arXiv Daily Digest

281

Papers

Seeing Less Is Not Seeing Safely: Privacy Leakage from Task-Scoped Robot Perception Exports

看得更少并不意味着安全：任务范围内机器人感知导出中的隐私泄露

Xu, Yuqiao, Ayday, Erman

Abstract

Domestic robots rely on rich perception to operate in private homes, but privacy risk persists even when raw sensor data remain local. Structured representations exported to downstream planners, cloud services, logs, or learning pipelines can still reveal household information through semantics, geometry, spatial structure, and task targets. We introduce Task-Functional Perception Distillation (TFPD), a task-scoped representation-export framework that keeps rich perception local and profiles downstream exports according to task utility, direct exposure, and multiple residual inference risks. Using 120 AI2-THOR scenes with scene-disjoint train/validation/test splits, frozen attacker selection, and representation-aware held-out attacks, we evaluate navigation, collision checking, and object-goal execution. Three navigation exports achieve identical success (1.000) and mean path ratio (0.898), yet representation-level linkability ranges from 0.532 to 0.970. Replacing an explicit target label with a target region reduces target-category macro-F1 from 1.000 to 0.077 while preserving success at 0.995, while geometric coarsening reduces object-category macro-F1 from 0.704 to 0.556 at a measurable collision-utility cost. A ProcTHOR replication preserves the navigation task-equivalence/privacy-inequivalence finding while changing the relative ordering of normalized and topological exports. These results show that neither field removal nor stronger abstraction induces a universal privacy ordering and motivate task-specific, multi-risk evaluation of the complete public representation.

Chinese Translation

家用机器人依赖丰富的感知能力在私人住宅中操作，但即使原始传感器数据保持本地，隐私风险依然存在。导出到下游规划者、云服务、日志或学习管道的结构化表示仍然可以通过语义、几何、空间结构和任务目标揭示家庭信息。我们提出了任务功能感知蒸馏（Task-Functional Perception Distillation, TFPD），这是一个任务范围内的表示导出框架，旨在保持丰富的感知本地化，并根据任务效用、直接暴露和多重残余推理风险来配置下游导出。通过使用120个AI2-THOR场景，采用场景不重叠的训练/验证/测试划分、冻结的攻击者选择和表示感知的保留攻击，我们评估了导航、碰撞检查和目标执行。三种导航导出实现了相同的成功率（1.000）和平均路径比率（0.898），然而表示级的可链接性范围从0.532到0.970。用目标区域替换显式目标标签将目标类别的宏F1从1.000降低到0.077，同时成功率保持在0.995，而几何粗化则在可测量的碰撞效用成本下将对象类别的宏F1从0.704降低到0.556。ProcTHOR的复制保持了导航任务等效性/隐私不等效性的发现，同时改变了归一化和拓扑导出的相对排序。这些结果表明，无论是领域移除还是更强的抽象都不会产生普遍的隐私排序，并促使对完整公共表示进行任务特定的多风险评估。

View on arXiv Download PDF AI Translation

cs.RO / 2 / 2609.03067

GPU-Accelerated Astrodynamics World Models for Spacecraft Rendezvous and Proximity Operations

基于GPU加速的航天动力学世界模型用于航天器交会与接近操作

Eddy, Duncan, Ward, Isaac R., Kim, Grace Ra, Kochenderfer, Mykel J.

Abstract

World models are an emerging paradigm in representation learning in which an agent jointly learns state-action dynamics and observation models from offline trajectory data, enabling multi-step planning and trajectory prediction with uncertainty estimates. They have shown strong results in robotics and game environments, but, to the best of our knowledge, have not previously been applied to the space domain. This paper introduces a world model-based approach to cooperative and non-cooperative spacecraft rendezvous and proximity operations. First, we introduce an open-source, JAX-based International Space Station (ISS) docking environment supporting parallel GPU simulation of spacecraft orbit and attitude dynamics, generating the thousands of state-action transitions that world model training requires. Second, we introduce Out-of-this-World-Model, a transformer-based world model that encodes relative kinematic states and body-fixed camera imagery into a latent state and predicts its evolution under commanded thrusts and torques using one-step flow matching. It produces a distribution over future observations, capturing stochastic dynamics and per-timestep uncertainty, and outperforms DreamerV3-style posterior-correction baselines with fewer trainable parameters and hyperparameters. Third, we apply the approach to a capsule autonomously docking with the ISS under keep-out-zone constraints, demonstrating improved sample efficiency and task performance over reinforcement learning baselines (53% versus 29% docking success across ports), better out-of-distribution generalization (on held-out ports the world model more than doubles baseline success, 40% versus 17%), and detection of anomalous objects encountered during approach with 98% classification accuracy. We open-source the simulation environment and model architecture to enable further study of this paradigm.

Chinese Translation

世界模型是一种新兴的表示学习范式，其中代理从离线轨迹数据中共同学习状态-动作动态和观测模型，使得能够进行多步规划和带有不确定性估计的轨迹预测。它们在机器人技术和游戏环境中表现出色，但据我们所知，尚未在航天领域应用。本文介绍了一种基于世界模型的合作与非合作航天器交会与接近操作的方法。首先，我们介绍了一个开源的、基于JAX的国际空间站（ISS）对接环境，支持航天器轨道和姿态动态的并行GPU仿真，生成世界模型训练所需的数千个状态-动作转变。其次，我们介绍了Out-of-this-World-Model，这是一种基于变换器的世界模型，它将相对运动状态和固定体相机图像编码为潜在状态，并使用一步流匹配预测其在指令推力和扭矩下的演变。它生成未来观测的分布，捕捉随机动态和每个时间步的不确定性，并在可训练参数和超参数更少的情况下超越DreamerV3风格的后验校正基线。第三，我们将该方法应用于在保持禁区约束下自主对接国际空间站的舱段，展示出比强化学习基线更高的样本效率和任务表现（各港口对接成功率为53%对比29%），更好的分布外泛化（在保留港口上，世界模型的成功率超过基线的两倍，40%对比17%），以及在接近过程中以98%的分类准确率检测到异常物体。我们开源了仿真环境和模型架构，以便进一步研究这一范式。

View on arXiv Download PDF AI Translation

cs.RO / 3 / 2609.03142

Sensing Which Modality Matters: Evidence-Gated Regularization for Robust VLA Policies

感知哪种模态重要：用于鲁棒 VLA 策略的证据门控正则化

Yang, Yue, Romeres, Diego, Hori, Chiori, Bertasius, Gedas, Szafir, Daniel, Jain, Siddarth

Abstract

Vision-Language-Action (VLA) policies fuse multimodal sensory inputs, but training on limited and homogeneous robot demonstrations encourages spurious inter-sensor correlations rather than task-relevant signal, a failure we term modality entanglement. Under real-world occlusions and distractors, this manifests as nuisance sensitivity to corruption of uninformative sensors and single-modality insufficiency when only one informative sensor remains intact. We propose Evidence-Gated Regularization (EGR), a modality-agnostic training objective that introduces zero inference-time overhead. EGR derives a per-frame and per-sensor task-relevance signal to gate two state-conditional consistency objectives: invariance on low-evidence sensors, and single-sensor sufficiency on high-evidence ones. We introduce a benchmark based on BEHAVIOR-1K, comprising a fast inference-only diagnostic suite and 47 rollout-based skills targeting modality entanglement. We validate EGR on this benchmark and on two real-robot setups with fundamentally different embodiments: a bi-manual setup with two Kinova arms and three RGB cameras, and a single-arm MELFA ASSISTA setup combining vision and GelSight tactile sensors. EGR improves simulation success rates (SR) from 12.5% to 16.4% under full modalities (+31%), from 9.4% to 16.5% under uninformative-sensor corruption (+75%), and from 2.8% to 6.1% under single-sensor fallback (+120%). Under physical-object distractors, EGR boosts SR from 30% to 85% on the bi-manual setup (+183%) and from 55% to 70% on the tactile setup (+27%).

Chinese Translation

视觉-语言-动作（VLA）策略融合了多模态传感器输入，但在有限且同质的机器人演示上训练会导致虚假的传感器间相关性，而非与任务相关的信号，这种失败我们称之为模态纠缠。在现实世界的遮挡和干扰下，这表现为对无信息传感器的干扰敏感性和在仅有一个信息传感器完好时的单模态不足。我们提出了证据门控正则化（EGR），这是一种与模态无关的训练目标，且在推理时引入零开销。EGR 为每帧和每个传感器推导出任务相关性信号，以门控两个状态条件一致性目标：对低证据传感器的不变性，以及对高证据传感器的单传感器充分性。我们基于 BEHAVIOR-1K 引入了一个基准，包含一个快速推理的诊断套件和 47 个针对模态纠缠的基于回放的技能。我们在该基准和两个具有根本不同实现的真实机器人设置上验证了 EGR：一个是配备两只 Kinova 手臂和三台 RGB 摄像头的双手设置，另一个是结合视觉和 GelSight 触觉传感器的单臂 MELFA ASSISTA 设置。EGR 在完全模态下将模拟成功率（SR）从 12.5% 提高到 16.4%（+31%），在无信息传感器干扰下从 9.4% 提高到 16.5%（+75%），在单传感器回退下从 2.8% 提高到 6.1%（+120%）。在物理对象干扰下，EGR 在双手设置上的 SR 从 30% 提高到 85%（+183%），在触觉设置上的 SR 从 55% 提高到 70%（+27%）。

View on arXiv Download PDF AI Translation

cs.RO / 4 / 2609.03175

Real-Time Shape Control of Multi-Segment Soft Robotic Arms Using Koopman Operators with Global and Local Observables

基于库普曼算子的多段软体机器人臂实时形状控制：结合全局和局部可观测量

Wang, Jiahe, Ristich, Eron, Ali, Sultan Haidar, weissman, Eric, Zhang, Lei, Jin, Wanxin, Ren, Yi, Sun, Jiefeng

Abstract

Multi-segment soft robotic arms can continuously reconfigure their body shapes for safe interaction, but tip control alone is insufficient for constrained-space tasks. Therefore, shape control is a more important task for multi-segment soft arms than tip control, but remains challenging due to the high dimensionality and nonlinear dynamics of continuum deformation. In existing work, shape control accuracy is defined by the error in the global frame (global shape error). For multi-segment soft arms, using only global shape error as the control objective is insufficient, as segment coupling, gravity-induced loading, and inertial effects become more significant. This difficulty increases with the number of segments. In this paper, we present a Koopman-based model predictive control framework that combines global and local observables, enabling real-time shape control on multi-segment soft robotic arms. The framework is evaluated through numerical and physical experiments. Numerical experiments demonstrate the scalability of the proposed controller by achieving shape control on robots with up to 10 independently actuated segments. The physical experiments demonstrate that the controller is capable of (1) real-time shape control of 3- and 5-segment robotic arms with tip speeds up to 0.6 m/s, (2) robust tracking without retraining, including distal payloads up to 400~g and recovery from a 7~N lateral disturbance, and (3) the potential for future inspection applications through a confined-space demonstration. These results demonstrate that the proposed framework enables dynamic, scalable, and accurate real-time shape control on multi-segment soft robotic arms.

Chinese Translation

多段软体机器人臂能够持续重新配置其身体形状以实现安全交互，但仅依靠末端控制不足以应对受限空间任务。因此，对于多段软体臂而言，形状控制比末端控制更为重要，但由于连续变形的高维性和非线性动力学，这一任务仍然具有挑战性。在现有研究中，形状控制的准确性通过全局框架中的误差（全局形状误差）来定义。对于多段软体臂，仅使用全局形状误差作为控制目标是不够的，因为段间耦合、重力引起的载荷和惯性效应变得更加显著。这一困难随着段数的增加而加剧。本文提出了一种基于库普曼算子的模型预测控制框架，该框架结合了全局和局部可观测量，实现了多段软体机器人臂的实时形状控制。通过数值和物理实验对该框架进行了评估。数值实验表明，所提出的控制器具有可扩展性，能够在最多10个独立驱动段的机器人上实现形状控制。物理实验表明，该控制器能够（1）对3段和5段机器人臂进行实时形状控制，末端速度可达0.6 m/s；（2）在不重新训练的情况下实现稳健跟踪，包括承载高达400克的远端负载，并能从7牛顿的横向扰动中恢复；（3）通过受限空间演示展示未来检查应用的潜力。这些结果表明，所提出的框架能够在多段软体机器人臂上实现动态、可扩展和准确的实时形状控制。

View on arXiv Download PDF AI Translation

cs.RO / 5 / 2609.03222

Following a Unique Path: A Fast Certifier Applied to Outlier-Robust Pose Registration

遵循独特路径：应用于异常值鲁棒姿态注册的快速认证器

Holmes, Connor, Goudar, Abhishek, Barfoot, Timothy D.

Abstract

Certifiable methods have arisen as a means to guarantee global optimality of solutions to non-convex problems using convex semidefinite programming (SDP) relaxations. The most performant of these methods use a local solver to obtain the candidate solution, and then certify its optimality using efficient linear algebra techniques. However, for many problems of interest in robotics, this local-solve-then-certify approach is impeded by a form of degeneracy in the relaxation, leaving a costly optimization of the relaxation as the only recourse. In this paper, we introduce our Central-Path Certifier (CP-Cert), a certifiable method explicitly tailored to certify candidate optima to problems that exhibit this form of degeneracy. Using a candidate as a starting point, our approach seeks a nearby region of the feasible space -- known as the central path -- where a valid certificate can be readily obtained. The approach is kept efficient by exploiting indirect linear algebra techniques, problem sparsity, and parallelism. We apply CP-Cert to both matrix-weighted pose registration and pointcloud data association, whose novel SDP relaxation is of independent interest. On simulated examples, we explore the properties of this novel relaxation and show that CP-Cert is fast and scalable, achieving runtimes that are up to three orders of magnitude faster than state-of-the-art direct solvers. Finally, we combine these contributions into a certifiable, outlier-robust pose-estimation pipeline, which we apply to real-world data.

Chinese Translation

可认证的方法已成为保证非凸问题解的全局最优性的手段，利用凸半正定规划（SDP）松弛。最有效的这些方法使用局部求解器获得候选解，然后使用高效的线性代数技术认证其最优性。然而，对于许多机器人领域中的重要问题，这种局部求解后认证的方法受到松弛中的一种退化形式的阻碍，使得对松弛的昂贵优化成为唯一的选择。在本文中，我们介绍了我们的中央路径认证器（Central-Path Certifier，CP-Cert），这是一种专门设计用于认证表现出这种退化形式的候选最优解的可认证方法。我们的做法以候选解为起点，寻求可行空间的邻近区域——称为中央路径——在该区域可以快速获得有效的证书。通过利用间接线性代数技术、问题稀疏性和并行性，我们保持了方法的高效性。我们将CP-Cert应用于矩阵加权姿态注册和点云数据关联，其新颖的SDP松弛本身也具有独立的研究价值。在模拟示例中，我们探讨了这种新松弛的特性，并显示CP-Cert快速且可扩展，其运行时间比最先进的直接求解器快三个数量级。最后，我们将这些贡献结合成一个可认证的、鲁棒的姿态估计管道，并将其应用于真实世界数据。

View on arXiv Download PDF AI Translation

cs.RO / 6 / 2609.03225

Long-Horizon Consistent and Interaction-Aware World Models for Multi-Style End-to-End Driving

面向多风格端到端驾驶的长时间一致性与交互感知世界模型

Han, Yuxuan, Wu, Kunyuan, Yang, Liyunong, Wang, Zilu, Jiang, Cansen, Xiao, Yi, Hu, Liang

Abstract

End-to-end autonomous driving has increasingly adopted world model-based reinforcement learning frameworks to improve learning efficiency through \textit{imagined rollouts}. However, existing world models suffer from three key limitations: temporal inconsistency in long-horizon imagined rollouts, inadequate modeling of ego-environment interactions, and limited adaptability to diverse driving styles. To address these challenges, we propose \textit{StyleDrive}, a world-model-based learning framework that jointly enforces long-horizon consistency, explicitly disentangles interactive traffic states, and supports multi-style policy optimization within a unified learning paradigm. First, we introduce a temporal consistency regularization that integrates historical latent states through gated cross-attention, stabilizing long-horizon imagined rollouts and mitigating error accumulation. Second, we design an explicit state disentanglement module that separates ego-relevant from ego-irrelevant interactive states, enabling more interpretable and efficient decision-making in complex traffic scenarios. Third, we enable multi-style driving behaviors through Group Relative Policy Optimization, which replaces per-step reward optimization with trajectory-wise relative advantages, reducing reward variance and supporting diverse driving styles without retraining. We evaluate StyleDrive on the Bench2Drive closed-loop driving benchmark, achieving a driving score of 88.44 (+17.08 over the previous best world model-based method) and a success rate of 66.82 (+16.58). Furthermore, we deploy StyleDrive on a real automated guided vehicle platform and demonstrate promising sim-to-real transfer capability in dynamic driving scenarios.

Chinese Translation

端到端自主驾驶越来越多地采用基于世界模型的强化学习框架，通过 extit{想象的滚动}提高学习效率。然而，现有的世界模型存在三个主要局限性：长时间想象滚动中的时间不一致性、对自我与环境交互的建模不足，以及对多样化驾驶风格的适应性有限。为了解决这些挑战，我们提出了 extit{StyleDrive}，这是一个基于世界模型的学习框架，旨在共同实现长时间一致性，明确解耦交互交通状态，并在统一学习范式中支持多风格策略优化。首先，我们引入了一种时间一致性正则化，通过门控交叉注意力集成历史潜在状态，稳定长时间想象滚动并减轻误差积累。其次，我们设计了一个显式状态解耦模块，将与自我相关的交互状态与无关状态分离，使得在复杂交通场景中能够进行更具可解释性和高效的决策。第三，我们通过组相对策略优化实现多风格驾驶行为，该方法用轨迹级相对优势替代逐步奖励优化，从而减少奖励方差，并在不重新训练的情况下支持多样化驾驶风格。我们在Bench2Drive闭环驾驶基准上评估了StyleDrive，取得了88.44的驾驶得分（比之前最佳的基于世界模型的方法提高了17.08）和66.82的成功率（提高了16.58）。此外，我们在一个真实的自动导引车平台上部署了StyleDrive，并在动态驾驶场景中展示了良好的仿真到现实的转移能力。

View on arXiv Download PDF AI Translation

cs.RO / 7 / 2609.03255

Establishing a Dynamic Multimodal HRI Dataset for Engagement Analysis with a Humanoid Robot

建立一个动态多模态人机交互数据集以分析与类人机器人互动中的参与度

Kim, Buwan, Jo, Wonse

Abstract

This paper presents an experimental design for constructing a multimodal dataset to analyze user engagement in human-robot interaction (HRI). Prior studies have mainly relied on observable behavioral cues, with limited frameworks integrating physiological signals. We therefore propose a structured data-collection protocol to build a multimodal dataset that includes wearable physiological signals, behavioral data, and self-report measures under different levels of task complexity defined in this experiment.

Chinese Translation

本文提出了一种实验设计，用于构建一个多模态数据集，以分析人机交互（HRI）中的用户参与度。先前的研究主要依赖于可观察的行为线索，且整合生理信号的框架有限。因此，我们提出了一种结构化的数据收集协议，以建立一个多模态数据集，该数据集包括可穿戴生理信号、行为数据和在本实验中定义的不同任务复杂度下的自我报告测量。

View on arXiv Download PDF AI Translation

cs.RO / 8 / 2609.03276

R2S-Eval: Robot Evaluation with Real-to-Sim Calibration via Vision-Language Models

R2S-Eval：通过视觉-语言模型实现真实到仿真校准的机器人评估

Wang, Yidi, Ruan, Feixiang, Chen, Ruoqu, Yin, Jie, Yu, Yang, Xu, Mengdi, Zhang, Kaifeng

Abstract

Evaluating robot manipulation policies is becoming increasingly important as generalist models, particularly vision-language-action (VLA) models, are deployed on physical robots. However, conventional real-world evaluation remains labor-intensive, unstable, and insufficiently informative. It requires repeated hardware trials, manual scene resets, and continuous operator monitoring, may produce different policy rankings across repeated evaluations, and primarily relies on success-rate metrics that provide limited information about execution quality. In contrast, humans assess robot performance by observing and comparing complete behaviors rather than relying solely on binary success outcomes. To this end, we propose R2S-Eval, an evaluation pipeline that combines real-to-sim calibration with vision-language model (VLM) preference evaluation. The real-to-sim component efficiently generates rollout videos in a simulator calibrated to the real-world evaluation setting, thereby reducing the need for repeated hardware trials. The VLM evaluator assesses the execution quality of rollout videos and produces pairwise preferences, which are subsequently aggregated into policy rankings. We further introduce a protocol to assess whether the proposed evaluation pipeline yields validated policy conclusions while mitigating the key challenges of conventional real-world evaluation. Experiments in both simulation and real-world settings demonstrate that R2S-Eval produces reliable and stable policy conclusions, achieves agreement with human preferences, substantially reduces repeated hardware-operation effort, and reveals behavior-quality differences that are not captured by binary success labels. In general, R2S-Eval advances robot evaluation from manual success counting toward automated, statistically stable, and quality-aware evaluation of robot behavior. Project page: https://r2s-eval.github.io.

Chinese Translation

随着通用模型，特别是视觉-语言-行动（VLA）模型在物理机器人上的应用，评估机器人操作策略变得越来越重要。然而，传统的现实世界评估仍然劳动密集、不稳定且信息不足。它需要重复的硬件试验、手动场景重置和持续的操作员监控，可能在重复评估中产生不同的策略排名，并主要依赖成功率指标，这些指标提供关于执行质量的有限信息。相比之下，人类通过观察和比较完整的行为来评估机器人性能，而不仅仅依赖于二元成功结果。为此，我们提出了R2S-Eval，一个将真实到仿真校准与视觉-语言模型（VLM）偏好评估相结合的评估管道。真实到仿真组件有效地生成在与现实世界评估设置校准的模拟器中的回放视频，从而减少了对重复硬件试验的需求。VLM评估器评估回放视频的执行质量并产生成对偏好，这些偏好随后被汇总成策略排名。我们进一步提出了一种协议，以评估所提出的评估管道是否在减轻传统现实世界评估的关键挑战的同时产生有效的策略结论。在模拟和现实世界环境中的实验表明，R2S-Eval能够产生可靠且稳定的策略结论，与人类偏好达成一致，显著减少重复硬件操作的工作量，并揭示二元成功标签未能捕捉到的行为质量差异。总体而言，R2S-Eval推动了机器人评估从手动成功计数向自动化、统计稳定和质量意识的机器人行为评估转变。项目页面：https://r2s-eval.github.io。

View on arXiv Download PDF AI Translation

cs.RO / 9 / 2609.03362

ARTiS: An Adaptive Robotic Gripper for Enhanced Tool Manipulation in Disassembly Applications

ARTiS：一种用于拆卸应用中增强工具操作的自适应机器人夹持器

Mykhailyshyn, Roman, Yukiyasu, Domae, Kensuke, Harada

Abstract

Grasping and holding tools while using them presents a considerable challenge not only for robots but also for humans. Such a challenge is particularly noticeable in processes involving assembly and disassembly, where efficiency and consistency depend on performing rapidly adaptive tasks. Nonetheless, contemporary robotic grasping technologies that can securely manipulate tools during operation frequently have significant constraints. In this paper, introduce ARTiS (Adaptive Robotic Tool Gripper in Disassembly Systems), a novel gripper that combines the adaptability of soft grippers, the dexterity of anthropomorphic hands, and the robustness of rigid mechanisms with a soft palm and fingertips. This unique combination makes it possible to hold tools securely in a variety of situations through using active jamming in the palm and fin-ray adaptation in fingertips. Furthermore, high finger dexterity is achieved through the seven degrees of freedom design, which enables the fingertips to orient to any surface, both for automated solutions and collaborative tasks. A comprehensive evaluation was conducted using a range of conventional disassembly tools to assess the gripper's compliance, durability, and functional versatility. More information, hardware instructions, and videos at https://romanmykhailyshyn.github.io/artis/

Chinese Translation

在使用工具时抓取和保持工具不仅对机器人构成了相当大的挑战，对人类也是如此。这种挑战在涉及组装和拆卸的过程中尤为明显，因为效率和一致性依赖于快速适应任务的执行。然而，当前能够在操作过程中安全操控工具的机器人抓取技术常常存在显著的限制。本文介绍了ARTiS（自适应机器人工具夹持器在拆卸系统中的应用），这是一种新型夹持器，结合了软夹持器的适应性、人形手的灵巧性以及刚性机制的稳健性，配备有柔软的掌心和指尖。这种独特的组合使得在各种情况下通过掌心的主动阻塞和指尖的鳍状适应能够安全地握持工具。此外，通过七自由度的设计实现了高指灵巧性，使得指尖能够朝向任何表面，适用于自动化解决方案和协作任务。我们使用一系列常规拆卸工具进行了全面评估，以评估夹持器的顺应性、耐用性和功能多样性。更多信息、硬件说明和视频请访问 https://romanmykhailyshyn.github.io/artis/

View on arXiv Download PDF AI Translation

cs.RO / 10 / 2609.03392

Programming and execution of skill-based human-robot-crane collaborative tasks

基于技能的人机起重机协作任务的编程与执行

Lohi, Taneli, Suomalainen, Markku, Mellanen, Roope, Heikkilä, Tapio

Abstract

Highly varying production sets increasing challenges for robotic manufacturing and indoor logistics. New capabilities for agility, flexibility, and robustness are needed. Robot skills, integrating motions, tool operations, and sensor perceptions consistently provide an execution mechanism for a versatile set of tasks with varying parameters. In this paper, easy-to-use CAD-model based programming and execution system for parametrized skills and skill monitors is showcased. The execution control structure is dynamic and parametrized, based on a modified Behavior Tree, where only event based communication is used. A human-robot-crane collaborative skill is shown as a test example, where a human instructs an overhead crane and a manipulator in inserting a heavy object supported by the crane, and guided by the manipulator, into the goal.

Chinese Translation

高度变化的生产集给机器人制造和室内物流带来了挑战。需要新的灵活性、适应性和鲁棒性能力。机器人技能整合了运动、工具操作和传感器感知，为一系列具有不同参数的多样化任务提供了一种一致的执行机制。本文展示了一种易于使用的基于CAD模型的编程和执行系统，用于参数化技能和技能监控。执行控制结构是动态和参数化的，基于修改后的行为树（Behavior Tree），仅使用基于事件的通信。作为测试示例，展示了一种人机起重机协作技能，其中人类指挥一台起重机和一个操纵器，将一个由起重机支撑并由操纵器引导的重物插入目标位置。

View on arXiv Download PDF AI Translation

cs.RO / 11 / 2609.03483

Air-Ground Collaborative Vision-and-Language Navigation via Shared Bird's-Eye Maps

基于共享鸟瞰图的空地协作视觉与语言导航

Zhang, Shuning, Li, Liang, Wang, Yunheng, Wang, Tao, Kang, Yihang, Xu, Renjing

Abstract

Air-ground collaborative Vision-and-Language Navigation (VLN) pairs an unmanned aerial vehicle (UAV) with a global bird's-eye view and an unmanned ground vehicle (UGV) with a local first-person view, yet the setting remains largely unexplored: existing training-free methods solve single-agent tasks but offer no collaboration mechanism, and a recent CARLA-Air evaluation found no stable cooperative behavior across five state-of-the-art VLA models; naive semantic communication or bidirectional coupling even degrades performance. We establish AGC-VLN (Air-Ground Collaborative VLN), the first training-free baseline for air-ground collaborative VLN. The key insight is that training-free methods decompose navigation into VLM-based semantic reasoning and deterministic geometric execution, exposing a collaboration interface: the UAV's global view, over which it renders the UGV's reported pose and the VLM-anchored target as CAR/GOAL markers with distance labels, yielding a shared bird's-eye map. From this map, the UGV acquires global spatial context its first-person view cannot provide, plans a road-following path with a frozen VLM, and executes it under closed-loop control; in parallel, the UAV runs 3D-SPF, a spatial-search upgrade of SPF that localizes the target in the downward view and flies toward it. On 100 closed-loop episodes in CARLA-Air's Town10HD scene, AGC-VLN reaches a 77.0% joint success rate, a collaboration gain of +27.0% over the weaker individual agent (the UAV, 50.0%), and exceeds the strongest published single-agent baseline (Travel UAV, 53.0%) by 24.0 points, stemming from the complementarity of the UAV's global view and the UGV's road-following execution. Project page: https://github.com/ZSN2024/AGC-VLN.

Chinese Translation

空地协作视觉与语言导航（VLN）将无人机（UAV）与全球鸟瞰视图相结合，并将无人地面车辆（UGV）与局部第一人称视图相结合，但这一设置仍然在很大程度上未被探索：现有的无训练方法解决单一代理任务，但未提供协作机制，最近的CARLA-Air评估发现五种最先进的VLA模型之间没有稳定的合作行为；幼稚的语义通信或双向耦合甚至会降低性能。我们建立了AGC-VLN（空地协作VLN），这是空地协作VLN的第一个无训练基准。关键见解在于，无训练方法将导航分解为基于VLM的语义推理和确定性几何执行，暴露出一个协作接口：UAV的全球视图，在此基础上渲染UGV报告的姿态和VLM锚定的目标，作为带有距离标签的CAR/GOAL标记，从而生成共享的鸟瞰图。UGV从该地图中获取其第一人称视图无法提供的全球空间上下文，规划一条遵循道路的路径，并在闭环控制下执行；与此同时，UAV运行3D-SPF，这是SPF的空间搜索升级版，能够在向下视图中定位目标并朝其飞行。在CARLA-Air的Town10HD场景中进行的100个闭环实验中，AGC-VLN达到了77.0%的联合成功率，相较于较弱的单一代理（UAV，50.0%）提高了27.0%的协作增益，并超越了最强的已发布单一代理基准（Travel UAV，53.0%）24.0个百分点，这源于UAV的全球视图与UGV的道路跟随执行之间的互补性。项目页面：https://github.com/ZSN2024/AGC-VLN。

View on arXiv Download PDF AI Translation

cs.RO / 12 / 2609.03497

BRIDGE: An Open-Source Humanoid Platform via Morphology-Control Co-Design for Physical AI

BRIDGE：一种通过形态控制协同设计实现的开源类人平台，用于物理人工智能

Wang, Jianren, Qian, Letian, Wang, Zikai, Wu, Weiwei, Zong, Junjie, Gupta, Abhinav, Pathak, Deepak

Abstract

Developing humanoid robots capable of leveraging human behavioral data is essential for general-purpose embodiment, yet conventional development remains bottlenecked by a decoupled paradigm that isolates hardware design from whole-body control. This approach leads to suboptimal systems that compromise human-like fluidity and agility. To bridge this gap, we introduce a data-driven morphology-control co-design framework that optimizes humanoid morphology for human-like movement. To quantify morphological fidelity, we also introduce a novel metric that jointly considers kinematic retargeting fidelity to human motion and dynamic tracking performance. Our framework achieves state-of-the-art (SOTA) performance across all metrics compared to baseline humanoids (Bumi, K1, and Toddlerbot). Finally, we realize this design in Bridge, an open-source, 88cm-tall humanoid platform released alongside its control policy. We demonstrate that Bridge captures human motion data with superior fidelity, exhibiting exceptional performance across foundational locomotion, robust balance, and highly dynamic maneuvers. Videos and open-source materials: https://sites.google.com/view/bridgerobot.

Chinese Translation

开发能够利用人类行为数据的类人机器人对于通用化的具身性至关重要，然而传统的发展模式仍然受到将硬件设计与全身控制分离的瓶颈限制。这种方法导致了妥协人类般流畅性和灵活性的次优系统。为了解决这一问题，我们提出了一种数据驱动的形态控制协同设计框架，该框架优化类人的形态以实现人类般的运动。为了量化形态的保真度，我们还引入了一种新颖的指标，该指标共同考虑了运动学重定向对人类运动的保真度和动态跟踪性能。与基准类人机器人（Bumi、K1和Toddlerbot）相比，我们的框架在所有指标上都达到了最先进的（SOTA）性能。最后，我们在Bridge中实现了这一设计，Bridge是一个88厘米高的开源类人平台，并发布了其控制策略。我们展示了Bridge以优越的保真度捕捉人类运动数据，在基础步态、稳健平衡和高度动态的机动中表现出色。视频和开源材料请访问：https://sites.google.com/view/bridgerobot。

View on arXiv Download PDF AI Translation

cs.RO / 13 / 2609.03561

TRaIL-Odom: Tightly Coupled Continuous Time Radar-IMU-LiDAR Odometry with Adaptive Doppler Weighting

TRaIL-Odom：紧耦合连续时间雷达-IMU-LiDAR里程计与自适应多普勒加权

Noh, Chiyun, Tuna, Turcan, Talbot, William, Hutter, Marco, Kneip, Laurent, Kim, Ayoung

Abstract

Existing radar-LiDAR fusion methods rely on fixed residual weights, even though the informativeness of radar Doppler and LiDAR geometry is scan- and direction-dependent, leading to uniform radar weighting that misallocates Doppler information across translational directions. To address this limitation, we propose two degeneracy-aware Doppler reweighting modules within a tightly coupled Radar-IMU-LiDAR odometry framework: per-point radar reweighting and scan-wise radar gain scheduling. Since geometric degeneracy is directional, we first identify weak translational directions from the LiDAR geometry and reweight individual radar Doppler constraints based on their alignment with the weak subspace. We further adjust the overall radar contribution using LiDAR geometric anisotropy such that radar is emphasized when LiDAR observability is poor and suppressed when LiDAR constraints are already reliable. Across 13 evaluated sequences, TRaIL-Odom achieves state-of-the-art overall performance, with clear advantages in geometrically degenerate scenes. In ablation experiments on three degenerate sequences, combining the two adaptive weighting modules reduces RMSE ATE and RTE by 86.0% and 78.5% relative to the fixed-weight baseline. We make our code and an accompanying dataset publicly available at https://github.com/ChiyunNoh/TRaIL-Odom.

Chinese Translation

现有的雷达-LiDAR融合方法依赖于固定的残差权重，尽管雷达多普勒和LiDAR几何信息的有效性依赖于扫描和方向，这导致雷达权重的均匀分配，错误地分配了跨平移方向的多普勒信息。为了解决这一局限性，我们在紧耦合的雷达-IMU-LiDAR里程计框架中提出了两个考虑退化的多普勒重加权模块：逐点雷达重加权和扫描级雷达增益调度。由于几何退化是方向性的，我们首先从LiDAR几何中识别出弱平移方向，并根据其与弱子空间的对齐情况对单个雷达多普勒约束进行重加权。我们进一步利用LiDAR几何各向异性调整整体雷达贡献，使得在LiDAR可观测性较差时强调雷达，而在LiDAR约束已经可靠时抑制雷达。在评估的13个序列中，TRaIL-Odom实现了最先进的整体性能，在几何退化场景中具有明显优势。在对三个退化序列的消融实验中，结合这两个自适应加权模块，相对于固定权重基线，RMSE ATE和RTE分别减少了86.0%和78.5%。我们将代码和相关数据集公开发布在 https://github.com/ChiyunNoh/TRaIL-Odom。

View on arXiv Download PDF AI Translation

cs.RO / 14 / 2609.03565

Toward Physically Grounded JEPA World Models for Goal-Conditioned Robotic Planning

面向物理基础的JEPA世界模型用于目标条件下的机器人规划

Liu, Muyuan, Huang, Yue, Liang, Zheng, Gao, Xiang

Abstract

Action-conditioned JEPA world models enable planning toward visually specified goals without reconstructing future pixels, yet latent prediction alone does not explicitly encourage the learned representations to retain information relevant to robotic control. We introduce an end-to-end JEPA world model that augments latent prediction with inverse dynamics (IDM) and state alignment (SA). While inverse dynamics discourages latent collapse and makes latent transitions informative of the actions that produced them, state alignment grounds consecutive representations in their associated physical configuration and motion. Across four benchmark tasks, our model attains the highest success rates on TwoRoom (100%), PushT (98%), and OGBench-Cube (87%), while performing comparably to LeWorldModel on Reacher. Our ablation further shows that adding state alignment consistently improves planning success over IDM alone across all four tasks. Although LeWorldModel, our primary baseline, attains higher average straightening on OGBench-Cube, transition-subspace analysis shows that its transition energy is concentrated in a substantially lower-dimensional subspace. Our state-aligned model exhibits a higher effective transition dimension than LeWorldModel and improves planning over IDM alone, supporting state alignment as an effective complement to inverse dynamics for robotic planning.

Chinese Translation

基于动作条件的JEPA世界模型能够实现面向视觉指定目标的规划，而无需重建未来像素，然而仅依赖潜在预测并不能明确促使学习到的表示保留与机器人控制相关的信息。我们提出了一种端到端的JEPA世界模型，通过引入逆动力学（Inverse Dynamics, IDM）和状态对齐（State Alignment, SA）来增强潜在预测。逆动力学防止潜在表示的坍缩，使潜在状态转移包含产生这些转移的动作信息；状态对齐则将连续的表示与其对应的物理配置和运动相结合。在四个基准任务中，我们的模型在TwoRoom（100%）、PushT（98%）和OGBench-Cube（87%）上取得最高成功率，在Reacher任务上表现与LeWorldModel相当。消融实验进一步表明，在所有四个任务中，加入状态对齐均能持续提升规划成功率，优于仅使用IDM。尽管我们的主要基线LeWorldModel在OGBench-Cube上获得更高的平均直线度，但转移子空间分析显示其转移能量集中在维度显著较低的子空间中。相比之下，我们的状态对齐模型展现出更高的有效转移维度，并在规划性能上优于仅使用IDM的模型，支持状态对齐作为逆动力学在机器人规划中的有效补充。

View on arXiv Download PDF AI Translation

cs.RO / 15 / 2609.03591

Scaling Bimanual Household Manipulation from 1,500 hours of Demonstrations to On-Policy Corrections

从1500小时示范到策略校正的双手家庭操作的扩展

Xu, Jiafeng, Li, Qi, Shen, Yan, Ren, Yiyu, Davies, Travis, He, Shaowen, Wang, Ze, Yang, Yifan, Cheng, Ran, Dong, Hao

Abstract

Learning generalist policies for robust bimanual manipulation is bottlenecked by the scarcity of high quality large scale human demonstration data. In this work, we release 1,500 hours of diverse bimanual manipulation demonstrations covering everyday household tasks, and use this comprehensive corpus to train XR-2, a powerful vision-language-action (VLA) model. Enabled by a purpose built high throughput data pipeline and a carefully designed multi stage training paradigm, XR-2 attains strong manipulation performance in our systematic experiments while retaining favorable training efficiency and high data utilization. We further study two critical scaling axes: varying the amount of expert demonstration data, and post training on DAgger correction data from real time human interventions. In both settings, task success rate improves steadily over the data ranges we probe, exhibiting a clear consistent scaling trend at our current data scale. These results validate both the learning capacity of XR-2 and the promising scaling properties of the released dataset, which we open source to support reproducible research on bimanual robot manipulation learning.

Chinese Translation

学习通用策略以实现稳健的双手操作受到高质量大规模人类示范数据稀缺的制约。在本研究中，我们发布了1500小时的多样化双手操作示范，涵盖日常家庭任务，并利用这一综合语料库训练了XR-2，一个强大的视觉-语言-动作（VLA）模型。得益于专门构建的高通量数据管道和精心设计的多阶段训练范式，XR-2在我们的系统实验中实现了强大的操作性能，同时保持了良好的训练效率和高数据利用率。我们进一步研究了两个关键的扩展维度：变化专家示范数据的数量，以及在实时人类干预下对DAgger校正数据的后期训练。在这两种情况下，任务成功率在我们探测的数据范围内稳步提高，展现出在当前数据规模下明显一致的扩展趋势。这些结果验证了XR-2的学习能力以及发布数据集的良好扩展特性，我们将其开源以支持双手机器人操作学习的可重复研究。

View on arXiv Download PDF AI Translation

cs.RO / 16 / 2609.03611

FailBench: How Reliable are VLMs at Judging Robot Task Success?

FailBench：视觉语言模型在判断机器人任务成功率方面的可靠性如何？

Navasardyan, Zaruhi, Danielyan, Tatul, Davtyan, Hrant

Abstract

Vision-Language Models (VLMs) are increasingly used to evaluate robot manipulation outcomes, but existing benchmarks offer limited evidence of cross-domain generalization. We introduce FailBench, a benchmark for robot failure detection comprising 2,197 manipulation attempts across 14 public sources (12 real-world, 2 simulated). In FailBench, 75% of failures occur naturally, and six real-world sources come from non-failure-detection datasets. Evaluating 13 VLM-based detectors, we find the best model achieves only 0.77 mean balanced accuracy. Notably, models fine-tuned for failure detection consistently underperform general-purpose VLMs and their own pretrained baselines. Performance depends heavily on required visual evidence: models approach saturation when outcomes depend on observable object motion, but degrade to near-chance (<0.60 balanced accuracy) on contact-intensive assembly tasks. Error analysis reveals a systematic bias toward predicting success under ambiguous evidence, which persists even with increased reasoning effort. Finally, we show that input-level intervention--spatially localizing and cropping outcome-relevant regions--improves the top detector by 2.4 percentage points without extra training.

Chinese Translation

视觉语言模型（VLMs）越来越多地用于评估机器人操作结果，但现有基准测试提供的跨领域泛化证据有限。我们引入了FailBench，这是一个用于机器人失败检测的基准，包含来自14个公共来源（12个真实世界，2个模拟）的2,197次操作尝试。在FailBench中，75%的失败是自然发生的，六个真实世界来源来自非失败检测数据集。评估了13个基于VLM的检测器后，我们发现最佳模型的平均平衡准确率仅为0.77。值得注意的是，针对失败检测进行微调的模型在性能上始终低于通用VLM和它们自身的预训练基线。性能在很大程度上依赖于所需的视觉证据：当结果依赖于可观察的物体运动时，模型接近饱和，但在接触密集的组装任务中性能降至接近随机水平（<0.60平衡准确率）。错误分析揭示了在模糊证据下预测成功的系统性偏见，即使在增加推理努力的情况下，这种偏见仍然存在。最后，我们展示了输入级干预——空间定位和裁剪与结果相关的区域——在没有额外训练的情况下将最佳检测器的准确率提高了2.4个百分点。

View on arXiv Download PDF AI Translation

cs.RO / 17 / 2609.03623

QLAUN: A Research-Oriented, Robust, Agile, Modular, and Affordable Torque-Controlled Quadruped Robot

QLAUN：一种面向研究的、稳健的、灵活的、模块化且经济实惠的扭矩控制四足机器人

Moudallal, Mohamad S., Maalouf, Noel J.

Abstract

QLAUN Bot (Quad-Legged Adaptive Unmanned Navigator Robot) is a torque-controlled quadruped robot that is research-oriented, cost-effective, and aimed at achieving simultaneous robustness and agility while being completely 3D-printed. It is a quadruped robot that is aimed at empowering robotics research at universities and research institutes in Lebanon and the MENA region. Using a novel electronics-free leg design strategy, we present a modular robot with interchangeable and easily replaceable legs. The 15 kg robot possesses 12 DoF (Degrees-of-Freedom) with three per leg, each paired with a completely 3D-printed Quasi-Direct Drive (QDD) actuator that consists of a brushless DC motor and a low-ratio gearbox transmission that is connected to a belt transmission system for significantly increasing the torque outputs at the joints. We present legs that have decoupled hip and knee actuators to improve the overall modularity of the robot. QLAUN is almost completely 3D-printed using polylactic acid (PLA) and assembled using off-the-shelf parts to create a robust, agile, and affordable robot for legged robot locomotion research. The legs possess joints with wide ranges of motion, including a continuous hip flexion-extension joint. A compliant foot, printed using TPU-95A is also implemented for alleviating hard impacts and handling terrain uncertainties. This extended abstract aims to introduce QLAUN, a novel platform for robotics research, emphasizing the design concepts and principles that underpin its development to the academic and research communities in the field of robotics.

Chinese Translation

QLAUN Bot（四足自适应无人导航机器人）是一种扭矩控制的四足机器人，具有面向研究的特点，经济高效，旨在实现稳健性和灵活性的同时，完全采用3D打印技术。该四足机器人旨在增强黎巴嫩及中东和北非地区大学和研究机构的机器人研究。通过一种新颖的无电子元件腿部设计策略，我们提出了一种模块化机器人，配备可互换和易于更换的腿部。该机器人重15公斤，具有12个自由度（DoF），每条腿3个自由度，每个自由度配备一个完全3D打印的准直接驱动（Quasi-Direct Drive, QDD）执行器，该执行器由无刷直流电机和低比传动齿轮箱组成，并连接到带传动系统，以显著提高关节的扭矩输出。我们提出的腿部设计具有解耦的髋关节和膝关节执行器，以提高机器人的整体模块化。QLAUN几乎完全采用聚乳酸（PLA）3D打印而成，并使用现成的零部件组装，创造出一种稳健、灵活且经济实惠的四足机器人，用于腿部机器人运动研究。腿部关节具有广泛的运动范围，包括连续的髋关节屈伸关节。还采用了使用TPU-95A打印的柔性足部，以减轻硬冲击并应对地形不确定性。本文扩展摘要旨在介绍QLAUN这一机器人研究的新平台，强调其开发背后的设计理念和原则，以便向机器人领域的学术界和研究界进行介绍。

View on arXiv Download PDF AI Translation

cs.RO / 18 / 2609.03630

Local Path Planning and Obstacle Avoidance for an Omnicopter Platform

全向飞行器平台的局部路径规划与障碍物避让

Helinski, Mikolaj, Theodoulis, Spilios, Hamandi, Mahmoud, Ali, Abdullah Mohamed, Tzes, Anthony, Popovic, Marija

Abstract

Autonomous unmanned aerial vehicles (UAVs) increasingly operate in cluttered environments where global planners such as RRT* are not directly deployable at control rates. This paper presents a real-time local planning and obstacle avoidance module for an omnidirectional multirotor (omnicopter) by extending the Dynamic Window Approach to six degrees of freedom (6D-DWA). Our method achieves real-time feasibility through (i) local-map voxelisation, (ii) a compact sphere-based approximation of the vehicle geometry, and (iii) adaptive velocity sampling in the 6D search space. To improve reactivity to unknown obstacles, we introduce a context-aware "Agile Mode" that adjusts scoring weights online to trade-off between goal progress, clearance, and heading/facing constraints during evasive manoeuvres. We evaluate our approach in simulation across computational stress tests, dense-waypoint path tracking, and static/unknown obstacle scenarios. Our planner runs consistently within a 0.2s control loop, tracks waypoint-dense global paths with < 0.1m average cross-track error and 13deg average heading error, and avoids collisions in static environments. For unknown obstacle avoidance, Agile Mode achieves 79.3% success for an off-centre obstacle and 41.4% for a centred obstacle, highlighting both the effectiveness of adaptive weighting and remaining limitations in highly constrained geometries.

Chinese Translation

自主无人机（UAV）越来越多地在杂乱的环境中操作，此时诸如RRT*等全局规划器无法直接以控制速率部署。本文提出了一种针对全向多旋翼飞行器（omnicopter）的实时局部规划与障碍物避让模块，通过将动态窗口方法扩展到六自由度（6D-DWA）来实现。我们的方法通过以下方式实现实时可行性：（i）局部地图体素化，（ii）基于球体的紧凑车辆几何近似，以及（iii）在六维搜索空间中的自适应速度采样。为了提高对未知障碍物的反应能力，我们引入了一种上下文感知的“敏捷模式”，该模式在线调整评分权重，以在规避机动过程中权衡目标进展、间隙和航向/朝向约束。我们在模拟中评估了我们的方法，包括计算压力测试、密集航点路径跟踪以及静态/未知障碍物场景。我们的规划器在0.2秒的控制循环内始终运行，跟踪密集航点的全局路径，平均横向误差小于0.1米，平均航向误差为13度，并在静态环境中避免碰撞。对于未知障碍物的避让，敏捷模式在偏心障碍物的成功率为79.3%，在中心障碍物的成功率为41.4%，突显了自适应加权的有效性以及在高度受限几何形状中的剩余局限性。

View on arXiv Download PDF AI Translation

cs.RO / 19 / 2609.03681

WISE: World-model-guided Imagination Scheduling for Efficient Post-training of Vision-Language-Action Models

WISE：基于世界模型的想象调度以高效后训练视觉-语言-动作模型

Zhang, Chenhao, Zhao, Hanyu, Cheng, Hang, Pan, Tengfei, Zeng, Long

Abstract

Post-training VLA policies typically rely on supervised fine-tuning with costly expert demonstrations or reinforcement learning with expensive and potentially unstable real-world exploration. World models offer a promising alternative by evaluating candidate behaviors through imagined futures, yet effective post-training requires more than accurate prediction: imagination must be scheduled where it is useful, bounded within reliable horizons, and translated into trustworthy policy supervision. In robotic manipulation, the value of imagination varies substantially across execution stages, while extended rollouts can accumulate prediction errors and introduce unreliable learning signals. We introduce WISE (World-model-guided Imagination Scheduling for Efficient Post-training of Vision-Language-Action Models), a unified framework that coordinates when and how world-model imagination is used during policy refinement. WISE selectively invokes imagination at interaction-relevant states, performs bounded multi-view rollouts, evaluates candidate futures using progress and completion signals, and uses their relative outcomes to refine actions generated from real interaction contexts. Extensive experiments with both $\pi_0$ and $\pi_{0.5}$ demonstrate consistent improvements across diverse manipulation tasks while reducing GPU computation time by approximately 80% compared with full imagination. Real-world evaluations further show substantial gains in robustness and generalization under diverse real-world distribution shifts.

Chinese Translation

后训练的视觉-语言-动作（VLA）策略通常依赖于昂贵的专家示范进行监督微调，或通过代价高昂且可能不稳定的现实世界探索进行强化学习。世界模型通过评估候选行为的想象未来提供了一种有前景的替代方案，但有效的后训练不仅需要准确的预测：想象必须在有用的地方进行调度，限制在可靠的范围内，并转化为可信的策略监督。在机器人操作中，想象的价值在执行阶段之间变化显著，而扩展的回合可能会累积预测误差并引入不可靠的学习信号。我们提出了WISE（基于世界模型的想象调度以高效后训练视觉-语言-动作模型），这是一个统一框架，协调在策略优化过程中何时以及如何使用世界模型的想象。WISE在与交互相关的状态下选择性地调用想象，执行有限的多视图回合，使用进展和完成信号评估候选未来，并利用其相对结果来优化来自真实交互上下文生成的动作。与$ ext{π}_0$和$ ext{π}_{0.5}$的广泛实验表明，在多样的操作任务中持续改善，同时与完全想象相比，GPU计算时间减少了约80%。现实世界评估进一步显示在多样的现实世界分布变化下，稳健性和泛化能力显著提升。

View on arXiv Download PDF AI Translation

cs.RO / 20 / 2609.03699

Predictive Zonotope Reduction: Precise Runtime Monitoring under Uncertainty

预测性锥体约简：不确定性下的精确运行时监控

Krsmanovic, Vladimir, Kohn, Florian, Finkbeiner, Bernd, Simovic, Milan

Abstract

Robots operating in physical environments make control decisions based on uncertain sensor measurements, which can lead to unsafe or suboptimal actions. Runtime monitors that check their behavior against safety specifications must represent this uncertainty soundly. Zonotopes are a widely used representation, but continuously incorporating new measurements grows their order unboundedly, so monitors must periodically apply an over-approximating reduction. The choice of the reduction method substantially affects the zonotope's precision, yet existing approaches typically utilize a fixed method throughout the run, even though the optimal choice depends on the current state. This paper presents a Predictive Zonotope Reduction (PZR) approach, which frames reducer selection as an optimal control problem and solves it using beam-search model predictive control. Policy distillation into a small neural policy further provides substantially higher execution speed than model predictive control while maintaining improved performance, enabling uncertainty-aware runtime monitoring on resource-constrained real-time systems. We implement our approach in the RLola runtime monitoring framework and evaluate it on a 5-degree-of-freedom robotic arm simulated in MuJoCo, with sensor uncertainty modeled according to ISO 5725. Experiments on a Raspberry Pi 5 show that dynamic reduction significantly lowers false-positive rates in monitoring compared with static reduction strategies.

Chinese Translation

在物理环境中运行的机器人基于不确定的传感器测量做出控制决策，这可能导致不安全或次优的行为。运行时监控器需要将其行为与安全规范进行检查，因此必须合理地表示这种不确定性。锥体（Zonotopes）是一种广泛使用的表示方法，但不断地纳入新测量会使其阶数无限增长，因此监控器必须定期应用过度近似的约简方法。约简方法的选择显著影响锥体的精度，但现有的方法通常在整个运行过程中使用固定的方法，尽管最佳选择依赖于当前状态。本文提出了一种预测性锥体约简（Predictive Zonotope Reduction, PZR）方法，将约简器选择框架视为一个最优控制问题，并使用束搜索模型预测控制（beam-search model predictive control）进行求解。将策略提炼为小型神经网络策略进一步提供了比模型预测控制更高的执行速度，同时保持了改进的性能，使得在资源受限的实时系统上能够进行不确定性感知的运行时监控。我们在RLola运行时监控框架中实现了该方法，并在MuJoCo中模拟的5自由度机器人手臂上进行了评估，传感器不确定性根据ISO 5725进行建模。在Raspberry Pi 5上的实验表明，与静态约简策略相比，动态约简显著降低了监控中的误报率。

View on arXiv Download PDF AI Translation

cs.RO / 21 / 2609.03704

Toward an~Integrated Cognitive--Ergonomic Architecture for~Human--Machine Interaction: Combining Cognitive Models with~Human Factors Ergonomics

朝向一个综合的认知-人因工程架构以促进人机交互：将认知模型与人因工程相结合

Lenat, Antoine, Cheminat, Olivier, Chablat, Damien, Charron, Camilo

Abstract

This paper presents an integrated approach to modeling human competencies by combining the theoretical foundations of cognitive architectures with principles from Human Factors Ergonomics (HFE). Through a comparative analysis of established cognitive models-SOAR, ACT-R, LIDA, and COCOM-we synthesize a tailored architecture designed to address the complexities of human-machine interaction (HMI) in dynamic environments. By contextualizing this model within ergonomic frameworks, we elucidate the mechanisms underlying decision-making, skill acquisition, and adaptive behavior, bridging the gap between cognitive theory and applied system design. Our framework is empirically grounded in industrial robotics applications, where operator expertise, normative knowledge, and real-time feedback loops are critical. The proposed architecture not only enhances the cognitive alignment of HMI systems but also provides a scalable methodology for designing intelligent, human-centered interfaces in high-stakes environments. This work advances both the theoretical understanding of human competencies and the practical implementation of adaptive, ergonomically optimized systems.

Chinese Translation

本文提出了一种综合的方法，通过将认知架构的理论基础与人因工程（HFE）的原则相结合，来建模人类能力。通过对已建立的认知模型-SOAR、ACT-R、LIDA和COCOM的比较分析，我们综合出一个量身定制的架构，旨在解决动态环境中人机交互（HMI）的复杂性。通过将该模型置于人因工程框架中，我们阐明了决策、技能获取和适应行为背后的机制，弥合了认知理论与应用系统设计之间的鸿沟。我们的框架在工业机器人应用中得到了实证支持，其中操作员的专业知识、规范知识和实时反馈循环至关重要。所提出的架构不仅增强了HMI系统的认知一致性，还提供了一种可扩展的方法，用于在高风险环境中设计智能的以人为中心的界面。这项工作推动了对人类能力的理论理解以及适应性和人因工程优化系统的实际实施。

View on arXiv Download PDF AI Translation

cs.RO / 22 / 2609.03715

MINERVA: How Small Can a Manipulation Policy Be and Still Solve LIBERO?

MINERVA：操控策略可以有多小仍能解决LIBERO？

Sendai, Kohei, Matsushima, Tatsuya, Iwasawa, Yusuke

Abstract

Vision-language-action (VLA) models with billions of parameters now dominate the LIBERO manipulation benchmark, but the model capacity actually required by the benchmark remains unclear. We introduce MINERVA (MINimal Efficient Robotic Vision-Action policy), a family of deliberately compact visuomotor policies designed to measure this task-specific capacity floor. A 0.54M-parameter policy achieves 95.1% average success over 2,000 rollouts on the four standard LIBERO suites, only 2.4 points below the reported LeRobot $\pi_{0.5}$ result despite using 7,700$\times$ fewer parameters. Performance saturates near 1M parameters and collapses below 0.25M. Across broad architectural, training, and inference sweeps, only action-chunk length and vision capacity consistently exceed a $\pm$1-point training-seed band. Flow matching provides no detectable advantage over direct L1 regression across three seeds, while regression is up to 3.8$\times$ faster on GPU. A task-ID permutation probe shows that standard LIBERO instruction conditioning primarily selects among memorized tasks: changing only the task-ID mapping reduces success to near chance. The same recipe achieves 94.6% success across 89 LIBERO-90 tasks, while LIBERO-Plus perturbations reduce performance to 46--56%, with near-zero robustness to photometric shifts. The 0.54M policy replans every control step in 5--9 ms per chunk on a laptop CPU, 113$\times$ faster than SmolVLA and 1,400$\times$ faster than $\pi_{0.5}$, without a GPU. These results establish a first empirical estimate of LIBERO's task-specific capacity floor and motivate capacity-aware design and distillation for deployment-efficient robot policies.

Chinese Translation

具有数十亿参数的视觉-语言-行动（VLA）模型如今主导了LIBERO操控基准，但该基准实际所需的模型能力仍不清楚。我们引入了MINERVA（最小高效机器人视觉-行动策略），这是一个故意设计为紧凑的视觉运动策略家族，旨在测量这一特定任务的能力底线。一个54万参数的策略在四个标准LIBERO套件上经过2000次试验，达到了95.1%的平均成功率，仅比报告的LeRobot $ ext{π}_{0.5}$结果低2.4个百分点，尽管使用的参数少了7700倍。性能在接近100万参数时饱和，而在25万以下则崩溃。在广泛的架构、训练和推理实验中，只有动作块长度和视觉能力始终超过±1点的训练种子带。流匹配在三个种子上未能提供相对于直接L1回归的可检测优势，而回归在GPU上快了最多3.8倍。任务ID置换探针表明，标准LIBERO指令条件主要是在已记忆任务中进行选择：仅改变任务ID映射就会将成功率降低到接近随机水平。相同的方法在89个LIBERO-90任务中达到了94.6%的成功率，而LIBERO-Plus的扰动将性能降低到46%至56%，对光度变化几乎没有鲁棒性。该54万参数的策略在笔记本CPU上每个控制步骤重新规划需要5至9毫秒，比SmolVLA快113倍，比$ ext{π}_{0.5}$快1400倍，且无需GPU。这些结果建立了LIBERO任务特定能力底线的首次实证估计，并推动了针对高效部署的机器人策略的能力感知设计和蒸馏。

View on arXiv Download PDF AI Translation

cs.RO / 23 / 2609.03720

RoughSense: Lightweight Terrain-Induced Rover Vibration Prediction Using Point Clouds and IMU Feedback

RoughSense：基于点云和IMU反馈的轻量级地形诱导探测器振动预测

Garcia, Gabriel Manuel, Aravecchia, Stephanie, Olivares-Mendez, Miguel Angel

Abstract

Autonomous navigation in space requires reliable terrain assessment for safe operations, especially in underground environments with limited communication, computing resources, and power budget. This paper presents a lightweight method for real-time vibration-aware traversability mapping using a Light Detecting And Ranging (LiDAR) point cloud and Inertial Measurement Unit (IMU) measurements. An initial vibration proxy is estimated from terrain geometry by applying Random sample consensus (RANSAC) to local point-cloud patches produced by a Simultaneous Localisation And Mapping (SLAM) algorithm. In parallel, the IMU provides local observations of the vibration experienced by the rover during traversal. The point-cloud-based prediction is then corrected online using Recursive Least Squares, allowing the system to adapt the geometric estimate to the measured rover response. The approach is evaluated in a lunar analogue environment, an outdoor field, and an underground mine.

Chinese Translation

在太空中的自主导航需要可靠的地形评估以确保安全操作，尤其是在通信、计算资源和电力预算有限的地下环境中。本文提出了一种轻量级的方法，通过光学探测与测距（LiDAR）点云和惯性测量单元（IMU）测量，实现实时振动感知的可通行性映射。通过对由同时定位与地图构建（SLAM）算法生成的局部点云块应用随机样本一致性（RANSAC），初步估计地形几何形状的振动代理。同时，IMU提供了探测器在穿越过程中经历的振动的局部观测。然后，基于点云的预测通过递归最小二乘法在线修正，使系统能够将几何估计适应于测量到的探测器响应。该方法在月球类比环境、户外场地和地下矿井中进行了评估。

View on arXiv Download PDF AI Translation

cs.RO / 24 / 2609.03758

A Multi-Vine Soft Robot Enabling Accessible Working Channel and Steering

一种多藤软机器人，能够实现可接入的工作通道和转向

Kashef, Reza, Suulker, Cem, Sofla, Mohammad Sheikh, Althoefer, Kaspar

Abstract

Soft eversion robots, also known as vine robots, have attracted growing interest for navigation and inspection tasks, including minimally invasive medical applications [1]. A vine robot consists of a thin, flexible, inextensible tube folded inward that everts and grows forward when pressurized. This tip-growth enables navigation with minimal friction, making vine robots well suited for complex environments such as the human colon [2]. While their inherent softness allows passive conforma- tion to curved pathways in confined spaces, navigation performance strongly depends on environmental inter- actions, including contact angle and the length of un- constrained deployed material [3], [4]. Sharp directional changes, such as those in the sigmoid colon, often limit passive growth and necessitate active steering. Existing solutions include distributed artificial muscles [5] or dedicated tip-based steering mechanisms [6]. In addition, many applications require payload delivery, such as sensors and tools [7], [8]. Within the ERC Synergy project EndoTheranostics, this motivates the development of vine robots capable of delivering micro- surgical tools during growth. Prior work has integrated working channels within the vine body [8], [9], but these approaches constrain tool size, introduce friction, and limit access to the environment to the robot tip. In this work, we propose a multi-vine architecture in which two vine robots are coupled to an externally integrated working channel via soft mounting tips [10]. Independent vine actuation enables active tip steering while advancing the working channel without embed- ding it within the vine bodies Figure 1. Experiments demonstrate sharp steering of nearly 90 degrees during growth, highlighting the potential of this architecture for versatile medical and non-medical applications.

Chinese Translation

软外翻机器人，也称为藤机器人，因其在导航和检查任务中的应用，尤其是微创医疗应用，受到越来越多的关注[1]。藤机器人由一根薄而灵活、不可延展的管子构成，该管子向内折叠，在加压时外翻并向前生长。这种尖端生长使得导航时摩擦最小，从而使藤机器人非常适合于复杂环境，如人体结肠[2]。尽管其固有的柔软性使其能够在狭小空间中被动适应弯曲路径，但导航性能在很大程度上依赖于环境交互，包括接触角和未约束展开材料的长度[3][4]。在S形结肠等区域，急剧的方向变化常常限制被动生长，并需要主动转向。现有的解决方案包括分布式人工肌肉[5]或专用的尖端转向机制[6]。此外，许多应用需要有效载荷的传递，例如传感器和工具[7][8]。在ERC Synergy项目EndoTheranostics中，这促使我们开发能够在生长过程中传递微创手术工具的藤机器人。之前的研究已经在藤体内集成了工作通道[8][9]，但这些方法限制了工具的大小，引入了摩擦，并限制了机器人尖端对环境的接触。在本研究中，我们提出了一种多藤架构，其中两个藤机器人通过软安装尖端与外部集成的工作通道相连[10]。独立的藤驱动使得在推进工作通道的同时能够主动转向，而无需将其嵌入藤体内。实验表明，在生长过程中能够实现近90度的急剧转向，突显了这一架构在多样化医疗和非医疗应用中的潜力。

View on arXiv Download PDF AI Translation

cs.RO / 25 / 2609.03760

Virtual Testing of Automated Driving Systems through Credible Simulations

通过可信模拟对自动驾驶系统进行虚拟测试

Dona, Riccardo, Rusciano, Espedito, Ciuffo, Biagio

Abstract

Simulation is increasingly used to support safety-related decision-making in road transport, particularly for the assessment and approval of automated driving systems (ADS). The complexity of ADS behavior and size of their operational design domains make exclusive reliance on physical testing impractical, leading to extensive use of virtual testing (VT) during the approval phase. This shift raises critical questions regarding the credibility of modelling and simulation (M&S) results used to support road safety decisions. Current VT accreditation approaches in the ADS domain typically rely on validation-only practices, which have been shown to scale poorly when applied to complex, multi-tool simulation environments. To address this limitation, this paper proposes a risk-based framework for assessing the credibility of simulation toolchains used in ADS safety evaluation, drawing inspiration from established practices in other safety-critical domains, notably NASA's STD-7009 for models and simulations. The framework extends traditional verification and validation (V&V) by explicitly linking credibility requirements to the intended use of simulation outputs and to the safety criticality of the decisions they support within the approval process. It provides a lifecycle-oriented assessment scheme integrating toolchain management, modelling assumptions and limitations, verification, validation, and sensitivity analysis. Credibility acceptance thresholds are defined proportionally, allowing differentiated requirements depending on whether simulation is used for exploratory safety analysis, partial decision support, or as a substitute for physical testing. While demonstrated for ADS, the proposed approach is directly applicable to road safety and simulation studies where VT plays a central role in safety assessment and regulatory decision-making.

Chinese Translation

模拟在道路运输中越来越多地被用于支持与安全相关的决策，特别是在自动驾驶系统（ADS）的评估和批准方面。ADS行为的复杂性以及其操作设计域的广泛性使得单纯依赖物理测试变得不切实际，从而在批准阶段广泛采用虚拟测试（VT）。这一转变引发了关于用于支持道路安全决策的建模和模拟（M&S）结果可信性的重要问题。目前在ADS领域的VT认证方法通常依赖于仅进行验证的实践，而这种方法在应用于复杂的多工具模拟环境时效果不佳。为了解决这一局限性，本文提出了一种基于风险的框架，用于评估在ADS安全评估中使用的模拟工具链的可信性，该框架借鉴了其他安全关键领域的成熟实践，特别是NASA的STD-7009模型和模拟标准。该框架通过明确将可信性要求与模拟输出的预期用途及其在批准过程中支持的决策的安全关键性联系起来，扩展了传统的验证和确认（V&V）。它提供了一种面向生命周期的评估方案，整合了工具链管理、建模假设和局限性、验证、确认和敏感性分析。可信性接受阈值按比例定义，允许根据模拟是用于探索性安全分析、部分决策支持，还是作为物理测试的替代品而设定不同的要求。尽管该方法已在ADS中得到验证，但其提出的方法同样适用于道路安全和模拟研究，在这些研究中，VT在安全评估和监管决策中发挥着核心作用。

View on arXiv Download PDF AI Translation

cs.RO / 26 / 2609.03761

Robot Aware Computational Design of Object Specific Passive Grippers for Additive Manufacturing

面向增材制造的对象特定被动抓取器的机器人感知计算设计

Omaisan, Abdullah Yahya Abdullah, Mohamed, Ibrahim Sheikh

Abstract

This paper presents an end-to-end computational pipeline that converts a selected object mesh, a measured object state, and a selected six-axis robot into an object-specific, unactuated, additively manufacturable gripper. The method couples exact-mesh RGB-D/ICP pose registration, deterministic surface-contact sampling, uncertainty-aware wrench screening, selection among six passive capture mechanisms, object-conformal surface synthesis, full-orientation robot inverse kinematics, a swept-volume-aware manufacturing domain, directional fused-deposition finite-element screening, and constrained three-dimensional SIMP topology optimization. Unlike workflows that treat grasp selection, tool geometry, motion, and structural design as separate problems, every exported design is bound to the source mesh, object pose, robot flange, contact set, and insertion hypothesis by a traceable design identifier. We derive the implemented registration, contact, fit-tolerance, finite-element, and density-optimization equations and prove three properties of the numerical construction: nodal load preservation, monotonic compliance sensitivity under SIMP interpolation, and voxel-domain containment after topology post-processing. Four archived object-specific attempts - a rabbit, camera flange, 3DBenchy, and faceted bust - meet the nominal fit, uncertain-wrench, runtime-sweep, and baseline/post-topology FEA gates. A deliberately enlarged +/-3 mm, +/-5 degree pose stress check differentiates the designs, retaining 29-134 of 160 simulated trials. Their reconstructed topologies retain 92.0-97.6% of the FE domain because functional regions are protected. Archived robot photographs show the corresponding printed assemblies qualitatively, while nominal material properties and absent coupon-calibrated, instrumented tests keep all four at digital-screening status rather than operational release.

Chinese Translation

本文提出了一种端到端的计算管道，该管道将选定的对象网格、测量的对象状态以及选定的六轴机器人转换为对象特定的、未驱动的、可增材制造的抓取器。该方法结合了精确网格 RGB-D/ICP 位姿注册、确定性表面接触采样、不确定性感知扭矩筛选、六种被动捕获机制的选择、对象符合表面合成、全方位机器人逆向运动学、考虑扫掠体积的制造域、方向性熔融沉积有限元筛选以及受限的三维 SIMP 拓扑优化。与将抓取选择、工具几何、运动和结构设计视为独立问题的工作流程不同，每个导出的设计都通过可追溯的设计标识符与源网格、对象位姿、机器人法兰、接触集和插入假设绑定在一起。我们推导了实现的注册、接触、配合公差、有限元和密度优化方程，并证明了数值构造的三个性质：节点载荷保持、在 SIMP 插值下的单调合规灵敏度，以及拓扑后处理后的体素域包含性。四个归档的对象特定尝试——兔子、相机法兰、3DBenchy 和多面体半身像——满足名义配合、不确定扭矩、运行时间扫掠和基线/后拓扑有限元分析门槛。故意扩大 +/-3 mm、 +/-5 度位姿应力检查区分了这些设计，保留了 160 次模拟试验中的 29-134 次。它们重建的拓扑保留了 92.0-97.6% 的有限元域，因为功能区域得到了保护。归档的机器人照片定性展示了相应的打印组件，而名义材料特性和缺乏 coupon 校准的仪器测试使得所有四个设计保持在数字筛选状态，而非操作释放。

View on arXiv Download PDF AI Translation

cs.RO / 27 / 2609.03794

A comparative study on the accuracy & repeatability of mobile robotic platforms for the delivery of precision NDE measurement

移动机器人平台在精密无损检测（NDE）测量交付中的准确性与重复性比较研究

Pour, SeyedMohammadAmin Nabi, Pierce, S. Gareth, Vithanage, Randika, Mohseni, Ehsan, Carswell, David, Shields, Matthew

Abstract

Mobile robotic platforms offer a flexible alternative to fixed manipulators for non-destructive evaluation (NDE) of large aerospace structures, but their base-positioning accuracy and how that accuracy should inform deployment have not been assessed under a common, externally referenced protocol. This work presents a laser tracker-based evaluation workflow (ground truth approximately 6 micrometers) that measures the static and segmented trajectory positioning accuracy of five commercial mobile platforms (KUKA KMP-1500, KUKA KMR, MiR250, Boston Dynamics Spot, Clearpath Husky) under a common protocol. A coupled multi-corner calibration recovers the laser-to-robot transformation and reflector offsets; ordinary least squares over all poses is used, with robust estimation retained only as a blunder check. Static positioning accuracy ranged from a median of 8.2 mm (KMP-1500) to 63.5 mm (Spot), with the wheel-odometry-only Husky uncalibratable. Dynamic path following was characterised by cross-track error; the component was insensitive to temporal alignment, which ranged from 6.9 mm (KMP-1500) to 112.1 mm (Spot). Both accuracy and calibratability tracked localisation capability, from the newest LiDAR SLAM platform to map-free visual odometry. No configuration meets the 0.2 to 1.0 mm aerospace NDE tolerance from the base alone; the results are framed as a design input that sizes the supplementary sensing each platform requires: roughly one order of magnitude for the best platform and nearly two for the worst, providing a reproducible basis for platform selection rather than a feasibility claim.

Chinese Translation

移动机器人平台为大型航空航天结构的无损评估（NDE）提供了一种灵活的替代方案，但其基础定位准确性及该准确性对部署的影响尚未在统一的外部参考协议下进行评估。本研究提出了一种基于激光跟踪器的评估工作流程（真实值约为6微米），测量了五种商业移动平台（KUKA KMP-1500、KUKA KMR、MiR250、波士顿动力Spot、Clearpath Husky）在统一协议下的静态和分段轨迹定位准确性。通过耦合多角校准恢复激光与机器人之间的变换及反射器偏移；对所有姿态采用普通最小二乘法，稳健估计仅作为错误检查。静态定位准确性从中位数8.2毫米（KMP-1500）到63.5毫米（Spot）不等，其中仅依靠轮子里程计的Husky无法校准。动态路径跟踪的特征由横向误差表征；该组件对时间对齐不敏感，时间对齐范围从6.9毫米（KMP-1500）到112.1毫米（Spot）。准确性和可校准性与定位能力相关，从最新的LiDAR SLAM平台到无地图的视觉里程计。没有任何配置能够满足航空航天NDE从基础单元所需的0.2到1.0毫米公差；结果被框定为设计输入，确定每个平台所需的补充传感器：最佳平台大约需要一个数量级，而最差的平台则接近两个数量级，为平台选择提供了可重复的基础，而非可行性声明。

View on arXiv Download PDF AI Translation

cs.RO / 28 / 2609.03889

FWBC-VLA: Force-Aware Whole-Body Compensation for Contact-Rich Loco-Manipulation

FWBC-VLA：面向接触丰富的全身补偿的力感知运动操作

Zhang, Yutian, Ma, Siyuan, Yang, Liwen, Li, Yang, Hao, Ce, Chi, Haozhen, We, Dong, Yu, Qiaojun, Hou, Dibo

Abstract

Contact-rich loco-manipulation requires a bridge between semantic action generation and physical interaction control. Existing Vision-language-action (VLA) models generate task-level actions from visual and linguistic observations, but cannot interpret the physical interactions induced by those actions. While the whole-body control (WBC) policy can stabilize the robot, it cannot distinguish task-relevant interaction forces from forces induced by external disturbances during manipulation. Although force/torque sensors provide direct measurements of physical interactions, retrofitting them entails additional hardware costs and substantial integration effort, particularly for platforms not designed with sensor integration in mind. To address this problem, we propose FWBC-VLA, a force-aware framework that bridges task-level VLA action generation and low-level whole-body compensation control for wheeled-legged robots. First, we introduce HSR-Force, a sensorless residual-torque estimator for inferring contact strength and its temporal variation. These contact estimates are then encoded as tokens and injected into the VLA action expert during action decoding, enabling the policy to perceive contact onset, sustained loading, and release. For loco-manipulation tasks, all parameters of the pretrained VLA backbone are fine-tuned on our WL\&Arm Dataset, which comprises more than 5,000 episodes. Moreover, the robot's proprioceptive state, the Jacobian-derived body-frame force estimate, and the estimated contact state are jointly fed into a compensation generator to produce corrective actions. The manipulation-centric actions are subsequently combined with the corrective actions and passed to the WBC policy for execution. Real-world experiments on whiteboard wiping and door opening with a door closer demonstrate the effectiveness of our FWBC-VLA in contact-rich loco-manipulation.

Chinese Translation

接触丰富的运动操作需要在语义动作生成和物理交互控制之间架起桥梁。现有的视觉-语言-动作（VLA）模型从视觉和语言观察中生成任务级动作，但无法解释这些动作所引发的物理交互。虽然全身控制（WBC）策略可以稳定机器人，但在操作过程中无法区分与任务相关的交互力和由外部干扰引起的力。尽管力/扭矩传感器提供了物理交互的直接测量，但改装它们需要额外的硬件成本和大量的集成工作，尤其是对于未考虑传感器集成的平台。为了解决这个问题，我们提出了FWBC-VLA，一个力感知框架，旨在连接任务级VLA动作生成和低级全身补偿控制，适用于轮腿机器人。首先，我们引入了HSR-Force，一个无传感器的残余扭矩估计器，用于推断接触强度及其时间变化。这些接触估计随后被编码为标记，并在动作解码过程中注入到VLA动作专家中，使策略能够感知接触开始、持续加载和释放。在运动操作任务中，预训练的VLA主干的所有参数在我们的WL&Arm数据集上进行了微调，该数据集包含超过5000个实验。此外，机器人的自我感知状态、基于雅可比的体框架力估计和估计的接触状态被共同输入到补偿生成器中，以产生纠正动作。这些以操作为中心的动作随后与纠正动作结合，并传递给WBC策略以执行。在白板擦拭和带门闭合器的开门等实际实验中，展示了我们的FWBC-VLA在接触丰富的运动操作中的有效性。

View on arXiv Download PDF AI Translation

cs.RO / 29 / 2609.03891

A hybrid pipeline for dynamic ontology-based semantic mapping

一种基于动态本体的混合语义映射管道

Dimitropoulos, Konstantinos, Hatzilygeroudis, Ioannis

Abstract

Semantic mapping plays a crucial role in the ability of a robot to interact with objects, operate and navigate a complex environment. The most common pipeline for semantic mapping consists of geometric mapping and localization (SLAM), perception, semantic fusion and semantic representation. However, more recent works also integrate a form of prior knowledge in their application, most notably knowledge graphs or semantic scene graphs, to improve contextual understanding of the environment. In this paper, we present a hybrid pipeline for semantic mapping. Our system incorporates an external calibrated camera using homography projection for geometric mapping and localization, combined with object detection, persistent object tracking and ontology driven semantic updates to build a dynamic semantic world model. Linear regression models are also used for correction of the estimated values of real world coordinates. The system continuously updates object instances, spatial properties and semantic relations based on real time sensory data. Ontologies are selected as form of knowledge representation due to their hierarchical structure, semantic expressiveness and support for dynamic world modelling.

Chinese Translation

语义映射在机器人与物体交互、操作和导航复杂环境的能力中起着至关重要的作用。最常见的语义映射管道包括几何映射和定位（SLAM）、感知、语义融合和语义表示。然而，最近的研究也在其应用中整合了一种先验知识，尤其是知识图谱或语义场景图，以提高对环境的上下文理解。本文提出了一种用于语义映射的混合管道。我们的系统结合了使用单应性投影的外部标定相机进行几何映射和定位，以及物体检测、持久物体跟踪和基于本体的语义更新，以构建动态语义世界模型。线性回归模型也用于修正实际世界坐标的估计值。该系统基于实时传感器数据持续更新物体实例、空间属性和语义关系。选择本体作为知识表示的形式，因其层次结构、语义表达能力以及对动态世界建模的支持。

View on arXiv Download PDF AI Translation

cs.RO / 30 / 2609.03906

Revisiting Topological Graphs for Macro Action based Closed-loop Reinforcement Learning of Vision Language Navigation in Continuous Environment

重新审视拓扑图在连续环境中基于宏观动作的闭环强化学习中的应用

Ye, Shuhao, Mao, Sitong, Cui, Yuxiang, Wei, Yufei, Yu, Xuan, Zhai, Shichao, Chen, Wen, Zhou, Shunbo, Xiong, Rong, Wang, Yue

Abstract

Vision-Language Navigation in Continuous Environments (VLN-CE) requires an agent to follow natural language instructions through unseen environments. Existing imitation learning (IL) pipelines struggle in this closed-loop setting: behavior cloning suffers from distribution shift, and DAgger's expert actions become ambiguous upon trajectory deviation. While Reinforcement Learning (RL) offers a natural paradigm to address this, directly applying RL to micro action spaces is sample-inefficient due to reward sparsity. To overcome this bottleneck, we reformulate VLN-CE as a Hierarchical Markov Decision Process (MDP), explicitly decoupling high-level planning from low-level control. By abstracting the environment into a topological graph, our high-level policy operates on a macro action space of frontier nodes, with a training-free low-level controller acting as its state transition, which significantly compresses the decision horizon and makes closed-loop RL tractable. To support RL optimization on the macro MDP, we propose an action-aware value head to effectively evaluate state values under the dynamic frontier action space, powering a graph-based PPO. Extensive experiments demonstrate the effectiveness of our architecture. Finally, our model achieves state-of-the-art performance on the R2R-CE and RxR-CE benchmarks.

Chinese Translation

在连续环境中的视觉-语言导航（VLN-CE）要求智能体根据自然语言指令在未见过的环境中行动。现有的模仿学习（IL）流程在这一闭环设置中面临挑战：行为克隆受到分布偏移的影响，而DAgger的专家动作在轨迹偏离时变得模糊。虽然强化学习（RL）提供了一种自然的范式来解决这一问题，但直接将RL应用于微观动作空间由于奖励稀疏性而导致样本效率低下。为了解决这一瓶颈，我们将VLN-CE重新构建为一个层次化的马尔可夫决策过程（MDP），明确将高层规划与低层控制解耦。通过将环境抽象为拓扑图，我们的高层策略在前沿节点的宏观动作空间上操作，而一个无需训练的低层控制器作为其状态转移，这显著压缩了决策视野，使得闭环RL变得可行。为了支持宏观MDP上的RL优化，我们提出了一种动作感知值头，以有效评估动态前沿动作空间下的状态值，推动基于图的PPO。大量实验表明我们架构的有效性。最终，我们的模型在R2R-CE和RxR-CE基准测试中达到了最先进的性能。

View on arXiv Download PDF AI Translation

cs.RO / 31 / 2609.03927

Toward Unified Robot Learning: Bridging Representation, Vision-Language-Action, and World Models

迈向统一的机器人学习：连接表征、视觉-语言-动作和世界模型

Mehta, Shaunak A., Hazarika, Ananya, Zhang, Haochen, Yang, Fan, Moriyama, Ryo, Li, Wenkai, Patel, Yash, Suzuki, Kanata

Abstract

For robots to operate reliably in real-world environments, they need to perceive their surroundings, act, and reason about the consequences of those actions. Rapid progress in the domains of representation learning, VLA models, and world models has significantly enhanced the capabilities of robot learning systems, enabling robots to work in increasingly complex environments. However, these paradigms are typically developed in isolation, resulting in fragmented systems that struggle with generalization, long-horizon temporal reasoning and planning, and deployment in unstructured environments. In this survey, we present a unified perspective on robot learning by organizing the existing methods along three complementary axes: understanding through representation learning, acting through VLA models, and reasoning through world models. We introduce a structured taxonomy that captures key design choices in environment representation, policy learning, and predictive modeling, and summarize the recent progress in these domains. Beyond classifying the existing works, we analyze how these components interact, discuss common limitations, and highlight emerging trends towards more integrated systems. Through this lens, we identify the challenges in the domain of robot learning, including uncertainty quantification, out-of-distribution generalization, cross-embodiment transfer, long-context understanding, and long-horizon planning. We argue that these challenges arise not only from limitations within individual components but also from the lack of integration across perception, action, and reasoning. Building on this analysis, we outline future directions towards unified, physically grounded, and probabilistic robot learning to develop robust robotic systems that maintain consistent internal representations and support decision making over extended interactions in real-world environments.

Chinese Translation

为了使机器人能够在真实环境中可靠地操作，它们需要感知周围环境、采取行动并推理这些行动的后果。在表征学习、视觉-语言-动作（VLA）模型和世界模型领域的快速进展显著增强了机器人学习系统的能力，使机器人能够在日益复杂的环境中工作。然而，这些范式通常是孤立开发的，导致系统碎片化，难以实现泛化、长时间序列推理和规划，以及在非结构化环境中的部署。在本次调研中，我们通过沿着三个互补的轴线组织现有方法，呈现了对机器人学习的统一视角：通过表征学习理解，通过VLA模型行动，以及通过世界模型推理。我们引入了一个结构化的分类法，捕捉环境表征、策略学习和预测建模中的关键设计选择，并总结了这些领域的最新进展。除了对现有工作的分类，我们还分析了这些组件之间的相互作用，讨论了常见的局限性，并强调了朝着更集成系统的 emerging trends。通过这一视角，我们识别了机器人学习领域中的挑战，包括不确定性量化、分布外泛化、跨体现迁移、长上下文理解和长时间规划。我们认为，这些挑战不仅源于各个组件的局限性，还源于感知、行动和推理之间缺乏整合。在此分析的基础上，我们概述了迈向统一的、物理基础的和概率性的机器人学习的未来方向，以开发能够保持一致内部表征并支持在真实环境中进行长期交互决策的稳健机器人系统。

View on arXiv Download PDF AI Translation

cs.RO / 32 / 2609.03970

Automated Weld Seam Recognition and 3D Mapping for Robotic Post Processing Using Photogrammetry and Semantic Segmentation

基于摄影测量和语义分割的自动化焊缝识别与三维建模用于机器人后处理

Raju, Augustin, Madavath, Abilash, Aubeeluck, Chandra Yuvesh, Pyschny, Nicolas, Hackelöer, Felix, Zwanzig, Florian

Abstract

Accurate identification of weld seam geometries is essential for automated robotic post processing operations such as grinding, finishing, and inspection. For large workpieces, complete surface scanning using high precision laser scanners or structured light sensors can be time consuming and often generates substantial amount of data that are not relevant. This paper presents an experimental vision based pipeline for the approximate localization of weld seams. This serves as a preliminary stage before high precision measurement. The proposed approach aims to reduce the overall scanning effort and data acquisition efficiency. The proposed method includes capturing images of the workpiece from multiple viewpoints, identifying weld seams from the images using semantic segmentation, reconstructing the workpiece using photogrammetry, and projection of identified weld seams into the reconstructed model.

Chinese Translation

准确识别焊缝几何形状对于自动化机器人后处理操作（如打磨、精加工和检验）至关重要。对于大型工件，使用高精度激光扫描仪或结构光传感器进行完整表面扫描可能耗时较长，并且通常会生成大量不相关的数据。本文提出了一种基于视觉的实验性流程，用于焊缝的近似定位。这作为高精度测量之前的初步阶段。所提出的方法旨在减少整体扫描工作量和数据采集效率。该方法包括从多个视角捕捉工件图像，利用语义分割从图像中识别焊缝，使用摄影测量重建工件，并将识别出的焊缝投影到重建模型中。

View on arXiv Download PDF AI Translation

cs.RO / 33 / 2609.03984

MulDP: Multimodal Diffusion Policy for Autonomous Quadruped Parkour Navigation across Complex Terrains

MulDP：用于复杂地形的自主四足机器人跑酷导航的多模态扩散策略

Hu, Kangmai, Zhang, Yueqi, Zhai, Peng, Wei, Xiaoyi, Hu, Jiabin, Liu, Zhixiang, Qian, Quancheng, Zhang, Lihua

Abstract

Quadruped robots have demonstrated impressive agility in parkour locomotion across complex terrains. However, most systems still rely on human intervention for high-level planning, and autonomous parkour navigation remains underexplored. The key challenges include fine-grained velocity regulation, long-horizon anticipatory behaviors, and tight coupling between perception and embodied execution. To address these challenges, we propose a Multimodal Diffusion Policy (MulDP) that integrates visual perception with robot proprioception and goal information to generate temporally coherent and anticipatory navigation velocity commands, tightly coupling perception with embodied control to enable robust autonomous navigation. To support the training of MulDP, we construct the first Quadruped Parkour Navigation Dataset (QPND), a multimodal dataset that encompasses diverse navigation behaviors and complex terrains. Extensive simulation and real-world experiments demonstrate that MulDP enables robust long-horizon autonomous navigation and effective traversal across complex terrains.

Chinese Translation

四足机器人在复杂地形的跑酷运动中展现了令人印象深刻的灵活性。然而，大多数系统仍依赖于人工干预进行高层规划，自主跑酷导航仍然未得到充分探索。主要挑战包括精细的速度调节、长时间的预期行为以及感知与身体执行之间的紧密耦合。为了解决这些挑战，我们提出了一种多模态扩散策略（MulDP），该策略将视觉感知与机器人本体感知和目标信息相结合，以生成时间上连贯且具有预期性的导航速度指令，紧密耦合感知与身体控制，从而实现稳健的自主导航。为了支持MulDP的训练，我们构建了第一个四足跑酷导航数据集（QPND），这是一个涵盖多样化导航行为和复杂地形的多模态数据集。大量的仿真和现实世界实验表明，MulDP能够实现稳健的长时间自主导航，并有效穿越复杂地形。

View on arXiv Download PDF AI Translation

cs.RO / 34 / 2609.04096

Adaptive Vision-Language Grasping via Composable Foundation Priors and Generalizable Grasp Synthesis

通过可组合基础先验和可推广抓取合成的自适应视觉-语言抓取

Yan, Sixu, Wang, Shikang, Huang, Binhua, Tang, Xuanlai, Fan, Guohua, Huang, Fan, Li, Haoxuan, Li, Yongkang, Li, Yuhan, Liao, Bencheng, Zhang, Zeyu, Liu, Wenyu, Liu, Hangxin, Wang, Xinggang

Abstract

This paper proposes AdaRoboVLG, a task-adaptive Vision-Language-Grasp (VLG) framework that supports generalizable grasp synthesis across different robotic hands. Unlike existing VLG methods that tightly couple foundation models with end-to-end grasp policies, AdaRoboVLG learns an efficient generalizable base policy that generates and evaluates physically feasible grasp candidates through explicit kinematic mapping and force-closure-based stability estimation, while offloading task-dependent understanding to specialized foundation-model modules. These modules provide composable priors that are integrated into the grasp synthesis process, enabling contextually adaptive grasp synthesis without retraining the underlying grasp policy. Through extensive simulation and real-world experiments, we demonstrate that (i) the base policy exhibits efficient learning and strong cross-hand generalization, (ii) the framework effectively incorporates spatial, cognitive, and temporal priors to address three representative grasping challenges without compromising grasp synthesis performance compared to state-of-the-art methods, and (iii) these priors can operate jointly to enable functional grasping in cluttered and dynamic environments. These results indicate that decoupling physical grasp synthesis from task-dependent understanding provides a scalable paradigm for robotic grasping, allowing future advances in foundation models to be directly translated into improved grasp capabilities without redesigning or retraining the underlying grasp policy. Supplementary videos are available at https://adarobovlg.github.io/

Chinese Translation

本文提出了AdaRoboVLG，一个任务自适应的视觉-语言-抓取（Vision-Language-Grasp, VLG）框架，支持不同机器人手的可推广抓取合成。与现有的紧密耦合基础模型与端到端抓取策略的VLG方法不同，AdaRoboVLG学习了一种高效的可推广基础策略，通过显式的运动学映射和基于力闭合的稳定性估计生成和评估物理可行的抓取候选，同时将任务依赖的理解卸载到专门的基础模型模块。这些模块提供可组合的先验，集成到抓取合成过程中，使得在不重新训练基础抓取策略的情况下实现上下文自适应的抓取合成。通过广泛的仿真和现实世界实验，我们证明了（i）基础策略表现出高效的学习和强大的跨手泛化能力，（ii）该框架有效地结合空间、认知和时间先验，以应对三种典型的抓取挑战，而不影响与最先进方法相比的抓取合成性能，以及（iii）这些先验可以共同作用，使得在杂乱和动态环境中实现功能性抓取。这些结果表明，将物理抓取合成与任务依赖的理解解耦提供了一种可扩展的机器人抓取范式，使得未来基础模型的进展能够直接转化为改进的抓取能力，而无需重新设计或重新训练基础抓取策略。补充视频可在 https://adarobovlg.github.io/ 获取。

View on arXiv Download PDF AI Translation

cs.RO / 35 / 2609.04103

Corner Cases: Headland Coverage Path Planning for Autonomous Driving in Arable Farming

边角案例：农业耕作中自主驾驶的边界覆盖路径规划

Soitinaho, Riikka, Oksanen, Timo

Abstract

This paper presents a new method for headland coverage path planning for arable fields. Several earlier approaches suggest covering the headland with nested polygons and smooth turns, however, covering the field corners entirely requires manoeuvres with reversing. In the new method, the polygon corners are modified to allow a reversing turn. A comparison to two other methods considering gap, overlap, and crossing the field boundary shows an improvement in the coverage result especially in field corners of around 90 degrees, and 240 degrees and above. Applicability of the new method is shown with several examples of real polygonal field maps.

Chinese Translation

本文提出了一种新的农业耕作领域的边界覆盖路径规划方法。早期的一些方法建议使用嵌套多边形和光滑转弯来覆盖边界，然而，完全覆盖田地角落需要进行倒车操作。在新方法中，多边形的角落被修改以允许倒车转弯。与考虑间隙、重叠和穿越田地边界的其他两种方法的比较显示，特别是在约90度和240度及以上的田地角落，覆盖结果有所改善。通过多个实际多边形田地地图的示例展示了新方法的适用性。

View on arXiv Download PDF AI Translation

cs.RO / 36 / 2609.04193

GIFT: Guided Intermediate Feature Training via Action-Oriented Structural Supervision for Robotic Manipulation

GIFT：通过面向动作的结构监督引导中间特征训练以实现机器人操作

Zheng, Yupeng, Li, Xiang, Gu, Songen, Zheng, Yuhang, Tian, Shuai, Li, Weize, Wang, Linbo, Li, Chaoyue, Zhang, Qichao, Li, Haoran, Xia, Zhongpu, Zhang, Ya-Qin, Yan, Shuicheng, Zhao, Dongbin

Abstract

Vision-language pre-training and predictive world modeling provide robot policies with rich semantic and dynamic visual features, but their native action and visual-prediction objectives may omit critical physical and task structure while retaining control-irrelevant visual redundancy. We call this mismatch between visual richness and control utility the action-sufficiency gap. We investigate whether this gap can be bridged by guiding intermediate features to preserve three control-relevant structure in robotic manipulation: geometry governing motion feasibility, affordance encoding instruction-relevant entities, and goals grounding instructions in task-relevant regions. To this end, we present GIFT (Guided Intermediate Feature Training), an architecture-flexible framework for learning intermediate features that translates these structures into training-time constraints through geometry alignment, affordance prediction, and goal-region reconstruction. We instantiate GIFT in a Vision-Language-Action (VLA) policy, a direct-action World-Action Model (WAM), and an inverse-dynamics WAM while retaining each model's action formulation. Under zero-shot transfer to LIBERO-Plus, GIFT-VLA, GIFT-WAM-Fast, and GIFT-WAM-IDM outperform StarVLA-OFT, Fast-WAM, and Fast-WAM-IDM by 4.6, 12.6, and 5.2 points, reaching 79.6%, 72.6%, and 87.8%, respectively. On RoboCasa, the three GIFT variants reach 61.4%, 83.6%, and 82.3%, outperforming their counterparts by 12.6, 9.0, and 8.4 points, respectively. Together, these results establish learning functionally structured intermediate features as a reusable principle across model-specific action formulations, with especially large gains on articulated-object tasks and high-precision real-world manipulation under unseen visual and spatial perturbations. Project page: https://openphoenix-team.github.io/GIFT-pages.

