← Back to Index
Daily Research Digest

arXiv Papers

2026-06-08
230
Papers
4
Categories
230
Translated
收藏清单 0
机器人学 (Robotics)
44
cs.RO / 1 / 2606.06556

Robots Need More than VLA and World Models

机器人需要的不仅仅是视觉-语言-动作模型和世界模型
Karcini, Elis, Mehrban, Faisal, Nguyen, Quang, Schwager, Mac, Ajoudani, Arash, Cadena, Cesar, Peters, Jan, Hutter, Marco, Bou-Ammar, Haitham
Abstract
Generalist robot intelligence is often framed as a policy-scaling problem: collect more robot demonstrations, train larger Vision-Language-Action (VLA) models, and expect broader generalisation. In this position paper, we argue that this framing is incomplete. The central bottleneck is not only policy learning, but the absence of mechanisms that convert the world's abundant unstructured behavioural data into grounded robot supervision. Human motion, internet video, simulation rollouts, and interactive demonstrations contain rich information about tasks, goals, contacts, failures, and physical constraints, yet most of this information is not directly usable by robot policies because it lacks embodiment-specific action labels, task semantics, and reward structure. We identify four missing components for the next generation of robotics: data interfaces for autolabelling unstructured behaviour, embodiment interfaces for retargeting human motion to robot actions, world-model interfaces for physics-grounded 3D reasoning, and reward interfaces for inferring task progress and success from video and language. We survey recent progress in robot foundation models, cross-embodiment datasets, learning from video, world models, and reward modelling, and propose a research agenda for building robotics systems that can learn not only from robot demonstrations, but from the broader physical world.
Chinese Translation
通用机器人智能通常被框架为一个策略扩展问题：收集更多的机器人演示，训练更大的视觉-语言-动作（VLA）模型，并期望更广泛的泛化。在这篇立场论文中，我们认为这种框架是不完整的。中央瓶颈不仅在于策略学习，还在于缺乏将世界丰富的非结构化行为数据转化为有根据的机器人监督的机制。人类运动、互联网视频、模拟回放和互动演示包含了关于任务、目标、接触、失败和物理约束的丰富信息，然而大多数这些信息由于缺乏特定于体现的动作标签、任务语义和奖励结构，无法被机器人策略直接使用。我们确定了下一代机器人所缺失的四个组成部分：用于自动标注非结构化行为的数据接口、用于将人类运动重新定向到机器人动作的体现接口、用于物理基础的三维推理的世界模型接口，以及用于从视频和语言推断任务进展和成功的奖励接口。我们回顾了机器人基础模型、跨体现数据集、从视频学习、世界模型和奖励建模的最新进展，并提出了一个研究议程，以构建能够不仅从机器人演示中学习，还能从更广泛的物理世界中学习的机器人系统。
View on arXiv Download PDF AI Translation
cs.RO / 2 / 2606.06569

PhyRoGen: Synthetic Generation of Physical Robot Manipulation Puzzles Using Procedural Content Generation

PhyRoGen：使用过程内容生成的物理机器人操控难题的合成生成
Droß, Lennart Julian, Orthey, Andreas, Toussaint, Marc
Abstract
Robot manipulation of physical puzzles is important for automatic assembly and disassembly tasks. However, to enable robots to solve physical puzzles, manipulation skills need to be learned, which requires large training datasets, the generation of which is often time consuming and tedious. To overcome this problem, we propose the Physical Robot Manipulation Puzzle Generation framework (PhyRoGen), which leverages procedural content generation (PCG) for automated generation of synthetic datasets of manipulation puzzles. PhyRoGen is a general-purpose puzzle generator, which can generate physical puzzles with interlocking object dependencies, where one articulated object must be manipulated before another can be moved. Based upon PhyRoGen, we define six concrete generators which we use to generate 24 physical puzzles. By using a benchmarking framework, we are able to solve all puzzles in 1 to 300 seconds using sampling-based planning algorithms. Finally, we demonstrate that every generated puzzle is manipulatable by using a KUKA LBR iiwa robot in a physical simulation. This shows that our framework is able to procedurally generate unique, solvable robot manipulation puzzles, which is a crucial ingredient to benchmark manipulation algorithms and to develop robust foundation models.
Chinese Translation
物理难题的机器人操控对于自动组装和拆卸任务至关重要。然而，为了使机器人能够解决物理难题，必须学习操控技能，这需要大量的训练数据集，而生成这些数据集通常既耗时又繁琐。为了解决这个问题，我们提出了物理机器人操控难题生成框架（PhyRoGen），该框架利用过程内容生成（PCG）自动生成合成的操控难题数据集。PhyRoGen是一个通用的难题生成器，能够生成具有互锁对象依赖关系的物理难题，其中一个关节对象必须在另一个对象可以移动之前被操控。基于PhyRoGen，我们定义了六个具体生成器，用于生成24个物理难题。通过使用基准测试框架，我们能够使用基于采样的规划算法在1到300秒内解决所有难题。最后，我们展示了每个生成的难题都可以通过在物理仿真中使用KUKA LBR iiwa机器人进行操控。这表明我们的框架能够程序性地生成独特且可解的机器人操控难题，这是基准测试操控算法和开发稳健基础模型的关键要素。
View on arXiv Download PDF AI Translation
cs.RO / 3 / 2606.06618

ChronoForest: Closed-Loop Multi-Tree Diffusion Planning for Efficient Bridge Search and Route Composition

ChronoForest：高效桥接搜索与路径组合的闭环多树扩散规划
Seo, Jungmin, Park, Jaesik
Abstract
How can we plan long-horizon routes that reach designated goals, visit required waypoints, and remain short when only short-horizon offline trajectories are available? This problem matters in offline navigation because collecting sufficiently rich long-horizon data is difficult, yet real agents must still solve long-range tasks with route-level efficiency rather than mere feasibility. The difficulty is twofold: at the microscopic level, composing many short-horizon segments creates a trade-off between search cost and path quality, while at the macroscopic level, waypoint ordering requires comparing pairwise travel costs among start, goal, and waypoint anchors that are unknown before planning and increasingly unreliable when estimated only from long-range temporal distance. In this paper, we propose ChronoForest, a closed-loop planning system that couples local bridge search and online route re-solving through an anchor-chaining tree diffusion planner and an online multi-tree orchestrator. ChronoForest uses temporal distance for short-range guidance and node evaluation, while using search-time bridge evidence to validate long-range anchor connectivity and repeatedly re-solve the route. On OGBench AntMaze-Stitch, ChronoForest achieves 99.8%, 99.3%, and 99.5% success on the medium, large, and giant splits and improves giant-stitch success by up to 34.5 points over prior reported diffusion-based results. On Hamiltonian route-composition benchmarks, online re-solving corrects poor temporal orderings and improves route quality while remaining substantially cheaper than exhaustive planning.
Chinese Translation
如何规划能够达到指定目标、访问必要途径点并保持简短的长时间路线，而仅有短时间离线轨迹可用？这个问题在离线导航中非常重要，因为收集足够丰富的长时间数据是困难的，但真实的代理仍然必须以路径级效率而不仅仅是可行性来解决长距离任务。这个难题有两个方面：在微观层面，组合多个短时间段会在搜索成本和路径质量之间产生权衡；而在宏观层面，途径点的排序需要比较起点、目标和途径点锚点之间的成对旅行成本，而这些在规划之前是未知的，并且在仅从长时间的时间距离估计时变得越来越不可靠。在本文中，我们提出了ChronoForest，一个闭环规划系统，通过锚链树扩散规划器和在线多树协调器，将局部桥接搜索与在线路线重新解决相结合。ChronoForest利用时间距离进行短距离指导和节点评估，同时使用搜索时间的桥接证据来验证长距离锚点的连通性，并反复重新解决路径。在OGBench AntMaze-Stitch上，ChronoForest在中型、大型和巨型分割中分别达到了99.8%、99.3%和99.5%的成功率，并在巨型拼接成功率上比之前报告的基于扩散的结果提高了多达34.5个百分点。在哈密顿路线组合基准测试中，在线重新解决纠正了较差的时间排序，提高了路径质量，同时仍然比穷举规划便宜得多。
View on arXiv Download PDF AI Translation
cs.RO / 4 / 2606.06627

What Matters When Cotraining Robot Manipulation Policies on Everyday Human Videos?

在日常人类视频上共同训练机器人操作策略时，哪些因素是重要的？
Li, Richard, Prakash, Aditya, Wen, Andrew, Gupta, Saurabh, Du, Yilun, Agrawal, Pulkit
Abstract
Human video datasets used for cotraining robot manipulation policies largely consist of curated demonstrations where motions are orchestrated to resemble robot behavior and 3D hand poses are captured with specialized hardware. A more plentiful source of data is everyday Internet video, but it is an open question what factors enable transfer from such videos to robots. We investigate this using a new dataset of 532 human videos with 28 hours of high-quality triangulated hand labels and natural motions. We find that hand pose quality affects transfer, but even with accurate hands, the inherent motion gap hinders transfer unless the vision and policy networks specialize to each embodiment. Our cotraining recipe yields consistent improvements, with an absolute success rate gain of $29.7\%$ in the low-robot-data regime across six manipulation tasks.
Chinese Translation
用于共同训练机器人操作策略的人类视频数据集主要由策划的演示组成，这些演示中的动作被精心安排以类似于机器人行为，并且三维手部姿势通过专用硬件捕获。一个更丰富的数据来源是日常互联网视频，但尚不清楚哪些因素能够促使这种视频向机器人转移。我们使用一个包含532个高质量三角化手部标签和自然动作的28小时人类视频的新数据集来研究这个问题。我们的研究发现，手部姿势的质量会影响转移，但即使手部准确，固有的动作差距也会阻碍转移，除非视觉和策略网络针对每种体现进行专业化。我们的共同训练方案在六个操作任务中，在低机器人数据条件下实现了绝对成功率提高29.7%的稳定改进。
View on arXiv Download PDF AI Translation
cs.RO / 5 / 2606.06686

On the Hardness of Optimal Motion on Trees

树上最优运动的难度
Geft, Tzvika
Abstract
This paper presents a simple framework that settles the complexity of Multi-Agent Path Finding (MAPF) on trees across standard objectives--distance, makespan, and flowtime--for both labeled and colored variants. In MAPF, agents occupy the vertices of a graph and must move to target vertices without collisions while optimizing a given objective. In the labeled case, the agents are distinct and have respective targets; in the colored case, agents of the same color are interchangeable. While many MAPF variants are known to be intractable, several basic cases on trees have remained open. We prove NP-hardness on trees for both labeled and 2-colored MAPF under all three objectives. In particular, we resolve the classical Pebble Motion problem, where one pebble moves at a time to an adjacent empty vertex and the goal is to minimize the total number of moves. Despite being one of the most basic discrete motion models, its complexity on trees had remained open for several decades. Moreover, for colored Pebble Motion, we give the first hardness result on any graph class, already with two colors, which is tight. All of these results are established through the hardness of Stack Rearrangement, itself posed as an open problem, which asks to optimally rearrange items stored in stacks, and which we also prove to be NP-hard. Notably, the connection to stacks yields hardness already on very simple trees--subdivided stars--across all problems. Together, these results reveal a common tractability barrier that permeates several fundamental motion models, thereby unifying and strengthening prior hardness results.
Chinese Translation
本文提出了一个简单的框架，解决了树上多智能体路径寻找（Multi-Agent Path Finding, MAPF）在标准目标——距离、完工时间和流动时间——下的复杂性问题，包括标记和着色变体。在MAPF中，智能体占据图的顶点，必须在不发生碰撞的情况下移动到目标顶点，同时优化给定的目标。在标记情况下，智能体是不同的，并且各自有目标；在着色情况下，相同颜色的智能体是可互换的。虽然已知许多MAPF变体是不可处理的，但树上的几个基本情况仍然悬而未决。我们证明了在所有三个目标下，标记和2着色MAPF在树上都是NP-困难的。特别地，我们解决了经典的棋子运动问题，其中一个棋子一次移动到相邻的空顶点，目标是最小化总移动次数。尽管这是最基本的离散运动模型之一，但其在树上的复杂性在数十年间仍未得到解决。此外，对于着色棋子运动，我们给出了在任何图类中关于两个颜色的第一个难度结果，这是紧密的。所有这些结果都是通过堆栈重排的难度建立的，堆栈重排本身被提出为一个开放问题，要求最优地重新排列存储在堆栈中的物品，我们也证明了它是NP-困难的。值得注意的是，与堆栈的联系使得在非常简单的树——细分星形图——上也产生了难度，涵盖了所有问题。总的来说，这些结果揭示了一个普遍的可处理性障碍，贯穿于几个基本运动模型，从而统一并加强了先前的难度结果。
View on arXiv Download PDF AI Translation
cs.RO / 6 / 2606.06704

Optimal Control Approach for Non-prehensile Ball Juggling Using a 7-DoF Manipulator

基于最优控制的非抓握球类杂技的7自由度机械臂控制方法
Ramadani, Joel, Rakčević, Vasilije, Laha, Riddhiman, Sachtler, Arne, Mesle, Valentin Le, Lilienthal, Achim J., Haddadin, Sami
Abstract
Non-prehensile object manipulation skills are important for real-world robot interactions, enabling highly dynamic tasks such as balancing a glass on a tray or the controlled sliding of items on a table. Among such tasks, those characterised by high-speed manipulation requirements and general sensitivity of the resulting hybrid dynamics are particularly hard to accomplish. Within these, juggling can be seen as a highly challenging maneuver to be solved. The key to robotic juggling is achieving dynamic stabilisation of an underactuated object. Since the object does not possess the ability of self-correction, its stability is entirely dependent on the forces applied to it. This creates a system that is sensitive to control inputs, where timing is critical to continuously counteract deviations and maintain the desired behavior. We develop a systematic method to control a 7-degree-of-freedom manipulator performing non-prehensile ball juggling with a tool. Our primary contribution is a model-based framework for generating juggling trajectories and stabilizing a periodic juggling motion for this hybrid system. The framework incorporates a two-stage optimal control approach to compute the underlying feasible motion patterns required for stable juggling. Offline-computed trajectories are then organised to enable real-time error correction without solving optimal control problems online. We demonstrate the effectiveness of the resulting controller by first evaluating its performance in a simulation environment and performing an experiment using a Franka Emika Panda robot.
Chinese Translation
非抓握物体操作技能对于现实世界中的机器人交互至关重要，使得机器人能够完成高度动态的任务，例如在托盘上平衡玻璃杯或在桌子上控制物品滑动。在这些任务中，具有高速操作要求和结果混合动力学的普遍敏感性特征的任务尤其难以完成。在这些任务中，杂技被视为一种极具挑战性的操作。机器人杂技的关键在于实现对欠驱动物体的动态稳定化。由于物体不具备自我修正的能力，其稳定性完全依赖于施加在其上的力。这就形成了一个对控制输入敏感的系统，其中时机对于持续抵消偏差和保持期望行为至关重要。我们开发了一种系统化的方法来控制一个执行非抓握球类杂技的7自由度机械臂。我们的主要贡献是为这一混合系统生成杂技轨迹并稳定周期性杂技运动的基于模型的框架。该框架结合了两阶段的最优控制方法，以计算稳定杂技所需的基本可行运动模式。离线计算的轨迹随后被组织，以便在不在线解决最优控制问题的情况下实现实时误差修正。我们通过首先在仿真环境中评估其性能，并使用Franka Emika Panda机器人进行实验，展示了所得到的控制器的有效性。
View on arXiv Download PDF AI Translation
cs.RO / 7 / 2606.06721

SCOUT: Semantic scene COverage via Uncertainty-guided Traversal

SCOUT：通过不确定性引导的遍历实现语义场景覆盖
Mao, Junyu, Ayoubi, Sara, Sharma, Vishnu D., Hadžić, Ilija, Andrews, Matthew
Abstract
Robots that operate over extended periods should not merely visit space; they should progressively understand it. Yet most 3D scene graph pipelines treat perception as a post-processing stage over a fixed dataset, decoupling scene representation from the decisions that determine what is observed in the first place. We present SCOUT, an online semantic exploration framework that closes this loop by coupling active traversal with probabilistic scene graph construction. Given a prior 2D occupancy map and posed RGB-D observations, SCOUT incrementally builds an uncertainty-aware 3D scene graph whose nodes maintain fused geometry and posterior beliefs over open-vocabulary object labels, while edges encode structural relations such as on, inside, belong, and next to. These beliefs are fed back to an uncertainty-guided traversal planner, which selects viewpoints by balancing expected semantic certainty gain, geometric coverage gain, and travel cost. In this way, the robot revisits ambiguous objects when additional evidence matters and expands into unseen free space when the scene remains incomplete. The resulting system treats semantic scene completeness as an operational objective rather than a passive by-product of semantic mapping, moving toward autonomous agents that can patrol, update, and reason about evolving indoor environments with minimal human intervention.
Chinese Translation
长时间运行的机器人不仅应访问空间，还应逐步理解空间。然而，大多数3D场景图管道将感知视为对固定数据集的后处理阶段，将场景表示与决定首次观察内容的决策解耦。我们提出了SCOUT，一个在线语义探索框架，通过将主动遍历与概率场景图构建结合，闭合了这一循环。在给定先前的2D占用图和已定位的RGB-D观测的情况下，SCOUT逐步构建一个关注不确定性的3D场景图，其节点维护融合的几何信息和对开放词汇对象标签的后验信念，而边则编码结构关系，如“在...上”、“在...内”、“属于”和“旁边”。这些信念被反馈给不确定性引导的遍历规划器，该规划器通过平衡预期的语义确定性增益、几何覆盖增益和旅行成本来选择视点。通过这种方式，机器人在额外证据重要时重新访问模糊对象，并在场景仍不完整时扩展到未见的自由空间。最终系统将语义场景的完整性视为一个操作目标，而不是语义映射的被动副产品，朝着能够以最小人类干预巡逻、更新和推理不断变化的室内环境的自主代理迈进。
View on arXiv Download PDF AI Translation
cs.RO / 8 / 2606.06727

IDDMBSE: Integrating Data-Driven and Model-Based Systems Engineering for Trusted Autonomous Cyber-Physical Systems

