splatting

SPARGen: Unifying Spatial Perception and Reasoning through Native Multimodal Generation

PISA: A Pseudo-Individual Source-Domain Feature Adaptation Framework for Test-Time Open-Vocabulary Object Detection

Self-Supervised Visual On-Policy Distillation

CSG-Mamba: A Convolutional Scoring Gating Vision State Space Network for Endoscopic Polyp Segmentation

SCVIB: Editable State-Conditioned Visual Instance Binding forMulti-Turn Personalized Localization

Concept Guidance: Precise, Training-Free Latent Control for Text-to-Image Generation

LightTeaNet: A Weakly Supervised Lightweight CNN for Multi-Label Tea Leaf Disease Detection and Localization

RankT2I: A Submodular Framework for Discovering Interpretable and Diverse Semantics in Text-to-Image Models

AppleScab-LT: A Longitudinal Real-Field Apple Scab Dataset for Temporal Disease Progression Analysis

Zero-Shot Skeleton-Based Action Anticipation

On the Robustness of Temporal Vision-Language Models for Surgical Endoscopy Videos

Learning to Forecast Crop Growth from Earth Observation Data

MAGneT-3D: Monocular and Domain-Generalizable Temporal 3D Detection

Seeing Red, Thinking Bad: Color Bias in Vision Language Models

Conditional Neural Optimal Transport for Predicting Cellular Phenotypes from Molecular Structure