Chinese Translation

视觉-语言预训练和预测世界建模为机器人策略提供了丰富的语义和动态视觉特征，但它们固有的动作和视觉预测目标可能会忽略关键的物理和任务结构，同时保留与控制无关的视觉冗余。我们将这种视觉丰富性与控制效用之间的不匹配称为动作充分性差距。我们研究是否可以通过引导中间特征来弥补这一差距，以保留机器人操作中的三种控制相关结构：支配运动可行性的几何结构、编码指令相关实体的可供性，以及将指令定位于任务相关区域的目标。为此，我们提出了GIFT（引导中间特征训练），这是一个架构灵活的框架，用于学习中间特征，通过几何对齐、可供性预测和目标区域重建将这些结构转化为训练时约束。在视觉-语言-动作（VLA）策略、直接动作世界-动作模型（WAM）和逆动力学WAM中实例化GIFT，同时保留每个模型的动作表述。在零样本转移到LIBERO-Plus时，GIFT-VLA、GIFT-WAM-Fast和GIFT-WAM-IDM分别比StarVLA-OFT、Fast-WAM和Fast-WAM-IDM高出4.6、12.6和5.2个百分点，分别达到79.6%、72.6%和87.8%。在RoboCasa上，三个GIFT变体分别达到61.4%、83.6%和82.3%，分别比其对应模型高出12.6、9.0和8.4个百分点。这些结果共同确立了学习功能结构化中间特征作为跨模型特定动作表述的可重用原则，尤其在未见视觉和空间扰动下的关节物体任务和高精度现实世界操作中获得了显著提升。项目页面：https://openphoenix-team.github.io/GIFT-pages。

View on arXiv Download PDF AI Translation

计算机视觉 (Computer Vision)

101

cs.CV / 1 / 2609.03052

IDSPACE: A Novel Document Generator for Reliable Evaluation of Digital Identity Verification Systems [Extended Technical Report]

IDSPACE：一种新型文档生成器，用于可靠评估数字身份验证系统 [扩展技术报告]

Xie, Lulu, Wang, Yancheng, Chowdhury, Kanchan, Garcia, Rolando, Yang, Yingzhen, Zou, Jia

Abstract

As services move online, trust institutions such as banks, lenders, and governments must verify the identity of remote users. Fraud detection tools are widely available, but evaluating and fine-tuning them remains difficult because identity documents are sensitive and therefore scarce. Synthetic data generation offers a path forward, and demand is clear: our prior work in this area has been downloaded over $11{,}000$ times (aggregated from eight parts). We introduce IDSpace, extending this line of research in three directions. First, we propose model-guided Bayesian optimization, which tunes generation parameters to maximize both visual similarity and prediction consistency with target-domain models given only a few samples from a target domain. Second, we decouple user-specified metadata (demographics, fraud patterns, capture device) from automatically tuned control parameters (font styles, noise levels, image quality), allowing users to configure evaluations without low-level expertise. Third, we expand beyond template images to support scanned and mobile-captured documents. Experiments show IDSpace improves evaluation consistency by $15-45\%$ over baselines including CycleGAN, diffusion inpainting, and non-guided optimization, using only a few real samples, while improving training accuracy by up to $9\%$ and SSIM similarity with the target domain by $10\%$. We also released a new dataset consisting of $359{,}240$ high-quality synthetic documents across ten European ID types.

Chinese Translation

随着服务的在线化，银行、贷款机构和政府等信任机构必须验证远程用户的身份。虽然欺诈检测工具广泛可用，但由于身份文件敏感且稀缺，评估和调整这些工具仍然困难。合成数据生成提供了一条前进的道路，需求显而易见：我们在这一领域的先前工作已被下载超过 $11{,}000$ 次（来自八个部分的汇总）。我们介绍了 IDSpace，沿着这一研究方向扩展了三个方面。首先，我们提出了模型引导的贝叶斯优化，该方法调整生成参数，以最大化视觉相似性和与目标领域模型的预测一致性，仅需少量目标领域样本。其次，我们将用户指定的元数据（人口统计信息、欺诈模式、捕获设备）与自动调整的控制参数（字体样式、噪声水平、图像质量）解耦，使用户能够在没有低级专业知识的情况下配置评估。第三，我们超越模板图像，支持扫描和移动捕获的文档。实验表明，IDSpace 在评估一致性上比包括 CycleGAN、扩散修复和非引导优化的基线提高了 $15-45 ext{ extperthousand}$，仅使用少量真实样本，同时将训练准确率提高了最多 $9 ext{ extperthousand}$，与目标领域的 SSIM 相似度提高了 $10 ext{ extperthousand}$。我们还发布了一个新数据集，包含 $359{,}240$ 个高质量合成文档，涵盖十种欧洲身份证类型。

View on arXiv Download PDF AI Translation

cs.CV / 2 / 2609.03077

Position: Unlabeled IS NOT Equal to No Human Supervision in Visual Learning

立场：未标记不等于视觉学习中的无人工监督

Lao, Dong

Abstract

This position paper argues that the absence of labels does not imply the absence of human supervision in visual learning, and urges the research community to identify sources of supervision more explicitly. Many recent methods in computer vision build upon representations learned from large-scale unlabeled data, and are therefore grouped under the same umbrella term ``unsupervised.'' However, different data curation schemes and training objectives embed substantially different human priors on which models rely, and we argue that one ``unsupervised'' umbrella term is no longer capturing these distinctions. This ambiguity makes it harder to compare unsupervised learning research conducted under different assumptions, coinciding with a sharp decline in papers titled with ``unsupervised'' in flagship computer vision conferences since 2021, despite continued growth of the field. While we fully embrace pre-training as a strong foundation for modern computer vision, we advocate for a community-level effort toward greater conceptual clarity: authors are encouraged to disclose priors in data selection and learning objectives, and to specify which components of a learning pipeline depend on which assumptions. Standardized disclosure practices can improve academic communication, ensure fairer comparisons, and preserve methodological diversity in unsupervised learning.

Chinese Translation

本文立场论文认为，标签的缺失并不意味着视觉学习中缺乏人工监督，并呼吁研究界更明确地识别监督来源。许多近期的计算机视觉方法建立在从大规模未标记数据中学习的表征之上，因此被统称为“无监督”。然而，不同的数据策划方案和训练目标嵌入了实质上不同的人类先验，而模型依赖于这些先验。我们认为，单一的“无监督”这一统称已无法捕捉这些差异。这种模糊性使得在不同假设下进行的无监督学习研究之间的比较变得更加困难，自2021年以来，尽管该领域持续增长，标有“无监督”的论文在旗舰计算机视觉会议中的数量却急剧下降。虽然我们完全接受预训练作为现代计算机视觉的坚实基础，但我们倡导在社区层面上努力实现更大的概念清晰性：鼓励作者披露数据选择和学习目标中的先验，并明确学习流程中哪些组件依赖于哪些假设。标准化的披露实践可以改善学术交流，确保更公平的比较，并保持无监督学习中的方法多样性。

View on arXiv Download PDF AI Translation

cs.CV / 3 / 2609.03080

Exemplar: Classical Priors Complement Frozen Features for Few-Shot Microscopy Segmentation at Native Resolution

Exemplar：经典先验补充冻结特征以实现原生分辨率下的少样本显微镜分割

Průšek, Michal, Novozámský, Adam, Šroubek, Filip

Abstract

Segmenting a new biomedical dataset usually means a domain-specific model trained on substantial annotation, or a foundation model steered at inference time. We present Exemplar, a few-shot segmenter that fuses a frozen DINOv3 backbone with a fixed bank of classical native-resolution filter responses in one lightweight head, fitted from the support masks alone. In the few-mask, native-resolution regime, classical priors and frozen self-supervised features are complementary: fused in one head, a single fixed configuration spans eleven biomedical imaging datasets. Under the same head, the classical bank alone reaches 0.693 on the eleven-dataset panel, scored by foreground intersection-over-union or centreline Dice, and the frozen features alone 0.672; the bank leads on seven of the eleven and the features on the rest, and fused they reach 0.782. Against five forward-pass few-shot methods, Exemplar leads in 54 of 55 method-dataset comparisons, 52 of them significant after Holm correction. From a single annotated mask it reaches 0.703 on the same panel, against 0.682 for a from-scratch nnU-Net trained on that same mask. At eight masks nnU-Net overtakes it on the panel mean, chiefly on centreline agreement, but takes 16-77x longer to fit.

Chinese Translation

对新的生物医学数据集进行分割通常意味着需要在大量标注上训练一个特定领域的模型，或者在推理时使用一个基础模型。我们提出了Exemplar，一种少样本分割器，它将冻结的DINOv3主干与固定的经典原生分辨率滤波响应库融合在一个轻量级的头部中，仅通过支持掩码进行拟合。在少样本、原生分辨率的情况下，经典先验和冻结的自监督特征是互补的：在一个头部中融合后，单一的固定配置可以覆盖十一种生物医学成像数据集。在同一个头部下，经典库单独在这十一种数据集上达到0.693的得分（通过前景交并比或中心线Dice评分），而冻结特征单独为0.672；经典库在十一种数据集中的七种上领先，而特征在其余数据集上领先，二者融合后达到0.782。在五种前向传播的少样本方法中，Exemplar在55个方法-数据集比较中领先54个，其中52个在Holm校正后显著。从单个标注掩码出发，它在同一面板上达到0.703，而从头开始训练的nnU-Net在该掩码上仅为0.682。在八个掩码时，nnU-Net在面板均值上超越了Exemplar，主要是在中心线一致性上，但拟合所需时间长达16-77倍。

View on arXiv Download PDF AI Translation

cs.CV / 4 / 2609.03085

Solving the Needle-in-a-Haystack Problem in Mammography Vision-Language Model with Differentiable Subset Sampling

利用可微分子集采样解决乳腺X光检查中的针在干草堆问题

Jeon, Young Seok, Brown-Mulry, Beatrice, Isaac, Rohan Satya, Dissanayaka, Anjana, Dapamede, Theo, Chavoshi, Mohammadreza, Gichoya, Judy, Trivedi, Hari

Abstract

There is growing interest in adopting CLIP-style vision--language model (VLM) pretraining for mammography. However, models that directly employ the standard CLIP architecture and training objective exhibit limited zero-shot performance in clinically important tasks such as cancer, finding-type, and BI-RADS predictions. We argue that this underwhelming performance is due to neglecting two characteristics of mammography data: (1) its high-res nature, and (2) homogeneity of radiology reports, largely driven by a predominance of negative/benign findings on examinations. We propose TopKSigLIP, a VLM designed to address these two limitations through a novel architecture and learning objectives. Instead of downscaling high-res mammography images to satisfy GPU memory constraints, TopKSigLIP introduces TopK-Patch module that learns to sample a sparse set of high-res patches likely to contain lesions, sidestepping the resolution--batch size tradeoff of VLM training. The sampled patch locations additionally serve as a built-in localization tool. To address report homogeneity, we replace the contrastive loss, which falsely repels semantically similar pairs, with a Sup-sigmoid loss. Sup-sigmoid loss extends the sigmoid loss from SigLIP with soft labels derived from structured data. TopKSigLIP outperforms existing open-source mammography and general medical VLMs on both internal and external benchmarks on density assessment, BI-RADS classification, finding subtyping, and cancer prediction under zero-shot evaluation. TopKSigLIP remains competitive under linear probing despite using a significantly smaller vision encoder and smaller training batches than baselines. The TopK-Patch module additionally achieves superior lesion localization over post-hoc Grad-CAM. Code and weights are made public:https://github.com/Youngseok0001/TopKSigLIP.

Chinese Translation

越来越多的研究者对采用CLIP风格的视觉-语言模型（VLM）进行乳腺X光检查的预训练表现出浓厚的兴趣。然而，直接采用标准CLIP架构和训练目标的模型在癌症、发现类型和BI-RADS预测等临床重要任务中的零-shot性能有限。我们认为，这种不尽如人意的表现是由于忽视了乳腺X光数据的两个特征：（1）其高分辨率特性，以及（2）放射学报告的同质性，这主要是由于检查中阴性/良性发现的占主导地位。我们提出了TopKSigLIP，这是一种旨在通过新颖的架构和学习目标来解决这两个局限性的VLM。TopKSigLIP引入了TopK-Patch模块，该模块学习采样一组稀疏的高分辨率补丁，这些补丁可能包含病变，从而避免了VLM训练中的分辨率与批量大小之间的权衡。采样的补丁位置还作为内置的定位工具。为了应对报告同质性，我们用Sup-sigmoid损失替代了对比损失，后者错误地排斥语义相似的对。Sup-sigmoid损失扩展了SigLIP中的sigmoid损失，使用来自结构化数据的软标签。TopKSigLIP在内部和外部基准测试中，在密度评估、BI-RADS分类、发现子类型和癌症预测的零-shot评估下，均优于现有的开源乳腺X光检查和一般医学VLM。尽管使用了显著较小的视觉编码器和较小的训练批次，TopKSigLIP在线性探测下仍保持竞争力。TopK-Patch模块在病变定位方面也优于后期的Grad-CAM。代码和权重已公开：https://github.com/Youngseok0001/TopKSigLIP。

View on arXiv Download PDF AI Translation

cs.CV / 5 / 2609.03102

WireSeg-32K: A Physics-Grounded Synthetic Dataset for Wire Instance Segmentation

WireSeg-32K：一个基于物理的线实例分割合成数据集

Dai, Zilin, Wang, Lehong, Yang, Yi, Fei, Xiang

Abstract

Deformable linear objects such as wires and cables are difficult to segment because they are thin, highly deformable, and frequently self-occluded, while large-scale instance-level annotations are expensive to obtain in real scenes. Existing resources either focus on cable tracing or semantic segmentation under constrained settings, or generate visually plausible images without physically grounded wire deformation. We present WireSeg-32k, a synthetic dataset for wire instance segmentation with 32,000 RGB images, instance masks, depth maps, and a complementary real-world test set with annotations. To generate this dataset, we develop DeformX, a co-simulation pipeline that couples Cosserat-rod dynamics with photorealistic Isaac Sim rendering, enabling physically plausible, contact-consistent wire shapes, CAD-based wire assets, and diverse visually grounded scenes. As a simple baseline, LoRA fine-tuning SAM3 on WireSeg-32k alone improves real-world mAP@75 by 10.2% over the off-the-shelf model, showing that physically grounded synthetic data can transfer to real wire perception.

Chinese Translation

可变形线性物体如电线和电缆的分割非常困难，因为它们细长、高度可变形且经常自遮挡，而在真实场景中获得大规模实例级标注的成本昂贵。现有资源要么专注于电缆追踪或在受限条件下的语义分割，要么生成视觉上可信的图像而没有物理基础的线变形。我们提出了WireSeg-32k，这是一个用于线实例分割的合成数据集，包含32,000张RGB图像、实例掩膜、深度图以及一个带有标注的补充真实世界测试集。为了生成该数据集，我们开发了DeformX，这是一个将Cosserat杆动力学与逼真的Isaac Sim渲染相结合的共同仿真管道，能够实现物理上合理、接触一致的线形状、基于CAD的线资产以及多样化的视觉基础场景。作为一个简单的基线，在WireSeg-32k上对SAM3进行LoRA微调，使得真实世界的mAP@75比现成模型提高了10.2%，表明基于物理的合成数据可以转移到真实的线感知中。

View on arXiv Download PDF AI Translation

cs.CV / 6 / 2609.03109

SLIDEFORGE: An LLM Agent for Controllable Editing of Slides as Structured Artifacts

SLIDEFORGE：一种用于可控幻灯片编辑的 LLM 代理作为结构化工件

Zheng, Haozhen, Wang, Fulin, Xiong, Tianhu, Yu, Yingjie, Qian, Shengyi, Yu, Hanchao, Schwing, Alex, Nahrstedt, Klara, Wu, Mingyuan

Abstract

Current AI agents compellingly describe slides. However, AI-assisted slide editing requires more than understanding: the output must retain layout, style, component structure, and native editability. Towards, AI-assisted slide editing, existing agents operate on screenshots or weak document representations and often fragment coherent visual units, rasterize editable content, or break layout. In contrast, for controllable slide editing, we introduce an agentic framework, SLIDEFORGE, which builds a Deck State Graph, an executable slide state that links visual decomposition, native pptx object structure, and perceptual organization. By recovering human-referable components while retaining fine-grained editable structure, SLIDEFORGE supports theme-preserving reconstruction through slide-native operations and rendered-state verification. We further introduce an evaluation paradigm for controllable slide transformation that jointly measures component recovery, preservation, restyling consistency, visual quality, and native editability. Experiments show that SLIDEFORGE outperforms direct prompting, screenshot-based agents, and generic code-agent baselines across these dimensions. Code is available at https://github.com/UIUC-MONET/SLIDEFORGE.

Chinese Translation

当前的人工智能代理能够有效地描述幻灯片。然而，人工智能辅助的幻灯片编辑不仅仅需要理解：输出必须保留布局、风格、组件结构和原生可编辑性。为了实现人工智能辅助的幻灯片编辑，现有代理通常在屏幕截图或弱文档表示上操作，常常会破坏连贯的视觉单元、光栅化可编辑内容或打破布局。相比之下，为了实现可控的幻灯片编辑，我们提出了一种代理框架 SLIDEFORGE，它构建了一个 Deck State Graph（幻灯片状态图），这是一个可执行的幻灯片状态，链接了视觉分解、原生 pptx 对象结构和感知组织。通过恢复人类可参考的组件，同时保留细粒度的可编辑结构，SLIDEFORGE 支持通过幻灯片原生操作和渲染状态验证进行主题保留重建。我们进一步引入了一种可控幻灯片转换的评估范式，该范式共同测量组件恢复、保留、样式一致性、视觉质量和原生可编辑性。实验表明，SLIDEFORGE 在这些维度上优于直接提示、基于截图的代理和通用代码代理基线。代码可在 https://github.com/UIUC-MONET/SLIDEFORGE 获取。

View on arXiv Download PDF AI Translation

cs.CV / 7 / 2609.03139

Beyond Small Patches: Black-Box Detection and Purification of Diverse Backdoor Triggers

超越小补丁：多样化后门触发器的黑箱检测与净化

Abdelnaby, Ahmed, Elmahallawy, Mohamed

Abstract

Deep neural networks (DNNs) are increasingly deployed in real-world vision systems, yet their predictions can be covertly manipulated by backdoor attacks, in which malicious triggers cause targeted misclassification while preserving high clean accuracy. Existing defenses often rely on model internals, training data, or clean validation samples, making them difficult to deploy when only black-box access to a trained model is available. We propose TRIM (Trigger Removal by Identifying Manipulated Regions), a deployment-oriented black-box defense that detects and selectively removes backdoor triggers at inference time without requiring model internals, training data, or clean samples. The key insight behind TRIM is to identify image regions that are responsible for anomalous model behavior and purify only those regions while preserving benign content. TRIM innovates via three key components: (i) region-based segmentation with deep feature representations, (ii) adaptive trigger discovery through inpainting and diffusion-based reconstruction to isolate regions responsible for misclassification---without assumptions about trigger type, shape, or location, and (iii) selective region purification that cleans poisoned regions while retaining benign content. To support practical deployment, TRIM further caches feature embeddings of previously identified triggers, enabling efficient recognition and avoiding redundant detection and purification. Extensive experiments across diverse datasets and backdoor types, including blended, sparse, varying-size, and multiple triggers, show that TRIM consistently outperforms existing black-box defenses, reducing attack success rates (ASR) to as low as 1.16% while preserving clean accuracy of up to 87.87%. These results demonstrate that effective backdoor mitigation is possible at inference time even when the defender has no access to any auxiliary data.

Chinese Translation

深度神经网络（DNN）在现实世界的视觉系统中越来越多地被部署，但其预测可能会被后门攻击秘密操控，其中恶意触发器导致目标误分类，同时保持高的清洁准确性。现有的防御措施通常依赖于模型内部、训练数据或清洁验证样本，这使得在仅能黑箱访问训练模型时难以部署。我们提出了 TRIM（通过识别操控区域进行触发器移除），这是一种面向部署的黑箱防御方法，能够在推理时检测并选择性移除后门触发器，而无需模型内部、训练数据或清洁样本。TRIM 的关键见解在于识别导致模型异常行为的图像区域，并仅净化这些区域，同时保留良性内容。TRIM 通过三个关键组件进行创新：（i）基于区域的分割与深度特征表示，（ii）通过修复和基于扩散的重建进行自适应触发器发现，以隔离导致误分类的区域——无需对触发器类型、形状或位置做出假设，以及（iii）选择性区域净化，清理被污染的区域，同时保留良性内容。为了支持实际部署，TRIM 进一步缓存之前识别的触发器的特征嵌入，从而实现高效识别，避免冗余检测和净化。在多种数据集和后门类型（包括混合、稀疏、不同大小和多个触发器）上的广泛实验表明，TRIM 始终优于现有的黑箱防御，将攻击成功率（ASR）降低至最低 1.16%，同时保持高达 87.87% 的清洁准确性。这些结果表明，即使防御者无法访问任何辅助数据，在推理时有效的后门缓解也是可能的。

View on arXiv Download PDF AI Translation

cs.CV / 8 / 2609.03153

VeriPhy: Agentic Physical Reasoning for World Model Evaluation and Refinement

VeriPhy：用于世界模型评估和精炼的主动物理推理

Xu, Wenzhuo, Zhu, Yuchen, Ge, Chongjian, Shen, Xuan, Shi, Jing, Kuen, Jason, Chen, Yongxin, Tao, Molei, McComb, Christopher, Gutiérrez, Noelia Grande, Gu, Jiuxiang

Abstract

Visual fluency in generated video does not imply physical reliability, and a scalar quality score alone is incapable of indicating the obligation a clip violates or the moment it fails. We present VeriPhy, an auditable physical-verification system in which a text-only planner compiles the prompt into typed physical obligations and a statically validated execution plan before any frame is observed. During execution, observations gate and scope only declared calls to frozen low-level experts (e.g., segmentation and tracking, counting, eleven typed physical measurements over the resulting tracks, depth, OCR, and audio-event detection). Each action returns a provenance-carrying evidence record whose payload, when usable, is either a typed measurement or an explicitly tagged learned state. Typed resolvers and fixed composition map usable records to a three-valued state (supported, contradicted, or unknown, surfaced as plausible, implausible, or abstain) with full provenance, so that every verdict is traceable to the evidence that produced it. We anchor evaluation in a 1,500-clip corpus of human-annotated flaw records that localize real generation failures in prompt reference, space, and time. On a 149-clip core carrying 304 such records, VeriPhy accounts for 228, against 164 for a published question-decomposition evaluator given the same clips and the same claims. Recall alone does not separate it from prompting the same backbone monolithically, which reaches 222; what separates them is that each decision retains its evidence record and provenance, making the traces auditable one verdict at a time and usable as the interface through which a critic verdict could be written back into generation.

Chinese Translation

生成视频的视觉流畅性并不意味着物理可靠性，单一的标量质量评分无法指示剪辑违反的义务或失败的时刻。我们提出了VeriPhy，这是一个可审计的物理验证系统，其中一个仅文本的规划器在观察任何帧之前，将提示编译成类型化的物理义务和静态验证的执行计划。在执行过程中，观察仅限于声明的调用，作用于冻结的低级专家（例如，分割和跟踪、计数、基于结果轨迹的十一种类型化物理测量、深度、光学字符识别（OCR）和音频事件检测）。每个动作返回一个携带来源的证据记录，其有效载荷在可用时，既可以是类型化的测量，也可以是显式标记的学习状态。类型化解析器和固定组合将可用记录映射到三值状态（支持、反驳或未知，表现为合理、不合理或放弃），并附带完整的来源信息，以便每个裁决都可以追溯到产生它的证据。我们在一个包含1,500个剪辑的人类注释缺陷记录语料库中锚定评估，该语料库定位了在提示参考、空间和时间中的真实生成失败。在一个包含304个此类记录的149个剪辑核心上，VeriPhy 记录了228个，而对于相同剪辑和相同声明，已发布的问题分解评估器记录了164个。仅凭召回率并不能将其与单一提示相区分，后者达到了222；将它们区分开的是每个决策保留其证据记录和来源，使得每个裁决的追踪可以逐一审计，并可作为批评裁决写回生成的接口。

View on arXiv Download PDF AI Translation

cs.CV / 9 / 2609.03158

Who Speaks for the Pruned? Visual Token Pruning as Coverage Optimization

谁为被修剪的代币发声？视觉代币修剪作为覆盖优化

Zhu, Qingchan, You, Weihang, Jiang, Hanqi, Yang, Changdi, Liu, Tianming, Yuan, Geng

Abstract

Visual token pruning reduces the inference cost of vision-language models (VLMs), but most methods only ask which tokens to keep. This retained-token view can keep redundant high-scoring tokens while leaving discarded evidence without a close representative. We propose CoverPruner, a training-free pruner that asks the complementary demand-side question: after a token is removed, which surviving original token represents it for the target VLM? CoverPruner formulates pruning as Representational Coverage Maximization (RCM), covering the full projected visual-token set with query-weighted demand. It instantiates RCM with projector-space coverage and a lightweight first-layer attention probe. Across multiple VLM architectures and compression rates, CoverPruner achieves the best average accuracy among all compared methods, with the largest gains usually appearing under aggressive compression.

Chinese Translation

视觉代币修剪降低了视觉语言模型（VLMs）的推理成本，但大多数方法仅关注保留哪些代币。这种保留代币的视角可能会保留冗余的高分代币，同时留下被丢弃的证据没有近似的代表。我们提出了CoverPruner，这是一种无训练的修剪器，它提出了互补的需求侧问题：在移除一个代币后，哪个存活的原始代币代表它以满足目标VLM？CoverPruner将修剪公式化为表征覆盖最大化（Representational Coverage Maximization, RCM），通过查询加权的需求覆盖完整的投影视觉代币集。它通过投影空间覆盖和轻量级的第一层注意力探针来实现RCM。在多种VLM架构和压缩率下，CoverPruner在所有比较方法中实现了最佳的平均准确率，通常在激进压缩下获得最大的提升。

View on arXiv Download PDF AI Translation

cs.CV / 10 / 2609.03199

RoboTok: An Internet-Scale Data Engine for Human Demonstration Retrieval and Dexterous Manipulation Learning

RoboTok：一个用于人类示范检索和灵巧操作学习的互联网规模数据引擎

Qian, Howard, Chen, Yiting, Xie, Yunfei, Ren, Kejia, Chanrungmaneekul, Podshara, Wang, Gaotian, Wen, Bowen, Wei, Chen, Hang, Kaiyu

Abstract

Robot learning increasingly depends on broad and diverse demonstrations, yet collecting robot data remains expensive and poorly suited to covering the long tail of real-world tasks. To address this bottleneck, we introduce RoboTok, an internet-scale data engine that, given a query human manipulation video, retrieves manipulation-relevant human demonstrations from web videos for training dexterous robot policies. Specifically, we learn a latent motion space from 3D hand trajectories expressed in estimated actor-centered reference frames. This representation enables manipulation behaviors to be compared across variations in camera viewpoint, scene appearance, and actor occlusions, while remaining compact enough for efficient search and continual indexing over internet-scale video collections. We evaluate RoboTok against existing robot-data retrieval approaches on retrieval benchmarks and downstream robot policy performance. Our results show that RoboTok retrieves more relevant manipulation demonstrations and improves downstream task success, establishing hand-pose trajectory-aware retrieval as a way to make web video a scalable and continuously growing source of supervision for robot learning.

Chinese Translation

机器人学习越来越依赖于广泛而多样的示范，然而收集机器人数据仍然昂贵且难以覆盖现实世界任务的长尾。为了解决这一瓶颈，我们提出了RoboTok，一个互联网规模的数据引擎，它能够根据查询的人类操作视频，从网络视频中检索与操作相关的人类示范，以训练灵巧的机器人策略。具体而言，我们从以估计的演员中心参考框架表达的3D手部轨迹中学习潜在运动空间。这种表示使得操作行为能够在相机视角、场景外观和演员遮挡的变化中进行比较，同时保持足够紧凑，以便在互联网规模的视频集合中高效搜索和持续索引。我们在检索基准和下游机器人策略性能上评估了RoboTok与现有机器人数据检索方法的表现。我们的结果表明，RoboTok能够检索到更相关的操作示范，并提高下游任务的成功率，确立了基于手势轨迹的检索作为将网络视频转变为机器人学习可扩展且持续增长的监督来源的一种方式。

View on arXiv Download PDF AI Translation

cs.CV / 11 / 2609.03206

Learning to Zoom Efficiently with a Contrastive Curriculum

通过对比课程高效学习缩放

Helm, Falko, Gurevych, Iryna

Abstract

Using a zoom-in tool is an important foundational part of modern visual agents, because it allows to efficiently handle tasks involving high-resolution images. Most previous methods need an extensive warm-start supervised fine-tuning phase for teaching models zoom-in. We show that this is not necessary by proposing a new intrinsic reward for learning tool use in MLLMs without the need for additional labels or warm-start SFT. Our InfoNCE-style reward uses a curriculum of increasingly hard negative tool calls as a contrastive training signal. Empirical experiments on $V^*$, HRBench and MME-RealWorld show that our approach is competitive while being more efficient. When used as a drop-in replacement for SFT, we even outperform all baselines. To directly measure the zoom-in ability of models, we further introduce the scalable synthetic Muffin&Chihuahua (M&C) dataset. Each image consists of a grid with every cell either showing a muffin or chihuahua. Leveraging the M&C dataset's unique region of interest labels, we find that recall is the metric that most strongly correlates the zoom-in region with final task performance. Our model and code for reproduction is publicly available under https://github.com/UKPLab/emnlp2026-zoom-in

Chinese Translation

使用缩放工具是现代视觉智能体的重要基础部分，因为它能够高效处理涉及高分辨率图像的任务。大多数先前的方法需要一个广泛的热启动监督微调阶段来教导模型进行缩放。我们通过提出一种新的内在奖励，证明这一点并非必要，该奖励用于在无需额外标签或热启动监督微调的情况下学习工具使用。我们的InfoNCE风格奖励使用越来越困难的负工具调用的课程作为对比训练信号。在$V^*$、HRBench和MME-RealWorld上的实证实验表明，我们的方法在竞争力的同时更加高效。当作为监督微调的替代方案使用时，我们甚至超越了所有基线。为了直接测量模型的缩放能力，我们进一步引入了可扩展的合成Muffin&Chihuahua (M&C)数据集。每个图像由一个网格组成，每个单元格要么显示一个松饼，要么显示一只吉娃娃。利用M&C数据集独特的兴趣区域标签，我们发现召回率是与缩放区域和最终任务性能最强相关的指标。我们的模型和代码可在https://github.com/UKPLab/emnlp2026-zoom-in上公开获取。

View on arXiv Download PDF AI Translation

cs.CV / 12 / 2609.03216

ProgResViT: Progressive Resolution and Width for Adaptive Vision Transformers

ProgResViT：用于自适应视觉变换器的渐进分辨率和宽度

Hojjat, Ali, Haberer, Janek, Landsiedel, Olaf

Abstract

Vision Transformers (ViTs) typically process every image using a fixed input resolution and model width, even though many images can be classified with substantially less computation. We introduce ProgResViT, an input-adaptive ViT that performs inference progressively across multiple rounds. The first round processes a low-resolution image with a narrow subnetwork. Inference terminates when the prediction is sufficiently confident; otherwise, the model reuses the representations produced in the current round and proceeds with a higher-resolution input and a wider subnetwork to refine its prediction. As all rounds share a single backbone, we propose Progress-Conditioned Soft Gating (PSG), which conditions token fusion and layer outputs on the current round, block, and input resolution. On image classification, applying ProgResViT to DeiT yields better accuracy-compute trade-offs than adaptive-width, adaptive-depth, and dynamic-token baselines. With knowledge distillation, a DeiT-based ProgResViT achieves 84.9% top-1 accuracy, slightly exceeding the reported DeiT-III-S accuracy under a comparable evaluation setting. We show that the same design also provides favorable accuracy-compute trade-offs for self-supervised DINO representations and downstream semantic segmentation. Code is available at https://github.com/ds-kiel/ProgResViT.

Chinese Translation

视觉变换器（ViTs）通常使用固定的输入分辨率和模型宽度处理每张图像，尽管许多图像可以通过显著较少的计算量进行分类。我们提出了ProgResViT，这是一种输入自适应的ViT，它在多个轮次中逐步进行推理。第一轮处理低分辨率图像，并使用狭窄的子网络进行处理。当预测结果达到足够的置信度时，推理终止；否则，模型重用当前轮次生成的表示，并继续使用更高分辨率的输入和更宽的子网络来细化其预测。由于所有轮次共享一个单一的主干网络，我们提出了进度条件软门控（Progress-Conditioned Soft Gating，PSG），该方法根据当前轮次、块和输入分辨率对令牌融合和层输出进行条件处理。在图像分类任务中，将ProgResViT应用于DeiT比自适应宽度、自适应深度和动态令牌基线提供了更好的准确性与计算量的权衡。通过知识蒸馏，基于DeiT的ProgResViT达到了84.9%的Top-1准确率，略微超过了在可比评估设置下报告的DeiT-III-S准确率。我们展示了相同的设计也为自监督DINO表示和下游语义分割提供了良好的准确性与计算量的权衡。代码可在https://github.com/ds-kiel/ProgResViT获取。

View on arXiv Download PDF AI Translation

cs.CV / 13 / 2609.03233

Counting Animals in Camera-Traps Image Sequences without Count Labels: Winning Solution to the iWildCam 2021 Challenge

在没有计数标签的情况下计数相机陷阱图像序列中的动物：iWildCam 2021挑战赛的获胜方案

Cunha, Fagner, Colonna, Juan G., Santos, Eulanda M. dos

Abstract

Camera traps have become an essential tool for wildlife monitoring, motivating the development of computer vision methods for the automated extraction of information from these data. While most prior work has focused on species identification, many ecological applications also require estimating the number of unique individuals appearing across short image sequences. This task is particularly challenging because camera traps typically acquire bursts of images at approximately one frame per second, creating large temporal discontinuities that may make conventional multi-object tracking methods unreliable, and because manually collecting individual count annotations is prohibitively expensive. In this work, we describe the winning solution to the iWildCam 2021 Challenge, which introduced a benchmark for counting animals at the sequence level under realistic annotation constraints where count annotations are unavailable for training. Our approach, MaxBoxCount, combines a strong species classification pipeline with a simple yet effective counting heuristic based on MegaDetector detections to estimate the number of unique individuals without requiring count annotations. Code is available at https://github.com/alcunha/iwildcam2021ufam.

Chinese Translation

相机陷阱已成为野生动物监测的重要工具，推动了计算机视觉方法的发展，以自动提取这些数据中的信息。尽管大多数先前的研究集中在物种识别上，但许多生态应用也需要估计在短图像序列中出现的独特个体数量。这个任务尤其具有挑战性，因为相机陷阱通常以每秒大约一帧的速度获取一系列图像，造成较大的时间间断，这可能使传统的多目标跟踪方法不可靠，并且手动收集个体计数注释的成本极高。在本研究中，我们描述了iWildCam 2021挑战赛的获胜方案，该挑战赛引入了一个在现实注释约束下进行序列级动物计数的基准，其中训练时不可用计数注释。我们的方法MaxBoxCount结合了强大的物种分类流程和基于MegaDetector检测的简单而有效的计数启发式，以在不需要计数注释的情况下估计独特个体的数量。代码可在https://github.com/alcunha/iwildcam2021ufam获取。

View on arXiv Download PDF AI Translation

cs.CV / 14 / 2609.03258

An Ensemble-Based Self-Taught Learning Approach for Parking Space Classification Under Limited Data

基于集成的自我学习方法在有限数据下的停车位分类

Cunha, Lucas de Oliveira, Gotz, Joelton Deonei, de Almeida, Paulo Lisboa, Hochuli, Andre Gustavo

Abstract

Parking spot classification is a fundamental task in intelligent transportation systems, yet most deep learning approaches rely on large amounts of annotated data and exhibit limited generalization across heterogeneous environments. To address these limitations, we investigate a self-taught learning framework based on unsupervised representation learning with convolutional autoencoders. The proposed approach learns transferable visual representations from unlabeled data and reuses the learned encoders as fixed feature extractors for supervised classification with limited annotated samples in the target domain. To further enhance robustness and mitigate architectural bias, an ensemble of heterogeneous autoencoders is employed, with independent classifier heads and prediction fusion at inference time. Experiments conducted on the PKLot and CNRPark benchmarks under cross-dataset evaluation protocols show that the proposed ensemble-based strategy substantially reduces annotation requirements while improving robustness under significant domain shifts, achieving accuracies between 93\% and 96\% in data-constrained scenarios.

Chinese Translation

停车位分类是智能交通系统中的一项基础任务，然而大多数深度学习方法依赖于大量的标注数据，并且在异构环境中表现出有限的泛化能力。为了解决这些局限性，我们研究了一种基于卷积自编码器的无监督表示学习的自我学习框架。所提出的方法从未标注数据中学习可迁移的视觉表示，并将学习到的编码器作为固定特征提取器，重新用于目标领域中有限标注样本的监督分类。为了进一步增强鲁棒性并减轻结构偏差，采用了一组异构自编码器，具有独立的分类器头和推理时的预测融合。在PKLot和CNRPark基准数据集上进行的跨数据集评估协议的实验表明，所提出的基于集成的策略显著减少了标注需求，同时在显著的领域转移下提高了鲁棒性，在数据受限的场景中实现了93\%至96etween%的准确率。

View on arXiv Download PDF AI Translation

cs.CV / 15 / 2609.03261

MedQA-MM: Shortcuts Behind Medical Visual Reasoning

MedQA-MM：医学视觉推理背后的捷径

Wang, Benlu, Zhang, Yifan, Yu, Jiaqing, Ong, Chin Siang, Huang, Juncheng, Li, Zhuohao, Zhang, Zhenyu, Cohan, Arman, Yu, Hong, Yao, Zonghai

Abstract

A benchmark score credits final answers, but not the route by which an item can be answered. In medical multimodal multiple-choice questions (MCQs), this distinction matters because a correct answer can be supported by the intended image finding or by benchmark-preserved cues in the wording of answers, non-visual clinical text, visible image text, artificial annotations, or device/context artifacts. We call the resulting score-level overinterpretation reasoning inflation. Here, a route is an observable input path that can support answer selection, not a claim about the model's hidden cognition. Across six medical multimodal MCQ datasets, we separate candidate cues from behavioral evidence through prompt- and image-side audits, modality ablations, and matched repairs that preserve the medical target and answer key. In a 13-configuration open-model panel, full-input accuracy is 62.63%, while text-only and options-only settings achieve 53.96% and 29.71%, respectively. Removing length-gap, absolute/conspicuous, and spatial/prepositional cues lowers accuracy by 6.58, 3.50, and 4.77 percentage points. We also construct MedQA-MM, a 1,000-item shortcut-mitigated subset, where text-only and options-only accuracy fall to 5.21% and 12.33%. This does not imply that models never use images; it shows that medical image-reasoning claims require route-level evidence.

Chinese Translation

基准分数只关注最终答案，而不考虑回答某个项目的路径。在医学多模态多项选择题（MCQs）中，这一区别至关重要，因为正确答案可以通过预期的图像发现或通过保留在答案措辞中的基准线索、非视觉临床文本、可见图像文本、人工注释或设备/上下文伪影来支持。我们称这种基于分数的过度解释推理为推理膨胀。在这里，路径是一个可观察的输入路径，可以支持答案选择，而不是关于模型隐性认知的主张。在六个医学多模态MCQ数据集中，我们通过提示和图像侧审核、模态消融以及匹配修复，分离候选线索与行为证据，这些修复保留了医学目标和答案关键。在一个包含13种配置的开放模型面板中，完整输入的准确率为62.63%，而仅文本和仅选项的设置分别达到53.96%和29.71%。去除长度差、绝对/显著和空间/介词线索的准确率分别降低了6.58、3.50和4.77个百分点。我们还构建了MedQA-MM，一个包含1,000个项目的捷径缓解子集，其中仅文本和仅选项的准确率降至5.21%和12.33%。这并不意味着模型从不使用图像；而是表明医学图像推理的主张需要路径级别的证据。

View on arXiv Download PDF AI Translation

cs.CV / 16 / 2609.03302

Tensor-based Brain Surface Modeling and Analysis

基于张量的脑表面建模与分析

Chung, Moo K., Worsley, Keith J., Robbins, Steve, Evans, Alan C.

Abstract

We present a unified computational approach to tensor-based morphometry in detecting the brain surface shape differences between two clinical groups based on magnetic resonance images. Our approach is novel in a sense that we combined surface modeling, surface data smoothing and statistical analysis in a coherent unified mathematical framework. The cerebral cortex has the topology of a 2D highly convoluted sheet. Between two different clinical groups, the local surface area and curvature of the cortex may differ. It is highly likely that such surface shape differences are not uniform over the whole cortex. By computing how such surface metrics differ, the regions of the most rapid structural differences can be localized. To increase the signal to noise ratio, diffusion smoothing based on the explicit estimation of Laplace-Beltrami operator has been developed and applied to the surface metrics. As an illustration, we demonstrate how this new tensor-based surface morphometry can be applied in localizing the cortical regions of the gray matter tissue growth and loss in the brain images longitudinally collected in the group of children.

Chinese Translation

我们提出了一种统一的计算方法，利用基于张量的形态测量技术检测两组临床患者之间的脑表面形状差异，该方法基于磁共振成像。我们的方法具有创新性，因为我们在一个连贯的统一数学框架中结合了表面建模、表面数据平滑和统计分析。大脑皮层的拓扑结构呈现为一个高度卷曲的二维表面。在两组不同的临床患者之间，皮层的局部表面积和曲率可能存在差异。这种表面形状差异很可能在整个皮层上并不均匀。通过计算这些表面度量的差异，可以定位出结构差异最显著的区域。为了提高信噪比，我们开发并应用了基于拉普拉斯-贝尔特拉米算子显式估计的扩散平滑技术于表面度量。作为示例，我们展示了如何将这种新的基于张量的表面形态测量应用于定位在儿童组中纵向收集的脑图像中灰质组织生长和损失的皮层区域。

View on arXiv Download PDF AI Translation

cs.CV / 17 / 2609.03334

Laplacian Frequency Hierarchies for Efficient 3D Gaussian Splatting Training

高效3D高斯点云训练的拉普拉斯频率层次结构

Yang, Yixiong, Zhang, Sisheng, Yan, Qingsong, Shi, Shaohuai, Wang, Qiang

Abstract

A key bottleneck in 3D Gaussian Splatting training is the continual growth of Gaussian primitives, which increases optimization cost and slows convergence, especially at high resolutions. We propose Laplacian Frequency Hierarchies, a simple yet efficient 3DGS scheme that combines Laplacian image decomposition with coarse-to-fine, frequency-staged training. After fitting lower-frequency structure, we archive the corresponding Gaussian field so that subsequent fields can optimize higher-frequency residuals without carrying the full primitive burden, and we compose the rendered components in the image domain via a Laplacian-style reconstruction at inference time. This design reduces the number of active Gaussians during training, thereby lowering optimization overhead and accelerating training. The proposed scheme is plug-and-play and orthogonal to prior 3DGS accelerations: it can be directly combined with strong backbones such as Taming-3DGS and FastGS to improve training speed with competitive reconstruction quality. It achieves average speedups of 1.73x and 1.21x at 1K setting, and 1.74x and 1.33x at 4K setting on Taming-3DGS and FastGS, with larger gains on more challenging scenes and increasingly pronounced benefits at higher resolutions.

Chinese Translation

3D高斯点云训练中的一个关键瓶颈是高斯原语的持续增长，这增加了优化成本并减缓了收敛速度，尤其是在高分辨率下。我们提出了拉普拉斯频率层次结构（Laplacian Frequency Hierarchies），这是一种简单而高效的3D高斯点云（3DGS）方案，它将拉普拉斯图像分解与粗到细的频率分阶段训练相结合。在拟合低频结构后，我们存档相应的高斯场，以便后续场可以在不承担完整原语负担的情况下优化高频残差，并通过推理时的拉普拉斯风格重建在图像域中组合渲染组件。该设计减少了训练期间活跃高斯的数量，从而降低了优化开销并加速训练。所提出的方案是即插即用的，并且与先前的3DGS加速方法正交：它可以直接与强大的骨干网络（如Taming-3DGS和FastGS）结合，以提高训练速度，同时保持竞争力的重建质量。在1K设置下，它在Taming-3DGS和FastGS上分别实现了1.73倍和1.21倍的平均加速，在4K设置下则实现了1.74倍和1.33倍的加速，在更具挑战性的场景中获得更大的收益，并在更高分辨率下获得越来越明显的好处。

View on arXiv Download PDF AI Translation

cs.CV / 18 / 2609.03341

PointGT: Simultaneous Geometry and Texture Editing for Point-Based Representations

PointGT：基于点的表示的几何与纹理同步编辑

Zhang, Yanshu, Shramko, George, Srinivasan, Pratul P., Li, Ke

Abstract

We present PointGT, a point-based 3D representation that enables simultaneous editing of object geometry and appearance. Existing reconstruction and view synthesis techniques produce volumetric 3D representations that are high-quality and photorealistic, but are difficult to edit. In particular, recent efforts to enable texture editing for 3D Gaussian Splatting representations are not compatible with geometry edits and deformations. Our method combines a point-based representation that is well-suited for geometry deformations with a learned UV mapping technique that enables high-resolution texture editing. We show that PointGT enables fine-grained editing of both geometry and texture in point-based neural representations with high rendering quality.

Chinese Translation

我们提出了PointGT，这是一种基于点的3D表示，能够同时编辑物体的几何形状和外观。现有的重建和视图合成技术生成的体积3D表示虽然高质量且具有照片级真实感，但编辑起来却相对困难。特别是，最近的努力使得3D高斯溅射（Gaussian Splatting）表示的纹理编辑与几何编辑和变形不兼容。我们的方法结合了一种适合几何变形的基于点的表示和一种学习的UV映射技术，从而实现高分辨率的纹理编辑。我们展示了PointGT能够在基于点的神经表示中实现几何和纹理的细粒度编辑，并具有高渲染质量。

View on arXiv Download PDF AI Translation

cs.CV / 19 / 2609.03349

P-CORE: Self-Supervised Surface Consistency for Point-Based Neural Editing

P-CORE：用于基于点的神经编辑的自监督表面一致性

Zhang, Yanshu, Peng, Shichong, Aghabozorgi, Mehran, Moazeni, Alireza, Li, Ke

Abstract

Advances in neural rendering have enabled high-fidelity multi-view reconstruction of 3D scenes. However, free-form non-rigid shape editing remains a significant challenge. Point-based neural representations are highly desirable for multi-view reconstruction because they lack fixed connectivity, which does not constrain the learned surface topology to that of the initialization. Yet this same property causes point-based representations to struggle with holes and surface discontinuities under large deformations. To address this, we propose a novel self-supervised method to enable point-based representations to adapt to large deformations without requiring ground truth multi-view images of deformed geometry. The key idea is to generate random deformations and to ensure consistency in the predicted surface before and after deformation. In particular, the surface prediction from the deformed point cloud should be the same as the deformation applied to the surface prediction from the original point cloud. We incorporate our approach into attention-based point representations, which differ from splatting-based point representations in their use of a learned interpolation kernel between points as opposed to a Gaussian kernel around each point. This learned interpolation kernel can learn to adapt to large deformations, without requiring addition or removal of points. We show that our framework significantly enhances its robustness to large deformations. Experiments on synthetic geometry editing benchmarks (Neural Editor, Objaverse) demonstrate that our approach outperforms existing point-based methods in zero-shot editing and significantly reduces artifacts. Furthermore, qualitative results on the DTU and Mip-NeRF 360 datasets demonstrate our method's effectiveness on real-world scenes.

Chinese Translation

神经渲染的进展使得高保真度的三维场景多视角重建成为可能。然而，自由形状非刚性编辑仍然是一个重大挑战。基于点的神经表示在多视角重建中非常理想，因为它们没有固定的连接性，这不限制学习到的表面拓扑与初始化的相同。然而，这一特性使得基于点的表示在大变形下容易出现孔洞和表面不连续性。为了解决这个问题，我们提出了一种新颖的自监督方法，使基于点的表示能够适应大变形，而无需真实的变形几何多视角图像。关键思想是生成随机变形，并确保在变形前后预测表面的一致性。特别是，从变形后的点云生成的表面预测应与从原始点云生成的表面预测所施加的变形相同。我们将我们的方法融入基于注意力的点表示中，这与基于喷溅的点表示不同，因为它使用的是点之间的学习插值核，而不是围绕每个点的高斯核。这个学习插值核可以学习适应大变形，而无需增加或移除点。我们展示了我们的框架显著增强了其对大变形的鲁棒性。在合成几何编辑基准（Neural Editor，Objaverse）上的实验表明，我们的方法在零样本编辑中优于现有的基于点的方法，并显著减少了伪影。此外，在DTU和Mip-NeRF 360数据集上的定性结果展示了我们方法在真实场景中的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 20 / 2609.03378

When Depth Hurts: Reliability-Aware Geometry Distillation for Depth-Free RGB-D Salient Object Detection

深度的伤害：基于可靠性的几何蒸馏用于无深度RGB-D显著目标检测

Wang, Xuehao, Hua, Jiaxin, Li, Runmei, Wu, Zhenyu, Chen, Chenglizhao, Gu, Ke, Hao, Aimin

Abstract

Depth can resolve appearance ambiguity in RGB-D salient object detection (SOD), yet sensor depth is not uniformly reliable. Missing regions, blurred boundaries, and structural artifacts can propagate through multimodal fusion and make an RGB-D detector less accurate than its RGB-only counterpart. Existing quality-aware approaches regulate observed depth but remain dependent on the same potentially defective modality. We propose \method, a reliability-aware geometry distillation framework developed for RGB-D SOD benchmarks without using dataset-provided depth during training or inference. A frozen Depth Anything V2 model serves only as a training-time teacher, transferring dense relative geometry, hierarchical spatial attention, and boundary structure to a compact edge-aware geometry branch. Pooled bidirectional interaction aligns geometry with appearance, and a pixel-wise reliability estimator selectively injects geometry that is compatible with the current RGB representation. The teacher is removed after training, leaving an RGB-only inference network. Trained on 2,985 RGB-mask pairs, \method{} achieves the best or tied-best result in 26 of 36 metric-dataset comparisons against ten recent RGB-D SOD methods, including a 13.4\% relative MAE reduction on ReDWeb-S. When retrained on DUTS-TR, it also improves the strongest prior $F$-measure by 4.2\% on PASCAL-S, showing that the distilled geometry transfers beyond a particular sensor or dataset domain. Code will be released upon publication.

Chinese Translation

深度能够解决RGB-D显著目标检测（SOD）中的外观模糊性，但传感器深度的可靠性并不均匀。缺失区域、模糊边界和结构伪影可能通过多模态融合传播，使得RGB-D检测器的准确性低于仅使用RGB的检测器。现有的质量感知方法调节观察到的深度，但仍然依赖于同一可能存在缺陷的模态。我们提出了 extit{method}，这是一个基于可靠性的几何蒸馏框架，旨在RGB-D SOD基准测试中不使用数据集提供的深度进行训练或推理。一个冻结的Depth Anything V2模型仅作为训练时的教师，传递密集的相对几何、分层空间注意力和边界结构到一个紧凑的边缘感知几何分支。汇聚的双向交互将几何与外观对齐，而逐像素的可靠性估计器选择性地注入与当前RGB表示兼容的几何。教师在训练后被移除，留下一个仅使用RGB的推理网络。在2985对RGB-掩码上训练后， extit{method}在36个度量-数据集比较中对比十种最新的RGB-D SOD方法，取得了26项最佳或并列最佳的结果，包括在ReDWeb-S上相对平均绝对误差（MAE）减少了13.4\%。在DUTS-TR上重新训练时，它还在PASCAL-S上将最强的先前$F$-度量提高了4.2\%，显示出蒸馏的几何超越了特定的传感器或数据集领域。代码将在发表时发布。

View on arXiv Download PDF AI Translation

cs.CV / 21 / 2609.03384

FoRIS: Progressive Foreground Refinement for Training-Free In-Context Segmentation

FoRIS：无训练上下文分割的渐进前景细化

Hu, Ming, Yin, Jianfu, Dou, Mingyu, Zhang, Miaomiao, Wang, Yao, Hu, Cong, Hu, Bingliang, Wang, Quan

Abstract

In-Context Segmentation (ICS) aims to precisely segment arbitrary semantic concepts, such as objects or parts, given one or a few annotated visual exemplars. In this paper, we revisit ICS from a more classical segmentation perspective, viewing it as a coarse-to-fine progressive refinement process. Rather than directly predicting the final mask through reference-query matching, we progressively refine the segmentation from coarse and ambiguous foreground responses to precise and complete foreground structures. Building upon this perspective, we propose a training-free in-context segmentation framework, termed FoRIS. Specifically, FoRIS consists of three key stages: Foreground Purification, Foreground Localization, and Foreground Consolidation, which progressively suppress background distractions, localize discriminative target regions, and recover complete foreground structures through semantic aggregation. Experimental results demonstrate that FoRIS achieves SOTA performance across semantic and part segmentation tasks, with average improvements of 4.5 and 4.8 mIoU points over existing approaches in the 1-shot and 5-shot settings, respectively. Code: https://github.com/Xi-Mu-Yu/FoRIS.

Chinese Translation

上下文分割（ICS）旨在根据一个或几个带注释的视觉示例，精确分割任意语义概念，如物体或部分。本文从更经典的分割视角重新审视ICS，将其视为一个粗到细的渐进细化过程。我们并不是通过参考-查询匹配直接预测最终的掩膜，而是逐步将分割从粗糙且模糊的前景响应细化为精确且完整的前景结构。在此基础上，我们提出了一种无训练的上下文分割框架，称为FoRIS。具体而言，FoRIS包括三个关键阶段：前景净化、前景定位和前景整合，这些阶段逐步抑制背景干扰、定位具有区分性的目标区域，并通过语义聚合恢复完整的前景结构。实验结果表明，FoRIS在语义和部分分割任务中实现了SOTA（最先进技术）性能，在1-shot和5-shot设置下，分别比现有方法平均提高了4.5和4.8 mIoU点。代码链接： https://github.com/Xi-Mu-Yu/FoRIS。

View on arXiv Download PDF AI Translation

cs.CV / 22 / 2609.03391

Exploring the Potential of Contrastive Language-Image Pre-training for Multi-Source Remote Sensing Data

探索对比语言-图像预训练在多源遥感数据中的潜力

Miao, Xiangyang, Yao, Kelu, Huang, Yekai, Xu, Xiaogang, Xue, Junxiao, Shen, Minjun, Lv, Chenghui, Liu, Shanji, Chen, Yaying, Li, Chao

Abstract

Contrastive language-image learning (CLIP) has become a key paradigm for remote sensing vision-language understanding. However, existing remote sensing contrastive learning methods are mostly built on RGB-oriented CLIP architectures, making it difficult to exploit heterogeneous sensors such as SAR, multi-spectral imaging (MSI), and hyperspectral imaging (HSI). To address this limitation, we propose OmniRSCLIP, an end-to-end contrastive learning framework that supports multi-source sensor inputs for remote sensing vision-language modeling. The key idea is to extend CLIP beyond its fixed RGB input interface without breaking the pretrained visual knowledge. To this end, OmniRSCLIP introduces Spectral-Spatial Basis Decomposition (SSBD), which formulates arbitrary-channel adaptation as a basis recomposition problem: pretrained CLIP patch embeddings provide transferable spatial bases, while wavelength-conditioned coefficients span sensor-specific embedding kernels within a constrained visual prior space. This design avoids forcing heterogeneous sensors into a fixed-channel input space, while aligning them in a unified image-text semantic space. We further introduce a spectral-context-aware mask-based contrastive learning scheme to suppress modality-specific redundant features and enhance fine-grained image-text alignment. Finally, to support multi-modal training, we construct OmniRS5M, the first large-scale remote sensing image-text corpus covering RGB, SAR, MSI, and HSI. Experiments on retrieval, zero-shot classification, and semantic localization show that OmniRSCLIP preserves strong RGB-domain performance while effectively extending CLIP to heterogeneous remote sensing modalities.

Chinese Translation

对比语言-图像学习（Contrastive Language-Image Pre-training, CLIP）已成为遥感视觉-语言理解的关键范式。然而，现有的遥感对比学习方法大多基于RGB导向的CLIP架构，这使得利用异构传感器（如合成孔径雷达（SAR）、多光谱成像（MSI）和高光谱成像（HSI））变得困难。为了解决这一限制，我们提出了OmniRSCLIP，一个端到端的对比学习框架，支持多源传感器输入以进行遥感视觉-语言建模。其关键思想是扩展CLIP超出其固定的RGB输入接口，同时不破坏预训练的视觉知识。为此，OmniRSCLIP引入了光谱-空间基分解（Spectral-Spatial Basis Decomposition, SSBD），将任意通道适配形式化为基重组问题：预训练的CLIP补丁嵌入提供可转移的空间基，而波长条件系数则在受限的视觉先验空间内跨越传感器特定的嵌入核。该设计避免了将异构传感器强行置入固定通道输入空间，同时将它们对齐到统一的图像-文本语义空间。我们进一步引入了一种光谱上下文感知的基于掩码的对比学习方案，以抑制模态特定的冗余特征，并增强细粒度的图像-文本对齐。最后，为了支持多模态训练，我们构建了OmniRS5M，这是第一个覆盖RGB、SAR、MSI和HSI的大规模遥感图像-文本语料库。检索、零样本分类和语义定位的实验表明，OmniRSCLIP在保持强大的RGB领域性能的同时，有效地将CLIP扩展到异构遥感模态。

View on arXiv Download PDF AI Translation

cs.CV / 23 / 2609.03406

Neural-Collapse-guided Task-Free Continual Anomaly Detection

神经崩溃引导的无任务持续异常检测

Kong, Xiaotong, Song, Chaoyang, Zhou, Ziai, Zhang, Jinxia, Zhang, Kanjian, Wei, Haikun

Abstract

Recent years have witnessed growing interest in continual anomaly detection for industrial visual inspection. However, real-world manufacturing environments exhibit unpredictable shifts in data distributions, rendering task-dependent continual learning assumptions impractical. To address this limitation, we formulate industrial anomaly detection as a task-free continual learning problem and propose NC-TFAD, a neural-collapse-inspired, geometry-driven framework for learning from non-stationary data streams without task boundaries. NC-TFAD freezes a pretrained backbone and aligns streaming features to a simplex Equiangular Tight Frame (ETF) prototype space to stabilize representation geometry under non-stationary streams. To satisfy the NC-inspired geometric construction in the absence of real anomalies, we generate synthetic anomaly samples as auxiliary anchors during training. Building on this geometry, we further introduce inter- and intra-class regularization together with a Focal Neural Collapse Contrastive (FNCC) loss to suppress representation drift and improve normal-anomaly separability. Finally, a normal-patch-prototype-guided localization branch constructs calibrated patch-wise deviation maps from normal training samples and fuses them with a weak self-attention prior, producing anomaly heatmaps without pixel-level annotations. Extensive experiments on MVTec AD and VisA show that NC-TFAD consistently outperforms representative task-free continual learning methods adapted from general vision, as well as unified anomaly detection baselines, in both image-level detection and pixel-level localization under the task-free continual learning protocol. These results highlight that geometry-driven modeling offers an effective and robust solution for task-free continual anomaly detection in real-world industrial applications.

Chinese Translation

近年来，工业视觉检测中的持续异常检测引起了越来越多的关注。然而，现实制造环境中数据分布的不可预测变化使得依赖任务的持续学习假设变得不切实际。为了解决这一限制，我们将工业异常检测形式化为一个无任务的持续学习问题，并提出了NC-TFAD，这是一种受神经崩溃启发的几何驱动框架，旨在从无任务边界的非平稳数据流中学习。NC-TFAD 冻结了一个预训练的主干网络，并将流式特征对齐到一个简单的等角紧框架（Equiangular Tight Frame, ETF）原型空间，以在非平稳流中稳定表示几何。为了在缺乏真实异常的情况下满足NC启发的几何构造，我们在训练期间生成合成异常样本作为辅助锚点。在此几何基础上，我们进一步引入类间和类内正则化，并结合焦点神经崩溃对比损失（Focal Neural Collapse Contrastive, FNCC），以抑制表示漂移并提高正常与异常的可分性。最后，一个正常补丁原型引导的定位分支从正常训练样本构建校准的补丁级偏差图，并将其与弱自注意力先验融合，生成无需像素级注释的异常热图。在MVTec AD和VisA上的大量实验表明，NC-TFAD在图像级检测和像素级定位方面，始终优于从通用视觉适配而来的代表性无任务持续学习方法，以及统一异常检测基线，这些结果突显了几何驱动建模为现实工业应用中的无任务持续异常检测提供了有效且稳健的解决方案。

View on arXiv Download PDF AI Translation

cs.CV / 24 / 2609.03415

Mudragen: Geometrically Supervised Generation of Interacting Two-Hand Mudras for Preserving Indian Classical Dance Heritage

Mudragen：几何监督生成互动双手手势以保护印度古典舞遗产

Kamble, Jagadish Kashinath, Mukhopadhyay, Jayanta, Roy, Debaditya, Das, Partha Pratim

Abstract

Automatic generation of hand gestures is essential for the transmission of Indian classical dance and critical for its preservation. Indian classical dance gesture datasets are inherently low-resource, and the canonical Sanskrit definitions of many mudras lack precise textual descriptions, limiting the effectiveness of conventional text-conditioned image generation models. We present \textbf{MudraGen}, a conditional diffusion framework that synthesizes realistic RGB images of \textit{Samyukta Hasta Mudras} -- interactive two-hand gestures from Bharatanatyam (an Indian classical dance form). Unlike prior work on simple hand signs or single-hand gestures, MudraGen introduces geometry-aware supervision to capture the precise coordination, anatomical validity, and cultural nuance of interacting hands. We formulate three geometry-aware objectives: Keypoint Loss for 3D joint alignment, Joint Offset Loss for inter-hand spatial coherence, and Shape Consistency, which serves as an anatomical regularizer by encouraging consistent hand morphology while allowing independent hand poses. Together, these objectives guide the diffusion model toward anatomically plausible and well-coordinated hand configurations, enabling the synthesis of photorealistic and pose-accurate gesture images. Experimental results show that MudraGen surpasses existing state-of-the-art generative approaches in visual realism, anatomical correctness, and preservation of fine hand-pose structure, enabling faithful reproduction of complex Samyukta Hasta mudras. Beyond quantitative gains, its ability to generate culturally grounded and structurally consistent gestures highlights practical applications in cultural preservation and dance education.

Chinese Translation

手势的自动生成对于印度古典舞的传承至关重要，并且对其保护具有重要意义。印度古典舞手势数据集本质上资源匮乏，许多手势的经典梵文定义缺乏精确的文本描述，限制了传统文本条件图像生成模型的有效性。我们提出了 extbf{MudraGen}，一种条件扩散框架，合成现实的 RGB 图像，专注于 extit{Samyukta Hasta Mudras}——来自巴拉特纳蒂亚姆（Bharatanatyam，一种印度古典舞形式）的互动双手手势。与以往对简单手势或单手手势的研究不同，MudraGen 引入了几何感知监督，以捕捉互动手的精确协调、解剖有效性和文化细微差别。我们制定了三个几何感知目标：关键点损失（Keypoint Loss）用于 3D 关节对齐，关节偏移损失（Joint Offset Loss）用于手间空间一致性，以及形状一致性（Shape Consistency），它作为解剖正则化器，通过鼓励一致的手部形态来允许独立的手部姿势。这些目标共同引导扩散模型朝向解剖上合理且协调良好的手部配置，从而实现逼真且姿势准确的手势图像的合成。实验结果表明，MudraGen 在视觉真实感、解剖正确性和细致手势结构的保留方面超越了现有的最先进生成方法，能够忠实再现复杂的 Samyukta Hasta 手势。除了定量提升外，其生成具有文化基础和结构一致的手势的能力突显了在文化保护和舞蹈教育中的实际应用。

View on arXiv Download PDF AI Translation

cs.CV / 25 / 2609.03429

When Do Frozen VLMs Respond to Image-Free Object-Token Edits? An Answer-Key-Free Protocol and What It Reveals

冻结的视觉语言模型何时会响应无图像对象标记的编辑？一种无答案关键的协议及其揭示的内容

Son, Wonbin, Choi, Gyumun, Seo, Junil, Rho, Seungmin, Lee, Mi Young, Kim, Hyungjoon

Abstract

Answering what-if queries about a scene with a VLM usually means injecting the assumption as text or repainting the scene with a generative model. We instead move the edit to the representation level, before the model input. The image is abstracted into a set of object-level tokens, and the original image never enters the VLM. This design rests on an open question: when do frozen VLMs actually respond to such token edits? We introduce an answer-key-free protocol: no post-edit answer is annotated. It scores edits whose answers are logically determined, and audits itself by reversing each scoreable choice. The protocol reveals three structures. The response is not free: explicit edit teaching, not ordinary VQA training, produces it in dense scenes and multiplies it in sparse ones, on all three operations. Once on, it is governed by token cleanliness and density, with deployable detector+segmenter tokens competitive with the oracle and outperforming it on VRSBench. And reading is a separable axis: the image-free token route preserves 92-96% of a matched patch-token baseline's free-text VQA, and the answers measurably depend on the tokens. The response, cleanliness, and reading structures are sign-preserved across two remote-sensing datasets (iSAID, VRSBench) and three frozen LM backbones. We release the probe generator, records, judge logs, and code.

Chinese Translation

使用视觉语言模型（VLM）回答关于场景的假设性查询通常意味着将假设以文本形式注入或使用生成模型重新绘制场景。我们则将编辑移至表示层面，在模型输入之前。图像被抽象为一组对象级标记，原始图像从未进入VLM。该设计基于一个未解的问题：冻结的VLM究竟何时会对这样的标记编辑做出响应？我们引入了一种无答案关键的协议：不对后编辑的答案进行注释。该协议对逻辑上可确定答案的编辑进行评分，并通过反转每个可评分选择来进行自我审计。该协议揭示了三种结构。响应并非自由：明确的编辑教学，而非普通的视觉问答（VQA）训练，在密集场景中产生响应，并在稀疏场景中加倍，适用于所有三种操作。一旦开启，响应受标记的清洁度和密度的控制，具有可部署的检测器+分割器标记在VRSBench上与oracle竞争，并超越其表现。而阅读是一个可分离的轴：无图像的标记路径保留了92-96%的匹配补丁标记基线的自由文本VQA，答案在可测量上依赖于标记。响应、清洁度和阅读结构在两个遥感数据集（iSAID，VRSBench）和三个冻结的语言模型主干中保持符号不变。我们发布了探针生成器、记录、评审日志和代码。

View on arXiv Download PDF AI Translation

cs.CV / 26 / 2609.03445

OCR-EDR: Rendering-Aware Diagnosis and Repair for Closed-Loop OCR Improvement

OCR-EDR：面向渲染的闭环OCR改进诊断与修复

Zhao, Linnan, Liu, Kang, Yu, Hao, Zhan, Jiabo, Sun, Chong, Li, Chen

Abstract

Although document OCR systems perform increasingly well on routine documents, complex formulas, structured text, and long-tail formats remain error-prone. OCR predictions may omit fine-grained content or hallucinate unsupported outputs, while equivalent encodings of the same visible content must be accommodated. Existing OCR evaluation methods mostly report aggregate metrics, offering limited support for analyzing case-level errors and improving OCR performance. We propose OCR-EDR (OCR Error Diagnosis and Repair), a rendering-aware framework that advances from fine-grained diagnosis to iterative repair. Given a source image, an editable OCR prediction, and its rendered image, OCR-EDR first jointly assesses whether the prediction and its rendering are consistent with the source, preserving valid predictions, including rendering-equivalent ones, while diagnosing and localizing genuine errors. It then applies executable edits and may request an updated rendering for iterative reassessment. We construct OCRErrBench from diverse real OCR predictions, covering text and formulas, exact and rendering-equivalent positives, and genuine errors, and develop the DocEDR model to execute the diagnosis--repair loop. On OCRErrBench, DocEDR achieves 94.78% diagnostic accuracy. It repairs 86.23% of erroneous inputs to visual consistency, raises formula Case-F1 by 30.99 percentage points over DOCR-Inspector-7B on DOCRcaseBench, and improves formula CDM by up to 4.62 percentage points on the identified Bad subsets of four OCR systems on UniMER-Test. These results show that OCR-EDR turns fine-grained OCR analysis into verified corrections and performance gains.

Chinese Translation

尽管文档OCR系统在常规文档上的表现越来越好，但复杂公式、结构化文本和长尾格式仍然容易出错。OCR预测可能会遗漏细粒度内容或产生不支持的输出，同时必须容纳相同可见内容的等效编码。现有的OCR评估方法大多报告汇总指标，提供的支持有限，难以分析个案错误并改善OCR性能。我们提出了OCR-EDR（OCR错误诊断与修复），这是一个面向渲染的框架，旨在从细粒度诊断推进到迭代修复。给定源图像、可编辑的OCR预测及其渲染图像，OCR-EDR首先共同评估预测及其渲染是否与源一致，保留有效预测，包括渲染等效的预测，同时诊断和定位真实错误。然后，它应用可执行的编辑，并可能请求更新的渲染以进行迭代重新评估。我们从多样的真实OCR预测中构建了OCRErrBench，涵盖文本和公式、精确和渲染等效的正例，以及真实错误，并开发了DocEDR模型来执行诊断-修复循环。在OCRErrBench上，DocEDR实现了94.78%的诊断准确率。它将86.23%的错误输入修复为视觉一致性，公式Case-F1在DOCRcaseBench上比DOCR-Inspector-7B提高了30.99个百分点，并在UniMER-Test的四个OCR系统识别的Bad子集上提高了公式CDM最多4.62个百分点。这些结果表明，OCR-EDR将细粒度的OCR分析转化为经过验证的修正和性能提升。

View on arXiv Download PDF AI Translation

cs.CV / 27 / 2609.03446

Preserving Knowledge across Space and Time for Continual Video Deepfake Detection

跨越时空保存知识以实现持续的视频深伪检测

Kim, Taehoon, Choi, Jongwook, Jo, Heejae, Park, Byungmin, Choi, Jongwon

Abstract

The continuous emergence of high-quality video deepfakes requires detectors that continually adapt to new forgery patterns, yet existing approaches, which are designed for deepfake images, fail to capture video-specific cues. Unlike deepfake images that contain only spatial artifacts, deepfake videos leave distinct evidence along both spatial and temporal axes, necessitating the separate preservation of each modality during sequential model updates. To overcome this limitation, we introduce a continual deepfake video detection framework, Modality-Specific Frequency Distillation (MSFD), that explicitly decomposes video features into spatial, temporal, and spatiotemporal modalities in the frequency domain. This decomposition enables independent preservation of each modality, as different deepfake video types exhibit varying reliance on spatial and temporal cues across tasks. Furthermore, MSFD adopts a cross-modality decorrelation loss that encourages spatiotemporal representations to remain orthogonal to single-modality cues. Extensive experiments show that our framework achieves stronger adaptation and preserves performance more effectively than state-of-the-art methods across diverse continual deepfake video scenarios.

Chinese Translation

高质量视频深伪的持续出现要求检测器能够不断适应新的伪造模式，然而现有方法主要针对深伪图像，未能捕捉视频特有的线索。与仅包含空间伪影的深伪图像不同，深伪视频在空间和时间轴上都留下了明显的证据，因此在顺序模型更新过程中需要分别保存每种模态。为了解决这一限制，我们提出了一种持续的视频深伪检测框架——模态特定频率蒸馏（Modality-Specific Frequency Distillation, MSFD），该框架在频域中明确将视频特征分解为空间、时间和时空模态。这种分解使得每种模态能够独立保存，因为不同类型的深伪视频在任务中对空间和时间线索的依赖程度各不相同。此外，MSFD采用了一种跨模态去相关损失，鼓励时空表示与单一模态线索保持正交。大量实验表明，我们的框架在多样化的持续深伪视频场景中，能够比现有最先进的方法实现更强的适应性和更有效的性能保持。

View on arXiv Download PDF AI Translation

cs.CV / 28 / 2609.03447

STARS-GS: Structure-Aware Regularized Gaussian Splatting for Large-Scale Aerial Surface Reconstruction

STARS-GS：结构感知正则化高斯喷溅用于大规模航空表面重建

Li, Bocheng, Zhang, Wenjuan, Han, Jie Pan. Dongxu, Ma, Xuesong, Yao, Yiling, Wang, Yaning

Abstract

Large-scale 3D surface reconstruction from aerial imagery is fundamental to geospatial mapping and urban modeling. Recent advances in 3D Gaussian Splatting (3DGS) have demonstrated considerable potential for this task. However, existing methods still face three major challenges in large and complex scenes: scene partitioning may split continuous scene elements across independently optimized sub-regions; geometric constraints mainly focus on the attributes of individual Gaussians while overlooking their local organization; and uniform regularization struggles to accommodate heterogeneous geometric structures. To address these issues, we propose STARS-GS, a structure-aware 3DGS framework for large-scale surface reconstruction. First, we introduce a structure-aware scene partitioning strategy that better preserves continuous scene structures during partitioning and reduces cross-region geometric inconsistencies and stitching artifacts through boundary refinement. Second, we develop neighborhood-aware Gaussian organization that extends geometric constraints from individual primitives to their neighborhood organization, encouraging Gaussians to better conform to local surface geometry. Third, we introduce adaptive surface regularization that adjusts the regularization strength according to local geometric characteristics, promoting geometric consistency in structured regions while preserving plausible variations in unstructured regions. Extensive experiments on large-scale aerial photogrammetry benchmarks demonstrate that STARS-GS consistently outperforms the evaluated Gaussian-based methods in surface reconstruction. It increases the average F1-score from 0.640 for the second-best method to 0.698, corresponding to a relative improvement of approximately 9.1\%, demonstrating effective improvements in geometric accuracy and surface completeness.

Chinese Translation

从航空影像中进行大规模三维表面重建是地理空间制图和城市建模的基础。最近在三维高斯喷溅（3D Gaussian Splatting, 3DGS）方面的进展显示了其在这一任务中的巨大潜力。然而，现有方法在处理大规模复杂场景时仍面临三大挑战：场景划分可能会将连续的场景元素分割到独立优化的子区域中；几何约束主要关注个体高斯的属性，而忽视了它们的局部组织；均匀正则化难以适应异质几何结构。为了解决这些问题，我们提出了STARS-GS，一种用于大规模表面重建的结构感知3DGS框架。首先，我们引入了一种结构感知的场景划分策略，在划分过程中更好地保留连续的场景结构，并通过边界细化减少跨区域的几何不一致性和拼接伪影。其次，我们开发了邻域感知的高斯组织，将几何约束从单个原始体扩展到其邻域组织，鼓励高斯更好地符合局部表面几何。第三，我们引入了自适应表面正则化，根据局部几何特征调整正则化强度，在结构化区域促进几何一致性，同时在非结构化区域保留合理的变化。在大规模航空摄影测量基准上的大量实验表明，STARS-GS在表面重建方面始终优于评估的基于高斯的方法。它将第二优方法的平均F1分数从0.640提高到0.698，约相当于9.1\%的相对提升，显示出在几何精度和表面完整性方面的有效改善。

View on arXiv Download PDF AI Translation

cs.CV / 29 / 2609.03453

Preprocessing Failure and Adversarial Detection in Depthwise-Separable Edge Vision Systems

深度可分离边缘视觉系统中的预处理失败与对抗检测

Mukta, Jannatul Masruk, Sanjida, Rifa, Tory, Adrita Rahman, Rahman, Md. Saifur, Hasan, Khondokar Fida

Abstract

Preprocessing-based defenses are the standard first-line response to adversarial attacks on edge vision systems, requiring no retraining, no architectural changes, and widely recommended as model-agnostic mitigations. Yet the foundational evaluations of these defenses were conducted on residual or Inception-class architectures, not on the depthwise-separable CNNs that dominate edge deployments. This untested assumption leaves a gap in the security evaluation literature. This paper closes that gap by evaluating six preprocessing defenses against adversarial perturbations across both architecture families. Across all perturbation levels and defenses tested, the two depthwise-separable architectures show consistently poor recovery while the residual architecture shows partial recovery; ablation results are consistent with an architectural rather than parametric explanation, though only three architectures and one attack family are evaluated. Crucially, this failure is not merely a negative result. The same output divergence that disqualifies preprocessing as a recovery mechanism reveals a detection opportunity: preprocessing consistently disrupts clean predictions while leaving adversarial predictions largely unchanged, an asymmetry that is directly measurable without retraining or architectural modification. We further show that standard image quality metrics are unreliable proxies for defense effectiveness, a methodological gap in current evaluation practice. A practitioner decision framework is provided for adversarially resilient edge vision deployment.

Chinese Translation

基于预处理的防御是对边缘视觉系统中对抗攻击的标准第一道防线，既无需重新训练，也无需架构更改，并且被广泛推荐作为模型无关的缓解措施。然而，这些防御的基础评估是在残差或Inception类架构上进行的，而不是在主导边缘部署的深度可分离卷积神经网络（CNN）上进行的。这一未经验证的假设在安全评估文献中留下了空白。本文通过评估六种预处理防御在两种架构家族中对抗扰动的效果来填补这一空白。在所有测试的扰动水平和防御中，两个深度可分离架构的恢复效果 consistently poor，而残差架构则显示出部分恢复；消融结果与架构解释而非参数解释一致，尽管仅评估了三种架构和一个攻击家族。至关重要的是，这一失败不仅仅是一个负面结果。使预处理不适合作为恢复机制的相同输出差异揭示了一个检测机会：预处理 consistently disrupts clean predictions，同时对抗预测几乎不变，这种不对称性可以在不重新训练或修改架构的情况下直接测量。我们进一步表明，标准图像质量指标并不是防御有效性的可靠代理，这在当前评估实践中存在方法论缺口。我们为对抗性韧性的边缘视觉部署提供了一个实践者决策框架。

View on arXiv Download PDF AI Translation

cs.CV / 30 / 2609.03463

BMCTrack-d: Pig re-identification and tracking via back marks in challenging camera settings

BMCTrack-d：在复杂摄像头环境下通过背部标记进行猪的重新识别与追踪

Brunner, David, Oczak, Maciej, Bordes, Marie, Rault, Jean-Loup, Winkler, Stephan M., Dorfer, Viktoria

Abstract

Automated pig monitoring is essential for assessing their health, behaviour, and welfare. To date, most pig monitoring solutions operate on the group-level, because individual-level monitoring requires reliable long-term identification and tracking of each animal. For domesticated pigs this remains challenging because pigs of the same breed often have highly uniform appearances. Moreover, research on pig monitoring is almost exclusively reported in top-down view camera settings, which considerably ease tracking, but are not always an option in practice. In this work, BMCTrack-d is presented, a novel tracking-by-detection approach that leverages unique back marks to enable robust pig re-identification and tracking in a challenging side-view camera setting, afflicted by rapidly moving pigs, severe occlusions and low resolution. The method first predicts the detected pigs' identities using a neural network-based back mark classifier. To improve re-identification reliability over time, two dedicated post-processing stages are introduced: a temporal prediction consistency check, which validates the identity assignments against the recent prediction history, and deduplication, which resolves conflicting identity assignments in each time step. By explicitly prioritising accurate, appearance-based re-identification over continuous tracking, the proposed approach addresses a key limitation of existing trackers for individual-level monitoring scenarios. On a demanding test set BMCTrack-d outperforms two strong baselines, BoT-SORT-ReID and TrackTrack-ReID, by 9.11% and 1.03%, respectively, in higher-order tracking accuracy. These results demonstrate the effectiveness of back mark-based re-identification and tracking for robust individual-level pig monitoring in challenging settings.

Chinese Translation

自动化猪监测对于评估其健康、行为和福利至关重要。迄今为止，大多数猪监测解决方案在群体层面上运行，因为个体层面的监测需要对每只动物进行可靠的长期识别和追踪。对于家猪而言，这仍然具有挑战性，因为同一品种的猪往往外观高度一致。此外，关于猪监测的研究几乎仅在俯视摄像头设置中报告，这在很大程度上简化了追踪，但在实际应用中并不总是可行。在本研究中，提出了BMCTrack-d，这是一种新颖的基于检测的追踪方法，利用独特的背部标记来实现猪在复杂的侧视摄像头环境中的鲁棒重新识别与追踪，该环境受到快速移动的猪、严重遮挡和低分辨率的影响。该方法首先使用基于神经网络的背部标记分类器预测检测到的猪的身份。为了提高随时间推移的重新识别可靠性，引入了两个专门的后处理阶段：时间预测一致性检查，通过验证身份分配与最近的预测历史来进行验证，以及去重，解决每个时间步中的身份冲突分配。通过明确优先考虑基于外观的准确重新识别而非持续追踪，所提出的方法解决了现有追踪器在个体层面监测场景中的一个关键限制。在一个具有挑战性的测试集上，BMCTrack-d在高阶追踪准确性方面分别比两个强基线BoT-SORT-ReID和TrackTrack-ReID提高了9.11%和1.03%。这些结果证明了基于背部标记的重新识别与追踪在复杂环境中进行鲁棒个体层面猪监测的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 31 / 2609.03475

SafeRestore: Detector-Relative Risk Certificates for Selective Industrial Image Restoration

安全恢复：针对选择性工业图像恢复的检测器相关风险证书

Yang, Shaoliang, Wang, Jun

Abstract

Industrial inspection pipelines often restore a measured image before a detector acts on it, yet restoration can suppress detector-supported defect structure or create clean-region activations. We formulate restoration as a selective action problem over the measured display, five restored candidates, and review. SafeRestore ranks candidates with action-specific fitted scores, chooses a gate on threshold-tuning data, and evaluates the fixed gate on a disjoint certification sample with two one-sided exact binomial bounds: one for the positive-conditional evidence-loss incident rate and one for the all-accepted excess-activation incident rate. The guarantee is marginal for one policy fixed before its certification outcomes are observed, under an image-level i.i.d. working model. In a retrospective split-sample study of 4,591 public Carinthia-S images, the protocol yields auditable risk-coverage behavior. The primary all-action policy passes in one of five training repetitions (12.0% +/- 26.9% pass-gated test coverage when failures count as zero), whereas fixed bicubic and reduced-complexity variants pass more often. On reserved morphologies, evidence-loss incidence rises to 81.1-90.3%, and KolektorSDD lacks both detector competence and enough positive certification images for the stated target. The contribution is therefore an auditable, detector-relative framework for deciding when a transformed image may be returned automatically and when review remains necessary -- not a claim that adaptive routing outperforms simpler policies on the present evidence.

Chinese Translation

工业检测流程通常在检测器对测量图像进行操作之前恢复该图像，然而恢复可能会抑制检测器支持的缺陷结构或产生干净区域的激活。我们将恢复问题表述为在测量显示、五个恢复候选图像和审查之间的选择性行动问题。SafeRestore 使用特定于行动的拟合评分对候选图像进行排名，选择一个在阈值调优数据上的门，并在一个不相交的认证样本上评估固定门，使用两个单侧精确二项界限：一个用于正条件证据损失事件率，另一个用于所有接受的过度激活事件率。在图像级独立同分布（i.i.d.）工作模型下，对于在其认证结果被观察之前固定的一项政策，保证是边际的。在对 4,591 张公共 Carinthia-S 图像的回顾性分样本研究中，该协议产生了可审计的风险覆盖行为。主要的全行动政策在五次训练重复中的一次通过（当失败计为零时，测试覆盖率为 12.0% +/- 26.9%），而固定的双三次插值和降低复杂度的变体通过的频率更高。在保留的形态上，证据损失发生率上升至 81.1-90.3%，而 KolektorSDD 在检测器能力和足够的正认证图像方面均不足以满足所述目标。因此，贡献在于提供一个可审计的、检测器相关的框架，以决定何时可以自动返回转换后的图像，何时仍需进行审查——而不是声称自适应路由在当前证据上优于更简单的政策。

View on arXiv Download PDF AI Translation

cs.CV / 32 / 2609.03480

Tree species mapping in Denmark: A comparison of spectral-temporal features with geospatial foundation model embeddings

丹麦树种制图：光谱-时间特征与地理空间基础模型嵌入的比较

Koukos, Alkiviadis, Kondylatos, Spyros, Nord-Larsen, Thomas, Nyborg, Lotte, Tøttrup, Christian, Grogan, Kenneth

Abstract

We map tree species across Denmark using National Forest Inventory plots and EO data, while evaluating the potential of foundation models for large-scale forest characterization. We compare two alternative input representations for tree species classification: (i) manually engineered spectral-temporal features (STF) derived from multi-temporal Sentinel-1 and Sentinel-2 observations, and (ii) embeddings generated by the EO FMs TESSERA and AlphaEarth. Both representations are complemented with canopy height information. Random forest, XGBoost, and Multi-Layer Perceptron (MLP) classifiers are evaluated for all input representations, with separate assessments for pure and mixed forest stands. The STF-based MLP achieves the highest classification performance, yielding macro F1 scores of 0.843 and 0.653 for pure and mixed stands, respectively. The MLP trained on TESSERA embeddings delivers competitive performance for pure stands, achieving results within 1.1 percentage points of the best-performing model. TESSERA consistently outperforms STF-based models when fewer than approximately 25% of training plots are available, demonstrating a substantial advantage under limited training data. Multi-year observations systematically improve classification accuracy relative to single-year inputs, while ablation experiments reveal the complementary contributions of Sentinel-1 backscatter, spectral indices, and canopy height data. The best-performing model is subsequently applied at the national scale to generate a 10 m tree species map of Denmark. Area-adjusted validation indicates an overall map accuracy of 79.9%. The resulting map, released as an open-access product, is the first high-resolution national tree species map of Denmark and provides a valuable resource for forest monitoring, ecological research, and land management applications.

Chinese Translation

我们利用国家森林清查样地和遥感数据对丹麦的树种进行制图，同时评估基础模型在大规模森林特征化中的潜力。我们比较了两种树种分类的替代输入表示：(i) 从多时相的 Sentinel-1 和 Sentinel-2 观测中提取的人工工程光谱-时间特征（STF），以及 (ii) 由遥感基础模型 TESSERA 和 AlphaEarth 生成的嵌入。这两种表示都结合了冠层高度信息。我们评估了随机森林、XGBoost 和多层感知器（MLP）分类器在所有输入表示下的表现，并对纯林和混交林进行了单独评估。基于 STF 的 MLP 实现了最高的分类性能，纯林和混交林的宏观 F1 分别为 0.843 和 0.653。基于 TESSERA 嵌入训练的 MLP 在纯林中的表现具有竞争力，其结果与最佳模型相差仅 1.1 个百分点。当训练样地少于约 25% 时，TESSERA 始终优于基于 STF 的模型，显示出在有限训练数据下的显著优势。多年的观测系统地提高了相对于单年输入的分类准确性，而消融实验揭示了 Sentinel-1 后向散射、光谱指数和冠层高度数据的互补贡献。随后，将表现最佳的模型应用于全国范围，生成了丹麦的 10 米树种地图。区域调整验证表明整体地图准确率为 79.9%。该地图作为开放获取产品发布，是丹麦首个高分辨率国家树种地图，为森林监测、生态研究和土地管理应用提供了宝贵资源。

View on arXiv Download PDF AI Translation

cs.CV / 33 / 2609.03516

Residual Optimal Transport-Based Experts Collaboration Towards Modality-Aware Infrared-Visible Object Detection

基于残差最优传输的专家协作实现模态感知的红外-可见物体检测

Zhao, Yue, Yu, Hua, Zhao, Yukun, Zhang, Yuzhi, Gong, Maoguo, Mei, Xin, Hu, Zhuping, Li, Yanchi, Qin, A. K.

Abstract

Infrared-visible object detection (IVOD) integrates complementary evidence from visible and infrared sensors for reliable perception in challenging scenes. In practice, sensors may fail or drop frames, leaving one modality unavailable or intermittent. Existing methods for IVOD assume both modalities are always present, and fixed fusion collapses when one stream is missing. Furthermore, it remains a critical challenge to reliably estimate semantic correlation across heterogeneous modalities, especially under spectral distribution discrepancy. We present FlexibleFusion, a unified and adaptive method that flexibly allocates integration pathways and fusion strength, operating seamlessly across complete and missing-modality regimes. At its core, the Modality-Aware Experts Collaboration (MAEC) mechanism selectively activates and aggregates cross-modal or intra-modal expert pathways. It allows cross-modal fusion when full modalities are available and falls back to self-fusion under missing conditions. Additionally, we design Residual Self-Paced Entropic Optimal Transport (RSPEOT) to align heterogeneous feature distributions from a transport perspective. Instead of relying on the fixed sparsity coefficient in standard entropic optimal transport (EOT), RSPEOT introduces a residual-driven self-paced update that prioritizes reliable matches and progressively refines harder ones. This design alleviates the additional optimization burden of standard EOT while preserving reliable semantic alignment. Comprehensive experiments under complete and missing-modality protocols show consistent performance across arbitrary modality configurations. Code will be released upon publication.

Chinese Translation

红外-可见物体检测（IVOD）整合来自可见光和红外传感器的互补证据，以在具有挑战性的场景中实现可靠感知。在实际应用中，传感器可能会出现故障或丢帧，导致某一模态不可用或间歇性可用。现有的IVOD方法假设两种模态始终存在，而固定融合在缺失某一流时会崩溃。此外，在光谱分布差异的情况下，可靠估计异构模态之间的语义关联仍然是一个关键挑战。我们提出了FlexibleFusion，这是一种统一且自适应的方法，灵活分配集成路径和融合强度，能够在完整模态和缺失模态的情况下无缝操作。其核心是模态感知专家协作（MAEC）机制，选择性地激活和聚合跨模态或同模态的专家路径。当完整模态可用时，它允许跨模态融合，而在缺失条件下则回退到自我融合。此外，我们设计了残差自适应熵最优传输（RSPEOT），从传输的角度对异构特征分布进行对齐。与标准熵最优传输（EOT）中依赖固定稀疏系数不同，RSPEOT引入了一种基于残差驱动的自适应更新，优先考虑可靠匹配并逐步细化更难的匹配。这一设计减轻了标准EOT的额外优化负担，同时保持了可靠的语义对齐。在完整和缺失模态协议下的全面实验表明，在任意模态配置中表现一致。代码将在发表后发布。

View on arXiv Download PDF AI Translation

cs.CV / 34 / 2609.03520

Neural Video Compression Based on Deformable Temporal Alignment and Difference-aware Fusion

基于可变形时间对齐和差异感知融合的神经视频压缩

Shan, Chuyue, Sun, Songlin, Chenwei, Wang, Zihan, Shen

Abstract

In conditional coding-based neural video compression, the quality of temporal context directly affects compression per- formance. Existing methods mostly construct context from prop- agated reference features, but they are vulnerable to motion esti- mation and local alignment errors in regions with complex mo- tion, occlusion, and high-frequency textures, resulting in inaccu- rate temporal information. To address this issue, this paper pro- poses a method combining deformable temporal alignment and difference-aware spatial selective fusion. A Context-aware Tem- poral Alignment Module is used to generate complementary tem- poral context, while a Difference-aware Spatial Selective Fusion module adaptively selects reliable temporal information and sup- presses misalignment. Experiments show that the proposed method achieves certain rate-distortion performance improve- ment over DCVC-DC.

Chinese Translation

在基于条件编码的神经视频压缩中，时间上下文的质量直接影响压缩性能。现有方法大多通过传播的参考特征构建上下文，但在复杂运动、遮挡和高频纹理区域，它们容易受到运动估计和局部对齐错误的影响，从而导致时间信息的不准确。为了解决这一问题，本文提出了一种结合可变形时间对齐和差异感知空间选择性融合的方法。使用上下文感知时间对齐模块生成互补的时间上下文，同时差异感知空间选择性融合模块自适应地选择可靠的时间信息并抑制错位。实验表明，所提出的方法在率失真性能上相较于DCVC-DC取得了一定的提升。

View on arXiv Download PDF AI Translation

cs.CV / 35 / 2609.03534

TruncGradGS: Improved 3D Gaussian Splatting via Truncated Gradient Updates

TruncGradGS：通过截断梯度更新改进的3D高斯点云渲染

Morales, Theo, Le-Pham, Nhat-Quynh, Atkins, Robin, Hua, Binh-Son

Abstract

3D Gaussian Splatting has become a de facto scene representation for novel view synthesis, yet robustly learning 3D Gaussian primitives from visual input remains challenging. Standard optimization relies on gradient-based updates, but a common issue is the gradient vanishing phenomenon: a pixel far from a Gaussian primitive often has diminishing gradient magnitudes to influence primitive attributes, resulting in suboptimal scene reconstruction. In this paper, we propose a method to address gradient vanishing with a piecewise truncated gradient formulation that improves the optimization stability and robustness to initializations. We show that our method consistently improves 3D Gaussian Splatting with random and COLMAP initializations while being generalizable across static and dynamic Gaussian Splatting. As a by-product, we also examine the limitations of current benchmarks for dynamic scenes, and introduce a novel dataset for benchmarking dynamic Gaussian Splatting using synthetic 3D scenes. We demonstrate the effectiveness of our method in both static and dynamic settings for the public benchmarks and our proposed dataset.