IDDMBSE：为可信的自主网络物理系统整合数据驱动与基于模型的系统工程
Baras, John S., Damera, Sai Sandeep, Matheu, Ryan, Enwerem, Clinton, Kumar, Praveen M. S.
Abstract
Autonomous cyber-physical systems (CPS) sit at the intersection of Model-Based Systems Engineering (MBSE) and data-driven Machine Learning and Artificial Intelligence (ML/AI), yet no integrated Systems Engineering (SE) methodology natively spans both. We address this gap with IDDMBSE, an Integrated Data-Driven and Model-Based Systems Engineering methodology that extends the rigorous MBSE V-process with a data-driven loop at every step, anchored in SysML, the autonomy stack, and a hybrid model-based plus data-driven trade-off architecture. We instantiate IDDMBSE as an interoperable, open-source tool chain: PERFECT, which maps SysML system architectures to executable ROS autonomy stacks for scalable performance evaluation; TRADES-X, which decomposes design-space exploration into a model-based optimization stage followed by a data-driven evaluation stage; and VERITAS, which combines formal, data-driven, and runtime verification into a single assurance workflow. We demonstrate IDDMBSE on a Trusted Autonomous Ground Robot across its development lifecycle, spanning sensor-suite selection, risk-sensitive path planning, behavior-tree task verification, conformal-prediction-based robust perception, and assured multi-robot coordination, all exercised in a contested-terrain Isaac Sim test range that we release with the tool chain. We close by sketching how IDDMBSE is being re-formulated on SysML v2 / KerML foundations to enable language-native composability and tighter ML/AI integration.
Chinese Translation
自主网络物理系统（CPS）位于基于模型的系统工程（MBSE）与数据驱动的机器学习和人工智能（ML/AI）之间的交叉点，但尚无一种集成的系统工程（SE）方法论能够原生涵盖这两者。我们通过IDDMBSE来填补这一空白，IDDMBSE是一种集成的数据驱动与基于模型的系统工程方法论，它在每个步骤中扩展了严谨的MBSE V过程，并在SysML、自治堆栈以及混合的基于模型和数据驱动的权衡架构的基础上建立了数据驱动循环。我们将IDDMBSE实例化为一个可互操作的开源工具链：PERFECT，它将SysML系统架构映射到可执行的ROS自治堆栈，以进行可扩展的性能评估；TRADES-X，它将设计空间探索分解为基于模型的优化阶段和随后的数据驱动评估阶段；以及VERITAS，它将形式验证、数据驱动验证和运行时验证结合成一个单一的保证工作流。我们在一个可信的自主地面机器人上展示了IDDMBSE，涵盖了其开发生命周期，包括传感器套件选择、风险敏感路径规划、行为树任务验证、基于符合性预测的鲁棒感知以及保证的多机器人协调，所有这些都在我们与工具链一起发布的竞争地形Isaac Sim测试场中进行。最后，我们概述了如何在SysML v2 / KerML基础上重新构建IDDMBSE，以实现语言原生的可组合性和更紧密的ML/AI集成。
View on arXiv Download PDF AI Translation
cs.RO / 9 / 2606.06761

AxisGuide: Grounding Robot Action Coordinate System in RGB Observations for Robust Visuomotor Manipulation

AxisGuide：在RGB观察中将机器人动作坐标系统与实际情况相结合，以实现稳健的视觉运动操控
Jang, Jiyun, Sung, Yujin, Joung, Woosung, Chae, Daewon, Lee, Sangwon, Kim, Sohwi, Kim, Jinkyu, Lee, Jungbeom
Abstract
Visuomotor manipulation policies trained via large-scale behavior cloning have achieved strong semantic scene understanding, yet often fail to reliably execute correct low-level actions under distribution shifts. For example, even in a simple pickup task with identical scene layouts, camera viewpoints, and illumination, performance can degrade substantially when the object is placed at unseen locations. We argue that this gap arises from insufficient action understanding, namely the inability to interpret the robot's base-frame action coordinate system in image space. To address this issue, we introduce AxisGuide, a lightweight guidance method that bridges semantic scene understanding and action-coordinate interpretation. Using camera parameters and end-effector poses, AxisGuide renders the robot base-frame axes in each camera view and augments RGB observations with a small set of cue channels that explicitly visualize the meaning of the +x, +y, and +z motions in image space. Extensive evaluations in both the LIBERO simulation and real-world environments demonstrate that AxisGuide yields substantial performance gains and improved generalization, highlighting the effectiveness of explicit action-coordinate cues for learning reliable and transferable generalist visuomotor policies.
Chinese Translation
通过大规模行为克隆训练的视觉运动操控策略已实现强大的语义场景理解，然而在分布变化下，往往无法可靠地执行正确的低级动作。例如，即使在具有相同场景布局、相机视角和照明条件的简单拾取任务中，当物体放置在未见位置时，性能也可能显著下降。我们认为，这一差距源于对动作理解的不足，即无法在图像空间中解释机器人的基坐标动作系统。为了解决这一问题，我们提出了AxisGuide，这是一种轻量级的指导方法，旨在桥接语义场景理解与动作坐标解释。通过使用相机参数和末端执行器姿态，AxisGuide在每个相机视图中渲染机器人基坐标轴，并通过一小组提示通道增强RGB观察，明确可视化图像空间中+x、+y和+z运动的含义。在LIBERO仿真和真实环境中的广泛评估表明，AxisGuide显著提高了性能和泛化能力，突显了明确的动作坐标提示在学习可靠且可转移的通用视觉运动策略中的有效性。
View on arXiv Download PDF AI Translation
cs.RO / 10 / 2606.06762

Multi-Robot Planning and Control from CCTV Camera Networks in a Real Warehouse

基于CCTV摄像头网络的多机器人规划与控制在真实仓库中的应用
Robinson, Luke, Ramtoula, Benjamin, Izaaryene, Anas, Newman, Paul, De Martini, Daniele
Abstract
Off-board control of mobile robots from cameras embedded in the environment offers a practical path to scalable autonomy, moving sensing and compute off the robots. We extend this idea from the single-robot case to coordinated fleets in a real warehouse, driving multiple robots with only a distributed CCTV network and edge compute. The system operates entirely in image space over an uncalibrated, pixel-wise topological camera graph, enabling wide-area operation with flexible camera placement. A hierarchical planner selects a camera sequence per robot and plans its image-space motion through each view, coordinating robots with a prioritised-then-joint strategy and treating overlapping camera regions as shared resources held by one robot at a time to prevent collisions and deadlocks. We validate the approach in a real warehouse with four robots and 30 cameras across six 27 m aisles, reporting mission times and coordination statistics. To our knowledge, this is the first field demonstration of multi-robot planning and coordination using only an external camera network and off-board compute, with robots carrying no task-specific navigation hardware.
Chinese Translation
通过环境中嵌入的摄像头对移动机器人进行离线控制，为可扩展自主性提供了一条实用路径，将传感和计算从机器人上移除。我们将这一理念从单机器人案例扩展到协调的机器人群体，在真实仓库中，仅依靠分布式CCTV网络和边缘计算驱动多个机器人。该系统完全在图像空间中操作，基于未校准的逐像素拓扑摄像头图，支持灵活的摄像头布置和广域操作。分层规划器为每个机器人选择摄像头序列，并规划其在每个视角下的图像空间运动，通过优先-再联合的策略协调机器人，将重叠的摄像头区域视为由一个机器人一次性占用的共享资源，以防止碰撞和死锁。我们在一个真实仓库中验证了该方法，使用四个机器人和30个摄像头覆盖六条27米的过道，报告了任务时间和协调统计数据。据我们所知，这是首次在仅使用外部摄像头网络和离线计算的情况下进行多机器人规划与协调的现场演示，且机器人未携带任何特定任务的导航硬件。
View on arXiv Download PDF AI Translation
cs.RO / 11 / 2606.06790

Learning All-Terrain Locomotion for a Planetary Rover with Actively Articulated Suspension

具有主动关节悬挂的行星探测车全地形运动学习
Bouton, Arthur, Hasseler, Tristan D., Paton, Michael, Brown, Travis, Levy, Jacob, Reid, William, Martin, Joshua, Nayar, Hari
Abstract
This paper presents ERNEST, a four-wheeled planetary rover concept equipped with a two-degree-of-freedom Active Gimbal Suspension that combines yaw and roll actuation to enable wheel reconfiguration, steering, and active load redistribution. A single neural network controller, trained to track a desired path across challenging terrain, fully unlocks the capabilities of this actuated suspension system for autonomous obstacle negotiation. A reinforcement learning framework is developed using the high-fidelity DARTS simulation engine, which combines rigid-contact dynamics and Bekker-Wong terramechanics, enabling the emergence of locomotion strategies adapted to loose-soil conditions. To obtain a single unified controller across heterogeneous terrains, a policy consolidation strategy merges the experience of terrain-specialized agents into one neural network, eliminating the need for explicit terrain classification and controller switching. The resulting controller operates on a combination of proprioceptive and exteroceptive feedback, including sparse stereo-derived terrain elevation, chassis attitude, joint states, and force-torque measurements. Zero-shot transfer to the physical rover is achieved through domain randomization, sensor noise injection, and model-to-real system identification. Experimental results demonstrate autonomous traversal of rock fields, a bump trap, a wheel-high step, sand ripples, and sandy slopes. On a 20{\deg} sandy slope, the learned controller reduces the cost of transport by 37% on dry sand despite the additional actuation, and achieves superior performance on wet sand where the passive suspension becomes completely immobilized.
Chinese Translation
本文提出了ERNEST，一个四轮行星探测车概念，配备了一个具有两个自由度的主动万向悬挂系统，该系统结合了偏航和滚动驱动，以实现轮子重配置、转向和主动负载重分配。一个单一的神经网络控制器经过训练，能够在具有挑战性的地形上跟踪期望路径，充分发挥该驱动悬挂系统在自主障碍物协商中的能力。我们开发了一个强化学习框架，使用高保真度的DARTS仿真引擎，该引擎结合了刚性接触动力学和Bekker-Wong土壤力学，使得适应松散土壤条件的运动策略得以出现。为了在异质地形上获得一个统一的控制器，策略整合策略将地形专业代理的经验合并为一个神经网络，消除了显式地形分类和控制器切换的需要。最终的控制器基于本体感知和外部感知反馈的组合进行操作，包括稀疏的立体图像生成的地形高度、底盘姿态、关节状态和力-扭矩测量。通过领域随机化、传感器噪声注入和模型到真实系统识别，实现了对物理探测车的零样本迁移。实验结果表明，能够自主穿越岩石场、障碍陷阱、轮高台阶、沙丘和沙坡。在20°的沙坡上，学习到的控制器在干沙上将运输成本降低了37%，尽管增加了驱动，但在湿沙上表现出更优越的性能，而被动悬挂则完全失去作用。
View on arXiv Download PDF AI Translation
cs.RO / 12 / 2606.06805

Lane Change Trajectory Planning for Personalized Driving Comfort and Mobility Efficiency

个性化驾驶舒适性与移动效率的变道轨迹规划
Dong, Haoxuan, Li, Dongjun, Song, Ziyou
Abstract
Lane changing entails simultaneous longitudinal and lateral motions that affect driving comfort and mobility efficiency. Because these motions are tightly coupled and subject to substantial inter-vehicle variability, trajectory planning for lane-change maneuvers is characterized by a highly personalized nature. This study proposes a neural network-driven planner that integrates a third-order polynomial trajectory generator with a learning module that infers optimal trajectory parameters across diverse driving conditions. Using a shared backbone with dual heads, one head ensures all-condition operational guarantees, while the other captures driver-specific preferences for comfort or mobility efficiency. A head-gated switching mechanism, realized through a statistical gate based on error-winner logistic regression, adaptively selects the appropriate head under varying driving conditions, which enables context-aware lane-change trajectory planning. Representative cases and Monte Carlo simulations show that the proposed planner achieves personalized comfort and mobility during lane changes, while the baseline ensures feasible trajectories under driving conditions where personalized data are insufficient or inaccessible.
Chinese Translation
变道涉及同时进行的纵向和横向运动，这些运动会影响驾驶舒适性和移动效率。由于这些运动紧密耦合并受到车辆间显著变异性的影响，变道操作的轨迹规划具有高度个性化的特征。本研究提出了一种基于神经网络的规划器，该规划器将三阶多项式轨迹生成器与一个学习模块相结合，该模块能够在不同驾驶条件下推断最佳轨迹参数。该规划器采用共享骨干网络，具有双头结构，其中一个头确保在所有条件下的操作保障，而另一个头则捕捉驾驶者对舒适性或移动效率的特定偏好。通过基于误差胜者逻辑回归的统计门实现的头部门控切换机制，在不同驾驶条件下自适应选择合适的头部，从而实现上下文感知的变道轨迹规划。代表性案例和蒙特卡洛仿真表明，所提出的规划器在变道过程中实现了个性化的舒适性和移动效率，而基线方法在个性化数据不足或不可获取的驾驶条件下确保了可行的轨迹。
View on arXiv Download PDF AI Translation
cs.RO / 13 / 2606.06829

Three-dimensional hydro-cluttered locomotion by an undulatory robot

一种波动机器人在三维水下杂物环境中的运动
Wang, Tianyu, Fernandez, Matthew, Tunnicliffe, Galen, Cornell, Nikolas, Duong, Justin, Dortilus, Donoven, Xu, Zhaochen J., Meza, Patricia, Lublinsky, Sean, Parikh, Darsh, Lin, Jianfeng, Grace, Emily, Goldman, Daniel I.
Abstract
Aquatic robots have expanded human access to underwater environments, yet many underwater spaces contain obstacles that can disrupt open-water locomotion. In "hydro-cluttered" environments, water is interspersed with rigid and flexible clutter, making body-obstacle contact unavoidable. Operating in these spaces requires robots that can regulate and exploit contact, but this regime remains difficult to model or simulate. Building on recent advances in mechanical intelligence in terradynamically capable limbless robotics, we develop principles for 3D aquatic locomotion using AquaMILR, an elongate limbless robot that combines bilateral cable-driven actuation, programmable body compliance, distributed depth regulation, corrosion-resistant enclosures, and onboard power and electronics for untethered field operation. Systematic robophysical experiments reveal that programmable body compliance regulates body deformation and converts body-environment interactions into fast, robust, forward progression across increasing hydro-clutter constraint strength. Depth regulation provides three-dimensional access, allowing the robot to bypass clutter, recover from obstruction, and continue through otherwise inaccessible routes. In potential jamming scenarios, emergent inertia-induced rolling acts as a spontaneous recovery mechanism, freeing the robot from clutter that would otherwise lead to failure and allowing locomotion to continue without additional control. Tests of the robot in an aquatic mangrove field demonstrate that these principles transfer to practical operation, enabling navigation and onboard visual inspection of inaccessible root zones. These results establish principles for hydro-cluttered locomotion and a design paradigm in which aquatic robots exploit environmental complexity as a locomotor resource.
Chinese Translation
水下机器人扩展了人类对水下环境的访问，但许多水下空间中存在障碍物，这些障碍物会干扰开放水域的运动。在“水下杂物”环境中，水与刚性和柔性杂物交错，使得身体与障碍物的接触不可避免。在这些空间中操作需要能够调节和利用接触的机器人，但这种机制仍然难以建模或模拟。在近期无肢机器人机械智能的进展基础上，我们开发了使用AquaMILR（一种细长的无肢机器人）的三维水下运动原理，该机器人结合了双边电缆驱动的驱动、可编程的身体顺应性、分布式深度调节、耐腐蚀外壳以及用于无绳现场操作的机载电源和电子设备。系统的机器人物理实验表明，可编程的身体顺应性调节身体变形，并将身体与环境的交互转化为在不断增加的水下杂物约束强度下的快速、稳健的前进。深度调节提供了三维访问，使机器人能够绕过杂物、从障碍中恢复并继续通过原本无法通行的路线。在潜在的卡滞场景中，出现的惯性诱导滚动作为自发恢复机制，解放了机器人，使其摆脱本可能导致失败的杂物，并允许在没有额外控制的情况下继续运动。在水下红树林区域对机器人的测试表明，这些原理可以转化为实际操作，使得机器人能够导航并对无法接触的根区进行机载视觉检查。这些结果建立了水下杂物运动的原理，并提出了一种设计范式，使水下机器人能够利用环境复杂性作为运动资源。
View on arXiv Download PDF AI Translation
cs.RO / 14 / 2606.06832

STRIPS-WM: Learning Grounded Propositional STRIPS-style World Models from Images

STRIPS-WM：从图像中学习基于图像的命题STRIPS风格世界模型
Ajith, Abhiroop, Chamzas, Constantinos
Abstract
Robots performing long-horizon visual manipulation observe high-dimensional images, but successful plans depend on action-relevant facts: what can be done now and what changes afterward. A useful planning representation should discard irrelevant visual details while preserving action applicability and effects. Classical task planners exploit this structure through symbolic operators with preconditions and effects, but obtaining such representations from raw visual experience remains challenging. We study a visual task-planning setting in which a robot receives only image transitions: the current image, executed high-level action, and the resulting image. At test time, given a start image and a goal image, the robot must produce a sequence of high-level actions that reaches the goal. To address this problem, we introduce STRIPS-WM, a framework for learning image-grounded STRIPS-style world models directly from visual transitions. STRIPS-WM first induces a finite abstract transition graph from images, then learns latent binary predicates and one grounded propositional operator per action label. The learned operators form a symbolic action model with sparse preconditions and add/delete effects. Finally, the learned predicates are distilled into a visual encoder, enabling classical planning directly from novel start and goal images. Experiments on visual rearrangement tasks show that STRIPS-WM improves image-to-plan success over the tested visual rollout, latent graph-search and latent-symbolic baselines.
Chinese Translation
执行长时间视觉操作的机器人观察高维图像，但成功的计划依赖于与动作相关的事实：现在可以做什么以及之后会发生什么变化。一个有用的规划表示应当丢弃无关的视觉细节，同时保留动作的适用性和效果。经典任务规划器通过具有前提条件和效果的符号操作符利用这种结构，但从原始视觉经验中获取此类表示仍然具有挑战性。我们研究了一种视觉任务规划设置，其中机器人仅接收图像转变：当前图像、执行的高层次动作和结果图像。在测试时，给定起始图像和目标图像，机器人必须生成一系列高层次动作以达到目标。为了解决这个问题，我们引入了STRIPS-WM，一个直接从视觉转变中学习图像基础的STRIPS风格世界模型的框架。STRIPS-WM首先从图像中诱导出有限的抽象转变图，然后为每个动作标签学习潜在的二元谓词和一个基于图像的命题操作符。所学习的操作符形成了一个具有稀疏前提条件和添加/删除效果的符号动作模型。最后，所学习的谓词被提炼为一个视觉编码器，使得能够直接从新的起始图像和目标图像进行经典规划。在视觉重排任务上的实验表明，STRIPS-WM在图像到计划的成功率上优于测试的视觉展开、潜在图搜索和潜在符号基线。
View on arXiv Download PDF AI Translation
cs.RO / 15 / 2606.06836

Think Like a Pilot: Fine-Grained Long-Horizon UAV Navigation

像飞行员一样思考：细粒度长时间范围无人机导航
Zheng, Xiangyi, Wang, Xiangyu, Liao, Qinan, Tang, Zimu, Liao, Yue, Lyu, Dongyue, Wang, Guodong, Liu, Junjie, Liu, Si
Abstract
Language-guided UAV agents must execute long-horizon semantic instructions while producing smooth, physically feasible continuous flight commands, yet existing Vision-Language Navigation (VLN) benchmarks typically use discrete or coarse actions and existing UAV Vision-Language-Action (VLA) tasks focus on short, atomic maneuvers. To address this gap in UAV task settings, we introduce \textbf{FLIGHT}, a \textbf{F}ine-grained \textbf{L}ong-horizon \textbf{I}nstruction-\textbf{G}uided benchmark for \textbf{H}ybrid UAV navigation and reasoning \textbf{T}asks, which combines multi-stage instructions with dense 6-DoF trajectory annotations across two dataset splits: Fine-grained VLN and Long-horizon Flow. To endow the UAV agent with the capability of real-time in-flight reasoning over task execution status and mission planning, while simultaneously accommodating high-frequency, real-time precise control, we further propose \textbf{FLIGHT VLA}, an asynchronous architecture that decouples a low-frequency Streaming Pilot Vision-Language Model (VLM) for task-state reasoning from a high-frequency diffusion action model for continuous control, supervised by explicit \textbf{Pilot Reasoning} texts that summarize the current flight state and anticipate the next subgoal. In closed-loop evaluation, FLIGHT VLA consistently surpasses representative VLN and VLA baselines on our FLIGHT benchmarks, achieving stronger multi-stage completion, subgoal adherence, and terminal control. Its trained Streaming Pilot Reasoning VLM further improves UAV video reasoning, validating the effectiveness of our design.
Chinese Translation
语言引导的无人机代理必须执行长时间范围的语义指令，同时生成平滑、物理可行的连续飞行指令。然而，现有的视觉-语言导航（Vision-Language Navigation, VLN）基准通常使用离散或粗略的动作，而现有的无人机视觉-语言-动作（Vision-Language-Action, VLA）任务则集中于短期、原子化的操作。为了解决无人机任务设置中的这一空白，我们提出了 extbf{FLIGHT}，一个 extbf{F}ine-grained extbf{L}ong-horizon extbf{I}nstruction- extbf{G}uided基准，用于 extbf{H}ybrid无人机导航和推理任务，结合了多阶段指令和密集的6自由度轨迹注释，分为两个数据集拆分：细粒度VLN和长时间范围流。为了赋予无人机代理实时飞行推理的能力，监控任务执行状态和任务规划，同时适应高频、实时精确控制，我们进一步提出了 extbf{FLIGHT VLA}，一种异步架构，将低频流媒体飞行员视觉-语言模型（Vision-Language Model, VLM）用于任务状态推理与高频扩散动作模型用于连续控制解耦，并由明确的 extbf{Pilot Reasoning}文本监督，这些文本总结当前飞行状态并预测下一个子目标。在闭环评估中，FLIGHT VLA在我们的FLIGHT基准上始终超越代表性的VLN和VLA基线，取得更强的多阶段完成、子目标遵循和终端控制。其训练的流媒体飞行员推理VLM进一步改善了无人机视频推理，验证了我们设计的有效性。
View on arXiv Download PDF AI Translation
cs.RO / 16 / 2606.06870

What Is My Robot Thinking? Design Considerations for Transparent and Trustworthy Shared Autonomy

我的机器人在想什么？透明和可信共享自主设计考虑
Belsare, Atharv, Karimi, Zohre, Mattson, Connor, Nakka, Rushiil, Brown, Daniel S.
Abstract
Assistive robots operating under shared autonomy must balance user control with autonomous assistance. Because robot actions depend on internal intent inference that is not directly observable, mismatches between inferred and intended goals can undermine coordination and trust. We investigate how interface-level transparency, including feedback modality (visual vs. auditory) and information richness (sparse vs. rich), shapes interaction in a vision-based shared autonomy system. In a user study with N=25 participants across two assistive manipulation tasks, we evaluate how these designs influence coordination and trust. Providing feedback significantly improves intent alignment and reduces corrective intervention, indicating that making the inferred goal legible accelerates convergence in shared control. Participants preferred visual over auditory feedback, while preferences for sparse versus rich information depended on task complexity. We also found that revealing the full belief distribution did not consistently improve alignment or trust. Together, these findings indicate that effective transparency enhances coordination primarily through goal legibility, while trust depends on task-appropriate information exposure rather than maximal disclosure. Based on these results, we outline guidelines for designing transparent shared autonomy systems.
Chinese Translation
在共享自主下运行的辅助机器人必须平衡用户控制与自主辅助之间的关系。由于机器人行为依赖于内部意图推断，而这种推断并不可直接观察，因此推断目标与预期目标之间的不匹配可能会削弱协调性和信任。我们研究了界面级透明度如何影响基于视觉的共享自主系统中的交互，包括反馈方式（视觉与听觉）和信息丰富性（稀疏与丰富）。在一项包含25名参与者的用户研究中，我们评估了这些设计如何影响协调性和信任。提供反馈显著改善了意图对齐，并减少了纠正干预，表明使推断目标清晰可见加速了共享控制中的收敛。参与者更偏好视觉反馈而非听觉反馈，而对稀疏信息与丰富信息的偏好则取决于任务复杂性。我们还发现，揭示完整的信念分布并未始终改善对齐或信任。综合这些发现表明，有效的透明度主要通过目标可读性增强协调性，而信任则依赖于适合任务的信息暴露，而非最大程度的披露。基于这些结果，我们概述了设计透明共享自主系统的指导方针。
View on arXiv Download PDF AI Translation
cs.RO / 17 / 2606.06877

Neuro-Symbolic Learning for Long-Horizon Task Planning Under Complex Logical Constraints

复杂逻辑约束下的长时间任务规划的神经符号学习
Du, Qiwei, Zhan, Zitong, Su, Shaoshu, Li, Bowen, Du, Yi, Zhao, Zhipeng, Fu, Taimeng, Scherer, Sebastian, Li, Jiaoyang, Wang, Chen
Abstract
Task planning often suffers from severe efficiency bottlenecks when robots must reason over long-horizon action sequences under complex logical constraints, including object affordances, spatial relationships, and sequential action dependencies. Recent neuro-symbolic methods improve planning efficiency by learning object-importance scores to prune task-irrelevant objects, but they typically rely on fixed offline supervision generated from full search spaces. This creates a train-test mismatch: at deployment, the planner operates in pruned search spaces induced by the model's own imperfect predictions, leading to exposure bias and degraded planning performance. To address this challenge, we formulate object-importance learning for task planning as an imperative learning-based bilevel optimization problem. The upper level optimizes a neural scorer, while the lower level solves a symbolic planning problem in the score-pruned search space. To stabilize this learning process, we introduce a 3R strategy into the lower-level planning, using parallel Repair, Restart, and Rollback recovery to provide reliable and adaptive feedback for upper-level learning. Experiments on three challenging benchmarks demonstrate state-of-the-art performance, including an 80.04% reduction in failure rate and a 57.14% reduction in planning time. We further validate the framework on a quadruped-based mobile manipulator in simulation and the real world, demonstrating its potential for efficient and deployable neuro-symbolic task planning.
Chinese Translation
在复杂逻辑约束下，任务规划通常在机器人必须对长时间动作序列进行推理时面临严重的效率瓶颈，这些约束包括物体可用性、空间关系和顺序动作依赖关系。最近的神经符号方法通过学习物体重要性评分来提高规划效率，从而修剪与任务无关的物体，但它们通常依赖于从完整搜索空间生成的固定离线监督。这导致了训练与测试之间的不匹配：在部署时，规划器在由模型自身不完美预测引发的修剪搜索空间中运行，导致暴露偏差和规划性能下降。为了解决这一挑战，我们将任务规划中的物体重要性学习形式化为一个基于学习的双层优化问题。上层优化一个神经评分器，而下层在评分修剪的搜索空间中解决符号规划问题。为了稳定这一学习过程，我们在下层规划中引入了3R策略，使用并行的修复、重启和回滚恢复，为上层学习提供可靠和自适应的反馈。在三个具有挑战性的基准测试上的实验表明了其最先进的性能，包括80.04%的失败率降低和57.14%的规划时间减少。我们进一步在仿真和现实世界中验证了该框架在基于四足机器人移动操控器上的应用，展示了其在高效和可部署的神经符号任务规划中的潜力。
View on arXiv Download PDF AI Translation
cs.RO / 18 / 2606.06878

A Cross-view Fusion Framework for Robust 6-DoF Grasp Pose Estimation

一种用于鲁棒6自由度抓取姿态估计的跨视图融合框架
Zhu, Kangjian, Jiang, Haobo, Qian, Jianjun, Xie, Jin
Abstract
In this paper, we propose a cross-view fusion framework that enhances the robustness of 6-DoF grasp pose estimation in corner views. Our framework alleviates occlusion by incorporating an auxiliary view and avoids the time-consuming, task-agnostic multi-view reconstruction through a post-fusion strategy. To enhance cross-view fusion, we propose a self-supervised contrastive learning strategy that leverages cross-view associations to regularize point cloud features. In brief, a cross-view point pair is considered a match if the two points correspond to the same 3D location, and a non-match if they represent distinct grasp directions. The learning strategy significantly enhances the spatial consistency and direction distinctiveness of point features, thereby facilitating cross-view fusion and improving estimation robustness. Furthermore, we propose a cross-view-aligned cylinder integration module to fuse grasp-relevant geometry into a comprehensive representation. Specifically, the module first aligns the cross-view points and features according to their similarity to enhance the robustness against noise. Subsequently, these points are registered into the cylindrical coordinate frame, emphasizing the rotation-symmetric geometry which is important for grasping. Finally, local self-attention and seed cross-attention layers are alternately employed, respectively enabling interactions within single views and across views, which supports fine-grained representation of grasp-relevant geometry. Our framework achieves strong performance on the GraspNet-1Billion benchmark and in real-world applications. Code is available at https://github.com/KJZhuAutomatic/Cross-view-Grasp.
Chinese Translation
在本文中，我们提出了一种跨视图融合框架，以增强角落视图中6自由度抓取姿态估计的鲁棒性。我们的框架通过引入辅助视图来缓解遮挡，并通过后融合策略避免了耗时的与任务无关的多视图重建。为了增强跨视图融合，我们提出了一种自监督对比学习策略，该策略利用跨视图关联来规范点云特征。简而言之，若两个点对应于同一3D位置，则视为匹配的跨视图点对；若它们表示不同的抓取方向，则视为不匹配。该学习策略显著增强了点特征的空间一致性和方向独特性，从而促进了跨视图融合并提高了估计的鲁棒性。此外，我们提出了一种跨视图对齐的圆柱体集成模块，将与抓取相关的几何信息融合成一个综合表示。具体而言，该模块首先根据点和特征的相似性对跨视图点进行对齐，以增强对噪声的鲁棒性。随后，这些点被注册到圆柱坐标系中，强调了抓取中重要的旋转对称几何。最后，局部自注意力层和种子跨注意力层交替使用，分别实现单视图内和跨视图间的交互，从而支持与抓取相关几何的细粒度表示。我们的框架在GraspNet-1Billion基准测试和实际应用中表现出色。代码可在 https://github.com/KJZhuAutomatic/Cross-view-Grasp 获取。
View on arXiv Download PDF AI Translation
cs.RO / 19 / 2606.06904

ActionMap: Robot Policy Learning via Voxel Action Heatmap

ActionMap：通过体素动作热图进行机器人策略学习
Yang, Pei, Ci, Hai, Chen, Yanzhe, Lv, Qi, Cai, Han, Shou, Mike Zheng
Abstract
Vision-language-action (VLA) models have advanced rapidly across backbones, training recipes, and data scale, yet the action decoder, which converts the backbone's hidden state into a continuous control signal, has barely changed and remains a single-point predictor across the majority of current VLAs. Whether implemented via autoregressive token bins, L1 regression, or flow-matching denoising, the resulting decoder treats the action space as unstructured, leaving the geometric proximity of neighboring actions unexploited during training. To advance this, we introduce ActionMap, a voxel heatmap action head that drops into an existing VLA in place of its native action decoder. For each new action, the head predicts a voxel heatmap over the action space, where each voxel directly stores the probability of the corresponding action. Across LIBERO simulation and real-world Franka manipulation, our heatmap head surpasses two architecturally distinct backbones at matched training steps (e.g., +8.2% over OpenVLA-OFT's L1 regression head on the LIBERO four-suite average), converges at comparable or faster rates on both backbones, and remains markedly more data-efficient at low training data. The cross-backbone consistency indicates that action representation is a real lever for VLA performance, distinct from further backbone or recipe scaling. Project Page: https://github.com/showlab/ActionMap.
Chinese Translation
视觉-语言-动作（VLA）模型在骨干网络、训练方案和数据规模方面迅速发展，然而，动作解码器几乎没有变化，仍然在大多数当前的VLA中作为单点预测器存在。无论是通过自回归令牌箱、L1回归还是流匹配去噪实现，最终的解码器都将动作空间视为无结构的，在训练过程中未能利用相邻动作的几何邻近性。为此，我们提出了ActionMap，一种体素热图动作头，替代现有VLA中的原生动作解码器。对于每个新动作，该头在动作空间上预测一个体素热图，其中每个体素直接存储对应动作的概率。在LIBERO仿真和真实世界的Franka操作中，我们的热图头在匹配训练步骤的情况下超越了两个架构上不同的骨干网络（例如，在LIBERO四套件平均上，相比OpenVLA-OFT的L1回归头提高了8.2%），在两个骨干网络上以可比或更快的速度收敛，并在低训练数据下显著更具数据效率。跨骨干网络的一致性表明，动作表示是提升VLA性能的真正杠杆，与进一步的骨干网络或方案扩展不同。项目页面：https://github.com/showlab/ActionMap。
View on arXiv Download PDF AI Translation
cs.RO / 20 / 2606.06944

T-GMP: Terrain-conditioned Generative Motion Priors for Versatile and Natural Humanoid Locomotion

T-GMP：基于地形条件的生成运动先验用于多样化和自然的人形运动
Guo, Junhong, Hu, Hao, Chen, Chen, Han, Haoxuan, Gong, Linao, Yang, Xin, He, Zhicheng, Su, Yao, He, Fenghua
Abstract
Achieving both anthropomorphic naturalness and robust terrain traversal remains a fundamental challenge in humanoid locomotion. Existing Reinforcement Learning (RL) approaches typically rely on fixed motion priors, limiting their adaptability to varying environments. We propose Terrain-conditioned Generative Motion Priors (T-GMP), a module that captures a terrain-conditioned latent motion manifold from a few expert state-terrain demonstrations using a Conditional Variational Autoencoder (CVAE). The learned priors enable smooth style transitions, facilitating a unified policy that adapts to terrain variations. We integrate T-GMP into an adversarial learning pipeline with our proposed Foothold Penalty, where a discriminator dynamically modulates naturalness constraints conditioned on local terrain features, guiding the generation of versatile and human-like motions. Experimental results demonstrate that our method outperforms existing baselines in traversal success rate and motion smoothness, while preserving biomimetically natural and physically coordinated motions.
Chinese Translation
实现人形运动的类人自然性和稳健的地形穿越仍然是一个基本挑战。现有的强化学习（Reinforcement Learning, RL）方法通常依赖于固定的运动先验，限制了它们对不同环境的适应能力。我们提出了基于地形条件的生成运动先验（Terrain-conditioned Generative Motion Priors, T-GMP），该模块利用条件变分自编码器（Conditional Variational Autoencoder, CVAE）从少量专家状态-地形示例中捕捉地形条件的潜在运动流形。学习到的先验使得风格过渡平滑，促进了一个能够适应地形变化的统一策略。我们将T-GMP集成到一个对抗学习管道中，并提出了足迹惩罚（Foothold Penalty），其中鉴别器动态调节基于局部地形特征的自然性约束，引导生成多样化和类人的运动。实验结果表明，我们的方法在穿越成功率和运动平滑性方面优于现有基线，同时保持生物仿真自然和物理协调的运动。
View on arXiv Download PDF AI Translation
cs.RO / 21 / 2606.06953