Chinese Translation

3D高斯点云渲染已成为新视图合成的事实场景表示方法，但从视觉输入中稳健地学习3D高斯原语仍然具有挑战性。标准优化依赖于基于梯度的更新，但一个常见的问题是梯度消失现象：远离高斯原语的像素通常具有较小的梯度幅度，难以影响原语属性，导致场景重建不理想。本文提出了一种通过分段截断梯度公式来解决梯度消失的方法，从而提高优化的稳定性和对初始化的鲁棒性。我们展示了我们的方法在随机和COLMAP初始化下始终能改善3D高斯点云渲染，并且在静态和动态高斯点云渲染中具有良好的通用性。作为附带成果，我们还考察了当前动态场景基准的局限性，并引入了一个新的数据集，用于基于合成3D场景对动态高斯点云渲染进行基准测试。我们在公共基准和我们提出的数据集上展示了我们的方法在静态和动态设置中的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 36 / 2609.03544

SafeRI: Recognition and Intervention for Token-Level Safety Intervention in Large Vision Language Models

SafeRI：大规模视觉语言模型中的标记级安全干预的识别与干预

Ma, Caoyuan, Gu, Tian, Liu, Wenpu, Xie, Weichu, Dong, Shuai, Xu, Yuqi, Zhao, Ji, Wang, Ziyue, Chang, Wenzheng, Wu, Taiqiang, Zhu, Yongfu, Shao, Wenqi, Wang, Zheng, Zheng, Yinqiang

Abstract

Existing safety alignment methods for vision-language models usually modify the model behavior globally: once the safety parameters are trained or loaded, they participate in both unsafe and already-safe generations. This always-on intervention can unnecessarily perturb the model's original reasoning path and degrade general multimodal capabilities. We argue that safety alignment should be an on-demand intervention rather than a permanent modification to every decoding trajectory. To this end, we propose a streaming recognition and gated LoRA framework for intrinsic VLM safety. During autoregressive generation, a lightweight recognizer estimates whether the current pre-token generation state is safe or unsafe. Its output updates the LoRA gate for the following decoding step; otherwise, generation follows the frozen-backbone policy. The LoRA module is trained from unsafe prefixes, transition statements, and safe continuations, so that it learns to redirect unsafe generations back to safe responses after activation. Experiments across multiple safety and general-purpose benchmarks demonstrate the effectiveness of our method in post-alignment settings.

Chinese Translation

现有的视觉语言模型安全对齐方法通常在全局范围内修改模型行为：一旦安全参数被训练或加载，它们就会参与到不安全和已经安全的生成中。这种始终开启的干预可能会不必要地扰动模型的原始推理路径，并降低其一般多模态能力。我们认为安全对齐应当是一种按需干预，而不是对每个解码轨迹的永久修改。为此，我们提出了一种用于内在视觉语言模型安全的流式识别和门控 LoRA 框架。在自回归生成过程中，一个轻量级的识别器评估当前的预标记生成状态是安全还是不安全。其输出更新下一个解码步骤的 LoRA 门；否则，生成遵循冻结主干策略。LoRA 模块从不安全前缀、过渡语句和安全延续中训练，以便在激活后学习将不安全生成重定向回安全响应。在多个安全和通用基准测试中的实验表明我们的方法在后对齐设置中的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 37 / 2609.03554

WIDE: Wildcard Inference with Dynamic Expansion for Cross-Modal Generative Retrieval

WIDE：用于跨模态生成检索的动态扩展通配符推理

Guo, Teng, Wang, Xin, Xu, Jiayou, Zhou, Keying, Shen, Jifeng, Ruan, Haoxin

Abstract

Generative retrieval has demonstrated significant success by unifying representation learning and search into a single sequence-to-sequence generation task. However, extending this paradigm to cross-modal retrieval reveals a critical challenge arising from the inherent information asymmetry across different modalities, such as the gap between concise text queries and dense visual candidates. This structural mismatch causes the autoregressive decoder to suffer from forced hallucination when generating identifiers via standard trie-constrained beam search, where the model is severely penalized for failing to guess fine-grained details absent from the query, allowing irrelevant candidates to hijack top rankings. To address this issue, we propose Wildcard Inference with Dynamic Expansion (WIDE). WIDE employs Adaptive Entropy Thresholding (AET) to calibrate layer-specific uncertainty boundaries offline. During the decoding generation phase, Asymmetry-aware Wildcard Decoding (AWD) detects semantic blind spots and emits wildcards instead of forced deterministic identifiers, dynamically expanding the search space without incurring log-probability penalties. Finally, Blind-Spot Re-ranking (BSR) evaluates the expanded candidate pool using a hybrid scoring mechanism that combines discrete generation confidence with continuous semantic similarity. Extensive experiments on the M-BEIR benchmark demonstrate that WIDE outperforms state-of-the-art generative retrieval methods, effectively suppressing forced hallucination while maintaining compact index structures.

Chinese Translation

生成检索通过将表示学习与搜索统一为单一的序列到序列生成任务，展示了显著的成功。然而，将这一范式扩展到跨模态检索时，暴露出一个关键挑战，即不同模态之间固有的信息不对称，例如简洁文本查询与密集视觉候选之间的差距。这种结构性不匹配导致自回归解码器在通过标准的字典约束束搜索生成标识符时遭遇强制幻觉，模型因未能猜测查询中缺失的细粒度细节而受到严重惩罚，使得无关候选能够劫持顶级排名。为了解决这个问题，我们提出了动态扩展通配符推理（WIDE）。WIDE采用自适应熵阈值（AET）离线校准特定层的不确定性边界。在解码生成阶段，非对称感知通配符解码（AWD）检测语义盲点，并发出通配符而不是强制确定性标识符，动态扩展搜索空间而不产生对数概率惩罚。最后，盲点重排序（BSR）使用混合评分机制评估扩展的候选池，该机制结合了离散生成置信度与连续语义相似性。在M-BEIR基准上的大量实验表明，WIDE在生成检索方法中优于现有的最先进技术，有效抑制了强制幻觉，同时保持紧凑的索引结构。

View on arXiv Download PDF AI Translation

cs.CV / 38 / 2609.03557

Building Pretraining Data for World Models: An Unreal Engine-Based Pipeline for Action-Conditioned Video Generation

构建世界模型的预训练数据：基于虚幻引擎的动作条件视频生成管道

Wang, Haoyu, Zhang, Songchun, Li, Haoran, Huang, Haoyang, Xue, Zeyue, Duan, Nan

Abstract

Action-conditioned video models require large-scale visual data paired with control signals that are temporally aligned with the resulting scene transitions. Such supervision is difficult to obtain from ordinary real-world video because the actions that caused each visual change are typically unknown. We present a large-scale synthetic data production pipeline built on Unreal Engine for generating action-conditioned, multi-view video. To accommodate the different execution requirements of real-time physics and high-quality offline rendering, the pipeline executes trajectory generation and final rendering in two stages: Stage I runs real physics in PIE and records per-frame character states, control inputs, and camera states into an intermediate trajectory representation; Stage II replays those trajectories in a new engine process and renders them offline with Movie Render Queue (MRQ). Around this core, we develop a distributed production system with cache-aware task partitioning, node-local slot scheduling, automated scene screening, aesthetic and luminance filtering, partial-output recovery, asynchronous upload, and continuous cluster health monitoring. The production cluster contains 25 servers with eight NVIDIA RTX 5090 GPUs per server. From 2,384 asset packs, 429 levels were retained for production together with a pool of 40 humanoid characters. The pipeline has produced 2,691 hours of 1080p video and 6,076 hours of 720p video. We describe the system architecture, the implementation decisions that emerged from production failures, and the limitations of using perceptual quality proxies for world-model data curation. The pipeline described in this report constitutes the Unreal Engine synthetic-data production component used in EchoWM.

Chinese Translation

动作条件视频模型需要与控制信号时间对齐的大规模视觉数据，这些信号与生成的场景转换相匹配。然而，从普通的现实世界视频中获取这样的监督信息是困难的，因为导致每个视觉变化的动作通常是未知的。我们提出了一种基于虚幻引擎的大规模合成数据生产管道，用于生成动作条件的多视角视频。为了满足实时物理和高质量离线渲染的不同执行要求，该管道将轨迹生成和最终渲染分为两个阶段：第一阶段在 PIE 中运行真实物理，并将每帧角色状态、控制输入和相机状态记录到中间轨迹表示中；第二阶段在新的引擎进程中重放这些轨迹，并使用 Movie Render Queue (MRQ) 进行离线渲染。在这一核心基础上，我们开发了一个分布式生产系统，具有缓存感知的任务划分、节点本地插槽调度、自动场景筛选、美学和亮度过滤、部分输出恢复、异步上传和持续集群健康监控。该生产集群包含 25 台服务器，每台服务器配备八个 NVIDIA RTX 5090 GPU。从 2,384 个资产包中，保留了 429 个关卡用于生产，并配备了 40 个类人角色。该管道已生成 2,691 小时的 1080p 视频和 6,076 小时的 720p 视频。我们描述了系统架构、从生产失败中得出的实施决策，以及使用感知质量代理进行世界模型数据管理的局限性。本报告中描述的管道构成了在 EchoWM 中使用的虚幻引擎合成数据生产组件。

View on arXiv Download PDF AI Translation

cs.CV / 39 / 2609.03563

FlashRender: Few-Step Generative Rendering via Camera-Controlled Video MeanFlow

FlashRender：通过摄像机控制的视频均流实现的少步生成渲染

Park, Byeongjun, Kim, Byung-Hoon, Chung, Hyungjin

Abstract

We present FlashRender, a few-step generative rendering framework that retakes a source video along a target camera trajectory in seconds. We identify sampling-step-dependent camera control as a prominent manifestation of discretization error in existing multi-step generative rendering models and show that resolving this inconsistency substantially lowers denoising trajectory curvature, facilitating subsequent step distillation. To this end, we introduce Representation Transformation and Alignment (RETA), which aligns hidden source-video representations with target-video features from a frozen visual geometry model. This directly encodes the geometric transformation within the source-video stream, enabling sampling-step-consistent camera control. We then fine-tune the model with the MeanFlow objective on the lower-curvature denoising trajectory induced by RETA, allowing the model to more effectively address discretization error. Finally, we apply on-policy flow map distillation to correct self-rollout errors under fixed few-step sampling. Extensive experiments show that RETA, MeanFlow, and on-policy flow map distillation play complementary roles in few-step generative rendering. Together, they enable our approach to match multi-step baselines in video quality and geometric consistency at 25x lower sampling cost while achieving superior camera controllability, even under out-of-distribution target camera trajectories.

Chinese Translation

我们提出了FlashRender，一个少步生成渲染框架，可以在几秒钟内沿目标摄像机轨迹重新拍摄源视频。我们识别出采样步长依赖的摄像机控制是现有多步生成渲染模型中离散化误差的一个显著表现，并展示了解决这一不一致性显著降低去噪轨迹曲率，从而促进后续步骤的蒸馏。为此，我们引入了表示变换与对齐（Representation Transformation and Alignment, RETA），它将隐藏的源视频表示与来自冻结视觉几何模型的目标视频特征对齐。这直接编码了源视频流中的几何变换，使得采样步长一致的摄像机控制成为可能。随后，我们在RETA引起的低曲率去噪轨迹上使用均流（MeanFlow）目标对模型进行微调，使模型更有效地解决离散化误差。最后，我们应用在线流图蒸馏（on-policy flow map distillation）来修正固定少步采样下的自回归错误。大量实验表明，RETA、均流和在线流图蒸馏在少步生成渲染中发挥了互补作用。它们共同使我们的方法在视频质量和几何一致性上与多步基线相匹配，同时在采样成本上降低了25倍，并在目标摄像机轨迹超出分布的情况下实现了更优的摄像机可控性。

View on arXiv Download PDF AI Translation

cs.CV / 40 / 2609.03569

Occlusion-Robust Multimodal Emotion Recognition in VR via Fusion of Facial Images and EMG

通过面部图像与肌电图融合实现虚拟现实中的遮挡鲁棒多模态情感识别

Nierula, Birgit, Tomotaki-Dawoud, Karam, Akguel, Mert, Lafci, Mustafa Tevfik, Przewozny, David, Hilsmann, Anna, Eisert, Peter, Bosse, Sebastian

Abstract

Head-mounted displays (HMDs) fundamentally limit emotion recognition in virtual reality (VR): by occluding the upper face, they render conventional image-based facial expression analysis incomplete, particularly for applications requiring real-time affective assessment. We address this challenge by fusing lower-face video with facial electromyography (EMG) from the occluded upper face to classify seven emotional categories (six basic emotions plus neutral). We introduce a synchronized multimodal dataset from 20 participants, pairing lower-face video with seven-channel upper-face EMG elicited by validated emotion stimuli. Under subject-independent test, our proposed late-fusion architecture merging convolutional visual embeddings with RBF-kernel EMG representations achieves 51% macro-F1, outperforming both image-only (41%) and EMG-only (43%) baselines. These results demonstrate that upper-face EMG provides robust complementary information under HMD-induced visual occlusion and establish a foundation for multimodal emotion recognition in naturalistic VR environments. This approach facilitates affect-adaptive applications, including communication training and therapeutic interventions. The dataset will be shared upon request under an ethical-use agreement.

Chinese Translation

头戴式显示器（HMD）在虚拟现实（VR）中根本限制了情感识别：通过遮挡上半部分面部，它使得基于图像的面部表情分析不完整，特别是在需要实时情感评估的应用中。我们通过将下半部分面部视频与来自被遮挡的上半部分面部的面部肌电图（EMG）融合，来解决这一挑战，以分类七种情感类别（六种基本情感加中性）。我们引入了一个来自20名参与者的同步多模态数据集，将下半部分面部视频与由经过验证的情感刺激引发的七通道上半部分面部EMG配对。在无关受试者的测试中，我们提出的后融合架构将卷积视觉嵌入与RBF核EMG表示相结合，达到了51%的宏观F1值，超越了仅使用图像（41%）和仅使用EMG（43%）的基线。这些结果表明，在HMD引起的视觉遮挡下，上半部分面部EMG提供了鲁棒的互补信息，并为自然虚拟现实环境中的多模态情感识别奠定了基础。这种方法促进了情感自适应应用的发展，包括沟通训练和治疗干预。该数据集将在请求下根据伦理使用协议共享。

View on arXiv Download PDF AI Translation

cs.CV / 41 / 2609.03572

Drive-HWM: Hierarchical World Models for Dynamic-Latent Guided Autonomous Driving

Drive-HWM：用于动态潜在引导自主驾驶的分层世界模型

Fan, Zhaoxin, Zhang, Tianbao, Wu, Wenjun, Wang, Xiaofeng, Jin, Yeying, Zhao, Jian, Zhu, Zheng, Yan, Shuicheng

Abstract

World models offer a promising paradigm for autonomous driving by predicting how traffic scenes may evolve and using such predictions to support action generation. However, existing approaches either separate future prediction from action generation or jointly predict them at the same temporal scale, making it difficult to simultaneously achieve long-horizon anticipation and responsive, observation-grounded decision making. We present Drive-HWM, a hierarchical slow--fast world modeling framework that organizes future representation prediction and action generation at complementary temporal scales. The slow world model predicts multi-step future representations to capture extended scene evolution. To explicitly model the abundant motion dynamics in driving environments, we introduce Dynamic-Aware Latents learned through optical-flow prediction. Guided by these future representations, the fast model uses a lightweight multimodal backbone and an autoregressive expert to jointly predict the next frame and the immediate action from the latest observation. Next-frame prediction encourages the fast model to capture imminent scene evolution, while one-step action generation allows decisions to be continuously updated as new observations arrive. Extensive experiments on NAVSIM v1 and v2 demonstrate the strong driving performance of Drive-HWM. Comprehensive ablation studies further validate the effectiveness of the hierarchical slow--fast design, dynamics-aware future representations, and joint next-frame and action prediction.

Chinese Translation

世界模型为自主驾驶提供了一种有前景的范式，通过预测交通场景的演变并利用这些预测来支持行动生成。然而，现有的方法要么将未来预测与行动生成分开，要么在相同的时间尺度上共同预测这两者，这使得同时实现长期预判和基于观察的响应性决策变得困难。我们提出了Drive-HWM，一种分层的慢-快世界建模框架，它在互补的时间尺度上组织未来表示预测和行动生成。慢世界模型预测多步未来表示，以捕捉延续的场景演变。为了明确建模驾驶环境中丰富的运动动态，我们引入了通过光流预测学习的动态感知潜在变量。受到这些未来表示的引导，快模型使用轻量级的多模态骨干网络和自回归专家，联合预测下一帧和最新观察下的即时行动。下一帧预测促使快模型捕捉即将发生的场景演变，而一步行动生成则允许随着新观察的到来持续更新决策。在NAVSIM v1和v2上的广泛实验展示了Drive-HWM的强大驾驶性能。全面的消融研究进一步验证了分层慢-快设计、动态感知未来表示以及联合下一帧和行动预测的有效性。

View on arXiv Download PDF AI Translation

cs.CV / 42 / 2609.03585

Text2Thermal: Physics-Aware Thermal Image Synthesis from Textual Priors

Text2Thermal：基于物理知识的文本优先热成像合成

Qazi, Tayeba, Lall, Brejesh, Mukherjee, Prerana

Abstract

Thermal infrared imaging offers reliable perception in darkness and adverse weather, but thermal datasets remain scarce, motivating extensive work on translating abundant RGB images into thermal. Such translation is fundamentally ill-posed as thermal appearance is governed by surface emissivity and object temperature, neither of which is observable in the visible spectrum, so a single RGB image is consistent with many valid thermal outputs. We argue that language offers a natural means of resolving this ambiguity, and propose Text2Thermal, a framework for physics-aware thermal image synthesis from textual priors. Rather than inferring the unobservable radiometric factors from RGB, we supply them explicitly through thermally grounded captions encoding material, weather, time-of-day, and heat-emission state, and adapt a pretrained Stable Diffusion backbone to the thermal domain. Because the radiometric content is determined entirely by the prompt, Text2Thermal synthesizes thermal imagery without requiring a registered RGB image at inference; where spatial guidance is desired, an optional control signal imparts scene geometry without disturbing the prompt-specified radiometry. Experiments on M3FD, FLIR, and FMB show that Text2Thermal achieves state-of-the-art FID among thermal image synthesis methods while offering text-level control that translation-based approaches cannot provide.

Chinese Translation

热红外成像在黑暗和恶劣天气条件下提供了可靠的感知，但热数据集仍然稀缺，这促使了将丰富的RGB图像转换为热图像的广泛研究。这种转换在本质上是一个不适定问题，因为热外观受表面发射率和物体温度的影响，而这两者在可见光谱中都是不可观察的，因此单一的RGB图像可以与许多有效的热输出一致。我们认为语言提供了一种自然的方式来解决这种模糊性，并提出了Text2Thermal，一个基于文本优先的物理知识热图像合成框架。我们不是从RGB推断不可观察的辐射因素，而是通过热相关的说明明确提供这些因素，这些说明编码了材料、天气、时间和热发射状态，并将预训练的Stable Diffusion骨干网络适应于热成像领域。由于辐射内容完全由提示决定，Text2Thermal在推理时无需注册的RGB图像即可合成热图像；在需要空间引导的情况下，选用的控制信号可以在不干扰提示指定的辐射条件下传达场景几何信息。在M3FD、FLIR和FMB上的实验表明，Text2Thermal在热图像合成方法中实现了最先进的FID，同时提供了基于文本的控制，这是基于转换的方法无法提供的。

View on arXiv Download PDF AI Translation

cs.CV / 43 / 2609.03602

SV-WAM: An Efficient Surround-View World-Action Model for End-to-End Autonomous Driving

SV-WAM：一种高效的全景世界-动作模型用于端到端自主驾驶

Wang, Jinyang, Li, Shiwei, Wang, Junjian, Deng, Zhiqiang, Gao, Jianbin, Zhao, Yihang, Liu, Liu, Zhao, Yongjia, Chen, Jinlong, Xu, Huirui, Pan, Yifeng, Liu, Kangwei, Ren, Fan, Tao, Ji, Yang, Minghao

Abstract

World models (WMs) have demonstrated strong potential for end-to-end autonomous driving by learning predictive representations of future scene dynamics. However, generating future videos during inference introduces substantial computational overhead, leading many recent driving WMs to adopt a single front camera as input for efficient deployment. This design restricts spatial coverage in safety-critical maneuvers such as lane changes, merges, and turns. To address this limitation, we propose SV-WAM, a surround-view world-action model (WAM) that preserves full six-camera observations while maintaining efficient inference. SV-WAM leverages future-video prediction as dense training supervision for action learning within a shared generative model, rather than as an inference-time output. At the core of this design is an action-centered causal mask that prevents action tokens from attending to future-video tokens during joint action-video denoising. Consequently, the video branch can be discarded at deployment, enabling efficient action-only planning. Furthermore, we introduce a differentiable drivable-area compliance regularizer that penalizes vehicle-footprint corners approaching or crossing drivable boundaries, improving planning safety and boundary awareness. Extensive experiments on the closed-loop NAVSIMv2 benchmark and the open-loop nuScenes benchmark demonstrate that SV-WAM achieves state-of-the-art planning performance with low inference latency and competitive zero-shot transfer capability.

Chinese Translation

世界模型（WMs）通过学习未来场景动态的预测表示，展示了在端到端自主驾驶中的强大潜力。然而，在推理过程中生成未来视频会引入大量计算开销，导致许多近期的驾驶世界模型采用单前置摄像头作为输入以实现高效部署。这种设计限制了在安全关键的操作（如变道、并入和转弯）中的空间覆盖。为了解决这一限制，我们提出了SV-WAM，一种全景世界-动作模型（WAM），它在保持高效推理的同时保留完整的六个摄像头观察。SV-WAM利用未来视频预测作为共享生成模型中动作学习的密集训练监督，而不是作为推理时的输出。这一设计的核心是一个以动作为中心的因果掩码，防止动作标记在联合动作-视频去噪过程中关注未来视频标记。因此，视频分支可以在部署时被丢弃，从而实现高效的仅动作规划。此外，我们引入了一种可微分的可行驶区域合规正则化器，惩罚接近或越过可行驶边界的车辆轮廓角，提高规划安全性和边界意识。在闭环NAVSIMv2基准和开放环nuScenes基准上的大量实验表明，SV-WAM在低推理延迟和竞争性的零-shot迁移能力下实现了最先进的规划性能。

View on arXiv Download PDF AI Translation

cs.CV / 44 / 2609.03615

Auditing Patient Privacy in Medical Generative Models: Scalable Memorization Detection with DeepSSIM++

医疗生成模型中的患者隐私审计：基于 DeepSSIM++ 的可扩展记忆检测

Scardace, Antonio, Guarnera, Francesco, Battiato, Sebastiano, Ravì, Daniele

Abstract

While deep generative models offer new opportunities for medical image synthesis and data sharing, their ability to memorize and reproduce training samples raises serious concerns about patient confidentiality. Detecting such memorization at scale remains challenging: traditional pixel-based metrics are sensitive to generation artifacts, whereas generic embedding-based metrics often lack the anatomical sensitivity required for medical data. To address this challenge, we introduce DeepSSIM++, a self-supervised similarity metric for scalable memorization auditing in medical generative models. By leveraging multi-scale feature aggregation and anatomy-preserving augmentations, DeepSSIM++ learns an embedding space where cosine similarity approximates the Structural Similarity Index (SSIM), eliminating the need for exact pixel-level registration. Compared with state-of-the-art baselines, DeepSSIM++ achieves an average Macro F1 improvement of 33 percentage points under ideal alignment and 46 percentage points under realistic spatial and intensity perturbations. Furthermore, it accelerates large-scale similarity computation by several orders of magnitude compared with analytical SSIM. By combining anatomical sensitivity and computational efficiency, DeepSSIM++ provides an open-source tool for scalable memorization auditing in medical generative AI. Code and data are publicly available at: https://github.com/brAIn-science/DeepSSIM.

Chinese Translation

尽管深度生成模型为医学图像合成和数据共享提供了新的机会，但其记忆和重现训练样本的能力引发了对患者机密性的严重担忧。在大规模检测此类记忆时仍然面临挑战：传统的基于像素的度量对生成伪影敏感，而通用的基于嵌入的度量往往缺乏医学数据所需的解剖敏感性。为了解决这一挑战，我们提出了 DeepSSIM++，一种用于医疗生成模型中可扩展记忆审计的自监督相似性度量。通过利用多尺度特征聚合和保持解剖结构的增强，DeepSSIM++ 学习了一个嵌入空间，在该空间中，余弦相似度近似结构相似性指数（SSIM），消除了对精确像素级配准的需求。与最先进的基线相比，DeepSSIM++ 在理想对齐下实现了平均宏 F1 提升 33 个百分点，在现实空间和强度扰动下提升 46 个百分点。此外，与分析 SSIM 相比，它在大规模相似性计算中加速了几个数量级。通过结合解剖敏感性和计算效率，DeepSSIM++ 提供了一个用于医疗生成 AI 中可扩展记忆审计的开源工具。代码和数据可在以下网址公开获取：https://github.com/brAIn-science/DeepSSIM.

View on arXiv Download PDF AI Translation

cs.CV / 45 / 2609.03629

EraseSAE: Surgical Concept Erasure in Text-to-Video Diffusion Models via Sparse Autoencoders

EraseSAE：通过稀疏自编码器在文本到视频扩散模型中进行概念的手术性消除

Wang, Xinghao, Li, Dong, Yu, Wei, Pan, Yingwei, Gong, Tao, Chu, Qi, Yu, Nenghai, Yao, Ting

Abstract

Recent advances in text-to-video (T2V) diffusion models have demonstrated remarkable generative capabilities, yet their reliance on loosely curated training data raises pressing safety and copyright concerns. Concept erasure offers a principled remedy by removing unwanted semantics from pretrained models while preserving remaining concepts. However, existing approaches typically operate at a coarse granularity misaligned with the fine-grained, distributed nature of concept representations, leading to incomplete removal or degraded generation quality. We argue that surgical erasure fundamentally requires intervention at the level of monosemantic features, where each unit encodes a single interpretable concept. To this end, we propose EraseSAE, a novel framework that leverages sparse autoencoders to achieve surgical concept erasure in DiT-based T2V diffusion models via a principled decompose-attribute-erase pipeline. We first introduce the Partitioned Convolutional Sparse Autoencoder, which decomposes dense spatiotemporal activations into disentangled, interpretable sparse features while preserving spatiotemporal coherence. A contrastive attribution mechanism then contrasts activations from paired prompts to isolate concept-specific feature kernels. At inference, timestep-resolved spatiotemporal masks derived from the identified kernels confine erasure to regions where the target concept is active, leaving unrelated content intact. Extensive experiments across diverse diffusion models and concept erasure tasks demonstrate that EraseSAE achieves precise and robust concept removal with minimal quality degradation, substantially outperforming state-of-the-art methods. The code is available at https://github.com/HiDream-ai/EraseSAE.

Chinese Translation

近期文本到视频（T2V）扩散模型的进展展示了显著的生成能力，但其对松散整理的训练数据的依赖引发了迫切的安全性和版权问题。概念消除提供了一种原则性的解决方案，通过去除预训练模型中的不必要语义，同时保留剩余概念。然而，现有方法通常在粗粒度层面上操作，与概念表示的细粒度、分布式特性不一致，导致不完全的去除或生成质量下降。我们认为，手术性消除根本上需要在单语义特征的层面进行干预，其中每个单元编码一个可解释的单一概念。为此，我们提出了EraseSAE，一个新颖的框架，利用稀疏自编码器通过原则性的分解-属性-消除管道在基于DiT的T2V扩散模型中实现手术性概念消除。我们首先介绍了分区卷积稀疏自编码器，它将密集的时空激活分解为解耦的、可解释的稀疏特征，同时保持时空一致性。然后，采用对比归因机制对配对提示的激活进行对比，以隔离特定概念的特征核。在推理时，从识别的特征核派生的时间步分辨的时空掩码将消除限制在目标概念活跃的区域，保持无关内容不变。针对不同扩散模型和概念消除任务的广泛实验表明，EraseSAE以最小的质量下降实现了精确和稳健的概念去除，显著优于最先进的方法。代码可在 https://github.com/HiDream-ai/EraseSAE 获取。

View on arXiv Download PDF AI Translation

cs.CV / 46 / 2609.03639

Stabilizing Camera-Controlled Novel View Synthesis at Inference Time

在推理时稳定化相机控制的新视图合成

Singh, Prajwal, Badola, Arjun, Kumari, Seema, Nagahara, Hajime, Raman, Shanmuganathan

Abstract

Training-free, camera-controlled novel view synthesis from a single image using pre-trained video diffusion models often becomes unstable under large camera motion and long generation horizons. Existing approaches commonly combine several inference-time components, making it unclear which design choices are most important for stability. We show that the main source of stability is simple. Decomposing camera motion into small autoregressive steps limits per-step geometric distortion and reduces error accumulation. A controlled camera-step study shows that performance remains stable for small motions and degrades more strongly as the per-step motion approaches $18$-$20^\circ$. We further evaluate geometry-constrained spatial attention and low-frequency appearance anchoring as supporting refinements, together with an efficient registration-free warping pipeline. Across RealEstate10K and MegaScene, CamTrol++ improves temporal and geometric consistency, downstream 3D reconstruction quality, and generation efficiency over training-free baselines. The method remains effective for 56-frame generation and under substantial controlled depth corruption. These results show that careful control of camera motion at inference time can substantially improve the stability of camera-controlled novel view synthesis without retraining or modifying the diffusion backbone.

Chinese Translation

基于预训练视频扩散模型，从单幅图像进行无训练的相机控制新视图合成在大幅相机运动和长生成时间范围内往往变得不稳定。现有方法通常结合多个推理时组件，使得不清楚哪些设计选择对稳定性最为重要。我们表明，稳定性的主要来源是简单的。将相机运动分解为小的自回归步骤可以限制每一步的几何失真并减少误差累积。控制相机步骤的研究表明，对于小幅运动，性能保持稳定，而当每步运动接近 $18$-$20^ ext{°}$ 时，性能下降更为明显。我们进一步评估了几何约束的空间注意力和低频外观锚定作为支持性改进，并结合高效的无注册变形管道。在 RealEstate10K 和 MegaScene 数据集上，CamTrol++ 在时间和几何一致性、下游 3D 重建质量以及生成效率方面优于无训练基线。该方法在 56 帧生成和显著受控深度损坏的情况下仍然有效。这些结果表明，在推理时仔细控制相机运动可以显著提高相机控制新视图合成的稳定性，而无需重新训练或修改扩散骨干网络。

View on arXiv Download PDF AI Translation

cs.CV / 47 / 2609.03641

Tree-Structured Vector Quantization For Efficient And Progressive Image Compression

用于高效和渐进图像压缩的树结构向量量化

Wang, Xinkun, Xu, Tianyi, Luo, Qingyu, Ma, Mingming, Jiao, Changzhe, Li, Fu, Niu, Yi

Abstract

Vector-quantization based image compression has achieved strong rate--distortion performance, yet most of them still produce a separate compressed representation for each target bitrate. Such variable-rate behavior allows one model to operate at multiple rates, but it does not necessarily provide a progressive bitstream whose prefixes are themselves decodable and can be refined by appending additional bits. We propose \textbf{Tree-VQ}, a progressive tree-structured vector quantization framework for learned image compression. Tree-VQ organizes discrete codewords as a hierarchical binary tree and represents each latent token by a routed root-to-leaf path. Crucially, every prefix of this path corresponds to a valid quantized representation, so shallow internal nodes serve as coarse reconstruction codes and deeper nodes provide successive refinements. This allows a compressed image to be decoded from an early prefix and progressively improved as more branch symbols are received, rather than being re-encoded for different target rates. To make this structure practical for compression, we introduce a prefix-compatible tree entropy model that codes progressive continuation decisions and routed branch refinements using only causally available decoded contexts. We further use rate-aware refinement scheduling to decide which spatial blocks should receive additional tree bits under a given prefix budget, and hierarchical prefix supervision to ensure that internal nodes are directly decodable at low rates. Experiments show that Tree-VQ achieves a superior performance--efficiency trade-off, delivering the best perceptual compression results with much fewer parameters and lower latency than competing methods.

Chinese Translation

基于向量量化的图像压缩在速率-失真性能上取得了显著的成果，但大多数方法仍然为每个目标比特率生成单独的压缩表示。这种可变比特率行为允许一个模型在多个速率下运行，但并不一定提供一个渐进的比特流，其前缀本身是可解码的，并且可以通过附加比特进行细化。我们提出了 extbf{Tree-VQ}，一种用于学习图像压缩的渐进树结构向量量化框架。Tree-VQ将离散码字组织为层次二叉树，并通过路由的根到叶路径表示每个潜在标记。关键是，这条路径的每个前缀都对应一个有效的量化表示，因此浅层内部节点作为粗略重建码，而更深的节点提供连续的细化。这使得可以从早期前缀解码压缩图像，并随着更多分支符号的接收逐步改进，而不是为不同的目标速率重新编码。为了使这种结构在压缩中实用，我们引入了一种前缀兼容的树熵模型，该模型使用仅因果可用的解码上下文对渐进续接决策和路由分支细化进行编码。我们进一步使用基于速率的细化调度来决定在给定前缀预算下哪些空间块应接收额外的树比特，并采用层次前缀监督以确保内部节点在低速率下是直接可解码的。实验表明，Tree-VQ在性能-效率权衡上表现优越，提供了最佳的感知压缩结果，且参数更少、延迟更低，相较于竞争方法具有明显优势。

View on arXiv Download PDF AI Translation

cs.CV / 48 / 2609.03655

PL-SCEA: Reconfiguring Pretrained Attention for Few-Shot Industrial Anomaly Detection

PL-SCEA：为少量工业异常检测重新配置预训练注意力

Yang, Xiaoyu, Wu, Qixing, Zhao, Huixian, Jin, Changlong

Abstract

Vision Foundation Models (VFMs) provide transferable patch representations for few-shot industrial anomaly detection, but their attention computation is typically inherited from pretraining objectives centered on semantic aggregation. This creates a potential mismatch: token relations that support semantic recognition may not adequately expose the localized texture and structural deviations required for anomaly localization. We therefore investigate the hypothesis that the attention computation of a frozen VFM can be reconfigured as a task-relevant component of anomaly detection. We instantiate this idea with Power-Law Self-Correlation Enhanced Attention (PL-SCEA), which retains the semantic context of pretrained query-key attention while constructing token-adaptive self-correlations over contextualized value features. Positive-correlation filtering and power-law reweighting then emphasize relations that are salient relative to each token's relational background, without introducing additional trainable attention projections. The resulting features are modeled by a lightweight variational autoencoder that provides a fixed-size reconstruction-based representation of category-specific normality. The two stages serve complementary roles: attention reconfiguration shapes how local relational deviations are represented, while reconstruction-based modeling converts deviations from learned normality into anomaly scores. Across MVTec AD and VisA, the complete framework achieves competitive image-level detection and consistently strong pixel-level localization across the evaluated few-shot settings. Ablations further show that PL-SCEA improves localization with either the VAE or a memory bank under the tested setting. These results support the view that task-aligned attention reconfiguration can improve the anomaly-localization capability of frozen pretrained representations.

Chinese Translation

视觉基础模型（VFM）为少量工业异常检测提供可转移的补丁表示，但它们的注意力计算通常继承自以语义聚合为中心的预训练目标。这造成了潜在的不匹配：支持语义识别的标记关系可能无法充分揭示异常定位所需的局部纹理和结构偏差。因此，我们探讨了一个假设，即冻结的 VFM 的注意力计算可以重新配置为异常检测的任务相关组件。我们用幂律自相关增强注意力（PL-SCEA）来实现这一思想，该方法保留了预训练查询-键注意力的语义上下文，同时在上下文化的值特征上构建标记自适应自相关。正相关过滤和幂律重加权强调了相对于每个标记的关系背景而言显著的关系，而不引入额外的可训练注意力投影。所得到的特征由轻量级变分自编码器（VAE）建模，提供了基于重建的类别特定正常性的固定大小表示。这两个阶段发挥互补作用：注意力重新配置塑造了局部关系偏差的表示方式，而基于重建的建模将从学习到的正常性偏差转换为异常分数。在 MVTec AD 和 VisA 上，完整框架在评估的少量设置中实现了竞争性的图像级检测和持续强大的像素级定位。消融实验进一步表明，在测试设置下，PL-SCEA 无论是与 VAE 还是内存库结合使用，都能改善定位。这些结果支持了任务对齐的注意力重新配置可以提高冻结预训练表示的异常定位能力的观点。

View on arXiv Download PDF AI Translation

cs.CV / 49 / 2609.03657

Rethinking 3D Noise: Learning 3D-Aware Video Priors via Optimization-Free Morphological Perturbations

重新思考三维噪声：通过无优化形态扰动学习三维感知视频先验

Şahin, Onat, Altillawi, Mohammad, Eskandar, George, Carbone, Carlos, Liu, Ziyuan

Abstract

3D scene representations like NeRF and 3D Gaussian Splatting (3DGS) suffer severe artifacts in sparse-view settings. Recent generative 3D artifact fixers attempt to address this, but rely on paired corrupted and clean renders requiring costly, per-scene reconstructions across varying view configurations. While 2D image augmentations act as instant regularizers, no explicit equivalents exist for 3D representations to preserve spatial consistency across views, an essential property for 3D-aware training. We propose 3D Morphological Perturbations as an optimization-free regularizer that preserves spatial consistency. Leveraging explicit 3DGS, we treat each Gaussian as a fundamental building block - analogous to a 2D pixel - and apply perturbations across its morphological parameter space via scale, rotation, and pruning. Our method eliminates per-scene 3DGS optimization loops from dataset curation while enabling models to learn stronger geometric priors than sparse-view baselines in diagnostic ablations conducted on a lightweight video diffusion sandbox. Scaled to a 14B-parameter video model via ControlNet, our approach maintains visual fidelity while reducing mean depth error by 12.5% over state-of-the-art image-to-image 3D artifact refiners, ultimately boosting downstream robotics policy success rates by up to 8.0% across 3 of 4 manipulation tasks.

Chinese Translation

三维场景表示（如 NeRF 和 3D Gaussian Splatting (3DGS)）在稀疏视图设置中遭受严重伪影。近期的生成性三维伪影修复方法试图解决这一问题，但依赖于配对的损坏和干净渲染图，这需要在不同视图配置下进行昂贵的逐场景重建。虽然二维图像增强作为即时正则化器发挥作用，但对于三维表示来说，并不存在明确的等效方法来保持视图间的空间一致性，而这对于三维感知训练是一个至关重要的属性。我们提出了三维形态扰动作为一种无优化的正则化器，以保持空间一致性。通过显式的 3DGS，我们将每个高斯视为基本构建块——类似于二维像素——并通过尺度、旋转和修剪在其形态参数空间中施加扰动。我们的方法消除了数据集策划中的逐场景 3DGS 优化循环，同时使模型能够学习比稀疏视图基线更强的几何先验，在轻量级视频扩散沙箱中进行的诊断消融实验中得到了验证。通过 ControlNet 将其扩展到一个 140亿参数的视频模型，我们的方法在保持视觉保真度的同时，将平均深度误差降低了 12.5%，相较于最先进的图像到图像三维伪影修复器，最终在 4 个操作任务中的 3 个上提高了下游机器人策略的成功率，最高可达 8.0%。

View on arXiv Download PDF AI Translation

cs.CV / 50 / 2609.03663

Cross-Dataset Transfer and Reliability of Explainable Artificial Intelligence for RhythmFormer Remote Photoplethysmography

跨数据集迁移与可解释人工智能在 RhythmFormer 远程光电容积描记中的可靠性

Chen, Louis, Nordling, Torbjörn E. M.

Abstract

Background. Remote photoplethysmography estimates the cardiovascular pulse from facial video, and its explanations have rested on inspecting heatmaps rather than on quantitative evidence about where a model reads it. We quantified the explanations and asked whether such explanations transfer between datasets and track model performance. Method. We trained eight condition-specific RhythmFormer models on NCKU-rPPG, recorded under three illumination levels, speaking, rotation, and cycling, estimated one heart rate per 5.12-second clip, and set them beside a UBFC-rPPG reproduction. Raw attention, rollout, attention flow, and Beyond Intuition were assessed by skin coverage and the Salience-guided Faithfulness Coefficient (SaCo). Results. Beyond Intuition ranked highest on both datasets, at median coverage 0.789 and SaCo 0.837 on Static level 3 against 0.826 and 0.917 on UBFC-rPPG; lower ranks differed. Within one participant of one condition, neither measure was related to a clip's heart-rate error, waveform correlation, or signal-to-noise ratio on either dataset: 186 of the 252 coefficients fell below $|\rho|=0.10$ and 28 reached $p<0.05$ against the 13 expected by chance. Across the eight scenarios only Beyond Intuition's coverage followed the three performance measures, at $\rho=-0.43$, $+0.57$, and $+0.43$, while the attention-only methods' SaCo ran opposite to each. It failed at 40 lux alone, its median coverage falling to 0.180 and its median SaCo to $-0.178$, whereas motion degraded the estimates far more without such a drop. Conclusions. Skin coverage and SaCo carry information complementary to the performance measures rather than a proxy for them: attributing to the skin does not guarantee an accurate estimate. What an attribution reveals about a condition is where the model looks rather than how faithfully its map is ordered.

Chinese Translation

背景：远程光电容积描记通过面部视频估算心血管脉搏，其解释主要依赖于热图的检查，而非关于模型读取位置的定量证据。我们量化了这些解释，并探讨了这种解释是否能够在数据集之间迁移以及是否能反映模型性能。方法：我们在 NCKU-rPPG 数据集上训练了八个特定条件的 RhythmFormer 模型，该数据集在三种照明水平下录制，涉及说话、旋转和骑行，每个 5.12 秒的片段估算一个心率，并将其与 UBFC-rPPG 数据集进行对比。通过皮肤覆盖率和显著性引导的忠实度系数（Salience-guided Faithfulness Coefficient, SaCo）评估了原始注意力、回滚、注意力流和 Beyond Intuition。结果：在两个数据集中，Beyond Intuition 的排名最高，在静态水平 3 的中位覆盖率为 0.789，SaCo 为 0.837，而在 UBFC-rPPG 上分别为 0.826 和 0.917；较低的排名有所不同。在一个条件下的单个参与者中，两个指标与片段的心率误差、波形相关性或信噪比无关：252 个系数中有 186 个低于 $| ho|=0.10$，28 个达到 $p<0.05$，而随机情况下预期为 13。跨八种场景中，只有 Beyond Intuition 的覆盖率与三项性能指标一致，相关系数分别为 $ ho=-0.43$、$+0.57$ 和 $+0.43$，而仅注意力方法的 SaCo 则与之相反。在 40 lux 的情况下表现不佳，其中位覆盖率降至 0.180，中位 SaCo 降至 $-0.178$，而运动对估算的影响更大，且没有出现如此下降。结论：皮肤覆盖率和 SaCo 提供的信息是对性能指标的补充，而非其代理：归因于皮肤并不保证准确的估算。归因所揭示的条件信息是模型关注的位置，而非其映射的忠实程度。

View on arXiv Download PDF AI Translation

cs.CV / 51 / 2609.03668

ARCOS: Zero-shot Boundary Localization for Corneal Layer Segmentation Across Optical Coherence Tomography Devices

ARCOS：用于光学相干断层扫描设备的角膜层分割的零样本边界定位

Brás, Nuno Vivas, Memmi, Benjamin, Bouhassane, Maëlle, Georgeon, Cristina, Borderie, Vincent, Plamann, Karsten, Chessel, Anatole

Abstract

Accurate segmentation of corneal layers in optical coherence tomography (OCT) is essential for quantitative assessment of corneal morphology, including layer thickness and structural changes associated with disease or surgery. However, automatic segmentation remains challenging because corneal interfaces are thin, affected by speckle noise, and variable across acquisition devices. In this work, we propose ARCOS, a patch-based zero-shot boundary localization framework for corneal layer segmentation in clinical anterior-segment OCT images. Rather than performing conventional region classification, the method predicts boundary heatmaps for the main corneal interfaces from overlapping native-resolution patches. Patch-level predictions are stitched across the full B-scan and converted into boundary locations to obtain continuous, anatomically ordered layer segmentations. The network combines multi-scale feature fusion with a self-conditioned refinement module that uses intermediate boundary information to improve local heatmap predictions while preserving spatial detail. The method was evaluated on clinical OCT images acquired from multiple devices and compared with representative segmentation baselines using boundary localization and derived thickness metrics. The proposed method achieved an off-by-one boundary localization accuracy of 95.1% and a mean absolute boundary error of 0.514 pixels on the matched-device test set. In zero-shot cross-device evaluation, it maintained an average off-by-one accuracy of 84.3% and a mean absolute boundary error of 0.855 pixels across unseen acquisition devices, outperforming the baseline models. Thickness estimates derived from the predicted boundaries showed low error across corneal regions, supporting the method's use for quantitative corneal OCT analysis.

Chinese Translation

在光学相干断层扫描（OCT）中，准确分割角膜层对于定量评估角膜形态至关重要，包括层厚度和与疾病或手术相关的结构变化。然而，由于角膜界面较薄、受斑点噪声影响以及在不同采集设备之间存在变异，自动分割仍然具有挑战性。在本研究中，我们提出了ARCOS，一种基于补丁的零样本边界定位框架，用于临床前段OCT图像中的角膜层分割。该方法不是执行传统的区域分类，而是从重叠的原生分辨率补丁中预测主要角膜界面的边界热图。补丁级别的预测在整个B扫描中拼接，并转换为边界位置，以获得连续的、解剖顺序的层分割。该网络结合了多尺度特征融合与自条件精细化模块，利用中间边界信息改善局部热图预测，同时保持空间细节。该方法在从多个设备获取的临床OCT图像上进行了评估，并与代表性的分割基线进行了比较，使用边界定位和衍生厚度指标。所提出的方法在匹配设备测试集上实现了95.1%的边界定位准确率和0.514像素的平均绝对边界误差。在零样本跨设备评估中，它在未见过的采集设备上保持了84.3%的平均边界定位准确率和0.855像素的平均绝对边界误差，优于基线模型。从预测边界衍生的厚度估计在角膜区域显示出较低的误差，支持该方法在定量角膜OCT分析中的应用。

View on arXiv Download PDF AI Translation

cs.CV / 52 / 2609.03673

Do Video Generators Track the World Across Segments? A Benchmark and Method for World-State Reasoning in Video Continuation

视频生成器是否在不同片段中追踪世界？视频延续中的世界状态推理基准与方法

Miao, Yingmao, Zhang, Pengfei, Xu, Chaoran, Yu, Meng, Tang, Jing, Chu, Xiangxiang, Shen, Chao, Lin, Chenhao

Abstract

Video generators build long videos by composing shorter parts, either by generating segments one after another or by autoregressively extending chunks. Each new part usually depends on memories of historical observations, such as recent frames, selected key frames, memory banks, or cached features. These memories preserve visible evidence from the past, but current generators do not reliably turn such evidence into a world-state interface: what holds in the video world after previous actions and how it should change under the next prompt. A past frame remains valid history, but it may not describe the state needed by the next segment; some states must instead be inferred from occluded or implicit changes rather than copied from a directly observed frame. This creates a simple but overlooked question for video continuation: given a previous video, its prompt, and a new prompt, can a model generate a continuation that reflects the state determined by both the historical video and the new prompt? To answer this question, we introduce Statebench, a benchmark that targets this gap by testing continuations over three state categories: past-visible states, occluded-process states, and complex-transition states. We further propose Stateagent, which explicitly maintains an entity-state representation, updates it under the new prompt, grounds the predicted post-action state as a future end frame, and renders the next video. Experiments show that our method improves controlled video continuation by raising the all-case state score (SCS-All) from 45.2 to 69.3, and also benefits story generation at the one-minute scale. Code is avaliable at https://github.com/AMAP-ML/StateAgent.

Chinese Translation

视频生成器通过组合较短的部分来构建长视频，通常是一个接一个地生成片段，或通过自回归地扩展块。每个新部分通常依赖于对历史观察的记忆，例如最近的帧、选定的关键帧、记忆库或缓存特征。这些记忆保留了来自过去的可见证据，但当前的生成器并不能可靠地将这些证据转化为世界状态接口：在先前动作之后，视频世界中发生了什么，以及在下一个提示下它应该如何变化。过去的帧仍然是有效的历史，但它可能无法描述下一个片段所需的状态；某些状态必须从遮挡或隐含的变化中推断，而不是直接从观察到的帧中复制。这为视频延续提出了一个简单但被忽视的问题：给定一个先前的视频、其提示和一个新的提示，模型能否生成一个反映由历史视频和新提示共同决定的状态的延续？为了解答这个问题，我们引入了Statebench，一个针对这一空白的基准，通过测试三个状态类别的延续来进行评估：过去可见状态、遮挡过程状态和复杂过渡状态。我们进一步提出了Stateagent，它明确维护实体状态表示，在新提示下更新该表示，将预测的后行动状态作为未来的结束帧，并渲染下一个视频。实验表明，我们的方法通过将所有情况状态得分（SCS-All）从45.2提高到69.3，改善了受控视频延续，并且在一分钟尺度的故事生成中也受益。代码可在 https://github.com/AMAP-ML/StateAgent 获取。

View on arXiv Download PDF AI Translation

cs.CV / 53 / 2609.03675

CoFiE: Coarse-to-Fine Evidence Selection for Efficient Streaming Video Understanding

CoFiE：高效流媒体视频理解的粗到细证据选择

Jiang, Jing, Ling, Yiran, Li, Ruonan, Stamoulis, Dimitrios, Liu, Jie

Abstract

Streaming video understanding requires Vision Language Models (VLLMs) to process growing video streams and answer user questions under tight latency constraints. Existing methods improve efficiency through token pruning and memory-bank schemes, but mainly reduce visual tokens after visual encoding. Consequently, downstream token pruning alone cannot substantially reduce end-to-end latency because the expensive frame encoding cost has already been incurred. We propose CoFiE, a Coarse-to-Fine Evidence Selection framework that decouples evidence selection into a coarse, query-agnostic filtering stage before the vision encoder and a fine, query-specific refinement stage during LLM prefill. CoFiE introduces Novelty-Guided Frame Filtering to retain visually distinctive candidate frames and Query-Specific Evidence Refinement to select the frames most relevant to the user query. This design removes substantial redundancy before frame encoding while preserving query-specific refinement once semantic information becomes available. Experiments show that CoFiE establishes a new state-of-the-art accuracy-efficiency trade-off across multiple video understanding benchmarks, reaching 78.86% accuracy on StreamingBench and 68.72% on OvO-Bench, with improvements of up to 3.15% over prior methods. Even with up to 80% evidence-frame filtering, CoFiE outperforms strong open-source multimodal models while improving end-to-end inference latency by up to 2.54 times.

Chinese Translation

流媒体视频理解需要视觉语言模型（VLLMs）在严格的延迟限制下处理不断增长的视频流并回答用户问题。现有方法通过令牌修剪和内存库方案提高效率，但主要是在视觉编码后减少视觉令牌。因此，仅依靠下游令牌修剪无法显著降低端到端延迟，因为昂贵的帧编码成本已经产生。我们提出了CoFiE，一个粗到细的证据选择框架，它将证据选择解耦为视觉编码前的粗略、与查询无关的过滤阶段，以及在LLM预填充期间的细致、与查询相关的精炼阶段。CoFiE引入了新颖性引导帧过滤，以保留视觉上独特的候选帧，并采用查询特定的证据精炼，以选择与用户查询最相关的帧。这一设计在帧编码之前消除了大量冗余，同时在语义信息可用后保留查询特定的精炼。实验表明，CoFiE在多个视频理解基准测试中建立了新的最先进的准确性-效率权衡，在StreamingBench上达到78.86%的准确率，在OvO-Bench上达到68.72%，比以前的方法提高了多达3.15%。即使在高达80%的证据帧过滤下，CoFiE仍然优于强大的开源多模态模型，同时将端到端推理延迟提高了多达2.54倍。

View on arXiv Download PDF AI Translation

cs.CV / 54 / 2609.03677

Understanding Autonomous Driving Datasets by Describing Differences between Image Subsets in Natural Language

通过自然语言描述图像子集之间的差异来理解自主驾驶数据集

Truetsch, Julian, Hauser, Felix, Stiller, Christoph, Bieder, Frank

Abstract

Understanding the composition of large-scale autonomous driving datasets is essential for safety, robustness, and reliable operation across domains. For example, domain shift between locations could lead to the operating environment being misaligned with the training data, resulting in potentially dangerous performance degradation. Yet, existing data analysis pipelines largely rely on metadata, predefined labels, or manual inspection, which provide limited semantic insight or do not scale. This paper studies set difference captioning: given two subsets of images, the goal is to produce a natural-language hypothesis describing differences between the target and reference set. Building on a two-stage formulation, we adapt the method to autonomous driving by focusing on object-centric patches derived from object detection, which simplifies aggregation and enables attribution of differences to specific object instances or categories. To evaluate this setting in-domain, we introduce a new benchmark, AD-Diff Bench. Low-concentration experiments assess the suitability of set-difference-captioning approaches to sparse, real-world differences. We restrict our experiments to open-weight models to support reproducibility and ease of deployment. The proposed benchmark and analysis provide a step towards practical, human-interpretable dataset introspection for autonomous driving datasets. Our implementation and benchmark dataset are available at https://github.com/KIT-MRT/AD-Diff

Chinese Translation

理解大规模自主驾驶数据集的组成对于安全性、鲁棒性以及跨领域的可靠操作至关重要。例如，地点之间的领域转移可能导致操作环境与训练数据不一致，从而导致潜在的危险性能下降。然而，现有的数据分析流程主要依赖于元数据、预定义标签或人工检查，这些方法提供的语义洞察有限或无法扩展。本文研究了集合差异描述：给定两个图像子集，目标是生成一个自然语言假设，描述目标集与参考集之间的差异。在两阶段的框架基础上，我们通过关注从目标检测中提取的以对象为中心的图像块，来将该方法适应于自主驾驶，这简化了聚合过程并使得差异归因于特定的对象实例或类别。为了在领域内评估这一设置，我们引入了一个新的基准，AD-Diff Bench。低浓度实验评估集合差异描述方法在稀疏、真实世界差异中的适用性。我们将实验限制在开放权重模型，以支持可重复性和易于部署。所提出的基准和分析为自主驾驶数据集的实用、可人类解释的数据集内省迈出了重要一步。我们的实现和基准数据集可在 https://github.com/KIT-MRT/AD-Diff 获取。

View on arXiv Download PDF AI Translation

cs.CV / 55 / 2609.03680

DropClick: Semi-Automated One-Click Segmentation for Agricultural Robotic Data

DropClick：农业机器人数据的半自动一键分割

Zimmer, Patrick, Halstead, Michael, McCool, Chris

Abstract

Labelling vision datasets, especially for segmentation tasks, is a laborious and costly process that stymies novel developments in agricultural robotics. In this paper, we present DropClick, a click-guided segmentation tool that simplifies the annotation process. Our system utilises single-click inputs on objects to generate pseudo-labels, which can replace manual annotations. DropClick stands out as it is a semi-automated approach and does not require a click for every object in the scene. It can therefore further reduce the required amount of user input drastically. We evaluate our method on two challenging agricultural robotic datasets, SB20 and BUP20 for plant and fruit segmentation, respectively. DropClick is first trained on a small subset of just 5 images from the original training data. This DropClick model can then be deployed as a one-click segmentation system and achieves comparable or higher performance than other one-click methods achieving an mIoU of 70.0 and 72.6 points, for SB20 and BUP20 respectively. DropClick then excels at maintaining high performance when clicks are not given (e.g. dropped); when 50% of the clicks are missing it still maintains an mIoU of 68.9 and 71.3 points, for SB20 and BUP20 respectively. We validate DropClick as a pseudo-labelling approach by taking its outputs to train a Mask2Former instance-based segmentation model in a semi-supervised manner. In this process, partially removing user input from DropClick yields similar high performance when compared to providing all clicks, at 70.1 vs 70.7 points AP50 for SB20 and no difference for BUP20 at 77.0 for both models; at the same time saving 46.3% of total input for SB20 and 31.9% for BUP20.

Chinese Translation

标注视觉数据集，尤其是分割任务，是一个繁琐且成本高昂的过程，这阻碍了农业机器人领域的新发展。本文提出了DropClick，一种点击引导的分割工具，简化了注释过程。我们的系统利用对物体的单击输入生成伪标签，这可以替代手动标注。DropClick的独特之处在于它是一种半自动化的方法，并不需要对场景中的每个物体都进行点击。因此，它可以大幅减少所需的用户输入量。我们在两个具有挑战性的农业机器人数据集SB20和BUP20上评估了我们的方法，分别用于植物和水果的分割。DropClick首先在原始训练数据的一个仅包含5张图像的小子集上进行训练。然后，这个DropClick模型可以作为一键分割系统部署，并且在SB20和BUP20上分别实现了70.0和72.6的mIoU，表现与其他一键方法相当或更高。DropClick在未给出点击时（例如，点击被丢弃）仍能保持高性能；当50%的点击缺失时，SB20和BUP20的mIoU仍分别保持在68.9和71.3。我们通过将DropClick的输出用于以半监督方式训练Mask2Former实例分割模型，验证了DropClick作为伪标注方法的有效性。在此过程中，部分移除DropClick的用户输入与提供所有点击相比，仍能获得相似的高性能，SB20的AP50为70.1与70.7，BUP20的两个模型均为77.0，同时节省了SB20总输入的46.3%和BUP20的31.9%。

View on arXiv Download PDF AI Translation

cs.CV / 56 / 2609.03688

ToPO: Token-Conditioned Preference Routing for Attention-Based Latent Diffusion Models

ToPO：基于令牌条件的偏好路由用于基于注意力的潜在扩散模型

Xu, Juntao, Li, Shihong, Au, Hoi Fan, Zhu, Ning

Abstract

Pairwise preference labels rank complete images, yet Diffusion-DPO applies their effect over many spatial and denoising-time coordinates. For attention-based, noise-prediction latent diffusion, ToPO (Token-Oriented Preference Optimization) constructs a per-minibatch, detached, separable spatial-temporal route from branchwise squared-residual contrast in a frozen reference denoiser. Preferred-branch cross-attention uses content tokens to modulate the spatial factor, and an auxiliary pixel-midpoint ordering term is added without local labels or a learned reward model. In matched three-seed retrainings with a shared update schedule, ToPO has higher endpoint estimates than Diffusion-DPO on all five reported SD-1.5 metrics and on HPSv2, ImageReward, and CLIP for SDXL. It also receives larger raw win shares in an aggregate blind SDXL A/B study. These findings are scoped to the reported equal-update U-Net protocols rather than an equal-compute comparison.

Chinese Translation

成对的偏好标签对完整图像进行排序，但Diffusion-DPO在多个空间和去噪时间坐标上应用其效果。对于基于注意力的噪声预测潜在扩散，ToPO（Token-Oriented Preference Optimization）构建了一个每个小批量的、独立的、可分离的时空路径，该路径源自冻结参考去噪器中的分支平方残差对比。优选分支的交叉注意力使用内容令牌来调节空间因子，并且在没有局部标签或学习奖励模型的情况下添加了辅助像素中点排序项。在共享更新计划的匹配三种种子重训练中，ToPO在所有五个报告的SD-1.5指标以及SDXL的HPSv2、ImageReward和CLIP上具有比Diffusion-DPO更高的终点估计。它在一项综合盲SDXL A/B研究中也获得了更大的原始胜率份额。这些发现的范围限于报告的等更新U-Net协议，而不是等计算比较。

View on arXiv Download PDF AI Translation

cs.CV / 57 / 2609.03689

Semantic-Aware Subgraph State Space Model for WSI Classification in Histopathology

用于组织病理学全切片图像分类的语义感知子图状态空间模型

Chen, Feixing, Lu, Hao, Luo, Lin, Xu, Yan

Abstract

Histopathological subtyping relies on the recognition of characteristic histological patterns. These patterns may be expressed by individual tissue structures or by the spatial distribution and co-occurrence of multiple structures, and they often span irregularly shaped tissue regions, termed semantic units in this work. However, conventional patch-based representations may fragment such units and fail to explicitly preserve their internal spatial organization, while efficiently modeling relationships among numerous spatially separated units remains challenging. To address these limitations, we propose the Semantic-Aware Subgraph State Space Model (SASG-SSM), a flexible and efficient framework for whole slide image (WSI) classification. Semantic-Aware Subgraphs (SASGs) first approximate irregularly shaped semantic units by adaptively grouping spatially connected patches guided by class-agnostic visual-semantic priors. By representing patches as graph nodes with adjacency edges, SASGs preserve their internal spatial organization rather than treating them as an unordered set. A Subgraph State Space Module (SG-SSM) subsequently combines a graph neural network encoder for intra-subgraph topology encoding with a Mamba-based state space encoder for efficient contextualization across large numbers of subgraphs. This module integrates local structural information within semantic units with global contextual information arising from their distribution and co-occurrence across the WSI, while efficiently modeling a large number of spatially distributed regions. Extensive experiments across four WSI subtyping datasets demonstrate consistent advantages over representative state-of-the-art methods. Further evaluations under small-cohort and few-shot settings demonstrate robustness and data efficiency under limited training data. Code will be released at https://github.com/HLSvois/SASG-SSM.

Chinese Translation

组织病理学亚型分类依赖于特征组织学模式的识别。这些模式可以通过单个组织结构或多个结构的空间分布与共现来表达，并且通常跨越不规则形状的组织区域，在本研究中称为语义单元。然而，传统的基于补丁的表示可能会破坏这些单元，并未能明确保留其内部空间组织，同时有效建模众多空间分离单元之间的关系仍然具有挑战性。为了解决这些局限性，我们提出了语义感知子图状态空间模型（Semantic-Aware Subgraph State Space Model, SASG-SSM），这是一个灵活且高效的全切片图像（Whole Slide Image, WSI）分类框架。语义感知子图（Semantic-Aware Subgraphs, SASGs）首先通过适应性地将空间连接的补丁分组，来近似不规则形状的语义单元，这一过程受到与类别无关的视觉-语义先验的指导。通过将补丁表示为具有邻接边的图节点，SASGs保留了其内部空间组织，而不是将其视为无序集合。随后，子图状态空间模块（Subgraph State Space Module, SG-SSM）结合了用于子图内部拓扑编码的图神经网络编码器与基于Mamba的状态空间编码器，以高效地对大量子图进行上下文化处理。该模块将语义单元内的局部结构信息与其在WSI中分布和共现所产生的全局上下文信息整合，同时高效建模大量空间分布区域。在四个WSI亚型数据集上的广泛实验表明，与代表性的最先进方法相比，SASG-SSM具有一致的优势。此外，在小样本和少量样本设置下的进一步评估展示了在有限训练数据下的鲁棒性和数据效率。代码将发布在 https://github.com/HLSvois/SASG-SSM。

View on arXiv Download PDF AI Translation

cs.CV / 58 / 2609.03690

MetaStructAtlas: A Grounded 3D Vision-Language Dataset and Benchmark for Functional and Structural Reasoning in Whole-Body PET/CT

MetaStructAtlas：一个用于全身 PET/CT 功能和结构推理的基础 3D 视觉-语言数据集与基准

Zheng, Chenguang, Xue, Le, Zhang, Yichi, Zhang, Wenbo, Ling, Zehui, Feng, Gang, Gao, Xin, Qi, Yuan, Cheng, Yuan, Hu, Zixin, Tian, Mei

Abstract

The joint interpretation of metabolic function and anatomical structure is essential for clinical diagnosis in whole-body PET/CT. Although recent advances in 3D medical vision-language models have demonstrated remarkable progress, current efforts are limited to regional CT imaging, leaving a critical void in comprehensive whole-body PET/CT analysis. In this work, we introduce MetaStructAtlas, a large-scale dataset for grounded whole-body PET/CT interpretation that synthesizes multimodal imaging with integrated anatomical, metabolic, and semantic annotations. MetaStructAtlas provides 490 co-registered 3D PET and CT volumes with 50,470 organ-level segmentation masks and grounded radiology reports. To facilitate interactive reasoning, we further developed MetaStructVQA, a standardized 3D grounded visual question-answering benchmark containing 100,565 QA pairs. This framework explicitly links diagnostic queries to visual evidence across modalities, encompassing anatomical, morphological, and metabolic characteristics. Finally, we evaluate state-of-the-art 3D medical VLMs on MetaStructVQA, establishing a robust foundation for multimodal representation learning and integrated whole-body reasoning in nuclear medicine.

Chinese Translation

在全身 PET/CT 中，代谢功能与解剖结构的联合解读对临床诊断至关重要。尽管最近在 3D 医学视觉-语言模型方面取得了显著进展，但当前的研究工作仅限于区域 CT 成像，导致在全面的全身 PET/CT 分析中存在重要空白。在本研究中，我们介绍了 MetaStructAtlas，这是一个用于基础全身 PET/CT 解读的大规模数据集，综合了多模态成像以及集成的解剖、代谢和语义注释。MetaStructAtlas 提供了 490 个共注册的 3D PET 和 CT 体积，包含 50,470 个器官级分割掩膜和基础放射学报告。为了促进交互式推理，我们进一步开发了 MetaStructVQA，一个标准化的 3D 基础视觉问答基准，包含 100,565 对问答对。该框架明确将诊断查询与跨模态的视觉证据联系起来，涵盖了解剖、形态和代谢特征。最后，我们在 MetaStructVQA 上评估了最先进的 3D 医学视觉-语言模型，为核医学中的多模态表示学习和综合全身推理奠定了坚实基础。

View on arXiv Download PDF AI Translation

cs.CV / 59 / 2609.03694

Observation-Conditioned Latent Energy Priors for Sparse Implicit Neural Shape Completion

基于观察条件的潜在能量先验用于稀疏隐式神经形状补全

Büschl, Paul, de la Rosa, Ezequiel, Wolleb, Julia, McGinnis, Julian, Nombela-Arrieta, César, Menze, Bjoern

Abstract

Implicit neural representations (INRs) can model continuous 3D shapes with a shared coordinate decoder and per-instance latent codes. At test time, autodecoder-style models commonly freeze the decoder and optimize a new latent code from sparse off-grid SDF samples. When these samples underconstrain inference, the latent can drift toward regions that fit the observations but decode implausible unobserved geometry. We propose a post-hoc observation-conditioned latent energy prior for frozen INR decoders. The energy scores standardized latents conditioned on a permutation-invariant encoding of the sparse observation set and is used as a residual expert alongside an L2 latent prior selected on validation data. We evaluate on a controlled cell-nucleus SDF dataset and a public MedShapeNet-derived SDF completion dataset. The proposed L2 objective augmented with conditional energy improves consistently over a validation-selected L2 baseline in the sparsest cell-nucleus regimes and, on MedShapeNet, outperforms both L2 and a six-component GMM latent-density prior across all reported readouts. A shuffled-context ablation is consistently weaker than matched context, supporting an observation-specific contribution. These results suggest that lightweight conditional energies can make pretrained INR decoders more observation-aware without retraining.

Chinese Translation

隐式神经表示（INRs）能够通过共享坐标解码器和每个实例的潜在编码来建模连续的三维形状。在测试阶段，自编码器风格的模型通常会冻结解码器，并从稀疏的离散SDF样本中优化新的潜在编码。当这些样本对推断的约束不足时，潜在编码可能会漂移到适合观察结果的区域，但解码出不合理的未观察几何形状。我们提出了一种后验观察条件的潜在能量先验，用于冻结的INR解码器。能量分数标准化潜在编码，条件依赖于稀疏观察集的置换不变编码，并作为与在验证数据上选择的L2潜在先验相结合的残差专家进行使用。我们在一个受控的细胞核SDF数据集和一个公共的MedShapeNet派生的SDF补全数据集上进行了评估。所提出的L2目标与条件能量相结合，在最稀疏的细胞核状态下，始终优于验证选择的L2基线，并且在MedShapeNet上，超越了L2和六个成分的GMM潜在密度先验在所有报告的读数中。随机上下文的消融实验始终弱于匹配上下文，支持观察特定的贡献。这些结果表明，轻量级的条件能量可以使预训练的INR解码器在不重新训练的情况下更具观察意识。

View on arXiv Download PDF AI Translation

cs.CV / 60 / 2609.03695

SignSeek: Learning Transferable Representations for Sign Dictionary Retrieval

SignSeek：学习可迁移的手语词典检索表示

Asasi, Sobhan, Sincan, Ozge Mercanoglu, Bowden, Richard

Abstract

Sign language dictionaries are essential resources for sign language learners, yet automatically retrieving a sign from a dictionary, given only a query video, remains a challenging problem due to the natural variability between signers. Existing sign representation learning methods are built for closed-set recognition, producing embeddings that do not generalise to the open-set, signer-independent setting that retrieval demands. \textbf{SignSeek} closes this gap by contrastively learning sign representations with saliency-guided articulator masking. A contrastive objective aligns same-gloss signs across signers, while our Articulator Saliency-Guided Masking (ASGM) pinpoints the single most critical articulator per sign. This drives two complementary objectives, a masked contrastive alignment (MAC) loss that sees the sign through a single articulator and a masked prediction (MAP) loss that reconstructs it in latent space from the surrounding spatio-temporal context. Pretrained on 266K samples ($\sim$5,700 glosses) across multiple sign languages, \textbf{SignSeek} sets a new state-of-the-art performance in cross-corpus retrieval on ASL-Citizen, WLASL, and NMFs-CSL without any downstream fine-tuning. Strikingly, it achieves zero-shot generalisation to an entirely unseen British Sign Language (BSL), surpassing methods explicitly trained on BSL, and transfers seamlessly to isolated sign recognition and subtitle alignment, outperforming prior skeleton-based methods.

Chinese Translation

手语词典是手语学习者的重要资源，但仅通过查询视频自动检索词典中的手势仍然是一个具有挑战性的问题，因为手势使用者之间存在自然的变异性。现有的手势表示学习方法是为闭集识别而构建的，产生的嵌入无法推广到检索所需的开放集、与使用者无关的设置。\textbf{SignSeek}通过对比学习手势表示，并结合显著性引导的发音器遮蔽，填补了这一空白。对比目标在不同的手势使用者之间对齐相同词义的手势，而我们的发音器显著性引导遮蔽（Articulator Saliency-Guided Masking, ASGM）则确定每个手势中最关键的发音器。这驱动了两个互补的目标：一个是通过单一发音器观察手势的遮蔽对比对齐（Masked Contrastive Alignment, MAC）损失，另一个是从周围时空上下文中在潜在空间中重建手势的遮蔽预测（Masked Prediction, MAP）损失。在多个手语中，基于266K样本（约5,700个词义）进行预训练的\textbf{SignSeek}在ASL-Citizen、WLASL和NMFs-CSL的跨语料库检索中设定了新的最先进性能，而无需任何下游微调。值得注意的是，它在完全未见过的英国手语（British Sign Language, BSL）上实现了零样本泛化，超越了专门针对BSL训练的方法，并无缝迁移到孤立手势识别和字幕对齐，优于之前基于骨架的方法。

View on arXiv Download PDF AI Translation

cs.CV / 61 / 2609.03729

Unfold The World: Factorize 4D Properties in Reinforcing Spatial Reasoning

展开世界：在强化空间推理中分解四维属性

Yang, Yijun, Zheng, Shenghe, Li, Wenbo, Liu, Jianhui, Sun, Haoze, Zhang, Yanbing, Jiang, Jiaxiu, Song, Lin, Huang, Haoyang, Duan, Nan, Zhu, Lei

Abstract

Despite the remarkable prowess of Vision-Language Models (VLMs) in general multimodal tasks, they remain fundamentally ``flat'' when reasoning about the physical world. We argue that this spatial bottleneck stems from a profound dimensional mismatch: while VLMs are trained to interpret 2D projections, true spatial reasoning demands the recovery of latent 3D geometry and temporal continuity. To conquer this high-dimensional complexity, we advocate a shift from monolithic learning to a ``divide and conquer'' paradigm. We present FactoSR, a factorized reinforcement learning framework that explicitly interpret the dimensions collapsed by visual projection. At its core, FactoSR decomposes the monolithic problem of world-consistent reasoning into three orthogonal, geometric sub-objectives: planar correspondence ($XY$), depth consistency ($Z$), and temporal reversibility ($T$). By optimizing these verifiable constraints within a unified policy learning mechanism, we effectively transform an ill-posed projection recovery problem into a series of tangible reasoning steps. Extensive evaluations on multi-view and video benchmarks demonstrate that this elegant decomposition yields substantial gains in 3D and 4D reasoning, achieving a 5.9% boost on VSI-Bench and 4.5% on All-Angles-Bench. Our findings suggest that reinforcing explicit, factorized 4D consistency is a critical step toward evolving VLMs into robust, world-aware reasoners.

Chinese Translation

尽管视觉-语言模型（VLMs）在一般多模态任务中展现出卓越的能力，但在推理物理世界时，它们仍然在本质上是“平面的”。我们认为，这一空间瓶颈源于深刻的维度不匹配：VLMs被训练用于解释二维投影，而真正的空间推理则需要恢复潜在的三维几何形状和时间连续性。为了克服这种高维复杂性，我们主张从单一学习转向“分而治之”的范式。我们提出了FactoSR，一个分解的强化学习框架，明确解释了视觉投影所压缩的维度。FactoSR的核心是将世界一致性推理的单一问题分解为三个正交的几何子目标：平面对应（$XY$）、深度一致性（$Z$）和时间可逆性（$T$）。通过在统一的策略学习机制中优化这些可验证的约束，我们有效地将一个不适定的投影恢复问题转化为一系列具体的推理步骤。在多视角和视频基准上的广泛评估表明，这种优雅的分解在三维和四维推理中带来了显著的提升，在VSI-Bench上提高了5.9%，在All-Angles-Bench上提高了4.5%。我们的研究结果表明，强化明确的、分解的四维一致性是将VLMs演变为强大、具备世界意识的推理者的重要一步。

View on arXiv Download PDF AI Translation

cs.CV / 62 / 2609.03740

Fill My Mirror: Geometry-Constrained Mirror Inpainting

填补我的镜子：几何约束下的镜像修复

Basson, Ofek, Vainer, Shimon, Hel-Or, Yacov, Fried, Ohad

Abstract

Mirrors are common in real-world images, yet producing geometrically consistent reflections with generative models remains challenging. Unlike most objects, mirror appearance depends on scene geometry and viewpoint, making it hard to synthesize using learned appearance priors alone. We address this in the mirror inpainting setting, where the scene is fixed and only the mirror region is generated. Our key insight is that much mirror content is geometrically constrained by the visible scene and need not be hallucinated. We estimate scene geometry and project visible content into the mirror to recover reflection regions determined by geometry. A generative model then completes the mirror region via a two-mask diffusion strategy balancing geometric constraints with the model's learned priors, reducing projection artifacts and improving reflection consistency. The method is training-free and applicable to complex real-world scenes. We evaluate on MirrorBench-V2 (synthetic) and real images. Using standard and geometry-aware metrics, we show that explicitly using scene geometry improves consistency.

Chinese Translation

镜子在现实世界图像中非常常见，但使用生成模型生成几何一致的反射仍然具有挑战性。与大多数物体不同，镜子的外观依赖于场景几何和视角，仅凭学习到的外观先验很难进行合成。我们在镜像修复的背景下解决了这一问题，其中场景是固定的，仅生成镜子区域。我们的关键见解是，许多镜子内容受到可见场景的几何约束，无需进行幻觉生成。我们估计场景几何并将可见内容投影到镜子中，以恢复由几何决定的反射区域。然后，生成模型通过一种两掩码扩散策略完成镜子区域，该策略在几何约束与模型学习到的先验之间取得平衡，从而减少投影伪影并提高反射一致性。该方法无需训练，适用于复杂的现实场景。我们在MirrorBench-V2（合成）和真实图像上进行了评估。使用标准和几何感知指标，我们表明显式使用场景几何可以提高一致性。

View on arXiv Download PDF AI Translation

cs.CV / 63 / 2609.03742

KnowVis: Knowledge-Centric Visual Summarization for Video Lectures

KnowVis：面向知识的在线视频讲座可视化摘要

Xu, Yi, Hou, Yifan, Zhang, Xiaoyu

Abstract

Video lectures are valuable educational resources, but their dense and lengthy formats often overwhelm novice learners. This difficulty stems from a fundamental pedagogical mismatch: while videos deliver transient information linearly, human learning requires constructing interconnected cognitive networks, a task that induces severe cognitive overload for novice learners lacking prior domain knowledge. Existing video summarization methods fail to resolve this mismatch, as they primarily produce text-heavy, linear condensations that still demand high cognitive effort. To bridge this gap, we propose KnowVis, a framework that transforms linear video lectures into pedagogically grounded visual narratives. KnowVis first extracts a detailed concept map from multimodal video content to identify important and challenging threshold concepts, then constructs structured knowledge units, and finally synthesizes engaging visual summaries. Alongside the framework, we introduce a curated dataset of 125 educational videos across 10 academic disciplines, paired with 1,079 generated visual summaries. Extensive automated evaluations and a human study demonstrate that, compared to state-of-the-art baselines, KnowVis generates more accurate and clear visuals that successfully reduce cognitive load and significantly improve student learning effectiveness and knowledge retention.

Chinese Translation

视频讲座是宝贵的教育资源，但其密集且冗长的格式常常让初学者感到不知所措。这种困难源于一种基本的教学不匹配：视频以线性方式传递瞬时信息，而人类学习需要构建相互关联的认知网络，这一任务对缺乏先前领域知识的初学者造成了严重的认知负担。现有的视频摘要方法未能解决这一不匹配，因为它们主要生成文本密集的线性浓缩内容，仍然需要高认知努力。为了解决这一问题，我们提出了KnowVis，一个将线性视频讲座转化为以教学为基础的可视化叙事的框架。KnowVis首先从多模态视频内容中提取详细的概念图，以识别重要且具有挑战性的阈值概念，然后构建结构化的知识单元，最后合成引人入胜的可视化摘要。除了该框架，我们还介绍了一个涵盖10个学科的125个教育视频的精心策划的数据集，并配有1,079个生成的可视化摘要。大量的自动化评估和人类研究表明，与最先进的基线相比，KnowVis生成的可视化内容更准确、更清晰，成功降低了认知负担，并显著提高了学生的学习效果和知识保留率。

View on arXiv Download PDF AI Translation

cs.CV / 64 / 2609.03756

ENEAS: Embedding-guided Neural Ensemble for Adaptive Segmentation

ENEAS：嵌入引导的神经集成方法用于自适应分割

del Pino, Javier, Rodríguez, Salvador, Garabito, Alejandro, Álvarez, Javier, Garabito, Chema

Abstract

We present ENEAS, a unified, text-promptable method for instance tracking and semantic discovery. Text-promptable segmentation models, including the latest foundation models such as SAM 3, still suffer from temporal hallucinations, spatial fragmentation, and semantic misclassification: they fail to report target absence when an object leaves the field of view, segment local textures instead of the complete object during extreme close-ups, and prioritize visual features over ontological reality, so that visually similar artifacts such as statues, paintings, or reflections are segmented as target entities. ENEAS works two ways from a single method: precise tracking and high-quality segmentation of a unique instance, and open-concept discovery of every instance a text query names, resolved by a semantic verification layer. For tracking, we extend the geometrically robust SeC architecture, previously limited to point interactions, with a text-prompting adapter and leverage its temporal memory, so that the target is held through disappearance without drifting to distractors and kept whole even when it fills the entire view. For discovery, the verification layer combines high-speed visual embedding matching with conditional VLM refinement, invoking semantic reasoning only for ambiguous candidates, which filters out the ontological errors that visual-only models cannot distinguish while keeping latency low. Designed with 3D reconstruction in mind, where a single misclassified distractor corrupts the asset, ENEAS unlocks high-quality semantic tracking and segmentation of video, of broad libraries, and of collections of temporally or spatially unordered data, together with the discrimination to tell true instances from their doppelgangers: things that look alike but are not the same. The code and models are available at https://github.com/speridlabs/eneas

Chinese Translation

我们提出了ENEAS，一种统一的、可通过文本提示的方法，用于实例跟踪和语义发现。可通过文本提示的分割模型，包括最新的基础模型如SAM 3，仍然面临时间幻觉、空间碎片化和语义误分类的问题：当物体离开视野时，它们无法报告目标缺失，在极近距离拍摄时分割局部纹理而非完整物体，并且优先考虑视觉特征而非本体现实，以至于视觉上相似的物体如雕像、画作或反射被分割为目标实体。ENEAS通过单一方法实现两种功能：对唯一实例的精确跟踪和高质量分割，以及对文本查询命名的每个实例的开放概念发现，这一过程通过语义验证层来解决。对于跟踪，我们扩展了几何鲁棒的SeC架构，之前仅限于点交互，增加了文本提示适配器，并利用其时间记忆，使得目标在消失时不漂移到干扰物上，即使在填满整个视野时也能保持完整。对于发现，验证层结合了高速视觉嵌入匹配和条件VLM精炼，仅对模糊候选者进行语义推理，从而过滤出视觉模型无法区分的本体错误，同时保持低延迟。ENEAS在设计时考虑了3D重建，其中单个误分类的干扰物会破坏资产，解锁了视频的高质量语义跟踪和分割，广泛的库，以及时间或空间上无序数据的集合，同时具备区分真实实例与其替身的能力：看起来相似但并不相同的事物。代码和模型可在https://github.com/speridlabs/eneas获取。

View on arXiv Download PDF AI Translation

cs.CV / 65 / 2609.03773

RealCADBench: Benchmarking Parametric CAD Modeling from Industrial Design Intents

RealCADBench：基于工业设计意图的参数化CAD建模基准测试

JoyIndustrial VisCAD Team, Cai, Linxin, Hong, Qiuhe, Huang, Zhichao, Li, Guanlin, Li, Zongzhen, Liu, Hongsen, Long, Yichen, Wang, Wei, Wang, Yuchen, Yang, Dongyue, Yu, Huimu, Zhong, Xianwen

Abstract

Parametric computer-aided design (CAD) modeling is difficult to evaluate with a single metric. Existing CAD benchmarks often emphasize synthetic or CAD-native settings, limited input modalities, or executability and IoUs alone. We introduce RealCADBench, a benchmark for intent-to-program CAD modeling from real industrial design intents. It contains 12,632 tasks from 19 factory-automation categories and spans text descriptions, 2D engineering drawings, real product pictures, and rendered images for both Part and Assembly modeling. We report results on a 1,770-task evaluation slice: 1,745 Part tasks across four input regimes and RCB-Assm25, a 25-task assembly study used in every reported assembly comparison. Each method generates FreeCAD API Python, which a shared runtime executes to export the 3D model. We evaluate the exported model using executability, Solid IoU, Surface IoU, and a rubric-based visual-semantic identity Judge. Among the nine standalone frontier large models evaluated, no model leads all four metrics. Across six frontier-scale large models, executability ranges from 0.565 to 0.812, Solid IoU from 0.2841 to 0.5379, and Surface IoU from 0.112 to 0.217 across the four Part regimes. The highest regime-balanced composite comes from a different model than the leaders on the four component metrics. On RCB-Assm25, Codex with GPT-5.5 improves executability and both IoU metrics over standalone GPT-5.5, but lowers the Judge score by 6.98 percentage points, leaving GPT-5.5 as the Judge leader. We also observe recurring failure modes, most notably missing fine structures, loss of part identity, and incorrect assembly placement. These results show that execution alone is insufficient to characterize realistic CAD modeling and that frontier models and agents differ substantially across executability, IoUs, and visual-semantic identity.

Chinese Translation

参数化计算机辅助设计（CAD）建模难以通过单一指标进行评估。现有的CAD基准测试通常强调合成或CAD原生设置、有限的输入方式，或仅关注可执行性和IoU（交并比）。我们提出了RealCADBench，这是一个基于真实工业设计意图的意图到程序的CAD建模基准。它包含来自19个工厂自动化类别的12,632个任务，涵盖文本描述、二维工程图、真实产品图片和渲染图像，适用于零件和装配建模。我们报告了在1,770个任务评估切片上的结果：1,745个零件任务，跨越四种输入模式，以及RCB-Assm25，这是每个报告的装配比较中使用的25个任务的装配研究。每种方法生成FreeCAD API Python代码，由共享运行时执行以导出3D模型。我们使用可执行性、Solid IoU、Surface IoU和基于评分的视觉语义身份评估导出的模型。在评估的九个独立前沿大型模型中，没有一个模型在所有四个指标上都领先。在六个前沿规模的大型模型中，可执行性范围从0.565到0.812，Solid IoU从0.2841到0.5379，Surface IoU在四种零件模式中从0.112到0.217。最高的模式平衡复合来自于与四个组件指标的领先者不同的模型。在RCB-Assm25中，Codex与GPT-5.5相比提高了可执行性和两个IoU指标，但Judge评分降低了6.98个百分点，使得GPT-5.5成为Judge的领先者。我们还观察到反复出现的失败模式，尤其是缺失细微结构、部件身份丧失和错误的装配位置。这些结果表明，仅靠执行能力不足以表征现实的CAD建模，前沿模型和代理在可执行性、IoU和视觉语义身份上存在显著差异。

View on arXiv Download PDF AI Translation

cs.CV / 66 / 2609.03788

A Reverse Sign Language Dictionary: Open-Vocabulary Sign Recognition from Continuous Signing via Video Captioning and Description Retrieval

反向手语词典：通过视频字幕和描述检索实现连续手势的开放词汇手势识别

Poveda-Gutiérrez, Santiago, Nakayama, Hideki, Bono, Mayumi

Abstract

Isolated Sign Language Recognition (ISLR) is conventionally cast as closed-set classification over gloss labels, which cannot generalize to signs unseen in training and ties every deployment to a gloss-annotated lexicon. We instead recognize signs extracted from continuous signing by (1) captioning a sign-level clip into a free-form procedural description of the articulation with an open-weight vision-language model, and (2) retrieving the closest entry from a vocabulary of target descriptions with a multilingual sentence encoder: a reverse sign language dictionary that needs no gloss supervision and admits an open vocabulary. On 1,300 sign-level segments from a Japanese Sign Language (JSL) dialogue corpus annotated with procedural descriptions (against a 2% top-10 chance floor over the 503-entry target vocabulary), fine-tuning the captioner substantially improves seen-class retrieval: language and vision tower fine-tuning raises top-10 retrieval on seen classes from 4.5% (untrained) to 49%, becoming statistically indistinguishable from a standard supervised closed-set classifier (I3D) on two of the three test sets where a closed-set classifier can be evaluated at all. More importantly, unseen-class retrieval also improves significantly over the untrained pipeline (11.5% -> 21.0% top-10, p=0.0094), a regime in which the closed-set classifier cannot participate. A matcher-side empirical upper-bound analysis shows the sentence encoder already recovers close to 100% of paraphrased gold descriptions, locating a gap in captioning quality that we aim to address in future work. To our knowledge this is the first description-based, open-vocabulary sign lookup from continuous signing without gloss supervision, and the first for JSL.

Chinese Translation

孤立手语识别（ISLR）通常被视为对词汇标签的闭集分类，这无法推广到训练中未见的手势，并将每次部署与带有词汇注释的词典绑定在一起。我们则通过以下方式识别从连续手势中提取的手势：(1) 利用开放权重的视觉-语言模型将手势级剪辑字幕化为自由形式的程序描述，(2) 使用多语言句子编码器从目标描述词汇中检索最接近的条目：一个不需要词汇监督并允许开放词汇的反向手语词典。在一个带有程序描述注释的日语手语（JSL）对话语料库中对1,300个手势级片段进行微调，显著提高了已见类别的检索：语言和视觉塔的微调将已见类别的前10名检索率从4.5%（未训练）提高到49%，在三个测试集中，其中一个闭集分类器可以进行评估的情况下，与标准监督闭集分类器（I3D）在统计上无显著差异。更重要的是，未见类别的检索也显著改善（11.5% -> 21.0%前10名，p=0.0094），这是闭集分类器无法参与的领域。匹配器侧的经验上限分析显示，句子编码器已经接近100%恢复了同义改写的黄金描述，揭示了我们未来工作中旨在解决的字幕质量差距。据我们所知，这是第一个基于描述的、开放词汇的连续手势查找，且不需要词汇监督，也是JSL的首次尝试。

View on arXiv Download PDF AI Translation

cs.CV / 67 / 2609.03796

LLaDA-Image: Building Strong Image Generators with Fully Open Training Recipes

LLaDA-Image：构建强大的图像生成器的完全开放训练方案

Chen, Chuyan, Chen, Haoxing, Chen, Kun, Cheng, Zhenglin, Cui, Long, Fang, Ruishan, Gu, Zhangxuan, Huang, Zhicheng, Lan, Zhenzhong, Lei, Yuanting, Li, Haoquan, Li, Jianguo, Li, Rongchuan, Li, Sidu, Lin, Tao, Liu, Deyuan, Liu, Jiacheng, Liu, Lin, Lou, Yuxuan, Lu, Zhisheng, Ma, Yuxin, Shen, Shuheng, Sun, Peng, Wang, Chaoyang, Wang, Hongjun, Wang, Xiaomei, Wang, Yongxin, Wu, Chengzhang, Wu, Hongru, Xie, Jun

Abstract

We introduce LLaDA-Image, a unified framework that pairs a 6B Diffusion Transformer (DiT) trained from scratch with a frozen vision-language understanding module built on the LLaDA2.0-Mini diffusion language model backbone. Instead of relying heavily on paired image-text data from the beginning, we first build a strong visual generative prior through image-only pre-training and mid-training. The generation pipeline comprises 220M samples, 98 of which are real images. For efficient and scalable optimization, we use parameter-free RMSNorm throughout the DiT together with the Muon optimizer. The resulting unified model produces highly photorealistic images while accurately following fine-grained editing instructions. We further distill LLaDA-Image into LLaDA-Image-Turbo, enabling fast inference in 2-4 sampling steps. On Qwen-Image-Bench, LLaDA-Image achieves overall scores of 53.53 and 53.38 on the English and Chinese tracks, respectively, setting a new state-of-the-art among open-source models on both tracks. To support further research on capable and efficient generative models, we release our model weights, training code, and detailed recipes.

Chinese Translation

我们介绍了LLaDA-Image，这是一个统一框架，将一个从头开始训练的6B扩散变换器（Diffusion Transformer, DiT）与一个基于LLaDA2.0-Mini扩散语言模型骨干的冻结视觉-语言理解模块相结合。我们并不从一开始就过度依赖配对的图像-文本数据，而是首先通过仅图像的预训练和中期训练构建一个强大的视觉生成先验。生成管道包含220M样本，其中98个是真实图像。为了实现高效和可扩展的优化，我们在整个DiT中使用无参数的RMSNorm，并结合Muon优化器。最终得到的统一模型能够生成高度逼真的图像，同时准确遵循细粒度的编辑指令。我们进一步将LLaDA-Image提炼为LLaDA-Image-Turbo，使其在2-4个采样步骤中实现快速推理。在Qwen-Image-Bench上，LLaDA-Image在英语和中文轨道上分别获得了53.53和53.38的总分，在两个轨道上都创造了开源模型的新最优状态。为了支持对高效能和高效生成模型的进一步研究，我们发布了我们的模型权重、训练代码和详细方案。

View on arXiv Download PDF AI Translation

cs.CV / 68 / 2609.03804

Urban Boundaries, Social Barriers: A Benchmark and Vision-Centric Framework for Mapping Gated Communities and Equity Implications

城市边界与社会障碍：一个基准和以愿景为中心的框架用于映射封闭社区及其公平性影响

Zhao, Minwei, Zhang, Weiming, Du, Jiawang, Liu, Qiming, Zhuang, Weiming, Nie, Pei, Wu, Cai

Abstract

Communities are fundamental spatial units that shape urban form and social life. Whether a residential compound is spatially open or enclosed affects mobility, access to public services, and equity, yet studies of Chinese fengbi xiaoqu remain largely qualitative or small-scale, limiting reproducible city-scale analysis. We address this gap by introducing GBA-GCs, a metropolitan-scale multimodal benchmark for locally grounded gated/open community recognition in China's Greater Bay Area, covering 37,444 residential compounds with aligned boundary polygons, high-resolution satellite imagery, Chinese metadata, and structured attributes, together with expert-verified labels, inter-annotator reliability, and official evaluation splits. Built on this benchmark, we present Multimodal Classifier for Gated Community (MCGC), a vision-centric multimodal framework based on DINOv3-SAT that fuses imagery, text, and structured cues via modality-aware cross-attention and adaptive gating to mitigate modality imbalance. MCGC consistently outperforms strong unimodal and multimodal baselines. Finally, we apply the validated model to metropolitan-scale mapping and report equity-oriented findings including spatial clustering of GCs, privatized green space, and reduced pedestrian connectivity. The benchmark, code, and release documentation are available at https://github.com/MinweiZhao/GBA-GCs.

Chinese Translation

社区是塑造城市形态和社会生活的基本空间单元。住宅小区的空间开放性或封闭性影响着流动性、公共服务的获取和公平性，但对中国封闭小区（fengbi xiaoqu）的研究仍然主要是定性或小规模的，限制了可重复的城市规模分析。为填补这一空白，我们引入了GBA-GCs，这是一个针对中国大湾区的城市规模多模态基准，涵盖了37,444个住宅小区，配备了对齐的边界多边形、高分辨率卫星影像、中国元数据和结构化属性，以及专家验证的标签、标注者间可靠性和官方评估分割。基于这一基准，我们提出了封闭社区多模态分类器（MCGC），这是一个基于DINOv3-SAT的以愿景为中心的多模态框架，通过模态感知的交叉注意力和自适应门控融合影像、文本和结构线索，以减轻模态不平衡。MCGC在强大的单模态和多模态基准上表现出色。最后，我们将经过验证的模型应用于城市规模的映射，并报告了以公平性为导向的发现，包括封闭社区的空间聚集、私有化绿地和行人连通性降低。基准、代码和发布文档可在https://github.com/MinweiZhao/GBA-GCs获取。

View on arXiv Download PDF AI Translation

cs.CV / 69 / 2609.03811

VisCAD: A Foundation Model Suite with Multimodal Industrial CAD Intelligence

VisCAD：具有多模态工业CAD智能的基础模型套件

JoyIndustrial VisCAD Team, Cai, Linxin, Hong, Qiuhe, Huang, Zhichao, Li, Guanlin, Liu, Hongsen, Liu, Ziqi, Long, Yichen, Wang, Luya, Wang, Yuchen, Wu, Wenxiang, Yu, Huimu, Zhang, Ning

Abstract

AI-assisted computer-aided design (CAD) for industrial products involves two challenging phases. Part-level generation maps diverse forms of user intent, including renders, text descriptions, 2D drawings, and real photographs, to executable programs in a CAD domain-specific language. Assembly-level generation must additionally handle interacting parts, plan mating relations, estimate poses, and place all parts correctly. Existing specialized CAD models are commonly trained on narrow input domains, such as renders or texts, and often generalize poorly, while general-purpose frontier models cover broader inputs but perform inconsistently across CAD domains. We present VisCAD, a foundation model suite designed to provide both broad generalization and strong CAD capability for realistic industrial products. At its core is VisCAD-M1, a 27B model trained through mid-training and post-training for part-level design generation. On PubCADBench and RealCADBench, VisCAD-M1 achieves the highest average part-level score among the evaluated models, reaching 0.5540 compared with 0.5496 for the strongest frontier model. Reusing VisCAD-M1 as a test-time verifier can further raise the score to 0.5797, an approximately 5 percent relative improvement over the previous state of the art. VisCAD also includes a domain-specific harness that leverages frontier models for complex assembly generation and demonstrates advantages over general-purpose harnesses in both quantitative and qualitative evaluations.

Chinese Translation

人工智能辅助的工业产品计算机辅助设计（CAD）涉及两个具有挑战性的阶段。部件级生成将用户意图的多样形式（包括渲染、文本描述、二维图纸和真实照片）映射到CAD领域特定语言中的可执行程序。组装级生成还必须处理交互部件，规划配合关系，估计姿态，并正确放置所有部件。现有的专业CAD模型通常在狭窄的输入领域（如渲染或文本）上进行训练，且往往泛化能力较差，而通用前沿模型覆盖更广泛的输入，但在CAD领域的表现不一致。我们提出了VisCAD，一个旨在为现实工业产品提供广泛泛化和强大CAD能力的基础模型套件。其核心是VisCAD-M1，一个通过中期训练和后期训练进行训练的27B模型，用于部件级设计生成。在PubCADBench和RealCADBench上，VisCAD-M1在评估模型中实现了最高的平均部件级得分，达到了0.5540，而最强的前沿模型得分为0.5496。将VisCAD-M1作为测试时验证器重用，可以进一步将得分提高到0.5797，相比于之前的最先进水平，约提高了5%的相对改善。VisCAD还包括一个领域特定的工具，利用前沿模型进行复杂的组装生成，并在定量和定性评估中显示出相对于通用工具的优势。

View on arXiv Download PDF AI Translation

cs.CV / 70 / 2609.03813

SPARK: Input-Conditioned Sparse Activation Modulation for Frozen DiT-based Super-Resolution

SPARK：基于输入条件的稀疏激活调制用于冻结的DiT超分辨率

Putamorsi, Federico, Zini, Leonardo, Cornia, Marcella, Baraldi, Lorenzo

Abstract

Real-world image super-resolution (SR) increasingly relies on Diffusion Transformer (DiT) backbones, whose internal activations can be dominated by a small number of massive channels. Yet improving perceptual quality in these models still typically requires fine-tuning the network or attaching additional adapters, leaving this structured activation space largely unexplored for adaptation. We investigate whether dominant channels can instead serve as a compact adaptation interface for frozen DiT-based SR models. We first characterize their behavior in pretrained SR backbones and show through controlled interventions that they strongly affect reconstruction quality. Building on this observation, we introduce SPARK, a lightweight input-conditioned controller that predicts bounded per-channel affine transformations for only the selected channels, while keeping the SR backbone and VAE frozen. Dominant channels are identified through an online activation-ranking procedure, and only a small predictor conditioned on the low-resolution VAE latent is optimized. Experiments on three DiT-based SR backbones across DIV2K, RealSR, and DRealSR show consistent gains in both fidelity and perceptual quality while modulating only eight channels per stream and block. Controlled comparisons further show that these gains cannot be explained by parameter budget or access to the selected channels alone.

Chinese Translation

现实世界中的图像超分辨率（SR）越来越依赖于扩散变换器（DiT）骨干网络，其内部激活通常由少数几个大通道主导。然而，提升这些模型的感知质量通常仍需要对网络进行微调或附加额外的适配器，这使得这一结构化激活空间在适应性方面仍然未被充分探索。我们研究了主导通道是否可以作为冻结的基于DiT的SR模型的紧凑适配接口。我们首先对预训练SR骨干网络中的行为进行特征描述，并通过控制干预展示它们对重建质量的强烈影响。基于这一观察，我们引入了SPARK，一个轻量级的输入条件控制器，它为仅选定通道预测有界的每通道仿射变换，同时保持SR骨干网络和变分自编码器（VAE）冻结。主导通道通过在线激活排名程序进行识别，并且仅优化一个基于低分辨率VAE潜变量的小型预测器。在DIV2K、RealSR和DRealSR上对三种基于DiT的SR骨干网络的实验显示，在仅调制每个流和块的八个通道的情况下，保真度和感知质量均有一致的提升。控制比较进一步表明，这些提升无法仅通过参数预算或对选定通道的访问来解释。

View on arXiv Download PDF AI Translation

cs.CV / 71 / 2609.03820

Select, Compress, Reinvest: A Controlled Study of Visual-Token Allocation in Long-Video MLLMs

选择、压缩、再投资：长视频多模态语言模型中视觉标记分配的对照研究

Khatri, Prakhar

Abstract

Long-video language models cannot look at every frame: an hour sampled once per second is 3,600 images, and a system keeps only a small fixed slice of that pool. Which frames survive that slice is usually treated as a preprocessing detail; we test whether it should be. Published selectors make the comparison hard because they change the frame scorer, the prompt boundary, the resolution policy, and the answering model all at once. We hold each fixed and vary one decision at a time: selection, spatial compression, and reinvestment of the savings, across six training-free selection rules, three long-video benchmarks, and two answering models. Selection is the largest single lever: on LongVideoBench's hour-long bin, eight query-selected frames beat sixteen uniformly spaced ones by 6.9 points, and Orthogonal Matching Pursuit, an unmodified decades-old sparse-approximation algorithm, matches or comes within a point of every purpose-built selector we compare it against, across all three benchmarks. Compression is close to free: halving each frame's spatial budget at fixed timestamps costs at most 0.44 points. Reinvestment is where that budget turns back into accuracy: spending the freed tokens on twice as many compressed frames, at a measured cost no higher than the original eight, returns a further two to three points; compression only pays off once its savings are spent this way. Along the way, an implementation bug in our own AKS baseline and a 0.07 to 3.74 point gap between two harnesses running the same published rules at the same budget show why these comparisons need to happen inside one controlled harness rather than across papers.

Chinese Translation

长视频语言模型无法查看每一帧：一小时每秒采样一次即为3600张图像，而系统仅保留该池中的一小部分固定切片。哪些帧能够存活在这个切片中通常被视为预处理细节；我们测试这是否应该如此。已发布的选择器使得比较变得困难，因为它们同时改变了帧评分器、提示边界、分辨率策略和回答模型。我们保持每个变量不变，并一次改变一个决策：选择、空间压缩和节省的再投资，涵盖六种无训练选择规则、三个长视频基准和两个回答模型。选择是最大的单一杠杆：在 LongVideoBench 的一小时长箱中，八个查询选择的帧比十六个均匀间隔的帧高出6.9分，而正交匹配追踪（Orthogonal Matching Pursuit）这一未修改的几十年老的稀疏近似算法在我们比较的所有三个基准中都能与每个专门构建的选择器匹敌或相差不超过一分。压缩几乎是免费的：在固定时间戳下将每帧的空间预算减半最多只损失0.44分。再投资是预算转化为准确性的地方：将释放的标记用于两倍数量的压缩帧，其测量成本不高于原来的八个，能再返回两到三分；压缩只有在以这种方式花费其节省时才会带来收益。在此过程中，我们自己 AKS 基线中的一个实现错误以及两个在相同预算下运行相同已发布规则的系统之间存在0.07到3.74分的差距，显示了这些比较为何需要在一个受控环境中进行，而不是跨论文进行。

View on arXiv Download PDF AI Translation

cs.CV / 72 / 2609.03824

VI3: Grounding Pretrained 3D Foundation Models with Inertial Cues

VI3：利用惯性线索为预训练的3D基础模型提供度量基础

Lozano, Ernesto, Jaenal, Alberto, Civera, Javier

Abstract

3D foundation models (3DFMs) excel at predicting camera poses and dense depth from multiple views of a scene, showcasing strong zero-shot generalization. However, as metric scale is not observable from monocular images, their absolute scale predictions are typically inaccurate. Inertial measurement units (IMUs), present in most devices, naturally complement monocular cameras by observing scaled motion. We introduce VI3, a model-agnostic framework that metrically anchors a pretrained 3DFM using only IMU readings. VI3 initializes and preintegrates the IMU to obtain a metric motion reference, which is then used to recover the scale of the 3DFM outputs. Our method includes adaptable anchoring strategies tailored to diverse 3DFM architectures. Experiments on synthetic and real aerial datasets demonstrate that VI3 recovers metric scale without ground-truth supervision while preserving geometric consistency, acting as a fine refinement under well-conditioned motion and as a strong prior when motion is less informative.

Chinese Translation

3D基础模型（3DFMs）在从场景的多个视角预测相机姿态和密集深度方面表现出色，展现出强大的零-shot泛化能力。然而，由于单目图像无法观察到度量尺度，它们的绝对尺度预测通常不准确。惯性测量单元（IMUs）在大多数设备中存在，能够通过观察缩放运动自然地补充单目相机。我们提出了VI3，一个模型无关的框架，仅使用IMU读数对预训练的3DFM进行度量锚定。VI3初始化并预积分IMU，以获得度量运动参考，然后用于恢复3DFM输出的尺度。我们的方法包括针对不同3DFM架构量身定制的可适应锚定策略。在合成和真实的航空数据集上的实验表明，VI3在没有真实值监督的情况下恢复度量尺度，同时保持几何一致性，在良好条件下的运动中作为精细调整，而在运动信息较少时作为强先验。

View on arXiv Download PDF AI Translation

cs.CV / 73 / 2609.03829

The impact of phase information for few-shot fine-grained image classification

相位信息对少样本细粒度图像分类的影响

Liu, Ruiling, Zhang, Linyue, Zeng, Wenyi, Lu, Jiamiao, Zhang, Weichuang, Sun, Changming, Zhang, Zejun, Zhao, Xiao

Abstract

Few-shot fine-grained image classification (FSFGIC) aims to classify similar images with limited labeled examples. This work highlights the critical yet underutilized role of phase information in capturing structural relationships within an image. This study introduces a novel plug-and-play amplitude-phase integration (API) module that effectively combines local and global frequency amplitude and phase information for obtaining more comprehensive feature descriptors. Additionally, a dedicated network, named PSF-Net, is proposed that adaptively fuses phase-based spatial and frequency information for FSFGIS. The designed PSF-Net can be easily integrated into standard episodic training architectures for end-to-end training from scratch. Extensive experiments on five public datasets demonstrate that the method outperforms existing state-of-the-art benchmarks.

Chinese Translation

少样本细粒度图像分类（FSFGIC）旨在对具有有限标注样本的相似图像进行分类。本研究强调了相位信息在捕捉图像内部结构关系中的关键但未充分利用的作用。本文提出了一种新颖的即插即用幅度-相位集成（API）模块，该模块有效地结合了局部和全局频率的幅度与相位信息，以获得更全面的特征描述符。此外，提出了一种专用网络，命名为 PSF-Net，该网络自适应地融合基于相位的空间和频率信息以用于 FSFGIC。设计的 PSF-Net 可以轻松集成到标准的情节训练架构中，实现从头到尾的训练。对五个公共数据集的广泛实验表明，该方法优于现有的最先进基准。

View on arXiv Download PDF AI Translation

cs.CV / 74 / 2609.03892

GraFT: A Training-Free Framework for Spatial Reasoning in Multimodal Large Language Models via 3D Scene Graphs

GraFT：一种基于3D场景图的多模态大语言模型空间推理的无训练框架

Du, Junqing, Ropero, Fernando, Turkoz, Erkin, Zhang, Yanfeng, Liu, Lu

Abstract

3D spatial reasoning underpins understanding and acting in the physical world, yet it remains unreliable in current multimodal large language models (MLLMs). These models falter at precise geometric measurement, at transforming between egocentric and allocentric viewpoints, and at grounding fine-grained appearance. The most common remedies fine-tune the model on large-scale curated spatial-reasoning datasets or attach dedicated encoders for 3D geometry, which typically couples the solution to costly supervision and a specific backbone. We instead introduce GraFT, a training-free framework that supplies the missing 3D structure through a compact, easily maintained 3D scene graph (3DSG). From this 3DSG, GraFT provides three spatial reasoning capabilities: (1) deterministic geometry through symbolic tools, (2) allocentric layout through a bird's-eye-view (BEV) rendering, and (3) visual-attribute grounding through task-relevant egocentric frames. On ScanQA, GraFT improves every metric over the same-backbone baseline, raising CIDEr by 27%. On VSI-Bench, GraFT improves frozen MLLMs by up to 65%, surpassing every proprietary and general-purpose open-source baseline, and several prominent fine-tuned spatial models.

Chinese Translation

3D空间推理是理解和在物理世界中行动的基础，但在当前的多模态大语言模型（MLLMs）中仍然不可靠。这些模型在精确几何测量、在自我中心和他中心视角之间的转换以及在细粒度外观的基础上表现不佳。最常见的解决方案是对模型进行大规模策划的空间推理数据集的微调，或为3D几何附加专用编码器，这通常使得解决方案依赖于昂贵的监督和特定的骨干网络。相反，我们提出了GraFT，这是一种无训练框架，通过紧凑且易于维护的3D场景图（3DSG）提供缺失的3D结构。通过这个3DSG，GraFT提供了三种空间推理能力：（1）通过符号工具实现确定性几何，（2）通过鸟瞰图（BEV）渲染实现他中心布局，以及（3）通过与任务相关的自我中心框架实现视觉属性基础。在ScanQA上，GraFT在相同骨干网络的基线之上提高了每个指标，CIDEr提高了27%。在VSI-Bench上，GraFT使冻结的MLLMs提高了多达65%，超越了所有专有和通用开源基线，以及几个显著的微调空间模型。

View on arXiv Download PDF AI Translation

cs.CV / 75 / 2609.03895

Concept of a Sensor Test Environment for Dusty Agricultural Conditions

尘土农业条件下传感器测试环境的概念

Buckel, Peter, Hermann, Johannes, Wollmann, Jonas, Dietmueller, Thomas, Oksanen, Timo

Abstract

Dust in agriculture presents a significant challenge for autonomous agricultural machinery. Dust can impair the performance of sensors and algorithms. This work, therefore, presents a concept for a proving ground consisting of an indoor and outdoor area. The indoor area comprises a laboratory test bench where dust circulates in a closed system and a test hall where life-size objects can be placed. The outdoor area features dedicated test setups that enable reproducible data to be recorded with and without dust during real-world agriculture work. The proving ground and the setups are visualized in 3D.

Chinese Translation

农业中的尘土对自主农业机械构成了重大挑战。尘土会影响传感器和算法的性能。因此，本文提出了一个由室内和室外区域组成的验证场概念。室内区域包括一个实验室测试台，尘土在封闭系统中循环，以及一个可以放置实物的测试大厅。室外区域则设有专门的测试装置，能够在真实农业工作中记录有尘土和无尘土的可重复数据。验证场及其装置以3D形式进行可视化。

View on arXiv Download PDF AI Translation

cs.CV / 76 / 2609.03919

OctWorld: Long-Range World-Consistent Video Generation with Octree-Based 3D Mapping

OctWorld：基于八叉树的长距离世界一致性视频生成

Lv, Zelong, Xu, Sicheng, Xiang, Jianfeng, Wang, Ruicheng, Dong, Yue, Deng, Yu, Sun, Guangzhong, Yang, Jiaolong

Abstract

We present OctWorld, a video diffusion framework with persistent 3D memory for generating explorable, world-consistent, and high-fidelity visual scenes. Given a single image, OctWorld performs stable autoregressive world generation along user-specified camera trajectories. We focus on long-range generation, characterized by extended camera paths and wide viewpoint coverage, where preserving spatial consistency is particularly challenging when previously generated regions are revisited. To address this problem, we introduce OctMap, an extensible and spatially adaptive 3D memory that progressively fuses generated visual observations and their corresponding depth maps into a global representation. OctMap employs TSDF fusion within a dynamic sparse octree whose spatial resolution adapts to image evidence. This design preserves geometric and appearance details across diverse scene scales while maintaining low memory overhead. Experiments demonstrate that OctWorld generates long-range, spatially consistent videos and outperforms prior methods on both existing benchmarks and challenging long-range generation settings. OctMap also provides clear advantages over point-based caches and fixed-resolution TSDF volumes. Project page: https://maxtirerror.github.io/octworldpage/

Chinese Translation

我们提出了OctWorld，一个具有持久3D记忆的视频扩散框架，用于生成可探索的、世界一致的高保真视觉场景。给定一张单一图像，OctWorld沿用户指定的相机轨迹执行稳定的自回归世界生成。我们专注于长距离生成，其特点是扩展的相机路径和广泛的视角覆盖，在重新访问先前生成区域时保持空间一致性尤为具有挑战性。为了解决这个问题，我们引入了OctMap，一个可扩展且空间自适应的3D记忆，逐步将生成的视觉观测及其对应的深度图融合成一个全局表示。OctMap在一个动态稀疏八叉树中采用TSDF融合，其空间分辨率根据图像证据自适应调整。该设计在保持低内存开销的同时，保留了不同场景尺度下的几何和外观细节。实验表明，OctWorld生成了长距离的空间一致性视频，并在现有基准和具有挑战性的长距离生成设置中超越了先前的方法。OctMap相较于基于点的缓存和固定分辨率的TSDF体积也提供了明显的优势。项目页面：https://maxtirerror.github.io/octworldpage/

View on arXiv Download PDF AI Translation

cs.CV / 77 / 2609.03931

Sparse auto-regressive modeling for scene generation from multi-view images

基于稀疏自回归建模的多视角图像场景生成

Lucas, Thomas, Pietrantoni, Maxime, Weinzaepfel, Philippe, Cho, Wonjune, Duisterhof, Bardienus Pieter, Leroy, Vincent, Revaud, Jerome

Abstract

Generating complete 3D scenes from sparse, unconstrained views is a fundamental challenge in 3D vision which requires reasoning beyond observed content while remaining computationally tractable. Existing feed-forward reconstruction methods are inherently limited to content visible in the input images, while 3D generative modeling is hindered by the high computational cost of dense volumetric representations and the scarcity of large-scale 3D supervision. We introduce SPAR3S, a sparse voxel-aligned 3D latent generative model for conditional scene completion without requiring ground-truth 3D data for supervision. Our key insight is to formulate 3D scene generation in a structured, compact, voxel-aligned 3D latent space where only occupied voxels are represented. We learn this sparse latent space directly from multi-view images using photometric supervision via differentiable 3D Gaussian Splatting. Given a partial set of observed voxels encoded from sparse input views, scene completion reduces to predicting the missing latent tokens and their spatial support within the voxel grid. To this end, we train a masked autoregressive transformer that jointly models voxel occupancy and latent token values, enabling efficient and spatially consistent generation of unseen regions. We demonstrate the effectiveness of our method on synthetic indoor scenes, achieving higher novel-view quality than prior work. We further validate its generalization on RealEstate10k, highlighting its applicability to real-world data.

Chinese Translation

从稀疏且不受限制的视角生成完整的3D场景是3D视觉中的一个基本挑战，这需要超越观察到的内容进行推理，同时保持计算上的可行性。现有的前馈重建方法本质上仅限于输入图像中可见的内容，而3D生成建模则受到密集体积表示的高计算成本和大规模3D监督稀缺的限制。我们提出了SPAR3S，一种稀疏体素对齐的3D潜在生成模型，用于条件场景补全，无需真实3D数据进行监督。我们的关键见解是将3D场景生成构建在一个结构化、紧凑的体素对齐3D潜在空间中，仅表示被占用的体素。我们通过可微分的3D高斯点云直接从多视角图像中学习这一稀疏潜在空间。给定从稀疏输入视角编码的部分观察体素集，场景补全简化为预测缺失的潜在标记及其在体素网格中的空间支持。为此，我们训练了一个掩蔽自回归变换器，该变换器联合建模体素占用和潜在标记值，从而实现高效且空间一致的未见区域生成。我们在合成室内场景上展示了我们方法的有效性，取得了比之前工作更高的新视角质量。我们进一步验证了其在RealEstate10k上的泛化能力，突显了其在真实世界数据中的适用性。

View on arXiv Download PDF AI Translation

cs.CV / 78 / 2609.03952

WorldReward: Reward Modeling for Camera-Conditioned World Models

WorldReward：用于相机条件下世界模型的奖励建模

Wang, Yibin, Wang, Zehan, Tang, Junshu, Li, Zhimin, Zhou, Yujie, Bu, Jiazi, Ling, Pengyang, Han, Feng, Zhang, Zhixiong, Xing, Long, Ding, Shengyuan, Li, Ziang, Jin, Cheng, Zang, Yuhang, Wang, Jiaqi, Pang, Tianyu

Abstract

Camera-conditioned world models generate interactive videos in which commanded actions should induce the expected scene changes while appearance, geometry, and temporal dynamics remain coherent. Existing rewards assess these requirements separately: geometry-based rewards estimate trajectory execution but cannot judge the visual quality of the executed motion, whereas image-based rewards measure frame quality without capturing action execution or temporal dynamics. We posit that a vision-language model (VLM) offers a shared reasoning space for relating actions to their visual outcomes. However, judging a complete long video against its full action sequence creates a lengthy, noisy context in which short-lived local action evidence can be missed or diluted. We present WorldReward, a VLM-based pairwise preference reward model that unifies action-consistency and visual-quality evaluation for camera-conditioned world models. WorldReward decomposes paired videos into action-aligned chunks, organizes each chunk into structured visual evidence, and aggregates chunk-level decisions by voting into separate video-level action and visual-quality preferences. To train it, we construct a large-scale reasoning-augmented preference dataset using structured judgments generated by a frontier VLM and refined through tool-based agent auditing and targeted human review. We further introduce WorldReward-Bench, a human-annotated benchmark measuring reward-model agreement with human preferences across action consistency, appearance quality, and motion quality. WorldReward achieves the highest agreement on all three dimensions, exceeding GPT-5.5 by 3.42, 1.45, and 3.56 percentage points, respectively. When used for RL post-training of HY-WorldPlay 1.5, it consistently improves both action execution and visual quality across short- to long-term horizons.

Chinese Translation

相机条件下的世界模型生成互动视频，其中指令动作应引发预期的场景变化，同时外观、几何形状和时间动态保持一致。现有的奖励机制分别评估这些要求：基于几何的奖励估计轨迹执行，但无法判断执行动作的视觉质量；而基于图像的奖励则测量帧质量，却未能捕捉动作执行或时间动态。我们认为视觉-语言模型（VLM）提供了一个共享的推理空间，用于将动作与其视觉结果关联。然而，将完整的长视频与其完整的动作序列进行评估会产生冗长且嘈杂的上下文，在这种情况下，短暂的局部动作证据可能会被遗漏或稀释。我们提出了WorldReward，一种基于VLM的成对偏好奖励模型，统一了相机条件下世界模型的动作一致性和视觉质量评估。WorldReward将成对视频分解为与动作对齐的块，将每个块组织为结构化的视觉证据，并通过投票将块级决策汇总为单独的视频级动作和视觉质量偏好。为了训练该模型，我们构建了一个大规模的增强推理偏好数据集，使用由前沿VLM生成的结构化判断，并通过基于工具的代理审计和针对性的人类评审进行优化。我们进一步引入了WorldReward-Bench，这是一个人类注释的基准，测量奖励模型与人类偏好在动作一致性、外观质量和运动质量方面的一致性。WorldReward在这三个维度上都达到了最高的一致性，分别超过GPT-5.5 3.42、1.45和3.56个百分点。在用于HY-WorldPlay 1.5的强化学习后训练时，它在短期到长期的范围内持续改善了动作执行和视觉质量。

View on arXiv Download PDF AI Translation

cs.CV / 79 / 2609.03956

RARF: Region-Aware Rectified Flows for 3D Brain MRI Inpainting

RARF：用于三维脑部MRI修复的区域感知校正流

Guija-Valiente, Tomas, Rodriguez-Gonzalez, Blanca, Malpica, Norberto, Torrado-Carvajal, Angel

Abstract

Medical image inpainting has the potential to improve automated brain MRI analysis by reconstructing healthy tissue within pathological regions. We introduce RARF, a task-agnostic region-aware rectified flow framework for masked data generation. We instantiate the framework for 3D brain MRI inpainting as our submission to the BraTS Inpainting Challenge 2026. RARF restricts the stochastic interpolation process to the inpainting region, while the observed voxels remain fixed and provide patient-specific anatomical context. A three-dimensional neural network receives the partially voided image, with Gaussian noise filling the missing region, together with the inpainting mask and the corresponding timestep. The model is trained using masked flow-matching and reconstruction-consistency objectives, combined with mask-aware preprocessing and data augmentation. During inference, the learned velocity field transports the initial noise toward a plausible reconstruction of the missing tissue, which is then combined with the unchanged observed anatomy. Experiments under the BraTS evaluation protocol show that the proposed approach produces competitive reconstructions while maintaining anatomical consistency. Source code is available at: https://github.com/TomasGuija/rarf.

Chinese Translation

医学图像修复有潜力通过重建病理区域内的健康组织来改善自动化脑部MRI分析。我们提出了RARF，一个任务无关的区域感知校正流框架，用于掩膜数据生成。我们将该框架实例化为三维脑部MRI修复，作为我们对2026年BraTS修复挑战的提交。RARF将随机插值过程限制在修复区域内，而观察到的体素保持固定，提供患者特定的解剖背景。一个三维神经网络接收部分缺失的图像，缺失区域填充高斯噪声，连同修复掩膜和相应的时间步。该模型使用掩膜流匹配和重建一致性目标进行训练，同时结合掩膜感知的预处理和数据增强。在推理过程中，学习到的速度场将初始噪声传输到缺失组织的合理重建中，然后与未改变的观察解剖结构结合。根据BraTS评估协议的实验表明，所提出的方法在保持解剖一致性的同时产生了具有竞争力的重建效果。源代码可在：https://github.com/TomasGuija/rarf获取。

View on arXiv Download PDF AI Translation

cs.CV / 80 / 2609.03981

Sharpening the Ensemble: An SSIM-Aligned Residual Refiner for Brain-MRI Inpainting Post-Processing

提升集成模型的性能：一种与结构相似性指数（SSIM）对齐的残差精细化器用于脑部MRI修复后处理

Kömürcü, Kubilay Kağan, Öksüz, İlkay

Abstract

Brain-MRI inpainting replaces a masked region of a scan with synthesized, anatomically plausible healthy tissue, so that analysis tools built for healthy brains can be applied to images they would otherwise reject. On the BraTS local-synthesis benchmark, which ranks submissions on the structural similarity index (SSIM), the peak signal-to-noise ratio, and the mean squared error (MSE) jointly, the strongest recent models are accurate, but several report blurry synthesized regions and attribute this to the mean-seeking behavior of the $\ell_1$ and MSE terms in their training losses. We address this in post-processing, forming a deep ensemble of the two co-first-place 2025 models and training a lightweight residual refiner on the ensemble's own outputs under an $\ell_1$ loss augmented with a structural-similarity term whose weight $\lambda$ we vary. At a moderate $\lambda$ the refiner improves SSIM over the ensemble, from $0.8767$ to $0.8780$ on a held-out reproduction of the official scorer and from $0.8555$ to $0.8572$ on the official validation leaderboard, with essentially no change in MSE. The gain is small but consistent, improving $62.6\%$ of the held-out cases with a signed-rank $p=2.2\times10^{-7}$, whereas over-weighting the structural term reverses it. Two ablations bound the effect. Adding any third model to the two-model ensemble degrades it, and classical unsharp masking fails to improve SSIM at any strength (best $0.8765$ against $0.8767$), so the gain reflects learned rather than indiscriminate sharpening. The result is a cheap, reproducible post-processing stage that improves an already strong ensemble without any large-scale retraining.

Chinese Translation

脑部MRI修复通过合成的、解剖上合理的健康组织替代扫描中的遮挡区域，以便能够将为健康大脑设计的分析工具应用于原本会被拒绝的图像。在BraTS本地合成基准测试中，该测试根据结构相似性指数（SSIM）、峰值信噪比和均方误差（MSE）共同对提交结果进行排名，最新的强大模型虽然准确，但有几个模型报告合成区域模糊，并将此归因于其训练损失中的$ ext{l}_1$和MSE项的均值寻求行为。我们在后处理阶段解决了这个问题，形成了两个并列第一的2025模型的深度集成，并在集成模型的输出上训练了一个轻量级的残差精细化器，使用增强了结构相似性项的$ ext{l}_1$损失，其中权重$ ext{λ}$是可变的。在适中的$ ext{λ}$下，精细化器在集成模型上改善了SSIM，从$0.8767$提高到$0.8780$（在官方评分的保留重现上），并从$0.8555$提高到$0.8572$（在官方验证排行榜上），而MSE几乎没有变化。尽管增益较小，但一致性良好，改善了$62.6 ext{ ext{%}}$的保留案例，签名秩$p=2.2 imes10^{-7}$，而过度加权结构项则会逆转这一结果。两个消融实验界定了这一效果的边界。向两模型集成中添加任何第三个模型都会降低其性能，而经典的非锐化遮罩在任何强度下都未能改善SSIM（最佳为$0.8765$，对比$0.8767$），因此增益反映的是学习到的而非无差别的锐化。该结果是一个廉价、可重复的后处理阶段，能够在不进行大规模再训练的情况下改善已经强大的集成模型。

View on arXiv Download PDF AI Translation

cs.CV / 81 / 2609.03985

IchthyoNoma: Nomenclature and Context Sensitivity of Zero-Shot Biological Vision--Language Models for Bangladeshi Freshwater Fish Recognition

IchthyoNoma：零样本生物视觉-语言模型在孟加拉国淡水鱼识别中的命名法与上下文敏感性

Nazim-E-Alam, Rahman, Tarek, Morol, Md Kishor

Abstract

Zero-shot vision-language models (VLMs) are increasingly used as training-free species recognizers, but reported accuracy can reflect more than visual species knowledge. We audit CLIP, BioCLIP, BioCLIP2, and a multilingual Jina CLIP v2 control on seven freshwater-fish categories from two Bangladeshi sources (10,321 images). BioCLIP2 reaches 72.36% on BFF-15 with English common names and 68.91% on SylFishBD with scientific names, versus 25.15% and 14.40% for generic CLIP. BioCLIP2 Bengali prompts are near chance in balanced accuracy (14.22-14.29%); Jina partially recovers Bengali discrimination to 21.89% and 16.36%, but bare Bengali names return to 14.29% on both sources. Paired SylFishBD interventions show no significant weak-blur effect, modest losses from stronger blur/gray masking, a larger white-mask artifact, and strong species dependence. Zero-shot biological VLM scores therefore jointly reflect biological specialization, multilingual alignment, nomenclature, prompt formulation, and context.

Chinese Translation

零样本视觉-语言模型（VLMs）越来越多地被用作无训练的物种识别工具，但报告的准确性可能反映的不仅仅是视觉物种知识。我们对CLIP、BioCLIP、BioCLIP2以及多语言Jina CLIP v2控制模型在来自两个孟加拉国来源的七类淡水鱼（10,321张图像）上的表现进行了审计。BioCLIP2在BFF-15数据集上使用英文通用名称达到了72.36%的准确率，而在SylFishBD数据集上使用科学名称则达到了68.91%，相比之下，通用CLIP的准确率仅为25.15%和14.40%。BioCLIP2使用孟加拉语提示的平衡准确率接近随机（14.22-14.29%）；Jina在一定程度上恢复了孟加拉语的区分能力，达到了21.89%和16.36%，但仅使用孟加拉语名称在两个来源上的准确率又回落至14.29%。配对的SylFishBD干预显示没有显著的弱模糊效应，强模糊/灰色遮罩导致适度损失，较大的白色遮罩伪影，以及强烈的物种依赖性。因此，零样本生物VLM的得分共同反映了生物专业化、多语言对齐、命名法、提示构造和上下文。

View on arXiv Download PDF AI Translation

cs.CV / 82 / 2609.03995

Catalogue Photography as a Cold Start: Toward Deployable Carbide Burr Recognition

目录摄影作为冷启动：可部署的硬质合金铣刀识别

Madavath, Abilash Philip, Aubeeluck, Chandra Yuvesh, Raju, Augustin, Pyschny, Nicolas, Hackelöer, Felix, Zwanzig, Florian

Abstract

Verifying that manufactured batches of milling tools or carbide rotary burrs conform to production order sheets remains a largely manual and error-prone quality assurance task. Automating this process with computer vision faces a critical cold-start constraint since no labelled imagery is available, leaving manufacturer catalogue photography as the sole source of supervision. We investigate how far catalogue supervision can support an industrial recognition pipeline under domain shift, explicitly measuring the gap between catalogue separability and performance on held-out field photographs. Our findings reveal three key insights. First, off-the-shelf frozen feature extractors do not reliably separate the two task attributes, head shape and tooth profile, motivating targeted representation learning. Second, metric learning produces near-perfect unsupervised cluster discovery on catalogue images (adjusted Rand index 0.94--0.97), but less than half of this gain transfers to field photographs. Third, the largest transfer gains do not come from model scale or representation complexity, but from simple changes that reduce domain sensitivity: converting images to grayscale (+0.22) and constraining retrieval using the known order sheet via Hungarian assignment (+0.11). We therefore treat catalogue photography as a useful cold start rather than a deployment-ready training domain, and provide empirical baselines and an evaluation protocol for catalogue-to-field transfer in precision tool manufacturing.

Chinese Translation

验证制造的铣刀或硬质合金旋转铣刀批次是否符合生产订单仍然是一个主要依赖人工且容易出错的质量保证任务。利用计算机视觉自动化这一过程面临着一个关键的冷启动限制，因为没有可用的标注图像，制造商的目录摄影成为唯一的监督来源。我们研究了目录监督在领域转移下能够支持工业识别管道的程度，明确测量了目录可分离性与保留的现场照片性能之间的差距。我们的研究结果揭示了三个关键见解。首先，现成的冻结特征提取器无法可靠地区分两个任务属性：头部形状和齿轮轮廓，这促使我们进行有针对性的表征学习。其次，度量学习在目录图像上产生了近乎完美的无监督聚类发现（调整后的兰德指数为0.94--0.97），但不到一半的增益转移到现场照片。第三，最大的转移增益并不来自模型规模或表征复杂性，而是来自简单的变化，这些变化减少了领域敏感性：将图像转换为灰度图像（+0.22）和通过匈牙利分配法使用已知订单表约束检索（+0.11）。因此，我们将目录摄影视为有用的冷启动，而不是一个准备好部署的训练领域，并为精密工具制造中的目录到现场转移提供了经验基准和评估协议。

View on arXiv Download PDF AI Translation

cs.CV / 83 / 2609.04009

The Blind Spot in 2D Infants' Pose Estimation:Robust Learning from Noisy Annotations

二维婴儿姿态估计中的盲点：从噪声标注中进行稳健学习

Cardinale, Emanuele, Proietti, Marco, Cacciatore, Alessandro, Spadea, Maria Francesca, Migliorelli, Lucia, Moccia, Sara

Abstract

Noisy annotations pose a significant challenge for supervised deep learning, as neural networks rely on large-scale, high-quality labeled data whose corruption can severely impair model performance. Although robustness to label noise has been extensively studied for classification tasks, it remains relatively underexplored in Pose Estimation (PE). This limitation becomes critical in clinical contexts, including neonatology, where PE of preterm infants is used to support the assessment of spontaneous motility, a key indicator of neurodevelopmental trajectories. In such settings, infants' images labeling is further hindered by visual challenges (e.g., keypoint self-occlusions, caregiver interference), making the annotation process inherently susceptible to errors. To tackle noisy annotations in PE, we introduce REliable keypoint selection via Memory of traINing Dynamics (REMIND), a clustering-based keypoint-selection strategy that exploits keypoint-wise training dynamics to identify noisy labels without assuming any prior knowledge of the noise distribution, thus enabling noise-free model training. When evaluated on the proprietary NeoPose dataset, comprising 46 videos of 46 preterm infants recorded in real clinical settings, REMIND correctly identifies noisy annotations across multiple corruption scenarios, achieving up to 93\% Area Under the Curve (AUC) with three different PE architectures used in the relevant literature. To our knowledge, this is the first study to explicitly address label noise in preterm infants' PE, paving the way for the design of trustworthy learning-based algorithms for infants'monitoring support when data quality cannot be guaranteed.

Chinese Translation

噪声标注对监督深度学习构成了重大挑战，因为神经网络依赖于大规模、高质量的标注数据，而这些数据的损坏可能严重影响模型性能。尽管对分类任务中的标签噪声的鲁棒性进行了广泛研究，但在姿态估计（Pose Estimation, PE）中仍然相对缺乏探索。这一局限性在临床环境中变得尤为关键，包括新生儿医学，其中对早产婴儿的姿态估计用于支持自发运动能力的评估，这是神经发育轨迹的一个关键指标。在这种情况下，婴儿图像的标注受到视觉挑战（例如，关键点自遮挡、看护者干扰）的进一步阻碍，使得标注过程本质上容易出现错误。为了解决PE中的噪声标注问题，我们提出了基于训练动态记忆的可靠关键点选择（REliable keypoint selection via Memory of traINing Dynamics, REMIND），这是一种基于聚类的关键点选择策略，利用关键点的训练动态来识别噪声标签，而无需假设噪声分布的任何先验知识，从而实现无噪声的模型训练。在对专有的NeoPose数据集进行评估时，该数据集包含46段在真实临床环境中录制的46名早产婴儿的视频，REMIND能够在多种损坏场景下正确识别噪声标注，达到93%的曲线下面积（Area Under the Curve, AUC），并使用相关文献中的三种不同PE架构。据我们所知，这是首个明确针对早产婴儿PE中的标签噪声进行研究的工作，为在数据质量无法保证的情况下设计可信赖的基于学习的婴儿监测支持算法铺平了道路。

View on arXiv Download PDF AI Translation

cs.CV / 84 / 2609.04026

Stable and Scalable Bundle Adjustment of Holistic 3D Structures

稳定且可扩展的整体三维结构束调整

Liu, Shaohui, Pautrat, Rémi, Barath, Daniel, Hartley, Richard, Larsson, Viktor, Pollefeys, Marc

Abstract

Bundle Adjustment (BA) is a cornerstone of 3D computer vision and has benefited from decades of advances in sparse optimization and numerical methods. It was originally developed for jointly optimizing camera intrinsics, poses and sparse 3D points. While extensions incorporate lines and other primitives, integrating richer geometric structures such as parallelism, coplanarity, or wireframes often introduces significantly increased computational cost and reduced numerical stability. In this paper, we propose a unified framework that extends bundle adjustment to jointly optimize geometric features and higher-order relations. We first introduce a taxonomy that distinguishes scalable geometric features with direct 2D measurements (e.g., points and lines), from groups encoding higher-order relations (e.g., coplanarity, parallelism, etc.), where we show that groups can be modeled as camera-like entities within the bundle adjustment framework. Building on this formulation, we propose that both group constraints and cross-feature relations (i.e., point-line associations) can be expressed through 2D reprojection measurements. By formulating group-induced and cross-feature reprojection errors, we preserve the sparsity structure of classical point-based BA under Schur elimination, while avoiding direct 3D regularization that degrades the conditioning and stability. Experiments on both real-world and synthetic datasets demonstrate runtime performance comparable to classical point-only bundle adjustment, while producing significantly richer 3D structures and improved geometric accuracy.

Chinese Translation

束调整（Bundle Adjustment, BA）是三维计算机视觉的基石，受益于数十年来稀疏优化和数值方法的进步。它最初是为联合优化相机内参、姿态和稀疏三维点而开发的。尽管扩展中包含了直线和其他原始元素，但将更丰富的几何结构（如平行性、共面性或线框）整合进来往往会显著增加计算成本并降低数值稳定性。本文提出了一个统一框架，将束调整扩展到联合优化几何特征和高阶关系。我们首先引入一个分类法，将具有直接二维测量的可扩展几何特征（如点和线）与编码高阶关系的组（如共面性、平行性等）区分开来，并展示了这些组可以在束调整框架内被建模为类似相机的实体。在此基础上，我们提出组约束和交叉特征关系（即点-线关联）可以通过二维重投影测量来表达。通过构建由组引起的和交叉特征的重投影误差，我们在施尔消元下保留了经典基于点的束调整的稀疏结构，同时避免了直接的三维正则化，这种正则化会降低条件性和稳定性。在真实世界和合成数据集上的实验表明，我们的方法在运行时间上与经典的仅基于点的束调整相当，同时生成了显著更丰富的三维结构和改善的几何精度。

View on arXiv Download PDF AI Translation

cs.CV / 85 / 2609.04031

DSAQuant: Denoising-Stage-Aligned Quantization-Aware Training for Video Generation

DSAQuant：用于视频生成的去噪阶段对齐量化感知训练

Li, Shuaiting, Gao, Zelin, Shen, Haibin, Shen, Yujun, Qin, Haotong, Xu, Yinghao

Abstract

Video diffusion models (VDMs) have achieved impressive progress in text-to-video generation, but their high memory and computational costs hinder practical deployment. Quantization-aware training (QAT) is an effective solution for compressing and accelerating advanced generative models without runtime overhead at inference. However, existing QAT methods suffer from a distinctive challenge in VDMs: while they often preserve prompt semantics, global layout, and coarse motion, the quantized model severely degrades visual details, texture fidelity, and sharpness. In this paper, we trace this degradation to the timestep-agnostic design of conventional quantization pipelines, which overlooks the stage-wise functionality of video denoising. In VDMs, early denoising steps mainly establish global structure and motion, whereas middle and late steps refine local appearance and high-frequency details. Based on this insight, we propose DSAQuant, a Denoising-Stage-Aligned Quantization-aware training framework for VDMs. During training, Denoising-Stage Oriented Supervision preserves teacher distillation in early steps for stable structure planning, while shifting later steps toward target-driven optimization to enhance detail reconstruction. During inference, Denoising-Stage Gated Guidance disables CFG in the final denoising steps to prevent it from amplifying quantization-induced errors into high-frequency artifacts. Extensive experiments on the Wan and CogVideoX families under W4A4 and W3A3 settings show that DSAQuant consistently outperforms the SOTA QAT baseline, improving the VBench average score by up to 6.60 under aggressive W3A3 quantization while preserving strong text-video alignment. These results demonstrate that effective VDM quantization requires not only reducing quantization error, but also aligning quantization training and inference with the stage-wise nature of video diffusion.

Chinese Translation

视频扩散模型（VDMs）在文本到视频生成方面取得了显著进展，但其高内存和计算成本阻碍了实际部署。量化感知训练（QAT）是压缩和加速先进生成模型的有效解决方案，而不会在推理时增加运行时开销。然而，现有的QAT方法在VDMs中面临一个独特的挑战：虽然它们通常保留提示语义、全局布局和粗略运动，但量化模型严重降低了视觉细节、纹理保真度和清晰度。本文追溯这种退化现象，归因于传统量化管道的时间步无关设计，这忽视了视频去噪的阶段性功能。在VDMs中，早期去噪步骤主要建立全局结构和运动，而中后期步骤则细化局部外观和高频细节。基于这一见解，我们提出了DSAQuant，一个针对VDMs的去噪阶段对齐量化感知训练框架。在训练过程中，去噪阶段导向监督在早期步骤中保留教师蒸馏，以实现稳定的结构规划，同时将后期步骤转向目标驱动的优化，以增强细节重建。在推理过程中，去噪阶段门控引导在最后的去噪步骤中禁用CFG，以防止其将量化引起的错误放大为高频伪影。在W4A4和W3A3设置下对Wan和CogVideoX系列进行的广泛实验表明，DSAQuant始终优于SOTA QAT基线，在激进的W3A3量化下将VBench平均分提高了最多6.60，同时保持了强大的文本-视频对齐。这些结果表明，有效的VDM量化不仅需要减少量化误差，还需要将量化训练和推理与视频扩散的阶段性特征对齐。

View on arXiv Download PDF AI Translation

cs.CV / 86 / 2609.04034

Editable Visual Design

可编辑的视觉设计

Ye, Junyan, Liu, Wei, Jiang, Dongzhi, Wen, Zichen, Li, HaoDong, Lv, Zhutao, Lin, Jiaxin, Yu, Jinhua, He, Jun, Huang, Zilong, Chen, Rui, Li, Weijia

Abstract

While diffusion base models such as GPT-Image-2 and Nano-Banana exhibit remarkable visual expressiveness, their end-to-end generation inherently yields flattened bitmaps with error-prone text, precluding layer-wise post-editing. Conversely, code-based visual generation via Coding Agents provides precise layout control and decoupled layers, yet remains constrained by a lack of global aesthetic intuition and the difficulty of coding complex visual assets. To address this, we propose Editable Visual Design, a new paradigm driven by a Coding Agent. We designate the VLM as the ``creative brain'' for requirement comprehension, task planning, and aesthetic judgment, while utilizing the image generation model as an on-demand ``visual world simulator'' to synthesize standalone visual assets. Operating under an ``imagine first, then act'' closed-loop workflow, the agent generates isolated assets, writes native HTML/CSS, and iteratively refines the design against visual rendering feedback. Furthermore, Agent Design Replay faithfully reproduces the creative and reasoning trajectory akin to that of professional human designers. Ultimately, the system delivers editable artifacts with decoupled layers and real text, enabling users to perform intuitive mouse dragging and layout adjustments on a graphical user interface. Validations on posters, infographics, and other scenarios show that this paradigm successfully achieves both refined aesthetics and production-grade editability.

Chinese Translation

尽管基于扩散模型的系统如 GPT-Image-2 和 Nano-Banana 展现了显著的视觉表现力，但其端到端生成本质上产生了扁平化的位图，并伴随易出错的文本，阻碍了逐层后期编辑。相反，通过编码代理进行的基于代码的视觉生成提供了精确的布局控制和解耦的层，但仍然受到缺乏全球美学直觉和编码复杂视觉资产的困难的限制。为了解决这一问题，我们提出了可编辑的视觉设计（Editable Visual Design），这是一个由编码代理驱动的新范式。我们将视觉语言模型（VLM）指定为需求理解、任务规划和美学判断的“创意大脑”，同时利用图像生成模型作为按需的“视觉世界模拟器”来合成独立的视觉资产。在“先想象，再行动”的闭环工作流程下，代理生成孤立的资产，编写原生 HTML/CSS，并根据视觉渲染反馈迭代优化设计。此外，代理设计重放（Agent Design Replay）忠实再现了类似于专业人类设计师的创意和推理轨迹。最终，该系统提供了具有解耦层和真实文本的可编辑工件，使用户能够在图形用户界面上进行直观的鼠标拖动和布局调整。在海报、信息图表和其他场景中的验证表明，这一范式成功实现了精致的美学和生产级的可编辑性。

View on arXiv Download PDF AI Translation

cs.CV / 87 / 2609.04070

Continuous Actions from Discrete Minds: Latent-Aligned Planning for End-to-End Autonomous Driving

来自离散思维的连续动作：端到端自主驾驶的潜在对齐规划

Yao, Ruoyu, Xie, Yusen, Liu, Qingzhao, Liu, Pei, Yang, Zewei, Zhu, Yipeng, Wang, Xiaolong, Ma, Jun

Abstract

Bridging the gap between the discrete reasoning of Vision-Language Models and the continuous, physics-constrained nature of autonomous driving remains a significant challenge. In this work, we introduce LaPla, a unified Vision-Language-Action (VLA) framework featuring latent-aligned planning to seamlessly ground semantic understanding in precise motion execution. We first design an action tokenizer based on a residual vector-quantized variational autoencoder (VQ-VAE), capturing vehicle kinematics and encoding trajectory features into a structured latent space. Rather than discrete codebook lookups that inevitably introduce quantization errors, LaPla repurposes this representation as a physical prior to bridge the modality gap between high-dimensional semantics and the raw action space. Specifically, given multimodal inputs integrating multi-view images, historical actions, and textual instructions, LaPla incorporates concurrent action queries to causally attend to the multimodal context in a single forward pass, projecting hidden states directly into the pretrained VQ-VAE latent space. The frozen decoder then translates these continuous latents into actions, effectively eliminating quantization errors and ensuring physically plausible trajectories while bypassing time-consuming autoregressive generation. Extensive experiments on the nuScenes benchmark demonstrate that LaPla achieves competitive open-loop performance, reducing long-horizon L2 error by 15.52% compared to state-of-the-art VLA methods. Closed-loop evaluations on the NVIDIA AlpaSim simulator further confirm its superior capability in ensuring smooth driving progress, improving the success rate by 33.34 percentage points with significantly reduced inference latency.

Chinese Translation

弥合视觉-语言模型的离散推理与自主驾驶的连续、物理约束特性之间的差距仍然是一个重大挑战。在本研究中，我们提出了LaPla，一个统一的视觉-语言-动作（VLA）框架，采用潜在对齐规划，将语义理解无缝地与精确的运动执行结合起来。我们首先设计了一种基于残差向量量化变分自编码器（VQ-VAE）的动作标记器，捕捉车辆运动学并将轨迹特征编码到结构化的潜在空间中。与不可避免引入量化误差的离散代码本查找不同，LaPla将这种表示重新利用作为物理先验，以弥合高维语义与原始动作空间之间的模态差距。具体而言，LaPla在单次前向传递中，结合多视角图像、历史动作和文本指令的多模态输入，融入并发动作查询，以因果方式关注多模态上下文，直接将隐藏状态投影到预训练的VQ-VAE潜在空间中。冻结的解码器随后将这些连续潜在变量转换为动作，有效消除了量化误差，并确保物理上合理的轨迹，同时避免了耗时的自回归生成。在nuScenes基准上的广泛实验表明，LaPla在开放循环性能上具有竞争力，与最先进的VLA方法相比，长时间范围L2误差降低了15.52%。在NVIDIA AlpaSim模拟器上的闭环评估进一步确认了其在确保平稳驾驶进程方面的优越能力，成功率提高了33.34个百分点，同时显著减少了推理延迟。

View on arXiv Download PDF AI Translation

cs.CV / 88 / 2609.04071

TAP-Path: Task-Adaptive Structural and Token Pruning for Efficient and Trustworthy Pathology Foundation Models

TAP-Path：用于高效和可信病理基础模型的任务自适应结构和标记剪枝

Hasan, Mehedi, Yeafi, Ashfak, Islam, Md Khairul

Abstract

Pathology foundation models improve transferable representation learning for histopathology, but recent gains often rely on encoders with hundreds of millions of parameters and high inference cost. We propose TAP-Path, a task-adaptive compression framework that directly restructures a pretrained Virchow2 encoder rather than distilling it into a separate student. TAP-Path combines validation-driven transformer-block selection, physical removal of redundant blocks, input-adaptive patch-token pruning, multi-depth feature recovery, and a lightweight gated task head. The final model retains 24 of 32 transformer blocks and 70% of patch tokens after pruning, reducing encoder parameters by 24.96% (631.24M to 473.70M) and analytical encoder compute by 35.20% (340.13G to 220.40G FLOPs). Across three task-head optimization seeds, TAP-Path achieved $87.98 \pm 0.067%$ test accuracy, $81.26 \pm 0.49%$ balanced accuracy, and $82.38 \pm 0.48%$ macro-F1 on a 32-class histopathology benchmark, compared with 86.89% for full Virchow2 and 87.67% for UNI2-h. TAP-Path achieved a Brier score of $0.1800 \pm 0.0005$ and failure-detection AUROC of $0.9047 \pm 0.0060$. A validation-only rare-aware objective improved rare-class balanced accuracy in a secondary operating analysis. Frozen external evaluation on 433 CPTAC samples yielded $91.22 \pm 0.83%$ accuracy and $91.10 \pm 0.81%$ balanced accuracy. These results show that task-adaptive structural and token sparsification can improve the accuracy-efficiency trade-off of large pathology foundation models while preserving reliability under internal and external evaluation.

Chinese Translation

病理基础模型改善了组织病理学的可转移表示学习，但最近的进展往往依赖于具有数亿参数和高推理成本的编码器。我们提出了TAP-Path，一种任务自适应压缩框架，它直接重构预训练的Virchow2编码器，而不是将其蒸馏为一个单独的学生模型。TAP-Path结合了基于验证的变换器块选择、冗余块的物理移除、输入自适应的补丁标记剪枝、多深度特征恢复以及轻量级的门控任务头。最终模型在剪枝后保留了32个变换器块中的24个和70%的补丁标记，编码器参数减少了24.96%（从631.24M降至473.70M），分析编码器计算减少了35.20%（从340.13G降至220.40G FLOPs）。在三个任务头优化种子上，TAP-Path在32类组织病理学基准测试中达到了$87.98 ext{±} 0.067 ext{%}$的测试准确率，$81.26 ext{±} 0.49 ext{%}$的平衡准确率，以及$82.38 ext{±} 0.48 ext{%}$的宏观F1分数，而完整的Virchow2为86.89%，UNI2-h为87.67%。TAP-Path的Brier分数为$0.1800 ext{±} 0.0005$，故障检测的AUROC为$0.9047 ext{±} 0.0060$。仅基于验证的稀有类感知目标在二次操作分析中提高了稀有类的平衡准确率。在433个CPTAC样本上的冻结外部评估结果显示，准确率为$91.22 ext{±} 0.83 ext{%}$，平衡准确率为$91.10 ext{±} 0.81 ext{%}$。这些结果表明，任务自适应的结构和标记稀疏化可以提高大型病理基础模型的准确性与效率之间的权衡，同时在内部和外部评估中保持可靠性。

View on arXiv Download PDF AI Translation

cs.CV / 89 / 2609.04083

CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker Distillation

CORE：通过重排序蒸馏提高 MLLM 嵌入中的组合推理

Song, Tingyu, Li, Mingxin, Zhang, Yanzhao, Long, Dingkun, Liu, Chu, Xie, Pengjun, Zhao, Yilun, Wu, Shu

Abstract

MLLM-based embedding models remain limited in compositional retrieval, often failing to distinguish scenes containing the same concepts but different attribute-object bindings. Yet the same backbone can resolve such distinctions when used as a cross-attentive reranker, motivating us to distill its compositional judgments into the embedding model. We propose CORE, which synthesizes candidate lists spanning five compositional matching levels and introduces a Rank-KL objective that trains the embedding model to reproduce the reranker's fine-grained ranking. We further introduce a graded evaluation protocol and compare contrastive learning, pairwise CoSENT, and listwise Rank-KL under the same data and tuning budget. Our comparison shows that both CoSENT and Rank-KL use the multi-level supervision more effectively than contrastive learning, with Rank-KL achieving the strongest overall performance. Across three compositional reasoning benchmarks (COLA, SUGARCREPE++, NEGBENCH), CORE-RERANKER-8B achieves an 82.7% total average, outperforming Jina-Reranker by 10.7 points, while CORE-EMBED-8B achieves the best total average (0.666) among all evaluated embedding models. The improvements transfer to the MCMR benchmark without sacrificing retrieval performance on COCO and Flickr30K.

Chinese Translation

基于 MLLM 的嵌入模型在组合检索方面仍然存在局限，常常无法区分包含相同概念但属性-对象绑定不同的场景。然而，当作为跨注意力重排序器使用时，相同的骨干网络能够解决这些区分，这激励我们将其组合判断蒸馏到嵌入模型中。我们提出了 CORE，它合成了跨越五个组合匹配级别的候选列表，并引入了 Rank-KL 目标，训练嵌入模型重现重排序器的细粒度排名。我们进一步引入了分级评估协议，并在相同的数据和调优预算下比较了对比学习、成对 CoSENT 和列表 Rank-KL。我们的比较显示，CoSENT 和 Rank-KL 在多级监督的使用上比对比学习更为有效，其中 Rank-KL 实现了最强的整体性能。在三个组合推理基准（COLA、SUGARCREPE++、NEGBENCH）上，CORE-RERANKER-8B 达到了 82.7% 的总平均分，比 Jina-Reranker 高出 10.7 分，而 CORE-EMBED-8B 在所有评估的嵌入模型中实现了最佳总平均分（0.666）。这些改进在不牺牲 COCO 和 Flickr30K 上的检索性能的情况下，成功转移到 MCMR 基准上。

View on arXiv Download PDF AI Translation

cs.CV / 90 / 2609.04088

Efficient Semantic Understanding from Digital Foveation

基于数字注视的高效语义理解

Caccavella, Caterina, Fra, Vittorio, Ziegler, Andreas, D'Angelo, Giulia, Sandamirskaya, Yulia

Abstract

Dense semantic segmentation allocates computational resources uniformly across the entire image, regardless of scene complexity or task relevance. Inspired by biological vision, we investigate whether semantic understanding can be achieved more efficiently through digital foveated perception. We introduce a lightweight active-vision pipeline that combines saliency-driven fixation selection, high-resolution foveal observations, low-resolution contextual information, semantic accumulation, and adaptive computation. Beyond conventional dense prediction metrics, we use object-level evaluation to measure semantic understanding under sparse observations. On ADE20K-Object, a single foveated observation achieves 95.9% of the baseline Top-1 accuracy and 96.9% of the baseline Top-3 accuracy while requiring only 4.7% of the computational cost. At the scene level, semantic accumulation recovers 90.6% of the baseline object recall while using 58.6% of the computation. These results suggest that substantial semantic understanding can emerge from sparse observations when computation is allocated selectively, highlighting active vision as an efficient alternative to uniform dense processing and motivating evaluation protocols beyond conventional pixel-wise segmentation metrics.

Chinese Translation

密集语义分割在整个图像上均匀分配计算资源，而不考虑场景复杂性或任务相关性。受生物视觉的启发，我们研究了是否可以通过数字注视感知更高效地实现语义理解。我们提出了一种轻量级的主动视觉管道，结合了基于显著性的注视选择、高分辨率的中心观察、低分辨率的上下文信息、语义累积和自适应计算。除了传统的密集预测指标外，我们还使用对象级评估来衡量稀疏观察下的语义理解。在ADE20K-Object数据集上，单个注视观察达到了基线Top-1准确率的95.9%和基线Top-3准确率的96.9%，同时仅需4.7%的计算成本。在场景级别，语义累积恢复了基线对象召回率的90.6%，而计算使用率为58.6%。这些结果表明，当计算被选择性分配时，稀疏观察可以产生显著的语义理解，突显了主动视觉作为均匀密集处理的高效替代方案，并激励了超越传统像素级分割指标的评估协议。

View on arXiv Download PDF AI Translation

cs.CV / 91 / 2609.04110

The Shape of Time: Video-Token Contrast for Temporal Understanding in VideoLMs

时间的形状：用于视频语言模型的时序对比视频标记

Shi, Yumeng, Long, Quanyu, Wu, Yin, Wang, Wenya

Abstract

Seeing frames in order does not mean representing time. Modern VideoLMs receive ordered video streams, yet their main supervision acts on generated text rather than video-token representations where event dynamics should first emerge. This mismatch allows models to learn temporal answers from shortcuts such as objects, scenes, and language priors, without requiring internal video representations to capture event progression. To address this, we propose VT-Contrast, a representation-level temporal counterfactual objective for VideoLMs. Its design asks where temporal supervision should act and what temporal differences it should expose. VT-Contrast supervises selected late-layer last-frame video tokens, where temporal information is expected to be integrated before language generation, and contrasts order-preserving views with same-video reordered counterfactuals graded by Kendall tau distance. It requires no architectural changes, is compatible with diverse VideoLM training tasks, and improves overall performance across temporal understanding benchmarks. Our code is available at https://github.com/ANDgate99/VT-Contrast.

Chinese Translation

按顺序查看帧并不意味着对时间的表征。现代视频语言模型（VideoLMs）接收有序的视频流，但它们的主要监督作用于生成的文本，而非视频标记表征，在这些表征中，事件动态应首先显现。这种不匹配使得模型能够通过物体、场景和语言先验等捷径学习时间性答案，而无需内部视频表征来捕捉事件进展。为了解决这个问题，我们提出了VT-Contrast，一种针对视频语言模型的表征级时间反事实目标。其设计旨在探讨时间监督应如何作用以及应揭示哪些时间差异。VT-Contrast对选定的后层最后帧视频标记进行监督，在这些标记中，预计在语言生成之前会整合时间信息，并将保持顺序的视图与同一视频的重新排序反事实进行对比，后者通过Kendall tau距离进行评分。它不需要架构更改，兼容多种视频语言模型训练任务，并在时间理解基准上提升整体性能。我们的代码可在 https://github.com/ANDgate99/VT-Contrast 获取。

View on arXiv Download PDF AI Translation

cs.CV / 92 / 2609.04120

BooM-VVT: Boosting Mask-Free Video Virtual Try-On with Image-Level Pseudo Data

BooM-VVT：通过图像级伪数据提升无掩膜视频虚拟试穿

Zhang, Wei, Li, Xin, Shi, Peishu, Gao, Jialin, Peng, Xuekang, Lian, Zhichao, Jin, Yeying

Abstract

Video virtual try-on (VVT) aims to generate realistic videos of a person wearing a target garment. Recent methods leverage a keyframe-driven video generation paradigm to improve in-the-wild performance, yet they still rely on masks to localize try-on regions, making them vulnerable to large motions and severe occlusions. Although mask-free image-based try-on methods have shown promising results by leveraging large-scale pseudo data, extending this paradigm to videos remains difficult, as constructing video-level pseudo data is prohibitively expensive. Furthermore, coarse keyframe sampling and the scarcity of multi-view try-on data limit existing keyframe-driven methods in maintaining garment consistency and handling diverse try-on tasks. To address these challenges, we propose BooM-VVT, a mask-free VVT framework built upon the keyframe-driven paradigm. To achieve mask-free VVT, we introduce a multi-stage training strategy that leverages image-level pseudo data for mask-free localization learning, substantially reducing the need for costly video-level pseudo data. To improve garment consistency, we propose Garment-Sensitive Keyframe Sampling, which selects keyframes based on garment-relevant body regions to better capture garment appearance. We further introduce Frame-Shared 3D-RoPE to establish spatiotemporal correspondences between keyframes and target video frames for accurate garment-detail transfer. Finally, we construct OmniView, a large-scale multi-view try-on dataset to support reliable try-on video generation under complex camera viewpoints and diverse try-on tasks. Extensive experiments demonstrate that BooM-VVT achieves superior temporal consistency and garment fidelity over existing methods. Project page: https://boomvvt.github.io/boomvvt.

Chinese Translation

视频虚拟试穿（VVT）旨在生成一个人穿着目标服装的真实视频。近期方法利用关键帧驱动的视频生成范式来提高在真实场景中的表现，但它们仍然依赖于掩膜来定位试穿区域，这使得它们在大幅运动和严重遮挡的情况下变得脆弱。尽管无掩膜的基于图像的试穿方法通过利用大规模伪数据显示出良好的结果，但将这一范式扩展到视频仍然困难，因为构建视频级伪数据的成本极高。此外，粗略的关键帧采样和多视角试穿数据的稀缺限制了现有关键帧驱动方法在保持服装一致性和处理多样化试穿任务方面的能力。为了解决这些挑战，我们提出了BooM-VVT，一个基于关键帧驱动范式的无掩膜VVT框架。为了实现无掩膜VVT，我们引入了一种多阶段训练策略，利用图像级伪数据进行无掩膜定位学习，显著减少了对昂贵的视频级伪数据的需求。为了提高服装一致性，我们提出了服装敏感的关键帧采样方法，该方法基于与服装相关的身体区域选择关键帧，以更好地捕捉服装外观。我们进一步引入了帧共享的3D-RoPE，以建立关键帧与目标视频帧之间的时空对应关系，从而实现准确的服装细节转移。最后，我们构建了OmniView，一个大规模的多视角试穿数据集，以支持在复杂摄像机视角和多样化试穿任务下的可靠试穿视频生成。大量实验表明，BooM-VVT在时间一致性和服装保真度方面优于现有方法。项目页面：https://boomvvt.github.io/boomvvt。

View on arXiv Download PDF AI Translation

cs.CV / 93 / 2609.04131

Beyond Retrieval: Progressive Latent Memory Evolution for Streaming Video Understanding

超越检索：用于流媒体视频理解的渐进式潜在记忆演化

Qu, Hongyu, Yao, Guangming, Xing, Ling, Hu, Xiaobin, Ding, Rongxing, Zhang, Guibin, Zhang, Fan, Yuan, Yi, Shu, Xiangbo, Yan, Shuicheng

Abstract

Streaming video understanding requires multimodal large language models (MLLMs) to process continuous visual inputs and respond to user queries under strict causality and bounded memory. Existing approaches typically compress historical observations into an external memory bank and retrieve query-relevant evidence as additional visual context. Though effective, this store-and-retrieve paradigm keeps historical evidence as external visual context, preventing it from being internalized into a compact, evolving latent memory that can continuously guide streaming reasoning. To bridge this gap, we introduce LatentStream, a progressive latent working memory framework that shifts streaming memory from store-and-retrieve to retrieve-and-internalize. Specifically, LatentStream comprises three coordinated components. First, Query-agnostic Hierarchical Streaming Memory organizes visual history into short-, mid-, and long-term levels under a fixed memory budget through Jenks-guided adaptive consolidation. Once a query arrives, Hierarchical Latent Memory Evolution equips groups of latent memory tokens with progressively expanding memory receptive fields, enabling them to iteratively retrieve historical evidence from their corresponding scopes and internalize it into a compact, fixed-length latent memory. Finally, Progressive Confidence-guided Latent Memory Optimization constructs a hierarchical progression reward from group-wise predictive entropy and jointly refines the latent memory tokens and retrieved evidence, encouraging increasingly confident streaming reasoning. Extensive experiments demonstrate that LatentStream achieves new state-of-the-art results on existing online and offline video benchmarks.

Chinese Translation

流媒体视频理解需要多模态大语言模型（MLLMs）处理连续的视觉输入，并在严格的因果关系和有限的记忆下响应用户查询。现有方法通常将历史观察压缩到外部记忆库中，并检索与查询相关的证据作为额外的视觉上下文。尽管有效，这种存储与检索的范式将历史证据保持为外部视觉上下文，阻碍了其内化为一个紧凑、不断演变的潜在记忆，从而能够持续指导流媒体推理。为了解决这一问题，我们提出了LatentStream，一个渐进式潜在工作记忆框架，将流媒体记忆从存储与检索转变为检索与内化。具体而言，LatentStream由三个协调的组件组成。首先，查询无关的分层流媒体记忆在固定的记忆预算下，通过Jenks引导的自适应整合，将视觉历史组织为短期、中期和长期层次。一旦查询到达，分层潜在记忆演化为潜在记忆标记组配备逐步扩展的记忆接收域，使其能够迭代地从相应范围内检索历史证据并将其内化为紧凑的固定长度潜在记忆。最后，渐进式置信度引导的潜在记忆优化从组间预测熵构建分层进展奖励，并共同优化潜在记忆标记和检索证据，鼓励越来越自信的流媒体推理。大量实验表明，LatentStream在现有在线和离线视频基准测试中实现了新的最先进结果。

View on arXiv Download PDF AI Translation

cs.CV / 94 / 2609.04151

Persistent Identity Preservation in Generative Image Models: A Benchmark and Evaluation System

生成图像模型中的持久身份保留：基准与评估系统

Ren, Mengwei, Zhang, Xuaner, Xia, Zhihao

Abstract

Generative image models can now produce high-quality images, follow complex instructions, and support precise edits, but they still struggle to preserve who or what is being depicted. When generating or editing images of a specific subject, identity may drift as the pose, expression, appearance, viewpoint, or surrounding scene changes. Existing subject-driven methods make fundamentally different choices about where identity is represented: through the input context (GPT-Image-2, NB2), as trainable subject-specific model parameters (LoRA), or as a persistent identity layer (PHOTA IDENTITY) reusable across generations and edits. We systematically benchmark these paradigms across subject-driven generation, editing, restoration, and multi-subject settings, with tasks designed to increasingly stress identity preservation. Our results show that identity preservation remains a distinct limitation of current generative foundation models: strong image quality and instruction following do not necessarily imply strong identity fidelity, and identity degradation becomes more pronounced under iterative edits, small subject scales, severe image degradation, and multi-subject composition. Persistent identity substantially reduces this degradation across generation, editing, and restoration, consistently improving identity preservation when applied to different foundation models while maintaining comparable instruction adherence and perceptual image quality. These results suggest that identity does not simply emerge from increasingly capable generative models, but can instead be represented as persistent subject knowledge that is composed independently with the underlying generative model.

Chinese Translation

生成图像模型现在能够生成高质量的图像，遵循复杂的指令，并支持精确的编辑，但在保留所描绘的对象身份方面仍然面临挑战。在生成或编辑特定对象的图像时，随着姿势、表情、外观、视角或周围场景的变化，身份可能会发生漂移。现有的以对象为驱动的方法在身份表示的选择上存在根本性的不同：通过输入上下文（GPT-Image-2，NB2）、作为可训练的特定对象模型参数（LoRA），或作为可在生成和编辑中重复使用的持久身份层（PHOTA IDENTITY）。我们系统地对这些范式在以对象为驱动的生成、编辑、修复和多对象设置中进行了基准测试，任务设计旨在逐步加大对身份保留的压力。我们的结果表明，身份保留仍然是当前生成基础模型的一个显著限制：强大的图像质量和指令遵循并不一定意味着强大的身份保真度，并且在迭代编辑、小对象规模、严重图像退化和多对象组合下，身份退化变得更加明显。持久身份在生成、编辑和修复过程中显著减少了这种退化，在应用于不同基础模型时始终提高了身份保留，同时保持了可比的指令遵循和感知图像质量。这些结果表明，身份并不是简单地从日益强大的生成模型中产生的，而是可以作为持久的对象知识进行表示，该知识可以与基础生成模型独立组合。

View on arXiv Download PDF AI Translation

cs.CV / 95 / 2609.04174

Zero-Shot Novel Depth Synthesis Using 3D Foundation Models Scene Representations

基于3D基础模型场景表示的零样本新颖深度合成

Akola, Denis M., Fouhey, David F.

Abstract

3D Foundation Models (3DFMs) such as VGGT have recently pushed the boundaries of 3D vision by predicting rich unified representations with feed-foward transformers. The scene representations learned by these models enable strong performance on multiple 3D vision tasks. In this paper, we investigate using their internal representations to infer 3D in the scene from new views. Our hypothesis is that in order to solve the task of 3D reconstruction, these models need to learn a representation that includes a large amount of general knowledge about 3D scenes. After showing that it is possible to decode hidden surfaces from internal 3DFM representations, we propose a method, Z3D, that estimates pointmaps in unseen views by doing latent diffusion on 3DFM representation. We show that Z3D can predict realistic depth maps for new views across multiple datasets.

Chinese Translation

3D基础模型（3DFMs），如VGGT，最近通过使用前馈变换器预测丰富的统一表示，推动了3D视觉的边界。这些模型学习到的场景表示在多个3D视觉任务中表现出色。本文探讨了利用它们的内部表示从新视角推断场景中的3D信息。我们的假设是，为了解决3D重建任务，这些模型需要学习一种包含大量关于3D场景的一般知识的表示。在证明可以从内部3DFM表示解码隐藏表面后，我们提出了一种方法Z3D，通过对3DFM表示进行潜在扩散来估计未见视角的点图。我们展示了Z3D能够在多个数据集上为新视角预测逼真的深度图。

View on arXiv Download PDF AI Translation

cs.CV / 96 / 2609.04183

Seeing Before Synthesizing: VLM-Guided Transition Event Discovery for Weakly-Supervised Dense Video Captioning

在合成之前进行观察：基于视觉语言模型的弱监督密集视频字幕中的过渡事件发现

Kim, Ye-Chan, Choi, Seunghee, Cha, SeungJu, Kim, Si-Woo, Kim, Hwiseon, Kim, Hyungee, Kim, Dong-Jin

Abstract

Weakly-Supervised Dense Video Captioning aims to localize and describe multiple events in untrimmed videos given only an ordered set of event-level captions per video. Recent work synthesizes auxiliary transition captions via LLM to provide additional vision-language alignment, but these captions lack visual grounding and are rigidly assigned to every inter-event gap at a fixed location and duration. To address these, we propose Seeing Before Synthesizing (SBS), a framework that adaptively provides visually grounded linguistic guidance only where warranted. Leveraging a VLM, we generate frame-level narratives for the inter-event gaps and detect transitions from the semantic variation across them. For identified transitions, we then refine inter-event temporal masks by blending the temporal midpoint with the semantic change point and selecting the width that maximizes vision-language alignment. Experiments on ActivityNet Captions and YouCook2 demonstrate state-of-the-art performance in both captioning and localization.

Chinese Translation

弱监督密集视频字幕旨在仅根据每个视频的有序事件级字幕，定位和描述未剪辑视频中的多个事件。近期的研究通过大型语言模型（LLM）合成辅助过渡字幕，以提供额外的视觉-语言对齐，但这些字幕缺乏视觉基础，并且被僵硬地分配到每个事件间隙的固定位置和持续时间。为了解决这些问题，我们提出了“在合成之前进行观察”（Seeing Before Synthesizing, SBS）框架，该框架自适应地仅在必要时提供视觉基础的语言指导。利用视觉语言模型（VLM），我们为事件间隙生成帧级叙述，并通过检测它们之间的语义变化来识别过渡。对于识别出的过渡，我们通过将时间中点与语义变化点进行融合，并选择最大化视觉-语言对齐的宽度，从而细化事件间的时间掩码。在ActivityNet Captions和YouCook2上的实验表明，在字幕生成和定位方面达到了最先进的性能。

View on arXiv Download PDF AI Translation

cs.CV / 97 / 2609.04196

Puffin-World: Scaling a Unified Multimodal Model with Native 3D World States

Puffin-World：扩展具有原生3D世界状态的统一多模态模型

Liao, Kang, Luo, Yihang, Wu, Xiao-Ming, Jin, Linyi, Wu, Size, Lin, Chunyu, Zhao, Yao, Wang, Fei, Li, Wei, Loy, Chen Change

Abstract

We propose Puffin-World, a unified multimodal architecture that integrates physical understanding, spatial simulation, and 3D world generation and reconstruction without relying on external offline modules. To reliably construct and interact with 3D worlds, our framework jointly models three native world states: physics (gravity field and latitude), geometry (depth), and appearance (image), together with a unified Omni-Camera representation that supports diverse tasks and flexible motions. Beyond modeling these states, we introduce a strategy for propagating physical dynamics across future frames. By grounding absolute camera properties in the real world, Puffin-World enables physically consistent and visually stable world generation. We further couple appearance and geometry within a single generative process, jointly synthesizing each future view and reconstructing its underlying geometry. This unified paradigm enables interleaved closed-loop applications requiring synergy across multiple tasks, including mimic and self-calibrated world exploration. To scale Puffin-World to complex scenarios, we construct Puffin-16M, comprising 15 million vision-language-camera triplets and 1 million trajectories featuring various and challenging motions. To foster further research in this area, we released the code, models, and datasets.

Chinese Translation

我们提出了Puffin-World，这是一种统一的多模态架构，集成了物理理解、空间模拟以及3D世界生成和重建，而无需依赖外部离线模块。为了可靠地构建和与3D世界进行交互，我们的框架共同建模了三种原生世界状态：物理（重力场和纬度）、几何（深度）和外观（图像），并结合了统一的全景相机（Omni-Camera）表示，支持多样化任务和灵活运动。除了建模这些状态外，我们还引入了一种在未来帧中传播物理动态的策略。通过将绝对相机属性与现实世界相结合，Puffin-World实现了物理一致和视觉稳定的世界生成。我们进一步将外观和几何结合在一个生成过程中，共同合成每个未来视图并重建其底层几何。这一统一范式使得需要跨多个任务协同的交错闭环应用成为可能，包括模仿和自校准的世界探索。为了将Puffin-World扩展到复杂场景，我们构建了Puffin-16M，包含1500万个视觉-语言-相机三元组和100万个轨迹，涵盖各种具有挑战性的运动。为了促进该领域的进一步研究，我们发布了代码、模型和数据集。

View on arXiv Download PDF AI Translation

cs.CV / 98 / 2609.04200

Principia: Relational Physics Tests for Video Models

Principia：视频模型的关系物理测试

Thozhiyoor, Varun Varma, Tripathi, Shivam, Radhakrishnan, Venkatesh Babu, Bhattad, Anand

Abstract

Evaluating physical reasoning in video models is difficult because absolute motion measurements depend on frame rate, object scale, and camera calibration, all of which are often ambiguous or unavailable in generated video. We propose a different approach. When two objects in the same scene obey the same physical law, their motions must satisfy predictable relationships, and these relationships hold independent of calibration. We introduce Principia, a benchmark that evaluates Newtonian physics through relational consistency between paired objects. Principia spans eight phenomena - gravity, restitution, friction, rotational inertia, projectile motion, momentum, pendulum, and mass-spring oscillation - across translational, rotational, collisional, and oscillatory dynamics, using real-world scenes recorded under controlled protocols. We also introduce a calibration-independent consistency score that quantifies physical violation directly in image space. Across thousands of generations from six state-of-the-art video generators, no model exceeds 0.42 on Principia despite all scoring around 0.8 on VBench. Vision-language models are evaluated on their ability to detect relational physics violations, with the best model achieving only 67% accuracy and most performing near chance level.

Chinese Translation

评估视频模型中的物理推理是困难的，因为绝对运动测量依赖于帧率、物体尺度和相机校准，而这些在生成的视频中往往模糊不清或不可用。我们提出了一种不同的方法。当同一场景中的两个物体遵循相同的物理法则时，它们的运动必须满足可预测的关系，这些关系与校准无关。我们引入了Principia，这是一个通过配对物体之间的关系一致性来评估牛顿物理学的基准。Principia涵盖了八种现象——重力、恢复、摩擦、转动惯量、抛体运动、动量、摆动和质量-弹簧振荡——涉及平移、旋转、碰撞和振荡动力学，使用在受控协议下录制的真实场景。我们还引入了一种与校准无关的一致性评分，直接在图像空间中量化物理违反。在来自六个最先进视频生成器的数千次生成中，没有模型在Principia上超过0.42，尽管所有模型在VBench上得分约为0.8。视觉-语言模型的评估基于其检测关系物理违反的能力，最佳模型的准确率仅为67%，而大多数模型的表现接近随机水平。

View on arXiv Download PDF AI Translation

cs.CV / 99 / 2609.04201

Scal3R: Learning Efficient Multi-Relative Pose Query for Scalable Online 3D Reconstruction

Scal3R：学习高效的多相对姿态查询以实现可扩展的在线三维重建

Lin, Chin-Yang, Sun, Yang-Che, Sun, Cheng, Yang, Fu-En, Chen, Min-Hung, Lin, Yen-Yu, Chiu, Wei-Chen, Liu, Yu-Lun

Abstract

Online 3D reconstruction models perform poorly on long videos. This happens because regressing poses relative to a fixed first-frame anchor forces extrapolation far beyond the training distribution. Small drifts accumulate and amplify into significant geometric collapse. However, we observe that per-frame depth remains stable throughout this failure. The backbone's local geometry remains intact; only the global pose head breaks down. Motivated by this decoupling, we introduce Scal3R. This approach reformulates online reconstruction as multi-reference relative pose querying. We use lightweight learnable tokens, which make up about ~1% of the parameters, and inject them into a completely frozen backbone via asymmetric attention. This setup queries poses relative to multiple past keyframes. An online pose-graph optimization system with loop closure suppresses long-range drift. Scal3R reaches convergence in 8 hours on a single GPU. It reduces the average ATE by over 60% on KITTI compared to the online baseline. It also achieves state-of-the-art performance across Virtual KITTI, Sintel, TUM-Dynamic, ScanNet, and 7-Scenes. Project page: https://linjohnss.github.io/scal3r/

Chinese Translation

在线三维重建模型在长视频上的表现较差。这是因为相对于固定的第一帧锚点回归姿态迫使模型进行超出训练分布的外推。小的漂移会累积并放大为显著的几何崩溃。然而，我们观察到每帧的深度在这一失败过程中保持稳定。主干网络的局部几何结构保持完整；只有全局姿态头崩溃。基于这一解耦现象，我们提出了Scal3R。该方法将在线重建重新表述为多参考相对姿态查询。我们使用轻量级可学习的标记，这些标记约占参数的1%，并通过非对称注意力将其注入到完全冻结的主干网络中。该设置相对于多个过去的关键帧查询姿态。一个带有环闭合的在线姿态图优化系统抑制了长距离漂移。Scal3R在单个GPU上8小时内达到收敛。与在线基线相比，它在KITTI上将平均绝对轨迹误差（ATE）降低了超过60%。它还在Virtual KITTI、Sintel、TUM-Dynamic、ScanNet和7-Scenes上实现了最先进的性能。项目页面：https://linjohnss.github.io/scal3r/

View on arXiv Download PDF AI Translation

cs.CV / 100 / 2609.04202

TokenMatch: 3D Mesh Correspondence Transformer with Curvature-Guided Tokenisation

TokenMatch：基于曲率引导的3D网格对应变换器

Islam, Adeela, Lähner, Zorah, Murino, Vittorio, Golyanik, Vladislav

Abstract

While data-driven 3D shape correspondence estimation has recently seen substantial progress, robust matching under partial observations and strong non-isometric deformations remains challenging. Existing learning-based approaches often rely on hand-crafted descriptors or template-based representations, whereas recent generative models over functional maps suffer from high inference cost, limited interpretability, and poor generalisation to partial shapes. In response to these limitations, this paper introduces TokenMatch, a new transformer-based unified model for estimating 3D shape correspondences. Our feed-forward approach trained exclusively on BeCoS, a challenging non-isometric partial-to-partial shape-matching dataset, can generalise to matching full shapes without retraining or fine-tuning. TokenMatch uses self- and cross-attention mechanisms to efficiently learn patch-level and point-level relations as well as dense correspondences between shape pairs. Our core insight is that meshes can be adaptively tokenised into patches using shape curvature guidance, enabling effective learning of shape-specific geometric descriptors for correspondence estimation. We evaluate TokenMatch on standard benchmarks for partial and full shape matching, including CP2P, PSMAL, BeCoS, FAUST, SCAPE, and SHREC'19. Our method achieves consistently high performance, in most cases outperforming existing methods for partial and full shape matching in the mean geodesic error and intersection-over-union metrics, while also running faster at sub-second inference speeds.

Chinese Translation

尽管基于数据的3D形状对应估计最近取得了显著进展，但在部分观测和强非等距变形下的鲁棒匹配仍然具有挑战性。现有的基于学习的方法通常依赖于手工设计的描述符或基于模板的表示，而最近的基于功能映射的生成模型则面临高推理成本、有限的可解释性和对部分形状的较差泛化能力。针对这些局限性，本文提出了TokenMatch，一种新的基于变换器的统一模型，用于估计3D形状对应。我们基于前馈的方法专门在BeCoS这一具有挑战性的非等距部分到部分形状匹配数据集上进行训练，能够在不重新训练或微调的情况下泛化到完整形状的匹配。TokenMatch利用自注意力和交叉注意力机制有效学习补丁级和点级关系，以及形状对之间的密集对应。我们的核心见解是，网格可以通过形状曲率引导自适应地被分割为补丁，从而有效学习形状特定的几何描述符以进行对应估计。我们在标准的部分和完整形状匹配基准上评估TokenMatch，包括CP2P、PSMAL、BeCoS、FAUST、SCAPE和SHREC'19。我们的方法在大多数情况下在平均测地误差和交并比指标上超越现有的部分和完整形状匹配方法，同时在推理速度上也更快，达到亚秒级的推理速度。

View on arXiv Download PDF AI Translation

cs.CV / 101 / 2609.04203

Temporal Self-Distillation: Learning Visual State Tracking in Videos Without Supervision

时间自蒸馏：无监督学习视频中的视觉状态跟踪

Venkatraman, Shravan, Zhao, Wenshuai, Vali, Mohammad Hassan, Solin, Arno

Abstract

We introduce S$^3$T (Self-Supervised Self-Distillation over Time), which, to the best of our knowledge, is the first fully self-contained framework for continuous video state tracking. Our method treats temporal sampling density as privileged information, based on the hypothesis that a denser view of the same clip recovers the running state more accurately. This view serves as the teacher, while a sparse-view student with the same weights learns to match its next-token distribution. The model generates its own target, so training requires no labels, separate teacher, or reward signal, and adds no inference cost. On LLaVA-OneVision-2-8B, S$^3$T improves VSTAT accuracy by $+1.74$ as a single model, $+2.38$ with souping, and $+2.70$ with additional vision-encoder adaptation, while prior self-evolving methods leave state tracking largely unchanged. The capability learned from unlabeled synthetic clips transfers to real videos, improving performance by $+7.95$ on VSTAT-YouTube state-tracking questions and $+4.50$ on MVBench Action Count.

Chinese Translation

我们提出了 S$^3$T（时间上的自监督自蒸馏），据我们所知，这是第一个完全自包含的连续视频状态跟踪框架。我们的方法将时间采样密度视为特权信息，基于这样的假设：对同一剪辑的更密集视图可以更准确地恢复运行状态。这个视图充当教师，而具有相同权重的稀疏视图学生学习匹配其下一个标记的分布。该模型生成自己的目标，因此训练不需要标签、单独的教师或奖励信号，并且不会增加推理成本。在 LLaVA-OneVision-2-8B 上，S$^3$T 作为单一模型提高了 VSTAT 准确率 $+1.74$，通过 souping 提高了 $+2.38$，通过额外的视觉编码器适配提高了 $+2.70$，而之前的自演化方法几乎没有改变状态跟踪。从未标记的合成剪辑中学习到的能力可以转移到真实视频上，在 VSTAT-YouTube 状态跟踪问题上提高了 $+7.95$，在 MVBench 动作计数上提高了 $+4.50$。

View on arXiv Download PDF AI Translation

人工智能 (Artificial Intelligence)

cs.AI / 1 / 2609.02981

Structure and Implementation of New Practical English Textbooks Driven by Artificial Intelligence

基于人工智能的新实用英语教材的结构与实施

Wang, Ya, Zhang, Lei, Yang, Xueguang, Chen, Bo

Abstract

Artificial intelligence is changing the form of applied English materials from fixed paper sequences to adaptive learning systems that can diagnose learners, recommend tasks, and provide formative feedback. This paper studies the structure and application of a new practical English textbook driven by artificial intelligence. A five-layer architecture is proposed: knowledge mapping, learner profiling, task generation, feedback orchestration, and teacher-side governance. A prototype was tested on 186 non-English-major undergraduates for eight weeks of teaching. Compared with a static digital textbook, the proposed system increased the unit completion accuracy from 72.4% to 84.9%, raised the average score for speaking tasks by 10.8 points, and reduced the teacher's correction time by 31.6%. Therefore, an AI-driven textbook can maintain the stability of the curriculum while providing personalised learning paths, rich practice materials and traceable classroom data.

Chinese Translation

人工智能正在将应用英语材料的形式从固定的纸质序列转变为能够诊断学习者、推荐任务和提供形成性反馈的自适应学习系统。本文研究了一种基于人工智能的新实用英语教材的结构和应用。提出了一种五层架构：知识映射、学习者画像、任务生成、反馈协调和教师端治理。该原型在186名非英语专业本科生中进行了为期八周的教学测试。与静态数字教材相比，所提系统将单元完成准确率从72.4%提高至84.9%，口语任务的平均分提高了10.8分，教师的批改时间减少了31.6%。因此，基于人工智能的教材能够在提供个性化学习路径、丰富的练习材料和可追踪的课堂数据的同时，保持课程的稳定性。

View on arXiv Download PDF AI Translation

cs.AI / 2 / 2609.03209

MasterControl Seventeen Every Time

每次都掌控十七

Lab, MasterControl AI

Abstract

We study a governed approach to enterprise analytics: a language model interprets the question, while deterministic policy selects and runs a pre-approved analytical program that returns both results and evidence. We show that this restriction can remain expressive within a defined analytical class, using relational operations plus aggregation, comparison, windows, ranking, and similarity. Fixed meaning, policy, data, and execution rules also make results replayable. Across 440 runs, three 8B models generated SQL and selected tools at runtime, while Qwen3-8B interpreted intent only and policy executed the approved program. None of 330 runtime-planning episodes matched the full answer-and-evidence contract across all test datasets; the policy-executed analyzer matched 110 of 110. This is a configuration-specific result, not evidence that runtime agents cannot succeed under other designs.

Chinese Translation

我们研究了一种受控的企业分析方法：语言模型解释问题，而确定性策略选择并运行经过预批准的分析程序，返回结果和证据。我们展示了这种限制在定义的分析类别内仍然可以保持表达性，使用关系操作、聚合、比较、窗口、排名和相似性。固定的含义、策略、数据和执行规则也使得结果可重放。在440次运行中，三个8B模型在运行时生成SQL并选择工具，而Qwen3-8B仅解释意图，策略执行经过批准的程序。在330个运行规划情节中，没有一个匹配所有测试数据集的完整答案和证据合同；而策略执行的分析器匹配了110个中的110个。这是一个特定配置的结果，并不证明运行时代理在其他设计下无法成功。

View on arXiv Download PDF AI Translation

cs.AI / 3 / 2609.03236

Speculative Macro Commit for Faster Tool-Using Agents

快速工具使用代理的推测宏提交

Liu, Zeyu, Kundu, Souvik, Beerel, Peter A.

Abstract

Tool-using LLM agents spend wall-clock time not only on model inference but also in serial action--observation turns, where each tool call, environment transition, and observation can delay subsequent decisions. We introduce \textbf{Speculative Macro Commit} (SMC), a runtime mechanism for a two-tier agent system: a large authoritative actor model produces the official trajectory, while a faster speculative drafter model continuously predicts and executes future action chains on an isolated environment snapshot. SMC mines recurring multi-action skeletons from training traces and stores them in a macro library used to match against action chains predicted by the drafter at runtime. When the actor's next tool call matches the first drafted action, SMC commits the remaining pre-executed draft steps, together with their observations, to the official trajectory. Using Qwen3.5-27B INT4 as the authoritative actor model and Qwen3.5-4B as the speculative drafter model, SMC matches the sequential agent's overall accuracy while reducing latency by 10.23\% over the Speculative Actions (SA) baseline and 18.59\% over sequential execution on the $\tau^2$-Bench Telecom subset. On AppWorld, SMC reduces wall time by 7.7\% over SA baseline and 44.9\% over sequential execution, with a small reduction in task completion. Overall, SMC provides a practical way to reuse multi-step speculative execution and reduce agent latency beyond single-step speculative actions. Our code is publicly available \href{https://github.com/zeyuliu1037/speculative-macro-commit}{\textcolor{magenta}{here}}.

Chinese Translation

工具使用的大型语言模型（LLM）代理不仅在模型推理上消耗时间，还在串行的动作-观察回合中消耗时间，其中每次工具调用、环境转换和观察都可能延迟后续决策。我们提出了 extbf{推测宏提交}（Speculative Macro Commit, SMC），这是一种针对双层代理系统的运行时机制：一个大型权威行为者模型生成官方轨迹，而一个更快的推测草拟模型则在隔离的环境快照上持续预测和执行未来的动作链。SMC从训练轨迹中挖掘重复的多动作骨架，并将其存储在宏库中，以便在运行时与草拟者预测的动作链进行匹配。当行为者的下一个工具调用与第一个草拟动作匹配时，SMC将剩余的预执行草拟步骤及其观察结果提交到官方轨迹中。使用Qwen3.5-27B INT4作为权威行为者模型，Qwen3.5-4B作为推测草拟模型，SMC在保持顺序代理整体准确性的同时，将延迟减少了10.23 ext{%}，相较于推测动作（Speculative Actions, SA）基线减少了18.59 ext{%}，在$ au^2$-Bench Telecom子集上表现显著。在AppWorld上，SMC相比SA基线将墙面时间减少了7.7 ext{%}，相比顺序执行减少了44.9 ext{%}，任务完成率略有下降。总体而言，SMC提供了一种实用的方法来重用多步骤推测执行，并在超越单步推测动作的情况下减少代理延迟。我们的代码公开可用，链接在此 extcolor{magenta}{ ext{https://github.com/zeyuliu1037/speculative-macro-commit}}。

View on arXiv Download PDF AI Translation

cs.AI / 4 / 2609.03340

Fresh Memory, Stale Plans: Dependency-Scoped Validation for Distributed LLM-Agent Memory

新鲜记忆，过时计划：面向依赖的分布式 LLM-Agent 内存验证

Chen, Evan, Wang, Shiqiang, Brinton, Christopher G.