LIMMT: Less is More for Motion Tracking

LIMMT：运动跟踪中的少即是多
Guan, Yu, Qi, Zekun, Lin, Chenghuai, Chen, Xuchuan, Liu, Dairu, Zhang, Wenyao, Wang, Jilong, Yu, Xinqiang, Wang, He, Yi, Li
Abstract
We argue that high-quality motion data can steer tracking policies toward better optimization trajectories early in training. In this work, we introduce LIMMT (Less Is More for Motion Tracking). To our knowledge, this is the first data-centric study for physics-based humanoid motion tracking. We go beyond simply removing low-quality and erroneous clips, but define motion data quality through three dimensions: physics feasibility, diversity, and complexity. We show that even training with under 3% of AMASS yields better tracking performance than training with the full dataset. We further conduct data cleaning on the estimated web-sourced mocap data. Extensive experiments and analyses validate the effectiveness of our framework.
Chinese Translation
我们认为，高质量的运动数据可以在训练早期引导跟踪策略朝着更好的优化轨迹发展。在本研究中，我们引入了LIMMT（Less Is More for Motion Tracking）。据我们所知，这是针对基于物理的人形运动跟踪的首个数据中心研究。我们不仅仅是简单地去除低质量和错误的片段，而是通过三个维度来定义运动数据质量：物理可行性、多样性和复杂性。我们展示了即使使用不到3%的AMASS数据进行训练，其跟踪性能也优于使用完整数据集的训练。我们进一步对估计的网络来源动作捕捉数据进行了数据清理。大量实验和分析验证了我们框架的有效性。
View on arXiv Download PDF AI Translation
cs.RO / 22 / 2606.06977

Compliance-Based Sensor Placement for Force Sensing on a Sensorized Prostate Phantom

基于合规性的传感器布置方法用于传感化前列腺模型的力传感
Tian, Sizhe, Adagolodjo, Yinoussa, Dequidt, Jeremie
Abstract
This work presents a compliance-based sensor placement method for force sensing on a sensorized prostate phantom designed for Digital Rectal Examination training. The phantom combines three internal pneumatic chambers, used as intrinsic pressure sensors, with ten surface displacement markers. A finite-element simulation dataset is generated by applying external forces at sampled surface locations, from which a compliance matrix relating force inputs to pressure and displacement responses is constructed. Based on this matrix, we propose a weighted greedy selection strategy that maximizes local force reconstructability while prioritizing the clinically relevant posterior contact region and avoiding marker placement directly within the Region of Interest. Compared with a global QR-based placement strategy, the proposed method increases the mean reconstructability score in the target region by 22.5%. These results suggest that region-aware sparse sensor placement can improve force observability in soft robotic medical phantoms while maintaining a limited and practical sensing configuration.
Chinese Translation
本研究提出了一种基于合规性的传感器布置方法，用于在为数字直肠检查培训设计的传感化前列腺模型上进行力传感。该模型结合了三个内部气动腔室，作为内在压力传感器，以及十个表面位移标记。通过在采样的表面位置施加外力，生成有限元仿真数据集，从中构建出一个将力输入与压力和位移响应相关联的合规矩阵。基于该矩阵，我们提出了一种加权贪婪选择策略，旨在最大化局部力重构能力，同时优先考虑临床相关的后接触区域，并避免在感兴趣区域内直接布置标记。与基于全局QR的布置策略相比，所提方法在目标区域内将平均重构能力评分提高了22.5%。这些结果表明，区域感知的稀疏传感器布置可以在保持有限且实用的传感配置的同时，提高软体机器人医疗模型中的力可观测性。
View on arXiv Download PDF AI Translation
cs.RO / 23 / 2606.06996

Mission-Level Runtime Assurance Framework for Autonomous Driving

自主驾驶的任务级运行保障框架
Tsai, Chieh, Hariri, Salim
Abstract
This paper studies runtime safety for autonomous driving when high-level driving commands become faulty or unreliable. Unlike conventional runtime-safety approaches that mainly focus on immediate vehicle safety, the proposed framework evaluates both driving safety and whether the vehicle can still successfully complete its mission before a command is executed. The framework extends highway-env with mission-level fault scenarios such as skipping required checkpoints, entering restricted areas, and generating future routes that can no longer complete the mission successfully. A runtime monitoring system is introduced to detect and reject unsafe or mission-infeasible commands before execution. For comparison, an adapted Simplex-Drive runtime-safety baseline with learning-based driving control, safety fallback control, and runtime controller switching is implemented using the public Simplex-Drive framework. Experimental results show that platform-level runtime safety alone cannot detect mission-level planning faults, while the proposed framework successfully rejects mission-infeasible commands and improves mission success under randomized fault conditions.
Chinese Translation
本文研究了在高层驾驶指令出现故障或不可靠时，自主驾驶的运行安全性。与传统的运行安全方法主要关注车辆的即时安全不同，所提出的框架在执行指令之前评估驾驶安全性以及车辆是否仍能成功完成其任务。该框架扩展了 highway-env，涵盖了任务级故障场景，例如跳过必要的检查点、进入限制区域以及生成无法成功完成任务的未来路线。引入了一种运行监控系统，以在执行之前检测并拒绝不安全或任务不可行的指令。为了进行比较，采用了基于公共 Simplex-Drive 框架的适应性 Simplex-Drive 运行安全基线，结合了基于学习的驾驶控制、安全回退控制和运行控制器切换。实验结果表明，仅依靠平台级运行安全无法检测到任务级规划故障，而所提出的框架能够成功拒绝任务不可行的指令，并在随机故障条件下提高任务成功率。
View on arXiv Download PDF AI Translation
cs.RO / 24 / 2606.07012

Task Editing for Generalizable 3D Visuomotor Policy Learning

用于可推广的3D视觉运动策略学习的任务编辑
Jiang, Jian-Jian, Yang, YiHan, Wei, Lan, Luo, Yuming, Wu, Xiao-Ming, Chen, Xuhang, Fan, Bin, Zhang, Dandan, Zheng, Wei-Shi
Abstract
3D visuomotor policies offer a promising direction for complex robotic manipulation, as depth maps and point clouds provide rich geometric information for spatial reasoning. However, their success often depends on large-scale real-world demonstrations, which are costly and time-consuming to collect. To this end, existing methods commonly use demonstration generation strategies to improve data efficiency by applying object-centric transformations to human-collected demonstrations, such as varying object poses or scales. While effective for local variation, these transformations largely preserve the original scene structure and skill sequence, limiting their ability to synthesize diverse scene-skill-object combinations for complex tasks. In this paper, we propose Task-Edit, a novel demonstration generation framework that generates diverse trajectories from a task-centric editing perspective. The key insight of Task-Edit is to decompose a task into scene, skill and object components, and flexibly recombine them. In this way, Task-Edit enables scalable demonstration generation and significantly improves generalization for long-horizon manipulation tasks. We evaluate Task-Edit through extensive real-world experiments and demonstrate three advantages: (1) Effectiveness: Task-Edit significantly improves 3D visuomotor policies across various real-world tasks and robot embodiments. (2) Generalizability: Task-Edit improves model generalization across different scenario setups. (3) Applicability: Task-Edit enables models to handle scenarios that are difficult to collect in the real world, including disturbance resistance, obstacle avoidance and unseen cluttered scenes.
Chinese Translation
3D视觉运动策略为复杂的机器人操作提供了一个有前景的方向，因为深度图和点云为空间推理提供了丰富的几何信息。然而，它们的成功往往依赖于大规模的现实世界演示，这些演示的收集成本高且耗时。为此，现有的方法通常采用演示生成策略，通过对人类收集的演示应用以物体为中心的变换来提高数据效率，例如改变物体的姿态或尺度。虽然这些变换在局部变化方面有效，但它们在很大程度上保留了原始场景结构和技能序列，限制了它们合成复杂任务中多样化场景-技能-物体组合的能力。在本文中，我们提出了Task-Edit，一个新颖的演示生成框架，从任务中心的编辑视角生成多样化的轨迹。Task-Edit的关键洞察是将任务分解为场景、技能和物体组件，并灵活地重新组合它们。通过这种方式，Task-Edit实现了可扩展的演示生成，并显著提高了长时间操作任务的泛化能力。我们通过广泛的现实世界实验评估Task-Edit，并展示了三个优势：（1）有效性：Task-Edit显著改善了各种现实世界任务和机器人形态下的3D视觉运动策略。（2）可推广性：Task-Edit提高了模型在不同场景设置下的泛化能力。（3）适用性：Task-Edit使模型能够处理在现实世界中难以收集的场景，包括抗干扰、避障和未见的杂乱场景。
View on arXiv Download PDF AI Translation
cs.RO / 25 / 2606.07013

A Multi-Operator Mixed-Reality Interface for Multi-Robot Control and Coordination: Co-Located and Private Workspace Collaboration

多操作员混合现实界面用于多机器人控制与协调：共处与私人工作空间协作
Adekoya, Omotoye Shamsudeen, Sgorbissa, Antonio, Recchiuto, Carmine Tommaso
Abstract
Multi-operator control of robot teams requires not only access to the same mission information, but also mechanisms for maintaining shared awareness and preventing conflicting interventions. Building on our previous HORUS interface (Holistic Operational Reality for Unified Systems) we present a mixed-reality interface that extends single-operator multi-robot supervision to collaborative multi-operator use. The system supports two complementary modes: a co-located shared workspace, in which operators observe and manipulate the same mini-map in the same physical location, and a private-workspace mode, in which operators work on the same mission through independently placed local workspaces. The architecture combines registration-driven scene construction, lightweight shared-session synchronization, and per-robot control leases to support collaborative monitoring, tasking, and teleoperation while preventing conflicting commands. We evaluated the approach in a human-subject study with 36 participants (18 pairs) controlling three Nova Carter mobile robots in two search environments. The performance of the objective task was comparable across the two modes, indicating that both modes supported effective mission execution. However, the co-located shared workspace significantly improved perceived collaboration, shared understanding, and handoff clarity, and was the preferred collaborative mode. These results indicate that physically co-locating the MR workspace improves how operators coordinate even when the underlying robot-control tools remain unchanged.
Chinese Translation
多操作员对机器人团队的控制不仅需要访问相同的任务信息，还需要维持共享意识和防止冲突干预的机制。在我们之前的HORUS界面（Holistic Operational Reality for Unified Systems）的基础上，我们提出了一种混合现实界面，将单操作员多机器人监督扩展到协作多操作员使用。该系统支持两种互补模式：共处共享工作空间，操作员在同一物理位置观察和操作相同的迷你地图；以及私人工作空间模式，操作员通过独立放置的本地工作空间在同一任务上进行工作。该架构结合了基于注册的场景构建、轻量级共享会话同步和每个机器人控制租赁，以支持协作监控、任务分配和远程操作，同时防止冲突命令。我们在一项包含36名参与者（18对）的以人为主体的研究中评估了该方法，参与者控制三台Nova Carter移动机器人在两个搜索环境中执行任务。两种模式下的目标任务表现相当，表明这两种模式都支持有效的任务执行。然而，共处共享工作空间显著提高了感知协作、共享理解和交接清晰度，并且是首选的协作模式。这些结果表明，物理共处的混合现实工作空间改善了操作员的协调方式，即使基础的机器人控制工具保持不变。
View on arXiv Download PDF AI Translation
cs.RO / 26 / 2606.07067

Extending Responsibility-Sensitive Safety for the Assessment of Offloaded Autonomous Driving Services

扩展责任敏感安全性以评估卸载的自主驾驶服务
Dehler, Robin, Thakur, Aryan, Buchholz, Michael
Abstract
Safety is a fundamental requirement in the development of autonomous driving (AD) systems. While function offloading has demonstrated significant benefits in terms of computational efficiency and energy consumption, its application to safety-critical AD functionality introduces new challenges. In particular, offloaded service compositions incur increased and variable response times due to wireless vehicle-to-everything (V2X) communication, which directly affects the vehicle's reaction time and thus its safety guarantees. In this paper, we address this challenge by extending the definitions of Responsibility-Sensitive Safety (RSS) to explicitly account for different response times of local and offloaded AD service compositions. Based on this extension, we propose an integration into function offloading, using the RSS safety constraints for offloading decision-making and fallback mechanisms. Offloaded service compositions are only permitted if the current traffic situation remains safe under the corresponding end-to-end response time. If this condition is violated, the system performs a controlled fallback to local execution. Furthermore, we introduce an enhanced fallback strategy that includes a warm-standby phase for offloaded services, enabling faster and safer transitions from offloaded to local services. The proposed approach is integrated into our AD stack and evaluated in both simulation and the real world. Experimental results demonstrate that the proposed method improves safety compared to state-of-the-art function offloading and safety frameworks, while preserving the benefits of distributed computation when safety conditions allow.
Chinese Translation
安全性是自主驾驶（AD）系统开发的基本要求。尽管功能卸载在计算效率和能耗方面表现出显著的优势，但其在安全关键的AD功能中的应用引入了新的挑战。特别是，由于无线车与一切（V2X）通信，卸载的服务组合会导致响应时间增加且不稳定，这直接影响车辆的反应时间，从而影响其安全保障。在本文中，我们通过扩展责任敏感安全性（RSS）的定义，明确考虑本地和卸载AD服务组合的不同响应时间，来应对这一挑战。基于这一扩展，我们提出将其整合到功能卸载中，利用RSS安全约束进行卸载决策和回退机制。仅当在相应的端到端响应时间下当前交通情况保持安全时，才允许卸载的服务组合。如果违反此条件，系统将执行受控回退到本地执行。此外，我们引入了一种增强的回退策略，包括卸载服务的热备份阶段，使得从卸载服务到本地服务的过渡更快且更安全。所提出的方法已集成到我们的AD堆栈中，并在模拟和现实世界中进行了评估。实验结果表明，与最先进的功能卸载和安全框架相比，所提出的方法在提高安全性的同时，保持了在安全条件允许时的分布式计算优势。
View on arXiv Download PDF AI Translation
cs.RO / 27 / 2606.07083

Predictive Style Matching: Natural and Robust Humanoid Locomotion

预测风格匹配：自然且稳健的人形运动
Nedelchev, Simeon, Chaikovskaia, Ekaterina, Davydenko, Egor, Zaliaev, Eduard, Gorbachev, Roman
Abstract
Reinforcement learning has become the prevailing approach to humanoid locomotion control: policies transfer reliably from simulation to hardware and recover gracefully from disturbances. Motion quality, however, still lags behind: task-only rewards often converge to stiff, asymmetric gaits, while motion imitation methods improve appearance but become more sensitive to external disturbances because reference signals can oppose the transient poses needed to regain balance. We propose Predictive Style Matching, in which an offline predictor maps the robot's lower-body state history and velocity commands to interpretable upper-body joint and gait targets that shape the rewards during training. Because the targets are state-conditioned rather than time-indexed and the predictor is used only at training time, the deployed controller inherits the proprioceptive interface and inference cost of a task-only RL baseline. On the Unitree G1, in both simulation and hardware, PSM reduces upper-body style error by roughly an order of magnitude over task-only RL while preserving its fall-recovery rate, whereas the motion-imitation baseline attains the lowest style error but fails to recover from disturbances about five times as often.
Chinese Translation
强化学习已成为人形运动控制的主要方法：策略能够可靠地从仿真转移到硬件，并能优雅地从干扰中恢复。然而，运动质量仍然滞后：仅基于任务的奖励往往收敛于僵硬且不对称的步态，而运动模仿方法虽然改善了外观，但对外部干扰变得更加敏感，因为参考信号可能与恢复平衡所需的瞬态姿态相抵触。我们提出了预测风格匹配（Predictive Style Matching），其中一个离线预测器将机器人下肢状态历史和速度命令映射到可解释的上肢关节和步态目标，这些目标在训练过程中塑造奖励。由于这些目标是状态条件而非时间索引，并且预测器仅在训练时使用，因此部署的控制器继承了仅基于任务的强化学习基线的本体感知接口和推理成本。在Unitree G1上，无论是在仿真还是硬件中，PSM将上肢风格误差减少了大约一个数量级，同时保持了其跌倒恢复率，而运动模仿基线虽然达到了最低的风格误差，但在恢复干扰方面失败的频率约为五倍。
View on arXiv Download PDF AI Translation
cs.RO / 28 / 2606.07089

Dreaming when Necessary: Advancing World Action Models with Adaptive Multi-Modal Reasoning

必要时的梦想：通过自适应多模态推理推进世界行动模型
Tang, Yinzhou, Xu, Jingbo, Shang, Yu, Song, Zihao, Gao, Chen, Wu, Wei, Li, Yong
Abstract
World Action Models (WAMs) offer a promising approach to embodied intelligence, yet existing methods rely heavily on video prediction as action priors and lack adaptive multimodal reasoning, limiting their effectiveness on long-horizon, complex tasks. We observe that WAMs require different multimodal reasoning modes under different execution contexts: textual reasoning is essential during task transitions to guide high-level action prediction, while visual reasoning is critical during fine-grained manipulation for precise control. Motivated by this observation, we propose \textbf{AdaWAM}, a world action model with adaptive multimodal reasoning abilities. AdaWAM integrates a lightweight dynamic router that autonomously triggers textual or visual reasoning as needed during task execution. Experiments on both simulated and real-world embodied tasks show that AdaWAM substantially improves inference efficiency while outperforming state-of-the-art embodied policies. Codes and demos are available at: https://adawam.github.io/.
Chinese Translation
世界行动模型（WAMs）为具身智能提供了一种有前景的方法，但现有方法过于依赖视频预测作为行动先验，并缺乏自适应多模态推理，限制了其在长时间跨度和复杂任务上的有效性。我们观察到，WAMs在不同执行上下文下需要不同的多模态推理模式：在任务过渡期间，文本推理对于指导高层次的行动预测至关重要，而在精细操作期间，视觉推理对于精确控制至关重要。基于这一观察，我们提出了 extbf{AdaWAM}，一种具有自适应多模态推理能力的世界行动模型。AdaWAM集成了一个轻量级动态路由器，能够在任务执行过程中根据需要自主触发文本或视觉推理。在模拟和现实世界的具身任务上的实验表明，AdaWAM显著提高了推理效率，同时超越了最先进的具身策略。代码和演示可在以下网址获取：https://adawam.github.io/。
View on arXiv Download PDF AI Translation
cs.RO / 29 / 2606.07107