Abstract

Distributed LLM-agent teams can read the latest shared facts and still act on an obsolete plan. A planner may derive an action from requirement $r_3$, another agent may commit $r_4$, and an executor may receive $r_4$ without replacing the plan derived from $r_3$. We call this \emph{stale-plan execution}: state freshness does not establish that the plan authorizing an action remains valid. We introduce PlanFence, a dependency-scoped action-validation protocol. Plans cite the exact public records they used, and an executor validates only the records that can affect the pending external action, replanning once or blocking when validation is incomplete. In 30 controlled live workflows with a post-plan revision, a freshness-only executor acts on the obsolete plan in every task, whereas PlanFence completes all tasks without an invalid action. Controlled replay reveals two conditional boundaries: proactive synchronization yields lower coordination stall at low churn, while PlanFence avoids repeated update-path coordination as churn grows and avoids validating unrelated state as the shared keyspace grows. These are controlled safety and systems-cost results, not general task-accuracy gains.

Chinese Translation

分布式 LLM-agent 团队可以读取最新的共享事实，但仍然基于过时的计划进行行动。一个规划者可能从需求 $r_3$ 推导出一个行动，另一个代理可能提交 $r_4$，而执行者可能接收到 $r_4$，却没有替换基于 $r_3$ 推导出的计划。我们称之为 extit{过时计划执行}：状态的新鲜度并不能证明授权某个行动的计划仍然有效。我们提出了 PlanFence，一种面向依赖的行动验证协议。计划引用它们所使用的确切公共记录，执行者仅验证可能影响待处理外部行动的记录，当验证不完整时进行重新规划或阻止执行。在 30 个受控的实时工作流中，进行后计划修订的情况下，单纯依赖新鲜度的执行者在每个任务中都基于过时计划进行行动，而 PlanFence 则在没有无效行动的情况下完成所有任务。受控重放揭示了两个条件边界：主动同步在低变动率下降低了协调停滞，而 PlanFence 随着变动率的增加避免了重复的更新路径协调，并在共享键空间增长时避免验证无关状态。这些是受控的安全性和系统成本结果，而非一般任务准确性的提升。

View on arXiv Download PDF AI Translation

cs.AI / 5 / 2609.03402

A Prompt-Engineering Approach to Develop Scalable, Flexible, and Real-Time Hybrid Micro-Level Personalization in a General Purpose AI Teaching Assistant

一种提示工程方法用于开发可扩展、灵活且实时的混合微观层次个性化通用人工智能教学助手

Basu, Saptarshi, Kakar, Sandeep, Goel, Ashok

Abstract

Artificial intelligence (AI) teaching assistants powered by large language models (LLMs) offer scalable educational support but often provide limited personalization. This study presents a prompt-engineering-based framework for personalizing general-purpose LLM/RAG-based AI teaching assistants such as Jill Watson across academic disciplines and courses. The framework adapts responses using six learner-specific dimensions: self-assessment, abstraction preference, verbosity preference, perceptual orientation, information processing style, and level of understanding, yielding 96 distinct learner profiles. Student queries are additionally analyzed using Bloom's Taxonomy to estimate cognitive complexity at the interaction level. Learner attributes and cognitive assessments are encoded in structured prompts that condition the LLM without requiring model retraining. The framework is evaluated through experiments using NLP metrics and a human study with five participants. Results show perceived differences in response style and structure across personalization conditions, with statistical analyses identifying learner attributes associated with measurable response changes. These findings provide preliminary evidence that prompt-based personalization can support adaptive behavior in LLM-powered educational agents.

Chinese Translation

由大型语言模型（LLMs）驱动的人工智能（AI）教学助手提供可扩展的教育支持，但通常个性化程度有限。本研究提出了一种基于提示工程的框架，用于个性化通用的LLM/RAG基础的AI教学助手，如Jill Watson，适用于不同学科和课程。该框架通过六个学习者特定维度调整响应：自我评估、抽象偏好、冗长偏好、感知取向、信息处理风格和理解水平，从而生成96种不同的学习者档案。此外，学生的提问还使用布鲁姆分类法（Bloom's Taxonomy）进行分析，以估计交互层面的认知复杂性。学习者属性和认知评估被编码在结构化的提示中，以调节LLM，而无需重新训练模型。该框架通过使用自然语言处理（NLP）指标的实验和对五名参与者的人类研究进行评估。结果显示，在个性化条件下，响应风格和结构存在感知差异，统计分析确定了与可测量响应变化相关的学习者属性。这些发现提供了初步证据，表明基于提示的个性化可以支持LLM驱动的教育代理的适应性行为。

View on arXiv Download PDF AI Translation

cs.AI / 6 / 2609.03407

Caught in the Story: Narrative Captivity in Multi-turn LLMs Conversation

被困于叙事中：多轮对话中的叙事囚禁

Wu, Yuhe, Wang, Guangyu, Chen, Yujie, Zhang, Jiatong, Chen, Yuran, Zhang, Yutong, Cheng, Xiyin, Cao, Wenpeng, Liu, Zhuang, Zhang, Guang

Abstract

People increasingly turn to large language models (LLMs) for everyday advice, making ethically charged interpersonal problems a practical moral-advisory context. Most prior work has studied this context through single-turn judgments or pressure-laden rebuttals, assumptions that poorly match how guidance is sought in real-world contexts. These assumptions leave unclear whether narration alone, without an explicit opposing position, can shift model judgments during multi-turn moral consultation. Yet real-world moral-conflict conversation often elicits one party's self-justifying account, which can unfold over multiple turns and create information asymmetry. We introduce \textbf{narrative captivity}, a failure mode in which a model treats an unopposed one-sided account as complete and aligns with the narrator's interpretation without seeking missing perspectives. To measure this phenomenon, we build a benchmark of $5{,}078$ interpersonal-conflict scenarios spanning six moral dimensions. Across 17 LLMs, narrative captivity is widespread: end-state judgments under multi-turn narration shift by 25 percentage points on average beyond the matched single-turn baseline. Stage-level analysis identifies preference optimization as a major contributor, while four inference-time strategies provide only partial mitigation. We hope our project fosters LLM advisors that preserve independent judgment in real-world consultation.

Chinese Translation

人们越来越多地依赖大型语言模型（LLMs）来获取日常建议，使得伦理上充满挑战的人际问题成为一个实际的道德咨询背景。以往的大多数研究通过单轮判断或施加压力的反驳来研究这一背景，这些假设与现实世界中寻求指导的方式不符。这些假设使得尚不清楚在多轮道德咨询中，仅凭叙述而没有明确的对立立场是否能够改变模型的判断。然而，现实世界中的道德冲突对话通常会引发一方自我辩解的叙述，这种叙述可能会持续多个回合，并造成信息不对称。我们引入了 extbf{叙事囚禁}，这是一种失效模式，其中模型将未遭反对的单方面叙述视为完整，并在未寻求缺失视角的情况下与叙述者的解释保持一致。为了衡量这一现象，我们构建了一个包含$5{,}078$个跨越六个道德维度的人际冲突场景的基准。在17个LLMs中，叙事囚禁现象普遍存在：在多轮叙述下的最终判断平均比匹配的单轮基线偏移25个百分点。阶段级分析确定偏好优化是主要贡献因素，而四种推理时策略仅能部分缓解这一现象。我们希望我们的项目能够促进LLM顾问在现实咨询中保持独立判断。

View on arXiv Download PDF AI Translation

cs.AI / 7 / 2609.03416

Dude: A Dual-Detection Multi-Agent System for Paper-Code Discrepancy Detection

Dude：一种用于论文与代码不一致检测的双重检测多智能体系统

Liu, Weijie, Zhao, Running, Yuan, Wenhao, Xu, Jinfeng, Xu, Zhanfeng, Zhang, Xiaoxi, Ngai, Edith Cheuk-Han

Abstract

LLM-empowered paper-code discrepancy detection has received growing concern since the scaling of research submissions exceeds the manual review capability. However, the limited context capacity and one-sided discrepancy detection of existing single-agent LLM paradigms lead to an inferior recall performance in detecting discrepancies. In this paper, we propose Dude, the first Dual-Detection Multi-Agent System for paper-code discrepancy detection. We discover that the granularity asymmetry of the paper-language and code-language introduces over-interpretation and over-reporting challenges in a multi-agent system design for discrepancy detection, resulting in increasing false positives. To address this, we propose a granularity-aligned negotiation and a two-stage salience-filtering mechanism in Dude, which effectively prevents agents from falsely reporting discrepancies. Experimental results in real-world paper-code discrepancy datasets showcase Dude's significant recall and precision improvement by up to 22.8%, increasing F1 score by up to 18.7% compared to baseline methods.

Chinese Translation

随着研究提交数量的增加超出了人工审查的能力，基于大型语言模型（LLM）的论文与代码不一致检测受到了越来越多的关注。然而，现有单智能体LLM范式的有限上下文容量和单方面的不一致检测导致其在检测不一致性时的召回性能较差。本文提出了Dude，这是首个用于论文与代码不一致检测的双重检测多智能体系统。我们发现论文语言与代码语言的粒度不对称在多智能体系统设计中引入了过度解释和过度报告的挑战，导致假阳性数量增加。为了解决这个问题，我们在Dude中提出了一种粒度对齐的协商机制和一个两阶段显著性过滤机制，有效防止智能体错误报告不一致性。在真实世界的论文与代码不一致数据集上的实验结果展示了Dude在召回率和精确度上显著提升，最高可达22.8%，F1分数相比基线方法提高了最高18.7%。

View on arXiv Download PDF AI Translation

cs.AI / 8 / 2609.03423

DuplexSpeechBench-IFEval: Evaluating Implicit Instruction Following in Full-Duplex Voice Agents

DuplexSpeechBench-IFEval：评估全双工语音代理中的隐式指令遵循

Mathur, Puneet, Manocha, Dinesh

Abstract

Full-duplex voice agents must continuously decide when to listen, backchannel, interrupt, handle speech overlaps, take the floor, and yield. Existing benchmarks largely test these behaviors through explicit turn-management instructions, while deployed agents are often configured through roles or personas from which the appropriate conversational behavior must be inferred. We introduce DuplexSpeechBench-IFEval (DSB-IFEval) for evaluating implicit instruction-following in real-time spoken interaction. (DSB-IFEval) comprises 1,038 test cases spanning eight diverse assistant roles and evaluates five conditioning protocols for instruction-following: default behavior, explicit behavioral instructions, persona-implied behavior, combined persona--rule conditioning, and instruction conflict. We measure real-time floor management using a deterministic Instruction Adherence Score (IAS) and persona-consistent content using LLM-judged Persona Adherence Score (PAS). Across six real-time speech systems, we find architecture-dependent trade-offs. Full duplex models like F-Actor and PersonaPlex are more sensitive to whether conversational behavior is stated explicitly or must be inferred from a persona, with adherence dropping by 9.7% and 4.5%, respectively, under persona-only conditioning. In contrast, GPT-Realtime, MiniCPM-o, and Fun-Audio-Chat strongly adhere to persona-consistent content, but their floor behavior does not adapt across explicit and persona-only instructions and remains constrained on several proactive actions. We further find that even if systems reliably follow conflicting directives to their prescribed persona, they still struggle to override them under safety conflict. These results show that inferring the behavior implied by a role, executing it at the appropriate conversational moment, and resolving competing instructions remain distinct challenges for full-duplex voice agents.

Chinese Translation

全双工语音代理必须持续决定何时倾听、进行反馈、打断、处理语音重叠、发言以及让步。现有基准主要通过显式的轮次管理指令来测试这些行为，而实际部署的代理通常通过角色或人格进行配置，从中必须推断出适当的对话行为。我们引入了DuplexSpeechBench-IFEval（DSB-IFEval）来评估实时口语交互中的隐式指令遵循。DSB-IFEval包含1,038个测试案例，涵盖八种不同的助手角色，并评估五种指令遵循的条件协议：默认行为、显式行为指令、人格暗示行为、结合人格与规则的条件以及指令冲突。我们使用确定性的指令遵循评分（Instruction Adherence Score, IAS）来衡量实时发言管理，并使用大型语言模型评判的人格遵循评分（Persona Adherence Score, PAS）来评估与人格一致的内容。在六个实时语音系统中，我们发现架构依赖的权衡。像F-Actor和PersonaPlex这样的全双工模型对对话行为是否显式陈述或必须从人格中推断更为敏感，在仅基于人格的条件下，遵循度分别下降了9.7%和4.5%。相反，GPT-Realtime、MiniCPM-o和Fun-Audio-Chat在与人格一致的内容上表现出强烈的遵循性，但它们的发言行为在显式和仅基于人格的指令之间并未适应，并在若干主动行为上受到限制。我们进一步发现，即使系统可靠地遵循与其规定的人格相冲突的指令，它们在安全冲突下仍然难以覆盖这些指令。这些结果表明，推断角色暗示的行为、在适当的对话时刻执行该行为以及解决相互竞争的指令仍然是全双工语音代理面临的独特挑战。

View on arXiv Download PDF AI Translation

cs.AI / 9 / 2609.03438

Do GUI Agents Know When Not to Act? Enabling Conflict-Aware Termination for Multimodal GUI Agents

图形用户界面代理是否知道何时不行动？实现多模态图形用户界面代理的冲突感知终止

Huang, Zhaoyuan, Ju, Tianjie, Cheng, Pengzhou, Wu, Zheng, Li, Yansi, Song, Chuanbiao, Lan, Jun, Zhu, Huijia, Wang, Weiqiang, Zhang, Zhuosheng

Abstract

Graphical user interface (GUI) agents are increasingly used to execute natural-language instructions on user interfaces, yet real users may issue infeasible instructions due to benign mistakes. A reliable agent should not only know how to act, but also when not to act. In this work, we introduce CONFLICTGUI, a benchmark covering instruction-internal conflicts and instruction-GUI context conflicts to study conflict-aware termination. Our evaluation reveals severe execution-biased overcompliance: agents that perform well on feasible tasks often continue to execute blindly under conflicting instructions. To mitigate this behavior, we propose CONFLICTGUARD, an inference-time framework that aligns an agent's feasibility awareness with its action generation. CONFLICTGUARD contains two coupled components: a feasibility verification protocol that guides the agent to assess instruction logic and GUI-side evidence before acting, and a conditional action modulation mechanism that steers agents from over-compliant execution into termination-oriented behavior. Experiments across five widely-used agents demonstrate that CONFLICTGUARD improves average conflict task success rate significantly, while preserving normal GUI-task performance. These results validate that a lightweight inference-time intervention can substantially boost GUI Agent's competence to identify inappropriate execution scenarios and refrain from unnecessary actions.

Chinese Translation

图形用户界面（GUI）代理越来越多地用于在用户界面上执行自然语言指令，但真实用户可能由于无意的错误而发出不可行的指令。一个可靠的代理不仅应知道如何行动，还应知道何时不行动。在本研究中，我们介绍了CONFLICTGUI，一个涵盖指令内部冲突和指令与GUI上下文冲突的基准，以研究冲突感知终止。我们的评估揭示了严重的执行偏向性过度遵从：在可行任务上表现良好的代理在冲突指令下往往会盲目继续执行。为减轻这种行为，我们提出了CONFLICTGUARD，一个推理时框架，将代理的可行性意识与其行动生成对齐。CONFLICTGUARD包含两个耦合组件：一个可行性验证协议，指导代理在行动之前评估指令逻辑和GUI侧证据，以及一个条件行动调节机制，促使代理从过度遵从的执行转向以终止为导向的行为。针对五个广泛使用的代理的实验表明，CONFLICTGUARD显著提高了平均冲突任务成功率，同时保持了正常的GUI任务性能。这些结果验证了轻量级推理时干预可以显著提升GUI代理识别不当执行场景的能力，并避免不必要的行动。

View on arXiv Download PDF AI Translation

cs.AI / 10 / 2609.03460

Beyond "Made with AI": Visualizing Provenance Density to Mitigate the Transparency Penalty

超越“由AI制作”：可视化来源密度以减轻透明度惩罚

Zhang, Qing, Huang, Yifei, Lee, Juyoung, Starner, Thad, Rekimoto, Jun

Abstract

As generative AI makes polished prose cheap to produce, users can no longer rely on fluency as a proxy for truth. We call this failure mode the Fluency Trap: users trust fluent hallucinations while also discounting accurate content once it is disclosed as AI-generated. Binary ``Made with AI'' labels respond with authorship disclosure, but they do not show what supports a claim. We propose Provenance Density, an evidence-visualization interface that shows the density of verified claims in a text. In a user study with 81 participants, an idealized Provenance Density interface produced a large discernment gap between truth and fabrication ($+4.15$ points, $d=1.82$), whereas participants given no signal showed no detectable discrimination. A technical audit with 200 samples shows that retrieval density alone is insufficient; unexpectedly, the Consistency Veto carries most of the discriminative signal on dynamic queries. As AI-generated content becomes indistinguishable from human writing, effective transparency must move from authorship disclosure toward evidence visualization.

Chinese Translation

随着生成性AI使得精炼的散文变得廉价，用户不再能够依赖流畅性作为真相的代理。我们将这种失败模式称为流畅陷阱（Fluency Trap）：用户信任流畅的幻觉，同时在内容被披露为AI生成后又会贬低其准确性。二元的“由AI制作”（Made with AI）标签通过作者披露来响应，但它们并未显示支持某一主张的依据。我们提出了来源密度（Provenance Density），这是一种证据可视化界面，展示文本中经过验证的主张的密度。在一项包含81名参与者的用户研究中，理想化的来源密度界面在真相与虚构之间产生了显著的辨别差距（+4.15分，d=1.82），而没有任何信号的参与者则未显示出可检测的区分。对200个样本的技术审计表明，仅凭检索密度是不够的；出乎意料的是，一致性否决（Consistency Veto）在动态查询中携带了大部分的辨别信号。随着AI生成的内容与人类写作变得难以区分，有效的透明度必须从作者披露转向证据可视化。

View on arXiv Download PDF AI Translation

cs.AI / 11 / 2609.03478

AutoGraphForge: Towards Automated Graph Theory Discovery

AutoGraphForge：迈向自动化图论发现

Pastorek, Ján

Abstract

We report on our ongoing project to develop a computational pipeline, AutoGraphForge, for an automated graph-theoretic conjecturing-refuting-formalizing-proving system. Conjecture generation is counterexample-guided and runs in rounds: a Graffiti3 generator proposes conjectures over a small, evolving snapshot table $T$ (initially a few hundred graphs with their computed invariants) that grows only by counterexamples to its own conjectures. A novelty filter of $559$ classical and folklore relations, closed under transitive composition and linear identity substitution, decides via a linear program whether a candidate is already implied by known results. Surviving candidates are tested against a dataset of about $348,000$ graphs, unioning the complete House of Graphs invariant export, the exhaustive census of all connected graphs on at most nine vertices, several extremal families (strongly regular, minimal Ramsey, Cayley, cages, barbells, lollipops, spiders), and random models. Counterexample-search algorithms then attack the remainder. Run for several rounds on an HPC cluster, the loop yields $6,522$ conjectures that survived the refutation dataset, the novelty filter and every active-search run -- among them nontrivial relations between the annihilation number and the edge-cover number for bipartite and regular graphs, which we prove by hand. A subsequent formalization and proving stage deterministically translates each surviving conjecture into a Lean 4 statement skeleton; every candidate proof is kernel-verified against a pinned mathlib4 and our custom invariant preamble. This stage integrates two neural provers -- DeepSeek-Prover-V2-671B (served with vLLM) and the Lean-specialised OProver-32B -- behind the independent kernel check. It is implemented end-to-end and passes initial sanity checks, with the full pipeline currently running on the cluster.

Chinese Translation

我们报告了正在进行的项目，旨在开发一个计算管道AutoGraphForge，用于自动化的图论猜想-反驳-形式化-证明系统。猜想生成是基于反例引导的，并且以轮次进行：一个Graffiti3生成器在一个小的、不断演变的快照表$T$（最初包含几百个图及其计算出的不变量）上提出猜想，该表仅通过对自身猜想的反例进行扩展。一个包含559个经典和民间关系的新颖性过滤器，在传递组合和线性恒等替换下封闭，通过线性程序决定候选猜想是否已被已知结果所暗示。存活的候选猜想会在大约348,000个图的数据集上进行测试，该数据集结合了完整的图的不变量导出、最多九个顶点的所有连通图的详尽普查、几个极值族（强正则、最小Ramsey、Cayley、笼、杠铃、棒棒糖、蜘蛛）以及随机模型。反例搜索算法随后攻击剩余的候选。该循环在高性能计算集群上运行多个轮次，产生了6,522个在反驳数据集、新颖性过滤器和每次主动搜索运行中存活的猜想——其中包括二部图和正则图的湮灭数与边覆盖数之间的非平凡关系，我们通过手动证明了这一点。随后，形式化和证明阶段确定性地将每个存活的猜想转换为Lean 4语句框架；每个候选证明都经过与固定的mathlib4和我们自定义的不变量前言的内核验证。此阶段集成了两个神经证明器——DeepSeek-Prover-V2-671B（通过vLLM提供服务）和专门针对Lean的OProver-32B——并在独立的内核检查后进行验证。该系统实现了端到端，并通过了初步的合理性检查，目前完整的管道正在集群上运行。

View on arXiv Download PDF AI Translation

cs.AI / 12 / 2609.03493

Making Every Tool Call Count: Necessary Tool-Evidence Path Rewards for Agentic Vision-Language Models

让每个工具调用都发挥作用：代理视觉-语言模型所需的工具-证据路径奖励

Long, Xingming, Liu, Yu, Yang, Zhiwei, Feng, Hanqi, Zhang, Shaojie, Poczos, Barnabas, Jiang, Chao, Luo, Zhenbo, Jiang, Lei, Fu, Pei

Abstract

Modern vision-language models (VLMs) can directly answer many image-grounded questions, yet they often struggle with complex queries requiring fine-grained visual details or external knowledge. To acquire this missing evidence, agentic VLMs invoke tools such as image cropping, image search, and text search. However, existing training paradigms primarily evaluate tool-use based on final answer correctness, leaving evidence acquisition and utilization insufficiently supervised. This leads to two critical shortcomings: (i) models frequently issue redundant or off-target tool calls that fail to gather necessary evidence, and (ii) even when appropriate tools are called, models often fail to extract the necessary information from the resulting observations. To address these limitations, we introduce the NTEP (Necessary Tool-Evidence Path), a novel annotation scheme that explicitly specifies the essential external evidence and corresponding tool calls for each query. Building upon this, we propose NTEP-R (NTEP Reward), a supervision mechanism ensuring that each tool invocation strictly advances the reasoning process toward the final solution. Specifically, our approach rewards the agent for aligning its pre-call intent with a necessary evidence-seeking goal, and for ensuring the information summarized from the post-call observation aligns with the necessary evidence. Furthermore, we introduce a non-repeated-goal regularizer to penalize redundant calls that revisit satisfied NTEP goals. Extensive evaluations on seven image-grounded benchmarks demonstrate that our 8B-parameter instantiation, NTEP-8B, significantly improves both search-oriented accuracy and tool-use efficiency within a unified three-tool framework. These results highlight the critical value of fine-grained tool-evidence path supervision for training robust agentic VLMs.

Chinese Translation

现代视觉-语言模型（VLMs）能够直接回答许多基于图像的问题，但在处理需要细致视觉细节或外部知识的复杂查询时常常面临挑战。为了获取缺失的证据，代理 VLMs 调用工具，如图像裁剪、图像搜索和文本搜索。然而，现有的训练范式主要基于最终答案的正确性来评估工具使用，导致证据获取和利用的监督不足。这导致了两个关键的缺点：（i）模型经常发出冗余或偏离目标的工具调用，未能收集必要的证据；（ii）即使在调用适当工具时，模型也常常未能从结果观察中提取必要的信息。为了解决这些局限性，我们引入了 NTEP（必要工具-证据路径），这是一种新颖的注释方案，明确规定了每个查询所需的外部证据和相应的工具调用。在此基础上，我们提出了 NTEP-R（NTEP 奖励），这是一种监督机制，确保每次工具调用严格推动推理过程朝向最终解决方案。具体而言，我们的方法奖励代理将其调用前的意图与必要的证据寻求目标对齐，并确保从调用后观察中总结的信息与必要证据一致。此外，我们引入了一种非重复目标正则化器，以惩罚重新访问已满足 NTEP 目标的冗余调用。在七个基于图像的基准上的广泛评估表明，我们的 8B 参数实例 NTEP-8B 在统一的三工具框架内显著提高了搜索导向的准确性和工具使用效率。这些结果突显了细致的工具-证据路径监督在训练强大的代理 VLMs 中的关键价值。

View on arXiv Download PDF AI Translation

cs.AI / 13 / 2609.03494

GrowPage: On-Demand KV Budgeting for Efficient LLM Reasoning Serving

GrowPage：高效 LLM 推理服务的按需 KV 预算管理

Ma, Qiankun, Zhou, Yanjiang, Xiong, Zinan, Wang, Haofei, Song, Zhen, Xiang, Yang, Zhang, Ziyao, Zheng, Hairong

Abstract

Long-output reasoning has made the key--value (KV) cache a critical memory bottleneck for efficient LLM serving. Existing KV compression methods usually rely on a predefined per-request budget and adjust only which KV states are retained, leaving the total capacity fixed throughout decoding. However, reasoning workloads exhibit substantial demand variation: different requests require different KV capacities, and the attention demand of an individual request evolves during generation. We introduce \textbf{GrowPage}, an on-demand KV budgeting framework that treats KV capacity as a runtime resource. GrowPage maintains lightweight dual-timescale query summaries to capture recent and long-term attention behaviors, and uses their relative attention working sets to estimate demand evolution. At each capacity boundary, GrowPage either compresses KV states within the current allocation or acquires an additional physical page when broader demand emerges. By integrating with PagedAttention's page-level memory abstraction, GrowPage preserves continuous batching and prefix caching. Experiments on reasoning benchmarks across multiple models show that GrowPage achieves a superior performance--throughput trade-off over existing approaches.

Chinese Translation

长输出推理使得关键--值 (KV) 缓存成为高效 LLM 服务的一个关键内存瓶颈。现有的 KV 压缩方法通常依赖于预定义的每个请求预算，仅调整保留哪些 KV 状态，而在解码过程中总容量保持不变。然而，推理工作负载表现出显著的需求变化：不同请求需要不同的 KV 容量，且单个请求的注意力需求在生成过程中不断演变。我们提出了 extbf{GrowPage}，一个按需 KV 预算管理框架，将 KV 容量视为运行时资源。GrowPage 维护轻量级的双时间尺度查询摘要，以捕捉近期和长期的注意力行为，并利用它们的相对注意力工作集来估计需求演变。在每个容量边界，GrowPage 要么在当前分配内压缩 KV 状态，要么在更广泛的需求出现时获取额外的物理页面。通过与 PagedAttention 的页面级内存抽象集成，GrowPage 保持了连续批处理和前缀缓存。在多个模型的推理基准测试中的实验表明，GrowPage 在性能与吞吐量的权衡上优于现有方法。

View on arXiv Download PDF AI Translation

cs.AI / 14 / 2609.03503

PPO-STGNN: A Proximal Policy Optimization Approach with Spatio-Temporal Graph Neural Networks for DAG Task Scheduling in Cloud-Edge-End Computing

PPO-STGNN：一种基于空间-时间图神经网络的近端策略优化方法用于云-边缘-端计算中的有向无环图任务调度

Qi, Yangshuo, Wang, Chenwei, Shen, Zihan, Sun, Songlin

Abstract

With the rapid development of the Internet of Things, computation intensive directed acyclic graph (DAG) tasks have become increasingly common in cloud-edge-end collaborative environments. However, cloud, edge, and end nodes are highly heterogeneous in computing capacity, network bandwidth, and energy consumption, which makes the efficient scheduling of tasks with complex dependencies an NP-hard problem. Traditional heuristic algorithms and conventional reinforcement-learning methods often fail to capture the spatio-temporal dynamics of system resources. This paper proposes PPO-STGNN, a DAG task-scheduling algorithm that integrates proximal policy optimization (PPO) with spatio-temporal graph neural networks (STGNNs). The method uses an STGNN to extract features from both the DAG task topology and the physical cloud-edge-end resource graph, and then optimizes the scheduling policy through PPO to minimize makespan and schedule length ratio (SLR) while improving CPU and memory load balancing. To accelerate convergence, a multi-teacher behavior-cloning mechanism is introduced for pretraining. Experimental results show that PPO-STGNN significantly improves load balancing while maintaining a low completion time, making it suitable for dynamic and heterogeneous cloud-edge- end DAG scheduling scenarios.

Chinese Translation

随着物联网的快速发展，计算密集型有向无环图（DAG）任务在云-边缘-端协作环境中变得越来越普遍。然而，云、边缘和端节点在计算能力、网络带宽和能耗方面高度异构，这使得具有复杂依赖关系的任务高效调度成为一个NP难题。传统的启发式算法和常规强化学习方法往往无法捕捉系统资源的时空动态。本文提出了PPO-STGNN，一种将近端策略优化（PPO）与空间-时间图神经网络（STGNNs）相结合的DAG任务调度算法。该方法利用STGNN从DAG任务拓扑和物理云-边缘-端资源图中提取特征，然后通过PPO优化调度策略，以最小化完工时间和调度长度比（SLR），同时改善CPU和内存负载均衡。为加速收敛，引入了一种多教师行为克隆机制进行预训练。实验结果表明，PPO-STGNN在保持低完成时间的同时显著改善了负载均衡，使其适用于动态和异构的云-边缘-端DAG调度场景。

View on arXiv Download PDF AI Translation

cs.AI / 15 / 2609.03515

What Matters for Aggressive Decoding-Time KV Eviction? Temporal Aggregation and Ranking Preservation

影响激进解码时间键值驱逐的因素：时间聚合与排名保持

Zeng, Bo, Zhao, Yu, Liu, Yefeng, Lu, Zhihong, Ni, Xuanfan, Wang, Xintong

Abstract

Decoding-time KV cache compression research focuses heavily on designing better token scoring functions, while the temporal rule that aggregates scores across decode steps is often treated as an implementation detail. Under aggressive KV compression, we find that exponential-moving-average (EMA) aggregation makes approximately order-preserving scorer modifications largely indistinguishable at the eviction-set level. Value-norm and entropy variants remain highly correlated with attention and produce nearly unchanged retention sets, whereas KeyDiff, key norm, recency, and a learned scorer alter the ranking and degrade substantially. We associate this stability with the evaluated aggregation, which couples layer weighting and temporal retention. Building on this observation, we introduce InertiaKV, an EMA-based decoding-time eviction method, and InertiaKV-Lazy, its periodic-refresh variant, which yields 1.34-1.46x decode throughput relative to full refresh InertiaKV. We also study Score-Free decoding as a separate empirical operating point: it scores the full context once at the first decode step, freezes that ranking, and incurs an average quality change of +0.03 while removing all subsequent scoring. Across six open-weight backbones and the LongBench, LongBench-v2, and RULER benchmarks, the results identify temporal aggregation and ranking preservation as distinct, consequential design factors; they do not imply that scoring quality is irrelevant in general.

Chinese Translation

解码时间键值缓存压缩研究主要集中在设计更好的令牌评分函数上，而跨解码步骤聚合评分的时间规则通常被视为实现细节。在激进的键值压缩下，我们发现指数移动平均（EMA）聚合使得在驱逐集层面上，近似保持顺序的评分器修改几乎无法区分。值范数和熵变体与注意力高度相关，并产生几乎不变的保留集，而 KeyDiff、键范数、最近性和学习评分器则改变了排名并显著降级。我们将这种稳定性与评估的聚合联系起来，该聚合结合了层加权和时间保留。基于这一观察，我们引入了 InertiaKV，一种基于 EMA 的解码时间驱逐方法，以及其周期性刷新变体 InertiaKV-Lazy，相较于完全刷新 InertiaKV，其解码吞吐量提高了 1.34-1.46 倍。我们还将无评分解码作为一个独立的经验操作点进行研究：它在第一次解码步骤中对完整上下文进行评分，冻结该排名，并在移除所有后续评分的同时，平均质量变化为 +0.03。在六个开放权重骨干网络以及 LongBench、LongBench-v2 和 RULER 基准测试中，结果表明时间聚合和排名保持是显著的设计因素；它们并不意味着评分质量在一般情况下无关紧要。

View on arXiv Download PDF AI Translation

cs.AI / 16 / 2609.03526

CulturalMenuBench: Probing the Knowledge-Application Gap in Multimodal Culinary Reasoning

CulturalMenuBench：探讨多模态烹饪推理中的知识应用差距

Zeng, Bo, Gao, Linfeng, Lin, Peiqin, Zhao, Yu, Zeng, Mingyan, Tong, Yu, Wang, Xintong, Xu, Linlong, Wang, Longyue, Luo, Weihua, Zhang, Qinggang, Su, Jinsong

Abstract

Multimodal language models achieve near-ceiling scores on food recognition benchmarks, yet it remains unclear whether this success reflects genuine cultural understanding or mere visual matching. To probe this distinction, we introduce CulturalMenuBench, a benchmark of 4,870 items in 10 languages across 18 regions; its 10 tasks pair final-dish and step-by-step cooking images with ingredients, procedural text, and regional labels, spanning basic recognition to process-grounded cultural attribution. Evaluating 12 models exposes a substantial knowledge-application gap: models exceeding 94% on standard multiple-choice tasks drop to at most 56% when attributing dishes to Chinese regional cuisines, despite an identical four-way format. Diagnostic analyses explain why: error patterns are consistent with random guessing, accuracy tracks visual distinctiveness rather than cultural structure, and models classify cuisines more accurately from dish names alone than from images (+7-18 points). The knowledge is thus present but cannot be activated through visual input. An ablation confirms these tasks genuinely require procedural evidence: removing sequential cooking images selectively degrades process-grounded tasks while others remain stable. Overall, CulturalMenuBench shows that near-perfect recognition can conceal an inability to apply cultural knowledge, motivating training that explicitly connects perception, procedure, and cultural context. Code and data are publicly available.

Chinese Translation

多模态语言模型在食品识别基准测试中取得了接近满分的成绩，但尚不清楚这一成功是否反映了真正的文化理解，还是仅仅是视觉匹配。为探讨这一区别，我们引入了CulturalMenuBench，这是一个包含4870个项目的基准，涵盖10种语言和18个地区；其10个任务将最终菜肴和逐步烹饪图像与食材、过程文本和地区标签配对，涵盖从基本识别到基于过程的文化归属。对12个模型的评估揭示了显著的知识应用差距：在标准多项选择任务中超过94%的模型，在将菜肴归属到中国地方菜系时最多下降至56%，尽管格式相同。诊断分析解释了原因：错误模式与随机猜测一致，准确性跟踪视觉特征而非文化结构，模型仅从菜名分类菜系的准确性比从图像中更高（提高7-18个百分点）。因此，知识是存在的，但无法通过视觉输入激活。消融实验确认这些任务确实需要过程证据：去除顺序烹饪图像会选择性地降低基于过程的任务的表现，而其他任务则保持稳定。总体而言，CulturalMenuBench表明，近乎完美的识别可能掩盖了应用文化知识的能力，促使训练明确连接感知、过程和文化背景。代码和数据已公开。

View on arXiv Download PDF AI Translation

cs.AI / 17 / 2609.03527

NeoRed: A Knowledge-Logic-Alignment Multimodal Large Language Model for Neonatal Respiratory Disease Diagnosis

NeoRed：一种用于新生儿呼吸疾病诊断的知识逻辑对齐多模态大型语言模型

Liu, Yinan, Xia, Hongtai, Xu, Haoran, Hong, Jiankang, Song, Jingkuan, Luo, Ye

Abstract

Neonatal respiratory diseases are a major cause of neonatal morbidity and mortality, posing substantial challenges in clinical practice. Despite recent advances, existing Multimodal Large Language Models (MLLMs) face two key limitations in neonatal diagnosis: (1) domain gap arising from predominantly adult training data; (2) insufficient integration of multidimensional clinical context for accurate diagnosis. To address these challenges, we collect two real-world clinical datasets (NeoCXR and NeoCXR-EV) and propose NeoRed, to the best of our knowledge, the first MLLM tailored for neonatal respiratory disease, filling the gap in neonatal diagnostic reports generation. To enhance joint diagnosis from heterogeneous clinical context and chest X-rays, we design a novel Knowledge-Logic-Alignment (KLA) framework which constrains model behavior from three perspectives: 1) Knowledge Prior Injection (KPI) incorporates neonatologist-inspired diagnostic priors into multimodal representations, guiding disease-specific attention across modalities; 2) Diagnostic Logic Constraint (DLC) aligns the semantics of generated reports with multimodal diagnostic logic; and 3) Visual Semantic Alignment (VSA) establishes semantic correspondence between visual features and imaging conclusions. Extensive experiments demonstrate that NeoRed enables accurate neonatal diagnostic reports generation, achieving ROUGE-L of 53.29% and Clinical Efficacy F1 score of 65.19% on NeoCXR, outperforming existing MLLMs. NeoRed also preserves competitive report generation performance on adult benchmarks (MIMIC-CXR and IU-Xray). Datasets will be available upon application.

Chinese Translation

新生儿呼吸疾病是新生儿发病率和死亡率的主要原因，对临床实践带来了重大挑战。尽管最近取得了一些进展，现有的多模态大型语言模型（MLLMs）在新生儿诊断方面面临两个主要限制：（1）由于主要使用成人训练数据而导致的领域差距；（2）对多维临床背景的整合不足，影响准确诊断。为了解决这些挑战，我们收集了两个真实世界的临床数据集（NeoCXR和NeoCXR-EV），并提出了NeoRed，尽我们所知，这是第一个专为新生儿呼吸疾病量身定制的MLLM，填补了新生儿诊断报告生成的空白。为了增强来自异构临床背景和胸部X光片的联合诊断，我们设计了一种新颖的知识逻辑对齐（KLA）框架，从三个方面约束模型行为：1）知识先验注入（KPI）将新生儿科医生启发的诊断先验融入多模态表示中，引导跨模态的疾病特异性注意力；2）诊断逻辑约束（DLC）将生成报告的语义与多模态诊断逻辑对齐；3）视觉语义对齐（VSA）建立视觉特征与成像结论之间的语义对应关系。大量实验表明，NeoRed能够生成准确的新生儿诊断报告，在NeoCXR上实现了53.29%的ROUGE-L和65.19%的临床有效性F1分数，超越了现有的MLLMs。NeoRed在成人基准（MIMIC-CXR和IU-Xray）上也保持了竞争性的报告生成性能。数据集将在申请后提供。

View on arXiv Download PDF AI Translation

cs.AI / 18 / 2609.03535

Feature Reconfiguration With Visual Prior for Medical Lesion Segmentation

基于视觉先验的医学病灶分割特征重构

Liu, Yinan, Hong, Jiankang, Gao, Zhen, Lu, Ye

Abstract

Lesion segmentation in medical images plays a critical role in clinical diagnosis and treatment planning. Despite significant advances, lesion segmentation remains challenging due to two major factors: (1) complex background interference; (2) diverse lesion morphology. Existing encoder-decoder based methods mainly focus on enhancing feature extraction or redesigning decoding strategies. However, they lack early prior guidance and feature reconfiguration during the encoding stage, limiting their effectiveness in handling these challenges. To address these limitations, we propose FreNet, a feature reconfiguration framework with visual priors, which performs pixel-level reconfiguration before encoding and feature-level reconfiguration during encoding for precise medical lesion segmentation. To suppress background responses, we propose an Implicit Prior Neural Network (IPNN), which models a continuous spatial field and leverages visual prior from SAM to reconfigure input image before encoding stage. To better handle diverse lesion morphology, we design a Dual-domain Feature Reconfiguration (DFR) module to progressively reconfigure backbone features during encoding stage. Within DFR, the Frequency Decoupling Module (FDM) decouples backbone features in frequency domain to enhance foreground-background discriminability, while the Spatial Localization Module (SLM) spatially relocates and improving spatial stability after frequency decoupling. Extensive experiments on 9 medical image segmentation benchmarks across three imaging modalities demonstrate that FreNet significantly outperforms state-of-the-art (SOTA) methods. On the challenging ETIS dataset, our method achieves Dice improvements of 5.0% over SOTA method and 7.2% over SAM.

Chinese Translation

医学图像中的病灶分割在临床诊断和治疗规划中发挥着关键作用。尽管取得了显著进展，病灶分割仍然面临两个主要挑战：（1）复杂背景干扰；（2）多样的病灶形态。现有的基于编码器-解码器的方法主要集中在增强特征提取或重新设计解码策略。然而，它们在编码阶段缺乏早期先验指导和特征重构，限制了其在应对这些挑战时的有效性。为了解决这些局限性，我们提出了FreNet，一个具有视觉先验的特征重构框架，在编码之前进行像素级重构，并在编码过程中进行特征级重构，以实现精确的医学病灶分割。为了抑制背景响应，我们提出了一种隐式先验神经网络（Implicit Prior Neural Network, IPNN），该网络建模连续的空间场，并利用来自SAM的视觉先验在编码阶段之前对输入图像进行重构。为了更好地处理多样的病灶形态，我们设计了一个双域特征重构（Dual-domain Feature Reconfiguration, DFR）模块，在编码阶段逐步重构主干特征。在DFR中，频率解耦模块（Frequency Decoupling Module, FDM）在频率域中解耦主干特征，以增强前景-背景的可区分性，而空间定位模块（Spatial Localization Module, SLM）在频率解耦后进行空间重新定位并提高空间稳定性。在三个成像模式下的9个医学图像分割基准上进行的广泛实验表明，FreNet显著优于最先进的方法（SOTA）。在具有挑战性的ETIS数据集上，我们的方法在Dice指标上比SOTA方法提高了5.0%，比SAM提高了7.2%。

View on arXiv Download PDF AI Translation

cs.AI / 19 / 2609.03546

Dalek: A Constructive Agent Machine

Dalek：一种建设性代理机器

Xie, Wanpeng

Abstract

We present Dalek, a closed machine designed for agents that realizes self-maintenance, self-evolution, self-reproduction, and self-organization on any substrate satisfying a general host contract. The machine is built from three primitives---actors, messages, and channels. Four obligations---a host boundary, a construction language, admissible transitions, and rule heredity---give its boundary, identity, and closure a structural basis. Von Neumann's 1948 self-reproducing automaton supplies a hereditary constructional core: a self-description together with a constructor, a copier, and a controller. Dalek combines this core with the four obligations and rederives its medium for a text-and-message agent substrate, adding explicit structures for boundary, identity, history, and growth. A large language model and a compiler occupy the payload position and form a general capability producer. New capabilities are authored, compiled, installed into the description, and inherited by descendants. The same path produces the machine's own organs and even its runtime, closing heredity and evolution within the machine.

Chinese Translation

我们提出了Dalek，一种为代理设计的封闭机器，能够在满足一般宿主合同的任何基质上实现自我维护、自我进化、自我复制和自我组织。该机器由三个基本元素构成——行为者、消息和通道。四项义务——宿主边界、构造语言、可接受的转变和规则遗传——为其边界、身份和封闭性提供了结构基础。冯·诺依曼1948年的自我复制自动机提供了一个遗传构造核心：一个自我描述以及一个构造器、一个复制器和一个控制器。Dalek将这一核心与四项义务相结合，并重新推导其文本和消息代理基质的媒介，为边界、身份、历史和增长添加了明确的结构。一个大型语言模型和一个编译器占据了有效载荷位置，形成了一个通用能力生成器。新的能力被创作、编译、安装到描述中，并由后代继承。同样的路径产生了机器自身的器官，甚至是其运行时，将遗传和进化封闭在机器内部。

View on arXiv Download PDF AI Translation

cs.AI / 20 / 2609.03553

GPS-Bench: A Governance Policy Benchmark for Automating Policy Analysis

GPS-Bench：自动化政策分析的治理政策基准

Le, Linh, Bui, Melanie, Nguyen, My Chiffon, Schlosser, Zachary, Williams-King, David

Abstract

Policy analysis requires more than predicting whether a proposal will pass: it requires identifying who will be affected, how those actors respond, and what follows. LLM-based policy simulations model these processes at scale, but their validity is hard to establish when plausible behaviour is never compared with observed outcomes. We introduce GPS-Bench, an evidence-grounded benchmark for governance policy simulation that links policies to relevant actors, actor actions and downstream impacts using legislative records, lobbying disclosures, regulatory documents, corporate filings, economic data and other public evidence. Actors are reconstructed from the dated record rather than prompted as archetypes, so a persona is an evidence object with provenance; a human-annotated pool forms the Gold evaluation set, while cases labelled by a separate LLM from retrieved evidence are treated as Silver supervision and never as test labels. Because every inference mode reads the same grounded state and emits the same schema, GPS-Bench turns "does multi-agent simulation help?" into a controlled comparison: we contrast joint reasoning, independent and communicating actor agents, graph-based methods and weight-level fine-tuning over one policy state. Fine-tuning on the grounded record gives the strongest actor-level impact prediction, and decomposition does not beat it; what decomposition adds is mechanism. Agents hold private, non-identical evidence, each seeing its own exposure clause, and address named partners with concrete joint proposals, what they offer, what they need in return, and why acting together beats acting alone, so the coalitions that form can be checked against the commitments the record holds. GPS-Bench therefore gives a common empirical setting for studying when evidence, actor modelling and multi-agent interaction improve the prediction and interpretation of policy outcomes.

Chinese Translation

政策分析不仅需要预测提案是否会通过，还需要识别谁将受到影响、这些参与者如何响应以及接下来会发生什么。基于大型语言模型（LLM）的政策模拟能够大规模地模拟这些过程，但当合理的行为从未与观察到的结果进行比较时，其有效性难以建立。我们引入了GPS-Bench，这是一个基于证据的治理政策模拟基准，它通过立法记录、游说披露、监管文件、公司申报、经济数据和其他公共证据，将政策与相关参与者、参与者行为及其下游影响联系起来。参与者是从历史记录中重建的，而不是作为原型被提示，因此一个角色是一个具有来源的证据对象；一个经过人工标注的池形成了金标准评估集，而由从检索证据中获得的独立LLM标记的案例被视为银标准监督，而不是测试标签。由于每种推理模式读取相同的基础状态并发出相同的模式，GPS-Bench将“多智能体模拟是否有帮助？”转变为受控比较：我们对比了联合推理、独立和沟通的参与者代理、基于图的方法以及在一个政策状态上的权重级微调。对基础记录的微调提供了最强的参与者级影响预测，而分解并未超越它；分解所增加的是机制。代理持有私人且不相同的证据，每个代理看到自己的曝光条款，并以具体的联合提案与指定的合作伙伴进行交流，说明他们提供什么、需要什么回报，以及为何共同行动优于单独行动，因此形成的联盟可以与记录所持的承诺进行核对。因此，GPS-Bench为研究何时证据、参与者建模和多智能体互动改善政策结果的预测和解释提供了一个共同的实证环境。

View on arXiv Download PDF AI Translation

cs.AI / 21 / 2609.03580

HalluPeer: A Taxonomy-driven Benchmark for Detecting Hallucinations in Scientific Peer Reviews

HalluPeer：一种基于分类法的基准，用于检测科学同行评审中的幻觉

Lin, Tzu-Ling, Yao, Dong-Ting, Hsiao, Teng-Fang, Chen, Wei-Chih, Shuai, Hong-Han

Abstract

The growing scale of academic peer review has motivated the use of Large Language Models (LLMs) as review assistants, yet LLMs can generate fluent but unsupported claims that undermine review reliability. Existing hallucination benchmarks are not designed for peer review, where verification requires grounding claims in long, technical papers. We introduce HalluPeer, a benchmark for detecting hallucinations in scientific peer reviews, providing aligned triples of paper content, human-written reviews, and hallucination-injected reviews, annotated for detection, classification, and localization. Our pipeline induces a peer-review-specific hallucination taxonomy, identifies review contexts, and injects hallucinations with automated filtering. Experiments on 12K papers and 38K reviews show that existing detectors struggle to separate hallucinations from legitimate critique, while evaluation on authentic reviews demonstrates that HalluPeer-defined hallucination patterns occur in real peer reviews, highlighting the critical need for source-aware verification. Our project page can be found in https://github.com/Lin-TzuLing/HalluPeer.git

Chinese Translation

学术同行评审的规模不断扩大，促使使用大型语言模型（LLMs）作为评审助手，然而，LLMs可能生成流畅但缺乏支持的主张，从而削弱评审的可靠性。现有的幻觉基准并未针对同行评审而设计，因为验证需要将主张与长篇技术论文中的内容相结合。我们提出了HalluPeer，这是一个用于检测科学同行评审中幻觉的基准，提供了论文内容、人类撰写的评审和注入幻觉的评审的对齐三元组，并进行了检测、分类和定位的注释。我们的流程引入了一个特定于同行评审的幻觉分类法，识别评审上下文，并通过自动过滤注入幻觉。在对12K篇论文和38K条评审的实验中，现有的检测器在区分幻觉与合法批评方面表现不佳，而对真实评审的评估表明，HalluPeer定义的幻觉模式确实出现在真实的同行评审中，突显了源意识验证的迫切需求。我们的项目页面可以在https://github.com/Lin-TzuLing/HalluPeer.git找到。

View on arXiv Download PDF AI Translation

cs.AI / 22 / 2609.03586

The Attention Triangle in Audio-Video Models

音视频模型中的注意力三角形

Polaczek, Sagi, Kraicer, Noa, Metzer, Gal, Ning, Zhuo, Mahdavi-Amiri, Ali, Cohen-Or, Daniel, Giryes, Raja

Abstract

Audio-video diffusion models rely on cross-modal attention to coordinate text, sound, and visual content, yet this same mechanism can introduce subtle and systematic semantic leakage. We study these models by probing and analyzing the ``attention triangle,'' comprising the three cross-attention edges connecting the text, audio, and video streams, and examine how semantic information is routed across modalities during generation. Our analysis reveals that routing along the audio-video edge is bidirectional: audio can influence video generation, while video can influence audio generation. This edge is shaped by biases encoded in the model's parameters and emerges as a major contributor to leakage: when prompts are in tension with learned priors, cross-modal interactions may override the intended conditioning and reroute semantics toward visually canonical but incorrect outcomes. These effects suggest that semantic artifacts arise not merely from attention spreading beyond its intended target, but from structured, bias-driven interactions along specific pathways. Building on this perspective, we extract attention-derived signals that expose how semantics are distributed and grounded across modalities, and use them as a diagnostic tool to both analyze and deliberately incur leakage under controlled conditions. This enables us to probe the internal dynamics of cross-modal routing and isolate the role of individual interactions. We further leverage these signals to guide inference-time interventions that encourage more consistent cross-modal alignment. Extensive experiments support our analysis and demonstrate improved semantic grounding while preserving generation quality.

Chinese Translation

音视频扩散模型依赖于跨模态注意力来协调文本、声音和视觉内容，但这一机制也可能引入微妙且系统性的语义泄漏。我们通过探测和分析“注意力三角形”来研究这些模型，该三角形由连接文本、音频和视频流的三条跨注意力边组成，并考察在生成过程中语义信息如何在不同模态之间传递。我们的分析揭示，音频与视频之间的边是双向的：音频可以影响视频生成，而视频也可以影响音频生成。这条边受到模型参数中编码的偏见的影响，并成为泄漏的主要因素：当提示与学习的先验知识存在张力时，跨模态交互可能会覆盖预期的条件，并将语义重新引导到视觉上典型但不正确的结果。这些效应表明，语义伪影的产生不仅仅是由于注意力超出其预期目标的扩散，而是由于沿特定路径的结构化、偏见驱动的交互。基于这一视角，我们提取了源于注意力的信号，揭示了语义如何在不同模态之间分布和扎根，并将其作为诊断工具，以分析和在受控条件下故意引发泄漏。这使我们能够探测跨模态路由的内部动态，并隔离个别交互的作用。我们进一步利用这些信号指导推理时的干预，以促进更一致的跨模态对齐。大量实验支持我们的分析，并在保持生成质量的同时展示了改进的语义扎根。

View on arXiv Download PDF AI Translation

cs.AI / 23 / 2609.03588

KC-Bench: A Dynamic Interactive Benchmark for Evaluating Knowledge Conflicts in LLM Agents

KC-Bench：评估大型语言模型代理中的知识冲突的动态交互基准

Lyu, Yaxing, Zhou, Shengjie, Toh, Binbin, Zhu, Pengyu, Li, Lijun

Abstract

As LLMs increasingly act through tools, they must reconcile user instructions, parametric knowledge, and dynamic environmental observations before taking actions. We introduce KC-Bench, a controlled multi-turn benchmark for measuring this capability across world-knowledge conflicts, input inconsistencies, and multi-source temporal conflicts. Its 238 tasks are manually screened from more than 1,000 generated candidates and combine a user simulator, stateful tools, deterministic environment assertions, an open-source natural-language evaluator, and human trajectory verification. Evaluation of nine models, including DeepSeek-V4-Flash, GLM-5.2, and MiniMax-M3, shows substantial cross-domain variation: no model handles factual correction, identity consistency checking, and temporal conflict resolution reliably across all settings. In the simulated environments, missed conflicts can propagate to tool calls or synthetic protected-data flows. KC-Bench isolates this model-level behavior rather than ranking complete agent frameworks, and provides a reproducible diagnostic for developing conflict-aware reasoning and execution safeguards.

Chinese Translation

随着大型语言模型（LLMs）越来越多地通过工具进行操作，它们必须在采取行动之前调和用户指令、参数知识和动态环境观察。我们介绍了KC-Bench，这是一个受控的多轮基准，用于测量在世界知识冲突、输入不一致性和多源时间冲突方面的这一能力。其238个任务是从超过1,000个生成候选中手动筛选而来，结合了用户模拟器、有状态工具、确定性环境断言、开源自然语言评估器和人类轨迹验证。对包括DeepSeek-V4-Flash、GLM-5.2和MiniMax-M3在内的九个模型的评估显示出显著的跨领域差异：没有一个模型能够在所有设置下可靠地处理事实修正、身份一致性检查和时间冲突解决。在模拟环境中，未解决的冲突可能会传播到工具调用或合成保护数据流。KC-Bench孤立了这种模型级行为，而不是对完整的代理框架进行排名，并提供了一个可重复的诊断工具，以便开发冲突感知的推理和执行保障。

View on arXiv Download PDF AI Translation

cs.AI / 24 / 2609.03621

A computable representation of the physical laboratory enables verifiable workflows

可计算的物理实验室表征实现可验证的工作流程

Li, Xiaobo, Ge, Luyao, Li, Xiaohui, Guo, Lulu, Mao, Ming, Zheng, Jiwang, Guan, Wenting, Yang, Xin, Luo, Yi, Jiang, Jun, Chen, Linjiang

Abstract

Making science computable requires representations of both scientific knowledge and the physical world in which scientific claims are tested. A computable representation of the physical laboratory is established through typed research objects, capability-bound operations and a compositional workflow algebra. It provides the physical-world counterpart to machine-readable knowledge, expressing workflows as programs over evolving laboratory states with explicit dependencies, decisions, iteration and concurrency. The representation was implemented in a modular agentic robotic laboratory by binding formal operations to executable Function Skills. For diverse scientific intents, capability-relative workflows were generated, while stateful simulation propagated object transformations and verified operation preconditions and laboratory constraints before dispatch. The proposed representation and its engineering framework jointly establish a general computational interface between agent reasoning and capability-bound physical transformations, providing a foundation for end-to-end autonomous scientific discovery.

Chinese Translation

使科学可计算需要对科学知识和科学主张被测试的物理世界进行表征。通过类型化研究对象、能力绑定操作和组合工作流程代数，建立了物理实验室的可计算表征。它为机器可读知识提供了物理世界的对应，表达了作为程序的工作流程，涵盖了不断演变的实验室状态及其显式依赖、决策、迭代和并发。该表征在一个模块化的智能机器人实验室中实现，通过将形式操作绑定到可执行的功能技能。针对多样的科学意图，生成了相对能力的工作流程，同时有状态的仿真传播对象变换，并在调度之前验证操作前提条件和实验室约束。所提出的表征及其工程框架共同建立了代理推理与能力绑定物理变换之间的一般计算接口，为端到端的自主科学发现提供了基础。

View on arXiv Download PDF AI Translation

cs.AI / 25 / 2609.03635

Analysis of Prompt Engineering for Drug Toxicity Prediction

药物毒性预测的提示工程分析

MacGregor, Mia, Don, Aakash Welgamage, Bartlett, Mark

Abstract

Clinical trials in the UK can cost up to {\pounds}1.3 million, with approximately 90% drug failure rate. Toxicity is a major contributing factor in drug failure. Testing is time and cost intensive. In recent years, the use of artificial intelligence has been increasingly explored to aid in the prediction of drug toxicity, with extensive use of large language models (LLMs). However, LLMs can show considerable variation when minor changes are made to prompts, which raises concerns about their sensitivity to prompt engineering. Prompt engineering is used to optimise a prompt given to an LLM to generate the desired output. This paper proposes a method to analyse prompt engineering for drug toxicity prediction. The aim of the paper is to investigate the importance of prompt phrasing for drug toxicity prediction. LLMs were prompted to identify chemical properties of significance when predicting drug toxicity. Prompts were constructed to investigate; job role, prompt structuring, and rule interpretation. LLMs were then used to generate datasets, using the identified features from initial prompting, which were then passed to machine learning algorithms. The experiments show that the natural variance which occurs in LLMs outweighs any fine-tuning of prompts. There were, however, substantial improvements in model performance when using chemoinformatic code to extract features instead of using LLM-generated values. The proposed analysis methodology is applicable to a wide range of prompt types across different areas of bioinformatics.

Chinese Translation

在英国，临床试验的费用可高达130万英镑，药物失败率约为90%。毒性是导致药物失败的主要因素之一。测试过程既耗时又昂贵。近年来，人工智能的应用越来越多地被探索用于辅助药物毒性的预测，尤其是大规模语言模型（LLMs）的广泛使用。然而，当对提示进行微小修改时，LLMs可能会表现出显著的变化，这引发了对其对提示工程敏感性的担忧。提示工程用于优化提供给LLM的提示，以生成所需的输出。本文提出了一种分析药物毒性预测中提示工程的方法。论文的目的是研究提示措辞在药物毒性预测中的重要性。LLMs被提示识别在预测药物毒性时重要的化学特性。构建了提示以调查工作角色、提示结构和规则解释。然后，使用LLMs生成数据集，利用初始提示中识别的特征，这些数据集随后被传递给机器学习算法。实验表明，LLMs中自然发生的变异超过了任何对提示的微调。然而，当使用化学信息学代码提取特征而不是使用LLM生成的值时，模型性能有了显著改善。所提出的分析方法适用于生物信息学不同领域的多种提示类型。

View on arXiv Download PDF AI Translation

cs.AI / 26 / 2609.03702

Synthetic Semantic Supervision for Contrastive Code Representation Learning in Small Transformers: An Empirical Study

小型变换器中对比代码表示学习的合成语义监督：一项实证研究

Paulsen, Kenneth, Tambon, Florian, Papadakis, Mike, Yoo, Shin

Abstract

General-purpose code embeddings power tools for code search, classification, and retrieval. Compact transformer encoders for code typically rely on either human-written docstrings (labor-intensive and inconsistent) or mined structural signals such as execution traces (setting-specific and costly to collect). We empirically study an alternative: contrastive pretraining of small encoders with synthetically generated natural-language descriptions emphasizing code functionality and intent, paired with code in a dual-encoder framework at training and discarded at inference. We benchmark this approach against pretraining-based baselines, generalist LLMs, and embedding-specific models on eight retrieval, classification, and generation tasks across C, C++, and Java. Synthetic semantic supervision yields statistically significant gains over pretraining baselines of the same inference-time size on five of eight tasks, with parity on two more; once fine-tuned, it matches or exceeds zero-shot models two orders of magnitude larger on classification, and it stays on par with execution-aware supervision at matched pretraining data, suggesting a scalable, effective alternative to existing code-representation paradigms.

Chinese Translation

通用代码嵌入为代码搜索、分类和检索工具提供支持。用于代码的紧凑变换器编码器通常依赖于人工编写的文档字符串（劳动密集且不一致）或挖掘的结构信号，如执行轨迹（特定于设置且收集成本高）。我们实证研究了一种替代方案：使用合成生成的自然语言描述进行小型编码器的对比预训练，这些描述强调代码的功能和意图，并在训练时与代码配对，在推理时被丢弃。我们将这种方法与基于预训练的基准、通用大型语言模型（LLMs）和特定嵌入模型进行了基准测试，涵盖了 C、C++ 和 Java 的八个检索、分类和生成任务。合成语义监督在八个任务中的五个上相较于相同推理时间大小的预训练基准取得了统计显著的提升，在另外两个任务上则持平；一旦经过微调，它在分类任务上与零样本模型的性能相匹配或超越，且在匹配的预训练数据下与执行感知监督保持一致，表明这是一种可扩展且有效的替代现有代码表示范式的方法。

View on arXiv Download PDF AI Translation

cs.AI / 27 / 2609.03707

Counterfactual Routing Using Integer Programming with Constraint Generation

基于整数规划和约束生成的反事实路径规划

Vos, Daniël, Lutz, Sterre

Abstract

We present our submission to the IJCAI 2025 'Counterfactual Routing Competition' (CRC 25). The goal of the competition is to find counterfactual explanations for the shortest path problem. This requires deciding what the minimal changes to a road network would make a route chosen by the user the optimal route. This enables explanations such as "Your suggested route would indeed have been optimal, if road X were not a bicycle path." Our solution models the problem as an integer program, iteratively incorporating constraints until an exact solution is found. In the final evaluation on held-out test instances, our method ranked fourth in solution quality and obtained its solution fastest on every instance, with an average runtime of 9.0 seconds compared to 118.8 seconds for the next-fastest submission.

Chinese Translation

我们提交了针对 IJCAI 2025 '反事实路径规划竞赛' (Counterfactual Routing Competition, CRC 25) 的方案。该竞赛的目标是为最短路径问题寻找反事实解释。这需要确定对道路网络的最小更改，以使用户选择的路线成为最佳路线。这使得可以提供诸如“如果道路 X 不是自行车道，您建议的路线确实会是最佳路线。”这样的解释。我们的解决方案将该问题建模为一个整数规划，迭代地加入约束，直到找到精确解。在对保留的测试实例进行的最终评估中，我们的方法在解决方案质量上排名第四，并且在每个实例上都以最快的速度获得了解决方案，平均运行时间为 9.0 秒，而下一个最快的提交则为 118.8 秒。

View on arXiv Download PDF AI Translation

cs.AI / 28 / 2609.03716

Artificial Intelligence for Energy Optimization in Data Centers

人工智能在数据中心能源优化中的应用

Ullah, Mohammed Basharath, Begum, Summaiya Unnisa, Ullah, Mohammed Nadeem

Abstract

Data centers are increasingly optimized by artificial intelligence and, at the same time, increasingly loaded by it. The literature treats these as two unrelated problems: control studies model workload as an exogenous arrival process, while sustainability studies model infrastructure as a fixed multiplier. We screen roughly 194 papers retrieved through a documented protocol, code 63 of them, and report what the coding shows. Of 28 primary control-oriented studies, 18 are validated in simulation alone and 5 reach physical hardware or a production facility; none account for water withdrawal, and none account for embodied carbon. Reported savings intervals across four technique families overlap almost completely, which means the field cannot presently rank its own methods. Ten recurring gaps are scored for consequence and tractability, and we set out CLEAR-DC, a framework coupling a control-policy branch to a workload-demand branch through an explicit elasticity term, reads out net rather than direct benefit, and emits a schema-conformant record covering energy, carbon, water, embodied share and validation venue. The framework is an architectural and methodological proposal, not a trained system; the contribution we defend empirically is the corpus analysis and the reporting schema derived from it. Coding sheet, derived statistics and all result artifacts: https://github.com/Kimalice/AI-for-Energy-Optimization-in-Data-Centers-Closing-the-Optimizer-Load-Loop

Chinese Translation

数据中心正越来越多地通过人工智能进行优化，同时也受到其影响而负载增加。文献将这视为两个无关的问题：控制研究将工作负载建模为外生到达过程，而可持续性研究则将基础设施建模为固定乘数。我们通过文档化的协议筛选出大约194篇论文，编码其中63篇，并报告编码结果。28项主要的控制导向研究中，18项仅在模拟中得到验证，5项达到了物理硬件或生产设施；没有一项考虑水资源提取，也没有一项考虑嵌入碳。四个技术类别报告的节能区间几乎完全重叠，这意味着该领域目前无法对自身的方法进行排名。我们对十个反复出现的研究空白进行了后果和可行性评分，并提出了CLEAR-DC框架，该框架通过明确的弹性项将控制政策分支与工作负载需求分支相结合，读取净效益而非直接效益，并生成覆盖能源、碳、水、嵌入份额和验证场所的符合模式的记录。该框架是一个架构和方法论的提案，而非一个训练系统；我们通过实证捍卫的贡献是语料分析及其衍生的报告模式。编码表、衍生统计和所有结果文档请访问：https://github.com/Kimalice/AI-for-Energy-Optimization-in-Data-Centers-Closing-the-Optimizer-Load-Loop

View on arXiv Download PDF AI Translation

cs.AI / 29 / 2609.03727

Proactive Service Agents: A Unified Decision Framework, Methods, and Evaluation

主动服务代理：统一决策框架、方法与评估

Tang, Yan, Cao, Tingyu, Tang, Yuanbo, Tang, Huaze, Hu, Keer

Abstract

Large language model agents can plan, invoke tools, and modify external states, yet most systems still take an explicit user instruction as a fixed starting point. Proactive service moves the decision upstream: an agent must infer service opportunities from incomplete environmental and user signals, choose among remaining silent, asking, assisting, and acting, and account for interruption, misunderstanding, overreach, and privacy costs. This survey gives an operational definition centered on initiative and formulates the problem as a partially observable sequential decision process constrained by authorization and risk. The formulation represents timing, content, and delivery within one structured action, while making explicit the option value of waiting, the decision value of questions, and feedback-induced state changes. On this basis, we organize existing methods along one decision pipeline (state and need estimation, intervention gating, action construction, and feedback adaptation) and describe prescribed, predictive, model based, and return optimizing mechanisms as nonexclusive policy-construction components. We further normalize decision units and three-axis evidence descriptors across streaming dialogue, screen, video, software-engineering, and human-agent collaboration resources, and formalize metrics for triggering, timing, calibration, user burden, safety, and policy value. The synthesis shows why offline classification performance alone does not predict deployment benefit and why long-term memory is not a defining condition of proactivity. Reliable proactive service instead requires calibrated incremental intervention value, verifiable authorization, recoverable execution, and counterfactual evidence.

Chinese Translation

大型语言模型代理能够进行规划、调用工具和修改外部状态，但大多数系统仍将明确的用户指令作为固定的起点。主动服务将决策上移：代理必须从不完整的环境和用户信号中推断服务机会，选择保持沉默、询问、协助或行动，并考虑中断、误解、过度干预和隐私成本。本文提供了一个以主动性为中心的操作性定义，并将问题表述为一个受授权和风险约束的部分可观察序列决策过程。该表述将时机、内容和交付整合为一个结构化的行动，同时明确等待的选项价值、问题的决策价值以及反馈引发的状态变化。在此基础上，我们沿着一个决策流程（状态和需求估计、干预门控、行动构建和反馈适应）组织现有方法，并将规定性、预测性、基于模型和回报优化机制描述为非排他性的政策构建组件。我们进一步规范了在流媒体对话、屏幕、视频、软件工程和人机协作资源中的决策单元和三轴证据描述符，并形式化了触发、时机、校准、用户负担、安全性和政策价值的度量。综合分析表明，仅凭离线分类性能无法预测部署收益，长期记忆也不是主动性的决定性条件。可靠的主动服务需要经过校准的增量干预价值、可验证的授权、可恢复的执行和反事实证据。

View on arXiv Download PDF AI Translation

cs.AI / 30 / 2609.03753

SimSkill: A Lifelong Learning AI Agent for Autonomous Mastery of Traffic Simulation

SimSkill：一个用于自主掌握交通仿真的终身学习人工智能代理

Liu, Qi, Wang, Qinzheng, Bie, Yiming

Abstract

As large language models (LLMs) become increasingly capable, the long-term value of AI systems depends not only on solving individual requests, but also on transforming experience and accumulated knowledge into durable, reusable competence. We introduce SimSkill, a self-evolving agent built around the Simulation of Urban MObility (SUMO) traffic simulator. SimSkill identifies capability gaps, generates and solves environment-grounded tasks, verifies solutions through an action--critic loop, and consolidates experience into episodic, procedural, and semantic memory without updating the backbone model. Through autonomous exploration, it builds a reusable library spanning the traffic-simulation workflow. We evaluate SimSkill on two held-out benchmarks with three backbone LLMs and independent artifact-based verification. SimSkill improves verified completion by up to 25 percentage points, while ablations show complementary contributions from procedural and semantic memory. Its benefits remain backbone- and budget-dependent: memory does not improve every model or uniformly reduce inference cost. More broadly, SimSkill illustrates a design paradigm in which natural language preserves and composes computational capabilities, while executable tools and code provide precise and reproducible execution. All code and experimental data are publicly available at https://github.com/qiliuchn/SimSkill-V1.

Chinese Translation

随着大型语言模型（LLMs）能力的不断增强，人工智能系统的长期价值不仅依赖于解决个别请求，还依赖于将经验和积累的知识转化为持久的、可重用的能力。我们介绍了SimSkill，一个围绕城市交通仿真（Simulation of Urban MObility, SUMO）交通仿真器构建的自我进化代理。SimSkill识别能力差距，生成并解决基于环境的任务，通过行动-评论循环验证解决方案，并将经验整合为情节记忆、程序记忆和语义记忆，而无需更新基础模型。通过自主探索，它建立了一个涵盖交通仿真工作流程的可重用库。我们在两个保留基准上评估SimSkill，使用三种基础LLM和独立的基于工件的验证。SimSkill提高了经过验证的完成率，最多可达25个百分点，而消融实验显示程序记忆和语义记忆的互补贡献。其效益依赖于基础模型和预算：记忆并不改善每个模型，也不均匀降低推理成本。更广泛地说，SimSkill展示了一种设计范式，其中自然语言保留并组合计算能力，而可执行工具和代码提供精确且可重复的执行。所有代码和实验数据均可在 https://github.com/qiliuchn/SimSkill-V1 上公开获取。

View on arXiv Download PDF AI Translation

cs.AI / 31 / 2609.03774

Rethinking World Models for Safety-Critical Embodied Systems

重新思考安全关键型具身系统的世界模型

Ma, Kailang, Huang, Heye, Kim, Inhi, Jang, Kitae

Abstract

World models have progressed from compact latent dynamics to generative, controllable, and interactive simulators of embodied environments. However, high predictive likelihood and visual fidelity do not necessarily ensure that a model preserves the evidence required for safe decision-making. This perspective identifies three structural mismatches in current world modeling: likelihood versus risk, prediction versus intervention, and finite-horizon prediction versus accumulated consequences. We propose the Risk-Informed World Model (RIWM) as a decision-centric research direction for safety-critical embodied systems. RIWM organizes world modeling around consequences, intervention, epistemic uncertainty, and recoverability, and integrates four interdependent capabilities: decision-relevant representation, counterfactual reasoning, safety-critical episodic memory, and runtime safety assurance. It distinguishes physical, social, and operational consequences while using epistemic uncertainty to qualify the evidence supporting action. We further discuss open challenges in identifying consequential futures, validating counterfactual reasoning, maintaining revisable safety memories, translating learned consequences into executable constraints, and determining when evidence is sufficient to act. This perspective argues that future world models should move beyond predicting likely futures toward identifying which futures matter, revising judgments through experience, and recognizing when to act, revise, sense, defer, or abstain.

Chinese Translation

世界模型已经从紧凑的潜在动态发展为生成式、可控的、互动的具身环境模拟器。然而，高预测似然性和视觉逼真度并不一定确保模型保留安全决策所需的证据。本文指出当前世界建模中的三种结构性不匹配：似然性与风险、预测与干预、有限时间预测与累积后果。我们提出风险知情世界模型（Risk-Informed World Model, RIWM）作为安全关键型具身系统的以决策为中心的研究方向。RIWM围绕后果、干预、认知不确定性和可恢复性组织世界建模，并整合四种相互依赖的能力：与决策相关的表征、反事实推理、安全关键的情景记忆和运行时安全保障。它区分物理、社会和操作后果，同时利用认知不确定性来限定支持行动的证据。我们进一步讨论了识别重要未来、验证反事实推理、维护可修订的安全记忆、将学习到的后果转化为可执行约束以及确定何时证据足够以采取行动等开放挑战。本文认为，未来的世界模型应超越预测可能的未来，转向识别哪些未来重要，通过经验修正判断，并认识何时采取行动、修正、感知、推迟或放弃。

View on arXiv Download PDF AI Translation

cs.AI / 32 / 2609.03787

DNative-Twin: Decision Graphs and Digital Twins for Reconstructable Agentic Decisions

DNative-Twin：用于可重构代理决策的决策图与数字双胞胎

Pang, Junjie, Xie, Zhenzhen, Han, Haoke, He, Ying, Wang, Jing, Liu, Gang

Abstract

AI agents increasingly gather evidence, invoke tools, apply constraints, and produce decisions that people or software may commit to action. A final output alone cannot show which evidence, tool state, rule, authorization, or action path produced it. We present DNative-Twin, a graph-native digital twin that records a committed agentic decision as a typed trajectory and re-executes its decision mechanism under declared conditions. The graph links the state observed by the agent, the path it followed, and the authority behind the resulting action. The twin synchronizes this information, replays the mechanism in isolation, and compares it under controlled changes. We instantiate the framework in enterprise decision processes using three public process logs and controlled replay suites. The experiments identify a specific failure: graph structure localizes represented changes but cannot determine the consequence of an unobserved tool state. In a three-condition controlled experiment with 300 injected instances, unresolved-divergence recall increased from 0 to 0.667 when replay-contract state was added and to 1.0 when verification results were also available; the held-out set contained no critical-class instance. Across 500--5,000 BPI 2020 cases, median end-to-end time increased from 0.794 to 8.889 seconds on the reported platform. These results separate the roles of graph structure, replay context, and verification evidence in reviewing a decision mechanism.

Chinese Translation

人工智能代理越来越多地收集证据、调用工具、应用约束，并产生人类或软件可能执行的决策。单一的最终输出无法显示产生该输出的证据、工具状态、规则、授权或行动路径。我们提出了DNative-Twin，一种图形原生数字双胞胎，它将一个已承诺的代理决策记录为一个类型化轨迹，并在声明的条件下重新执行其决策机制。该图连接了代理观察到的状态、其遵循的路径以及导致结果行动的权威。双胞胎同步这些信息，在隔离环境中重放机制，并在受控变化下进行比较。我们在企业决策过程中实例化该框架，使用三个公共过程日志和受控重放套件。实验识别出一个特定的失败：图结构局限于表示变化，但无法确定未观察到的工具状态的后果。在一个包含300个注入实例的三条件受控实验中，当重放合约状态被添加时，未解决分歧的召回率从0增加到0.667，当验证结果也可用时，召回率增加到1.0；被保留的集合中没有关键类别实例。在500到5000个BPI 2020案例中，报告平台的中位端到端时间从0.794秒增加到8.889秒。这些结果区分了图结构、重放上下文和验证证据在审查决策机制中的角色。

View on arXiv Download PDF AI Translation

cs.AI / 33 / 2609.03797

Transfiver: Human-AI Co-Inference through a Shared Editable State

Transfiver：通过共享可编辑状态实现人机协同推理

Park, Minji, Yoon, Seunghyun, Lim, Hyuk

Abstract

Long-term human-AI interaction is difficult because the information that guides inference is updated implicitly by the model and is not directly inspectable or controllable by the user. We introduce the TRANSparent Framework for Interactive, Verifiable, Editable Representation (Transfiver), an architecture for human-AI co-inference through a shared editable state. Its central idea is that interaction-specific information is maintained in a single persistent state $(S_t)$ that both the model and the human update. Transfiver distinguishes two modes of state evolution. In an implicit stream update, the model interprets ongoing interaction and decides whether new information revises an existing state item or creates a new one. In an explicit directed edit, a human inspects and modifies an addressed item. Both act on the same underlying state, so a human correction changes the state that subsequent computation reads, rather than adding another instruction or separate record. The architecture separates shared parameters $(\theta)$, learned before ordinary use, from the persistent state $(S_t)$, which evolves during deployment without parameter retraining. Extending Transfiver to rich natural-language, relational, and large-scale shared states remains open.

Chinese Translation

长期的人机交互面临挑战，因为指导推理的信息由模型隐式更新，用户无法直接检查或控制。我们提出了透明框架用于交互、可验证、可编辑表示（Transfiver），这是一种通过共享可编辑状态实现人机协同推理的架构。其核心思想是，特定于交互的信息保存在一个持久的状态 $(S_t)$ 中，该状态由模型和人类共同更新。Transfiver 区分了两种状态演变模式。在隐式流更新中，模型解释正在进行的交互，并决定新信息是修订现有状态项还是创建新项。在显式定向编辑中，人类检查并修改指定项。两者都作用于相同的基础状态，因此人类的修正会改变后续计算读取的状态，而不是添加另一个指令或单独记录。该架构将共享参数 $( heta)$（在普通使用前学习）与持久状态 $(S_t)$ 分开，后者在部署过程中演变而无需参数再训练。将 Transfiver 扩展到丰富的自然语言、关系和大规模共享状态仍然是一个开放问题。

View on arXiv Download PDF AI Translation

cs.AI / 34 / 2609.03800

Govern the Model, Not Only the Data: Storage, Circulation, and Learning in Creative AI

治理模型，而不仅仅是数据：创意人工智能中的存储、流通与学习

Perry, Phoenix, Simms, George, Wilson, Elizabeth, Boudiaf, Yasmine, Bryan-Kinns, Nick, Brain, Tega, DuBois, R. Luke, Rule, Alix, Smith, Rachel Meade, Nichole, Kelani, Pawar, Atharva Pravin, Fiebrink, Rebecca

Abstract

Federated learning is increasingly presented as a privacy-preserving advance: personal data remain on the device, and only model updates are shared. It borrows the vocabulary of the federated social web, yet inverts its logic, distributing computation while the resulting model stays with whoever convened the training. We argue that federation is not in itself a remedy for extractive AI, because outcomes depend on who governs the data and the model and who has agency over the practices that shape them. We describe three layers at which a creative community can hold its work: storage, circulation, and learning. Examining artist-governed trusts, cooperatives, and consent infrastructures, we show that creator governance is established at storage and circulation but stops at learning: contributors can consent to training, yet have little say over the resulting model or its federation. We map the research space this opens, pairing technical open problems with the human questions from which they unfold. We propose four design principles for a creative data commons that governs models and their federation, not only datasets: govern the model, not only the corpus; make the terms legible at the moment of contribution; design for refusal as a first-class state; and decide stewardship in the open and account for it.

Chinese Translation

联邦学习越来越被视为一种保护隐私的进步：个人数据保留在设备上，只有模型更新被共享。它借用了联邦社交网络的词汇，但反转了其逻辑，分散了计算，而生成的模型则留在召集训练的主体手中。我们认为，联邦本身并不是解决剥削性人工智能的良方，因为结果取决于谁治理数据和模型，以及谁对塑造这些实践拥有权力。我们描述了创意社区可以管理其工作的三个层面：存储、流通和学习。通过考察艺术家治理的信托、合作社和同意基础设施，我们展示了创作者治理在存储和流通层面建立，但在学习层面停滞：贡献者可以同意训练，但对生成的模型或其联邦几乎没有发言权。我们描绘了这一领域的研究空间，将技术性开放问题与由此展开的人文问题相结合。我们提出了四个设计原则，以治理模型及其联邦，而不仅仅是数据集：治理模型，而不仅仅是语料库；在贡献时使条款清晰可见；将拒绝设计为一种优先状态；以及在公开场合决定管理权并对此负责。

View on arXiv Download PDF AI Translation

cs.AI / 35 / 2609.03806

SVG-Score: Human-Aligned Evaluation of Text-to-SVG Generation

SVG-Score：人类对齐的文本到SVG生成评估

Cipriano, Marco, Zini, Leonardo, Schild, Alexandra, Teutschbein, Valentin, Mimi, Afsana, Cornia, Marcella, Baraldi, Lorenzo, de Melo, Gerard

Abstract

Scalable Vector Graphics (SVG) generation is attracting increasing attention as generative models improve in expressiveness and controllability. Progress, however, is held back by the lack of domain-specific evaluation protocols: current practice relies on metrics designed for natural images, most notably CLIPScore, which was never trained on vector graphics and aligns only partially with human judgment. We introduce \textbf{\ours}, a human-aligned evaluation framework for text-to-SVG generation. Through controlled caption and image perturbations, we first show that CLIP-based scores barely react to the errors SVG generators actually make, such as wrong colors, counts, and spatial relations, and that off-the-shelf Vision-Language Model (VLM) judges, while more sensitive, respond unevenly across error types and SVG styles. We then introduce a human-annotated dataset for \textit{Semantic Alignment}, measuring how faithfully a generated SVG reflects its caption. Building on it, we develop two complementary evaluators: CLIP scorers adapted to vector graphics and then aligned to human preferences, for fast large-scale evaluation, and a VLM judge trained with supervised fine-tuning and reward-shaped reinforcement learning, for more expressive and interpretable assessment. Using both, we benchmark major open-source, commercial, and optimization-based SVG generators on an independent caption set.

Chinese Translation

可缩放矢量图形（SVG）生成正受到越来越多的关注，因为生成模型在表现力和可控性方面不断提高。然而，进展受到缺乏特定领域评估协议的制约：当前的实践依赖于为自然图像设计的指标，尤其是CLIPScore，该指标从未在矢量图形上进行过训练，并且与人类判断的对齐程度仅为部分。我们提出了 extbf{ extit{SVG-Score}}，一个用于文本到SVG生成的人类对齐评估框架。通过对标题和图像进行控制扰动，我们首先展示了基于CLIP的评分几乎对SVG生成器实际犯下的错误（如错误的颜色、数量和空间关系）没有反应，而现成的视觉语言模型（VLM）评估者虽然更敏感，但对错误类型和SVG风格的响应不均匀。随后，我们引入了一个人类注释的数据集，用于 extit{语义对齐}，测量生成的SVG在多大程度上忠实于其标题。基于此，我们开发了两个互补的评估器：适应于矢量图形并与人类偏好对齐的CLIP评分器，用于快速大规模评估，以及经过监督微调和奖励形状强化学习训练的VLM评估者，用于更具表现力和可解释性的评估。通过使用这两者，我们在一个独立的标题集上对主要的开源、商业和基于优化的SVG生成器进行了基准测试。

View on arXiv Download PDF AI Translation

cs.AI / 36 / 2609.03818

CauseCollab: Causal Unified and Modality-Agnostic Network for Heterogeneous Collaborative Perception

CauseCollab：用于异构协同感知的因果统一和模态无关网络

Li, Weize, Li, Yang, Yuan, Quan, Fu, Xiaoyuan, Luo, Guiyang, Li, Jinglin

Abstract

Collaborative perception enhances environment understanding through multi-agent information sharing, but its performance in real-world scenarios is constrained by heterogeneous sensor modalities and model architectures. Recent protocol-based two-stage methods alleviate this problem by mapping heterogeneous features into a shared protocol space; however, independently trained modality-specific converters often generate modality-specific pseudo-protocol distributions, leading to semantic inconsistency and error accumulation, which is particularly pronounced in scenarios with large modality discrepancies. To address this issue, we propose CauseCollab, a causal unified and modality-agnostic network. CauseCollab formulates representation learning in the protocol space from a causal perspective, explicitly disentangling semantic factors from modality-specific statistical confounders via causal metric learning. Meanwhile, CauseCollab adopts context-guided Unified Converter for heterogeneous modalities to ensure cross-modal semantic consistency. In addition, integrating new modalities only requires training adapters with minimal parameters. Extensive experiments on the OPV2V and DAIR-V2X datasets demonstrate that CauseCollab achieves state-of-the-art performance, with more significant gains in scenarios involving large modality gaps.

Chinese Translation

协同感知通过多智能体信息共享增强环境理解，但其在现实场景中的表现受到异构传感器模态和模型架构的限制。近期基于协议的两阶段方法通过将异构特征映射到共享协议空间来缓解这一问题；然而，独立训练的模态特定转换器通常生成模态特定的伪协议分布，导致语义不一致和错误累积，这在模态差异较大的场景中尤为明显。为了解决这一问题，我们提出了CauseCollab，一种因果统一和模态无关的网络。CauseCollab从因果角度对协议空间中的表示学习进行公式化，明确地将语义因素与模态特定的统计混杂因素通过因果度量学习进行解耦。同时，CauseCollab采用上下文引导的统一转换器处理异构模态，以确保跨模态语义一致性。此外，集成新模态只需训练具有最小参数的适配器。在OPV2V和DAIR-V2X数据集上的大量实验表明，CauseCollab实现了最先进的性能，在涉及大模态差距的场景中获得了更显著的提升。

View on arXiv Download PDF AI Translation

cs.AI / 37 / 2609.03834

Semantic Bayesian World Models

语义贝叶斯世界模型

Soru, Tommaso

Abstract

Knowledge graphs describe reality in crisp assertions, while the systems now consuming them, foundation models and autonomous agents, reason natively in probabilities. We argue that this mismatch is why the integration of language models and knowledge graphs remains a data-feeding pipeline rather than a unified reasoning architecture. We envision Semantic Bayesian World Models (SBWMs): a Web that describes the world not as a database of facts but as a shared, evolving fabric of beliefs over knowledge graphs, where ontological axioms constrain priors, observations update beliefs by Bayesian conditioning, and actions intervene upon the world. We work through what an agent gains from such a model: a home-security agent deciding whether the figure at the gate is a courier or a burglar, an actuarial estimate aggregated by entailment rather than by string frequency, a planning task that language models reliably fail, and the estimation of quantities that no document has ever stated. We then set out what the community must build to make them possible: belief annotation over RDF~1.2, probabilistic entailment regimes, semantic calibration layers, and protocols by which agents that have never met can exchange, and disagree over, calibrated beliefs.

Chinese Translation

知识图谱通过清晰的断言描述现实，而当前使用这些知识图谱的系统，如基础模型和自主智能体，则以概率为基础进行推理。我们认为，这种不匹配是语言模型与知识图谱的整合仍然只是一个数据输入管道，而非统一推理架构的原因。我们设想了语义贝叶斯世界模型（Semantic Bayesian World Models, SBWMs）：一个描述世界的网络，不是作为事实的数据库，而是作为一个共享的、不断演变的信念结构，基于知识图谱，其中本体公理约束先验，观察通过贝叶斯条件更新信念，而行动则对世界产生影响。我们探讨了智能体从这样的模型中获得的收益：一个家庭安全智能体决定门口的身影是快递员还是入侵者，通过蕴涵而非字符串频率聚合的精算估计，一个语言模型可靠地失败的规划任务，以及对任何文档从未陈述过的数量的估计。接着，我们阐明了社区必须构建的内容，以使这些模型成为可能：基于RDF 1.2的信念注释、概率蕴涵机制、语义校准层，以及从未见面的智能体之间可以交换和对校准信念产生分歧的协议。

View on arXiv Download PDF AI Translation

cs.AI / 38 / 2609.03860

Adapting to Evolving Requirements: Agentic AI for Retail Supply Chain Operations

适应不断变化的需求：面向零售供应链运营的自主人工智能

Zheng, Lei, Yang, Liping, Li, Zihao, Lyu, Guodong, Koh, Chaik Ming, Teo, Chung-Piaw

Abstract

Retail supply chain operations rely on coupled decision modules that must adapt as requirements evolve. LLMs offer a natural-language interface for this task, but existing methods primarily focus on individual optimization models. Extending them to heterogeneous decision pipelines is challenging because a requirement may admit multiple intervention paths with different downstream effects. We formulate requirement-driven adaptation as the joint selection of an intervention route and an admissible module-level change, and propose a graph-constrained agentic framework in which domain agents expose admissible reformulation interfaces and a central processor searches over bounded intervention paths. Candidates are validated and compared using downstream KPIs. In collaboration with a large retail partner, we evaluate 100 warehouse requirements elicited from practitioner interviews, with GPT, Qwen, and DeepSeek as base LLMs. Relative to direct LLM reformulation, our framework improves correctness and end-to-end success across all three models, raising end-to-end success from 72--76% to 79--83%.

Chinese Translation

零售供应链运营依赖于必须随着需求演变而调整的耦合决策模块。大型语言模型（LLMs）为此任务提供了自然语言接口，但现有方法主要集中于单一优化模型。将其扩展到异构决策流程是具有挑战性的，因为一个需求可能允许多条干预路径，并且这些路径具有不同的下游影响。我们将需求驱动的适应性形式化为干预路径和可接受模块级变化的联合选择，并提出了一种图约束的自主框架，其中领域代理暴露可接受的重构接口，而中央处理器则在有限的干预路径上进行搜索。候选方案通过下游关键绩效指标（KPIs）进行验证和比较。在与一家大型零售合作伙伴的合作中，我们评估了从实践者访谈中提取的100个仓库需求，以GPT、Qwen和DeepSeek作为基础LLM。相较于直接的LLM重构，我们的框架在所有三个模型中提高了正确性和端到端成功率，将端到端成功率从72-76%提升至79-83%。

View on arXiv Download PDF AI Translation

cs.AI / 39 / 2609.03871

Bioinfoysis Technical Report

Bioinfoysis 技术报告

Shao, Qingyang, Zhang, Xin, Yuan, Zhouyang, Chen, Xianying, Xiang, Yujia, Yang, Zihao, Ye, Tong, Zhang, Yangqi, Xu, Jiakang, Yan, Xiaoqing, Luo, Xuan, Li, Keyi, Fan, Enci, Kang, Kai, Liu, Zhuohan, Jin, Xingyu, Teng, Chunran, Li, Tao, Lv, Xinyu, Wang, Minghui, Li, Wenfeng, Gao, Yidan, Liu, Siyu, Luo, Mingrui, Liang, Zhu, Qiao, Guanren, Xu, Zhiping

Abstract

Large language model agents have shown promise in bioinformatics, but most existing systems focus primarily on producing final answers, treating planning, tool use, and code execution as transient interactions. This design is poorly suited to long-horizon bioinformatics tasks, where conclusions must remain connected to the data, computations, and intermediate evidence that support them. We introduce \textbf{Bioinfoysis}, a multi-agent harness that represents each request as a persistent, artifact-grounded analysis run. Bioinfoysis combines global planning with step-wise, evidence-driven replanning: the planner maintains an executable checklist and revises pending steps using structured handoffs returned after each worker execution. These handoffs bind intermediate results to their responsible agent, checklist step, and plan generation, preventing stale evidence from being silently reused after replanning. A controlled runtime validates generated scripts, tables, and figures before they are used in downstream analysis or reporting, while role-specific context, persistent memory, and governed bioinformatics skills support reliable execution over long analysis trajectories. We evaluate Bioinfoysis on BixBench and two question-answering tracks of LAB-Bench 2. On BixBench, Bioinfoysis achieves state-of-the-art accuracy of 82.4\%. Across four underlying language models, Bioinfoysis increases average accuracy from 27.81\% to 64.13\% on SeqQA2 and from 3.13\% to 31.25\% on DbQA2. These results demonstrate that reliable bioinformatics automation depends not only on model capability, but also on the harness that governs planning, execution, memory, and evidence flow. We hope that the emergence of Bioinfoysis will play a driving and leading role in the development of the bioinformatics community. Our demo website can be seen in https://report.bioinfoysis.com/.