Coarse-to-Control: Action-Token Planning for Vision-Language-Action Models

粗到控：用于视觉-语言-动作模型的动作令牌规划
Wu, Jinhao, Zhang, Shiduo, Liu, Yicheng, Yu, Xiaopeng, Li, Sixian, Wang, Siyin, Zhao, Hang, Huo, Jing, Gao, Yang, Gong, Jingjing, Qiu, Xipeng, Jiang, Yu-Gang
Abstract
Most vision-language-action (VLA) models map observations directly to actions without explicit intermediate planning, which limits performance on long-horizon tasks where early mistakes compound. We propose Coarse-to-Control, a plan-execute VLA that introduces planning natively in the action-token space. The key idea is to let the policy first predict a compact sequence of coarse action tokens that summarize the intended future trajectory, and then generate executable action tokens conditioned on this plan. Because both planning and execution share a unified discrete action vocabulary, the plan stays close to the control manifold and provides directly actionable guidance rather than an abstract hint that must be translated back to motor commands. Experiments on LIBERO, SimplerEnv-WidowX, and real-world manipulation tasks show that action-token planning consistently improves over direct action generation, with the largest gains on long-horizon multi-stage tasks.
Chinese Translation
大多数视觉-语言-动作（VLA）模型直接将观察映射到动作，而没有明确的中间规划，这限制了它们在长时间任务中的表现，因为早期的错误会累积。我们提出了粗到控（Coarse-to-Control），一种在动作令牌空间中原生引入规划的计划-执行VLA。其关键思想是让策略首先预测一系列紧凑的粗略动作令牌，以总结预期的未来轨迹，然后根据该计划生成可执行的动作令牌。由于规划和执行共享统一的离散动作词汇，计划保持接近控制流形，并提供直接可操作的指导，而不是必须转换回运动指令的抽象提示。在LIBERO、SimplerEnv-WidowX和真实世界操作任务上的实验表明，动作令牌规划在直接动作生成上持续取得改进，尤其在长时间多阶段任务中表现出最大的增益。
View on arXiv Download PDF AI Translation
cs.RO / 30 / 2606.07118

QuadVerse: An Integrated Framework Aligning Visual-Physical Reality for Quadruped Simulation

QuadVerse：一个整合框架，用于对齐四足动物仿真的视觉-物理现实
Chen, Yuxiang, Wang, Yuanhao, Zhang, Ziheng, Zhang, Meng, Liu, Yu, Jia, Yufei, Wang, Tiancai, Zhou, Erjin, Xie, Jin
Abstract
Simulation is central to robot learning, yet the sim-to-real gap remains a major bottleneck.Existing approaches often tackle visual or dynamic gaps separately, overlooking how these individual mismatches accumulate and propagate throughout the robot's state evolution.In this paper, we introduce QuadVerse, an integrated framework that uses reconstructed scenes as a calibration substrate for aligning visual perception, physical interaction, and actuator dynamics.From captured RGB videos, we reconstruct geometry-constrained 3D Gaussian Splatting (3DGS) scenes that support batched photorealistic ego-view rendering and collision-ready semantic mesh extraction. The meshes further enable contact calibration by initializing spatially varying friction priors and refining them through trajectory-based posterior search.To address remaining actuator discrepancies, QuadVerse trains a residual dynamics compensator by replaying real-world trajectories on the contact-calibrated terrain, reducing the entanglement between terrain-induced contact errors and actuator non-idealities.Experiments show that QuadVerse improves reconstruction quality and locomotion tracking over relevant baselines.Leveraging this foundation, we demonstrate robust zero-shot visual-navigation policy deployment without task-specific real-world rollouts.
Chinese Translation
仿真在机器人学习中至关重要，但仿真与现实之间的差距仍然是一个主要瓶颈。现有方法通常分别处理视觉或动态差距，忽视了这些个体不匹配如何在机器人状态演变过程中累积和传播。在本文中，我们介绍了QuadVerse，一个整合框架，它利用重建场景作为校准基底，以对齐视觉感知、物理交互和执行器动态。通过捕获的RGB视频，我们重建了几何约束的3D高斯点云（3D Gaussian Splatting，3DGS）场景，支持批量的逼真自我视角渲染和准备碰撞的语义网格提取。这些网格进一步通过初始化空间变化的摩擦先验并通过基于轨迹的后验搜索进行优化，从而实现接触校准。为了解决剩余的执行器差异，QuadVerse通过在接触校准的地形上重放真实世界的轨迹来训练残差动态补偿器，减少地形引起的接触误差与执行器非理想性之间的纠缠。实验表明，QuadVerse在重建质量和运动跟踪方面优于相关基线。在此基础上，我们展示了在没有特定任务的真实世界回放情况下，稳健的零-shot视觉导航策略部署。
View on arXiv Download PDF AI Translation
cs.RO / 31 / 2606.07170

Test-Time Trajectory Optimization for Autonomous Driving

自主驾驶的测试时轨迹优化
Xu, Yihong, Zablocki, Eloi, Yin, Yuan, Ramzi, Elias, Kirby, Ellington, Boulch, Alexandre, Cord, Matthieu
Abstract
End-to-end planners for autonomous driving typically generate a set of candidate trajectories, score each one, and return the highest-scoring candidate. However, the scorer is applied only after the proposals are generated and cannot influence the set of trajectories: a weak set of candidates limits planning performance regardless of the scorer's quality. We instead treat the scorer as a learned trajectory-level reward function and search for trajectories that maximize it. Our method, TOAD, runs the Cross-Entropy Method at test time, warm-started from the planner's proposals. It requires no retraining and is plug-and-play for existing planners. Across six base planners, TOAD improves results on NAVSIM-v1 (94.7 PDMS), NAVSIM-v2 (56.3 EPDMS), and the closed-loop HUGSIM benchmark. The code will be made publicly available via the project page: https://valeoai.github.io/TOAD/.
Chinese Translation
端到端的自主驾驶规划器通常生成一组候选轨迹，对每个轨迹进行评分，并返回得分最高的候选轨迹。然而，评分器仅在提案生成后应用，无法影响轨迹集：即使评分器质量再高，弱的候选集也会限制规划性能。我们将评分器视为一个学习的轨迹级奖励函数，并搜索最大化该函数的轨迹。我们的方法TOAD在测试时运行交叉熵方法，从规划器的提案中进行热启动。它不需要重新训练，并且可以与现有规划器无缝集成。在六个基础规划器上，TOAD在NAVSIM-v1（94.7 PDMS）、NAVSIM-v2（56.3 EPDMS）和闭环HUGSIM基准测试中改善了结果。代码将通过项目页面公开发布：https://valeoai.github.io/TOAD/
View on arXiv Download PDF AI Translation
cs.RO / 32 / 2606.07186

A Causal Probabilistic Framework for Perception-Informed Closed-Loop Simulation of Autonomous Driving

基于因果概率框架的感知驱动闭环仿真在自动驾驶中的应用
Fei, Zhennan, Johansson, Rickard, Andersson, Mikael, Eng, Matthias, Eriksson, Mattias, Kianfar, Kaveh, Rahrovani, Sadegh, van der Ploeg, Chris, Borth, Michael, Buermann, Maren, Braat, Michiel, Goossens, Henk, Han, Zijian, Vakilzadeh, Majid Khorsand, de Campos, Gabriel Rodrigues
Abstract
Software-in-the-loop (SIL) simulation is a cornerstone for the validation of modern automotive safety functions. However, many current frameworks utilize ideal sensing, which bypasses the functional insufficiencies of perception algorithms, leading to over-optimistic safety assessments. This paper proposes a perception-informed SIL testing methodology that bridges the gap between ground-truth simulation and real-world perception behavior. We present a framework for incorporating causal probabilistic models into standardized, scenario-based simulation toolchains, applicable to both Advanced Driver Assistance Systems (ADAS) and Autonomous Driving Systems (ADS). Our approach enables the systematic injection of realistic perception errors, such as loss of detection, sizing inaccuracies, and positioning offsets, derived from physical triggering conditions like fog, rain, and object-merging scenarios. By evaluating these ``faults'' within a standardized simulation environment, we demonstrate that perception-informed testing reveals latent operational risks that ideal SIL environments fail to capture, providing a scalable pathway for SOTIF (ISO 21448) validation.
Chinese Translation
软件在环（SIL）仿真是现代汽车安全功能验证的基石。然而，许多现有框架利用理想传感器，这绕过了感知算法的功能不足，导致过于乐观的安全评估。本文提出了一种感知驱动的SIL测试方法，弥合了真实模拟与现实感知行为之间的差距。我们提出了一个将因果概率模型纳入标准化场景基础仿真工具链的框架，适用于高级驾驶辅助系统（ADAS）和自动驾驶系统（ADS）。我们的方法能够系统性地注入现实感知错误，例如检测丢失、尺寸不准确和定位偏差，这些错误源自物理触发条件，如雾、雨和物体合并场景。通过在标准化仿真环境中评估这些“故障”，我们证明了感知驱动的测试揭示了理想SIL环境未能捕捉的潜在操作风险，为SOTIF（ISO 21448）验证提供了一条可扩展的路径。
View on arXiv Download PDF AI Translation
cs.RO / 33 / 2606.07193

Shield-Loco: Shielding Locomotion Policies with Predictive Safety Filtering

Shield-Loco：通过预测安全过滤保护运动策略
Shirwatkar, Aditya, Sanokowski, Sebastian, Kolathaya, Shishir, Johnson, Aaron, Khadiv, Majid
Abstract
Reinforcement learning (RL) policies enable dynamic legged locomotion but lack mechanisms to avoid violations of safety constraints that are absent during training. Large-scale offline safe learning is impractical for covering all edge cases. Existing safety frameworks either rely on reduced-order models that cannot reason about whole-body behaviors or require conservative recovery controllers that degrade task performance. We propose a predictive safety filter that post-hoc filters the nominal contact locations fed to the RL policy. When a collision is predicted, a sampling-based optimizer asynchronously searches for safer contact sequences using a full-physics model, while a learned value function bootstraps long-horizon returns. Our three algorithmic components (geometric projection of sampled contacts, momentum-augmented updates, and replica-exchange) make the optimization tractable in a discontinuous contact landscape. We validate the filter on a quadruped robot in dense, cluttered environments, both in simulation and in the real world, showing substantial reductions in safety violations with minimal deviation from the nominal input.
Chinese Translation
强化学习（RL）策略能够实现动态的四足运动，但缺乏避免在训练期间缺失的安全约束违规的机制。大规模离线安全学习在覆盖所有边缘案例时是不切实际的。现有的安全框架要么依赖于无法推理整体行为的降阶模型，要么需要保守的恢复控制器，这会降低任务性能。我们提出了一种预测安全过滤器，该过滤器对输入到RL策略的名义接触位置进行事后过滤。当预测到碰撞时，基于采样的优化器异步搜索更安全的接触序列，使用全物理模型，同时学习的价值函数为长时间回报提供引导。我们的三个算法组件（采样接触的几何投影、动量增强更新和复制交换）使得在不连续接触环境中优化变得可行。我们在密集、杂乱的环境中对四足机器人验证了该过滤器，包括模拟和现实世界，显示出在名义输入的最小偏差下，安全违规现象显著减少。
View on arXiv Download PDF AI Translation
cs.RO / 34 / 2606.07211

An Abstract Architecture for Explainable Autonomy in Hazardous Environments

危险环境中可解释自主性的抽象架构
Luckcuck, Matt, Taylor, Hazel M, Farrell, Marie
Abstract
Autonomous robotic systems are being proposed for use in hazardous environments, often to reduce the risks to human workers. In the immediate future, it is likely that human workers will continue to use and direct these autonomous robots, much like other computerised tools but with more sophisticated decision-making. Therefore, one important area on which to focus engineering effort is ensuring that these users trust the system. Recent literature suggests that explainability is closely related to how trustworthy a system is. Like safety and security properties, explainability should be designed into a system, instead of being added afterwards. This paper presents an abstract architecture that supports an autonomous system explaining its behaviour (explainable autonomy), providing a design template for implementing explainable autonomous systems. We present a worked example of how our architecture could be applied in the civil nuclear industry, where both workers and regulators need to trust the system's decision-making capabilities.
Chinese Translation
自主机器人系统被提议用于危险环境中，通常是为了降低对人类工人的风险。在不久的将来，人类工人可能会继续使用和指导这些自主机器人，类似于其他计算工具，但具有更复杂的决策能力。因此，工程努力的一个重要领域是确保用户信任该系统。近期文献表明，可解释性与系统的可信度密切相关。与安全性和保密性特性一样，可解释性应当在系统设计中考虑，而不是事后添加。本文提出了一种支持自主系统解释其行为（可解释自主性）的抽象架构，为实现可解释自主系统提供了设计模板。我们展示了一个具体示例，说明我们的架构如何应用于民用核能行业，在该行业中，工人和监管者都需要信任系统的决策能力。
View on arXiv Download PDF AI Translation
cs.RO / 35 / 2606.07217

Robotic Policy Adaptation via Weight-Space Meta-Learning

通过权重空间元学习进行机器人策略适应
Bianchi, Christian, Yousefi, Siamak, Sampieri, Alessio, Roberti, Andrea, Rigazio, Luca, Galasso, Fabio, Franco, Luca
Abstract
Vision-Language-Action (VLA) models are emerging as a promising paradigm for robotic manipulation, enabling general-purpose policies trained from large corpora of demonstrations and action labels. However, adapting these models to new tasks still typically requires task-specific demonstrations, action annotations, and additional fine-tuning, making deployment costly and difficult to scale. We propose WIZARD, a weight-space meta-learning framework that sidesteps task-specific fine-tuning by generating task-specific LoRA parameters for a frozen VLA policy. Given only a language instruction and a short demonstration video, WIZARD predicts the corresponding adaptation weights in a single forward pass, without target-task action labels or test-time optimization. During meta-training, WIZARD learns to map task evidence directly to expert LoRA updates, capturing relationships between tasks in weight space. Experiments on LIBERO show that WIZARD improves performance by up to ~2x on unseen dataset collections and up to ~14x on unseen tasks. On a Franka Emika Panda, WIZARD consistently improves over a real-domain adapted baseline, showing that generated adapters provide task-level specialization beyond simulation.
Chinese Translation
视觉-语言-动作（VLA）模型作为一种有前景的机器人操作范式，能够从大量示范和动作标签中训练出通用策略。然而，将这些模型适应于新任务通常仍然需要特定于任务的示范、动作注释和额外的微调，这使得部署成本高且难以扩展。我们提出了WIZARD，一个权重空间元学习框架，通过为冻结的VLA策略生成特定于任务的LoRA参数，避免了特定任务的微调。仅给定语言指令和一段简短的示范视频，WIZARD能够在一次前向传播中预测相应的适应权重，而无需目标任务的动作标签或测试时优化。在元训练过程中，WIZARD学习将任务证据直接映射到专家LoRA更新，捕捉权重空间中任务之间的关系。在LIBERO上的实验表明，WIZARD在未见数据集集合上的性能提高了约2倍，在未见任务上的性能提高了约14倍。在Franka Emika Panda上，WIZARD始终优于经过真实领域适应的基线，显示生成的适配器提供了超越仿真的任务级专业化。
View on arXiv Download PDF AI Translation
cs.RO / 36 / 2606.07244

Beyond Waypoints: A Trajectory-Centric Waypointing Paradigm for Vision-Language Navigation

超越航点：一种以轨迹为中心的视觉-语言导航航点范式
Shi, Haoxiang, Deng, Xiang, Zhang, Haoyu, Chu, Qiaohui, Wang, Yaowei, Nie, Liqiang
Abstract
Vision-Language Navigation in Continuous Environments (VLN-CE) requires agents to follow natural-language instructions while navigating in real-world-like environments. Most VLN-CE approach\-es adopt a three-stage framework: a waypoint predictor proposes navigable waypoints, and a navigator selects the best waypoint, with a low-level controller executing the movement to it. However, this decoupled paradigm often leads to unreachable waypoints or inconsistencies between planning and control. In this work, instead of predicting isolated waypoints, we introduce a novel paradigm called Trajectory Waypoint, which grounds each candidate waypoint in an executable trajectory. To realize this, we design a Trajectory Waypoint Predictor formulated as a TSDF-guided diffusion policy, which steers trajectory generation away from obstacles, inherently ensuring the reachability of the predicted waypoints. We further propose a trajectory-enhanced navigator that injects the associated trajectory as additional information for planning, enabling strict consistency between high-level semantic decisions and low-level execution. Extensive experiments on the VLN-CE benchmark show that our Trajectory Waypoint paradigm achieves superior performance over the baselines.
Chinese Translation
在连续环境中的视觉-语言导航（VLN-CE）要求智能体在类似真实世界的环境中遵循自然语言指令进行导航。大多数VLN-CE方法采用三阶段框架：航点预测器提出可导航的航点，导航器选择最佳航点，低级控制器执行移动到该航点。然而，这种解耦的范式常常导致无法到达的航点或规划与控制之间的不一致。在本研究中，我们引入了一种新颖的范式，称为轨迹航点（Trajectory Waypoint），它将每个候选航点与可执行的轨迹相结合，而不是预测孤立的航点。为实现这一目标，我们设计了一个轨迹航点预测器，其形式为基于TSDF的扩散策略，能够引导轨迹生成避开障碍物，从而内在地确保预测航点的可达性。我们进一步提出了一种轨迹增强导航器，将相关轨迹作为额外信息注入规划中，从而实现高层语义决策与低层执行之间的严格一致性。在VLN-CE基准上的广泛实验表明，我们的轨迹航点范式在性能上优于基线方法。
View on arXiv Download PDF AI Translation
cs.RO / 37 / 2606.07304

CAPE: Contrastive Action-conditioned Parallel Encoding for Embodied Planning

CAPE：对比动作条件的并行编码用于具身规划
Chen, Cong, Wang, Haowen, Zhang, Zhixiang, Ren, Pei, Che, Zhengping
Abstract
Embodied agents need to predict the future consequences of candidate actions in order to plan effectively before execution. Existing visual dynamics models learn by reconstructing future visual states or rolling out dense latent representations, which spreads learning capacity across visually salient but planning-irrelevant content rather than the action-conditioned changes that drive manipulation outcomes. We propose CAPE, a Contrastive Action-conditioned Parallel Encoding framework that learns visual dynamics by distinguishing the future outcomes induced by different action sequences. Given an initial observation and a candidate action sequence, CAPE decodes the full future latent trajectory in a single forward pass and is trained with a Goal-Convergent Contrastive Objective that aligns predictions corresponding to the same future outcome while separating those corresponding to different outcomes. On real-world DROID and zero-shot transfer to RoboCasa, CAPE substantially outperforms prior baselines on future-state retrieval, offline action matching, and closed-loop planning, while notably reducing planning-time inference cost at long prediction horizons.
Chinese Translation
具身智能体需要预测候选动作的未来结果，以便在执行前有效规划。现有的视觉动态模型通过重建未来视觉状态或展开密集的潜在表示进行学习，这使得学习能力分散在视觉显著但与规划无关的内容上，而不是驱动操作结果的动作条件变化。我们提出了CAPE，一种对比动作条件的并行编码框架，通过区分不同动作序列引发的未来结果来学习视觉动态。给定初始观察和候选动作序列，CAPE在一次前向传递中解码完整的未来潜在轨迹，并使用目标收敛对比目标进行训练，该目标将对应于相同未来结果的预测对齐，同时将对应于不同结果的预测分离。在真实世界的DROID和零样本迁移到RoboCasa的实验中，CAPE在未来状态检索、离线动作匹配和闭环规划方面显著优于先前的基准，同时在较长预测时间范围内显著降低了规划时间推理成本。
View on arXiv Download PDF AI Translation
cs.RO / 38 / 2606.07383

RhinoVLA Technical Report

RhinoVLA技术报告
Intelligence, Huixi, :, Zhang, Chen, Zhou, Chenyang, Ding, Guanglei, He, Guanghui, Gao, Haibin, Chen, Jiajia, Zhang, Jianyong, Yu, Lianyi, Xu, Ningyi, Xu, Ping, Li, Qingchen, Hu, Yingjun, Zhang, Yijia, Liu, Yuxi
Abstract
Vision-Language-Action (VLA) models have shown strong potential for robotic manipulation, but real-time deployment on edge hardware remains challenging. In this work, we identify VLM visual and context tokens as a major source of deployment latency: for GEMM-dominated projection operators, computation grows linearly with the number of input tokens when model dimensions are fixed. Motivated by this observation, we propose RhinoVLA, a deployment-oriented VLA model co-designed with the Huixi R1 edge SoC. RhinoVLA adopts a token-efficient Qwen3-VL backbone and a continuous Action Expert, reducing the VLM-side token and computation burden while preserving pretrained multimodal capability. To support cross-robot learning, RhinoVLA further introduces a unified interface that combines View Registry, 72D physical state-action slot space, and robotinstance LoRA, allowing heterogeneous robot observations and action schemas to be aligned under a shared policy. On the deployment side, RhinoVLA is optimized through hardware-aware compilation, mixed-precision execution, and parallel visual encoding. Experiments show that RhinoVLA achieves downstream performance comparable to {\pi}0.5 at a similar parameter scale, while reaching 11.69 Hz end-to-end inference on Huixi R1, meeting the 10 Hz real-time closedloop control target. The project will be open-sourced at https://github.com/HuixiAI/RhinoVLA.
Chinese Translation
视觉-语言-动作（VLA）模型在机器人操作中显示出强大的潜力，但在边缘硬件上的实时部署仍然面临挑战。在本研究中，我们识别出VLM视觉和上下文标记是导致部署延迟的主要来源：对于以GEMM为主的投影算子，当模型维度固定时，计算量随输入标记数量线性增长。基于这一观察，我们提出了RhinoVLA，一种与Huixi R1边缘SoC共同设计的面向部署的VLA模型。RhinoVLA采用了高效的Qwen3-VL主干网络和连续的动作专家，减少了VLM侧的标记和计算负担，同时保留了预训练的多模态能力。为了支持跨机器人学习，RhinoVLA进一步引入了一个统一接口，结合了视图注册、72D物理状态-动作槽空间和机器人实例LoRA，使异构机器人观察和动作模式能够在共享策略下对齐。在部署方面，RhinoVLA通过硬件感知编译、混合精度执行和并行视觉编码进行了优化。实验表明，RhinoVLA在相似参数规模下实现了与{ ext{π}}0.5相当的下游性能，同时在Huixi R1上达到了11.69 Hz的端到端推理，满足10 Hz实时闭环控制目标。该项目将开源于https://github.com/HuixiAI/RhinoVLA。
View on arXiv Download PDF AI Translation
cs.RO / 39 / 2606.07386

Spline Policy: A Structured Representation for Robot Policies

样条策略：机器人策略的结构化表示
Tian, Mengze, Li, Yiming, Liu, Sichao, Ijspeert, Auke, Calinon, Sylvain
Abstract
Modern imitation-learning policies for robot manipulation often represent actions as fixed-resolution action chunks, which are simple and effective but expose limited geometric and temporal structure before execution. This paper studies Spline Policy (SP), a structured representation that replaces action chunks with spline parameters while keeping the policy backbone unchanged. The predicted spline can be decoded as a compact continuous trajectory, queried at different temporal resolutions, constrained or edited in parameter space, and passed to downstream controllers. For quadratic spline outputs, the same representation can also be converted into a state-dependent vector field through an analytical distance-field construction. Under the regularity and projection assumptions of this construction, the induced dynamics do not increase the distance to the generated spline, yielding a principled local corrective mechanism around the predicted motion. The spline output further supports uncertainty propagation from observations to spline parameters, trajectories, and flow fields, and can be combined with classical control mechanisms such as null-space collision avoidance without retraining the policy backbone. We instantiate SP with diffusion, flow-matching, transformer-based, and vision-language-action backbones. Experiments in low-dimensional motion learning, simulated manipulation under matched backbones, dexterous manipulation, and real-robot case studies show that SP remains compatible with modern policy learners while exposing useful motion-structure properties, including compact decoding, temporal resampling, local correction around predicted motions, uncertainty evaluation, and controller compatibility.
Chinese Translation
现代模仿学习策略在机器人操作中通常将动作表示为固定分辨率的动作块，这种方法简单有效，但在执行前暴露的几何和时间结构有限。本文研究了样条策略（Spline Policy, SP），这是一种结构化表示，它用样条参数替代动作块，同时保持策略骨干不变。预测的样条可以解码为紧凑的连续轨迹，可以在不同的时间分辨率下查询，在参数空间中受到约束或编辑，并传递给下游控制器。对于二次样条输出，同样的表示还可以通过分析距离场构造转换为状态依赖的向量场。在该构造的规则性和投影假设下，诱导的动力学不会增加生成样条的距离，从而在预测运动周围产生一个有原则的局部修正机制。样条输出进一步支持从观测到样条参数、轨迹和流场的不确定性传播，并且可以与经典控制机制（如零空间碰撞避免）结合，而无需重新训练策略骨干。我们用扩散、流匹配、基于变换器和视觉-语言-动作的骨干实例化了SP。在低维运动学习、匹配骨干下的模拟操作、灵巧操作和真实机器人案例研究中的实验表明，SP与现代策略学习者兼容，同时展现出有用的运动结构特性，包括紧凑解码、时间重采样、预测运动周围的局部修正、不确定性评估和控制器兼容性。
View on arXiv Download PDF AI Translation
cs.RO / 40 / 2606.07389

Simulation-Driven Imitation Learning for Biosignals-Free Shared-Autonomy Prosthetic Grasping

基于仿真的模仿学习用于无生物信号的共享自主假肢抓取
Shi, Kaijie, Lu, Wanglong, Chen, Huiling, da Fonseca, Vinicius Prado, Zou, Ting, Zhao, Hanli, Jiang, Xianta
Abstract
Biosignals-free shared-autonomy control of upper-limb prosthetic hands aims to enable natural and low-effort manipulation without relying on EMG or other physiological signals. Recent imitation-learning-based approaches have shown promising results, but their scalability is limited by the cost and variability of collecting large amounts of real-world human demonstration data. In this work, we present a scalable simulation framework that automatically generates diverse reach-to-grasp demonstrations from a wrist-mounted virtual camera. The framework combines physically feasible grasp synthesis, natural reaching trajectories retargeting, and reach--grasp--lift execution in procedurally generated indoor environments. It records wrist-view observations, proprioception, and actions to build a large-scale demonstration dataset for imitation learning. Through extensive simulation benchmarks, we evaluate object and scene generalization and compare several representative state-of-the-art imitation learning methods. Results show that the simulated demonstrations are sufficiently rich and consistent for effective policy learning. In three realistic settings, the learned sim-to-real policy achieves over 90\% grasp success, surpasses baseline methods, and exhibits stronger generalization, highlighting the promise of simulation-driven training for biosignals-free shared-autonomy prosthetic grasping. The demonstrations are available at \href{https://sites.google.com/view/sim-prosthetic-grasp/home}{https://sites.google.com/view/sim-prosthetic-grasp/home}.
Chinese Translation
无生物信号的上肢假肢手的共享自主控制旨在实现自然且低努力的操作，而无需依赖肌电图（EMG）或其他生理信号。近期基于模仿学习的方法显示出良好的效果，但其可扩展性受到收集大量真实人类示范数据的成本和变异性的限制。在本研究中，我们提出了一个可扩展的仿真框架，该框架能够自动生成来自腕部安装虚拟摄像头的多样化抓取示范。该框架结合了物理可行的抓取合成、自然的抓取轨迹重定向以及在程序生成的室内环境中执行的抓取-提升操作。它记录腕部视角的观察、身体感知和动作，以构建一个用于模仿学习的大规模示范数据集。通过广泛的仿真基准测试，我们评估了物体和场景的泛化能力，并比较了几种具有代表性的最先进的模仿学习方法。结果表明，模拟示范在有效策略学习方面足够丰富且一致。在三个现实场景中，学习到的从仿真到现实的策略实现了超过90%的抓取成功率，超越了基线方法，并表现出更强的泛化能力，突显了基于仿真训练在无生物信号共享自主假肢抓取中的潜力。示范数据可在此获取：https://sites.google.com/view/sim-prosthetic-grasp/home。
View on arXiv Download PDF AI Translation
cs.RO / 41 / 2606.07424

Rapid co-design of Buoyancy-assisted robots for Challenging Locomotion using Gaussian Evolutionary Specialists

基于高斯进化专家的浮力辅助机器人快速共设计以应对挑战性运动
Sinha, Ankit, Sontakke, Nitish, Hong, Dennis, Tanaka, Yusuke, Ha, Sehoon
Abstract
Designing high-performance legged robots requires jointly optimizing morphology and control. Model-free Reinforcement Learning (RL) offers an alternative to model-predictive control for developing robust controllers without explicitly specifying robot dynamics. Thus, we have seen theuse of RL to train controllers and evaluate designs for robot morphology optimization. While RL has shown success inlocomotion, using it in the co-design inner loop is expensive due to repeated policy training. Universal policies conditioned on morphology offer a promising alternative, but suffer from behavioral diversity collapse, converging to a single strategy that performs sub-optimally across designs. On the other hand, end-to-end Mixture-of-Experts (MoE) architectures fail due to a collapse in its representation. We propose Gaussian Evolutionary Specialists (GES), a framework that decouples design-space partitioning from policy learning to capture diverse behaviors explicitly. GES assigns specialist policies to evolving Gaussian regions and iteratively refines them via training, probing, and territory expansion. The resulting specialists are integrated into a design sampling loop, replacing costly re-training with direct evaluation. When tested on the Buoyancy-Assisted Light Legged Unit (BALLU), GES discovers designs with 5 - 25% higher performance than naive universal policies. On hardware, a GES optimized design overcomes a 24 cm tall obstacle - 3x improvement over the baseline BALLU design. Moreover, GES curtails design optimization time by 37%.
Chinese Translation
设计高性能的腿式机器人需要对形态和控制进行联合优化。无模型强化学习（Reinforcement Learning, RL）为开发稳健的控制器提供了一种替代模型预测控制的方法，无需明确指定机器人动力学。因此，我们看到RL被用于训练控制器并评估机器人形态优化的设计。尽管RL在运动方面取得了成功，但在共设计的内部循环中使用它的成本较高，因为需要重复进行策略训练。基于形态条件的通用策略提供了一种有前景的替代方案，但由于行为多样性崩溃，最终收敛到单一策略，导致在不同设计中表现不佳。另一方面，端到端的专家混合（Mixture-of-Experts, MoE）架构由于其表示的崩溃而失败。我们提出了高斯进化专家（Gaussian Evolutionary Specialists, GES）框架，该框架将设计空间划分与策略学习解耦，以明确捕捉多样化行为。GES将专家策略分配给不断演化的高斯区域，并通过训练、探测和领土扩展进行迭代优化。最终的专家被整合到设计采样循环中，用直接评估替代昂贵的重新训练。在对浮力辅助轻型腿单元（Buoyancy-Assisted Light Legged Unit, BALLU）的测试中，GES发现的设计性能比简单的通用策略高出5%至25%。在硬件测试中，经过GES优化的设计能够克服24厘米高的障碍，比基线BALLU设计提高了3倍。此外，GES将设计优化时间缩短了37%。
View on arXiv Download PDF AI Translation
cs.RO / 42 / 2606.07437

Re-imagining ISO 26262 in the Age of Autonomous Vehicles: Enhancing Controllability through Transferability and Predictability

在自动驾驶汽车时代重新构想ISO 26262：通过可转移性和可预测性增强可控性
Shinde, Chaitanya, Hajieghrary, Hadi, Schmitt, Paul, Shoemaker, Adam, Seifert, Bodo, Kenner, Steve
Abstract
The ISO 26262 standard defines functional safety for road vehicles through risk assessments based on Severity, Exposure, and Controllability, grounded in a human-driven vehicle paradigm. In the context of autonomous vehicles (AVs), the absence of a human driver necessitates revisiting these principles. This paper decomposes the Controllability placeholder into two auditable evidence dimensions of ISO 26262 by introducing two measurable sub-concepts: Transferability and Predictability. Transferability extends Controllability to capture AV systems' ability to hand off control to dedicated fallback safety mechanisms, while Predictability captures how easily external agents can anticipate AV behavior. Predictability is formally defined from human-robot interaction-inspired principles, and a mathematical framework is provided to quantify it. A designed-versus-achievable gap is introduced to distinguish architectural fallback claims from scene-conditioned achievable fallback capability. The proposed metrics align with ISO 26262 and ISO/PAS 21448 (SOTIF), rendering fallback and interaction claims falsifiable and traceable across ODD slices. These dimensions complement rather than replace existing standards, and the enhancements preserve the structure of ISO 26262 while extending its applicability to driverless automated systems operating at SAE Levels 4 and 5.
Chinese Translation
ISO 26262标准通过基于严重性、暴露和可控性的风险评估定义了道路车辆的功能安全，基于人类驾驶的车辆范式。在自动驾驶汽车（AVs）的背景下，缺乏人类驾驶员需要重新审视这些原则。本文将可控性这一占位符分解为ISO 26262的两个可审核证据维度，通过引入两个可测量的子概念：可转移性和可预测性。可转移性扩展了可控性，以捕捉AV系统将控制权交给专用后备安全机制的能力，而可预测性则捕捉外部代理预测AV行为的难易程度。可预测性从人机交互启发的原则中正式定义，并提供了一个数学框架来量化它。引入了设计与可实现之间的差距，以区分架构后备声明与场景条件下可实现的后备能力。所提出的度量标准与ISO 26262和ISO/PAS 21448（SOTIF）对齐，使得后备和交互声明在ODD切片中可被证伪和追踪。这些维度并不是替代现有标准，而是互补，增强了ISO 26262的结构，同时扩展了其在SAE 4级和5级自动驾驶系统中的适用性。
View on arXiv Download PDF AI Translation
cs.RO / 43 / 2606.07464

Planning-aligned Token Compression for Long-Context Autonomous Driving

规划对齐的长时上下文自主驾驶令牌压缩
Liang, Zhixuan, Chen, Yuxiao, You, Yurong, Karkus, Peter, Ding, Wenhao, Li, Boyi, Popov, Alexander, Wang, Yan, Igl, Maximilian, Li, Yiming, Xu, Danfei, Smolyanskiy, Nikolai, Ivanovic, Boris, Luo, Ping, Pavone, Marco
Abstract
Monolithic vision-action models represent an emerging paradigm in autonomous driving. However, this architecture produces token sequences that quickly exceed real-time computational budgets when encoding extended temporal context for complex interactions. While approaches like linear transformers and external memory try to make the context lightweight, token compression is most compatible with the architecture as it requires no backbone modifications. Yet existing compression adopts rule-based heuristics like temporal decay, decoupled from planning, risking loss of decision-critical information. We propose COMPACT-VA, a planning-aligned working memory framework built on conditional VQ-VAE, compressing extended context into bounded representations. Compression is conditioned on both historical trajectory and a learned planning intent that the posterior encoder distills from future trajectories during training, while the prior encoder learns to predict it from compressed observations. The compressed memory, concatenated with the predicted latent, feeds the policy for end-to-end optimization, planning with retained decision-critical information. We evaluate on high-signal dynamic scenarios where historical context is most critical for behavior correctness (e.g., stop, yield, or proceed), and accordingly design behavioral metrics. Under comparable token budgets, we achieve $>$6% improvement (68.3%) on success rates with consistent gains across metrics. Ablations validate planning-aligned coupling effectiveness. Closed-loop evaluation confirms that COMPACT-VA maintained general driving performance with 3.3* speedup and 2.7* memory reduction over uncompressed processing.
Chinese Translation
单体视觉-动作模型代表了自主驾驶中的一种新兴范式。然而，当对复杂交互进行扩展时间上下文编码时，这种架构产生的令牌序列很快就会超出实时计算预算。尽管线性变换器和外部记忆等方法试图使上下文轻量化，但令牌压缩与该架构最为兼容，因为它不需要对主干进行修改。然而，现有的压缩方法采用了基于规则的启发式方法，如时间衰减，这与规划脱钩，存在丢失决策关键信息的风险。我们提出了COMPACT-VA，一个基于条件VQ-VAE的规划对齐工作记忆框架，将扩展上下文压缩为有界表示。压缩是基于历史轨迹和从未来轨迹中提炼的学习规划意图进行条件化的，而先前编码器则学习从压缩观察中预测这一意图。压缩后的记忆与预测的潜在变量连接，供策略进行端到端优化，从而在保留决策关键信息的情况下进行规划。我们在历史上下文对行为正确性至关重要的高信号动态场景中进行评估（例如，停车、让行或继续行驶），并相应设计行为指标。在可比较的令牌预算下，我们在成功率上实现了超过6%的提升（68.3%），并在各项指标上均取得了一致的增益。消融实验验证了规划对齐耦合的有效性。闭环评估确认，COMPACT-VA在未压缩处理的基础上实现了3.3倍的速度提升和2.7倍的内存减少，同时保持了整体驾驶性能。
View on arXiv Download PDF AI Translation
cs.RO / 44 / 2606.07506