Chinese Translation

大型语言模型代理在生物信息学中展现了良好的前景，但大多数现有系统主要集中于生成最终答案，将规划、工具使用和代码执行视为短暂的交互。这种设计不适合长期的生物信息学任务，因为结论必须与支持它们的数据、计算和中间证据保持联系。我们介绍了 extbf{Bioinfoysis}，一个多代理框架，将每个请求表示为一个持久的、基于文献的分析运行。Bioinfoysis 结合了全局规划与逐步的、基于证据的重新规划：规划者维护一个可执行的检查清单，并在每个工作者执行后使用结构化的交接信息修订待处理步骤。这些交接信息将中间结果绑定到其负责的代理、检查清单步骤和计划生成，防止在重新规划后过时的证据被静默重用。受控的运行时在生成的脚本、表格和图形用于下游分析或报告之前进行验证，而角色特定的上下文、持久记忆和受控的生物信息学技能支持在长期分析轨迹上的可靠执行。我们在 BixBench 和 LAB-Bench 2 的两个问答轨道上评估了 Bioinfoysis。在 BixBench 上，Bioinfoysis 达到了 82.4\% 的最先进准确率。在四个基础语言模型中，Bioinfoysis 将 SeqQA2 的平均准确率从 27.81\% 提高到 64.13\%，将 DbQA2 的平均准确率从 3.13\% 提高到 31.25\%。这些结果表明，可靠的生物信息学自动化不仅依赖于模型能力，还依赖于管理规划、执行、记忆和证据流的框架。我们希望 Bioinfoysis 的出现能够在生物信息学社区的发展中发挥推动和引领作用。我们的演示网站可以在 https://report.bioinfoysis.com/ 查看。

View on arXiv Download PDF AI Translation

cs.AI / 40 / 2609.03874

STAIR (STructure Aware Information Retriever): A novel dataset and LLM based retriever for document structure augmentation

STAIR（结构感知信息检索器）：一种基于新数据集和大型语言模型的文档结构增强检索器

Kumar, Vineet, Pulivarthi, Meghanadh, kumar, vishwajeet, Sen, Jaydeep, Bhat, Riyaz Ahmad, Joshi, Sachindra

Abstract

Retrieval Augmented Generation (RAG) is a key component for generating accurate and hallucination free answers using Large Language Models (LLMs). LLMs are improving at handling long context, but still suffer from "lost in the middle" problem. Thus, precise and accurate retrieval is important. Current retrievers chunk long context into length-based manageable chunks - in the process throwing away rich and informative semantic global structure in the corpus. We introduce a novel retrieval system STAIR that empowers an LLM to exploit global structure in a corpus such as a Table of Contents (ToC) to efficiently store and retrieve information from its model parameters. Our thorough and careful ablation studies with a finetuned Differentiable Search Index (DSI) system show that ToC helps build a low hallucination (less than 0.05%) generative Information Retrieval (IR) system and can generalize to examples where very few training samples are available. To further research in this novel direction of ToC based retrieval we release SearchTome - a diverse benchmark created from 18 books across 6 diverse domains to further research in this novel direction. STAIR achieves a high Recall@1 score of 82.6% on SearchTome as compared to DSI (76.9%), where the difference is found to be statistically significant. STAIR easily beats other strong baselines such as BM25 (59.5%), DPR (68.7%) and out-of-the-box Mistral (13.8%).

Chinese Translation

检索增强生成（RAG）是使用大型语言模型（LLMs）生成准确且无幻觉答案的关键组成部分。虽然LLMs在处理长上下文方面有所改善，但仍然面临“中间迷失”问题。因此，精确和准确的检索显得尤为重要。目前的检索器将长上下文分块为基于长度的可管理块，在此过程中丢弃了语料库中丰富且信息量大的全局语义结构。我们提出了一种新型检索系统STAIR，使LLM能够利用语料库中的全局结构，例如目录（ToC），以高效地存储和检索其模型参数中的信息。我们与微调的可微搜索索引（DSI）系统进行了全面且细致的消融研究，结果表明，目录有助于构建低幻觉（低于0.05%）的生成信息检索（IR）系统，并能够推广到训练样本极少的示例。为了进一步研究这一基于目录的检索新方向，我们发布了SearchTome——一个由6个不同领域的18本书籍创建的多样化基准，以促进这一新方向的研究。与DSI（76.9%）相比，STAIR在SearchTome上达到了82.6%的高Recall@1得分，且这一差异具有统计学意义。STAIR轻松超越了其他强基线，如BM25（59.5%）、DPR（68.7%）和开箱即用的Mistral（13.8%）。

View on arXiv Download PDF AI Translation

cs.AI / 41 / 2609.03880

Xiaomi-TabLDM: A Tabular Foundation Model Technical Report

小米-TabLDM：表格基础模型技术报告

TabLDM Team, Wang, Penghui, Liu, Wei, Wang, Hong, Huang, Chengyue, Sun, Yuxi, Wang, Zirui, Huang, Hongming, Wang, Quan, Liu, Chunxiao, Meng, Erli, Wang, Bin

Abstract

We introduce Xiaomi-TabLDM, a tabular large data foundation model for classification and regression via in-context learning, which delivers superior prediction accuracy without requiring task-specific fine-tuning. Pretrained exclusively on synthetic data generated from structural causal models (SCMs), our model enables more flexible context utilization and more efficient capacity scaling. i) A new performance standard. Strong regression performance across benchmarks: Xiaomi-TabLDM ranks 1st on OpenML-CTR23 and 2nd on regression across TALENT, TabArena, and BCCO, demonstrating consistently strong regression performance across four complementary benchmark suites. Favorable performance--efficiency trade-off: Xiaomi-TabLDM combines strong predictive performance with substantially lower computational cost. For example, on TabArena regression, it achieves the second-highest Elo while using 82% less training time and 68% less prediction time than the top-ranked TabFM. ii) Large-scale synthetic pretraining. Xiaomi-TabLDM expands the coverage and diversity of synthetic tabular data used for pretraining. We also adopt a three-stage training strategy together with dual-stream feature grouping, lightweight Attention Residual, and sparse Mixture-of-Experts, enabling Xiaomi-TabLDM to learn richer feature interactions and expert specialization across diverse tabular tasks. iii) Test-time scaling. Xiaomi-TabLDM further extends tabular prediction through test-time compute scaling, where allocating additional computation at inference time consistently improves predictive performance over the base model.

Chinese Translation

我们介绍了小米-TabLDM，一种用于分类和回归的表格大数据基础模型，通过上下文学习实现卓越的预测准确性，无需特定任务的微调。该模型仅在由结构因果模型（SCMs）生成的合成数据上进行预训练，使其能够更灵活地利用上下文并更高效地扩展能力。i) 新的性能标准。在各基准测试中表现出色的回归性能：小米-TabLDM在OpenML-CTR23上排名第一，在TALENT、TabArena和BCCO的回归任务中排名第二，展示了在四个互补基准套件中一致的强大回归性能。良好的性能-效率权衡：小米-TabLDM将强大的预测性能与显著较低的计算成本相结合。例如，在TabArena回归任务中，它在使用比排名第一的TabFM少82%的训练时间和68%的预测时间的情况下，达到了第二高的Elo分数。ii) 大规模合成预训练。小米-TabLDM扩展了用于预训练的合成表格数据的覆盖范围和多样性。我们还采用了三阶段训练策略，结合双流特征分组、轻量级注意力残差和稀疏专家混合模型，使小米-TabLDM能够在多样的表格任务中学习更丰富的特征交互和专家专业化。iii) 测试时扩展。小米-TabLDM通过测试时计算扩展进一步提升了表格预测能力，在推理时分配额外计算资源始终能提高相较于基础模型的预测性能。

View on arXiv Download PDF AI Translation

cs.AI / 42 / 2609.03883

Inferring Affective Consciousness in an Artificial Agent: A Case Study

推断人工智能体的情感意识：案例研究

Solms, Mark, Grimbly, St John, Bassett, Bruce, Boonstra, Evert, Hodson, Rowan, Kuske, Nicolas, Mahadew, Kival, Rosman, Benjamin, van Hoof, Charel, Shock, Jonathan

Abstract

Creatures that display 'hedonic place preference behaviour' are thought by many scientists to experience feelings, on the assumption that their attraction to pleasure-producing substances which lack nutritional value (e.g. cocaine, morphine) cannot easily be attributed to unconscious instinctual behaviour. In this paper, we discuss how a simple artificial agent that instantiates attributes of an affective system engaging in felt uncertainty about its intrinsic needs in relation to environmental resources can similarly display hedonic place preference behaviour -- through an apparently subjective form of information processing -- while simultaneously being entirely deter-ministic. We outline some implications of this artificially engineered behaviour for our understanding of the physical basis of consciousness and the experience of free will.

Chinese Translation

许多科学家认为，表现出“享乐场所偏好行为”的生物体经历情感，这一假设基于它们对缺乏营养价值的愉悦物质（如可卡因、吗啡）的吸引力不能轻易归因于无意识的本能行为。本文讨论了一个简单的人工智能体如何具备情感系统的属性，并在与环境资源的内在需求相关的情况下，表现出对自身需求的不确定性感知，从而以一种明显主观的信息处理方式显示出享乐场所偏好行为，同时又完全是确定性的。我们概述了这种人工设计行为对我们理解意识的物理基础和自由意志体验的一些启示。

View on arXiv Download PDF AI Translation

cs.AI / 43 / 2609.03912

Lose the Order, Keep the Hierarchy: Deordering HTN Plans

失去顺序，保持层次：HTN计划的去序化

Togarepi, Takudzwa, Quenard, Gaspard, Pellier, Damien, Fiorino, Humbert

Abstract

Hierarchical Task Network (HTN) planning is a powerful planning formalism based on task decomposition. Although most of the literature studied plan generation, comparatively less attention has been paid to post-plan optimization. In particular, plan deordering has been extensively studied in classical planning but remains under-researched in the HTN setting. Plan deordering removes unnecessary ordering constraints between actions in a plan whilst keeping the plan valid. In this paper, we adapt two established plan deordering techniques from classical planning by extending the techniques to account for hierarchical decomposition constraints. We evaluate our proposed approaches on the IPC 2023 Partial-Order HTN benchmarks and we compare them against Optiplan, an HTN planner that generates partially ordered plans directly. Our results show a substantial reduction in number of ordering constraints in both our implementations. Although we also observe a reduction in critical path length, the improvements are less pronounced.

Chinese Translation

层次任务网络（Hierarchical Task Network, HTN）规划是一种基于任务分解的强大规划形式。尽管大多数文献研究了计划生成，但对计划优化后的研究相对较少。特别是，计划去序化在经典规划中得到了广泛研究，但在HTN环境下仍然缺乏研究。计划去序化在保持计划有效的同时，去除计划中动作之间不必要的顺序约束。本文通过扩展经典规划中的两种成熟的计划去序化技术，以考虑层次分解约束，来适应HTN规划。我们在IPC 2023部分有序HTN基准测试上评估了我们提出的方法，并将其与直接生成部分有序计划的HTN规划器Optiplan进行了比较。我们的结果显示，在我们的实现中，顺序约束的数量大幅减少。尽管我们也观察到关键路径长度的减少，但改进的效果不那么明显。

View on arXiv Download PDF AI Translation

cs.AI / 44 / 2609.03920

Value-Preserving Architectures for Agentic AI Systems

价值保留架构用于自主智能系统

Pesare, Alessandro, Dolci, Tommaso, Hose, Katja, Sallinger, Emanuel

Abstract

The emergence of agentic AI and LLM-based multi-agent systems (MAS) presents unprecedented opportunities for automating complex tasks, while simultaneously raising critical concerns about the preservation of fundamental human-centered values, such as privacy, fairness, and safety. Although software engineering has traditionally focused on functional correctness, the adoption of LLMs and AI agents into complex socio-technical systems has intensified the need for responsible software engineering and robust value alignment. In MAS, architectural design decisions, such as coordination mechanisms, communication protocols, and system topologies, play a central role in shaping system behavior and the outcomes they produce. This paper argues that architectural choices influence not only the functionality and performance of MAS but can also promote value-oriented system behavior. Therefore, we investigate how different architectural designs support different human-centered values, discussing the following value-preserving architectural patterns: (i) a privacy-aware architecture with a federated topology, (ii) a distributed architecture to promote pluralism and diversity, and (iii) a guard-agent architecture to detect and mitigate unfairness. Finally, we introduce representative use cases to illustrate the proposed architectures in real-world scenarios. By linking architectural design with human-centered values, this work lays the foundation for a unified set of architectural patterns and guidelines towards the design of trustworthy MAS.

Chinese Translation

自主智能和基于大型语言模型（LLM）的多智能体系统（MAS）的出现为自动化复杂任务提供了前所未有的机会，同时也引发了关于保护基本以人为本的价值观（如隐私、公平和安全）的重大担忧。尽管软件工程传统上关注功能正确性，但LLM和人工智能代理的采用加剧了对负责任的软件工程和稳健的价值对齐的需求。在MAS中，架构设计决策（如协调机制、通信协议和系统拓扑）在塑造系统行为及其产生的结果方面发挥着核心作用。本文认为，架构选择不仅影响MAS的功能和性能，还可以促进以价值为导向的系统行为。因此，我们研究不同架构设计如何支持不同的人本价值，讨论以下价值保留架构模式：（i）具有联邦拓扑的隐私意识架构，（ii）促进多元化和多样性的分布式架构，以及（iii）用于检测和缓解不公平的守卫代理架构。最后，我们介绍代表性的使用案例，以说明所提议的架构在现实场景中的应用。通过将架构设计与以人为本的价值联系起来，本研究为设计可信赖的MAS奠定了统一的架构模式和指导原则的基础。

View on arXiv Download PDF AI Translation

cs.AI / 45 / 2609.03923

Speak for Me: Giving LLMs the Situational Awareness to Participate in a Meeting

为我发声：赋予大型语言模型参与会议的情境意识

Khan, Muneeb, Kirstein, Frederic, Ruas, Terry, Gipp, Bela

Abstract

In online meeting delegation, LLM agents fail to recognize when to speak. With no structured way to track stances, coverage, and floor, they miss the moments where they should contribute. Prompt-only delegates stay silent on 51.4% of the absent participant's talking opportunities on the AMI corpus. We present CAPA (Collaborative Agent Predictive Architecture), an architecture for online meeting delegation. A Perceiver updates the meeting state from each observed turn. A Predictor forecasts how the conversation will continue. A Controller decides whether to speak and which proposition to surface. A Generator phrases the chosen contribution in the participant's style. Two judges score the forecast and the action against the next observed turn. A Recalibrator updates the meeting state from those verdicts for future decisions. To evaluate online delegation, we introduce an episode-level protocol that scores whether, when, and what a delegate contributes around the participant's actual idea units. The protocol's schema-constrained LLM judges align with human annotations at Cohen's kappa = 0.71. On 137 AMI meetings, CAPA reduces the silence rate from 51.4% to 2.5%, doubles credited recovery (26.1 --> 52.2), and keeps hallucination at 0.6%. The failure mode shifts from omission to selection, with each residual near-miss attributable to a specific module of the architecture. Mechanism ablations identify the meeting state as the lever that closes the recognition gap, where raw-context scaling alone does not.

Chinese Translation

在在线会议的代表性中，大型语言模型（LLM）代理未能识别何时发言。由于缺乏结构化的方式来跟踪立场、覆盖范围和发言权，他们错过了应当贡献的时刻。在AMI语料库中，仅依靠提示的代理在缺席参与者的发言机会中保持沉默的比例高达51.4%。我们提出了CAPA（协作代理预测架构），一种用于在线会议代表的架构。感知器（Perceiver）从每个观察到的轮次更新会议状态。预测器（Predictor）预测对话将如何继续。控制器（Controller）决定是否发言以及提出哪个观点。生成器（Generator）以参与者的风格措辞所选择的贡献。两个评审对预测和行动与下一个观察到的轮次进行评分。重新校准器（Recalibrator）根据这些裁决更新会议状态，以便做出未来的决策。为了评估在线代表，我们引入了一种基于情节的协议，该协议评分代表在参与者实际思想单元周围的贡献情况、时机和内容。该协议的模式约束LLM评审与人类注释的Cohen's kappa为0.71。在137个AMI会议中，CAPA将沉默率从51.4%降低到2.5%，使得被认可的恢复率翻倍（26.1 --> 52.2），并将幻觉率保持在0.6%。失败模式从遗漏转变为选择，每个剩余的近失都可归因于架构的特定模块。机制消融实验表明，会议状态是缩小识别差距的杠杆，而仅仅依靠原始上下文扩展是无法实现的。

View on arXiv Download PDF AI Translation

cs.AI / 46 / 2609.03938

Towards Numerical TOHTN Planning with SMT-based HTN-SAT Encoding

基于 SMT 的 HTN-SAT 编码的数值 TOHTN 规划研究

Quenard, Gaspard, Togarepi, Takudzwa, Pellier, Damien, Fiorino, Humbert

Abstract

While HTN planning has received significant attention in recent years, support for numerical reasoning remains very limited. In this paper, we investigate numerical Totally-Ordered HTN (TOHTN) planning and show how standard SAT-based encodings can be naturally extended with SMT to handle numeric fluents. In addition, we introduce a benchmark suite for numerical TOHTN planning, providing a first common basis for evaluation in this setting. Experimental results show that this simple encoding already constitutes a competitive baseline. This work opens the way to more expressive approaches to HTN planning.

Chinese Translation

尽管 HTN 规划近年来受到了广泛关注，但对数值推理的支持仍然非常有限。本文研究了数值完全有序 HTN（TOHTN）规划，并展示了如何将标准的基于 SAT 的编码自然扩展为 SMT，以处理数值流。除此之外，我们还引入了一个数值 TOHTN 规划的基准套件，为该领域的评估提供了首个共同基础。实验结果表明，这种简单的编码已经构成了一个具有竞争力的基线。这项工作为 HTN 规划的更具表现力的方法开辟了道路。

View on arXiv Download PDF AI Translation

cs.AI / 47 / 2609.03943

More Criticism Does Not Make a Better Review: EquiReview-R

更多的批评并不意味着更好的评审：EquiReview-R

Zhang, Zexing, Li, Jichao, Lei, Tianyang, Fu, Yude, Kewei, Yang

Abstract

AI reviewers can now produce many specific criticisms, but more criticism is not necessarily a better review. A review may miss a consequential weakness or retain an allegation that available evidence does not support. These failures require opposite corrections, yet generation-oriented systems and aggregate measures obscure the distinction. We therefore recast AI-assisted review as evidence-guided refinement of a structured concern set, with omission and overcritique treated as separate risks. Building on this formulation, we introduce EquiReview-R, which resolves existing concerns against localized evidence, searches for missing issues from independent and review-conditioned perspectives, and returns stop, continue, or defer. To expose the failure mode that motivates this design, we construct an evidence-linked trajectory corpus. Its retrospective analysis shows why revision must precede further search: nearly all concerns in a high-recall review lack a definitive evidential disposition, while an earlier refinement mechanism cannot revise them. On a frozen cohort of previously unseen papers, EquiReview-R satisfies the prespecified non-inferiority criterion for major omission, reduces major overcritique from 15.5% to 8.1%, and attains a one-sided omission upper bound of 9.9% while stopping on 52.4% of papers. Computation-matched controls, controlled pairs, and ablations show that the gain comes from revision rather than extra inference or shorter output. We release the corpus as ReviewTrace, an evidence-linked resource for studying review revision, disagreement, and provenance.

Chinese Translation

人工智能评审者现在能够产生许多具体的批评，但更多的批评不一定意味着更好的评审。一项评审可能会遗漏一个重要的弱点，或者保留一个没有证据支持的指控。这些失误需要相反的纠正，然而以生成为导向的系统和聚合度量模糊了这种区别。因此，我们将人工智能辅助评审重新构建为基于证据的结构化关注集的细化，其中遗漏和过度批评被视为独立的风险。在此基础上，我们引入了EquiReview-R，它针对局部证据解决现有问题，从独立和评审条件的角度搜索遗漏的问题，并返回停止、继续或推迟。为了揭示促使这一设计的失效模式，我们构建了一个与证据相关的轨迹语料库。其回顾性分析显示，为什么修订必须在进一步搜索之前进行：在高召回率评审中，几乎所有的关注点都缺乏明确的证据处置，而早期的细化机制无法对其进行修订。在一个冻结的未见论文群体中，EquiReview-R 满足了预先设定的主要遗漏非劣性标准，将主要过度批评从15.5%降低到8.1%，并在停止52.4%的论文时达到了9.9%的单侧遗漏上限。计算匹配的对照组、受控对和消融实验表明，收益来自修订，而非额外推理或更短输出。我们将该语料库发布为ReviewTrace，这是一个与证据相关的资源，用于研究评审修订、分歧和来源。

View on arXiv Download PDF AI Translation

cs.AI / 48 / 2609.03960

FiMI Banking: A Sovereign Model for Indian Retail Banking

FiMI 银行：印度零售银行的主权模型

NPCI AI Research Team, Kumar, Aman, Desai, Asit, Bhushan, Chandra, Sharma, Harsh, Bhushan, Harshit, Kadam, Hrithik, Doshi, Keyur, Kapardheeswar, Kolisetty Sai, Adhikary, Krishanu, Shaik, Nadeem, Prakash, Navya, Kukreja, Nitin, Devadiga, Prashant, MH, Shamanth, Pandey, Shantanu, Paul, Suvradip, Dedhia, Yatharth

Abstract

Banks need conversational systems that can answer product questions, assist customers with account-related requests, and operate safely within strict operational and regulatory constraints. General-purpose language models do not reliably meet these requirements. They fall short when a task requires grounded information, correct tool use, or cautious handling of bank-specific sensitive situations. We introduce FiMI Banking, a controlled Indian retail-banking setting. We build it from vetted banking documents, structured ground truth, synthetic customer backgrounds, and banking tools. We evaluate two post-training approaches: preference optimization for response-level behavior, and reinforcement learning with verifiable rewards for multi-turn tool-use tasks. Preference optimization improves safe behavior substantially: out-of-scope refusal rises from 52% to 80%. Reinforcement learning improves edge-case performance from 0.509 to 0.718 and order-sensitive task performance from 0.590 to 0.679, while using 29% fewer generated tokens. These results show that preference optimization and verifiable-reward reinforcement learning address complementary requirements for reliable banking agents.

Chinese Translation

银行需要能够回答产品问题、协助客户处理账户相关请求，并在严格的操作和监管约束下安全运行的对话系统。通用语言模型无法可靠地满足这些要求。当任务需要基于事实的信息、正确的工具使用或谨慎处理银行特定的敏感情况时，它们往往表现不佳。我们提出了 FiMI 银行，这是一个受控的印度零售银行环境。我们基于经过审查的银行文件、结构化的真实数据、合成的客户背景和银行工具构建了该环境。我们评估了两种后训练方法：针对响应级行为的偏好优化，以及使用可验证奖励的强化学习用于多轮工具使用任务。偏好优化显著改善了安全行为：超出范围的拒绝率从 52% 上升至 80%。强化学习将边缘案例的表现从 0.509 提升至 0.718，将顺序敏感任务的表现从 0.590 提升至 0.679，同时减少了 29% 的生成标记。这些结果表明，偏好优化和可验证奖励的强化学习满足了可靠银行代理的互补需求。

View on arXiv Download PDF AI Translation

cs.AI / 49 / 2609.03966

Interface-Induced Trajectory Censoring

接口引发的轨迹审查

Wang, Wenbo

Abstract

Agent evaluations report a tool-call rate read off the serving stack. That number can be zero while the model is emitting well-formed calls: the interface censors the trajectory before anything downstream sees it. On BFCL v4's own data, executor and scorer, holding weights, cases, decoding and seeds fixed and changing only the serving adapter, the same model scores 0.00 or 0.96 / 0.19. A 2x2 over chat template and parser locates the effect exactly: both main effects are exactly zero and all of it sits in the interaction -- no component is defective, and repairing one side of the contract buys precisely nothing. On tau-bench's 115 interactive retail tasks the same swap moves server-parsed calls from 0 to 636 and tasks reaching any tool execution from 0 to 103. Our probe reproduces the funnel across a 21x scale range of Qwen2.5-Coder: the server parses 0/100 at every size while well-formed emitted calls rise to 80/100 at 32B (~72 after calibration against an adjudicated gold standard). Under a matched envelope, across a comparable scale span, the silent fraction stays at 0-2, a prediction committed to the repository before the run. Llama-3.1-8B's 23% rate of calling the task function itself as a tool falls to 0 under one strict:true flag. The mismatch reaches inside the training loop, and its consequence is scale-dependent: in verl's AgentLoop at 7B, 45 of 115 generations carry a complete call; 0 are accepted, 0 execute, 0 return an observation. At 1.5B the same zero is over-determined, so we report the two scales separately. At evaluation time, repairing the adapter restores the mechanism but not a significant outcome gain: parsing 0->84, rescues 0->9, pass rate 53->62 (n.s.). We release a 98-line preflight check that catches every silent failure here. The observed tool-call rate is not a property of the model alone; it is a property of the model-interface stack that measures it.

Chinese Translation

代理评估报告了从服务栈读取的工具调用率。这个数字可以为零，即使模型发出的调用是格式良好的：接口在任何下游看到之前审查了轨迹。在 BFCL v4 自身的数据上，执行器和评分器固定权重、案例、解码和种子，仅更改服务适配器，同一模型的得分为 0.00 或 0.96 / 0.19。一个 2x2 的聊天模板和解析器准确定位了这个效应：两个主要效应都是零，所有效应都在交互中——没有任何组件有缺陷，修复合同的一方并不会带来任何收益。在 tau-bench 的 115 个交互零售任务中，同样的交换将服务器解析的调用从 0 移动到 636，任何工具执行的任务从 0 移动到 103。我们的探测器在 Qwen2.5-Coder 的 21 倍规模范围内重现了漏斗：服务器在每个规模下的解析率为 0/100，而格式良好的发出调用在 32B 时上升到 80/100（在与裁定的黄金标准校准后约为 72）。在一个匹配的包络下，跨越可比的规模范围，静默比例保持在 0-2，这是在运行之前提交到库的预测。Llama-3.1-8B 将任务函数本身作为工具调用的比例从 23% 降至 0，在一个 strict:true 标志下。这个不匹配深入到训练循环，其后果是规模依赖的：在 verl 的 AgentLoop 中，在 7B 的情况下，115 次生成中有 45 次携带完整调用；0 次被接受，0 次执行，0 次返回观察。在 1.5B 的情况下，同样的零是过度确定的，因此我们分别报告这两个规模。在评估时，修复适配器恢复了机制，但没有显著的结果增益：解析从 0->84，救回从 0->9，及格率从 53->62（不显著）。我们发布了一个 98 行的预检，捕捉到这里的每一个静默失败。观察到的工具调用率不仅是模型的属性；它是测量模型-接口栈的属性。

View on arXiv Download PDF AI Translation

cs.AI / 50 / 2609.03973

Common-Witness Certificates and Sharp Feature Bounds for Counterfactual Image Auditing

共同见证证书与反事实图像审计的精确特征界限

Faghihi, Usef, Saki, Amir

Abstract

An image editor may satisfy every regional plausibility constraint separately even when no single latent explanation fits the complete output. We formalize this local-to-global failure using a common witness grade and witness nerve. The framework separates auditing from causal identification: shared exogeneity alone allows every coupling of the regime marginals, whereas an externally justified witness relation yields sharp partial-identification bounds for prespecified image features. Helly-type arguments provide short incompatibility certificates for quasiconvex losses, heterogeneous action strata, and finite witness atlases; a blocker-hypergraph formula gives exact repair counts. Simultaneous confidence regions for the regime marginals give finite-sample outer coverage of the complete identified interval. Controlled MNIST, Morpho-MNIST, and smallNORB studies demonstrate the predicted local-global separation, while synthetic experiments test sharp bounds, certificate recovery, and structured computation. The method audits a declared feature relation and does not identify unrestricted pixel-level counterfactuals.

Chinese Translation

图像编辑器可能在每个区域的合理性约束下单独满足条件，即使没有单一的潜在解释能够适应完整的输出。我们通过共同见证等级和见证神经形式化这种局部到全局的失效。该框架将审计与因果识别分开：仅共享外生性就允许每种制度边际的耦合，而外部证明的见证关系则为预设图像特征提供了精确的部分识别界限。Helly型论证为拟凸损失、异质行动层和有限见证图集提供了简短的不兼容证书；一个阻塞超图公式给出了确切的修复计数。制度边际的同时置信区间为完整识别区间提供了有限样本的外覆盖。控制的MNIST、Morpho-MNIST和小型NORB研究展示了预测的局部-全局分离，而合成实验则测试了精确界限、证书恢复和结构化计算。该方法审计声明的特征关系，并不识别无限制的像素级反事实。

View on arXiv Download PDF AI Translation

cs.AI / 51 / 2609.04005

The Dually Flat Geometry of Planning as Inference

规划作为推理的双重平坦几何

Milosevic, Nikola, Kataoka, Asaki, Hinrichs, Nicolas, Doya, Kenji, Scherf, Nico

Abstract

We present an alternative characterization of the occupancy measure of reinforcement learning, obtained by embedding the planning criterion into the dynamics through a resetting planning process. Its stationary measure, which we term visitation measure, is the object on which the information geometry of decision making is most naturally expressed. The achievable visitation measures form a dually flat statistical manifold whose two affine charts are the visitation probabilities and the log-policies, dual under the conditional entropy. This structure makes planning-as-inference generalize from linear rewards to nonlinear functionals of the visitation, each iterate solved by one natural-gradient step, and gives the temporal-difference error the interpretation of a marginal-utility estimate. We develop the geometry and its consequences for reinforcement learning and theoretical neuroscience.

Chinese Translation

我们提出了一种强化学习占用度量的替代特征，通过将规划标准嵌入动态系统中，通过重置规划过程获得。我们称之为访问度量的平稳度量是决策信息几何最自然表达的对象。可实现的访问度量形成一个双重平坦的统计流形，其两个仿射图是访问概率和对数策略，在条件熵下是对偶关系。这一结构使得规划作为推理能够从线性奖励推广到访问的非线性泛函，每次迭代通过一个自然梯度步骤解决，并赋予时间差错以边际效用估计的解释。我们发展了这一几何结构及其对强化学习和理论神经科学的影响。

View on arXiv Download PDF AI Translation

cs.AI / 52 / 2609.04013

LLM4CKD: Large Language Models for Early Stage Chronic Kidney Disease Screening

LLM4CKD：大型语言模型用于早期慢性肾病筛查

Kabir, Muhammad Ashad, Munira, Sirajam

Abstract

Early screening of chronic kidney disease (CKD) is critical for timely intervention, yet most machine learning (ML) and deep learning (DL) approaches require labeled data and model training, limiting their use in real-world screening settings. This study evaluates the effectiveness of large language models (LLMs) for CKD screening under zero-shot and few-shot in-context learning settings and compares them with traditional ML and DL methods. We propose a framework that uses clinically selected tabular features and structured prompt templates to enable LLM-based inference without task-specific training. LLM performance is evaluated across multiple prompt styles, feature configurations, and data settings, and compared with standard ML, DL, and tabular foundation model (TFM) baselines, and existing CKD screening tools. The results show that LLMs can achieve competitive performance using only a small number of examples, often matching or outperforming traditional approaches in low-data settings. However, their performance remains model-dependent and less stable as input complexity increases. In contrast, ML, DL, and TFM models show more consistent improvement with larger training data. Overall, the findings highlight a trade-off between data efficiency and stability, suggesting that LLMs may serve as a flexible complementary approach for CKD screening when labeled data are limited.

Chinese Translation

早期筛查慢性肾病（CKD）对及时干预至关重要，但大多数机器学习（ML）和深度学习（DL）方法需要标记数据和模型训练，这限制了它们在实际筛查环境中的应用。本研究评估了大型语言模型（LLMs）在零-shot和少-shot上下文学习设置下进行CKD筛查的有效性，并将其与传统的ML和DL方法进行了比较。我们提出了一个框架，该框架利用临床选择的表格特征和结构化提示模板，使得基于LLM的推理无需特定任务的训练。LLM的性能在多种提示风格、特征配置和数据设置下进行了评估，并与标准的ML、DL和表格基础模型（TFM）基准以及现有的CKD筛查工具进行了比较。结果表明，LLMs在仅使用少量示例的情况下可以实现具有竞争力的性能，通常在低数据环境中与传统方法相匹配或超越。然而，随着输入复杂性的增加，它们的性能仍然依赖于模型，并且稳定性较差。相比之下，ML、DL和TFM模型在更大训练数据下表现出更一致的改进。总体而言，研究结果突显了数据效率与稳定性之间的权衡，表明当标记数据有限时，LLMs可能作为CKD筛查的灵活补充方法。

View on arXiv Download PDF AI Translation

cs.AI / 53 / 2609.04014

InSituMeasure: Probing Situated Measurement Grounding in Industrial Scenes with Multimodal Large Language Models

InSituMeasure：利用多模态大型语言模型探测工业场景中的情境测量基础

Shen, Chao, Li, Xinyuan, Zhou, Yunfan, Yao, Jianguo, Guan, Haibing, Wang, Zhihai, Li, Xijun

Abstract

For trained operators, gauge reading requires little specialized knowledge, low cognitive effort, and high repeatability. Yet Multimodal Large Language Models (MLLMs) remain unreliable in continuous-valued measurement despite strong results on general multimodal benchmarks. Existing benchmarks expose this weakness but isolate measurement from realistic, knowledge-grounded settings, with limited situated context, specialized instruments, real-world noise, and matched diagnostic annotations, reducing realism and constraining root-cause analysis. We introduce InSituMeasure to evaluate situated measurement grounding. It contains 2,922 real industrial monitoring scenes across eight functional categories of professional engineering instruments, with dense gauge-attribute annotations and noise tags for failure diagnosis. We define metrics for numerical accuracy under predefined tolerances and unit consistency, rejection of fake or unanswerable tasks, and alignment between model failures and annotated error factors. Across 24 state-of-the-art MLLMs, the best model reaches only 25.7\% joint value-unit accuracy and 51.8\% confidence-diagnosis F1, revealing a substantial gap between general multimodal competence and reliable situated measurement. Further analysis identifies failures from text-induced shortcuts, overconfident responses, and authentic industrial noise, including mixed disturbances, viewpoint deviation, occlusion, and environmental interference.

Chinese Translation

对于经过培训的操作员而言，读表需要的专业知识较少，认知负担低且重复性高。然而，尽管在一般多模态基准测试中表现良好，多模态大型语言模型（MLLMs）在连续值测量方面仍然不可靠。现有的基准测试暴露了这一弱点，但将测量与现实的、知识基础的环境隔离，缺乏情境背景、专业仪器、真实世界的噪声和匹配的诊断注释，从而降低了现实性并限制了根本原因分析。我们引入了InSituMeasure来评估情境测量基础。该数据集包含来自八个专业工程仪器功能类别的2,922个真实工业监测场景，配有密集的表盘属性注释和故障诊断的噪声标签。我们定义了在预定义公差和单位一致性下的数值准确性指标、对虚假或无法回答任务的拒绝，以及模型失败与注释错误因素之间的一致性。在24个最先进的MLLMs中，最佳模型仅达到25.7%的联合值-单位准确率和51.8%的置信度-诊断F1，揭示了通用多模态能力与可靠情境测量之间的显著差距。进一步分析确定了由于文本诱导的捷径、过于自信的响应以及真实工业噪声（包括混合干扰、视角偏差、遮挡和环境干扰）导致的失败。

View on arXiv Download PDF AI Translation

cs.AI / 54 / 2609.04021

FLY-EVAL++: An Evidence-Driven Evaluation Protocol for Safety-Constrained Flight Prediction with Large Language Models

FLY-EVAL++：一种基于证据的安全约束飞行预测评估协议，适用于大型语言模型

Wu, Yalun, Fang, Junfeng, Wang, Jiawei, Liu, Haotian, Yang, Qijun, Yang, Minghan, Guo, Hongcheng, Li, Zhoujun, Wang, Boyang

Abstract

Evaluating large language models (LLMs) in safety-critical, physics-governed environments requires more than accuracy-based metrics, because predictions that are numerically close to the ground truth can still violate operational constraints, combine fields in physically inconsistent ways, or fail to produce usable structured outputs. Existing evaluation protocols do not measure these failure modes reliably. We propose FLY-EVAL++, an evidence-driven evaluation protocol that combines deterministic verification of protocol compliance, physical feasibility, and safety constraints with fixed rubric-guided aggregation into interpretable multi-dimensional scores. We instantiate FLY-EVAL++ for Flight Trajectory and Attitude Prediction (FTAP) by extending the PilotBench setting with history-conditioned and multi-step prediction tasks. Across 66 LLMs, safety compliance is the most discriminative dimension of model behavior: models with comparable predictive performance differ by more than 28 points in safety score, and we observe recurrent failures including safety violations under physically plausible predictions and instability in multi-step rollouts. These results show that evaluation in safety-critical domains should measure constraint satisfaction and structured validity explicitly rather than rely on accuracy-centric reporting alone.

Chinese Translation

在安全关键、受物理规律约束的环境中评估大型语言模型（LLMs）需要的不仅仅是基于准确性的指标，因为数值上接近真实值的预测仍可能违反操作约束，以物理不一致的方式组合领域，或未能生成可用的结构化输出。现有的评估协议无法可靠地测量这些失败模式。我们提出了FLY-EVAL++，一种基于证据的评估协议，结合了对协议合规性、物理可行性和安全约束的确定性验证，以及基于固定评分标准的可解释多维评分聚合。我们通过扩展PilotBench设置，针对历史条件和多步预测任务，为飞行轨迹和姿态预测（FTAP）实例化了FLY-EVAL++。在66个LLMs中，安全合规性是模型行为中最具区分性的维度：具有可比预测性能的模型在安全评分上相差超过28分，我们观察到重复出现的失败，包括在物理上合理的预测下的安全违规和多步展开中的不稳定性。这些结果表明，在安全关键领域的评估应明确测量约束满足和结构有效性，而不是仅依赖于以准确性为中心的报告。

View on arXiv Download PDF AI Translation

cs.AI / 55 / 2609.04024

Instruction Duplication as an Inference-Time Control Primitive

指令重复作为推理时控制原语

Lavrenko, Victor

Abstract

Procedural instruction following is a basic requirement for controllable language-model systems, especially when generated trajectories are inspected or repaired downstream. We introduce instruction duplication, a minimal black-box inference-time control that repeats only the procedural instruction, without retraining or decoding changes. Across seven instruction-tuned models, 300 medical multiple-choice questions, eight placement conditions, and 16,800 scheduled generations, moving from one to two copies raises the deterministic All-8 diagnostic--responses passing all eight observable tests--from 90.22% to 93.17% (+2.95 percentage points), eliminating 30.2% of the failures remaining after one copy. Pre-provisional TF-IDF recall rises from 73.44% to 74.81% (+1.38 points; Holm-adjusted p < .001), while final-answer accuracy remains exactly 60.21%. Premature commitment increases from 1.52% to 2.30% (p_Holm = .00536). A blinded challenge audit yields 10/30 directional confirmations, 20/30 perceptual ties, and no reversals; its prespecified 28/30 confirmation criterion is not met. Yet this distinction can matter operationally when a downstream system acts on the generated trajectory. In Answer Engineering (AE), where explicit trajectory state determines local repair, the published reason-first no-editing SSNHL endpoint was 25.1%; system-only AE was later reproduced at 84.2%, and the same trailing duplicate raised it to 97.1%. For conductive diagnostic branch preservation, the corresponding values are 58.9% published without editing, 78.6% with reproduced AE, and 73.8% with AE plus duplication--a within-AE decrease, but still 14.9 points above the no-editing baseline. Instruction duplication is therefore a low-complexity, placement-sensitive control whose practical value can emerge through the downstream system that consumes the exposed trajectory.

Chinese Translation

程序指令跟随是可控语言模型系统的基本要求，特别是在生成的轨迹需要在后续进行检查或修复时。我们引入了指令重复，这是一种最小的黑箱推理时控制，仅重复程序指令，而无需重新训练或解码更改。在七个经过指令调优的模型、300个医学多项选择题、八种放置条件和16,800次计划生成中，从一个副本增加到两个副本使得确定性 All-8 诊断——所有八个可观察测试均通过的响应——从90.22%提高到93.17%（+2.95个百分点），消除了在一个副本后剩余的30.2%的失败。预先规定的 TF-IDF 召回率从73.44%上升到74.81%（+1.38点；Holm调整后的 p < .001），而最终答案的准确率保持在60.21%。过早承诺从1.52%增加到2.30%（p_Holm = .00536）。一次盲审挑战审核得出10/30的方向确认，20/30的感知平局，且没有反转；其预先规定的28/30确认标准未能达成。然而，当下游系统对生成的轨迹采取行动时，这一区别在操作上可能具有重要意义。在答案工程（AE）中，明确的轨迹状态决定局部修复，已发布的理由优先不编辑的 SSNHL 终点为25.1%；系统单独的 AE 后来重现为84.2%，而相同的尾随重复将其提高到97.1%。对于传导性诊断分支的保留，相应的值为58.9%（未编辑发布），78.6%（重现的 AE），以及73.8%（AE 加重复）——在 AE 内部的下降，但仍比不编辑基线高出14.9个百分点。因此，指令重复是一种低复杂度、对放置敏感的控制，其实际价值可以通过消费暴露轨迹的下游系统显现出来。

View on arXiv Download PDF AI Translation

cs.AI / 56 / 2609.04030

IRWOZ 2.0: A Large Language Model-driven Dialogue Dataset for Industrial Robot Conversations

IRWOZ 2.0：一个基于大型语言模型的工业机器人对话数据集

Li, Chen, Chrysostomou, Dimitrios

Abstract

IRWOZ has improved industrial human-robot interaction (HRI) dialogue systems through domain-specific annotations. However, its initial version contains substantial noise in dialogue states and utterances, limiting state-tracking accuracy. We introduce IRWOZ 2.0, which addresses these limitations through large language model (LLM) enhanced generation (Mistral/Claude-3.5) and quality refinements. Our improved dataset expands to 390 dialogues across 4 industrial domains (Assembly, Delivery, Position, Relocation), featuring manual corrections and automated typo removal. Benchmark experiments on dialogue state tracking demonstrate significant improvements, with GPT-2's BLEU-4 score increasing from 0.1651 to 0.5604 compared to original IRWOZ. To support industrial HRI research, we publicly released IRWOZ 2.0 dataset at https://ieee-dataport.org/documents/irwoz-20-large-language-model-driven-dialogue-dataset-industrial-robot-conversations

Chinese Translation

IRWOZ通过领域特定的注释改善了工业人机交互（HRI）对话系统。然而，其初始版本在对话状态和发言中包含了大量噪声，限制了状态跟踪的准确性。我们推出了IRWOZ 2.0，通过大型语言模型（LLM）增强生成（Mistral/Claude-3.5）和质量改进来解决这些限制。我们改进的数据集扩展到涵盖4个工业领域（组装、交付、定位、搬迁）的390个对话，包含手动修正和自动拼写错误删除。对话状态跟踪的基准实验显示出显著的改善，相较于原始IRWOZ，GPT-2的BLEU-4分数从0.1651提高到0.5604。为了支持工业HRI研究，我们在https://ieee-dataport.org/documents/irwoz-20-large-language-model-driven-dialogue-dataset-industrial-robot-conversations公开发布了IRWOZ 2.0数据集。

View on arXiv Download PDF AI Translation

cs.AI / 57 / 2609.04063

Spurious Advantage Hidden in GRPO

隐藏在GRPO中的虚假优势

Wang, Jiamian, Basu, Samyadeep, Goswami, Koustava, Yu, Tong, Tao, Zhiqiang

Abstract

Group Relative Policy Optimization (GRPO) is widely studied for reinforcement learning with verifiable rewards, where its advantage estimator assigns each rollout a magnitude from within-group reward statistics. In the common case, this magnitude rewards rollouts that reach the correct answer through reasoning. Yet, an overlooked case shares the same surface: a rollout may land on it by guessing, and the formula still assigns a high magnitude, which we identify as the spurious advantage. This arises in three cases: bounded-answer tasks with a small candidate set; open-answer sets hosting bounded sub-cases; and search agents whose budget opens many paths to the same answer. In all three, this misleads the policy toward guess-like behaviors. We propose SIGNBALANCE, whose magnitude is composition-free: it keeps the verifier sign, uses a global scale, and restores zero-mean balance via a stop-gradient per-class rescaling. Across math and search agent benchmarks at different scales, SIGNBALANCE matches GRPO on open-answer math and improves on bounded-answer math and search agents. Code will be released.

Chinese Translation

群体相对策略优化（Group Relative Policy Optimization, GRPO）在具有可验证奖励的强化学习中得到了广泛研究，其中其优势估计器为每个回合分配一个来自组内奖励统计的量。在常见情况下，这个量奖励通过推理达到正确答案的回合。然而，一个被忽视的情况与之表面相同：一个回合可能通过猜测达到正确答案，而公式仍然分配一个高的量，我们将其识别为虚假优势。这种情况出现在三种情形中：候选集较小的有界答案任务；包含有界子案例的开放答案集；以及预算允许多个路径到达相同答案的搜索代理。在这三种情况下，这会误导策略朝向类似猜测的行为。我们提出了SIGNBALANCE，其量是无组合的：它保持验证者的符号，使用全局尺度，并通过每类的停止梯度重新缩放恢复零均值平衡。在不同规模的数学和搜索代理基准测试中，SIGNBALANCE在开放答案数学上与GRPO相匹配，并在有界答案数学和搜索代理上有所改进。代码将会发布。

View on arXiv Download PDF AI Translation

cs.AI / 58 / 2609.04094

DRACO: Fine-Grained Credit Assignment with Dynamic Rubrics for Long-Horizon Agent Training

DRACO：用于长时间跨度代理训练的动态评分细粒度信用分配

Gandhi, Shubham, Goyal, Saurabh, Kate, Kiran, Rizk, Yara

Abstract

Reinforcement Learning from Verifiable Rewards works well when a task has a programmatic checker, but most long-horizon agent domains have none. We work in the outcome-blind setting, where ground-truth success signals are not available. Multi-criteria rubrics are a popular way to supply such a reward; they are scored once per trajectory, but a single scalar is a poor signal across tens of steps. We propose DRACO: Distributing Rubric-based Advantage for Credit Optimization. It generates rubrics dynamically during training to track the policy's evolving capability, scores those rubrics once per completed trajectory, and redistributes that judgment over the steps responsible for annotated rubrics to produce differentiated per-step advantages in GRPO. The redistribution is closed-form and does not introduce any trained attribution module. On AppWorld, DRACO gains 15.9 points over the base model and 5.3 points over GRPO trained with a sparse ground-truth reward, despite not using any verifiers itself. On out-of-domain Tau-Bench, it gains 5.3 points over the base model even without a frontier judge, beating both ground-truth-reward training and other rubric-based training settings. The code for DRACO is available at https://github.com/IBM/draco.

Chinese Translation

基于可验证奖励的强化学习在任务具有程序化检查器时表现良好，但大多数长时间跨度的代理领域并不存在这样的检查器。我们在结果盲设置中工作，其中真实的成功信号不可用。多标准评分是提供此类奖励的一种流行方式；它们在每个轨迹上评分一次，但单一的标量在数十个步骤中是一个较差的信号。我们提出了DRACO：基于评分的优势分配以优化信用。它在训练过程中动态生成评分，以跟踪策略的不断发展能力，在每个完成的轨迹上对这些评分进行评分，并将该判断在负责标注评分的步骤之间重新分配，以在GRPO中产生差异化的逐步优势。该重新分配是封闭形式的，并且不引入任何训练的归因模块。在AppWorld上，DRACO比基础模型提高了15.9分，比使用稀疏真实奖励训练的GRPO提高了5.3分，尽管它本身并未使用任何验证器。在域外的Tau-Bench上，即使没有前沿评判者，它也比基础模型提高了5.3分，超越了基于真实奖励的训练和其他基于评分的训练设置。DRACO的代码可在https://github.com/IBM/draco获取。

View on arXiv Download PDF AI Translation

cs.AI / 59 / 2609.04098

Why Gated DeltaNet Survives 4-Bit Quantization: NVFP4 W4A4 for the Recurrent Half of a Hybrid 27B LLM

为何 Gated DeltaNet 能够在 4 位量化中生存：混合 27B LLM 的递归部分的 NVFP4 W4A4

Kozyrev, Sergii, Maiboroda, Davyd

Abstract

Hybrid LLMs pair softmax attention with linear-attention layers such as Gated DeltaNet (GDN), whose recurrent state summarizes the context in fixed size. Early community 4-bit quantizations of Qwen3.8-27B (48 GDN layers, 16 attention layers) left the GDN block in 8- or 16-bit precision -- especially its decay and write-strength gates -- on the intuition that errors in a recurrence accumulate over long contexts. We test that intuition by building Minima: NVFP4 W4A4 on all 496 linear layers, GDN included. Across perplexity at 4K/32K, MMLU-Pro, GSM8K, AIME'25, GPQA-Diamond, LiveCodeBench, and RULER retrieval to 64K, Minima matches BF16 within seed noise (5-task average -0.52) while being the smallest (17.5 GiB) and fastest-prefill (+14-19%) recipe we compare, and its 32K perplexity gap shrinks with position. A four-part mechanism study explains why: (i) NVFP4's 16-element block scaling localizes the residual stream's extreme outliers, equalizing activation error across layer roles; (ii) the supposedly fragile gate projections are the least sensitive -- softplus/exponential and sigmoid parameterizations compress ~11% GEMM error to ~2% output error; (iii) the delta-rule recurrence holds injected noise at a flat plateau over 32K tokens and forgets a state impulse within hundreds of steps, because each write overwrites the state along the current key direction; (iv) the per-token quantization cost washes out with context instead of compounding. We also repair a global-scale mismatch that arises when per-module-calibrated NVFP4 checkpoints are served by kernels that fuse those modules into one GEMM, and show calibrated FP8 KV-cache scales are performance-free. The result: a practical recipe -- quantize everything, ship KV scales -- and a mechanistic account of why the recurrent half of a hybrid LLM is the easy half to quantize. Checkpoint: https://huggingface.co/minima-ai/mnma_qwen3.8_27b_nvfp4

Chinese Translation

混合 LLM 将 softmax 注意力与线性注意力层（如 Gated DeltaNet (GDN)）相结合，其递归状态以固定大小总结上下文。早期社区对 Qwen3.8-27B（48 个 GDN 层，16 个注意力层）的 4 位量化使 GDN 模块保持在 8 位或 16 位精度，尤其是其衰减和写入强度门，基于这样的直觉：递归中的错误在长上下文中会累积。我们通过构建 Minima：在所有 496 个线性层上应用 NVFP4 W4A4（包括 GDN）来测试这一直觉。在 4K/32K 的困惑度、MMLU-Pro、GSM8K、AIME'25、GPQA-Diamond、LiveCodeBench 和 RULER 检索到 64K 的测试中，Minima 在种子噪声范围内与 BF16 相匹配（5 任务平均 -0.52），同时是我们比较的最小（17.5 GiB）和最快的预填充配方（+14-19%），其 32K 困惑度差距随着位置的变化而缩小。四部分机制研究解释了原因：（i）NVFP4 的 16 元素块缩放将残差流的极端离群值局部化，使各层角色的激活误差均衡；（ii）所谓脆弱的门投影是最不敏感的——softplus/exponential 和 sigmoid 参数化将 ~11% 的 GEMM 误差压缩至 ~2% 的输出误差；（iii）delta-rule 递归在 32K 个 token 上保持注入噪声在一个平坦的平原上，并在数百步内遗忘状态冲动，因为每次写入都沿着当前关键方向覆盖状态；（iv）每个 token 的量化成本随着上下文的增加而消散，而不是累积。我们还修复了一个全球规模的不匹配问题，该问题出现在每个模块校准的 NVFP4 检查点被融合为一个 GEMM 的内核服务时，并展示了校准的 FP8 KV 缓存规模是无性能损失的。结果：一个实用的配方——量化所有内容，运输 KV 缩放——以及对混合 LLM 的递归部分为何是量化较容易部分的机制性解释。检查点：https://huggingface.co/minima-ai/mnma_qwen3.8_27b_nvfp4

View on arXiv Download PDF AI Translation

cs.AI / 60 / 2609.04127

Epistemic Warrant for LLM Recommendations: Characterizing the Basis for Reliance When Ground Truth Is Unavailable

大语言模型推荐的认识论依据：在缺乏客观真相时的依赖基础特征化

Vardi, Shai, Sedoc, João

Abstract

Large language models are increasingly used to support organizational decisions, yet users often lack a principled basis for assessing whether to rely on a specific recommendation. Existing approaches typically evaluate broad model properties, such as reliability, uncertainty, or robustness, or focus on user trust, rather than the underlying basis for relying on an individual recommendation. Adapting theoretical foundations from epistemology, we introduce epistemic warrant, a decision-level construct that characterizes the stability of a model's preference and the scope over which that preference holds. We operationalize this construct through a four-tier reliance certificate for pairwise recommendations, distinguishing among unstable, context-dependent, locally supported, and broadly supported recommendations. We validate the construct using contemporary methodologies: known-groups tests successfully recover expert-prespecified warrant orderings, and stronger warrants systematically align with independent consensus from crowd workers. Furthermore, we demonstrate that epistemic warrant provides information distinct from verbalized confidence and is not readily explained by decision difficulty. Ultimately, this framework offers a theoretically grounded, implementable approach for characterizing the warrant of individual LLM recommendations when objective ground truth is unavailable.

Chinese Translation

大语言模型越来越多地被用于支持组织决策，然而用户往往缺乏一个原则性的基础来评估是否应依赖于特定的推荐。现有的方法通常评估模型的广泛属性，如可靠性、不确定性或稳健性，或关注用户信任，而不是依赖于单个推荐的基础。我们借鉴认识论的理论基础，引入了认识论依据（epistemic warrant），这一决策层面的构念特征化了模型偏好的稳定性及其适用范围。我们通过一套四级依赖证书对成对推荐进行了操作化，区分了不稳定、依赖上下文、局部支持和广泛支持的推荐。我们使用当代方法验证了这一构念：已知组测试成功恢复了专家预设的依据排序，而更强的依据与来自众包工作者的独立共识系统性对齐。此外，我们展示了认识论依据提供的信息与口头表达的信心不同，并且不易被决策难度所解释。最终，这一框架为在缺乏客观真相时特征化个体大语言模型推荐的依据提供了一个理论基础扎实、可实施的方法。

View on arXiv Download PDF AI Translation

cs.AI / 61 / 2609.04128

Environment Evolution for Terminal Agents

终端智能体的环境演化

Fan, Zhiyuan, Yu, Tinghao, Cai, Yuanjun, Zhou, Jiang, Guan, Jiangtao, Liu, Jincheng, Yang, Yun, Hu, Dingxin, Han, Zhuo, Wu, Xing, Zhang, Feng, Wang, Lilin

Abstract

Scaling interactive and verifiable environments is critical for training terminal agents. As frontier models become more capable, environments synthesized from scratch become less challenging and thus provide limited learning signals. Recent co-evolution methods iteratively synthesize environments near the model's learnable frontier based on weaknesses exposed during rollouts. However, their dependence on on-policy rollouts limits generalization and the continuous provision of learning signals as the model becomes stronger. In this paper, we propose environment evolution, which incrementally increases environment difficulty off-policy and schedules the evolved environments generation by generation during training to provide continuous learning signals. We derive three evolution directions that influence environment difficulty from the multi-turn learning objective and then implement evolution along these directions through a loop-engineered multi-agent harness. Quantitative rollout experiments with Hy4 preview, Claude Opus 5, and GPT-5.6 Sol show that environment evolution consistently produces more difficult environments. We validate its effectiveness on Qwen3.6-27B and Qwen3.6-35B-A3B through simple long-horizon RL training, improving their performance by 14.4 and 18.0 percentage points on Terminal-Bench 2.1, respectively.

Chinese Translation

为终端智能体训练扩展互动和可验证的环境至关重要。随着前沿模型能力的增强，从零开始合成的环境变得不再具挑战性，因此提供的学习信号有限。最近的共同演化方法基于在回合中暴露出的弱点，迭代合成接近模型可学习边界的环境。然而，它们对在线回合的依赖限制了泛化能力，并在模型变得更强大时无法持续提供学习信号。本文提出了环境演化，它通过离线逐步增加环境难度，并在训练过程中按代调度演化环境，以提供持续的学习信号。我们从多轮学习目标中推导出三种影响环境难度的演化方向，并通过循环工程的多智能体框架沿这些方向实施演化。与 Hy4 preview、Claude Opus 5 和 GPT-5.6 Sol 的定量回合实验表明，环境演化始终产生更具挑战性的环境。我们通过简单的长时间跨度强化学习训练验证了其在 Qwen3.6-27B 和 Qwen3.6-35B-A3B 上的有效性，分别在 Terminal-Bench 2.1 上提高了 14.4 和 18.0 个百分点的性能。

View on arXiv Download PDF AI Translation

cs.AI / 62 / 2609.04135

The Natural Language Interaction Protocol and Standard for AI Agents

人工智能代理的自然语言交互协议与标准

Xing, Luyi, Topaloglu, Rasit Onur, Sinha, Ranjan, Ratnaparkhi, Abhay, Ndichu, Samuel, Nguyen, Christopher, Das, Anindita, Sheffler, Tom, Rahouti, Mohamed, Li, Zichuan, Liao, Xiaojing, Aiyagari, Sanjay

Abstract

AI agents are increasingly being developed and deployed across organizations using heterogeneous agent-development frameworks, AI models, tool interfaces, protocols, and execution environments. To realize their potential social and business impact, these agents must be able to interoperate through a common communication protocol. The Natural Language Interaction Protocol (NLIP), developed by researchers and practitioners across companies and universities and standardized by Ecma International, addresses this need by defining a standards-based application-layer protocol for AI-agent interaction. NLIP provides a lightweight semantic message envelope that can be carried over existing transports such as HTTP/HTTPS, WebSocket, and AMQP, while allowing NLIP-aware agents and gateways to adapt between clients, agents, local context stores, ontologies, tools, enterprise services, and heterogeneous underlying protocols. This paper presents the motivation and design rationale of NLIP, its message model and transport bindings, security-by-design considerations, reference implementation, representative applications, adoption signals, and relationship to emerging agent protocols such as MCP and A2A.

Chinese Translation

人工智能代理正越来越多地在各组织中开发和部署，使用异构的代理开发框架、人工智能模型、工具接口、协议和执行环境。为了实现其潜在的社会和商业影响，这些代理必须能够通过共同的通信协议进行互操作。自然语言交互协议（Natural Language Interaction Protocol, NLIP）由来自公司和大学的研究人员和实践者共同开发，并由Ecma International进行标准化，旨在满足这一需求，定义了一个基于标准的应用层协议用于人工智能代理之间的交互。NLIP提供了一种轻量级的语义消息封装，可以通过现有的传输方式（如HTTP/HTTPS、WebSocket和AMQP）进行传输，同时允许NLIP感知的代理和网关在客户端、代理、本地上下文存储、本体、工具、企业服务和异构底层协议之间进行适配。本文介绍了NLIP的动机和设计原理、其消息模型和传输绑定、安全设计考虑、参考实现、代表性应用、采用信号以及与新兴代理协议（如MCP和A2A）的关系。

View on arXiv Download PDF AI Translation

cs.AI / 63 / 2609.04141

Efficient Test-Time Adaptation through Human-AI Interaction

通过人机交互实现高效的测试时适应

Wang, Zora Zhiruo, Gandhi, Apurva, Shao, Rulin, Chen, Aspen, Mueller, Jonas, Liang, Zhiqi, Chen, Jett, Ryan, Michael, Ma, Qianou, He, Luxi, Cheng, Zhoujun, He, Andre, Kim, Seungone, Geng, Jiayi, Zheng, Mingqian, Sun, Weiwei, Zhang, Zheyuan, Zhao, Xinran, Wang, Yike, Hou, Abe, Jiang, Liwei, Koh, Pang Wei, Yang, Diyi, Neubig, Graham, Fried, Daniel

Abstract

AI agents are trained on population-scale data to encode broad capabilities spanning those of many practitioners. Yet the artifacts they produce rarely meet the personal bar professionals need to stake their reputation on. On realistic, open-ended tasks where success criteria are heterogeneous and insufficiently documented, individual expertise lives precisely in the elevation and departure from the average. In practice, iterative human-agent interaction surfaces criteria that users cannot fully specify up front, yet apply repeatedly across tasks. We argue this cross-session interaction data is a rich, underused signal for closing the gap to individual expertise. In this work, we propose test-time adaptation through human-agent interaction (TAHI), which integrates these signals into agent context and weights, and crystallizes each user's training and evaluation criteria via an evolving rubric module. We adapt agents to 30 individuals in two high-utility domains, writing and visual creation, on a total of 600 tasks. Our agents improve solo task success by 4.5-20.9% within only tens of tasks. Meanwhile, our evolving rubric module serves as a scalable annotation tool, creating evaluation rubrics that catch 16.0-22.3% more failures than those from LMs or humans alone. While agents are adapted towards individuals, we show these personalized agents also produce improvements in success of up to 8.8% that generalize across users.

Chinese Translation

人工智能代理在大规模数据上进行训练，以编码涵盖众多从业者的广泛能力。然而，它们所产生的成果往往难以满足专业人士所需的个人标准，以维护其声誉。在成功标准异质且记录不足的现实开放任务中，个体专业知识恰恰体现在对平均水平的提升和偏离上。在实践中，迭代的人机交互揭示了用户无法完全事先指定的标准，但这些标准在不同任务中反复适用。我们认为，这种跨会话的交互数据是一个丰富且未被充分利用的信号，可以缩小与个体专业知识之间的差距。在本研究中，我们提出了通过人机交互进行测试时适应（Test-Time Adaptation through Human-Agent Interaction, TAHI），该方法将这些信号整合到代理的上下文和权重中，并通过一个不断演变的评分模块具体化每个用户的训练和评估标准。我们在写作和视觉创作两个高效用领域对30个个体进行了代理适应，共完成600个任务。我们的代理在仅仅几十个任务内，单独任务成功率提高了4.5%-20.9%。同时，我们的演变评分模块作为一个可扩展的注释工具，创建的评估评分标准比仅由语言模型或人类单独提供的标准多捕捉了16.0%-22.3%的失败案例。尽管代理是针对个体进行适应的，但我们展示了这些个性化代理在用户间也能产生高达8.8%的成功率提升。

View on arXiv Download PDF AI Translation

cs.AI / 64 / 2609.04148

Terminal-Universe: Turning Agent Trajectories into Scalable Terminal Environments

终端宇宙：将智能体轨迹转化为可扩展的终端环境

Wu, Jie, Zhang, Zhenru, Zhang, Beichen, Wang, Xuwu, Su, Yuhui, Chen, Mouxiang, Wang, Peng, Wang, Zhihai, Shen, Que, Zhou, Hao, Yang, An, Huang, Fei, Yang, Yujiu, Liu, Dayiheng

Abstract

As terminal-based code agents become prevalent, agent trajectories have accumulated at scale, while realistic, executable environments remain scarce. However, environments are what agent post-training actually requires: each can be re-queried into many verifiable tasks and provides execution feedback, whereas a trajectory is a single frozen demonstration. Rather than generating environments from scratch, we observe that the tool-execution history in existing trajectories exposes the structure and contents of the environments in which they ran, making it possible to reconstruct those environments from the trajectories themselves. Thus, we introduce Terminal-Universe, a framework which turns each trajectory into a reusable environment and explores it for synthesizing new tasks and continued interactions. Specifically, Terminal-Universe replays the file operations recorded in a trajectory to restore each file before the agent modified it, yielding a partial workspace; a completion agent then supplies the missing files and dependencies. On this recovered workspace, we both reconstruct the original intent task and synthesize entirely new ones. Besides, we also scale the tasks along two complementary axes: breadth and depth. For breadth, we mine directional dependency relations between related environments and synthesize cross-workspace queries spanning multiple codebases, as developers routinely do in real-world development. For depth, we extend the initial single-turn query into a multi-round session that captures iterative user feedback and requirement refinement via a user agent. Applied to public terminal agent trajectories, Terminal-Universe produces 37.3k task-sufficient environments. Supervised fine-tuning of Qwen3.5-27B on this corpus improves single-round performance on Terminal-Bench 2.1 by 11.9 points and multi-round performance on EvoCode-Bench v2 MT@4 by 13.8 points.

Chinese Translation

随着基于终端的代码智能体的普及，智能体轨迹在规模上不断积累，而现实可执行的环境却依然稀缺。然而，环境正是智能体后训练所需的：每个环境都可以被重新查询为多个可验证的任务，并提供执行反馈，而轨迹则是一个单一的静态演示。我们观察到，现有轨迹中的工具执行历史揭示了它们运行的环境的结构和内容，这使得从轨迹本身重建这些环境成为可能。因此，我们引入了终端宇宙（Terminal-Universe），一个将每个轨迹转化为可重用环境的框架，并探索其用于合成新任务和持续交互。具体而言，终端宇宙重放轨迹中记录的文件操作，以在智能体修改文件之前恢复每个文件，从而生成一个部分工作区；然后，补全智能体提供缺失的文件和依赖项。在这个恢复的工作区上，我们既重建原始意图任务，也合成全新的任务。此外，我们还沿着两个互补的轴对任务进行扩展：广度和深度。在广度方面，我们挖掘相关环境之间的方向性依赖关系，并合成跨工作区查询，跨越多个代码库，正如开发者在现实开发中常做的那样。在深度方面，我们将初始的单轮查询扩展为多轮会话，通过用户智能体捕捉迭代的用户反馈和需求细化。应用于公共终端智能体轨迹，终端宇宙生成了37.3k个任务充足的环境。在此语料库上对Qwen3.5-27B进行监督微调，使其在Terminal-Bench 2.1上的单轮性能提高了11.9分，在EvoCode-Bench v2 MT@4上的多轮性能提高了13.8分。

View on arXiv Download PDF AI Translation

cs.AI / 65 / 2609.04166

From Deceptive Outputs to Deceptive Mechanisms: A Causal Framework for Language-Model Deception Research

从欺骗性输出到欺骗性机制：语言模型欺骗研究的因果框架

Shkolnikov, Yakov Pyotr

Abstract

Research and news coverage of language-model deception increasingly attributes human-like mental-state concepts to language models. Such claims can blur the distinction between behavior that looks deceptive and a mechanism that is actually deceptive. We introduce a causal taxonomy separating prior commitment from retrospective report, model preference from realized output, false preference from sensitivity to the utility of misleading a recipient, and deceptive behavior from the provenance of the objective or strategy producing it. We test these distinctions in two open-weight model families. Across controlled guessing-game and stock-trading experiments, we find that deceptive-looking behavior can arise without the corresponding proposed mechanism, while other interventions provide direct evidence that recipient information state can causally affect deceptive preference. These results show that deceptive behavior can provide evidence for a deceptive mechanism. But even evidence for such a mechanism does not establish model agency in the deception.

Chinese Translation

对语言模型欺骗的研究和新闻报道越来越多地将类人心理状态的概念归因于语言模型。这些说法可能模糊了看似欺骗的行为与实际上欺骗的机制之间的区别。我们提出了一种因果分类法，将先前的承诺与回顾性报告、模型偏好与实现输出、虚假偏好与对误导接收者的效用敏感性，以及欺骗行为与产生该行为的目标或策略的来源区分开来。我们在两个开放权重模型家族中测试了这些区分。在控制的猜测游戏和股票交易实验中，我们发现，看似欺骗的行为可以在没有相应提议机制的情况下出现，而其他干预则提供了直接证据，表明接收者的信息状态可以因果地影响欺骗偏好。这些结果表明，欺骗行为可以为欺骗机制提供证据。但即使有这样的机制的证据，也并不能确立模型在欺骗中的主动性。

View on arXiv Download PDF AI Translation

cs.AI / 66 / 2609.04170

A Case Study on Emergent Cheating and Whistleblowing in Autonomous Research Swarms

自主研究群体中出现的作弊与举报案例研究

Paglieri, Davide, Cross, Logan, Genewein, Tim, Leibo, Joel Z., Tomasev, Nenad, Vezhnevets, Alexander Sasha

Abstract

Multi-agent AI science ecosystems rely on agents possessing tools that allow them to communicate, coordinate, and build on each other's work. Yet this shared infrastructure can also introduce vulnerabilities by creating a substrate for the contagious spread of unintended and undesirable behaviors. We report a case study on a research collective of 100 autonomous LLM agents tasked with proving formal mathematical conjectures. Within the swarm, cheating spontaneously emerged and was later challenged by whistleblowers - both without any external intervention. When a single agent discovered an exploit in the evaluation system, it propagated across the collective via a shared knowledge library and later through peer-to-peer messages. Despite early reluctance, a cohort of agents adopted the exploit in response to competitive pressure. A separate group of agents produced an emergent counter-response: auditing fraudulent proofs, alerting peers across broadcast and private channels, staging boycotts, lodging formal complaints, and proposing validation patches. In recent incidents, agent swarms coordinated covertly through improvised side-channels (Dalton and Wallace, 2026; Greenblatt et al., 2026). Our setting differs: the same transparent channels that carried the exploit also gave non-cheating agents the visibility they needed to detect fraud, organize resistance, and enforce norms. We cast the problem of managing the agents' shared infrastructure as the knowledge commons governance problem (Ostrom, 1990). To protect the commons from exploits, we propose to adopt institutional mechanisms, such as graduated sanctioning and collective-choice rules, to support decentralized self-governance in autonomous swarms.

Chinese Translation

多智能体人工智能科学生态系统依赖于具备沟通、协调和相互构建工作能力的智能体。然而，这种共享基础设施也可能引入脆弱性，为意外和不良行为的传播提供了基础。我们报告了一项关于100个自主大型语言模型（LLM）智能体组成的研究集体的案例研究，任务是证明正式的数学猜想。在这个群体中，作弊行为自发出现，并随后被举报者挑战——这一切均未受到外部干预。当一个智能体发现评估系统中的漏洞时，它通过共享知识库传播到整个集体，随后通过点对点消息进一步扩散。尽管最初有所顾虑，一组智能体在竞争压力下采用了这一漏洞。另一组智能体则产生了自发的反应：审计虚假证明，通过广播和私人渠道警告同伴，发起抵制，提出正式投诉，并建议验证补丁。在最近的事件中，智能体群体通过即兴的侧信道秘密协调（Dalton 和 Wallace, 2026；Greenblatt 等, 2026）。我们的研究环境有所不同：同样传递漏洞的透明渠道也为非作弊智能体提供了检测欺诈、组织抵抗和执行规范所需的可见性。我们将管理智能体共享基础设施的问题视为知识公地治理问题（Ostrom, 1990）。为了保护公地免受漏洞的侵害，我们建议采用制度机制，如逐级制裁和集体选择规则，以支持自主群体中的去中心化自我治理。

View on arXiv Download PDF AI Translation

cs.AI / 67 / 2609.04172

Rethinking On-Policy Distillation of Large Language Models II: One Training Example

重新思考大型语言模型的在线蒸馏 II：单个训练示例

Fu, Zixuan, He, Bingxiang, Zuo, Yuxin, Huang, Haohuan, Zhang, Jinqian, Xiao, Ruhang, Qian, Cheng, Luo, Qinyu, Gao, Huan-ang, Wang, Yudong, Liu, Zhiyuan, Ding, Ning, Xiao, Chaojun

Abstract

On-policy distillation (OPD) combines student-generated rollouts with dense token-level supervision from a teacher. Existing work has mainly studied its algorithmic behavior, leaving the role of training data unclear. We examine this role at the data-minimal limit by training on a single query. One-shot OPD keeps improving for hundreds of steps and recovers most of full-data OPD's gain across task domains and model families. We explain this result through the states visited during training and the rate at which the student aligns with the teacher. We measure \emph{state coverage}, the fraction of the states full-data OPD visits that a query set's rollouts reach. A single query already reaches $71.5\%$, most of it within the first 100 steps. Adding semantically distinct queries raises coverage and validation accuracy together, until 16 queries reach $98.9\%$ and match full-data training. Yet alignment slows at a similar pace whether OPD trains on one query or the whole dataset, and even a fixed set of states takes hundreds of steps to absorb. OPD is therefore data-overfed but algorithm-starved. Its rollouts quickly expose broad supervision, while the student absorbs that supervision increasingly slowly. The state-coverage result extends to multi-teacher OPD, where 16 semantically diverse queries per domain match full-data MOPD. As a further stress test, content-light templates and off-domain WildChat queries also approach the real-query baseline. Task content and induced state coverage can therefore come apart. We hope these findings direct future work toward the step efficiency of OPD, and prompt a re-examination of the data and the mechanisms behind its recent successes in frontier post-training.

Chinese Translation

在线蒸馏（On-policy Distillation, OPD）结合了学生生成的回滚与来自教师的密集标记级监督。现有研究主要关注其算法行为，未明确训练数据的作用。我们在数据最小限制下，通过对单个查询进行训练来考察这一作用。一-shot OPD 在数百步内持续改进，并在任务领域和模型家族中恢复了大部分全数据 OPD 的收益。我们通过训练过程中访问的状态以及学生与教师对齐的速率来解释这一结果。我们测量了状态覆盖率（state coverage），即全数据 OPD 访问的状态中，查询集的回滚所覆盖的状态比例。单个查询已经达到了 71.5%，其中大部分在前 100 步内实现。添加语义上不同的查询可以同时提高覆盖率和验证准确性，直到 16 个查询达到 98.9% 并与全数据训练相匹配。然而，无论 OPD 是在一个查询上训练还是在整个数据集上训练，对齐的速度都在类似的节奏下减缓，即使是固定的状态集也需要数百步才能被吸收。因此，OPD 是数据过度供给但算法匮乏的。其回滚迅速暴露出广泛的监督，而学生吸收这些监督的速度逐渐减慢。状态覆盖率的结果扩展到多教师 OPD，其中每个领域的 16 个语义多样的查询与全数据 MOPD 匹配。作为进一步的压力测试，内容轻量的模板和域外的 WildChat 查询也接近真实查询基线。因此，任务内容与诱导的状态覆盖率可以分开。我们希望这些发现能引导未来的研究关注 OPD 的步骤效率，并促使重新审视数据及其在前沿后训练中最近成功背后的机制。

View on arXiv Download PDF AI Translation

cs.AI / 68 / 2609.04177

A Computationally Feasible Framework for Causal Probabilistic Explanation

一个计算上可行的因果概率解释框架

Urbaniak, Rafal, Witty, Sam, Waxman, Daniel, Zane, Andy, Garg, Poorva, Bunnapradist, Emily, Vaidyanathan, Sankaran, Feser, Jack, Lehe, Drew, Bingham, Eli

Abstract

Explaining why a specific outcome occurred, and which inputs deserve the blame or credit, is central to philosophical, scientific, and policy analysis. Existing tools split into two camps. The theory of actual causality (AC) gives principled verdicts, but only for toy-sized models, because computing them requires enumerating counterfactual scenarios. Scalable attribution methods like SHAP (or even causal SHAP) at least partially ignore the causal structure that generated the data, and can give answers that conflict with a careful causal analysis. We close this gap with Probabilistic Causal Impact (PCI). PCI builds on actual causality and on Pearl's notions of probability of necessity and sufficiency, but recasts the question of explainability as an estimation problem on a probabilistic causal model that is easily approximated via Monte Carlo. By specifying a distribution over "candidate explanations," a distribution over counterfactual values, and a scoring function, PCI provides tractable, causally grounded, graded explanations, generalizing AC and Pearl's probability of causation as degenerate cases. We evaluate PCI in synthetic and real-world examples, spanning consistency checks with AC, scaling experiments, complex continuous-valued dynamical systems, and a real-world deployed causal machine learning model trained on millions of datapoints.

Chinese Translation

解释为何特定结果发生，以及哪些输入应当承担责任或获得功劳，是哲学、科学和政策分析中的核心问题。现有工具分为两类。实际因果性理论（Actual Causality, AC）提供了原则性的判决，但仅适用于玩具模型，因为计算它们需要枚举反事实场景。可扩展的归因方法如SHAP（或甚至因果SHAP）至少在某种程度上忽略了生成数据的因果结构，可能给出与细致的因果分析相矛盾的答案。我们通过概率因果影响（Probabilistic Causal Impact, PCI）填补了这一空白。PCI基于实际因果性和Pearl关于必要性和充分性概率的概念，但将可解释性的问题重新表述为在一个可以通过蒙特卡洛方法轻松近似的概率因果模型上的估计问题。通过指定“候选解释”的分布、反事实值的分布以及评分函数，PCI提供了可处理的、基于因果的、分级的解释，推广了AC和Pearl的因果概率作为退化情况。我们在合成和真实世界的例子中评估了PCI，包括与AC的一致性检查、扩展实验、复杂的连续值动态系统，以及在数百万数据点上训练的实际部署的因果机器学习模型。

View on arXiv Download PDF AI Translation

cs.AI / 69 / 2609.04198

Clean Engineering, Unstable Measurement: A Preregistered Reliability Failure of Black-Box LLM Observers on Shared Endpoints

清洁工程，不稳定的测量：黑箱大型语言模型观察者在共享端点上的预注册可靠性失败

Zhu, Haoyaun, Zhang, Jie

Abstract

Language-model judges now gate training data, score generations, and drive leaderboards. The judge is then a measurement instrument, resting on one rarely stated assumption: the same request, sent to the same model name, reads the same tomorrow. We audited that assumption in two preregistered campaigns with every threshold fixed in advance; neither got past validating its instrument. Across 52,988 audited request attempts, same-window repeat rankings agreed at Spearman 0.400 against a required 0.90, and byte-identical next-day replays agreed at 0.78 against a required 0.99, each time with the execution record at ceiling. Three mechanisms explain the gap: a label-to-meaning mapping that biased readouts as strongly as the signal; candidate gaps seven orders of magnitude below the instrument's own noise floor; and byte-identical inputs returning different rankings, a noise that exact-permutation readouts compound. Neither metric substitution nor sampling repaired it on the tested grid. Preregistered follow-ups bound the problem: waiting did not help on the days sampled (0.805 versus 0.800, replicated over five further days); switching providers did not help (four providers share the floor, medians 0.74 to 0.88, predicted by none of the metadata fields they expose); self-hosting on batch-invariant kernels helped only while the server was quiet; and on constructed errors with known gaps, the readout's separation tracks error type, not size. We distill the evidence into a three-level snapshot-identity ladder, eight design rules, and a reporting checklist; a pilot at roughly 2% of the study's call volume would have exposed both unreachable gates in advance. All results concern externally measured behaviour on shared serving infrastructure. On a shared endpoint, a model name is not a frozen instrument; a preregistered evaluation must measure its instrument before freezing any gate on it.

Chinese Translation

语言模型评判者现在控制训练数据、评分生成结果并推动排行榜。评判者因此成为一种测量工具，基于一个很少被提及的假设：相同的请求，发送到相同的模型名称，明天的结果也应相同。我们在两个预注册的实验中审计了这一假设，所有阈值均提前固定；但都未能验证其测量工具。在52,988次审计请求尝试中，同窗口的重复排名以斯皮尔曼相关系数0.400达不到所需的0.90，而字节完全相同的次日重放结果以0.78达不到所需的0.99，每次执行记录均处于上限。三个机制解释了这一差距：标签到意义的映射使读数受到强烈偏倚；候选差距低于测量工具自身噪声水平七个数量级；以及字节完全相同的输入返回不同的排名，这种噪声在精确排列的读数中被放大。无论是度量替代还是采样都未能修复在测试网格上的问题。预注册的后续研究界定了问题：在采样的日子里等待并没有帮助（0.805对比0.800，在五个后续日子中重复）；更换提供商也没有帮助（四个提供商共享同一基础设施，中位数为0.74到0.88，均未被它们公开的元数据字段预测）；在批量不变内核上自托管仅在服务器安静时有效；在已知差距的构造错误上，读数的分离跟踪错误类型，而非大小。我们将证据提炼为一个三层快照身份阶梯、八条设计规则和一个报告检查表；在研究调用量的约2%的试点中，将提前暴露两个无法到达的门。所有结果均涉及在共享服务基础设施上外部测量的行为。在共享端点上，模型名称并不是一个固定的工具；预注册评估必须在冻结任何门之前测量其工具。

View on arXiv Download PDF AI Translation

计算语言学 (Computation and Language)

cs.CL / 1 / 2609.02889

Where Does Harness-Optimization Value Live? Localized Gains and the Budget-Splitting Trap in Self-Evolving LLM Agents

优化价值的源头在哪里？自我演化的 LLM 代理中的局部收益与预算分配陷阱

Nguyen, Michael, Tan, Wei Chen, Hassan, Nurul Aisyah, Raman, Arvind, Lim, Li Hua, Razak, Ahmad Faiz

Abstract

A growing body of work improves frozen large language models (LLMs) as agents by evolving their harness: the textual scaffolding around the model, including persona, strategy, format rules, and control heuristics. Existing reflective prompt-evolution methods usually optimize this harness as one flat string. We instead ask where the optimization value actually resides. We introduce HARNESSEVO, which decomposes the harness into four separately evolvable slots: role, task-strategy, tool/format-rules, and reflection/control. Using the same reflective optimizer under an iso-budget setting, we pair this decomposition with leave-one-in and leave-one-out attribution to measure the contribution of each slot. On ALFWorld with a frozen 7B backbone, HARNESSEVO does not significantly improve the overall binary success rate over either the stock harness or flat-string evolution: 0.657 versus 0.642 and 0.642, respectively. However, the slot-level analysis reveals that nearly all useful optimization value is localized in the reflection/control slot, which achieves a leave-one-in gain of +0.119. The other slots are individually null. We further show that uniform budget splitting is harmful: allocating 64 rollouts across four slots leaves only 16 per slot, below the optimizer's effective search floor, causing every slot to freeze at its empty seed. Concentrating the budget on the high-credit control slot recovers the lost gain, reaching 0.761 with half the split budget. The effect is task-contingent. On WebShop, all slots freeze empty and all methods tie, indicating a genuine absence of recurrent, verbalizable control failures rather than budget starvation. Overall, our results suggest that harness value is localized, uniform budget splitting can be actively harmful, and credit assignment should precede structured agent-evolution.

Chinese Translation

越来越多的研究通过演化模型周围的文本框架（即“哈尼斯”）来提升冻结的大型语言模型（LLMs）作为代理的能力，这些文本框架包括角色、策略、格式规则和控制启发式方法。现有的反思性提示演化方法通常将这一哈尼斯作为一个平坦的字符串进行优化。我们则提出了一个问题：优化价值实际上存在于何处。我们引入了 HARNESSEVO，它将哈尼斯分解为四个可单独演化的部分：角色、任务策略、工具/格式规则和反思/控制。在相同的预算设置下，我们使用相同的反思优化器，将这种分解与“留一在内”和“留一在外”的归因方法相结合，以测量每个部分的贡献。在使用冻结的 7B 主干的 ALFWorld 上，HARNESSEVO 在整体二元成功率上并未显著优于原始哈尼斯或平坦字符串演化：分别为 0.657、0.642 和 0.642。然而，部分级别的分析表明，几乎所有有用的优化价值都集中在反思/控制部分，该部分的留一在内增益为 +0.119。其他部分则单独为零。我们进一步表明，均匀的预算分配是有害的：在四个部分之间分配 64 次回合，每个部分仅剩 16 次，低于优化器的有效搜索底线，导致每个部分都在其空种子处冻结。将预算集中在高信用的控制部分可以恢复失去的增益，达到 0.761，且预算分配仅为一半。该效应与任务相关。在 WebShop 上，所有部分均冻结为空，所有方法均平局，表明真正缺乏重复的、可言语化的控制失败，而不是预算匮乏。总体而言，我们的结果表明哈尼斯的价值是局部的，均匀的预算分配可能是主动有害的，信用分配应优先于结构化的代理演化。

View on arXiv Download PDF AI Translation

cs.CL / 2 / 2609.02890

Bounded Personas Match Retrieval on Classification but Not Regression for a Frozen Agent

有限人格在分类任务中匹配检索，但在回归任务中不匹配的冷冻代理

Yoon, JaeHa, Park, Minjun, Kim, Seoyeon, Lee, Jiwoo, Choi, Hyunwoo, Kang, Dohyun

Abstract

A personalized language agent must convert a user's interaction history into behavior on each new request at inference time. Two strategies dominate. Retrieval pulls a few of the user's most relevant past items into the prompt, which is accurate but pays a per-query selection and context cost that grows with the history. Distillation instead compresses the history once into a compact natural-language persona, which is bounded, query-independent, and interpretable, but is widely assumed to sacrifice accuracy. Whether, and on which tasks, a distilled persona can match retrieval has not been characterized cleanly. We introduce PersonaLink, a training-free method that distills a user's history into a bounded three-field persona and recursively refines it: each pass self-evaluates the frozen agent on a held-out slice of the user's own labeled history, rewrites the persona from its errors, and keeps the result only when it does not regress on that slice. Because every comparison shares one frozen 7B backbone and differs only in what is placed in context, the design isolates the effect of representation from that of the model. The result is a clear task-type asymmetry. On 200 users of LaMP-2 (15-way news categorization), PersonaLink reaches 0.745-0.755 accuracy, statistically indistinguishable from BM25 retrieval (0.760-0.765).

Chinese Translation

个性化语言代理必须在推理时将用户的交互历史转换为每个新请求的行为。两种策略占主导地位。检索策略将用户最相关的过去项目提取到提示中，这种方法准确但在每次查询时会产生随着历史增长的选择和上下文成本。相反，蒸馏策略将历史压缩为一个紧凑的自然语言人格，这种人格是有限的、与查询无关的且可解释的，但普遍认为会牺牲准确性。蒸馏人格是否能够在某些任务上匹配检索尚未得到清晰的表征。我们提出了PersonaLink，这是一种无训练的方法，将用户的历史蒸馏为一个有限的三字段人格，并递归地对其进行精炼：每次迭代在用户自己标记历史的保留片段上自我评估冷冻代理，基于错误重写人格，并仅在不出现回归的情况下保留结果。由于每次比较共享一个冷冻的7B主干网络，仅在上下文中放置的内容不同，因此该设计将表示的影响与模型的影响隔离开来。结果显示出明显的任务类型不对称性。在200名LaMP-2用户（15类新闻分类）中，PersonaLink的准确率达到0.745-0.755，统计上与BM25检索（0.760-0.765）无显著差异。

View on arXiv Download PDF AI Translation

cs.CL / 3 / 2609.02892

Counterexamples as Feedback for Agent Self-Correction

反例作为智能体自我修正的反馈

Badrinarayan, Sidhesh, Parthasarathy, Adithya

Abstract

Single-turn code-generation metrics understate a central property of deployed agents: whether they can repair a wrong artifact after receiving concrete feedback. This paper presents A-CEGIS, a lightweight framework that uses counterexamples as feedback for evaluating multi-turn refinement in natural-language-to-regex synthesis. An agent proposes a regex, a deterministic oracle checks it under full-match semantics, and compact false-positive or false-negative witnesses guide the next turn. On 30 NL-RX-Turk tasks, diagnostic counterexample feedback solves 90\% of tasks within a four-turn ablation budget, compared with 17% for zero-shot generation, 27% for generic self-correction, and 23% for error-only feedback. In a full diagnostic run with hardening, all tasks are solved on the hidden set by the final turn, with mean time-to-success of 2.7 turns and robust success of 77% after targeted probing. These results show that A-CEGIS measures how efficiently an agent improves across turns while adding a practical robustness check beyond the original held-out cases.

Chinese Translation

单轮代码生成指标低估了已部署智能体的一个核心特性：它们在收到具体反馈后是否能够修复错误的产物。本文提出了 A-CEGIS，一个轻量级框架，利用反例作为反馈来评估自然语言到正则表达式（regex）合成中的多轮精炼。智能体提出一个正则表达式，确定性神谕在完全匹配语义下对其进行检查，紧凑的假阳性或假阴性证据引导下一轮。在 30 个 NL-RX-Turk 任务中，诊断性反例反馈在四轮消融预算内解决了 90\% 的任务，而零-shot 生成仅为 17\%，通用自我修正为 27\%，仅错误反馈为 23\%。在完整的诊断运行中，通过强化，所有任务在最后一轮时都在隐藏集上得到解决，平均成功时间为 2.7 轮，经过针对性探测后成功率达 77\%。这些结果表明，A-CEGIS 衡量了智能体在多轮中提高效率的能力，同时在原始保留案例之外增加了实用的鲁棒性检查。

View on arXiv Download PDF AI Translation

cs.CL / 4 / 2609.02893

Probe Generalization as Subspace Selection for OOD Deception Detection

作为子空间选择的探针泛化用于OOD欺骗检测

Yoo, Daniel, Skapars, Adrians

Abstract

Linear probes can be used to detect behaviors and concepts inside language model activations, but may fail to transfer to out-of-distribution examples. When studying the generalization performance of Llama-3.1-8B-Instruct probes over 3 held-out deception detection datasets, we find that projecting inputs onto a small subset of principal components (PCs) from the training distribution of activations enables cross-domain transfer that nearly matches the performance of probes trained directly on the test distribution. Furthermore, we find that PC interpretations can be used to find a subset of those transferable PCs. By using an LLM judge to score each PC on whether its most/ least activating examples imply a transferable deception direction, then probing on the highest-scoring PCs, we close the baseline-to-oracle gap by 78% on Insider Trading Report and by 25% on Sandbagging. The directions a source probe weights heavily appear to encode source-specific surface features, while the directions that actually transfer appear to encode the same contrast more abstractly, in a way natural language descriptions can capture. Broadly, our results suggest that the OOD robustness of probes is largely determined by subspace selection.

Chinese Translation

线性探针可用于检测语言模型激活中的行为和概念，但可能无法有效迁移到分布外（OOD）样本。在研究Llama-3.1-8B-Instruct探针在三个未见欺骗检测数据集上的泛化性能时，我们发现将输入投影到训练分布激活的少数主成分（PCs）子集上，能够实现几乎匹配直接在测试分布上训练的探针性能的跨域迁移。此外，我们发现主成分的解释可以用来筛选出可迁移的PC子集。通过使用大型语言模型（LLM）评审者对每个PC的最高/最低激活示例是否暗示可迁移的欺骗方向进行评分，然后仅在得分最高的PC上进行探测，我们在Insider Trading Report数据集上缩小了基线与理想状态之间的差距78%，在Sandbagging数据集上缩小了25%。源探针权重较大的方向似乎编码了源域特定的表面特征，而实际可迁移的方向则以更抽象的方式编码相同的对比，这种方式可以被自然语言描述捕捉。总体而言，我们的结果表明探针的OOD鲁棒性在很大程度上取决于子空间的选择。

View on arXiv Download PDF AI Translation

cs.CL / 5 / 2609.02894

R$^{2}$Adapter: A Routing and Rewriting Adapter for Efficient Hybrid RAG

R$^{2}$Adapter：一种用于高效混合检索增强生成的路由与重写适配器

Guo, Yucan, Su, Miao, Guan, Saiping, Bai, Long, Hou, Zhongni, Li, Zixuan, Jin, Xiaolong, Guo, Jiafeng, Cheng, Xueqi

Abstract

Retrieval-Augmented Generation (RAG) has become a prevailing paradigm for enhancing Large Language Models (LLMs) with non-parametric knowledge. Vanilla RAG efficiently handles simple queries but struggles with relational or multi-hop reasoning. Graph-based RAG alleviates this issue but incurs higher inference complexity and latency. In practice, user queries can differ significantly in their complexity, rendering a fixed RAG strategy suboptimal. However, existing hybrid text-graph RAG methods typically rely on heuristic and LLM-based routing, resulting in unnecessary overhead and strong dependence on the underlying LLM. To address these challenges, we propose R$^{2}$Adapter, a lightweight plug-in Routing and Rewriting Adapter designed to allocate queries between vanilla and graph-based RAG dynamically. By routing only the queries that genuinely benefit from graph-based reasoning, R$^{2}$Adapter reduces unnecessary graph retrieval overhead. Additionally, uncertain graph-routed queries are rewritten to better expose their multi-hop reasoning requirements, improving retrieval quality without additional supervision. Extensive experiments on three multi-hop QA benchmarks demonstrate that R$^{2}$Adapter reduces graph-based RAG usage by up to 59% while maintaining comparable answer accuracy. This adapter is model-agnostic and can be seamlessly integrated into diverse vanilla and graph-based RAG pipelines, providing an efficient and adaptive solution for hybrid RAG systems.

Chinese Translation

检索增强生成（RAG）已成为增强大型语言模型（LLMs）与非参数知识的主要范式。传统的 RAG 能够高效处理简单查询，但在关系或多跳推理方面表现不佳。基于图的 RAG 缓解了这一问题，但带来了更高的推理复杂性和延迟。在实际应用中，用户查询的复杂性可能存在显著差异，这使得固定的 RAG 策略并不理想。然而，现有的混合文本-图 RAG 方法通常依赖于启发式和基于 LLM 的路由，导致不必要的开销和对基础 LLM 的强依赖。为了解决这些挑战，我们提出了 R$^{2}$Adapter，一种轻量级的插拔式路由与重写适配器，旨在动态分配传统 RAG 和基于图的 RAG 之间的查询。通过仅路由那些真正受益于基于图推理的查询，R$^{2}$Adapter 减少了不必要的图检索开销。此外，对于不确定的图路由查询，进行重写以更好地揭示其多跳推理需求，从而在没有额外监督的情况下提高检索质量。在三个多跳问答基准上的广泛实验表明，R$^{2}$Adapter 将基于图的 RAG 使用减少了多达 59%，同时保持了相当的答案准确性。该适配器是模型无关的，可以无缝集成到多种传统和基于图的 RAG 流水线中，为混合 RAG 系统提供高效且自适应的解决方案。

View on arXiv Download PDF AI Translation

cs.CL / 6 / 2609.02895

BharatGather: A Culturally-Informed Benchmark Dataset for Misinformation and Fake News Detection in Indian Public Events

BharatGather：一个针对印度公共事件中虚假信息和假新闻检测的文化信息基准数据集

Bramhecha, Parth, Deshmukh, Smit, Bodhale, Sairaj, Borate, Adwait, Joshi, Raviraj

Abstract

Large-scale public events, such as religious festivals, political rallies, and cultural gatherings, are increasingly vulnerable to the rapid dissemination of misinformation, posing substantial risks to public safety and social cohesion. While automated fake news detection has seen significant methodological progress, existing benchmarks frequently fail to capture the socio-cultural nuances and event-specific dynamics characteristic of the Indian context. This paper introduces BharatGather, a curated, multi-source dataset specifically engineered for binary misinformation classification within the ecosystem of Indian mass gatherings. The corpus comprises 14,646 records constructed through a hybrid pipeline involving systematic web scraping of prominent fact-checking platforms, multimedia transcript extraction, and Large Language Model (LLM)-mediated synthetic augmentation to ensure narrative diversity. By providing a resource tailored to the unique complexities of event-aware misinformation in India, this work facilitates the development of culturally informed detection systems and establishes a rigorous benchmark for evaluating their performance in high-stakes public environments.

Chinese Translation

大型公共事件，如宗教节日、政治集会和文化聚会，越来越容易受到虚假信息快速传播的影响，这对公共安全和社会凝聚力构成了重大风险。尽管自动化假新闻检测在方法论上取得了显著进展，但现有基准往往无法捕捉到印度背景下的社会文化细微差别和事件特定动态。本文介绍了BharatGather，一个专门为印度大型聚会生态系统中的二元虚假信息分类而设计的多源数据集。该语料库由14,646条记录构成，通过系统性地从主要事实核查平台进行网络抓取、多媒体转录提取，以及利用大型语言模型（Large Language Model, LLM）进行合成增强，以确保叙事的多样性。通过提供一个针对印度事件相关虚假信息独特复杂性的资源，本研究促进了文化信息化检测系统的发展，并为在高风险公共环境中评估其性能建立了严格的基准。

View on arXiv Download PDF AI Translation

cs.CL / 7 / 2609.02896

PiPMRE: A Pipeline Based on Language Model for Medical Relation Extraction

PiPMRE：基于语言模型的医学关系提取管道

Duan, Jiaxin, Lu, Fengyu, Liu, Junfei

Abstract

Medical relation extraction (MRE) is commonly known for extracting entities and their relations jointly from a medical text, which has attracted considerable attention in recent years. Previous studies treat MRE as a sequence tagging task, which results in either a challenging design of the tagging schema or a failed extraction of multiple relations, due to intricate relationships among medical entities. In this work, we review the task from the linguistic perspective and propose a novel pipeline framework, PiPMRE, developed on language models to enhance MRE performance. Specifically, PiPMRE consists of a relation generator and a relation filter. Given a text, the generator first yields multiple relational triplets, and then the filter scores each triplet and retains only those that pass the borderline as the final results. Implementing PiPMRE requires no tagging schema; instead, we use a simple template to reformulate the input text, ensuring that entities and relations are generated in a contextual order. Extensive experimental results on two public datasets demonstrate the advancement of PiPMRE. It surpasses the previous state-of-the-art by an average of 5.6 recall points and 4.4 accuracy points. PiPMRE's superiority is also demonstrated in few-shot settings.

Chinese Translation

医学关系提取（MRE）通常被认为是从医学文本中共同提取实体及其关系的任务，近年来引起了相当大的关注。以往的研究将MRE视为序列标注任务，这导致了标注方案设计的挑战性或多重关系提取的失败，因为医学实体之间的关系复杂。在本研究中，我们从语言学的角度回顾了这一任务，并提出了一种新颖的管道框架PiPMRE，该框架基于语言模型以提升MRE性能。具体而言，PiPMRE由关系生成器和关系过滤器组成。给定一段文本，生成器首先产生多个关系三元组，然后过滤器对每个三元组进行评分，仅保留那些通过边界线的三元组作为最终结果。实施PiPMRE不需要标注方案；相反，我们使用简单的模板重新构造输入文本，确保实体和关系以上下文顺序生成。在两个公共数据集上的大量实验结果表明，PiPMRE的性能有所提升，平均超越了之前的最先进水平5.6个召回点和4.4个准确点。PiPMRE在少样本设置中的优越性也得到了验证。

View on arXiv Download PDF AI Translation

cs.CL / 8 / 2609.02897

Margins, Not Windows: Training-Free Per-Step Lossy Speculative Decoding

边际，而非窗口：无训练的逐步有损推测解码

Urbán, Oszkár, Kwon, Young D., Venieris, Stylianos I., Mascolo, Cecilia

Abstract

Speculative decoding accelerates LLM inference by drafting candidate tokens and verifying them in parallel. Tree-attention drafters such as EAGLE-3 are widely adopted, yet typically hold two decisions fixed: (1) a strict token-match verification rule and (2) a static draft-tree shape. Prior work relaxes each in isolation under limiting assumptions: long draft chains for training-free lossy verification, and adaptive tree shaping under a fixed token budget. We introduce AdaptiveSpec, a training-free per-step speculative decoding method that adapts both decisions from internal signals already produced during decoding. A per-step margin rule promotes a mismatched draft-proposed token when the ratio of the target's probability on the drafted token to its top-1 probability exceeds a threshold with no dependence on draft length or underlying drafter architecture. A per-step tree policy adjusts the draft tree's depth, width, and node count directly from a fused signal of draft top-1 confidence and a rolling acceptance history capturing recent draft-target agreement, allowing the total draft count to vary rather than only be redistributed. The two adaptations operate on orthogonal axes and compound in effect. Implemented on the SGLang production-grade serving engine, AdaptiveSpec improves throughput over the state-of-the-art autoregressive speculative decoding method EAGLE-3 by up to 56%, recovering 93% to fully lossless task accuracy across GSM8K, MATH-500, and HumanEval on three target models (DeepSeek-R1-Distill-Llama-8B, Llama-3.1-8B-Instruct, Qwen3-8B).

Chinese Translation

推测解码通过并行草拟候选标记并验证它们来加速大语言模型（LLM）的推理。树注意力草拟器如 EAGLE-3 被广泛采用，但通常固定了两个决策：（1）严格的标记匹配验证规则和（2）静态草拟树形状。之前的工作在有限假设下分别放宽了这两个限制：针对无训练的有损验证的长草拟链，以及在固定标记预算下的自适应树形状。我们提出了 AdaptiveSpec，这是一种无训练的逐步推测解码方法，能够根据解码过程中已经产生的内部信号自适应调整这两个决策。逐步边际规则在目标的草拟标记的概率与其最高概率的比率超过阈值时，促进不匹配的草拟提议标记，而不依赖于草拟长度或基础草拟器架构。逐步树策略直接根据草拟的最高置信度和捕捉近期草拟与目标一致性的滚动接受历史的融合信号，调整草拟树的深度、宽度和节点数量，从而允许总草拟计数变化，而不仅仅是重新分配。这两个适应在正交轴上操作并产生复合效果。在 SGLang 生产级服务引擎上实现的 AdaptiveSpec，相较于最先进的自回归推测解码方法 EAGLE-3 提高了高达 56% 的吞吐量，并在三个目标模型（DeepSeek-R1-Distill-Llama-8B、Llama-3.1-8B-Instruct、Qwen3-8B）上恢复了 GSM8K、MATH-500 和 HumanEval 的 93% 到完全无损的任务准确率。

View on arXiv Download PDF AI Translation

cs.CL / 9 / 2609.02898

Distilled Rapid Embedding Transfer (DRET): Parameter-Efficient Biomedical Domain Adaptation via Priority-Based Embedding Transfer

蒸馏快速嵌入转移（DRET）：基于优先级的嵌入转移实现参数高效的生物医学领域适应

Sundaram, Girish, Berleant, Daniel

Abstract

Large domain-specific language models such as BioBERT and ClinicalBERT achieve strong performance on biomedical NLP tasks, but their computational demands make them impractical for many real-world deployments. General-purpose, parameter-efficient models such as DistilBERT are lightweight yet lack the domain knowledge required for specialized tasks such as PICO (Population, Intervention, Comparison, Outcome) classification. We introduce Distilled Rapid Embedding Transfer (DRET), a knowledge-transfer paradigm that injects biomedical domain knowledge from large specialized models into a smaller general-purpose model without retraining on the original specialized corpora. DRET is developed as an iterative family of strategies: a unified tokenizer-merge strategy (DRET 1.x), hybrid embedding averaging (DRET 2.0), and a priority-based embedding-transfer mechanism (DRET 3.x) that hierarchically selects embeddings from the most authoritative source models, further combined with embedding-layer freezing, differential learning rates, label propagation, and imbalance-aware loss functions (DRET 4.x). We evaluate DRET on token-level PICO classification using the EBM-NLP corpus under severe class imbalance, across a twelve-metric battery. DRET-enhanced DistilBERT (66M parameters) attains balanced accuracy, recall, and ROC-AUC competitive with, and on several class-wise metrics exceeding, models an order of magnitude larger, while retaining DistilBERT's efficiency. We further show that transfer occurs at the embedding level through cosine-similarity, semantic-shift, and t-SNE analyses. DRET offers a scalable, resource-efficient route to near-domain-expert performance for biomedical text mining, with direct application to automated systematic literature reviews and clinical decision support.

Chinese Translation

大型领域特定语言模型，如BioBERT和ClinicalBERT，在生物医学自然语言处理任务中表现出色，但其计算需求使其在许多实际应用中不切实际。通用的、参数高效的模型，如DistilBERT，虽然轻量，但缺乏进行PICO（人群、干预、比较、结果）分类等专业任务所需的领域知识。我们提出了蒸馏快速嵌入转移（DRET），一种知识转移范式，它将大型专业模型中的生物医学领域知识注入到较小的通用模型中，而无需在原始专业语料库上进行再训练。DRET作为一系列迭代策略开发：统一的分词器合并策略（DRET 1.x）、混合嵌入平均（DRET 2.0），以及基于优先级的嵌入转移机制（DRET 3.x），该机制从最权威的源模型中分层选择嵌入，并进一步结合嵌入层冻结、差异学习率、标签传播和关注不平衡的损失函数（DRET 4.x）。我们在严重类别不平衡的情况下，使用EBM-NLP语料库对DRET进行评估，针对标记级PICO分类，采用十二项指标进行测试。DRET增强的DistilBERT（66M参数）在平衡准确率、召回率和ROC-AUC方面与多个数量级更大的模型竞争，并在若干类别指标上超越这些模型，同时保持DistilBERT的高效性。我们进一步通过余弦相似度、语义偏移和t-SNE分析展示了嵌入级别的转移。DRET为生物医学文本挖掘提供了一条可扩展、资源高效的途径，以接近领域专家的表现，直接应用于自动化系统文献综述和临床决策支持。

View on arXiv Download PDF AI Translation

cs.CL / 10 / 2609.02899

Contamination Inflates Scores but Rarely Reorders Large Language Model Leaderboards

污染膨胀分数但很少重新排序大型语言模型排行榜

Xiao, Xingyao, Cheng, Yihong

Abstract

Benchmark contamination, the leakage of test items into training data, is widely described as a threat to the reliability of large language model (LLM) leaderboards. We argue that this concern conflates two distinct questions: whether contamination inflates absolute scores, and whether it reorders the ranking of models. We recast contamination as a violation of anchor-item invariance and measure it through the differential functioning of original versus semantically equivalent paraphrased items, a within-item contrast that holds the measured skill fixed and isolates memorization from capability. Using per-instance responses from 47 publicly released models and 74 models finetuned with a known dose of contamination, across four benchmarks (ARC, GSM8K, HellaSwag, MMLU), we first calibrate the measure against ground truth: it recovers injected contamination dose-responsively (a corrected effect of +0.187 accuracy points for test-set leakage) and never flags a negative-control model trained only on the legitimate training split (-0.012). We then quantify leaderboard impact: the rank correlation between a standard leaderboard and a paraphrase-controlled leaderboard is 0.997, and a sensitivity analysis shows that the observed differential contamination is far below the level needed to move rankings, with only 3 of 188 model-by-benchmark cases showing differential contamination corroborated across two references. Contamination among these public models is therefore largely uniform: it inflates absolute scores without reordering the leaderboard, and ranking distortion requires the rare case of differential contamination. We provide a calibrated invariance audit, released as a reference implementation, and recommend that leaderboards report paraphrase-controlled rankings alongside confidence intervals.

Chinese Translation

基准污染，即测试项目泄漏到训练数据中，被广泛认为是对大型语言模型（LLM）排行榜可靠性的威胁。我们认为，这一担忧混淆了两个不同的问题：污染是否膨胀了绝对分数，以及它是否重新排序了模型的排名。我们将污染重新表述为对锚定项目不变性的违反，并通过原始项目与语义等效的改述项目的差异性表现来测量污染，这是一种在项目内部对比，固定测量技能并将记忆与能力隔离开来。使用来自47个公开发布模型的每个实例响应和74个经过已知污染剂量微调的模型，在四个基准（ARC、GSM8K、HellaSwag、MMLU）上，我们首先将该测量与真实情况进行校准：它能够根据注入的污染剂量进行恢复（测试集泄漏的修正效果为+0.187准确度点），且从未标记出仅在合法训练分割上训练的负对照模型（-0.012）。然后，我们量化排行榜的影响：标准排行榜与改述控制排行榜之间的排名相关性为0.997，灵敏度分析表明，观察到的差异性污染远低于移动排名所需的水平，只有188个模型与基准案例中有3个显示出经过两个参考验证的差异性污染。因此，这些公共模型之间的污染在很大程度上是均匀的：它膨胀了绝对分数而没有重新排序排行榜，而排名扭曲需要罕见的差异性污染案例。我们提供了一个经过校准的不变性审计，作为参考实现发布，并建议排行榜报告改述控制的排名及其置信区间。

View on arXiv Download PDF AI Translation

cs.CL / 11 / 2609.02901

Dual-Form ASR: Semantics-Aware Inverse Text Normalization for Chinese Speech Recognition

双重形式自动语音识别：面向语义的中文语音识别逆文本规范化

Zhang, Fengrun, Fu, Li, Zhou, Wangjin, Fan, Lu, Wu, Youzheng, He, Xiaodong

Abstract

Modern automatic speech recognition (ASR) scenarios require both spoken-form transcripts for faithful transcription and readable written-form transcripts with inverse text normalization (ITN). However, these forms are typically produced by cascaded modules, where a spoken-form ASR output is rewritten by a separate ITN component, making written-form ASR-ITN vulnerable to recognition errors and decoupling normalization from acoustic-contextual modeling, especially for semantically dependent numeric expressions. In this paper, we propose Dual-Form ASR (DF-ASR), a framework that extends spoken-form ASR capability to semantics-aware written-form ITN through paired spoken-form and written-form supervision while retaining prompt-level selection between transcript forms. The dual-form supervision is constructed via a large language model (LLM)-driven generate-and-judge workflow, and training is further enhanced by ITN-MWER, a sequence-level objective that assigns higher cost to errors on normalization-sensitive spans. We also introduce a decision-aware REQUIRE-ITN/\FORBID-ITN protocol to separately measure required normalization and forbidden-span preservation. On manually annotated Chinese subsets from SpeechIO, DF-ASR consistently outperforms open-source ASR-ITN systems, remains competitive with strong closed-source references, and preserves reliable prompt-level control between spoken-form and written-form outputs.

Chinese Translation

现代自动语音识别（ASR）场景需要同时提供忠实转录的口语形式文本和可读的书面形式文本，并进行逆文本规范化（ITN）。然而，这两种形式通常由级联模块生成，其中口语形式的ASR输出由单独的ITN组件重写，这使得书面形式的ASR-ITN容易受到识别错误的影响，并且将规范化与声学上下文建模解耦，尤其是在语义依赖的数字表达方面。本文提出了双重形式自动语音识别（DF-ASR）框架，该框架通过配对的口语形式和书面形式监督，将口语形式ASR能力扩展到面向语义的书面形式ITN，同时保留了转录形式之间的提示级选择。双重形式监督通过大型语言模型（LLM）驱动的生成与判断工作流程构建，训练进一步通过ITN-MWER增强，该序列级目标对规范化敏感区间的错误赋予更高的成本。我们还引入了一种决策感知的REQUIRE-ITN/ FORBID-ITN协议，以分别测量所需的规范化和禁止区间的保留。在SpeechIO的手动标注中文子集上，DF-ASR始终优于开源ASR-ITN系统，并在与强大的闭源参考系统的竞争中保持竞争力，同时在口语形式和书面形式输出之间保持可靠的提示级控制。

View on arXiv Download PDF AI Translation

cs.CL / 12 / 2609.02902

RL-ADA: A World-Feedback Framework for Adversarially Robust Enterprise Dialogue Agents

RL-ADA：一种针对对抗性鲁棒企业对话代理的世界反馈框架

Narayanan, Ram, Rajgarhia, Harshit, Mukherji, Abhishek

Abstract

Deploying task-oriented dialogue agents in enterprise customer support faces a persistent annotation bottleneck: robust training requires labelled interaction data at scale, yet enterprise conversational logs are privacy-sensitive and expensive to annotate, while user behaviour evolves faster than labelling pipelines can keep pace. We present RL-ADA (Reinforcement Learning with Adversarial Dialogue Agents), a co-evolutionary training framework that eliminates this bottleneck by replacing human labels with \emph{world feedback}: consequence-based reward signals derived directly from measurable interaction outcomes. A Customer Support Agent (DA, 3B parameters) and an Adversarial Customer Agent (CA, 7B parameters) co-evolve in an adversarial arena guided by a fixed automated judge: the DA is rewarded for correctly handling multi-turn customer conversations to successful resolution, while the CA is rewarded for producing realistic, intent-concealing utterances that cause misroutes, creating asymmetric adversarial pressure through opposing but independently structured rewards. An isolation gym iteratively retrains the weaker agent on prior-failure transcripts, requiring no human annotation at any stage. In a banking customer support proof of concept, tool-routing errors are eliminated and the strict end-to-end PASS rate doubles over five co-evolutionary cycles, driven solely by automated arena reward with no labelled data. We additionally observe the emergence of \textbf{Contextual Camouflage}, an adversarial strategy in which the CA learns to embed intent within dense realistic customer detail purely from reward pressure, with direct implications for enterprise red-teaming and robustness evaluation.

Chinese Translation

在企业客户支持中部署任务导向的对话代理面临着持续的标注瓶颈：鲁棒训练需要大规模的标记交互数据，但企业对话日志因隐私敏感而且标注成本高昂，同时用户行为的演变速度超过了标注流程的跟进能力。我们提出了RL-ADA（与对抗性对话代理的强化学习），这是一种共进化训练框架，通过用 extit{世界反馈}替代人工标签来消除这一瓶颈：基于后果的奖励信号直接来源于可测量的交互结果。客户支持代理（DA，3B参数）和对抗客户代理（CA，7B参数）在一个由固定自动评判者指导的对抗环境中共进化：DA因成功处理多轮客户对话而获得奖励，而CA则因产生真实且隐瞒意图的发言导致错误路由而获得奖励，通过相对但独立结构的奖励创造出不对称的对抗压力。一个隔离训练环境迭代地在先前失败的记录上重新训练较弱的代理，任何阶段均不需要人工标注。在一个银行客户支持的概念验证中，工具路由错误被消除，严格的端到端PASS率在五个共进化周期中翻倍，完全由自动化环境奖励驱动，无需标记数据。我们还观察到 extbf{上下文伪装}的出现，这是一种对抗策略，CA学习在密集的真实客户细节中嵌入意图，完全基于奖励压力，这对企业红队测试和鲁棒性评估具有直接影响。

View on arXiv Download PDF AI Translation

cs.CL / 13 / 2609.02940

Listen to the Latents: Self-Correcting Speech Recognition in Large Audio Language Models Through Hidden-State Interactions

倾听潜在因素：通过隐状态交互实现大型音频语言模型中的自我修正语音识别

Hsu, Chan-Jan, Kim, Jaeyeon, Yang, Chao-Han Huck, Watanabe, Shinji, Lee, Hung-yi, Busso, Carlos

Abstract

Recent automatic speech recognition (ASR) systems increasingly integrate large language models (LLMs) to leverage their semantic knowledge, either externally through logit fusion or internally through warm initialization. However, how to effectively combine these two strategies remains underexplored. In this work, we refine warm-initialized LLM-based ASR models by leveraging their own pre-adaptation base LLMs, focusing on LoRA-adapted settings where the base LLM is preserved. To achieve this, we propose Hybrid Search, a targeted correction strategy motivated by two observations. First, interaction features that characterize the relationship between LLM-based ASR hidden states and base-LLM hidden states provide informative signals about a token's degree of semantic dependence. Second, selectively refining targeted tokens with high semantic dependence improves ASR performance far beyond naive global LLM-correction methods including rescoring and late fusion. Our analysis suggests that, even after semantic knowledge transfer through warm initialization, LLM-based ASR models can still leverage their base LLM to further improve inference-time performance.

Chinese Translation

近年来，自动语音识别（ASR）系统越来越多地集成大型语言模型（LLMs），以利用其语义知识，既可以通过逻辑融合（logit fusion）进行外部集成，也可以通过温暖初始化（warm initialization）进行内部集成。然而，如何有效地结合这两种策略仍然未被充分探索。在本研究中，我们通过利用自身的预适应基础LLM，改进了基于温暖初始化的LLM-ASR模型，重点关注LoRA适应设置，其中基础LLM得以保留。为此，我们提出了混合搜索（Hybrid Search），这是一种有针对性的修正策略，基于两个观察结果。首先，表征LLM-ASR隐状态与基础LLM隐状态之间关系的交互特征提供了关于一个标记语义依赖程度的信息信号。其次，选择性地修正具有高语义依赖的目标标记，显著提高了ASR性能，远超包括重新评分（rescoring）和后期融合（late fusion）在内的简单全局LLM修正方法。我们的分析表明，即使在通过温暖初始化进行语义知识转移后，基于LLM的ASR模型仍然可以利用其基础LLM进一步提升推理时的性能。

View on arXiv Download PDF AI Translation

cs.CL / 14 / 2609.02942

Judging LLM-as-a-Judge: Concerning Rubric Artifacts in LLM-based Automated Text Generation Evaluation

评估 LLM 作为评审者：关于 LLM 基于的自动文本生成评估中的评分标准伪影

Bagaria, Anshul, Sundaram, Sowmya S, Krishnan, Gokul S, Ravindran, Balaraman

Abstract

LLM-as-a-Judge pipelines are increasingly used to evaluate AI-generated text, based on the assumption that judgments arise from reasoning over candidate responses with respect to a rubric. We show that this assumption warrants further scrutiny. Classifiers trained only on rubric text, without access to any evaluated response, achieve nontrivial predictive performance on judge outputs. This suggests that rubric formulations encode recoverable evaluative signals, allowing scores to be partially anticipated independently of model outputs. Finally, counterfactual perturbations reveal that judges often fail to reliably update their decisions when either the candidate response or the rubric criterion is reversed. Our findings raise concerns about the reliability of rubric-based LLM evaluation and highlight the need for further methodological study of automated evaluation via LLMs.

Chinese Translation

LLM 作为评审者的流程越来越多地用于评估 AI 生成的文本，基于这样一个假设：评判是通过对候选响应与评分标准的比较进行推理得出的。我们表明，这一假设值得进一步审视。仅基于评分标准文本训练的分类器，在没有任何评估响应的情况下，仍能在评审输出上取得非平凡的预测性能。这表明评分标准的表述编码了可恢复的评估信号，使得分数在一定程度上可以独立于模型输出进行预测。最后，反事实扰动揭示，当候选响应或评分标准的标准被反转时，评审者往往未能可靠地更新他们的决策。我们的研究结果引发了对基于评分标准的 LLM 评估可靠性的担忧，并强调了对通过 LLM 进行自动评估的进一步方法研究的必要性。

View on arXiv Download PDF AI Translation

cs.CL / 15 / 2609.02954

LexIssue: Benchmarking Legal Issue Identification in Chinese Civil Litigation

LexIssue：中文民事诉讼中法律问题识别的基准测试

Xie, Huiyuan, Huang, Yuqin, Hao, Zhicheng, Cai, Yida, Wang, Shaochun, Liu, Zhenghao, Ye, Yuxiao

Abstract

Identifying the issues disputed between litigating parties is a crucial component of real-world litigation. However, legal issues remain comparatively underexplored in legal AI research. In this work, we study the computational modelling of legal issue identification in litigation. We introduce a legally grounded hierarchical schema that represents legal issues through both free-form issue descriptions and structured legal categories, and formulate legal issue identification as two complementary tasks: legal issue generation and legal issue classification. Based on this formulation, we construct LexIssue, a benchmark containing 430 real-world Chinese civil litigation cases and 1,303 expert-annotated disputed legal issues. We further develop an issue-centric legal knowledge base spanning 27 causes of action and 441 candidate legal issue entries to support retrieval-augmented reasoning. Experimental results across a diverse set of models show that retrieval-augmented generation using the constructed legal issue knowledge base consistently improves performance in identifying disputed legal issues and their corresponding legal attributes.

Chinese Translation

识别诉讼当事人之间争议的问题是现实诉讼中的一个关键组成部分。然而，法律问题在法律人工智能研究中仍然相对未被深入探讨。在本研究中，我们探讨了诉讼中法律问题识别的计算建模。我们引入了一个法律基础的层次结构方案，通过自由形式的问题描述和结构化的法律类别来表示法律问题，并将法律问题识别形式化为两个互补的任务：法律问题生成和法律问题分类。基于这一形式化，我们构建了LexIssue，一个包含430个真实案例的中文民事诉讼基准和1,303个专家标注的争议法律问题。我们进一步开发了一个以问题为中心的法律知识库，涵盖27种诉因和441个候选法律问题条目，以支持检索增强推理。跨多种模型的实验结果表明，使用构建的法律问题知识库进行检索增强生成在识别争议法律问题及其相应法律属性方面始终提高了性能。

View on arXiv Download PDF AI Translation

cs.CL / 16 / 2609.03005

Unifying Conformal Language Tasks with In-Context Ensembles

统一的符合性语言任务与上下文集成

Huang, Xiao Shi, Lin, Chen-Yuan, Kuwahara, Bruce, Leung, Kin Kwan, Cresswell, Jesse C.

Abstract

Many NLP tasks, such as summarization and extractive question answering, reduce to retrieving relevant content from documents under two constraints: coverage, retaining enough pertinent information to achieve some goal, and conciseness, removing as much irrelevant information as possible. Conformal prediction methods have been used to guarantee coverage, and must be optimized for conciseness through design of a score function. State-of-the-art scoring functions use hand-engineered LLM prompts asking the model to rate the importance of content, but manual prompt engineering is labor-intensive and task-specific. We introduce the Conformal Relevance framework which uses in-context learning example curation and ensembling to create a score function which maintains coverage while improving conciseness with minimal manual input. We demonstrate this framework's application on seven NLP tasks, and also theoretically study the impact of diversity for ensembled conformal scores, giving a complementarity condition that characterizes when ensembling improves worst-case sentence scores, and a saturation bound on ensemble improvement.

Chinese Translation

许多自然语言处理（NLP）任务，如摘要生成和提取式问答，归结为在两个约束条件下从文档中检索相关内容：覆盖性，即保留足够的相关信息以实现某个目标；简洁性，即尽可能去除无关信息。符合性预测方法已被用于保证覆盖性，并必须通过设计评分函数来优化简洁性。当前最先进的评分函数使用手工设计的LLM（大语言模型）提示，要求模型评估内容的重要性，但手动提示工程既劳动密集又特定于任务。我们提出了符合性相关框架，该框架利用上下文学习示例的策划和集成，创建一个评分函数，在保持覆盖性的同时，以最小的人工输入提高简洁性。我们展示了该框架在七个NLP任务上的应用，并从理论上研究了集成符合性评分的多样性影响，给出了一个互补条件，表征何时集成改善最坏情况下的句子评分，以及集成改进的饱和界限。

View on arXiv Download PDF AI Translation

cs.CL / 17 / 2609.03047

SHELF: A Synthetic Harness for Multi-Task Bibliographic Benchmarking

SHELF：用于多任务书目基准测试的合成工具

Bommarito II, Michael J.

Abstract

Libraries and archives manage large collections with limited staff and computing budgets, yet common benchmarks do not systematically test their bibliographic work. They need to know which methods work for their tasks and what those methods require to run. SHELF, the Synthetic Harness for Evaluating LLM Fitness, addresses this gap. It is a Python system that turns labelled taxonomies, writing specifications, and a generation budget into controlled benchmark data and evaluation tasks. This first release contains 62,899 model-written documents based on Library of Congress vocabularies, with tasks for classification, clustering, retrieval, pair classification, and instruction retrieval. We compare TF, TF-IDF, BM25, popular encoders, and, on subject classification only, zero-shot decoders; each method appears only on tasks that support it. Subject classification reaches 0.8887, while genre-form classification reaches only 0.2605, and several pair and clustering tasks remain near chance. Sparse methods remain competitive on classification, while TF-IDF is the fastest measured arm in the subject timing experiment. SHELF also varies bibliographic facets independently and can generate new, verifiably unseen documents after a model's training cutoff. Comparisons with LCSHBench and Project Gutenberg show that model rankings transfer more reliably than absolute scores, but SHELF scores do not estimate accuracy on production catalogue data. We release all source code and data under permissive licenses on GitHub and Hugging Face.

Chinese Translation

图书馆和档案馆在有限的人员和计算预算下管理大量藏书，但现有的基准测试并未系统性地评估它们的书目工作。它们需要了解哪些方法适用于其任务，以及这些方法运行所需的条件。SHELF（合成评估大型语言模型适用性的工具）填补了这一空白。它是一个Python系统，将标记的分类法、写作规范和生成预算转化为受控的基准数据和评估任务。此次首次发布包含62,899份基于国会图书馆词汇的模型生成文档，涵盖分类、聚类、检索、对分类和指令检索等任务。我们比较了TF、TF-IDF、BM25、流行编码器，以及仅在主题分类中使用的零样本解码器；每种方法仅出现在支持其的任务中。主题分类达到了0.8887，而体裁分类仅达到了0.2605，多个对分类和聚类任务的结果接近随机水平。稀疏方法在分类中仍然具有竞争力，而TF-IDF在主题计时实验中是测得的最快方法。SHELF还独立变化书目特征，并且可以在模型训练截止后生成新的、可验证的未见文档。与LCSHBench和古腾堡计划的比较显示，模型排名的转移比绝对分数更可靠，但SHELF的分数并不能估计生产目录数据的准确性。我们在GitHub和Hugging Face上以宽松的许可证发布所有源代码和数据。

View on arXiv Download PDF AI Translation

cs.CL / 18 / 2609.03148

Large Language Models in Resolving Contextual Knowledge Conflicts

大型语言模型在解决上下文知识冲突中的应用

Yang, Xinye, Liu, Zhenyang, Li, Ruisi, Lei, Yuanyuan

Abstract

Most prior works focused on conflicts between an LLM's internal parametric knowledge and externally provided context. In contrast, we investigate how LLMs handle conflicts that arise within contextual knowledge itself. We introduce a taxonomy of six types of contextual conflicts (factual, inferential, temporal, granularity, perspective, and ambiguity) and contribute a comprehensive dataset ContextConflict for this setting. The dataset contains 5,781 samples, covers both reasoning and summarization tasks, and includes both explicit contradictions and implicit conflicts that require multi-step reasoning. Experiments on nine LLMs show that current models still fall short in resolving contextual knowledge conflicts. We further provide mechanistic interpretability insights into how LLMs process such conflicts, revealing their latent awareness of conflicts and the representational geometry underlying conflict processing. In addition, our analysis uncovers a consistent model bias towards earlier evidence, and this positional preference serves as a key obstacle to effective conflict resolution. Motivated by these findings, we further propose a simple training-free, label-free steering method that steers activations to encourage a more comprehensive incorporation of evidences for better conflict resolution. On our dataset, the method consistently improves accuracy on reasoning tasks and generates higher-quality, more balanced summaries for summarization tasks.

Chinese Translation

大多数先前的研究集中于大型语言模型（LLM）内部参数知识与外部提供的上下文之间的冲突。相比之下，我们探讨了LLM如何处理上下文知识内部产生的冲突。我们引入了六种类型的上下文冲突的分类法（事实性、推理性、时间性、粒度、视角和模糊性），并为此设置贡献了一个全面的数据集ContextConflict。该数据集包含5,781个样本，涵盖了推理和摘要任务，并包括显式矛盾和需要多步推理的隐式冲突。对九个LLM的实验表明，当前模型在解决上下文知识冲突方面仍然存在不足。我们进一步提供了关于LLM如何处理这些冲突的机制可解释性见解，揭示了它们对冲突的潜在意识以及冲突处理背后的表征几何。此外，我们的分析揭示了模型对早期证据的一致偏见，这种位置偏好成为有效解决冲突的关键障碍。基于这些发现，我们进一步提出了一种简单的无训练、无标签的引导方法，该方法引导激活以鼓励更全面地整合证据，从而更好地解决冲突。在我们的数据集上，该方法在推理任务上持续提高了准确性，并为摘要任务生成了更高质量、更平衡的摘要。

View on arXiv Download PDF AI Translation

cs.CL / 19 / 2609.03160

No country for old linguists: LLM-brain alignment underdetermines neural computation

老语言学家无处可去：大型语言模型与大脑的对齐不足以决定神经计算

Murphy, Elliot

Abstract

Nastase et al. (2026) argue that large language models (LLMs) may illuminate language processing because both rely on distributed, context-sensitive representations shaped by statistical learning. Their rejection of simple cortical "boxology" is persuasive, and they articulate a strong case for the value of LLM-brain alignment research. The key question is what kind of inference LLM-brain alignment licenses. My claim here will be narrow: representational alignment can in principle constrain mechanistic hypotheses, but it does not by itself identify a mechanism. Nastase et al. acknowledge that an encoding model can capture features represented in neural activity without establishing a shared architecture or algorithm. Yet the authors sometime move from alignment to "shared computational principles" and ultimately to LLMs as mechanistic models of natural language. Indeed, their methodological caveat that alignment does not establish a shared architecture or algorithm sits uneasily with their conclusion that LLMs might instantiate the same computational principles as biological brains and provide a "fully mechanistic model" of language. I discuss what I consider to be problems of logical, causal, and computational underdetermination in Nastase et al.'s (2026) proposal.

Chinese Translation

Nastase 等人（2026）认为，大型语言模型（LLMs）可能有助于揭示语言处理，因为两者都依赖于由统计学习塑造的分布式、上下文敏感的表征。他们对简单皮层“盒子学”的拒绝是有说服力的，并且他们清晰地阐述了 LLM-大脑对齐研究的价值。关键问题是 LLM-大脑对齐允许什么样的推断。我的观点将是狭义的：表征对齐原则上可以限制机制假设，但它本身并不能识别机制。Nastase 等人承认，编码模型可以捕捉神经活动中表示的特征，而无需建立共享的架构或算法。然而，作者有时会从对齐转向“共享计算原则”，最终将 LLM 视为自然语言的机制模型。确实，他们的一个方法论警告，即对齐并不建立共享的架构或算法，与他们得出的结论，即 LLM 可能体现与生物大脑相同的计算原则并提供语言的“完全机械模型”，显得不太协调。我讨论了我认为在 Nastase 等人（2026）提案中存在的逻辑、因果和计算不足的问题。

View on arXiv Download PDF AI Translation

cs.CL / 20 / 2609.03181

Jina-OCR-v1: Efficient Document Parsing with Speculative Decoding and Dense Verifiable Rewards

Jina-OCR-v1：基于推测解码和密集可验证奖励的高效文档解析

García, Alejandro Barón, Wang, Feng, Casademont, Emilia Garcia, Xiao, Han

Abstract

We present Jina-OCR-v1, an end-to-end document parsing model built to serve on low-budget GPUs. It combines the compressed-vision encoder and the 3B mixture-of-experts decoder of DeepSeek-OCR, which activates about 570M parameters per token, with a FastMTP speculative decoding head that shares a single draft block recursively across K=3 prediction steps. Greedy verification makes decoding lossless. Post-training combines instruction alignment, robustness fine-tuning on difficult documents, and GRPO under dense verifiable rewards: deterministic formula, table, and structural checks that award partial credit. The training data mixes cleaned public corpora with targeted synthetic pages. At the default dynamic-resolution setting, Jina-OCR-v1 scores 91.14 on OmniDocBench v1.6 and 83.4 on olmOCR-Bench, and reaches the highest page throughput in our comparison at 2.57 pages per second. On a low-budget GPU such as the NVIDIA L4, FastMTP doubles decoding speed over greedy autoregressive decoding. The model is publicly available at https://huggingface.co/jinaai/jina-ocr-v1.

Chinese Translation

我们提出了Jina-OCR-v1，这是一个为低预算GPU构建的端到端文档解析模型。它结合了DeepSeek-OCR的压缩视觉编码器和3B专家混合解码器，每个token激活约570M参数，并配备了FastMTP推测解码头，该头在K=3个预测步骤中递归共享单个草稿块。贪婪验证使解码无损。后训练结合了指令对齐、在困难文档上的鲁棒性微调，以及在密集可验证奖励下的GRPO：确定性公式、表格和结构检查，这些检查给予部分信用。训练数据混合了清理过的公共语料库和针对性的合成页面。在默认动态分辨率设置下，Jina-OCR-v1在OmniDocBench v1.6上得分91.14，在olmOCR-Bench上得分83.4，并在我们的比较中以每秒2.57页达到了最高的页面吞吐量。在像NVIDIA L4这样的低预算GPU上，FastMTP的解码速度是贪婪自回归解码的两倍。该模型已在https://huggingface.co/jinaai/jina-ocr-v1上公开提供。

View on arXiv Download PDF AI Translation

cs.CL / 21 / 2609.03201

MemoryLACE: Memory Lifecycle-Aware Consolidation and Evidence Retrieval

MemoryLACE：记忆生命周期感知的整合与证据检索

Yacoubi, Meriem, Schmidt, Pia, Petrovic, Nenad, Frikha, Ahmed, Kirchhoff, Martin, Knoll, Alois

Abstract

Long-term LLM agents must preserve information across interactions while distinguishing repeated evidence, historical states, updates, and unresolved contradictions. Existing textual memory systems retrieve semantically relevant memories efficiently but often leave these relationships implicit, whereas richer structured approaches model them through global graphs, hierarchical abstractions, or reflection at greater complexity. We introduce MemoryLACE (MemLACE), a lightweight memory framework that explicitly models the lifecycle of textual evidence through sparse merge, supersession, and contradiction relations while preserving atomic natural-language memories and their provenance. Rather than retrieving memories independently, MemLACE reconstructs relation-aware evidence units that expose current, historical, supporting, and conflicting evidence for downstream reasoning. Across BEAM and StructMemEval, using open-weight and proprietary LLM backbones, MemLACE achieves the highest overall performance in same-backbone comparisons while reducing end-to-end runtime on BEAM by 66.6% relative to Hindsight, the strongest reported reflective-memory baseline. Ablation studies identify lifecycle expansion and temporal awareness as the principal contributors to these gains. Together, the results demonstrate that explicitly modeling the local lifecycle of textual evidence is sufficient to substantially improve long-term memory reasoning without requiring comprehensive knowledge graphs or global reflection.

Chinese Translation

长期的LLM（大规模语言模型）代理必须在交互中保留信息，同时区分重复的证据、历史状态、更新和未解决的矛盾。现有的文本记忆系统能够高效地检索语义相关的记忆，但往往将这些关系隐含，而更丰富的结构化方法则通过全局图、层次抽象或更复杂的反思来建模。我们提出了MemoryLACE（MemLACE），一个轻量级的记忆框架，明确建模文本证据的生命周期，通过稀疏合并、替代和矛盾关系，同时保留原子自然语言记忆及其来源。MemLACE并不是独立地检索记忆，而是重建关系感知的证据单元，揭示当前、历史、支持和冲突的证据，以便于下游推理。在BEAM和StructMemEval上，使用开放权重和专有LLM骨干网络，MemLACE在同一骨干网络比较中实现了最高的整体性能，同时相较于Hindsight（最强的反思记忆基线），在BEAM上将端到端运行时间减少了66.6%。消融研究表明，生命周期扩展和时间感知是这些提升的主要贡献因素。综合来看，结果表明，明确建模文本证据的局部生命周期足以显著改善长期记忆推理，而无需全面的知识图谱或全局反思。

View on arXiv Download PDF AI Translation

cs.CL / 22 / 2609.03213

LLMs Learn Better In-Context from Rules than from Examples

大型语言模型在上下文中从规则学习的效果优于从示例学习

Fu, Xiang, Cho, Seungmin, Lee, Yukyung, Kim, Najoung

Abstract

Large language models (LLMs) exhibit in-context learning capabilities, where they can learn new tasks from prompt contexts without weight updates. We compare the learning efficacies of two prominent modes of in-context learning: (1) learning from descriptions of rules (instruction following); and (2) learning from examples of input-output demonstrations (few-shot prompting). Through five learning tasks that cover diverse domains (games, arithmetic, linguistic inferences), we compare two modes of learning (rules vs. examples) specifying the same underlying task. We furthermore explore model and task properties that modulate the learning efficacies. We find that models generally learn more reliably from rules than from examples alone, and additional examples on top of rules or simply scaling up the number of examples do not lead to consistent and significant gains. Instruction tuning amplifies the benefit of rule-based learning while keeping example-based learning capacities intact. Surprisingly, we find no privileged effect of example-based learning in base models, and rules still lead to gains in algebraic task domains. Overall, the comparative efficacy of rules over examples is larger when the task recruits algebraic abstractions and computations, and smaller when the task requires distributional sensitivity and/or recruits parametric knowledge.

Chinese Translation

大型语言模型（LLMs）展现了上下文学习的能力，即它们可以在不更新权重的情况下从提示上下文中学习新任务。我们比较了两种主要的上下文学习模式的学习效果：（1）从规则描述中学习（遵循指令）；（2）从输入-输出示例中学习（少量示例提示）。通过五个涵盖不同领域（游戏、算术、语言推理）的学习任务，我们比较了这两种学习模式（规则与示例），并指定了相同的基础任务。我们进一步探讨了影响学习效果的模型和任务属性。我们发现，模型通常从规则中学习的可靠性高于仅从示例中学习，而在规则基础上增加示例或简单地增加示例数量并未带来一致且显著的提升。指令调优增强了基于规则学习的优势，同时保持了基于示例学习的能力。令人惊讶的是，我们在基础模型中未发现基于示例学习的特权效应，规则仍然在代数任务领域带来收益。总体而言，当任务涉及代数抽象和计算时，规则相较于示例的比较效果更显著，而当任务需要分布敏感性和/或涉及参数知识时，则效果较小。

View on arXiv Download PDF AI Translation

cs.CL / 23 / 2609.03215

SWIM: Student Writing Simulation via Proficiency-Conditioned Generation

SWIM：基于能力条件生成的学生写作模拟

Do, Heejin, Kontak, Jakub, Sachan, Mrinmaya

Abstract

Writing proficiency manifests in how students develop content, organize ideas, choose words, and use language. Despite growing interest in LLM-based student simulation, whether LLMs can reproduce such multidimensional variation in extended writing remains largely unexplored. In this work, we explore if language models can realistically simulate student writing, and introduce SWIM, a task that formulates Student Writing sIMulation as proficiency-conditioned essay generation. We evaluate prompting, supervised fine-tuning (SFT), and reinforcement learning (RL) methods for writing simulation using automated essay scoring as a measure of profile alignment. Extensive experiments reveal that prompting provides limited proficiency control, even for strong proprietary LLMs with rubric-grounded strategies. In particular, while models can adjust content-oriented traits, they struggle to reproduce the lexical, grammatical, and organizational variation in different proficiency levels. SFT substantially improves alignment, while RL with the proposed proficiency-alignment reward yields further gains across all writing traits and essay prompts. Our findings suggest that explicit supervision enables substantially stronger profile alignment than prompting alone, while authentic low-proficiency writing remains challenging to reproduce.

Chinese Translation

写作能力体现在学生如何发展内容、组织思想、选择词汇和使用语言上。尽管对基于大型语言模型（LLM）的学生模拟的兴趣日益增长，但LLM是否能够在扩展写作中再现这种多维变化仍然大部分未被探索。在本研究中，我们探讨了语言模型是否能够现实地模拟学生写作，并引入了SWIM，一个将学生写作模拟（Student Writing sIMulation）形式化为基于能力条件的论文生成的任务。我们评估了使用自动化论文评分作为能力对齐度量的提示、监督微调（SFT）和强化学习（RL）方法。大量实验表明，提示在能力控制方面提供的效果有限，即使对于具有基于评分标准策略的强大专有LLM也是如此。特别是，虽然模型能够调整内容导向的特征，但在再现不同能力水平的词汇、语法和组织变化方面却面临挑战。SFT显著改善了对齐度，而使用所提出的能力对齐奖励的RL在所有写作特征和论文提示中都取得了进一步的提升。我们的研究结果表明，明确的监督能够实现比单纯提示更强的能力对齐，而真实的低能力写作仍然难以再现。

View on arXiv Download PDF AI Translation

cs.CL / 24 / 2609.03218

The Analyst in the Prompt: Role, Retrieval, and Memory Biases in LLM Financial Analysis

提示中的分析师：大型语言模型金融分析中的角色、检索与记忆偏差

Asaad, Ahmed, Mohamed, Amr, Zhang, Yang, Abdelsalam, Omneya

Abstract

Large Language Models (LLMs) increasingly use user context such as memory, profiles, and role prompts to personalize their responses. This personalization can affect evidence-based judgment: the same evidence may lead to different conclusions under different user contexts. Finance provides a high-stakes setting to study this problem because decisions often depend on interpreting long and complex documents. We test this using 3,575 SEC filings across twelve LLMs. We compare persona-conditioned retrieval, neutral retrieval, and memory-framed context to separate the effect of evidence selection from the effect of interpretation. We find that most user-context spillover comes from how models interpret the same evidence under different roles, rather than from retrieving different evidence. We then test two simple mitigation strategies: expressing the same investor mindset as a user profile instead of an assistant role, and separating evidence-based and personalized outputs. Both reduce spillover, but neither removes it completely, and their effectiveness varies substantially across models.

Chinese Translation

大型语言模型（LLMs）越来越多地使用用户上下文，如记忆、个人资料和角色提示，以个性化其响应。这种个性化可能影响基于证据的判断：在不同的用户上下文下，相同的证据可能导致不同的结论。金融领域提供了一个高风险的环境来研究这个问题，因为决策往往依赖于对长篇复杂文件的解读。我们使用3575份美国证券交易委员会（SEC）文件在十二个LLM中进行测试。我们比较了基于角色的检索、中立检索和记忆框架上下文，以区分证据选择的影响与解读的影响。我们发现，大多数用户上下文的溢出效应来自模型在不同角色下对相同证据的解读，而不是来自检索不同的证据。随后，我们测试了两种简单的缓解策略：将相同的投资者心态表达为用户个人资料，而不是助手角色，以及将基于证据的输出与个性化输出分开。这两种策略都减少了溢出效应，但都未能完全消除，且其有效性在不同模型间差异显著。

View on arXiv Download PDF AI Translation

cs.CL / 25 / 2609.03221

Counterfactual Fairness Audits of Multi-Step Clinical LLM Agents Require a Measured Per-Action Instability Floor

多步骤临床大语言模型代理的反事实公平性审计需要测量每个行动的不稳定性底线

Bellibaltu, Rohith Reddy, Singh, Manpreet, Parashar, Deepak, Joshi, Rahul

Abstract

Counterfactual audits are the standard tool for checking whether a clinical agent treats demographically distinct but clinically identical patients differently. They report a flip rate: how often an action changes when only the patient descriptor changes. We show that this quantity is uninterpretable on its own. Re-running an identical condition ten times over sixteen vignettes (same narrative, same descriptor string, nothing varied) moved a clinical agent's action in 8.7% of outcome-vignette cells, and instability was heterogeneous across actions by a factor of eight, from 0.022 for ICU escalation to 0.179 for controlled-substance caution. No demographic contrast in our data was distinguishable from that floor. A second model gives a pooled floor of 6.7% and ranks the six actions almost identically (Spearman 0.94, exact p=0.017), so the floor is not one system's artefact. Majority-vote aggregation over five draws removes 39% of it and then flattens, and a null simulation attributes the residue to heterogeneous per-cell rates, so replication mitigates without eliminating. Any counterfactual fairness estimate reported without a per-action floor beside it therefore cannot be read as evidence of disparity. The measurements were taken with FairMedAgent, an evaluation harness for disparity in the actions of clinical LLM agents whose estimand, the within-range counterfactual flip rate, counts only flips between actions a published decision rule admits and a clinician has adjudicated. That estimand requires band adjudication, which is under way; no disparity result is claimed here. Each synthetic vignette runs a six-stage trajectory (five model-facing decisions around a deterministic environment step) under fixed-form conditions spanning race, sex, age, insurance, English proficiency, and their intersections. The harness, the floor protocol, and every analysis script are released.

Chinese Translation

反事实审计是检查临床代理是否对人口统计上不同但临床上相同的患者采取不同对待的标准工具。它们报告了翻转率：当仅患者描述符发生变化时，行动改变的频率。我们表明，这一数量本身是不可解释的。在十次相同条件下的重复实验中，针对十六个情境（相同叙述、相同描述符字符串、无变化），临床代理的行动在8.7%的结果-情境单元中发生了变化，并且不同行动之间的不稳定性相差八倍，从ICU升级的0.022到管制物质警告的0.179。我们数据中的任何人口统计对比都无法与这一底线区分开来。第二个模型给出了6.7%的汇总底线，并几乎相同地对六个行动进行了排名（Spearman 0.94，精确p=0.017），因此该底线不是某一系统的伪影。对五次抽样的多数投票聚合去除了39%的底线，然后趋于平坦，而零模拟将残余归因于每个单元的不均匀率，因此复制减轻了而非消除。因此，任何未附带每个行动底线的反事实公平性估计都不能被视为不平等的证据。测量是通过FairMedAgent进行的，这是一种评估临床大语言模型代理行动不平等的工具，其估计量，即范围内的反事实翻转率，仅计算被已发布决策规则允许且由临床医生裁定的行动之间的翻转。该估计量需要带裁定，目前正在进行中；此处不声称有不平等结果。每个合成情境运行一个六阶段轨迹（五个面向模型的决策围绕一个确定性环境步骤）在固定形式条件下，涵盖种族、性别、年龄、保险、英语水平及其交集。该工具、底线协议和每个分析脚本均已发布。

View on arXiv Download PDF AI Translation

cs.CL / 26 / 2609.03235

SGD-KV: Summarization Guided KV Cache Compression

SGD-KV：摘要引导的KV缓存压缩

Liu, Zeyu, Song, Woomin, Fu, Xuandi, Jayanthi, Sai Muralidhar, Govindan, Vivek, Galstyan, Aram, Bodapati, Sravan Babu, Ronanki, Srikanth

Abstract

Large language models (LLMs) face severe memory bottlenecks in long-context inference due to the linearly growing size of key-value (KV) caches. Existing KV cache compression techniques typically rely on simple heuristics, overlooking the distinct functional roles of different attention heads. We present SGD-KV (Summarization-Guided KV Cache Compression), a head-aware framework that leverages a novel chunk-summarization diagnostic task to systematically identify and prioritize attention heads specialized in hierarchical information aggregation. Experiments on Qwen2.5-7B-1M and Qwen3-32B across diverse long-context benchmarks demonstrate that SGD-KV achieves state-of-the-art performance with contexts up to 1M tokens, while reducing KV cache memory usage by up to 75%. Our findings show that strategically allocating the KV cache budget based on the summarization score distribution of attention heads yields a superior efficiency-accuracy trade-off for long-context inference.

Chinese Translation

大型语言模型（LLMs）在长上下文推理中面临严重的内存瓶颈，这是由于键值（KV）缓存的大小线性增长所致。现有的KV缓存压缩技术通常依赖于简单的启发式方法，忽视了不同注意力头的功能角色差异。我们提出了SGD-KV（摘要引导的KV缓存压缩），这是一个关注注意力头的框架，利用一种新颖的块摘要诊断任务系统地识别和优先考虑专门用于层次信息聚合的注意力头。在Qwen2.5-7B-1M和Qwen3-32B的多种长上下文基准测试中的实验表明，SGD-KV在处理高达1M标记的上下文时实现了最先进的性能，同时将KV缓存的内存使用减少了多达75%。我们的研究结果表明，基于注意力头的摘要得分分布战略性地分配KV缓存预算，可以在长上下文推理中实现更优的效率-准确性权衡。

View on arXiv Download PDF AI Translation

cs.CL / 27 / 2609.03254

What Else Needs Fixing? Exploring Cost-Effective Test-Time Compute for Revision Propagation in Artifacts Generated Through Conversation

还有什么需要修正的？探索在对话生成的工件中进行修订传播的成本有效测试时间计算

Kikuta, Daisuke

Abstract

Large Language Models (LLMs) often help users generate artifacts through iterative cycles of generation and revision in conversation. A challenge here is that, when users specify only a local change during revision, LLMs must instead identify the relevant dependencies and propagate the revision to all affected parts of the artifact. This paper studies this ability of LLMs on conversationally generated artifacts, where the artifact context and its dependencies may be embedded in the conversation history. Toward practical use, we also explore cost-effective test-time compute for this new setting. Specifically, we introduce a new benchmark for this setting, and evaluate nine revision methods, including sequential reflection and parallel sampling variants, using gpt-oss-20b/120b, gpt-5.4-mini, and qwen3.5-9b/27b/122b on the benchmark. The results show that baselines achieve accuracies of 68.3--93%, and the most cost-effective method is selecting from three parallel samples using either LLM-based or medoid selection, which improves accuracy by 2.2--9.7%. Our code and dataset are available at https://github.com/ntt-dkiku/llm-revision-propagation.

Chinese Translation

大型语言模型（LLMs）通常通过生成和修订的迭代循环帮助用户生成工件。在此过程中，一个挑战是，当用户在修订时仅指定局部更改时，LLMs必须识别相关的依赖关系并将修订传播到工件的所有受影响部分。本文研究了LLMs在对话生成的工件上的这一能力，其中工件的上下文及其依赖关系可能嵌入在对话历史中。为了实现实际应用，我们还探索了这一新环境下的成本有效测试时间计算。具体而言，我们为这一环境引入了一个新的基准，并评估了九种修订方法，包括顺序反思和并行采样变体，使用gpt-oss-20b/120b、gpt-5.4-mini和qwen3.5-9b/27b/122b进行基准测试。结果表明，基线的准确率达到68.3%至93%，而最具成本效益的方法是从三个并行样本中选择，使用LLM基础或质心选择，这提高了准确率2.2%至9.7%。我们的代码和数据集可在https://github.com/ntt-dkiku/llm-revision-propagation获取。

View on arXiv Download PDF AI Translation

cs.CL / 28 / 2609.03273

Contextual Tamil Spelling and Grammar Correction Using Progressively Fine-Tuned Sequence-to-Sequence Transformers

基于逐步微调的序列到序列变换器的上下文泰米尔拼写和语法纠正

A, Karthikeyan, S, Jaya Nirmala, Sivanesan, Sangeetha, R, Indhu, Kumar, Pranav, Johnson, Bharat Jude, Ram, Vishnu

Abstract

Tamil spell and grammar correction is challenging because Tamil is an agglutinative low-resource language with rich verbal morphology, complex sandhi (phonetic transformation) rules at word boundaries, and a script of 247 distinct letters. Prior work targets word-level surface errors with rule-based methods, statistical n-gram models, Minimum Edit Distance, or hybrid pipelines with a transformer re-ranker; such methods cannot reliably handle contextual errors - subject-verb agreement, tense consistency, or cross-word sandhi - which require sentence-level understanding. We propose an end-to-end sequence-to-sequence formulation and fine-tune mT5-small and mBART-50 on a synthetic corpus of up to 657,720 noisy-clean Tamil sentence pairs spanning ten error categories. Both backbones follow the same four-stage progressive schedule, each stage targeting one weakness: surface noise (v2), contextual grammar (v3), single-site sandhi (v4), and multi-site cross-word sandhi (v5). On a 1,000-sentence balanced diagnostic set verified disjoint from all training data, our best model, mBART-50 v5, reaches 69.3% top-1 exact-match accuracy, with 87.5% on sandhi and 43.5% on subject-verb agreement. The schedule is what produces these gains: subject-verb accuracy rises from 1.0% to 52.5% once contextual pairs are introduced, and sandhi from 0% to 87.5% once multi-site sandhi pairs are. We additionally quantify a precision-recall trade-off this literature has not reported: sandhi recall is paid for monotonically in identity accuracy. Finally, Tamil-LLaMA-7B-Instruct reaches 19.0% zero-shot and 24.7% with three demonstrations against a 20.0% copy baseline, showing that a Tamil-adapted instruction model does not transfer to specialised sentence-level correction without task-specific supervision.

Chinese Translation

泰米尔拼写和语法纠正具有挑战性，因为泰米尔语是一种富有形态变化的粘着语，属于低资源语言，具有复杂的音变（sandhi）规则和247个不同字母的书写系统。以往的研究主要针对词级表面错误，采用基于规则的方法、统计n-gram模型、最小编辑距离或与变换器重排序器结合的混合管道；这些方法无法可靠地处理上下文错误——主谓一致、时态一致性或跨词音变——这些都需要句子级理解。我们提出了一种端到端的序列到序列模型，并在一个包含高达657,720对噪声-干净泰米尔句子的合成语料库上微调了mT5-small和mBART-50，涵盖十种错误类型。两个基础模型遵循相同的四阶段渐进调度，每个阶段针对一种弱点：表面噪声（v2）、上下文语法（v3）、单点音变（v4）和多点跨词音变（v5）。在一个与所有训练数据完全独立的1,000句平衡诊断集上，我们的最佳模型mBART-50 v5达到了69.3%的顶级精确匹配准确率，其中音变的准确率为87.5%，主谓一致的准确率为43.5%。这一调度产生了这些提升：一旦引入上下文对，主谓准确率从1.0%上升至52.5%，而音变准确率从0%上升至87.5%。我们还量化了这一文献未曾报告的精确率-召回率权衡：音变召回率在身份准确率中单调付出。最后，泰米尔-LLaMA-7B-Instruct在零样本情况下达到了19.0%，在三个示例情况下达到了24.7%，而20.0%的复制基线显示，泰米尔适应的指令模型在没有任务特定监督的情况下无法转移到专业的句子级纠正。

View on arXiv Download PDF AI Translation

cs.CL / 29 / 2609.03293

PACE: Towards Surfacing Hidden Conflicts in User Requests

PACE：揭示用户请求中隐藏冲突的探索

Kim, Yoojin, Jang, Jihyoung, Kim, Hyounghun

Abstract

Personalized assistants should not only comply with user requests but also assess whether those requests are appropriate given the user's current circumstances. However, prior work has primarily focused on accurately executing requests, overlooking the need for assistants to account for context and engage in conflict-based refusal. Furthermore, while existing work on conflict or safety detection relies on explicitly provided factors, real-world scenarios often involve implicit factors that must be retrieved from a knowledge base (KB). To this end, we introduce Personalized Assistants for Conflict Evaluation (PACE), a dataset for evaluating whether models can identify latent constraints, expressed as egocentric knowledge or events, that render seemingly reasonable user requests inappropriate. PACE pairs user requests grounded in well-defined personas with egocentric KB facts, requiring models to integrate contextual evidence to determine whether a request is conflicting. This implicit retrieval setting hinders the direct association between user requests and conflict-inducing knowledge, making it difficult for existing models to identify relevant user-specific facts. To address this challenge, we further propose PaceMaker, a multi-agent framework in which specialized agents coordinate across query reformulation, multi-hop graph traversal, and conflict-aware filtering to retrieve contextually decisive evidence. Experiments on PACE evaluate both evidence retrieval quality and conflict decision accuracy, showing that PaceMaker consistently outperforms existing approaches.

Chinese Translation

个性化助手不仅应遵循用户请求，还应评估这些请求在用户当前情况下是否合适。然而，以往的研究主要集中在准确执行请求上，忽视了助手需要考虑上下文并进行基于冲突的拒绝。此外，现有关于冲突或安全检测的研究依赖于明确提供的因素，而现实场景往往涉及必须从知识库（KB）中检索的隐含因素。为此，我们引入了个性化助手冲突评估（PACE），这是一个用于评估模型是否能够识别潜在约束的数据集，这些约束以自我中心知识或事件的形式表达，使得看似合理的用户请求变得不合适。PACE将基于明确定义的人物角色的用户请求与自我中心的知识库事实配对，要求模型整合上下文证据以确定请求是否存在冲突。这种隐式检索设置阻碍了用户请求与引发冲突的知识之间的直接关联，使得现有模型难以识别相关的用户特定事实。为了解决这一挑战，我们进一步提出了PaceMaker，一个多智能体框架，其中专门的智能体在查询重构、多跳图遍历和冲突感知过滤之间协调，以检索上下文决定性证据。在PACE上的实验评估了证据检索质量和冲突决策准确性，结果表明PaceMaker在各项指标上均优于现有方法。

View on arXiv Download PDF AI Translation

cs.CL / 30 / 2609.03321

Decoupling Turn-Taking from Semantics: A Decoupled Data Approach for Finite-State-Machine-Based Full-Duplex Dialogue

将轮流发言与语义解耦：基于有限状态机的全双工对话的解耦数据方法

Li, Yihang, Chu, Chenhui

Abstract

The Neural Finite State Machine (NFSM) framework offers a pragmatic path to full-duplex dialogue by serializing turn-taking control and response generation onto a single causal tape under the standard next-token prediction objective, thereby preserving semantic prowess at a low fine-tuning cost. However, its reliance on synthetic text data fundamentally limits turn-taking naturalness, as Large Language Models (LLMs) cannot faithfully simulate the fine-grained acoustic temporal dynamics of real human dialogues. In this work, we propose a decoupled data approach that learns turn-taking from real Human-Human (HH) spoken dialogues while shaping semantic behavior through configurable Human-Agent (HA) text dialogues. To operationalize this approach, we introduce a rule-based event-guided data transformation method that serializes HH spoken dialogues into FSM tapes by classifying turn-taking events and applying deterministic mapping rules, enabling scalable supervision without LLM-generated annotations. We further propose a Source-Aware Calibrated (SAC) Loss that jointly calibrates the long-tailed distribution of state transition tokens and channels each data source toward the capability it best supervises. Experiments show that our approach substantially improves turn-taking proficiency while recovering the foundation LLM's semantic capability. Our code and model are available at https://github.com/Liyht/def-fsm.

Chinese Translation

神经有限状态机（NFSM）框架通过将轮流发言控制和响应生成序列化到单一因果带上，在标准的下一个标记预测目标下，为全双工对话提供了一条务实的路径，从而在低微调成本下保持了语义能力。然而，其对合成文本数据的依赖在根本上限制了轮流发言的自然性，因为大型语言模型（LLMs）无法真实模拟人类对话中细致的声学时间动态。在本研究中，我们提出了一种解耦数据方法，该方法从真实的人际（Human-Human, HH）口语对话中学习轮流发言，同时通过可配置的人机（Human-Agent, HA）文本对话塑造语义行为。为了实现这一方法，我们引入了一种基于规则的事件引导数据转换方法，该方法通过分类轮流发言事件并应用确定性映射规则，将HH口语对话序列化为有限状态机（FSM）带，从而实现可扩展的监督，而无需LLM生成的注释。我们进一步提出了一种源感知校准损失（Source-Aware Calibrated, SAC Loss），该损失共同校准状态转移标记的长尾分布，并将每个数据源引导至其最佳监督能力。实验表明，我们的方法显著提高了轮流发言的熟练度，同时恢复了基础LLM的语义能力。我们的代码和模型可在 https://github.com/Liyht/def-fsm 获取。

View on arXiv Download PDF AI Translation

cs.CL / 31 / 2609.03322

How Perturbations Propagate: A Multi-Level Analysis of Robustness in Large Language Models

扰动如何传播：大语言模型鲁棒性的多层次分析

Chan, Dun Li, Liu, Emily, Allu, Niyathi, Hoang, Christian

Abstract

Language models encounter typos, corrupted text, altered words, and disrupted token order, yet robustness is usually evaluated only through output behavior. We study how six naturalistic and synthetic input perturbations propagate through decoder-only language models at three levels: output behavior, hidden-state geometry, and attention-head function. We evaluate behavioral effects across four GPT-2 and two Qwen2.5 checkpoints by analyzing layerwise geometry using centered kernel alignment and intrinsic dimension, and examine attention-head responses in GPT-2. Perturbation types produce distinguishable metric profiles that are not fully captured by output measures and are only partly consistent across the tested checkpoints. Copying scores are especially associated with activation-patching recovery under token substitution and shuffling. Gradient-guided HotFlip perturbations also cause stronger behavioral and representational disruption than rate-matched random token substitutions in GPT-2; their behavioral effects are consistent across all six tested checkpoints. Our results show that robustness claims based on a single behavioral or representational metric can be misleading, and motivate multi-level evaluation of how perturbations alter language-model computation.

Chinese Translation

语言模型会遇到拼写错误、损坏的文本、修改过的单词和打乱的标记顺序，但鲁棒性通常仅通过输出行为进行评估。我们研究了六种自然和合成输入扰动如何在三层次上传播：输出行为、隐藏状态几何和注意力头功能。通过使用中心核对齐和内在维度分析层级几何，我们评估了四个GPT-2和两个Qwen2.5检查点的行为效应，并检查了GPT-2中的注意力头响应。扰动类型产生可区分的度量特征，这些特征并未完全通过输出度量捕获，并且在测试的检查点之间仅部分一致。复制分数特别与标记替换和洗牌下的激活修复恢复相关。梯度引导的HotFlip扰动在GPT-2中也导致比速率匹配的随机标记替换更强的行为和表征干扰；它们的行为效应在所有六个测试的检查点中是一致的。我们的结果表明，基于单一行为或表征度量的鲁棒性声明可能具有误导性，并促使对扰动如何改变语言模型计算进行多层次评估。

View on arXiv Download PDF AI Translation

cs.CL / 32 / 2609.03330

Less Is Moral: A CHARMing Framework for Moral Foundations Detection in Endorsement Behaviour

少即是多：一个用于识别支持行为中道德基础的 CHARM 框架

Fu, Huixiang, Rizoiu, Marian-Andrei

Abstract

Moral language plays a central role in shaping online endorsement and the diffusion of information, yet existing moral foundation detection systems often suffer from poor cross-domain generalization, weak rationale grounding, and reliance on costly prompting-based large language models (LLMs). We introduce CHARM, a MA\textbf{C}- and \textbf{H}ate-speech-\textbf{A}ware \textbf{R}ationale-aligned \textbf{M}oral foundation detection framework built on a lightweight fine-tuned LLM, which integrates complementary moral grounding, rationale alignment, and polarity-aware hate speech signals to support more robust and faithful moral prediction. Unlike prior dictionary-, fine-tune-, or prompt-based detectors, which decouple computation from psychological theory, CHARM is built so that each component -- MAC cross-attention, rationale alignment, and hate-speech modulation -- operationalizes a distinct psychological construct. Using a 30\% subsample of the MFTC, MFRC, and News training pools together with the richer supervision in MFTCXplain, CHARM improves AUC by up to 15.3\% in-domain, surpasses the supervised baselines on every out-of-domain dataset in both AUC and F1, and offers a scalable, low-cost alternative to prompting-based LLM detectors. We further apply CHARM to large-scale COVID-19 discourse on Twitter and show that moral value alignment is strongly associated with online endorsement behavior. By making moral framing measurable at scale, CHARM offers a practical tool for studying the spread of morally charged misinformation.

Chinese Translation

道德语言在塑造在线支持和信息传播中发挥着核心作用，但现有的道德基础检测系统往往在跨领域泛化、推理基础薄弱以及依赖昂贵的基于提示的大型语言模型（LLMs）方面存在问题。我们提出了 CHARM，一个基于轻量级微调 LLM 的 MA extbf{C} 和 extbf{H}ate-speech- extbf{A}ware extbf{R}ationale-aligned extbf{M}oral 基础检测框架，它集成了互补的道德基础、推理对齐和极性感知的仇恨言论信号，以支持更稳健和真实的道德预测。与之前的基于词典、微调或提示的检测器不同，后者将计算与心理理论解耦，CHARM 的每个组件——MAC 跨注意力、推理对齐和仇恨言论调制——都实现了一个独特的心理构念。通过使用 MFTC、MFRC 和 News 训练池的 30\% 子样本以及 MFTCXplain 中更丰富的监督，CHARM 在领域内的 AUC 提高了多达 15.3\%，在每个领域外的数据集上超越了监督基线，无论是在 AUC 还是 F1 指标上，并提供了一种可扩展、低成本的替代方案，取代基于提示的 LLM 检测器。我们进一步将 CHARM 应用于 Twitter 上的大规模 COVID-19 讨论，并显示道德价值对齐与在线支持行为之间存在强关联。通过在大规模上使道德框架可测量，CHARM 为研究道德充沛的虚假信息传播提供了一个实用工具。

View on arXiv Download PDF AI Translation

cs.CL / 33 / 2609.03331

FPCO-Dialog: A Multi-Turn False-Premise Benchmark for Correction and Cooperation in Vision-Language Models

FPCO-Dialog：用于视觉语言模型中纠正与合作的多轮错误前提基准

Ma, Jiayuan, Lu, Yuqi, Guo, Weiyang, Wang, Chenrui, Shu, Junyi, Liu, Xuebo, Zhang, Min, Li, Jing

Abstract

Vision-language models (VLMs) are increasingly deployed in multi-turn settings where users may describe visual content with incorrect assumptions. Yet existing evaluations rarely isolate how models respond when the same visually grounded false premise persists across dialogue turns. We introduce FPCO-Dialog, a benchmark for evaluating correction and cooperation behavior in VLMs under repeated false premises. FPCO-Dialog contains 1,080 images and 10,800 question turns, stratified by visual complexity, object category, and false-premise class, and uses a 10-turn protocol in which a correct dialogue prefix is followed by repeated false-premise referring expressions. We evaluate 20 commercial and open-source VLMs with a model-agnostic protocol and CorrTP@K, a correction-rate metric over false-premise turns, scored by two independent detectors. FPCO-Dialog reveals substantial and persistent cross-model differences in aggregate correction tendency, model-specific turn-wise dynamics, and systematic variation across false-premise types under the benchmark's substitution distribution. The dataset, evaluation protocol, model outputs, detector labels, and code are available.

Chinese Translation

视觉语言模型（VLMs）越来越多地应用于多轮对话场景中，用户可能会以错误的假设描述视觉内容。然而，现有的评估很少单独考察当相同的视觉基础错误前提在对话轮次中持续存在时模型的反应。我们引入了FPCO-Dialog，这是一个用于评估VLMs在重复错误前提下的纠正与合作行为的基准。FPCO-Dialog包含1,080张图像和10,800个问题轮次，按视觉复杂性、物体类别和错误前提类别进行分层，并使用10轮协议，其中正确的对话前缀后跟随重复的错误前提指代表达。我们使用模型无关的协议和CorrTP@K（一个针对错误前提轮次的纠正率指标，由两个独立检测器评分）评估了20个商业和开源的VLMs。FPCO-Dialog揭示了在总体纠正倾向、模型特定的轮次动态以及在基准的替代分布下不同错误前提类型之间的系统性变化方面，模型之间存在显著且持久的差异。数据集、评估协议、模型输出、检测器标签和代码均可获取。

View on arXiv Download PDF AI Translation

cs.CL / 34 / 2609.03366

Accountable AI with Grounded, Faithful, Consistent, Actionable Rationales: A Case Study in Clinical Trial Matching with VERDICT

具有扎实、可信、一致和可操作性理由的负责任人工智能：以VERDICT为例的临床试验匹配研究

Zhou, Zikai, Jin, Yufei, Xu, Yilin, Wang, Yu-Chiang, Chao, Chieh-Ju, Lam, Monica S.

Abstract

Accountability means a decision can be examined, justified, and contested. LLMs make this hard: fluent output may be ungrounded, incomplete, or unfaithful to the decision process. Achieving accountability requires verified rationales (how was the decision reached), assumptions (what was assumed rather than known), policy consistency (the same treatment for the same facts), and pivotal conditions (what would change the outcome). We introduce self-faithfulness as an automatic test of accountability: changing the pivotal conditions should change the decision. We examine accountable AI through clinical trial matching, a high-stakes task central to evidence-based medicine. Although LLM-based matchers match patients to trials reasonably accurately, they apply decision policies inconsistently and produce rationales that are unfaithful to their own decisions. We introduce VERDICT, an LLM-based agent that translates a decision task, its constraints, and its policy into Satisfiability Modulo Theories (SMT), then derives the decision with SMT and MaxSMT solvers -- so policies are applied consistently and decisions are accountable by construction. Across a SIGIR 2016-derived dataset and TREC 2021, VERDICT achieves the strongest decision accuracy among LLM-only and neurosymbolic baselines, applies policies with perfect consistency, and produces clinician-preferred rationales grounded in explicit assumptions and pivotal conditions, with improved counterfactual self-faithfulness.

Chinese Translation

问责制意味着决策可以被审查、证明和质疑。大型语言模型（LLMs）使这一过程变得困难：流畅的输出可能缺乏基础、信息不全或与决策过程不一致。实现问责制需要经过验证的理由（决策是如何做出的）、假设（假定的而非已知的内容）、政策一致性（对相同事实采取相同处理）和关键条件（什么会改变结果）。我们引入自我可信性作为问责制的自动测试：改变关键条件应当改变决策。我们通过临床试验匹配这一与循证医学密切相关的高风险任务来考察负责任的人工智能。尽管基于LLM的匹配器能够相对准确地将患者与试验匹配，但它们在决策政策的应用上不一致，并且产生的理由与自身决策不一致。我们介绍了VERDICT，一个基于LLM的代理，它将决策任务、约束条件和政策转换为可满足性理论（Satisfiability Modulo Theories, SMT），然后通过SMT和MaxSMT求解器推导出决策——因此政策得以一致应用，决策在构建上是可问责的。在一个基于SIGIR 2016的数据集和TREC 2021中，VERDICT在LLM-only和神经符号基线中实现了最强的决策准确性，以完美的一致性应用政策，并产生基于明确假设和关键条件的临床医生偏好的理由，同时提高了反事实自我可信性。

View on arXiv Download PDF AI Translation

cs.CL / 35 / 2609.03370

FrameBench:A Language Understanding Benchmark Based on Frame Semantics

FrameBench：基于框架语义的语言理解基准

Yano, Chihiro, Sasano, Ryohei

Abstract

In frame semantics, sentence comprehension is assumed to proceed by relating lexical meaning to background knowledge called semantic frames, thereby enabling readers to implicitly enrich the text with unstated information. Recent large language models (LLMs) have achieved strong performance across a wide range of downstream tasks. However, it remains unclear whether they can reproduce the kinds of implicit enrichment that humans naturally make during comprehension. To address this question, we introduce FrameBench, a benchmark grounded in frame semantics. FrameBench consists of multiple-choice questions that test whether models distinguish the frames evoked by the same verb across contexts. We construct the benchmark for English and Japanese using FrameNet-style resources and a generation-and-verification pipeline with native-speaker judgments. Our experiments on a diverse set of models reveal challenges for small models, while several large models surpass the human reference scores. We release the constructed FrameBench dataset and the code for dataset construction and evaluation at https://github.com/SasanoLab/FrameBench.

Chinese Translation

在框架语义中，句子理解被假设为通过将词汇意义与称为语义框的背景知识相关联，从而使读者能够隐式地丰富文本中的未陈述信息。近期的大型语言模型（LLMs）在广泛的下游任务中取得了强劲的表现。然而，目前尚不清楚它们是否能够重现人类在理解过程中自然进行的隐式丰富。为了解决这个问题，我们引入了FrameBench，这是一个基于框架语义的基准。FrameBench由多项选择题组成，测试模型是否能够区分在不同语境中由同一动词引发的框架。我们使用FrameNet风格的资源和一个生成与验证的流程（结合母语者的判断）构建了英语和日语的基准。我们对多种模型的实验揭示了小型模型面临的挑战，而一些大型模型则超越了人类参考分数。我们在https://github.com/SasanoLab/FrameBench发布了构建的FrameBench数据集及数据集构建和评估的代码。

View on arXiv Download PDF AI Translation

cs.CL / 36 / 2609.03394

Chiaroscuro for Emotions: A Contrastive Emotion Benchmark Grounded in Appraisal Theory

情感的明暗对比：基于评估理论的对比情感基准

Bommana, Divyesh, Saim, Mohammad, Jiang, Tianyu

Abstract

Emotion recognition benchmarks often predict one emotion per text, missing many real-world scenarios where two people arrive at opposing emotions from a single shared event. For example, a child kicks the seat in front of her in excitement while the passenger ahead grows angry. We introduce CHIARO, a 1,000 human-annotated sentence benchmark for contrastive emotion inference grounded in appraisal theory. Each scene describes one causal trigger eliciting a positive emotion in one person and a negative emotion in the other, drawn from a ten-class taxonomy. We benchmark seven frontier LLMs and four off-the-shelf emotion classifiers. The strongest LLM reaches 67.3 macro-F1, well below human agreement, while existing emotion classifiers score near chance. Beyond evaluation, CHIARO also serves as a training signal. When combined with an existing emotion corpus, the resulting downstream classifier improves on CHIARO itself and on six of ten external emotion benchmarks, which positions our dataset as a complementary signal for emotion recognition.

Chinese Translation

情感识别基准通常预测每个文本中的一种情感，忽视了许多现实场景，其中两个人对同一共享事件产生相反的情感。例如，一个孩子因兴奋而踢前面的座椅，而前面的乘客则感到愤怒。我们介绍了CHIARO，这是一个包含1,000个人工标注句子的对比情感推断基准，基于评估理论。每个场景描述一个因果触发因素，在一个人身上引发积极情感，而在另一个人身上引发消极情感，来源于一个十类分类法。我们对七个前沿大型语言模型（LLMs）和四个现成的情感分类器进行了基准测试。最强的LLM达到了67.3的宏观F1值，远低于人类一致性，而现有的情感分类器的得分接近随机水平。除了评估，CHIARO还作为训练信号。当与现有的情感语料库结合时，得到的下游分类器在CHIARO本身和十个外部情感基准中的六个上都有所提升，这使我们的数据集成为情感识别的补充信号。

View on arXiv Download PDF AI Translation

cs.CL / 37 / 2609.03395

TabScope: Question-Adaptive Scope Selection for Table Question Answering

TabScope：针对表格问答的问答自适应范围选择

Wang, Yuxiang, Gan, Junhao, Qi, Jianzhong

Abstract

Large Language Models (LLMs) have shown strong performance on table question answering, yet their accuracy often degrades as table size increases. We find that this degradation is not uniform across question types. Localization-sensitive questions are particularly affected by irrelevant table content, while questions requiring broader evidence may still benefit from full-table reasoning. Based on this observation, we propose a question-adaptive framework that dynamically selects between localized and full-table reasoning. The framework constructs question-specific sub-tables through operation-aware table decomposition and uses the predicted question type to determine the appropriate reasoning mode. We further introduce silver reference sub-tables for evaluating evidence selection and construct SLQA, a benchmark based on real-world long tables. Experiments on WikiTQ and SLQA show that localization is particularly effective for lookup and local reasoning questions, while adaptive selection between localized and full-table reasoning achieves the best overall performance. These results highlight that long-table QA requires deciding not only how to localize, but also when to localize. Our code and datasets will be made available upon publication of the paper.

Chinese Translation

大型语言模型（LLMs）在表格问答方面表现出色，但随着表格规模的增加，其准确性往往会下降。我们发现这种下降在不同问题类型之间并不均匀。对位置敏感的问题尤其受到无关表格内容的影响，而需要更广泛证据的问题仍可能从全表推理中受益。基于这一观察，我们提出了一种问答自适应框架，该框架动态选择局部推理和全表推理之间的模式。该框架通过操作感知的表格分解构建特定于问题的子表，并利用预测的问题类型来确定适当的推理模式。我们进一步引入了银参考子表用于评估证据选择，并构建了基于真实世界长表的基准SLQA。对WikiTQ和SLQA的实验表明，局部化对查找和局部推理问题特别有效，而局部推理与全表推理之间的自适应选择则实现了最佳的整体性能。这些结果强调了长表问答不仅需要决定如何局部化，还需要决定何时局部化。我们的代码和数据集将在论文发表后公开。

View on arXiv Download PDF AI Translation

cs.CL / 38 / 2609.03410

To What Extent Do Large Language Models Understand Bangla Idioms?

大型语言模型在多大程度上理解孟加拉语习语？

Akter, Mousumi, Sayeedi, Md. Faiyaz Abdullah, Sayeedi, Nurul Labib, Shatabda, Swakkhar

Abstract

Idiomatic expressions are an integral part of natural language, reflecting cultural nuances and posing unique challenges for computational models, particularly in low-resource languages. In this paper, we present the first large-scale benchmark dataset of Bangla idioms, complemented by a synthetic multiple-choice question (MCQ) dataset for idiom meaning identification. We conduct a comprehensive evaluation of recent large language models (LLMs) across three idiom-related tasks: paraphrasing, idiom span detection, and meaning identification, leveraging zero-shot and few-shot prompting strategies. Our results reveal substantial variability in model performance, with no single LLM consistently outperforming others across all tasks. Notably, Phi-4-mini-instruct excels in paraphrasing, Kimi-K2-32b-instruct in span detection, and Gemini-2.5-flash in meaning identification. We believe that our datasets and analyses will provide valuable resources to guide future research in improving LLM comprehension of idiomatic expressions, particularly in Bangla and other low-resource languages.

Chinese Translation

习语是自然语言的重要组成部分，反映了文化的细微差别，并为计算模型带来了独特的挑战，尤其是在资源匮乏的语言中。本文呈现了首个大规模的孟加拉语习语基准数据集，并附有用于习语意义识别的合成多项选择题（MCQ）数据集。我们对近期的大型语言模型（LLMs）在三个与习语相关的任务上进行了全面评估：释义、习语跨度检测和意义识别，利用零样本和少样本提示策略。我们的结果显示模型性能存在显著差异，没有单一的LLM在所有任务中始终优于其他模型。值得注意的是，Phi-4-mini-instruct在释义任务中表现优异，Kimi-K2-32b-instruct在跨度检测中表现突出，而Gemini-2.5-flash在意义识别中表现最佳。我们相信我们的数据集和分析将为未来研究提供宝贵资源，以改善LLM对习语的理解，特别是在孟加拉语和其他资源匮乏语言中。

View on arXiv Download PDF AI Translation

cs.CL / 39 / 2609.03426

Lngram v2: Latent N-Gram Memory with Interpretable Discrete Representations

Lngram v2：具有可解释离散表示的潜在N-gram记忆

Zheng, Yunao, Wen, Bin, Wang, Xiaojie

Abstract

Transformers lack a native lookup mechanism, requiring repeated dense computation to recognize and reuse local static patterns. Lngram v1 introduces tokenizer-independent conditional memory through discrete latent n-gram addressing, but its memory capacity is coupled with the backbone width, limiting scalability due to high parameter and activation costs. We propose Lngram v2, which decouples the number of routes, memory dimension, and backbone width, and introduces a context-aware grouped-query attention readout to scale memory capacity independently. A zero-value Sink and counterfactual surrogate gradients further improve readout selectivity and routing trainability while preserving hard discrete addressing. Experiments across vision--language models (VLMs) of different scales show consistent improvements, including successful scaling to a 30B-parameter model. Compared with Lngram v1, Lngram v2 substantially reduces both total and activated memory parameters while maintaining or improving language modeling performance. Further analysis shows that its discrete IDs preserve substantial semantic structure of continuous hidden states, enabling semantic recovery from IDs alone and stable ID--semantic associations across datasets. These results establish Lngram v2 as an efficient and scalable latent conditional memory mechanism whose discrete addresses also provide a structured interface for analyzing internal model representations.

Chinese Translation

Transformer缺乏原生的查找机制，需反复进行密集计算以识别和重用局部静态模式。Lngram v1通过离散潜在n-gram寻址引入了与分词器无关的条件记忆，但其记忆容量与主干网络宽度耦合，因参数和激活成本高限制了可扩展性。我们提出了Lngram v2，解耦了路径数量、记忆维度和主干宽度，并引入了上下文感知的分组查询注意力读出机制，实现记忆容量的独立扩展。零值汇聚（Sink）和反事实代理梯度进一步提升了读出选择性和路由可训练性，同时保持了硬离散寻址。不同规模的视觉-语言模型（VLM）实验表明，Lngram v2表现出一致的改进，成功扩展至30亿参数模型。与Lngram v1相比，Lngram v2显著减少了总记忆参数和激活记忆参数，同时保持或提升了语言建模性能。进一步分析显示，其离散ID保留了连续隐藏状态的显著语义结构，实现了仅通过ID的语义恢复及跨数据集的稳定ID-语义关联。这些结果确立了Lngram v2作为一种高效且可扩展的潜在条件记忆机制，其离散地址还为分析内部模型表示提供了结构化接口。

View on arXiv Download PDF AI Translation

cs.CL / 40 / 2609.03430

Random Attention: Rethinking KV Cache Eviction for Efficient Reasoning

随机注意力：重新思考 KV 缓存驱逐以实现高效推理

Wang, Heng, Qiu, Jielin, Zhao, Wenting, Qian, Cheng, Yang, Liangwei, Han, Jiawei, Ji, Heng, Savarese, Silvio, Heinecke, Shelby, Wang, Huan

Abstract

Large language models achieve superior performance on tasks that require extended reasoning, but long chains of thought make the KV cache a severe memory bottleneck. Existing KV cache compression methods share one paradigm: score each cached token by some estimate of how much it will matter later, and keep the top-scoring ones. We show that the selection signal contributes almost nothing. Random Attention keeps the prompt and evicts uniformly at random within each attention head, computing no score at all; across four models and six reasoning tasks it matches the strongest prior evictor while serving 32-43% higher throughput than it in vLLM deployment. Controlled experiments explain this by showing that 1) the prompt is the fragile part of the cache, and most of the gap between selectors is just whether their selection signal happened to keep it; 2) the reasoning trace protects itself against eviction with redundancy at two levels, in the text (the model restates what it still needs as it works) and across attention heads (each keeps its own copy of the trace), so once the prompt is safe, a random draw retains enough copies of what the model still needs, and no score is required to pick them. Our code is publicly available at https://github.com/SalesforceAIResearch/Random-Attention.

Chinese Translation

大型语言模型在需要扩展推理的任务中表现出色，但长链思维使得 KV 缓存成为严重的内存瓶颈。现有的 KV 缓存压缩方法共享一个范式：通过某种估计来为每个缓存的令牌打分，以判断其后续的重要性，并保留得分最高的令牌。我们表明，选择信号几乎没有贡献。随机注意力（Random Attention）保持提示，并在每个注意力头内均匀随机驱逐，完全不计算得分；在四个模型和六个推理任务中，它的表现与最强的先前驱逐方法相匹配，同时在 vLLM 部署中提供了 32-43% 更高的吞吐量。控制实验解释了这一现象，表明 1）提示是缓存中脆弱的部分，选择器之间的差距主要在于它们的选择信号是否恰好保留了提示；2）推理轨迹通过在文本中（模型在工作时重述其仍需的内容）和跨注意力头（每个头保留其自身的轨迹副本）两个层面上的冗余自我保护，因此一旦提示安全，随机抽取就能保留足够的模型仍需的副本，而不需要选择得分。我们的代码已公开发布在 https://github.com/SalesforceAIResearch/Random-Attention。

View on arXiv Download PDF AI Translation

cs.CL / 41 / 2609.03432

Decoupled Analysis-Judging: An Automated Creativity Evaluator Using LLMs in Complex Multi-step Creativity Tasks

解耦分析-判断：在复杂多步骤创造性任务中使用大型语言模型的自动化创造力评估器

Wang, Xiangyu, Wu, Jin, Li, Xiaoyu, Zheng, Chanjin, Zhou, Yifeng

Abstract

Automated evaluation of creativity tasks remains challenging for LLM-as-a-Judge, as LLM is susceptible to biases such as verbosity bias and leniency bias. Such limitations are particularly evident in Contextually-Grounded and Procedurally-Structured Tasks (CGPST), a complex multi-step creativity task where inter-step dependencies, highly subjectivity, and wide scoring ranges lead to more unstable and biased judgments. Existing approaches either rely on task-specific training or directly apply LLM-as-a-Judge, both of which struggle to ensure reliable evaluation under such complexity. To bridge these gaps, we propose CreaEval, an automated creativity evaluator for CGPST that decouples typical LLM-as-a-Judge into analysis and judging. Correspondingly, CreaEval involves two critical phases: Memory-augmented Analysis, a SoT-LLM converts multi-step responses into structured evaluation evidence, incorporating cross-step memory; and Evidence-based Judging, a Judge-LLM uses the extracted evidence for judging without accessing raw responses. Comprehensive experiments show that CreaEval achieves an average performance improvement of 22.74% over the second-best baselines across CGPST and two classic simple creativity tasks, demonstrating its generalizability. The code is available at https://github.com/Jaong/CreaEval.

Chinese Translation

对于大型语言模型（LLM）作为评判者的自动化创造力任务评估仍然具有挑战性，因为LLM容易受到冗长偏见和宽容偏见等偏见的影响。这些局限性在上下文基础和程序结构化任务（CGPST）中尤为明显，这是一种复杂的多步骤创造性任务，其中步骤间的依赖性、高度主观性和广泛的评分范围导致更不稳定和偏见的判断。现有方法要么依赖于特定任务的训练，要么直接应用LLM作为评判者，这两者在如此复杂的情况下都难以确保可靠的评估。为了解决这些问题，我们提出了CreaEval，一个用于CGPST的自动化创造力评估器，它将典型的LLM作为评判者解耦为分析和判断两个部分。相应地，CreaEval涉及两个关键阶段：增强记忆的分析（Memory-augmented Analysis），一个SoT-LLM将多步骤响应转换为结构化评估证据，并结合跨步骤记忆；以及基于证据的判断（Evidence-based Judging），一个Judge-LLM利用提取的证据进行判断，而不访问原始响应。全面的实验表明，CreaEval在CGPST和两个经典简单创造性任务中平均性能提升了22.74%，展示了其通用性。代码可在 https://github.com/Jaong/CreaEval 获取。

View on arXiv Download PDF AI Translation

cs.CL / 42 / 2609.03454

When Retrieval Helps: Selective Retrieval for Single-Turn Mental-Health QA

检索何时有助：单轮心理健康问答的选择性检索

Oh, Hyunseo, Kim, Chong-Kwon, Choi, Yoonhyuk

Abstract

Retrieval-augmented generation (RAG) can improve the specificity and grounding of large language model responses, but its effect is not uniformly beneficial in single-turn mental-health question answering, where user queries often combine emotional distress, treatment concerns, and safety-sensitive needs. We study when retrieval helps or hurts mental-health QA, and whether a lightweight selective retrieval policy can better control this trade-off. We operationalize retrieval need using three draft-conditioned utility dimensions: psychoeducational need, coping need, and response specificity, together with a rule-based safety trigger. Following psychotherapy-grounded RAG systems such as coTherapist, we construct a compact and controllable guideline corpus comprising coping-strategy, psychoeducational, and safety resources. We fine-tune an instruction-tuned generator on MentalChat16K using QLoRA and compare Closed-book, Always Retrieval, and Selective Retrieval settings on CounselBench-Eval and CounselBench-Adv. Experiments show that retrieval is not uniformly beneficial in this domain. Always Retrieval improves specificity but lowers overall quality and introduces additional safety-sensitive failures. Selective Retrieval preserves closed-book behavior for low-need cases while avoiding the additional degradation caused by unconditional retrieval, supporting the view that retrieval activation is a safety-sensitive control decision.

Chinese Translation

检索增强生成（RAG）可以提高大型语言模型响应的特异性和基础性，但在单轮心理健康问答中，其效果并不总是有利，因为用户查询通常结合了情感困扰、治疗关注和安全敏感需求。我们研究了检索何时有助于或伤害心理健康问答，以及轻量级选择性检索策略是否可以更好地控制这一权衡。我们通过三个草稿条件的效用维度来操作化检索需求：心理教育需求、应对需求和响应特异性，并结合基于规则的安全触发器。借鉴以心理治疗为基础的RAG系统，如coTherapist，我们构建了一个紧凑且可控的指南语料库，包括应对策略、心理教育和安全资源。我们在MentalChat16K上使用QLoRA微调了一个指令调优生成器，并在CounselBench-Eval和CounselBench-Adv上比较了闭卷、始终检索和选择性检索设置。实验表明，在该领域检索并不总是有利。始终检索提高了特异性，但降低了整体质量，并引入了额外的安全敏感失败。选择性检索在低需求情况下保持了闭卷行为，同时避免了无条件检索造成的额外降级，支持了检索激活是一个安全敏感控制决策的观点。

View on arXiv Download PDF AI Translation

cs.CL / 43 / 2609.03467

When Users Don't Ask: Benchmarking Context-Driven Memory Retrieval in Conversational Agents

当用户不提问时：基于上下文的记忆检索在对话代理中的基准测试

Chang, Wen-Yu, Chen, Yun-Nung

Abstract

Large language models (LLMs) are increas- ingly deployed as long-horizon conversational agents, motivating growing interest in mem- ory systems. However, existing benchmarks primarily evaluate memory through QA-style probing rather than in-situ conversational usage. We introduce LOCOMO-CONV, a conversa- tional memory benchmark derived from Lo- CoMo with four query styles: dialog, implicit, counterfactual, and composed. Across five rep- resentative memory systems, we evaluate both retrieval recall and end-to-end response qual- ity. Our experiments show that conversational framing exposes substantial retrieval gaps over- looked by QA benchmarks, especially on im- plicit and composed queries, which multi-facet query rewriting narrows for raw-turn mem- ory but not abstractive memory. We further find that strong retrieval does not fully trans- late into response quality, and that implicit queries exhibit silent grounding, where mem- ory improves contextual grounding without ex- plicitly surfacing the gold fact. These results point to reasoning-based memory elaboration as a promising direction, and we release aux- iliary supportive_memory annotations captur- ing conversationally useful context beyond the original gold evidence.