Affordance-Based Hierarchical Reinforcement Learning for Quadruped Pedipulation

基于可供性层次强化学习的四足机器人物体操控
Girgin, Tuba, Castelblanco, Jose, Rodriguez, Gabriel, Girgin, Emre, Kilic, Cagri
Abstract
The object manipulation capabilities of quadruped robots is an open research challenge. While previous studies have focused on low-level policy learning, task execution still relies on expert-designed high-level trajectories. Autonomous selection of both an affordable interaction point on the target object and an affordable robot base pose removes the need for pre-designed trajectories. This study proposes a three-level hierarchical reinforcement learning (RL) framework that utilizes pose affordances to guide the navigation policy, while the navigation policy drives the locomotion policy. In addition, the pedipulation policy is guided by interaction-point affordances, enabling object-centric pose alignment of the quadruped robot and effective end-effector manipulation planning. We train the proposed framework in the IsaacSim ecosystem and evaluate it in both simulation and real-world settings. We investigate the effectiveness of pose affordance across multiple scenarios in simulation while various object interaction tasks are validated on real-world setting forming an object-interaction dataset. The results show that the proposed framework can autonomously identify candidate poses based on their affordance and successfully execute object manipulation tasks in the real world without human guidance.
Chinese Translation
四足机器人在物体操控方面的能力仍然是一个开放的研究挑战。尽管之前的研究集中在低级策略学习上，但任务执行仍依赖于专家设计的高级轨迹。自主选择目标物体上的可供交互点和可供机器人基座姿态的能力消除了对预先设计轨迹的需求。本研究提出了一种三层次的层次强化学习（RL）框架，该框架利用姿态可供性来指导导航策略，而导航策略又驱动运动策略。此外，操控策略受到交互点可供性的指导，使四足机器人能够实现以物体为中心的姿态对齐和有效的末端执行器操控规划。我们在IsaacSim生态系统中训练所提出的框架，并在模拟和现实环境中进行评估。我们在模拟中调查了姿态可供性在多种场景下的有效性，同时在现实环境中验证了各种物体交互任务，形成了一个物体交互数据集。结果表明，所提出的框架能够自主识别基于可供性的候选姿态，并成功在现实世界中执行物体操控任务，而无需人工指导。
View on arXiv Download PDF AI Translation
计算机视觉 (Computer Vision)
90
cs.CV / 1 / 2606.06520

Applying Deep Learning for cockpit segmentation in the context of mixed reality

在混合现实背景下应用深度学习进行驾驶舱分割
Sousa, Alexandre Leles, Nielson, Pedro de Oliveira, Rodrigues, Erick Oliveira, Santos, Rafael Francisco dos, Vitor, Giovani Bernardes
Abstract
Computer vision is an area that has been growing continuously. With the advance of technologies with a first-person view, new development opportunities have emerged inside the area. Mixed reality promotes virtual environments with objects from the physical world shown in real time. For that, it's necessary to be concerned with the immersion of the user in this simulated environment, increasingly seeking to bring it closer to a possible desired reality. This paper proposes the development of image processing in order to perform the segmentation of images to identify what is foreground and background in order to facilitate the union of virtual and real images. Thus, the present work obtain real images of the user using the off-highway truck simulator CAT793F, through a camera, to be able to perform the segmentation of such images with artificial intelligence techniques.The convolutional neural network architectures "U-net" and "DeepLabV3+" are applied to perform image segmentation. As a result, metrics with around 90% accuracy were presented and and the best model was determined.
Chinese Translation
计算机视觉是一个持续增长的领域。随着第一人称视角技术的进步，该领域出现了新的发展机遇。混合现实促进了虚拟环境的创建，实时展示来自物理世界的对象。为此，必须关注用户在该模拟环境中的沉浸感，越来越多地寻求将其与可能的理想现实相结合。本文提出了图像处理的发展，以实现图像的分割，从而识别前景和背景，以便促进虚拟图像和真实图像的结合。因此，本研究通过摄像头获取用户的真实图像，使用越野卡车模拟器CAT793F，以便能够利用人工智能技术对这些图像进行分割。采用卷积神经网络架构“U-net”和“DeepLabV3+”进行图像分割。结果显示，准确率约为90%的指标被提出，并确定了最佳模型。
View on arXiv Download PDF AI Translation
cs.CV / 2 / 2606.06532

GOPAgen: Motion-Aware and Efficient Agentic Long-Video Understanding with Structural Memory and Hierarchical Reasoning

GOPAgen：具有结构记忆和层次推理的运动感知高效代理长视频理解
Chi, Haozhe, Jin, Yang, Mu, Yadong
Abstract
Despite significant progress in agentic long video understanding, existing methods still lack detailed motion comprehension coupled with an efficient memory architecture. In this paper, we propose GOPAgen, a novel approach that first integrates video codec into the video understanding framework via a meticulously designed motion agent trained on Groups of Pictures (GOPs) from video codec. We further develop a GOP tree reasoning algorithm, which is naturally aligned with video codec and enhances the model's ability to understand local detailed motions in videos. Additionally, we carefully design a structural memory mechanism that integrates local motion information with detailed captions in structural pages, and propose an efficient coarse-to-fine zoom-in algorithm to fully exploit the structural memory. Furthermore, we incorporate a motion vector database into the framework to enable efficient retrieval of motion vectors at different granularities. Overall, our method achieves superior Video Question Answering (VQA) performance on various video understanding benchmarks, including MotionBench and Egoschema, thereby demonstrating the superiority of our proposed framework.
Chinese Translation
尽管在代理长视频理解方面取得了显著进展，但现有方法仍然缺乏详细的运动理解以及高效的记忆架构。本文提出了GOPAgen，这是一种新颖的方法，首先通过精心设计的运动代理将视频编码器集成到视频理解框架中，该运动代理基于视频编码中的图像组（Groups of Pictures, GOPs）进行训练。我们进一步开发了一种GOP树推理算法，该算法与视频编码自然对齐，并增强了模型理解视频中局部详细运动的能力。此外，我们精心设计了一种结构记忆机制，将局部运动信息与结构页面中的详细说明集成，并提出了一种高效的粗到细放大算法，以充分利用结构记忆。此外，我们将运动矢量数据库纳入框架，以实现不同粒度的运动矢量的高效检索。总体而言，我们的方法在包括MotionBench和Egoschema在内的各种视频理解基准上实现了卓越的视频问答（Video Question Answering, VQA）性能，从而证明了我们提出的框架的优越性。
View on arXiv Download PDF AI Translation
cs.CV / 3 / 2606.06536

Attention-Guided Autoencoder Fusion for Insulator Defect Detection Using UAV Transmission-Line Imaging

基于注意力引导的自编码器融合用于无人机传输线路绝缘子缺陷检测
Allam, Malak, Shaban, Khaled, Hamdi, Ali
Abstract
Automated defect detection in high-voltage transmission-line insulators remains challenging due to severe class imbalance, large scale variation, and the small spatial extent of defect instances in Unmanned Aerial Vehicle (UAV) imagery. To address these challenges, this paper proposes AE-YOLO, an Attention-Guided AutoEncoder-Enhanced YOLO framework for robust insulator defect detection. The architecture integrates lightweight bottleneck autoencoders within a Feature Pyramid Network-Path Aggregation Network (FPN-PAN) neck. This preserves anomaly-sensitive information during multi-scale feature fusion. Convolutional Block Attention Modules (CBAM) are used throughout the backbone, enhancing feature discrimination and suppressing background interference. The framework also introduces a variance-maximizing autoencoder regularization strategy, which encourages diverse, defect-discriminative latent representations. The network trains using a unified objective that combines focal loss, Complete IoU (CIoU) loss, and autoencoder regularization to address foreground-background imbalance and improve localization accuracy. During inference, Weighted Boxes Fusion (WBF) combines predictions from YOLOv8, YOLOv10, and YOLO11. An autoencoder-guided confidence boosting mechanism improves sensitivity to rare defect categories. Experiments on the Insulator-Defect Detection dataset show that AE-YOLO with an EfficientNetV2 backbone achieves 95.10 percent mAP at 0.5, 96.40 percent precision, and 93.80 percent recall. This performance surpasses the strongest YOLO-family baseline by 5.0 points in mAP at 0.5 and 6.7 points in recall. These results confirm the effectiveness and adaptability of the framework. The model is a practical and scalable solution for UAV-based transmission-line inspection and defect monitoring.
Chinese Translation
高压传输线路绝缘子中的自动缺陷检测仍然面临严重的类别不平衡、大规模变化以及无人机（UAV）图像中缺陷实例的空间范围小等挑战。为了解决这些问题，本文提出了AE-YOLO，一种基于注意力引导的自编码器增强YOLO框架，用于稳健的绝缘子缺陷检测。该架构在特征金字塔网络-路径聚合网络（Feature Pyramid Network-Path Aggregation Network, FPN-PAN）颈部集成了轻量级瓶颈自编码器。这在多尺度特征融合过程中保留了对异常敏感的信息。整个主干网络中使用了卷积块注意力模块（Convolutional Block Attention Modules, CBAM），增强了特征区分能力并抑制了背景干扰。该框架还引入了一种方差最大化自编码器正则化策略，鼓励多样化的、缺陷区分性的潜在表示。网络使用统一目标进行训练，该目标结合了焦点损失、完全IoU（Complete IoU, CIoU）损失和自编码器正则化，以解决前景-背景不平衡并提高定位精度。在推理过程中，加权框融合（Weighted Boxes Fusion, WBF）结合了YOLOv8、YOLOv10和YOLO11的预测。自编码器引导的置信度提升机制提高了对稀有缺陷类别的敏感性。在绝缘子缺陷检测数据集上的实验表明，采用EfficientNetV2主干的AE-YOLO在0.5时达到了95.10%的mAP、96.40%的精度和93.80%的召回率。这一性能在mAP 0.5时超过了最强YOLO系列基线5.0个百分点，召回率提高了6.7个百分点。这些结果确认了该框架的有效性和适应性。该模型是一个实用且可扩展的解决方案，适用于基于无人机的传输线路检查和缺陷监测。
View on arXiv Download PDF AI Translation
cs.CV / 4 / 2606.06538

WorldBench: A Challenging and Visually Diverse Multimodal Reasoning Benchmark

WorldBench：一个具有挑战性和视觉多样性的多模态推理基准
Yin, Yida, Krishnakumar, Harish, Lee, Chung Peng, Zeng, Boya, Chai, Wenhao, Tong, Shengbang, Chen, Wenhu, Xu, Hu, Fu, Xingyu, Sarch, Gabriel, Korolova, Aleksandra, Liu, Zhuang
Abstract
In real-world applications, models are expected to perform reliably across diverse settings. Yet, many existing multimodal benchmarks expand task types without capturing the visual diversity needed to handle open-ended visual inputs. We present WorldBench, a challenging and visually diverse reasoning benchmark to evaluate Multimodal Large Language Models (MLLMs). We build a taxonomy of thousands of visual concepts across multiple domains (e.g., living things). Guided by this taxonomy, we curate a broad collection of images from search engines and existing datasets to comprehensively represent the visual world. Through structured trial-and-error, we manually design challenging questions that frontier MLLMs fail to answer. On quantitative and human evaluations, WorldBench achieves higher visual diversity than any existing diverse benchmark. Evaluating 15 MLLMs on WorldBench reveals weaknesses in visual understanding: even the strongest model reaches only 64.0% accuracy, while some models perform marginally above chance-level. We hope our work highlights the importance of visual diversity in building multimodal benchmarks.
Chinese Translation
在现实世界的应用中，模型需要在多种环境下可靠地执行。然而，许多现有的多模态基准扩展了任务类型，却未能捕捉到处理开放式视觉输入所需的视觉多样性。我们提出了WorldBench，一个具有挑战性和视觉多样性的推理基准，用于评估多模态大型语言模型（MLLMs）。我们构建了一个涵盖多个领域（例如，生物体）数千个视觉概念的分类体系。在这一分类体系的指导下，我们从搜索引擎和现有数据集中精心策划了一系列广泛的图像，以全面代表视觉世界。通过结构化的试错过程，我们手动设计了具有挑战性的问题，这些问题是前沿的MLLMs无法回答的。在定量和人工评估中，WorldBench的视觉多样性超过了任何现有的多样性基准。对15个MLLMs在WorldBench上的评估揭示了视觉理解的弱点：即使是最强的模型也仅达到64.0%的准确率，而一些模型的表现仅略高于随机水平。我们希望我们的工作强调了在构建多模态基准中视觉多样性的重要性。
View on arXiv Download PDF AI Translation
cs.CV / 5 / 2606.06539

Synthetic Benchmarks Overstate Forward-Forward Scaling: Real-Data Limits of Layer-Local Training

合成基准高估了前向-前向缩放：层局部训练的真实数据限制
Chen, Yucheng
Abstract
Forward-Forward (FF) learning [Hinton, 2022] replaces backpropagation with strictly layer-local goodness updates. Recent FF-CNN work has narrowed the gap to BP on 32x32 benchmarks, raising the question of whether layer-local training is becoming a viable alternative at realistic scale. To probe this rigorously, we develop DTG-FF -- dynamic temperature goodness, decoupled normalization, and multi-layer fusion -- as an instrument that sets FF-family state of the art across nine real-data benchmarks (91.8% CIFAR-10 and the first FF baseline at ImageNet-100 224x224), and use it to audit how far layer-local training actually scales. (1) Real-data scaling. Under identical recipe and backbone, an architecture-matched BP-DeepSup baseline beats DTG-FF by 2.40/5.93 pp on CIFAR-10/CIFAR-100, and the gap widens with class count. At 224x224 the same instrument reaches only 49.4% -- the first FF baseline at this scale, versus typical BP above 75% [Tian et al., 2020] -- exposing a real-data ceiling invisible at 32x32. (2) Synthetic vs. real K-conflict. DTG-FF increasingly outperforms BP as class count K grows on synthetic teacher-student tasks, yet on real images the FF-BP gap reverses sign and widens with K. A within-dataset CIFAR-100 coarse vs. fine probe isolates label-hierarchy from image distribution: synthetic K-sweeps confound output dimensionality with fine-grained discrimination difficulty and thereby overstate FF transferability. (3) Systems audit. FF can be implemented without storing depth-wide activations, but on commodity 8 GB hardware standard BP+gradient-accumulation reaches 4.18 GB / 157 imgs/s versus DTG-FF's 7.90 GB / 138 imgs/s, so a memory-based justification for FF at this scale is not supported under fair baselines.
Chinese Translation
前向-前向（FF）学习 [Hinton, 2022] 用严格的层局部良好性更新替代了反向传播。最近的 FF-CNN 研究在 32x32 基准上缩小了与反向传播的差距，提出了层局部训练是否在现实规模上成为可行替代方案的问题。为了严谨地探讨这一问题，我们开发了 DTG-FF——动态温度良好性、解耦归一化和多层融合——作为一种工具，在九个真实数据基准上设定了 FF 系列的最新水平（CIFAR-10 达到 91.8% 和 ImageNet-100 224x224 的首个 FF 基线），并利用它审计层局部训练的实际扩展性。(1) 真实数据扩展。在相同的配方和骨干网络下，与架构匹配的 BP-DeepSup 基线在 CIFAR-10/CIFAR-100 上分别比 DTG-FF 高出 2.40/5.93 个百分点，且随着类别数量的增加，差距进一步扩大。在 224x224 的情况下，同样的工具仅达到 49.4%——在该规模下的首个 FF 基线，而典型的反向传播则超过 75% [Tian et al., 2020]——揭示了在 32x32 上不可见的真实数据上限。(2) 合成与真实 K 冲突。随着类别数量 K 的增加，DTG-FF 在合成教师-学生任务上逐渐超越反向传播，但在真实图像上，FF-BP 差距反转并随着 K 的增加而扩大。一个在数据集内的 CIFAR-100 粗略与精细探测将标签层次与图像分布隔离：合成 K 扫描混淆了输出维度与细粒度区分难度，从而高估了 FF 的可迁移性。(3) 系统审计。FF 可以在不存储深度宽激活的情况下实现，但在普通的 8 GB 硬件上，标准的 BP+梯度累积达到 4.18 GB / 157 图像/秒，而 DTG-FF 则为 7.90 GB / 138 图像/秒，因此在这一规模下，基于内存的 FF 理由在公平基线下并不成立。
View on arXiv Download PDF AI Translation
cs.CV / 6 / 2606.06601

Direct 3D-Aware Object Insertion via Decomposed Visual Proxies

通过分解视觉代理实现直接的3D感知物体插入
Gong, Jingbo, Wang, Yikai, Lan, Yushi, Wan, Yuhao, Ouyang, Ziheng, Zhao, Rui, Cheng, Ming-Ming, Hou, Qibin, Loy, Chen Change
Abstract
Object insertion aims to seamlessly composite a reference object into a specified region of a background image. Recent diffusion-based methods achieve high visual quality but formulate insertion as a simple 2D inpainting task, providing no explicit control over the object's 3D pose and limiting their practical applicability. We propose DIRECT (Decomposed Injection for Reference Composition and Target-integration), a novel framework that integrates interactive pose manipulation with high-fidelity 2D image synthesis to enable pose-controllable object insertion. Our method decomposes the insertion conditions into three complementary components: appearance guidance capturing visual details from the reference object, geometry guidance derived from the user-adjusted 3D proxy, and context guidance from the target background. By injecting them through separate pathways, DIRECT avoids feature entanglement and simultaneously preserves reference appearance, follows the user-specified pose, and adapts the object to the target scene. We also introduce an automated data construction pipeline to improve the diversity and quality of training data. Experiments show that DIRECT outperforms previous methods in both geometric controllability and visual quality.
Chinese Translation
物体插入旨在将参考物体无缝合成到背景图像的指定区域。近期基于扩散的方法实现了高视觉质量，但将插入任务简化为简单的2D修补，未能对物体的3D姿态进行明确控制，从而限制了其实际应用性。我们提出了DIRECT（Decomposed Injection for Reference Composition and Target-integration），这是一个新颖的框架，将交互式姿态操控与高保真2D图像合成相结合，以实现可控姿态的物体插入。我们的方法将插入条件分解为三个互补组件：外观引导，从参考物体捕捉视觉细节；几何引导，源自用户调整的3D代理；以及上下文引导，来自目标背景。通过独立通道注入这些组件，DIRECT避免了特征纠缠，同时保持参考外观，遵循用户指定的姿态，并将物体适应于目标场景。我们还引入了一个自动化数据构建管道，以提高训练数据的多样性和质量。实验表明，DIRECT在几何可控性和视觉质量方面优于先前的方法。
View on arXiv Download PDF AI Translation
cs.CV / 7 / 2606.06631