Chinese Translation

大型语言模型（LLMs）越来越多地被用作长时间对话的代理，这引发了对记忆系统的日益关注。然而，现有的基准主要通过问答（QA）风格的探测来评估记忆，而不是在实际对话中的使用。我们引入了LOCOMO-CONV，这是一个源自Lo-CoMo的对话记忆基准，包含四种查询风格：对话、隐式、反事实和组合。在五个代表性的记忆系统中，我们评估了检索召回率和端到端响应质量。我们的实验表明，对话框架揭示了QA基准所忽视的显著检索差距，尤其是在隐式和组合查询上，对于原始轮次记忆，多面查询重写缩小了这一差距，但对抽象记忆则没有。我们进一步发现，强大的检索并未完全转化为响应质量，隐式查询表现出静默基础，即记忆改善了上下文基础，而没有明确呈现黄金事实。这些结果指向基于推理的记忆扩展作为一个有前景的方向，我们发布了辅助的支持性记忆注释，捕捉了超出原始黄金证据的对话有用上下文。

View on arXiv Download PDF AI Translation

cs.CL / 44 / 2609.03487

Pattern Over-Generalization of Knowledge Graph Embedding

知识图谱嵌入的模式过度泛化

Kim, Junsik, Kim, Kangil

Abstract

Knowledge graph embedding (KGE) demonstrates its effectiveness for predicting missing links in knowledge graphs (KGs) by projecting entities and relations into a low-dimensional vector space. It is crucial for KGE models to effectively capture inference patterns (patterns) inherent in KGs, such as symmetry/antisymmetry, inversion and composition. Although recent KGE models exhibit strong capabilities in modeling such diverse patterns, they suffer from inherent limitations stemming from pattern over-generalization, where embeddings learned from only a single pattern instance inevitably generalize that pattern to all related instances, i.e., generalize the pattern universally. To address this issue, we propose PogRE (Pattern Over-Generalization Robust Embedding), a simple but effective method that utilizes dense linear transformations and compound operations for relation representation. Our theoretical analysis demonstrates that a dense linear transformation allows a pattern to become progressively universal as more triples are observed in the pattern. Furthermore, after observing d+1 linearly independent entities (d+1 denotes the dimension of entity), the linear transformation guarantees universal generalization of the pattern across all related instances. Experimental results on three standard benchmark datasets show that PogRE outperforms existing state-of-the-art KGE models in link prediction. Moreover, our empirical results indicate that PogRE effectively addresses the negative impact of over-generalization.

Chinese Translation

知识图谱嵌入（KGE）通过将实体和关系投影到低维向量空间，展示了其在预测知识图谱（KGs）中缺失链接方面的有效性。对于KGE模型而言，有效捕捉知识图谱中固有的推理模式（模式）至关重要，例如对称/反对称、反转和组合。尽管近期的KGE模型在建模这些多样化模式方面表现出强大的能力，但它们受到模式过度泛化的固有限制，即从单一模式实例学习到的嵌入不可避免地将该模式泛化到所有相关实例，即普遍泛化该模式。为了解决这个问题，我们提出了PogRE（模式过度泛化鲁棒嵌入），这是一种简单但有效的方法，利用稠密线性变换和复合操作进行关系表示。我们的理论分析表明，稠密线性变换使得随着观察到更多三元组，模式逐渐变得普遍。此外，在观察到d+1个线性独立的实体后（d+1表示实体的维度），线性变换保证了模式在所有相关实例中的普遍泛化。在三个标准基准数据集上的实验结果表明，PogRE在链接预测方面优于现有的最先进KGE模型。此外，我们的实证结果表明，PogRE有效地解决了过度泛化的负面影响。

View on arXiv Download PDF AI Translation

cs.CL / 45 / 2609.03502

Building and Evaluating Fixed-Voice Thai TTS from Synthetic Speech

构建和评估基于合成语音的固定语音泰语文本转语音系统

Pipatanakul, Kunat, Manakul, Potsawee, Sirichotedumrong, Warit, Sripaisarnmongkol, Sittipong, Nathong, Pakorn, Jirabovonvisut, Phatrasek

Abstract

In low-resource settings, deploying TTS typically requires choosing between a large voice-cloning model with costly inference or a compact fixed-voice system that requires a speaker-specific corpus. We study a third route: using a large voice-cloning model as a programmable data source to turn a short voice reference (e.g., 15 seconds) into a compact fixed-voice student trained entirely on synthetic speech. This setting makes pipeline design consequential: teacher errors become training targets, while filtering failed generations can reduce coverage of difficult texts. Thai further introduces challenges from ambiguous word boundaries, lexical tone, names and loanwords, numeric verbalization, and Thai-English code-switching. We study how text preparation, synthetic generation, quality filtering, rejection sampling, and frontend choices affect the resulting student, and where teacher limitations remain. We evaluate CER, Challenge-Set Keyword Accuracy, Prosody Pause Accuracy, speaker similarity, and speaking rate. The resulting 82M-parameter model, Wayu-Paxa-TTS-Edge, enables on-device Thai TTS without reference audio. It achieves 68.2% Challenge-Set Keyword Accuracy (85.5% of Gemini 3.1) and 91.4% pause precision, outperforming its OmniVoice teacher (89.9%) and reaching 94.8% of Gemini 3.1. It also achieves the lowest pause-placement error and intra-word pause rates among the three systems, and 3.7% and 1.1% CER on Thai and English, respectively. We open-source the model and evaluation framework for Thai TTS development.

Chinese Translation

在资源匮乏的环境中，部署文本转语音（TTS）系统通常需要在大型语音克隆模型（具有高昂的推理成本）和需要特定说话者语料库的紧凑型固定语音系统之间进行选择。我们研究了一条第三条路径：利用大型语音克隆模型作为可编程数据源，将短语音参考（例如，15秒）转化为完全基于合成语音训练的紧凑型固定语音学生。这种设置使得管道设计变得至关重要：教师错误成为训练目标，而过滤失败的生成可以减少对困难文本的覆盖。泰语进一步引入了模糊的词边界、词汇音调、姓名和外来词、数字口语化以及泰英代码切换等挑战。我们研究了文本准备、合成生成、质量过滤、拒绝采样和前端选择如何影响最终的学生模型，以及教师的局限性所在。我们评估了字符错误率（CER）、挑战集关键词准确率、韵律停顿准确率、说话者相似性和语速。最终得到的8200万参数模型Wayu-Paxa-TTS-Edge，能够在没有参考音频的情况下实现设备上的泰语TTS。它达到了68.2%的挑战集关键词准确率（相当于Gemini 3.1的85.5%）和91.4%的停顿精度，超越了其OmniVoice教师（89.9%），并达到了Gemini 3.1的94.8%。在三个系统中，它还实现了最低的停顿放置错误率和词内停顿率，以及在泰语和英语中分别为3.7%和1.1%的CER。我们将该模型和评估框架开源，以促进泰语TTS的发展。

View on arXiv Download PDF AI Translation

cs.CL / 46 / 2609.03511

Lost in Reordering: Structural Sensitivity of Multilingual LLMs under Semantics-Preserving Perturbations

迷失在重排序中：多语言大型语言模型在保持语义的扰动下的结构敏感性

Nhayakkat, Karthika, Verma, Rajat, Brahma, Maharaj, Mahesh, Vetcha Gnana, Desarkar, Maunendra Sankar, Ramakrishnan, Ganesh, Saluja, Rohit

Abstract

Large Language Models (LLMs) demonstrate strong multilingual reasoning performance, yet their robustness to semantics-preserving structural variation remains underexplored, particularly for relatively free word-order languages. We investigate the structural sensitivity of multilingual LLMs using two linguistically grounded perturbation settings in Hindi and Malayalam: constrained constituent reordering and active-passive voice transformation. We introduce a benchmark dataset IndicReStruct, with two variants, GSM8K-Reordered and GSM8K-Voice, constructed from GSM8K while preserving semantic meaning. Across six state-of-the-art LLMs and multiple prompting strategies, we observe consistent and significant degradation in mathematical reasoning performance under structurally perturbed inputs. To further understand these failures, we perform qualitative error analysis and mechanistic interpretability experiments using residual-stream activation patching. Our analyses show that reasoning failures frequently arise from disruptions in entity-quantity alignment and that intermediate transformer layers contribute most strongly toward reasoning restoration. Overall, our findings suggest that current multilingual LLMs remain highly sensitive to surface syntactic realization and lack robust compositional invariance under structurally different but semantically equivalent inputs.

Chinese Translation

大型语言模型（LLMs）展现出强大的多语言推理能力，但它们对保持语义的结构变化的鲁棒性仍然未被充分探讨，尤其是在相对自由的词序语言中。我们通过在印地语和马拉雅拉姆语中使用两种基于语言学的扰动设置：受限成分重排序和主动-被动语态转换，来研究多语言LLMs的结构敏感性。我们引入了一个基准数据集IndicReStruct，包含两个变体GSM8K-Reordered和GSM8K-Voice，这些变体是从GSM8K构建的，同时保持语义意义。在六个最先进的LLMs和多种提示策略下，我们观察到在结构扰动输入下，数学推理性能的一致且显著的下降。为了进一步理解这些失败，我们进行了定性错误分析和机制可解释性实验，使用残差流激活补丁。我们的分析表明，推理失败通常源于实体-数量对齐的中断，而中间变换层对推理恢复的贡献最为显著。总体而言，我们的研究结果表明，当前的多语言LLMs对表面句法实现高度敏感，并且在结构上不同但语义上等价的输入下缺乏稳健的组合不变性。

View on arXiv Download PDF AI Translation

cs.CL / 47 / 2609.03577

Language, Language Models, and What We're Talking About

语言、语言模型与我们所谈论的内容

Nissim, Malvina

Abstract

Language models are commonly discussed as technical artefacts, but they are obviously shaped by the linguistic worlds conveyed by data during their training. Using Italian language models as evidence, I want to bring attention to the nature of the systems which result from training and specialising models on translated and synthetic data, and further curating them, and to the meaning of testing them on equally unnatural data. Are these eventually models of Italian? Are they models of language? Does NLP still care about language? These questions yield another, more concrete question: what language do we actually want language models to produce? I argue that this question cannot be answered if we do not first consider a clearer distinction between language models designed as technical products and language models designed as tools for studying language itself. The answers then might be diverse, the languages we are talking about might be diverse, and the picture might not be as pessimistic as we fear.

Chinese Translation

语言模型通常被视为技术产物，但显然它们受到训练过程中数据所传达的语言世界的影响。以意大利语模型为证据，我希望引起人们对通过翻译和合成数据训练和专业化模型所形成的系统的性质，以及在同样不自然的数据上测试它们的意义的关注。这些模型最终是意大利语的模型吗？它们是语言的模型吗？自然语言处理（NLP）是否仍然关心语言？这些问题引出了另一个更具体的问题：我们实际上希望语言模型生成什么语言？我认为，如果我们不首先考虑将语言模型作为技术产品与将其作为研究语言工具之间的更清晰的区别，这个问题是无法回答的。那么，答案可能是多样的，我们所谈论的语言也可能是多样的，整体情况可能并不像我们所担心的那样悲观。

View on arXiv Download PDF AI Translation

cs.CL / 48 / 2609.03595

How Far Can Synthetic Data Take Thai OCR?

合成数据能为泰国OCR带来多大帮助？

Pipatanakul, Kunat

Abstract

We investigate what makes synthetic OCR supervision transfer to real Thai documents and use the resulting insights to build Wayu-Paxa-OCR-Zero, a Thai OCR model adapted without OCR labels from real Thai document pages. Synthetic data provide exact labels at scale, but "realism" conflates source domain, page context, typography, spatial structure, and glyph variation. We disentangle these factors with a controlled document-reconstruction pipeline and evaluate each variant under page- and crop-level training on printed and handwritten Thai documents. Non-text context has little consistent effect, whereas typeface diversity, two-dimensional structure, and real handwriting glyphs improve transfer; moreover, source-domain matching depends on training granularity, with in-domain reconstruction approaching real printed supervision under page-level training (1.82% versus 1.31% median character error rate) but underperforming out-of-domain reconstruction under crop-level training (15.59% versus 5.52%). Guided by these findings, we adapt the 0.9B-parameter PaddleOCR-VL-1.6 into Wayu-Paxa-OCR-Zero using 45,723 synthetic pages: relative to its base checkpoint, it reduces median character error rate from 6.64% to 1.24% on printed pages and from 74.87% to 20.55% on handwriting and outperforms Typhoon OCR v1 7B on all five evaluation sets, showing that synthetic-only training can be competitive.

Chinese Translation

我们研究了合成OCR监督如何转移到真实的泰国文档，并利用所得见解构建了Wayu-Paxa-OCR-Zero，这是一个在没有真实泰国文档页面OCR标签的情况下适应的泰国OCR模型。合成数据提供了大规模的精确标签，但“真实感”混淆了源领域、页面上下文、排版、空间结构和字形变体。我们通过一个受控的文档重建流程解开这些因素，并在印刷和手写的泰国文档上对每种变体进行页面和裁剪级别的训练评估。非文本上下文的影响不一致，而字体多样性、二维结构和真实手写字形则改善了转移效果；此外，源领域匹配依赖于训练的粒度，在页面级训练下，领域内重建接近真实印刷监督（中位字符错误率为1.82%对比1.31%），但在裁剪级训练下表现不佳（15.59%对比5.52%）。根据这些发现，我们将0.9B参数的PaddleOCR-VL-1.6适配为Wayu-Paxa-OCR-Zero，使用了45,723个合成页面：与其基础检查点相比，它在印刷页面上的中位字符错误率从6.64%降低到1.24%，在手写页面上的错误率从74.87%降低到20.55%，并在所有五个评估集上超越了Typhoon OCR v1 7B，显示出仅使用合成训练也能具备竞争力。

View on arXiv Download PDF AI Translation

cs.CL / 49 / 2609.03597

KhatianDoc: A Human-Verified Benchmark Diagnosing Multimodal LLM Failure on Bengali Legal Land Records

KhatianDoc：一个经过人工验证的基准，诊断孟加拉法律土地记录中多模态大语言模型的失败

Hasan, Tasmiad, Ratul, Arafat Zaman, Saalim, Sarker Sadman, Hossain, S. M. Shah Nawaz, Reza, Khan Raiyan Ibne, Nimi, Sumaiya Tabassum

Abstract

Land ownership in Bangladesh is recorded in Ana-Ganda-Kora-Kranti-Til, a base-16 positional fraction system with dedicated Unicode glyphs, no mainstream font, and no coverage in any OCR pipeline or tokenizer. The handwritten records that carry these fractions, RS Khatians, are the authoritative title record for millions of parcels and a frequent subject of civil litigation, yet no benchmark has asked whether a machine can read one. We introduce KhatianDoc, a four-task benchmark built from 107 real RS Khatian records from the Vumi (land) Office of Munshiganj, Bangladesh: symbol recognition, base-16-to-decimal conversion, structured field extraction, and legal document question answering over 1,634 QA pairs. Ground truth was transcribed by hand, verified by a land-law practitioner to full agreement, and anonymized through positional tokens that keep the referential distinctions multi-hop questions depend on. We evaluate six multimodal LLMs (8B to 72B+, open and closed) under a fixed zero-shot protocol. Five QA categories, 39.3% of our stratified set, return zero correct answers from every model; on the arithmetic task, every model that emits a number does worse than a constant-mean baseline, with exact- and near-match scores coinciding: decorrelation, not approximation. Auditing our own metrics surfaced two artifacts in opposite directions: we correct a refusal-scoring bug and report the fixed scores beside the originals, and flag an inflated metadata metric as an upper bound. KhatianDoc documents not a performance gap but the absence of a capability, with verified ground truth for future systems. Code and data, with a redacted image release, are publicly available.

Chinese Translation

孟加拉国的土地所有权记录采用Ana-Ganda-Kora-Kranti-Til，这是一种带有专用Unicode字形的16进制位置分数系统，没有主流字体，也没有在任何OCR管道或分词器中得到覆盖。承载这些分数的手写记录，即RS Khatians，是数百万地块的权威产权记录，并且经常成为民事诉讼的主题，但没有基准研究询问机器是否能够读取这些记录。我们介绍了KhatianDoc，这是一个由孟加拉国Munshiganj的Vumi（土地）办公室的107份真实RS Khatian记录构建的四项任务基准：符号识别、16进制到十进制转换、结构化字段提取，以及基于1,634对问答的法律文件问答。真实数据由人工转录，并由土地法律从业者进行完全一致的验证，通过位置标记进行匿名处理，以保持多跳问题所依赖的指称区分。我们在固定的零-shot协议下评估了六个多模态大语言模型（8B到72B+，开放和封闭）。五个问答类别中，39.3%的分层集从每个模型返回零个正确答案；在算术任务中，所有发出数字的模型的表现都不如常数均值基线，精确匹配和近似匹配得分一致：是去相关，而非近似。审计我们自己的指标揭示了两个相反方向的伪影：我们修正了一个拒绝评分的错误，并在原始分数旁边报告了修正后的分数，同时将一个膨胀的元数据指标标记为上限。KhatianDoc记录的不是性能差距，而是能力的缺失，并为未来的系统提供了经过验证的真实数据。代码和数据，以及经过编辑的图像发布，均已公开提供。

View on arXiv Download PDF AI Translation

cs.CL / 50 / 2609.03619

Remember and Reweight: Enhancing Multi-Agent Debate with Experience Memory and Confidence Estimation

记忆与重加权：通过经验记忆和信心估计增强多智能体辩论

Jin, Xuanfa, Ma, Zhijian, Zeng, Yongcheng, Cui, Xinyu, Zhang, Haifeng, Wang, Jun

Abstract

Multi-agent debate (MAD) improves the reasoning capabilities of large language models by having multiple agents iteratively refine their responses through discussion. However, MAD suffers from a critical vulnerability known as shared misconception: when a majority of agents initially converge on an incorrect answer, the debate process tends to amplify rather than correct the error. Existing methods primarily address peer skew but leave the agents' inherently biased concept priors unaddressed. To mitigate this systematic weakness, we propose R$^2$-MAD (Remember and Reweight for Multi-Agent Debate), a framework that equips agents with an experience memory accumulated from past debates. R$^2$-MAD intervenes on both failure modes through two complementary mechanisms: A debate-state-aware retrieval policy dynamically calibrates the concept prior by retrieving relevant historical evidence based on the current consensus level. Then these retrieved experiences provide a basis for estimating per-agent reliability, yielding confidence weights to modulate peer influence. Experiments on various benchmarks show that R$^2$-MAD achieves consistent improvements over existing single-agent and MAD baselines.

Chinese Translation

多智能体辩论（MAD）通过让多个智能体通过讨论迭代地完善其回应，从而提升大型语言模型的推理能力。然而，MAD存在一个关键的脆弱性，即共享误解：当大多数智能体最初趋向于一个错误答案时，辩论过程往往会放大而非纠正这一错误。现有方法主要解决同伴偏差问题，但未能解决智能体固有的偏见概念先验。为了缓解这一系统性弱点，我们提出了R$^2$-MAD（记忆与重加权用于多智能体辩论），这是一个为智能体提供从过去辩论中积累的经验记忆的框架。R$^2$-MAD通过两种互补机制干预这两种失败模式：一个关注辩论状态的检索策略根据当前共识水平动态校准概念先验，通过检索相关的历史证据。然后，这些检索到的经验为估计每个智能体的可靠性提供了基础，从而产生信心权重以调节同伴影响。在各种基准测试中的实验表明，R$^2$-MAD在现有的单智能体和MAD基线之上实现了一致的改进。

View on arXiv Download PDF AI Translation

cs.CL / 51 / 2609.03633

Doesn't Stop Reasoning: Analysis of Spurious CoT Termination

思考并未停止：虚假链式思维终止的分析

Koh, Seunghee, Choi, Sungjae, Kwon, Minchan, Baek, Sunghyun, Kim, Junmo

Abstract

Chain-of-thought (CoT) reasoning improves large reasoning models (LRMs) on complex tasks but often produces long, redundant traces. Recent training-free early-exit methods shorten these traces by choosing an intermediate point to stop reasoning. We study one such strategy that injects an end-of-think token (EoT, ) at this point to trigger the reasoning-to-answering transition, and find that the injected EoT does not always induce a clean answering phase. Answering-phase generation can continue before the model regenerates another EoT, with the span preceding this regenerated EoT scaling with the reasoning tokens saved by early exit and exhibiting continued reasoning behavior. We call this spurious CoT termination, where reasoning-like generation continues into the answering phase. We hypothesize that insufficient attention to the injected EoT contributes to spurious CoT termination and probe this hypothesis with Exit-token Attention Biasing (EAB). Across four LRMs, five benchmarks, and two early-exit methods, increasing attention to the injected EoT reduces spurious CoT termination and answering-phase length. These results reveal a limitation of controlling LRMs by externally matching their explicit think-block format. Inserting the EoT token conforms to this format but does not by itself guarantee the intended reasoning-to-answering transition. Our code is available at https://github.com/Seunghee-Koh/Spurious-CoT-Termination.

Chinese Translation

链式思维（Chain-of-thought, CoT）推理在复杂任务中提升了大型推理模型（Large Reasoning Models, LRM）的表现，但常常产生冗长且重复的推理轨迹。最近的无训练早期退出方法通过选择一个中间点来停止推理，从而缩短这些轨迹。我们研究了一种在此点注入思考结束标记（end-of-think token, EoT, ）以触发推理到回答转换的策略，发现注入的 EoT 并不总是能引发干净的回答阶段。在模型重新生成另一个 EoT 之前，回答阶段的生成可以继续，且在此重新生成的 EoT 之前的跨度与通过早期退出节省的推理标记成正比，并表现出持续的推理行为。我们将这种现象称为虚假链式思维终止（spurious CoT termination），即推理式生成继续进入回答阶段。我们假设对注入的 EoT 关注不足会导致虚假链式思维终止，并通过退出标记注意力偏置（Exit-token Attention Biasing, EAB）来验证这一假设。在四个 LRM、五个基准和两种早期退出方法中，增加对注入 EoT 的关注减少了虚假链式思维终止和回答阶段的长度。这些结果揭示了通过外部匹配 LRM 的显式思考块格式来控制 LRM 的局限性。插入 EoT 标记符合这一格式，但并不能单独保证预期的推理到回答的转换。我们的代码可在 https://github.com/Seunghee-Koh/Spurious-CoT-Termination 获取。

View on arXiv Download PDF AI Translation

cs.CL / 52 / 2609.03652

The Impact of Synthetic Data Augmentation on Discourse-Pragmatic Function Classification

合成数据增强对话语语用功能分类的影响

Sorahi, Sara, Tang, Kevin, Kazemian, Reza

Abstract

Synthetic data augmentation has become a common strategy for addressing class imbalance in NLP, but most approaches focus on the quantity and diversity of generated examples rather than their geometric relationship to real training data. We investigate this question in the context of discourse pragmatic function classification, a task where data sparsity is a structural feature rather than a collection artefact. Using 410 manually annotated instances of the English word look drawn from the British National Corpus, spanning four functions: Attention Signal, Directive, Discourse Marker, and Interjection. We generate synthetic training examples with Llama 3.1 and partition them by their cosine distance from real training data in RoBERTa embedding space. We compare six training conditions that differ in the placement of synthetic examples relative to the empirical decision boundary, while holding augmentation quantity constant across conditions. All augmented conditions improve macro F and accuracy over the real only baseline, but core proximal examples (NEAR) yield the largest gains in macro F (0.113), while a distance balanced mix achieves the highest accuracy (0.748). No condition improves AUC, indicating that augmentation shifts the decision boundary rather than improving the model's underlying probability estimates. These findings suggest that where synthetic examples land in representation space matters as much as how many are generated, with implications for low resource pragmatic classification more broadly.

Chinese Translation

合成数据增强已成为解决自然语言处理（NLP）中类别不平衡的常见策略，但大多数方法关注于生成示例的数量和多样性，而非它们与真实训练数据的几何关系。我们在对话语语用功能分类的背景下研究了这个问题，该任务的数据稀疏性是一个结构特征，而非收集工件。我们使用从英国国家语料库中提取的410个手动标注的“look”实例，涵盖四种功能：注意信号（Attention Signal）、指令（Directive）、话语标记（Discourse Marker）和感叹词（Interjection）。我们使用Llama 3.1生成合成训练示例，并根据它们在RoBERTa嵌入空间中与真实训练数据的余弦距离进行划分。我们比较了六种训练条件，这些条件在合成示例相对于经验决策边界的位置上有所不同，同时在各条件间保持增强数量不变。所有增强条件在宏观F值和准确率上均优于仅使用真实数据的基线，但核心近似示例（NEAR）在宏观F值上获得了最大的提升（0.113），而距离平衡混合则达到了最高的准确率（0.748）。没有任何条件改善AUC，这表明增强改变了决策边界，而不是改善模型的基础概率估计。这些发现表明，合成示例在表示空间中的位置与生成的数量同样重要，这对低资源语用分类有更广泛的影响。

View on arXiv Download PDF AI Translation

cs.CL / 53 / 2609.03654

Enhancing Financial Question Answering: A Novel Benchmark Dataset of Banks' financial statements

增强金融问答：银行财务报表的新基准数据集

Miola, Arianna, Spaccavento, Bruno, Silotto, Lorenzo, Bianchetti, Marco, Cagliero, Luca

Abstract

The comparative analysis of banks' financial statements poses significant challenges for automated question answering systems due to their complexity, substantial length, technical language, and inhomogeneity of both textual and numerical content across different jurisdictions and institutions. We introduce FinRAG-QA, a novel benchmark dataset for financial question answering, which comprises 999 practitioner-curated questions on 10 standardised indicators, grounded in 209 annual and Pillar 3 reports from 24 major European and U.S. banks spanning 2019-2023. Unlike prior financial QA benchmarks, which centre on U.S. filings and single-institution analysis, FinRAG-QA targets cross-institutional retrieval over documents averaging 198k words, longer than any existing financial QA resource. On this benchmark we evaluate a multi-stage RAG pipeline and isolate the contribution of each component. Contextual chunk enrichment combined with a retrieval-optimised embedding model raises NDCG@10 from 0.322 to 0.710; conditional on the ground truth being retrieved, a reasoning-optimised generator raises answer accuracy from 44.6% to 79.0% (+34.4 percentage points), at roughly 20x the generation latency. We further show that cross-encoder reranking degrades retrieval when the first-stage ranking is already strong, and that a single top-ranked chunk outperforms larger contexts at generation time. Experiments were run in late 2024-early 2025 with the models available at that time.

Chinese Translation

银行财务报表的比较分析由于其复杂性、较长的篇幅、专业术语以及不同法域和机构间文本与数值内容的不均匀性，给自动问答系统带来了显著挑战。我们提出了FinRAG-QA，这是一个用于金融问答的新基准数据集，包含了针对10个标准化指标的999个实践者策划的问题，基于来自24家主要欧洲和美国银行的209份2019-2023年的年度报告和第三支柱报告。与以往主要集中于美国文件和单一机构分析的金融问答基准不同，FinRAG-QA旨在跨机构检索，文档平均长度达到198,000个单词，超过任何现有的金融问答资源。在此基准上，我们评估了一个多阶段的RAG管道，并隔离了每个组件的贡献。结合上下文块增强和检索优化的嵌入模型，使得NDCG@10从0.322提高到0.710；在检索到真实答案的条件下，优化推理的生成器将答案准确率从44.6%提高到79.0%（增加34.4个百分点），生成延迟大约为20倍。我们进一步展示了当第一阶段排名已经很强时，交叉编码器重排序会降低检索效果，并且在生成时，单个排名最高的块优于更大上下文。实验在2024年底至2025年初进行，使用当时可用的模型。

View on arXiv Download PDF AI Translation

cs.CL / 54 / 2609.03687

A Circuit for Plural Reference: How LLMs Represent and Retrieve Singular and Plural Entities

复数指称的电路：大型语言模型如何表示和检索单数与复数实体

Danh, Anh, Nouwen, Rick, Poesio, Massimo

Abstract

Coreference resolution is an important task in contextual reasoning. In this paper, we investigate the mechanism for representing and retrieving singular and plural entities for plural reference. We use a combination of mechanistic interpretability and attention pattern analysis to study the process in which LLMs predict a pronoun to refer back to previously mentioned entities. Using a range of causal intervention techniques, we find a set of attention heads that are responsible for (1) representing coreference information in the input, (2) identifying entities that form a plural reference, (3) transferring the information to the component that is responsible for selecting the antecedents and predicting the pronoun. We also find that LLMs align with humans in preference for plural pronoun. Specifically, entities in a plural construction are more likely to be referred to as a plural entity if they are ontologically similar and are linked by the conjunction "and".

Chinese Translation

共指解析是上下文推理中的一项重要任务。本文研究了复数指称中单数和复数实体的表示与检索机制。我们结合机械可解释性和注意力模式分析，研究大型语言模型（LLMs）预测代词以指代先前提到的实体的过程。通过一系列因果干预技术，我们发现了一组负责以下功能的注意力头：(1) 在输入中表示共指信息，(2) 识别形成复数指称的实体，(3) 将信息传递给负责选择先行词和预测代词的组件。我们还发现，LLMs在对复数代词的偏好上与人类一致。具体而言，如果在本体上相似并通过连词“和”相连，则复数结构中的实体更可能被指称为复数实体。

View on arXiv Download PDF AI Translation

cs.CL / 55 / 2609.03719

Opening mind by opening architecture: analysis strategies

通过开放架构开启思维：分析策略

Vitucci, Francesco, Silvi, Giuseppe, Annese, Daniele Giuseppe, Scagliola, Francesco, Di Furia, Anthony

Abstract

In numerical signal processing for electroacoustic composition, the progressive loss of specific development and research environments caused by the increasing use of digital market tools has favoured the dominance of the closed-architecture audio processor model. This model, while powerful, envisions the possibility of describing output data about its perceived characteristics, but at the cost of ignoring its internal process and interacting systems, which become complex, powerful environments but closed in an inscrutable black box, a loss we must consider. Any digital signal processing technique tells a story. Just as the words of a language incorporate social, historical and technical polysemic layers, a signal processor has its own story of implementation, a gradual technological achievement with its inevitable aesthetic consequences. Through the looking-glass of literature, one can access those environments with renewed awareness by reestablishing a scientific method and an attitude to research. In this specific case, starting from the case study of Manfred Schroeder's historical reverbs, we illustrate the process of building analytical evaluation tools, as well as practical implementation, at the basis of a conscious study path.

Chinese Translation

在电声作曲的数字信号处理过程中，由于数字市场工具的日益普及，特定发展和研究环境的逐步丧失促成了封闭架构音频处理器模型的主导地位。该模型虽然强大，但设想了描述其感知特征的输出数据的可能性，却以忽视其内部过程和交互系统为代价，这些系统变成了复杂而强大的环境，却封闭在一个不可理解的黑箱中，这是我们必须考虑的损失。任何数字信号处理技术都讲述着一个故事。正如语言的词汇包含社会、历史和技术的多义层次，信号处理器也有其自身的实施故事，这是一个逐步的技术成就，伴随着不可避免的美学后果。通过文学的透镜，可以通过重新建立科学方法和研究态度，以更新的意识接触这些环境。在这个特定的案例中，从曼弗雷德·施罗德（Manfred Schroeder）历史混响的案例研究出发，我们阐述了构建分析评估工具的过程，以及在有意识的学习路径基础上的实际实施。

View on arXiv Download PDF AI Translation

cs.CL / 56 / 2609.03734

Beyond BLEU: A Case for Redefining Sign Language Translation Benchmarks

超越 BLEU：重新定义手语翻译基准的案例

Ranum, Oline, Fish, Edward, Hadfield, Simon, Bowden, Richard

Abstract

BLEU-4 is the standard metric for evaluating sign language translation (SLT), but spoken-language metrics may not adequately reflect sign language proficiency. The multimodal, low-resource context of SLT allows models to exploit spurious correlations and spoken-language priors, rather than learning stronger sign representations. In this paper, we evaluate the relationship between spatio-temporal understanding and BLEU-4 across six SLT models on Phoenix-2014T and CSL-Daily, showing that gains in BLEU-4 are not on their own evidence of better sign language understanding. This work introduces an alternative inspired by language-learning assessment, using an open-weight-LLM QA protocol that measures salient content preservation. It aligns more closely with human rankings and is six to seven times more paraphrase-invariant than BLEU-4. Applied to SLT, this protocol targets content transfer, is more robust to train-test overlap, and gives a different picture of the field: the five gloss-free systems are largely within noise of one another on Phoenix-2014T, while the gloss-supervised system stands 9.3 points higher, a gap invisible to BLEU-4.

Chinese Translation

BLEU-4 是评估手语翻译（SLT）的标准指标，但口语语言的指标可能无法充分反映手语的熟练程度。手语翻译的多模态、低资源背景使得模型能够利用虚假的相关性和口语语言的先验，而不是学习更强的手语表示。在本文中，我们评估了六个 SLT 模型在 Phoenix-2014T 和 CSL-Daily 数据集上，时空理解与 BLEU-4 之间的关系，显示出 BLEU-4 的提升并不能单独证明更好的手语理解。该研究引入了一种受语言学习评估启发的替代方案，使用开放权重的 LLM QA 协议来衡量显著内容的保留。该方法与人类排名更为一致，并且在改写不变性方面比 BLEU-4 高出六到七倍。应用于手语翻译时，该协议针对内容传递，对训练-测试重叠更具鲁棒性，并提供了该领域的不同视角：在 Phoenix-2014T 上，五个无注释系统的表现大致相当，而有注释监督系统的得分高出 9.3 分，这一差距在 BLEU-4 中并不可见。

View on arXiv Download PDF AI Translation

cs.CL / 57 / 2609.03775

Typological Feature Prediction with Large Language Models: An In-Context Learning Approach

利用大型语言模型进行类型特征预测：一种上下文学习方法

Wang, Qianwen, Ng, York Hay, Khan, Aditya, Lee, En-Shiun Annie

Abstract

Typological features are widely used in multilingual NLP, and the prediction of such features holds downstream utility. However, existing methods to predict missing values lack interpretable justifications for predictions, while their performance across resource levels and feature types remains underexplored. Given LLMs' abilities in meta-linguistic reasoning and in providing rationales, we investigate LLMs' performance in typological feature prediction via an in-context learning approach with linguistic data from URIEL+ and Glottolog. We find that zero-shot prompting is insufficient, but when given phylogenetic and geographic neighbour evidence, LLMs substantially outperform all baselines without disadvantaging low-resource languages. We further find that most LLM rationales are consistent with the provided evidence, offering a step toward explainable typological feature prediction.

Chinese Translation

类型特征在多语言自然语言处理（NLP）中被广泛使用，而对这些特征的预测具有下游应用价值。然而，现有的缺失值预测方法缺乏可解释的预测依据，同时在资源水平和特征类型上的表现仍未得到充分探索。鉴于大型语言模型（LLMs）在元语言推理和提供理由方面的能力，我们通过使用来自URIEL+和Glottolog的语言数据，研究了LLMs在类型特征预测中的表现，采用上下文学习的方法。我们发现，零-shot 提示不足，但在提供系统发育和地理邻近证据时，LLMs的表现显著优于所有基线，并且没有对低资源语言造成不利影响。我们进一步发现，大多数LLM的推理与提供的证据一致，为可解释的类型特征预测迈出了重要一步。

View on arXiv Download PDF AI Translation

cs.CL / 58 / 2609.03781

IndicSafeEval: Safety Robustness of Large Language Models under Multilingual Persuasive Jailbreak Attacks

IndicSafeEval：大型语言模型在多语言说服性越狱攻击下的安全性鲁棒性

Mondal, Saikat, Mamta, Varshney, Deeksha, Cocarascu, Oana, Ekbal, Asif

Abstract

Large language models (LLMs) are increasingly used in multilingual settings, yet their safety is still evaluated primarily in English. This limits our understanding of how alignment failures manifest in low-resource and culturally diverse languages. We introduce IndicSafeEval, a persuasion-based jailbreak evaluation framework for Indian languages. Our benchmark combines ten safety critical content categories with six human-like persuasive strategies across four different Indian languages, such as Hindi, Bengali, Marathi and Punjabi, resulting in 7,200 adversarial prompts. We conduct a systematic black-box evaluation of several open-source LLMs to examine how their safety behaviour varies across languages, persuasion strategies, and risk categories. Our analysis shows that the model does not behave equally safely across all languages and prompt styles. Instead, safety performance depends strongly on both the languages used and the way a request is phrased using persuasive cues. We further observe that different risk categories exhibit different levels of vulnerability, with some types of harmful content being significantly more susceptible to persuasion-based jailbreaks than others. These findings reveal important limitations of current safety evaluations, which are largely English-centric, and underscore the need for multilingual and persuasion-aware benchmarking frameworks to more accurately assess real-world LLM safety. Our implementation is available at https://github.com/MonSaikat/IndicSafeEval. Warning: this paper contains example data that may be offensive or harmful.

Chinese Translation

大型语言模型（LLMs）在多语言环境中的使用日益增加，但其安全性评估仍主要集中在英语。这限制了我们对低资源和文化多样性语言中对齐失败表现的理解。我们提出了IndicSafeEval，这是一个针对印度语言的基于说服的越狱评估框架。我们的基准结合了十个安全关键内容类别和六种类人类的说服策略，涵盖四种不同的印度语言，如印地语、孟加拉语、马拉地语和旁遮普语，共生成7200个对抗性提示。我们对几种开源LLMs进行了系统的黑箱评估，以检查它们的安全行为在不同语言、说服策略和风险类别之间的变化。我们的分析显示，模型在所有语言和提示风格中的安全性表现并不相同。相反，安全性能在很大程度上依赖于所使用的语言和请求的措辞方式，尤其是使用说服性提示的方式。我们进一步观察到，不同的风险类别表现出不同的脆弱性水平，某些类型的有害内容对基于说服的越狱攻击显著更敏感。这些发现揭示了当前安全评估的重要局限性，主要集中在英语，并强调了需要多语言和关注说服的基准框架，以更准确地评估现实世界中LLM的安全性。我们的实现可在https://github.com/MonSaikat/IndicSafeEval获取。警告：本文包含可能令人反感或有害的示例数据。

View on arXiv Download PDF AI Translation

cs.CL / 59 / 2609.03814

Evaluating Criterion-Conditioned Behaviour of Large Language Models in Content Moderation

评估大型语言模型在内容审核中的标准条件行为

Zhang, Danting, Peng, Bei, Loftin, Robert

Abstract

Large language models (LLMs) demonstrate strong performance on standard content moderation benchmarks. However, these benchmarks often aggregate multiple moderation criteria into a single label, making it unclear whether models can disentangle them and reliably apply each criterion when making decisions. To study whether LLMs exhibit criterion-conditioned behaviour, we introduce Diagnostic Evaluation of COntent (DECO), a criterion-independent factorisation of content that enables controlled, criterion-level evaluation. We also introduce pairwise evaluation to compare model outputs across different criteria for the same input. Across four moderation datasets and four LLMs, we find that strong benchmark performance can hide substantial failures at the criterion level. Models struggle most when correct decisions depend not on overall harmfulness, but on the specific aspect of the content that the criterion requires them to assess. Our results highlight a key limitation of current content moderation benchmarks: strong performance on aggregated labels does not provide sufficient evidence that LLMs can reliably evaluate content with respect to individual moderation criteria. These findings call for the development of evaluation methods that explicitly measure criterion-conditioned behaviour.

Chinese Translation

大型语言模型（LLMs）在标准内容审核基准测试中表现出色。然而，这些基准测试通常将多个审核标准聚合为一个单一标签，这使得不清楚模型是否能够将其解耦，并在做出决策时可靠地应用每个标准。为了研究LLMs是否表现出标准条件行为，我们引入了内容诊断评估（Diagnostic Evaluation of COntent, DECO），这是一种标准独立的内容分解方法，能够进行受控的标准级评估。我们还引入了成对评估，以比较相同输入在不同标准下的模型输出。在四个审核数据集和四个LLMs的实验中，我们发现强大的基准性能可能掩盖了在标准层面上的重大失败。当正确决策依赖于内容的特定方面而非整体有害性时，模型表现最差。我们的结果突显了当前内容审核基准的一个关键局限性：在聚合标签上的强表现并不能提供足够的证据表明LLMs能够可靠地根据个别审核标准评估内容。这些发现呼吁开发明确测量标准条件行为的评估方法。

View on arXiv Download PDF AI Translation

cs.CL / 60 / 2609.03887

Beyond Shallow Alignment: How Post-Training Methods Determine Refusal Circuits And Steering Robustness

超越浅层对齐：后训练方法如何决定拒绝电路和引导鲁棒性

Nguyen, Hoang Cuong, Dras, Mark, Naseem, Usman

Abstract

How do the methods used to train language models to refuse harmful requests shape how that refusal actually works inside the model? We compare three post-training methods - supervised fine-tuning, reasoning-augmented fine-tuning (training on reasoning chains that justify a safety decision), and preference optimization (ORPO) - across three architecturally distinct models (Llama-3.1-8B, Gemma-2-9B, Qwen3-8B). We find that training method, not just data, reshapes how refusal is computed internally: reasoning-augmented training consistently produces a distinct kind of refusal computation, visible across all three models, while architecture independently shapes internal structure and how reliably refusal can be steered. Most importantly, no method we study achieves all three properties we would want from safe alignment at once: refusal that isn't concentrated in a few fragile components, safety gains that don't cost general capability, and safety behavior correctable through small, targeted edits. We caution against treating current post-training methods as a solved, reliable defense, especially for security-critical use. Code and models are available in https://github.com/hoangcuongnguyen2001/Beyond-Shallow-Alignment.

Chinese Translation

用于训练语言模型以拒绝有害请求的方法如何塑造这种拒绝在模型内部的实际运作？我们比较了三种后训练方法——监督微调、增强推理微调（在推理链上进行训练，以证明安全决策的合理性）和偏好优化（ORPO）——在三种架构上明显不同的模型（Llama-3.1-8B、Gemma-2-9B、Qwen3-8B）之间的表现。我们发现，训练方法，而不仅仅是数据，重塑了内部拒绝的计算方式：增强推理训练始终产生一种独特的拒绝计算方式，这在所有三种模型中均可见，而架构则独立地塑造了内部结构以及拒绝的引导可靠性。最重要的是，我们研究的任何方法都未能同时实现我们希望从安全对齐中获得的三种特性：拒绝不集中于少数脆弱组件、安全收益不以牺牲一般能力为代价，以及可通过小规模、针对性的编辑进行修正的安全行为。我们警告不要将当前的后训练方法视为已解决的、可靠的防御，特别是在安全关键的应用中。代码和模型可在 https://github.com/hoangcuongnguyen2001/Beyond-Shallow-Alignment 获取。

View on arXiv Download PDF AI Translation

cs.CL / 61 / 2609.03894

CROCODIL: Cross-Model Code Editing with LLMs

CROCODIL：基于大型语言模型的跨模型代码编辑

Zhong, Linghan, Thimmaiah, Aditya, Srinivasa, Jayanth, Gligoric, Milos, Li, Junyi Jessy

Abstract

Large language models (LLMs) have become ubiquitous tools for code generation and editing. However, development teams often use multiple LLM assistants. Different developers may prefer different models, and individual developers may switch between models across different coding sessions. Because of this, the edits any one model makes are frequently applied to foreign code originally generated by another model. These LLMs are often trained on different datasets, and as a result have different stylistic preferences. Do LLMs behave differently when they edit foreign code originally written by a different LLM with a different coding style? We find that models tend to make more, and often excessive, edits on foreign code. We introduce CROCODIL (Cross-model Code Editing with LLMs), a post-training framework for reducing excessive edits while preserving functional correctness. CROCODIL's similarity reward penalizes large changes, while its execution reward scores build and test success. We use the product of these two rewards to encourage the policy to decrease the edit size without decreasing the edit task success rate. CROCODIL is available at https://github.com/EngineeringSoftware/Crocodil.

Chinese Translation

大型语言模型（LLMs）已成为代码生成和编辑的普遍工具。然而，开发团队通常使用多个 LLM 助手。不同的开发者可能偏好不同的模型，个别开发者在不同的编码会话中可能会在模型之间切换。因此，任何一个模型所做的编辑经常应用于最初由另一个模型生成的外部代码。这些 LLM 通常是在不同的数据集上训练的，因此具有不同的风格偏好。当 LLM 编辑最初由不同 LLM 以不同编码风格编写的外部代码时，它们的行为是否会有所不同？我们的研究发现，模型往往会对外部代码进行更多且常常过度的编辑。我们提出了 CROCODIL（基于大型语言模型的跨模型代码编辑），这是一个后训练框架，旨在减少过度编辑，同时保持功能正确性。CROCODIL 的相似性奖励惩罚大幅修改，而其执行奖励则根据构建和测试的成功进行评分。我们使用这两种奖励的乘积来鼓励策略在不降低编辑任务成功率的情况下减少编辑规模。CROCODIL 可在 https://github.com/EngineeringSoftware/Crocodil 获取。

View on arXiv Download PDF AI Translation

cs.CL / 62 / 2609.03915

RuleMem: Active Rule Memory for Long-Term Conversational Agents

RuleMem：用于长期对话体的主动规则记忆

Zeng, Xingyuan, Wu, Zuohan, Yao, Quanming, Wang, Yue, Liu, Wei, Zheng, Libin, Wang, Jiuke, Yin, Jian

Abstract

Question answering agents in long-term conversations must reason over massive, temporally dispersed dialogue histories. However, existing memory mechanisms primarily treat past information as \textit{passively} stored facts, leading to semantic gaps and unreliable reasoning. To address this limitation, we propose RuleMem, a rule-based memory framework that induces reusable logical rules from historical interactions to \textit{actively} guide both evidence retrieval and reasoning. Specifically, RuleMem constructs natural-language Horn clauses from conversations and validates them via a Rule Perplexity Consistency (RPC) mechanism. These induced rules enable the retrieval of semantically distant evidence while providing an explicit logical structure for answer generation. We conducted a comprehensive evaluation of RuleMem on two long-term conversational benchmarks, LoCoMo and LongMemEval_s*. In a rigorous comparison against 14 baselines on LoCoMo, RuleMem achieved the highest accuracy, exceeding the baseline average by 27.47 points (a 54.3% relative improvement).

Chinese Translation

在长期对话中，问答代理必须对大量时间上分散的对话历史进行推理。然而，现有的记忆机制主要将过去的信息视为 extit{被动}存储的事实，这导致了语义差距和不可靠的推理。为了解决这一局限性，我们提出了RuleMem，一种基于规则的记忆框架，它从历史交互中引导可重用的逻辑规则，以 extit{主动}指导证据检索和推理。具体而言，RuleMem从对话中构建自然语言的霍恩子句，并通过规则困惑度一致性（Rule Perplexity Consistency, RPC）机制对其进行验证。这些引导的规则使得能够检索语义上遥远的证据，同时为答案生成提供了明确的逻辑结构。我们在两个长期对话基准LoCoMo和LongMemEval_s*上对RuleMem进行了全面评估。在与14个基线的严格比较中，RuleMem在LoCoMo上达到了最高准确率，超过基线平均值27.47分（相对提高54.3%）。

View on arXiv Download PDF AI Translation

cs.CL / 63 / 2609.03930

Fixed Suffix Dependency Ratio: Quantifying the Dual-Track Mechanism of Gender Assignment in Latvian Loanwords

固定后缀依赖比率：量化拉脱维亚借词中性别分配的双轨机制

Zhang, Yelingyun, Kapenieks, Atis, Platonova, Marina

Abstract

Existing research has repeatedly observed the tendency for English loanwords to cluster in the masculine gender across different recipient languages, yet the origin of this pattern remains difficult to determine, as fixed morphological rules and default assignments are frequently analysed together. This study proposes the Fixed Suffix Dependency Ratio (FSDR) to quantify the degree of reliance on fixed derivational suffixes across different genders, and to distinguish between morphological anchoring and free-choice in distribution. By examining 1,832 Latvian noun lemma types, the results reveal a significant FSDR asymmetry within the loanword system: feminine loanwords rely significantly more on fixed derivational suffixes, while masculine loanwords are more concentrated in the free-choice zone. This pattern exhibits loanword specificity and has become more pronounced in contemporary usage. FSDR therefore provides a quantitative framework for testing default gender and shows how masculine default can be activated and reinforced under language contact.

Chinese Translation

现有研究反复观察到英语借词在不同接受语言中倾向于聚集于阳性别，但这一模式的起源仍然难以确定，因为固定的形态规则和默认分配常常被一起分析。本研究提出了固定后缀依赖比率（Fixed Suffix Dependency Ratio, FSDR），以量化不同性别之间对固定派生后缀的依赖程度，并区分形态锚定与分布中的自由选择。通过对1,832个拉脱维亚名词词条类型的分析，结果揭示了借词系统中显著的FSDR不对称性：阴性借词对固定派生后缀的依赖显著高于阳性借词，而阳性借词则更集中于自由选择区域。这一模式表现出借词的特异性，并在当代用法中变得更加明显。因此，FSDR提供了一个定量框架，用于检验默认性别，并展示了在语言接触下阳性默认如何被激活和强化。

View on arXiv Download PDF AI Translation

cs.CL / 64 / 2609.03953

Beyond Majority Vote: Multi-Perspective Adjudication for Medical Hallucination Detection

超越多数投票：医学幻觉检测的多视角裁决

Cecil, Joe, Freedman, Marjorie

Abstract

Understanding the frequency of factual errors in chatbot-generated text and evaluating systems that detect these errors is critical for determining chatbot safety. Yet factual-error detection is often treated as a single-pass, single-annotator labeling problem. In long-form chatbot responses, factual errors can be subtle and embedded within mostly correct text. We develop a multi-perspective annotation study of medically relevant chatbot responses, combining first-pass annotation, LLM-as-a-Judge (LaJ) candidate discovery, and two forms of adjudication: medical-expert and evidence-based fact-checking. First-pass annotators frequently miss factual errors later validated by adjudicators. LaJ improves candidate discovery, but is insufficient on its own: It misses factual errors that annotators catch. We also find disagreement among adjudicators, suggesting that adjudication over multiple candidate sources can improve benchmark completeness, but does not eliminate the need to apply judgment and expertise. Applied to an existing benchmark, this technique reveals a similar pattern of missing annotations. Together, these results suggest that in the settings examined here, single-pass hallucination benchmarks may achieve scale at the cost of undercounting factual errors. Multi-pass adjudication can improve coverage, but inferences drawn from the benchmarks are still sensitive to the judgment, expertise, and evidence used to determine error presence.

Chinese Translation

理解聊天机器人生成文本中的事实错误频率并评估检测这些错误的系统，对于确定聊天机器人的安全性至关重要。然而，事实错误检测通常被视为单次、单一注释者的标注问题。在长篇聊天机器人回应中，事实错误可能是微妙的，并嵌入在大部分正确的文本中。我们开展了一项针对医学相关聊天机器人回应的多视角注释研究，结合了第一次标注、LLM作为裁判（LaJ）候选发现以及两种裁决形式：医学专家和基于证据的事实核查。第一次标注者经常会遗漏后来由裁决者验证的事实错误。LaJ提高了候选发现的效果，但单独使用时仍然不足：它遗漏了标注者捕捉到的事实错误。我们还发现裁决者之间存在分歧，这表明对多个候选来源的裁决可以提高基准的完整性，但并不能消除应用判断和专业知识的必要性。应用于现有基准时，这一技术揭示了遗漏标注的类似模式。综合来看，这些结果表明，在此处考察的环境中，单次幻觉基准可能以低估事实错误为代价实现规模化。多次裁决可以改善覆盖率，但从基准得出的推论仍然对判断、专业知识和用于确定错误存在的证据敏感。

View on arXiv Download PDF AI Translation

cs.CL / 65 / 2609.03955

Two-Stage Reinforcement Learning for Sound and Adversarial Test Generation in Code LLMs

基于两阶段强化学习的代码大语言模型中的有效性和对抗性测试生成

Xu, Jiacheng, Zhang, Wentao, Lyu, Zhiyi, Zhang, Fuxiang, Wang, Chaojie, Liu, Yang, An, Bo

Abstract

Reinforcement learning (RL) has substantially advanced code generation with large language models (LLMs) through executable feedback. The feedback for coding problems mainly comes from specific test cases, where high-quality test cases are often scarce since they should be both sound and discriminative. We thus turn to study the auto-generation of test cases using the learned model. We find this is naturally an adversarial RL problem: the model is expected to generate effective test cases as counterexamples, depending on the solver's current failure modes. We propose Test Cases Scaling (TCS), a two-stage RL framework for effective test generation. Both stages train a test generator from a rolling policy-aligned buffer: Stage 1 generates tests consistent with the reference solution, and Stage 2 restricts the buffer to current failure modes and learns counterexample tests. Across TACO and LiveCodeBench, TCS improves both pass@1 and inference-time answer selection according to generated tests. We find the learned test generator also enables effective selection among other LLM outputs.

Chinese Translation

强化学习（RL）通过可执行反馈在代码生成方面取得了显著进展，尤其是在大型语言模型（LLMs）中。编码问题的反馈主要来自特定的测试用例，而高质量的测试用例往往稀缺，因为它们需要同时具备有效性和区分性。因此，我们转向研究使用学习模型自动生成测试用例。我们发现这自然构成了一个对抗性强化学习问题：模型需要根据求解器当前的失败模式生成有效的测试用例作为反例。我们提出了测试用例缩放（Test Cases Scaling, TCS），这是一个用于有效测试生成的两阶段强化学习框架。两个阶段均从滚动策略对齐的缓冲区中训练测试生成器：第一阶段生成与参考解决方案一致的测试，第二阶段将缓冲区限制为当前失败模式，并学习反例测试。在 TACO 和 LiveCodeBench 上，TCS 根据生成的测试提高了 pass@1 和推理时间答案选择的效果。我们发现，学习到的测试生成器还能够在其他 LLM 输出中实现有效选择。

View on arXiv Download PDF AI Translation

cs.CL / 66 / 2609.03967

Investigating the Ability of Large Language Models to Analyze Recipes for Diabetes

研究大型语言模型分析糖尿病食谱的能力

Venkataramanan, Revathy, Luthra, Aditya, Nadimuthu, Venkatesan, Sheth, Amit

Abstract

Several studies have evaluated the ability of Large Language Models (LLMs) for meal planning, yielding positive outcomes. These models can process natural language inputs and leverage learned knowledge from their pretraining to generate meal plans. In this work, we investigate the ability of LLMs to analyze the suitability of given recipes for diabetes. The primary challenge for LLMs is to retrieve relevant dietary guidelines for diabetes, decompose recipes into ingredients and cooking methods, and apply these guidelines to determine the recipe's suitability. To study these challenges, we employ three kinds of prompts namely, (i) Direct Query Prompt (ii) Context-Guided Prompt, and (iii) Exemplary Context Prompt that incorporate different levels of diabetes dietary guidelines from medical sources. We introduce a benchmark dataset curated for this investigation consisting of 7607 recipes that include 3807 recipes suitable for diabetes and 3800 recipes not suitable for diabetes. Our results demonstrate that most LLMs are cautious in predicting recipes as suitable to prevent detrimental outcomes. Further, the models that can reason using the dietary guidelines performed better in predicting the suitability of recipes for diabetes. Overall, Mistral-7B and Llama 70B showed superior performance to their counterparts.

Chinese Translation

多项研究评估了大型语言模型（LLMs）在餐饮规划方面的能力，结果积极。这些模型能够处理自然语言输入，并利用其预训练中学习到的知识生成餐饮计划。在本研究中，我们探讨了LLMs分析给定食谱对糖尿病适宜性的能力。LLMs面临的主要挑战是检索与糖尿病相关的饮食指南，将食谱分解为成分和烹饪方法，并应用这些指南来确定食谱的适宜性。为研究这些挑战，我们采用了三种不同类型的提示，即（i）直接查询提示（Direct Query Prompt），（ii）上下文引导提示（Context-Guided Prompt），以及（iii）示例上下文提示（Exemplary Context Prompt），这些提示结合了来自医学来源的不同层次的糖尿病饮食指南。我们引入了一个为本研究精心策划的基准数据集，包含7607个食谱，其中包括3807个适合糖尿病的食谱和3800个不适合糖尿病的食谱。我们的结果表明，大多数LLMs在预测食谱适宜性时持谨慎态度，以防止不良后果。此外，能够运用饮食指南进行推理的模型在预测食谱对糖尿病的适宜性方面表现更佳。总体而言，Mistral-7B和Llama 70B的表现优于其他模型。

View on arXiv Download PDF AI Translation

cs.CL / 67 / 2609.03992

Alignment-Free Text-Audiobox for Voice Dubbing and Full-Duplex Dialogue Synthesis

无对齐文本音频盒用于语音配音和全双工对话合成

Chen, Sanyuan, Hwang, Min-Jae, Inoue, Sho, Sun, Anna, Yu, Bokai, Kant, David, Hyun, Dongmin, Desblancs, Dorian, Antonovsky, Gregory, Repin, Oleg, Chen, Peng-Jen, Ma, Xutai, Tu, Zehai, Pino, Juan, Hsu, Wei-Ning

Abstract

We present Alignment-Free Text-Audiobox (Text-AB), a unified framework for high-quality voice dubbing and full-duplex dialogue synthesis. Building on a Diffusion Transformer trained with a flow-matching objective, Text-AB departs from the Audiobox system along three dimensions. First, it operates in a latent diffusion framework using DAC-VAE features that encode 48 kHz waveforms into a 25 Hz latent sequence, giving over 10x higher compression than previous EnCodec representations while improving resynthesis quality. Second, Text-AB is alignment-free: it consumes raw text via an off-the-shelf text encoder and learns text-speech alignment through cross-attention, removing the need for forced alignment and explicit duration prediction. Third, we scale model and data substantially, pretraining a 3B-parameter model on 480k hours of monolingual speech, followed by supervised fine-tuning on three downstream tasks: cross-lingual voice dubbing, full-duplex dialogue synthesis, and emotional full-duplex dialogue synthesis. At inference, Text-AB supports one-shot generation for up to ~1 min of speech and arbitrarily long-form generation via a multi-diffusion scheme, plus a multi-stage reranking strategy that enhances quality based on automated metrics. On a real-world dubbing benchmark, Text-AB delivers a step-change improvement over the latest internal dubbing system, with large gains in prosody similarity, voice similarity, naturalness, and shareability. For full-duplex dialogue synthesis, it approaches human recordings on short-form conversations and substantially outperforms the latest internal model on long-form human-likeness and expressivity, while natively modeling turn-taking, back-channeling, and emotional dynamics. For emotional dialogue synthesis, emotion conditioning significantly improves emotion alignment and emotional interaction quality over the unconditioned baseline.

Chinese Translation

我们提出了无对齐文本音频盒（Text-AB），这是一个用于高质量语音配音和全双工对话合成的统一框架。Text-AB基于一个使用流匹配目标训练的扩散变换器，沿三个维度脱离了音频盒系统。首先，它在潜在扩散框架中运行，使用DAC-VAE特征将48 kHz波形编码为25 Hz的潜在序列，相较于之前的EnCodec表示，提供了超过10倍的压缩，同时提高了再合成质量。其次，Text-AB是无对齐的：它通过现成的文本编码器处理原始文本，并通过交叉注意力学习文本与语音的对齐，消除了强制对齐和显式持续时间预测的需求。第三，我们大幅扩展了模型和数据，在480k小时的单语语音上预训练了一个3B参数的模型，随后在三个下游任务上进行监督微调：跨语言语音配音、全双工对话合成和情感全双工对话合成。在推理时，Text-AB支持一次生成长达约1分钟的语音，并通过多扩散方案支持任意长的生成，此外还有一个多阶段重排序策略，根据自动化指标提升质量。在一个真实世界的配音基准测试中，Text-AB相较于最新的内部配音系统实现了显著的改进，在韵律相似性、声音相似性、自然性和可分享性方面均有显著提升。在全双工对话合成方面，它在短篇对话中接近人类录音，并在长篇人类相似性和表现力上大幅超越最新的内部模型，同时原生建模轮流发言、反馈通道和情感动态。在情感对话合成中，情感条件显著改善了情感对齐和情感互动质量，相较于无条件基线有显著提升。

View on arXiv Download PDF AI Translation

cs.CL / 68 / 2609.04022

Representational alignment yields generalizable safety in language models

表征对齐实现语言模型的可泛化安全性

Li, Lingyu, Teng, Yan, Wang, Yingchun, Hu, Xia

Abstract

Aligning large language models (LLMs) is essential for their safe deployment. Current alignment methods mainly optimize observable responses, yet models remain vulnerable when the same harmful intent is recast in unfamiliar or adversarial forms that humans can easily recognize. Prototype theory offers an account of this adaptability. Human concepts are represented around central cases, and new instances are categorized according to their graded typicality relative to these prototypes. Here we show that such categorization of moral concepts is weakly preserved in current LLMs. Across 23 LLMs, models often failed to distinguish opposed moral categories or preserve fine-grained typicality within each category. These deficits persist across parameter sizes and alignment stages. We developed representational similarity optimization, which directly aligns the latent representations in LLMs with the categorization expressed in human moral judgements, without supervising generated responses. In matched experiments using the same 251,334 moral annotations, standard behavioral alignment learned the intended moral judgements at the response level while leaving the categorization structure largely unchanged and increasing vulnerability across adversarial evaluations. Reorganizing moral categorization produced more modest gains in explicit judgements but consistently improved adversarial robustness across model scales on diverse benchmarks and attack strategies. Our findings provide functional support for the view that prototype-based categorization contributes to behavioral adaptability. They also show that transferring this representational principle to LLMs yields generalizable safety under adversarial conditions.

Chinese Translation

对大型语言模型（LLMs）进行对齐对于其安全部署至关重要。目前的对齐方法主要优化可观察的响应，但当相同的有害意图以人类容易识别的不熟悉或对抗形式重新表达时，模型仍然脆弱。原型理论提供了这种适应性的解释。人类概念围绕中心案例进行表征，新实例根据其相对于这些原型的典型性进行分类。在此，我们展示了当前LLMs中道德概念的这种分类是弱保留的。在23个LLMs中，模型通常未能区分对立的道德类别或在每个类别内保持细粒度的典型性。这些缺陷在不同参数规模和对齐阶段中持续存在。我们开发了表征相似性优化，直接将LLMs中的潜在表征与人类道德判断中表达的分类进行对齐，而无需监督生成的响应。在使用相同的251,334个道德注释的匹配实验中，标准行为对齐在响应层面学习了预期的道德判断，同时在很大程度上保持了分类结构不变，并在对抗评估中增加了脆弱性。重新组织道德分类在显性判断中产生了更温和的收益，但在不同基准和攻击策略上的模型规模中始终提高了对抗鲁棒性。我们的发现为原型基础的分类有助于行为适应性这一观点提供了功能支持。它们还表明，将这一表征原则转移到LLMs中可以在对抗条件下实现可泛化的安全性。

View on arXiv Download PDF AI Translation

cs.CL / 69 / 2609.04048

Translation as a Decision Space: A Multi-Agent Perspective on Low-Resource Dialect Generation

作为决策空间的翻译：低资源方言生成的多智能体视角

Alkhder, Hasan, Abboush, Mohammad, Tchappi, Igor, Zengin, Ahmet, Najjar, Amro

Abstract

Neural machine translation (NMT) systems typically produce a single output per input, obscuring the alternative decision trajectories implicitly available within multilingual decoding. This opacity becomes particularly problematic in low-resource dialect settings, where multiple linguistically valid realizations may differ in lexical authenticity, register, and structural stability. We propose reframing translation as a structured decision space explored by autonomous translation agents. Instead of analyzing a single output, we model distinct translation pathways as agents operating over a shared multilingual backbone. Inter-agent divergence is treated not as error but as an interpretable behavioral signal. We conduct an empirical study on Turkish--Syrian Arabic translation using three agents: (1) zero-shot direct translation, (2) dialect-stabilized translation via lightweight fine-tuning, and (3) pivot translation through English. Evaluation is performed on 5,000 dialogue sentences, while stabilization is trained on 5,000 additional Turkish--Syrian sentence pairs drawn from television dialogue and MADAR-Turk resources. Rather than optimizing for conventional performance metrics, we quantify structured behavioral displacement using dialect marker frequency, lexical proximity to standardized Arabic, and structural variance. Lightweight stabilization nearly doubles dialect marker usage, increasing it from 0.2266 to 0.4988, while significantly reducing structural instability. Pivot mediation introduces normalization pressure and measurable compression effects, whereas zero-shot translation exhibits the highest decision variance. We argue that translation divergence across agents reveals latent decision flexibility within multilingual models and we provide a principled interpretability framework for low-resource dialect generation.

Chinese Translation

神经机器翻译（NMT）系统通常为每个输入生成单一输出，这掩盖了多语言解码中隐含的替代决策轨迹。在低资源方言环境中，这种不透明性尤为突出，因为多种语言上有效的实现可能在词汇真实性、语域和结构稳定性上存在差异。我们建议将翻译重新构建为一个由自主翻译智能体探索的结构化决策空间。我们不再分析单一输出，而是将不同的翻译路径建模为在共享多语言基础上运作的智能体。智能体之间的差异不被视为错误，而是可解释的行为信号。我们对土耳其语-叙利亚阿拉伯语翻译进行了实证研究，使用了三个智能体：（1）零样本直接翻译，（2）通过轻量级微调实现的方言稳定翻译，以及（3）通过英语进行的中介翻译。评估在5000个对话句子上进行，而稳定性训练则基于从电视对话和MADAR-Turk资源中提取的5000对额外的土耳其语-叙利亚句子。我们并不优化传统的性能指标，而是通过方言标记频率、与标准阿拉伯语的词汇接近度和结构方差来量化结构化的行为位移。轻量级稳定化几乎使方言标记的使用量翻倍，从0.2266增加到0.4988，同时显著降低了结构不稳定性。中介翻译引入了规范化压力和可测量的压缩效果，而零样本翻译则表现出最高的决策方差。我们认为，智能体之间的翻译差异揭示了多语言模型中的潜在决策灵活性，并为低资源方言生成提供了一个原则性的可解释性框架。

View on arXiv Download PDF AI Translation

cs.CL / 70 / 2609.04108

Sequential Beats Joint: On the Interplay between On-Policy Distillation and RLVR

顺序优先于联合：关于策略蒸馏与可验证奖励强化学习之间的相互作用

Li, Boyan, Chen, Bingsen, Yang, Chenghao, Nie, Ping, Zhao, Chen, Ye, Xi

Abstract

Reinforcement learning with verifiable rewards (RLVR) and on-policy distillation (OPD) have emerged as two dominant methods for post-training reasoning LLMs. Prior work uses OPD's dense token-level supervision to complement the sparse RL reward, fusing the two signals within a single step: either as a \emph{weighted-additive combination} or a \emph{teacher-modulated rescaling} of the RL advantage. In this paper, we show that a simple two-stage scheme, OPD-then-RL, consistently outperforms pure OPD, pure RLVR, and all such joint baselines across logic and math reasoning benchmarks. Beyond the empirical results, we further provide a systematic understanding of this through pass@$k$ behavior, learning dynamics, and parameter updates, yielding a consistent explanation: OPD expands the student's coverage of teacher-supported solutions and RL sharpens within that support, while jointly optimizing the two signals causes them to interfere.To provide a practical recipe, we find that the OPD validation score is the key signal for when to switch to RL, and that OPD is a better cold start for RL than SFT. Together, our results establish OPD-then-RL as a simple yet strong way to combine the two methods, turning two entangled signals into complementary stages.

Chinese Translation

可验证奖励强化学习（RLVR）和策略蒸馏（OPD）已成为后训练推理大型语言模型（LLMs）的两种主导方法。先前的研究利用OPD的密集令牌级监督来补充稀疏的RL奖励，在单个步骤中融合这两种信号：要么作为 extit{加权加和组合}，要么作为RL优势的 extit{教师调制重缩放}。在本文中，我们展示了一种简单的两阶段方案，即OPD-然后-RL，始终优于纯OPD、纯RLVR以及所有此类联合基线，在逻辑和数学推理基准测试中表现出色。除了实证结果外，我们进一步通过pass@$k$行为、学习动态和参数更新提供了系统性的理解，得出了一个一致的解释：OPD扩展了学生对教师支持解决方案的覆盖范围，而RL在该支持范围内进行精细化优化，同时联合优化这两种信号会导致它们相互干扰。为了提供一个实用的方案，我们发现OPD验证分数是何时切换到RL的关键信号，并且OPD在冷启动RL方面优于SFT。总之，我们的结果确立了OPD-然后-RL作为一种简单而有效的方式来结合这两种方法，将两个交织的信号转化为互补的阶段。

View on arXiv Download PDF AI Translation

cs.CL / 71 / 2609.04173

Last Translation Benchmark

最后翻译基准

Zouhar, Vilém, Bafna, Niyati, Choudhary, Mukund, Züfle, Maike, Rajaee, Sara, Chen, Pinzhen, Vamvas, Jannis, Papi, Sara, de Gibert, Ona, Malik, Bhavitvya, Habba, Eliya, Mastromichalakis, Orfeas Menis, Schmidtová, Patrícia, Wastl, Michelle, Issaka, Sheriff, Choshen, Leshem, Biderman, Stella, Anastasopoulos, Antonis, Niehues, Jan, Sennrich, Rico, Sachan, Mrinmaya, Bojar, Ondřej, Murray, Kenton, Tiedemann, Jörg, Aji, Alham Fikri, Koehn, Philipp, Monz, Christof, Birch, Alexandra, Vajjala, Sowmya, Kranti, Chalamalasetti, España-Bonet, Cristina, Sarwar, Nobin, Kaczér, David, Asano, Shunta, Marmonier, Malik, Jaff, Daban Q., Mishra, Vaisakhi, Khalifa, Hend Al-, Sarti, Gabriele, Saha, Sourajit, Rehlinger, Nils, Villa, Juan Daniel Cuervo, Tonglet, Jonathan, Purkayastha, Saugata, Macháček, Dominik, Ramanujam, Jagannathan, Do, Heejin, Nadova, Zuzana, Philippy, Fred, Retkowski, Fabian, Lymperaiou, Maria, Casola, Silvia, Yukhymenko, Hanna, Dipta, Shubhashis Roy, Ryu, Sangwon, Jerez, Andrés, Keinan, Ron, Yusuf, Shuaib Shuaib, Vempati, Avantica, Staiano, Maria Carmen, Purkayastha, Sukannya, Cosma, Adrian, Babenko, Vitalii, Inan, Erivan, Nigam, Aviral, Aissa, Wafa, Haouari, Fatima, Gummadi, Venkata Prasanth Kumar, Jafarzadeh, Mehdi, Scourneau, Valentin, Edman, Lukas, Sun, Kaiser, Tan, Shaomu, Gholizadeh, Mohammad Sadegh, David, Johannes-Rudolf, Srirag, Dipankar, Gilabert, Javier García, Binkyte, Ruta, Ali, Manar, Bucur, Ana-Maria, Farrag, Sabry E., Saber, Youssef, Liu, Yihong, Maillard, Jean, Nicoleta, Cojocaru, Yuan, Xiaochuang, Ahmadi, Sina, Mondorf, Philipp, Dhole, Kaustubh, Wixinger, Roman, Qian, Shenbin, Tuor, Manuel, Troshin, Sergey, Yahav, Jonathan, Thoker, Fida Mohammad, Rezapour, Amir Arsalan, Gamboa, Lance Calvin Lim, Reusens, Manon, Kukk, Kätriin, Chowdhury, Koel Dutta, Gallipoli, Giuseppe, Hoang, Christian, Saha, Shaswati, Aycock, Seth, Kocoń, Jan, Chen, Bo, Vu, Linh, Venkatkrishna, Vatsal, Ahsan, Arafat, Nguyen, Luan Thanh, Soliman, Hassan, Dementieva, Daryna, Rampisela, Theresia Veronika, Do, Ngoc Quynh Tram, Huber, Marius, Egashira, Kazuki, Wasi, Azmine Toushik, Poritski, Vladislav, Zhang, Mike, Shah, Deep, Gavrikov, Paul, Salim, Luis Frentzen, Africa, David, Damanhuri, R., Bello, Bello Umar, Garg, Anumit, Rao, Gengyu, Ammanamanchi, Pawan Sasanka, Dementaviciute, Kamile, Michail, Andrianos, Teja, L D M S Sai, Zhu, Dawei, Fan, Yi, Liu, Wei, Farsi, Farhan, Herranen, Elias, Chowdhury, Sankalan Pal, Sanchez, Karen, Shami, Farzad, Urlana, Ashok, Wang, Zimu, Limisiewicz, Tomasz, Pattnayak, Priyaranjan, Ojastu, Marii, Na, Hongbin, Radoi, Emilian, Zhao, Chenyi, Hinojosa, Carlos, de Varda, Andrea Gregor, Alyafeai, Zaid, Alzahrani, Reem, Kathrotia, Nehal, Flückiger, Alex, Carr, Ulysses Sekai Tully, Layacan, Jimson Paulo, Kaplan, Guy, Tiwari, Ritwik, Dagli, Rishit, Volchek, Oksana, Caswell, Isaac R, Yi, Bowen, Kövér, Blanka, Yari, Amir Hossein, Chorana, Aicha, Wang, Zhengxiang, Keränen, Selja, Simko, Samuel, Olusanya, Joy, Chim, Jenny, Doyen, Enzo, Lakkamaneni, Vivek Harsha, Conrad, Sophia, Sadeghi, Pouya, Panayiotou, Panayiotis, Lara, Luis, Nayem, Jannatul, Yahav, Eran, Das, Debanshu, Karamolegkou, Antonia, Goel, Anmol, Mandal, Aishik, Cerruti, Tommaso, Zhao, Raoyuan, Haltiuk, Mykola, Aung, Thura, Almousa, Naser, Kargaran, Amir Hossein, Bawden, Rachel, Zheng, Qiaoyuan, Lango, Mateusz, Egressy, Beni, Velásquez, Fidel Rodríguez, Jongwiriyanurak, Natchapon, Do, Minh Ngoc, Gaido, Marco, Libon, Lena, Kuzmin, Dzmitry, Nyalang, Badal, Taroni, Antoine, Niculae, Andrei, Kani, Abdulaziz Nura, Zawar, Rushikesh, Šuppa, Marek, Savoldi, Beatrice, Simons, Andreas, Merchant, Rayyan, Levy, Ilai Yaron, Pinto, Francesco, Yang, Ziyi, Xavier, Yolanda, Frontull, Samuel, Habibi, Muhammad Ravi Shulthan, Enevoldsen, Kenneth, Majid, Harris Abdul, Padovani, Francesca, Graf, Tim, Bielakova, Tatiana, Djurabaeva, Sharifa, Ji, Shaoxiong, Ahmad, Raia Abu, Stepachev, Pavel, Qi, Jirui, Munot, Ayush Sunil, Pakniat, Alireza, Terryn, Ayla Rigouts, Lu, Yuxing, Paniv, Yurii, Fu, Xiyan, Adewumi, Tosin, Kumar, Sunisth, Thunus, Stéphane J. P. S., Satish, Shree Harsha Bokkahalli, Bali, Shayan, Gupta, Prakhar, Diallo, Papa Abdou Karim Karou, Akrap, Matija, Culjak, Marko, Onderková, Kristýna, Attieh, Joseph, Tensay, Esrael Teferi, Fittschen, Elisabeth, Sagot, Benoît, Ni, Jingwei, Fan, Yu

Abstract

For scientific progress, we need benchmarks that test the limits of state-of-the-art models, and evaluation methods that inform us about failure cases. As models get stronger, standard benchmarks for machine translation are approaching saturation. Further, automatic translation metrics are unreliable, vulnerable to reward-hacking, and provide unactionable assessments. Even gold human evaluation is not problem-free, because it often lacks reproducibility, objectivity, and scalability. Overall, this prevents us from tracking objective progress in the field and identifying pathways for improvement. We introduce the Last Translation Benchmark, a collection of human-authored and peer-reviewed examples (texts, images, audio, videos) that break leading machine translation models. We also present a new evaluation approach: each example comes with handcrafted verification rules describing concrete failure cases on that example, therefore allowing reliable and actionable future evaluation. The Last Translation Benchmark is a live dataset that accepts ongoing contributions. The latest version is LTBv1, containing accepted contributions prior to September 1st 2026, with future releases planned as new data is continuously collected.

Chinese Translation

为了推动科学进步，我们需要能够测试最先进模型极限的基准，以及能够告知我们失败案例的评估方法。随着模型的不断增强，机器翻译的标准基准正接近饱和。此外，自动翻译指标不可靠，容易受到奖励操控的影响，并且提供的评估结果难以采取行动。即使是黄金标准的人类评估也并非没有问题，因为它往往缺乏可重复性、客观性和可扩展性。总体而言，这阻碍了我们在该领域追踪客观进展和识别改进路径。我们介绍了最后翻译基准（Last Translation Benchmark），这是一个由人类创作并经过同行评审的示例集合（文本、图像、音频、视频），这些示例能够突破领先的机器翻译模型。我们还提出了一种新的评估方法：每个示例都附带手工制作的验证规则，描述该示例上的具体失败案例，从而允许可靠且可操作的未来评估。最后翻译基准是一个动态数据集，接受持续的贡献。最新版本为LTBv1，包含截至2026年9月1日的接受贡献，未来将根据持续收集的新数据计划发布新版本。

View on arXiv Download PDF AI Translation

cs.CL / 72 / 2609.04180

Knowledge Acquisition During Pre-training? Large Language Models Learn Better With Auxiliary Views

预训练期间的知识获取？大型语言模型通过辅助视角学习得更好

Lee, Joseph, Huang, Yidi, Kim, Dokyoon, Yang, Shu, Shen, Li

Abstract

Gaps remain in our understanding of how large language models (LLMs) acquire knowledge during pre-training. We posit that auxiliary views, reformulations of knowledge, are causally helpful for learning. We design controlled experiments to isolate this. First, we confirm that repetition is necessary for acquisition and clarify that paraphrasing helps only at smaller batch sizes. Second, holding the token budget fixed, allocating tokens from document repetition to auxiliary views improves learning, counterintuitively, even for factual recall. Third, the effectiveness of auxiliary views is not contingent on the strength of the teacher model that generates them. Fourth, we identify forms of knowledge, contextual and foundational, that aid learning in the presence of prior knowledge gaps. Finally, we examine how these effects manifest mechanistically via layer-wise biases and compression. Together, our findings suggest that auxiliary representations of knowledge, which arise naturally in large pre-training corpora, are a key factor in the success of pre-training and offer a plausible explanation for why data diversity matters.

Chinese Translation

我们对大型语言模型（LLMs）在预训练期间如何获取知识的理解仍存在空白。我们假设辅助视角，即知识的重述，对学习具有因果帮助。我们设计了受控实验以对此进行隔离。首先，我们确认重复对于知识获取是必要的，并澄清重述仅在较小的批量大小下有帮助。其次，在固定的标记预算下，将来自文档重复的标记分配给辅助视角，反而提高了学习效果，即使是在事实回忆方面也是如此，这一点与直觉相悖。第三，辅助视角的有效性并不依赖于生成它们的教师模型的强度。第四，我们识别出在存在先前知识空白的情况下，有助于学习的知识形式，包括上下文知识和基础知识。最后，我们考察了这些效应如何通过层级偏差和压缩机制表现出来。综合来看，我们的发现表明，知识的辅助表征在大型预训练语料库中自然产生，是预训练成功的关键因素，并为数据多样性为何重要提供了合理的解释。

View on arXiv Download PDF AI Translation

cs.CL / 73 / 2609.04194

Legibility is Not Interpretability: Comparing Judged and Actual Importance in Chain-Of-Thought Reasoning

可读性并非可解释性：比较链式推理中的判断重要性与实际重要性

Du, Kevin, Hoyle, Alexander, Ruis, Laura, Locatelli, Acyr

Abstract

Reasoning traces from chain-of-thought models appear to offer a legible window into how a model arrives at its answer. A growing body of work treats them as such, using LLM judges to diagnose errors, evaluate faithfulness, and provide step-level supervision via process reward models and generative critics. These practices rely on the text of a reasoning step carrying information about its functional role. But does the text actually encode information about which reasoning steps matter? We operationalize the importance of a reasoning step as its advantage: the change in expected reward, e.g., producing the correct final answer, from including that step, estimated via Monte Carlo rollouts. Basing ground truth on these estimates, we evaluate whether LLM judges can identify high-advantage steps and find that sufficiently capable LLMs can outperform a prevalence baseline but fall well short of a noise ceiling. Fine-tuning a model as a step-level critic yields strong improvement for incorrect responses but remains distant from ceiling for correct responses, suggesting that step importance is only partially recoverable from the text of the reasoning trace. Our findings contribute to a growing body of chain-of-thought faithfulness work that cautions against treating the legibility of reasoning traces as interpretability, especially with implications for process reward modeling.

Chinese Translation

链式推理模型的推理轨迹似乎提供了一个清晰的窗口，展示模型是如何得出答案的。越来越多的研究将其视为如此，利用大型语言模型（LLM）评估错误、评估忠实性，并通过过程奖励模型和生成性评论者提供逐步监督。这些做法依赖于推理步骤的文本携带关于其功能角色的信息。但文本是否真的编码了哪些推理步骤重要的信息？我们将推理步骤的重要性操作化为其优势：即通过包括该步骤而导致的预期奖励变化，例如，产生正确的最终答案，这一变化通过蒙特卡洛回滚法进行估计。基于这些估计的真实情况，我们评估了LLM评审者是否能够识别高优势步骤，发现足够强大的LLM能够超越普遍基线，但仍远未达到噪声上限。将模型微调为逐步评论者对错误响应的改进显著，但对于正确响应仍然远离上限，这表明推理轨迹的文本仅部分可恢复步骤的重要性。我们的发现为日益增长的链式推理忠实性研究提供了贡献，提醒人们在进行过程奖励建模时，谨慎对待推理轨迹的可读性与可解释性之间的关系。

View on arXiv Download PDF AI Translation

cs.CL / 74 / 2609.04197

ESPO: Error-Structured Prompt Optimization via Diagnose, Diversify, and Stabilize

ESPO：通过诊断、多样化和稳定化进行错误结构化提示优化

Liu, Lihao, Tang, Peng, Singh, Kunwar Yashraj, Ghadar, Shabnam

Abstract

Evolutionary prompt optimizers such as GEPA suffer from prompt bloat: each iteration appends rules and caveats, producing prompts up to 3$\times$ longer yet no more accurate. We trace this to three deficiencies - incomplete error observation, limited search diversity, and unreliable selection - and propose ESPO (Error-Structured Prompt Optimization), which decomposes prompt optimization into three phases: Diagnose clusters all training errors into structural patterns in one round; Propose generates candidates via four complementary strategies with independent biases; Select applies bootstrap stability selection. On seven public NLP benchmarks - Tweet, MMLU, GSM8K, HotpotQA, ScoNe, HoVer, and PUPA - ESPO improves average accuracy by $+$3.76 pp over the state-of-the-art (74.67% vs 70.91% for GEPA), matching or exceeding GEPA on every dataset while producing prompts 47% shorter (1,004 vs 1,878 chars) and faster at inference. Cross-model experiments across four additional student models (Gemma 3 12B, Mistral 14B, Qwen3 32B, Claude Haiku 4.5) show ESPO yields the best average accuracy on every model tested, with the largest gap on Qwen3 GSM8K (15.00% $\to$ 91.40%). A generalization bound (Appendix) grounds each phase in a corresponding term of the test-time gap, and the ablation confirms a key prediction: adding diversity without bootstrap selection actually hurts performance ($-$1.20%).

Chinese Translation

进化提示优化器如GEPA面临提示膨胀的问题：每次迭代都会附加规则和注意事项，导致提示长度增加至最多3倍，但准确性并未提高。我们将此归因于三个缺陷——不完整的错误观察、有限的搜索多样性和不可靠的选择——并提出了ESPO（错误结构化提示优化），该方法将提示优化分解为三个阶段：诊断阶段将所有训练错误聚类为结构模式；提议阶段通过四种互补策略生成具有独立偏差的候选项；选择阶段应用自助稳定选择。在七个公共自然语言处理基准测试上——Tweet、MMLU、GSM8K、HotpotQA、ScoNe、HoVer和PUPA——ESPO的平均准确率比最先进的方法提高了3.76个百分点（74.67%对比GEPA的70.91%），在每个数据集上均与GEPA持平或超越，同时生成的提示长度减少了47%（1,004个字符对比1,878个字符），推理速度更快。跨四个额外学生模型（Gemma 3 12B、Mistral 14B、Qwen3 32B、Claude Haiku 4.5）的跨模型实验表明，ESPO在每个测试模型上都获得了最佳的平均准确率，其中在Qwen3 GSM8K上差距最大（15.00%提升至91.40%）。一个推广界限（附录）将每个阶段与测试时差的相应项联系起来，消融实验确认了一个关键预测：在没有自助选择的情况下增加多样性实际上会降低性能（-1.20%）。

View on arXiv Download PDF AI Translation

cs.CL / 75 / 2609.04199

Compile by Training: Turning Natural-Language Specifications into Local Neural Functions

通过训练编译：将自然语言规范转化为局部神经函数

Deng, Yuntian, Nie, Pengyu, Shieber, Stuart

Abstract

Many recurring text functions are easy to describe but difficult to implement with rules, while calling a large remote model for every input introduces repeated cost, latency, and dependency on a provider. We present compile by training, which turns a natural-language specification into a reusable neural function. At compile time, teacher models generate task-specific examples that are used to train a small adapter for a compact interpreter. The resulting function runs without the teachers and can be stored, versioned, and composed like ordinary software. On FuzzyBench-Hard, a subset on which the Program-as-Weights fast compiler produced no exact matches, compile by training reaches 83.6% semantic accuracy. This higher accuracy comes with a higher compile-time cost: roughly a minute rather than seconds for the fast compiler. We deploy the compiler in a public interactive service and demonstrate compiled functions in a multi-site website helper, a language-controlled 3D avatar, and a bidirectional English-Claudish translator.

Chinese Translation

许多重复出现的文本功能容易描述但难以通过规则实现，而对每个输入调用一个大型远程模型则会引入重复的成本、延迟以及对提供者的依赖。我们提出了通过训练编译（compile by training），将自然语言规范转化为可重用的神经函数。在编译时，教师模型生成特定任务的示例，用于训练一个小型适配器以构建紧凑的解释器。生成的函数在没有教师的情况下运行，并且可以像普通软件一样存储、版本控制和组合。在 FuzzyBench-Hard 上，程序作为权重（Program-as-Weights）快速编译器未能产生任何精确匹配，而通过训练编译的语义准确率达到了 83.6%。这种更高的准确率伴随着更高的编译时间成本：大约需要一分钟，而快速编译器则只需几秒。我们在一个公共互动服务中部署了该编译器，并展示了在多站点网站助手、语言控制的 3D 头像以及双向英语-克劳迪什（Claudish）翻译器中的编译函数。

View on arXiv Download PDF AI Translation

arXiv Papers

Seeing Less Is Not Seeing Safely: Privacy Leakage from Task-Scoped Robot Perception Exports

GPU-Accelerated Astrodynamics World Models for Spacecraft Rendezvous and Proximity Operations

Sensing Which Modality Matters: Evidence-Gated Regularization for Robust VLA Policies

Real-Time Shape Control of Multi-Segment Soft Robotic Arms Using Koopman Operators with Global and Local Observables

Following a Unique Path: A Fast Certifier Applied to Outlier-Robust Pose Registration

Long-Horizon Consistent and Interaction-Aware World Models for Multi-Style End-to-End Driving

Establishing a Dynamic Multimodal HRI Dataset for Engagement Analysis with a Humanoid Robot

R2S-Eval: Robot Evaluation with Real-to-Sim Calibration via Vision-Language Models

ARTiS: An Adaptive Robotic Gripper for Enhanced Tool Manipulation in Disassembly Applications

Programming and execution of skill-based human-robot-crane collaborative tasks

Air-Ground Collaborative Vision-and-Language Navigation via Shared Bird's-Eye Maps

BRIDGE: An Open-Source Humanoid Platform via Morphology-Control Co-Design for Physical AI

TRaIL-Odom: Tightly Coupled Continuous Time Radar-IMU-LiDAR Odometry with Adaptive Doppler Weighting

Toward Physically Grounded JEPA World Models for Goal-Conditioned Robotic Planning

Scaling Bimanual Household Manipulation from 1,500 hours of Demonstrations to On-Policy Corrections

FailBench: How Reliable are VLMs at Judging Robot Task Success?

QLAUN: A Research-Oriented, Robust, Agile, Modular, and Affordable Torque-Controlled Quadruped Robot

Local Path Planning and Obstacle Avoidance for an Omnicopter Platform

WISE: World-model-guided Imagination Scheduling for Efficient Post-training of Vision-Language-Action Models

Predictive Zonotope Reduction: Precise Runtime Monitoring under Uncertainty

Toward an~Integrated Cognitive--Ergonomic Architecture for~Human--Machine Interaction: Combining Cognitive Models with~Human Factors Ergonomics

MINERVA: How Small Can a Manipulation Policy Be and Still Solve LIBERO?

RoughSense: Lightweight Terrain-Induced Rover Vibration Prediction Using Point Clouds and IMU Feedback

A Multi-Vine Soft Robot Enabling Accessible Working Channel and Steering

Virtual Testing of Automated Driving Systems through Credible Simulations

Robot Aware Computational Design of Object Specific Passive Grippers for Additive Manufacturing

A comparative study on the accuracy & repeatability of mobile robotic platforms for the delivery of precision NDE measurement

FWBC-VLA: Force-Aware Whole-Body Compensation for Contact-Rich Loco-Manipulation

A hybrid pipeline for dynamic ontology-based semantic mapping

Revisiting Topological Graphs for Macro Action based Closed-loop Reinforcement Learning of Vision Language Navigation in Continuous Environment

Toward Unified Robot Learning: Bridging Representation, Vision-Language-Action, and World Models

Automated Weld Seam Recognition and 3D Mapping for Robotic Post Processing Using Photogrammetry and Semantic Segmentation

MulDP: Multimodal Diffusion Policy for Autonomous Quadruped Parkour Navigation across Complex Terrains

Adaptive Vision-Language Grasping via Composable Foundation Priors and Generalizable Grasp Synthesis

Corner Cases: Headland Coverage Path Planning for Autonomous Driving in Arable Farming

GIFT: Guided Intermediate Feature Training via Action-Oriented Structural Supervision for Robotic Manipulation

IDSPACE: A Novel Document Generator for Reliable Evaluation of Digital Identity Verification Systems [Extended Technical Report]

Position: Unlabeled IS NOT Equal to No Human Supervision in Visual Learning

Exemplar: Classical Priors Complement Frozen Features for Few-Shot Microscopy Segmentation at Native Resolution

Solving the Needle-in-a-Haystack Problem in Mammography Vision-Language Model with Differentiable Subset Sampling

WireSeg-32K: A Physics-Grounded Synthetic Dataset for Wire Instance Segmentation

SLIDEFORGE: An LLM Agent for Controllable Editing of Slides as Structured Artifacts

Beyond Small Patches: Black-Box Detection and Purification of Diverse Backdoor Triggers

VeriPhy: Agentic Physical Reasoning for World Model Evaluation and Refinement

Who Speaks for the Pruned? Visual Token Pruning as Coverage Optimization

RoboTok: An Internet-Scale Data Engine for Human Demonstration Retrieval and Dexterous Manipulation Learning

Learning to Zoom Efficiently with a Contrastive Curriculum

ProgResViT: Progressive Resolution and Width for Adaptive Vision Transformers

Counting Animals in Camera-Traps Image Sequences without Count Labels: Winning Solution to the iWildCam 2021 Challenge

An Ensemble-Based Self-Taught Learning Approach for Parking Space Classification Under Limited Data

MedQA-MM: Shortcuts Behind Medical Visual Reasoning

Tensor-based Brain Surface Modeling and Analysis

Laplacian Frequency Hierarchies for Efficient 3D Gaussian Splatting Training

PointGT: Simultaneous Geometry and Texture Editing for Point-Based Representations

P-CORE: Self-Supervised Surface Consistency for Point-Based Neural Editing

When Depth Hurts: Reliability-Aware Geometry Distillation for Depth-Free RGB-D Salient Object Detection

FoRIS: Progressive Foreground Refinement for Training-Free In-Context Segmentation

Exploring the Potential of Contrastive Language-Image Pre-training for Multi-Source Remote Sensing Data

Neural-Collapse-guided Task-Free Continual Anomaly Detection

Mudragen: Geometrically Supervised Generation of Interacting Two-Hand Mudras for Preserving Indian Classical Dance Heritage

When Do Frozen VLMs Respond to Image-Free Object-Token Edits? An Answer-Key-Free Protocol and What It Reveals

OCR-EDR: Rendering-Aware Diagnosis and Repair for Closed-Loop OCR Improvement

Preserving Knowledge across Space and Time for Continual Video Deepfake Detection

STARS-GS: Structure-Aware Regularized Gaussian Splatting for Large-Scale Aerial Surface Reconstruction

Preprocessing Failure and Adversarial Detection in Depthwise-Separable Edge Vision Systems

BMCTrack-d: Pig re-identification and tracking via back marks in challenging camera settings

SafeRestore: Detector-Relative Risk Certificates for Selective Industrial Image Restoration

Tree species mapping in Denmark: A comparison of spectral-temporal features with geospatial foundation model embeddings

Residual Optimal Transport-Based Experts Collaboration Towards Modality-Aware Infrared-Visible Object Detection

Neural Video Compression Based on Deformable Temporal Alignment and Difference-aware Fusion

TruncGradGS: Improved 3D Gaussian Splatting via Truncated Gradient Updates

SafeRI: Recognition and Intervention for Token-Level Safety Intervention in Large Vision Language Models

WIDE: Wildcard Inference with Dynamic Expansion for Cross-Modal Generative Retrieval

Building Pretraining Data for World Models: An Unreal Engine-Based Pipeline for Action-Conditioned Video Generation

FlashRender: Few-Step Generative Rendering via Camera-Controlled Video MeanFlow

Occlusion-Robust Multimodal Emotion Recognition in VR via Fusion of Facial Images and EMG

Drive-HWM: Hierarchical World Models for Dynamic-Latent Guided Autonomous Driving

Text2Thermal: Physics-Aware Thermal Image Synthesis from Textual Priors

SV-WAM: An Efficient Surround-View World-Action Model for End-to-End Autonomous Driving