From Pixels to Newtons: Predicting In Vivo Joint Contact Forces from Monocular Video

从像素到牛顿：从单目视频预测体内关节接触力
Lauer, Jessy
Abstract
Joint contact forces govern implant longevity, cartilage health, and rehabilitation outcomes, shaping who develops osteoarthritis, who recovers well from joint replacement, and who benefits from biomechanical interventions. Yet they remain measurable only invasively, in a few dozen patients with instrumented implants. I present a physics-free pipeline to predict instantaneous 3D hip and knee contact forces from an uncalibrated monocular video: no markers, force plates, electromyography, subject-specific imaging, or musculoskeletal model. Parametric body meshes are recovered per frame, encoded as kinematic features, and decoded into forces by a transformer whose pose stream is adaptively modulated at every layer by body shape, joint, side, activity text, and self-supervised video tokens (V-JEPA 2), unifying hip and knee in a single model. Under leave-one-subject-out cross-validation across 26 patients and 25 activity categories from the in vivo OrthoLoad database, the pipeline matches the accuracy of subject-specific musculoskeletal simulations ($0.32 \pm 0.08$ BW RMSE for hip; $0.23 \pm 0.03$ BW for knee) and resolves peak force changes smaller than those reported for gait retraining and osteoarthritis progression. Applied zero-shot to an independent instrumented cohort, it rivals or outperforms prior published methods. Even without curated activity labels, video features alone preserve accuracy and enable end-to-end inference on raw footage. Driven by the predictor, a generative motion prior produces biomechanically plausible variants with reduced peak loading, rediscovering strategies from the predictive simulation literature. This pipeline establishes uncalibrated monocular video as a viable modality for estimating joint loading, opening a path toward retrospective analysis of archived clinical recordings, primary-care screening, and at-home rehabilitation tracking.
Chinese Translation
关节接触力决定了植入物的使用寿命、软骨健康和康复结果，影响谁会发展为骨关节炎、谁能从关节置换中良好恢复，以及谁能从生物力学干预中受益。然而，这些力目前仅能通过侵入性方法在少数几十名装有仪器的患者中进行测量。本文提出了一种无物理模型的流程，通过未校准的单目视频预测瞬时的三维髋关节和膝关节接触力：不需要标记、力板、肌电图、个体特征成像或肌肉骨骼模型。每帧恢复参数化的身体网格，编码为运动学特征，并通过一个变换器解码为力，该变换器的姿态流在每一层都根据身体形状、关节、侧别、活动文本和自监督视频标记（V-JEPA 2）自适应调节，从而将髋关节和膝关节统一在一个模型中。在对26名患者和来自体内OrthoLoad数据库的25个活动类别进行的留一法交叉验证中，该流程的准确性与个体特征的肌肉骨骼模拟相匹配（髋关节的$0.32 ext{ BW} ext{ RMSE} ext{，} ext{ } ext{ } 0.08$；膝关节的$0.23 ext{ BW} ext{，} ext{ } ext{ } 0.03$），并能够解析小于步态重训练和骨关节炎进展报告的峰值力变化。该方法在一个独立的仪器化队列中应用零样本学习，与先前发布的方法相媲美或表现更优。即使没有经过整理的活动标签，仅凭视频特征也能保持准确性，并实现对原始视频的端到端推断。在预测器的驱动下，一个生成运动先验产生了生物力学上合理的变体，减少了峰值负荷，重新发现了预测模拟文献中的策略。该流程确立了未校准的单目视频作为估计关节负荷的可行方式，为对存档临床记录的回顾性分析、初级护理筛查和居家康复跟踪开辟了道路。
View on arXiv Download PDF AI Translation
cs.CV / 8 / 2606.06664

Inside the Visual Mind: Neuroscience-Motivated Concept Circuits for Interpreting and Steering Vision Transformers

视觉思维内部：基于神经科学的概念电路用于解释和引导视觉变换器
Li, Tang, Chen, Yanlin, Ma, Mengmeng, Peng, Xi
Abstract
Despite high accuracy, Vision Transformer (ViT) predictions can be driven by spurious cues, raising the need to understand their inner workings before safe deployment. Sparse autoencoders (SAEs) provide a promising lens for decomposing model representations into human-interpretable concepts, yet adapting SAE-based interpretation to ViTs remains challenging due to limited control over concept coverage and subjective, non-scalable feature interpretation. To fill the gaps, motivated by neuroscience-inspired principles, we propose ViSAE, a mechanistic interpretability toolbox for understanding ViT inner workings through concept circuits. ViSAE consists of three components: (1) A probing suite with 64K images and a 16K visually grounded concept vocabulary, improving concept coverage efficiency by 20x over ImageNet and interpretation accuracy by 28.7% over existing concept sets. (2) Top-down concept reading and Bottom-up circuit tracing algorithms that automatically recover ViT inner workings via concept circuits. (3) Applications for auditing and steering ViT behavior. Through concept editing, ViSAE improves the worst-group accuracy on WaterBirds by 48.2%, outperforming existing methods by 23.8%. Our data and code: https://github.com/deep-real/ViSAE.
Chinese Translation
尽管视觉变换器（Vision Transformer, ViT）预测的准确性很高，但其预测可能受到虚假线索的影响，这就需要在安全部署之前理解其内部工作机制。稀疏自编码器（Sparse Autoencoders, SAEs）为将模型表示分解为人类可解释的概念提供了一个有前景的视角，但将基于SAE的解释方法适配到ViT上仍然具有挑战性，因为对概念覆盖的控制有限，并且特征解释主观且不可扩展。为填补这一空白，我们提出了ViSAE，一个基于神经科学启发原则的机械可解释性工具箱，通过概念电路理解ViT的内部工作机制。ViSAE由三个部分组成：（1）一个探测套件，包含64K图像和一个16K视觉基础概念词汇，概念覆盖效率比ImageNet提高了20倍，解释准确性比现有概念集提高了28.7%。 （2）自上而下的概念读取和自下而上的电路追踪算法，能够通过概念电路自动恢复ViT的内部工作机制。 （3）用于审计和引导ViT行为的应用。通过概念编辑，ViSAE将WaterBirds数据集上最差组的准确率提高了48.2%，超越现有方法23.8%。我们的数据和代码可在：https://github.com/deep-real/ViSAE.
View on arXiv Download PDF AI Translation
cs.CV / 9 / 2606.06666

Architecture-Adaptive Uncertainty Fusion for Deepfake Detection

面向架构自适应的不确定性融合用于深伪检测
Sharma, Ritesh, Ghasemigol, Mohammad, Motai, Yuichi
Abstract
Deepfake detection systems achieve near-perfect accuracy on benchmarks, yet forensic deployment demands reliable prediction uncertainty. Existing uncertainty quantification (UQ) methods rely on single sources and ignore that optimal uncertainty composition varies across architectures. We propose Correlation-Optimized Fusion (COF), an architecture-adaptive framework that fuses five complementary uncertainty sources -- epistemic, aleatoric, calibration, conformal, and distributional -- by maximizing Pearson correlation between fused uncertainty scores and prediction errors via constrained optimization on the probability simplex. COF requires no model modifications and only 42 s of weight optimization, compared to 20--45 h for a 5-model Deep Ensemble. Evaluation across eleven architectures on FaceForensics++ reveals a fundamental trade-off: under matched train/evaluation protocol, non-linear methods achieve approximately 5--6% higher in-domain correlation than COF (mean r = 0.438), but this reverses under distribution shift. On CelebDF, COF outperforms Random Forest in 9/11 architectures with up to 7.3x higher correlation (MaxViT-B: r = 0.249 vs. 0.034); RF degrades 85% cross-domain to r = 0.071, whereas COF retains substantially more signal (74% drop to r = 0.116). Cross-dataset evaluation on CelebDF and DFDC reveals catastrophic generalization failure across all methods: in-domain correlations of 0.41--0.47 collapse to near-zero externally (mean degradation 90.7%), with seven of eleven architectures exhibiting uncertainty inversion. These results establish COF as a practical, interpretable framework for controlled-distribution deployment and identify domain-adaptive UQ as the central open challenge for forensic deployment.
Chinese Translation
深伪检测系统在基准测试中实现了近乎完美的准确性，但法医学应用要求可靠的预测不确定性。现有的不确定性量化（UQ）方法依赖于单一来源，忽略了最佳不确定性组合在不同架构间的变化。我们提出了相关性优化融合（Correlation-Optimized Fusion, COF），这是一种架构自适应框架，通过在概率单纯形上进行约束优化，最大化融合的不确定性评分与预测误差之间的皮尔逊相关性，融合五种互补的不确定性来源——认知不确定性、随机不确定性、校准不确定性、符合性不确定性和分布不确定性。与需要20至45小时的5模型深度集成（Deep Ensemble）相比，COF不需要模型修改，仅需42秒的权重优化。在FaceForensics++上对十一种架构的评估揭示了一个基本的权衡：在匹配的训练/评估协议下，非线性方法的领域内相关性比COF（平均r = 0.438）高出约5-6%，但在分布变化下则相反。在CelebDF上，COF在11种架构中的9种上优于随机森林（Random Forest），相关性高出最多7.3倍（MaxViT-B: r = 0.249 vs. 0.034）；随机森林在跨领域时降级85%，相关性降至r = 0.071，而COF则保留了更多信号（降幅74%至r = 0.116）。在CelebDF和DFDC上的跨数据集评估揭示了所有方法的灾难性泛化失败：领域内相关性从0.41-0.47崩溃至接近零（平均降幅90.7%），其中七种架构表现出不确定性反转。这些结果确立了COF作为一种实用且可解释的框架，用于受控分布的部署，并将领域自适应UQ识别为法医学应用的核心开放挑战。
View on arXiv Download PDF AI Translation
cs.CV / 10 / 2606.06671

JA-SIREN: Deterministic Initialization for Sinusoidal Networks via Spectral Matching

JA-SIREN：通过谱匹配实现正弦网络的确定性初始化
Alsakabi, Mohammed, Hu, Kejia, Dolan, John M., Tonguz, Ozan K.
Abstract
Existing implicit neural representation (INR) approaches suffer from stochastic initialization that does not guarantee consistent or high-quality performance across runs, with variations reaching more than 2.5 dB (78%) in image regression. This variation is problematic for scientific computing and simulation, where result reproducibility is crucial. To address this problem, we present Jacobi-Anger Sinusoidal Representation Network (JA-SIREN), a deterministic initialization scheme for sinusoidal networks grounded in classical spectral analysis. By computing the Discrete Sine Transform (DST) of the target signal and leveraging the Jacobi-Anger expansion, we derive closed-form weights for a two-layer sinusoidal MLP that analytically match the network's initial spectral response to the target signal, requiring no random seed or additional hyperparameter tuning. On the Kodak dataset, JA-SIREN achieves a mean PSNR of 67.18 dB, a 21.30 dB improvement over the best baseline. This is achieved with zero run-to-run variance, confirming that spectrally-informed initialization is a more effective and reproducible alternative to stochastic initialization for sinusoidal INRs.
Chinese Translation
现有的隐式神经表示（INR）方法受到随机初始化的影响，这种初始化无法保证在多次运行中表现一致或高质量，图像回归中的变化可达2.5 dB（78%）以上。这种变化在科学计算和仿真中是一个问题，因为结果的可重复性至关重要。为了解决这个问题，我们提出了雅可比-安格正弦表示网络（JA-SIREN），这是一种基于经典谱分析的正弦网络确定性初始化方案。通过计算目标信号的离散正弦变换（DST）并利用雅可比-安格展开，我们推导出一个两层正弦多层感知器的闭式权重，这些权重在解析上将网络的初始谱响应与目标信号匹配，无需随机种子或额外的超参数调整。在Kodak数据集上，JA-SIREN实现了67.18 dB的平均PSNR，比最佳基线提高了21.30 dB。这是在零运行间方差的情况下实现的，确认了谱信息驱动的初始化是正弦INR中比随机初始化更有效且可重复的替代方案。
View on arXiv Download PDF AI Translation
cs.CV / 11 / 2606.06684

Adaptive Band Selection for Hyperspectral Classification with Spatially Disjoint Evaluation

具有空间不重叠评估的高光谱分类自适应波段选择
El-Hajri, Ikram, Karrakchou, Ouassim, Mousist, Alejandro
Abstract
Hyperspectral band selection methods based on differentiable selectors can be sensitive to initialization and to extracting a final discrete subset, while prescribed band counts limit flexibility. We propose SGBR-HC (Spectral-Group Band Ranking with Hard-Concrete initialization), a two-stage method that uses a supervised spectral ranking to initialize trainable sparse gates rather than treating ranking as a fixed selection rule, letting the number of selected bands be determined by training. Stage-1 scores candidate bands from training pixels by class discriminability and spectral diversity; this ranking seeds the gate logits for Stage-2, which trains the sparse gates jointly with a spatial classifier. Under spatially disjoint evaluation on Pavia University and Houston 2013, verified by retraining a fresh classifier on the selected bands, SGBR-HC achieves the highest mean overall accuracy and Cohen's kappa with approximately twenty bands. Bypassing Stage-1 degrades OA by 8.84 pp on Pavia University and 22.15 pp on Houston 2013, confirming the ranking prior's role. Random pixel splits inflate OA on Pavia University by 30.56 pp, underscoring spatial leakage as a critical evaluation confound.
Chinese Translation
基于可微选择器的高光谱波段选择方法对初始化和提取最终离散子集可能敏感，而预设的波段数量限制了灵活性。我们提出了SGBR-HC（硬混凝土初始化的光谱组波段排名），这是一种两阶段方法，利用监督光谱排名来初始化可训练的稀疏门，而不是将排名视为固定选择规则，从而让所选波段的数量由训练决定。第一阶段通过类别可区分性和光谱多样性对训练像素中的候选波段进行评分；该排名为第二阶段的门逻辑提供种子，该阶段与空间分类器共同训练稀疏门。在Pavia大学和2013年休斯顿的空间不重叠评估下，通过在所选波段上重新训练一个新的分类器进行验证，SGBR-HC在大约二十个波段下实现了最高的平均整体准确率和Cohen's kappa。跳过第一阶段使Pavia大学的整体准确率下降了8.84个百分点，休斯顿2013下降了22.15个百分点，确认了排名先验的作用。随机像素划分使Pavia大学的整体准确率提高了30.56个百分点，强调了空间泄漏作为一个关键评估干扰因素。
View on arXiv Download PDF AI Translation
cs.CV / 12 / 2606.06685

RigPAPR: Rig-Based Animation of Static Neural Point Clouds from a Fixed-Viewpoint Video

RigPAPR：基于骨架的静态神经点云动画生成，源自固定视角视频
Peng, Shichong, Zhang, Yanshu, Li, Ke
Abstract
Static neural point reconstructions capture a subject at high fidelity from posed images. Given such a reconstruction, we aim to animate it to follow a monocular fixed-viewpoint driving video of the subject, whether captured or produced by image-to-video (I2V) generation, and to recover a rigged, re-posable 3D asset. Existing methods deform Gaussian splats through direct linear blend skinning (LBS) or mesh proxies, both of which are prone to joint-boundary artifacts under articulation, even with per-primitive corrections. We trace the artifact to the representation: each splat carries an individual shape calibrated in the canonical pose to tile with its neighbours. Under rigid LBS, each splat moves with its bone but cannot bend, so the canonical tiling breaks at joint boundaries into gaps and spikes. Proximity attention point rendering (PAPR) instead carries no per-primitive shape; each pixel is recomposed at render time from the deformed primitives' positions, so the surface re-forms naturally with the articulation. We present RigPAPR, which auto-rigs a static PAPR cloud and drives it under direct LBS from a single fixed-viewpoint video, without mesh proxy, pose-dependent correction, or category template. On synthetic subjects, RigPAPR matches the strongest baseline at the supervised view and exceeds mesh-based and Gaussian-splatting baselines at novel views by 3+dB PSNR, with cleaner joint-boundary renderings of both synthetic and real subjects.
Chinese Translation
静态神经点重建能够从姿态图像中高保真地捕捉一个主体。给定这样的重建，我们旨在使其动画化，以跟随主体的单目固定视角驱动视频，无论是捕获的还是通过图像到视频（I2V）生成的，并恢复一个可重新定位的骨架3D资产。现有方法通过直接线性混合蒙皮（LBS）或网格代理来变形高斯点，这两者在关节运动时都容易出现关节边界伪影，即使进行了逐原始体素的修正。我们将伪影归因于表示：每个点携带一个在规范姿态下校准的个体形状，以与其邻居拼接。在刚性LBS下，每个点随其骨骼移动，但无法弯曲，因此在关节边界处的规范拼接会断裂，形成间隙和尖刺。相对注意点渲染（PAPR）则不携带逐原始体素形状；每个像素在渲染时根据变形原始体素的位置重新组合，因此表面在关节运动时自然重构。我们提出了RigPAPR，它能够自动为静态PAPR点云建立骨架，并在没有网格代理、姿态相关修正或类别模板的情况下，从单个固定视角视频下通过直接LBS驱动。在合成主体上，RigPAPR在监督视图下与最强基线相匹配，并在新视图下超过基于网格和高斯点的基线3+dB PSNR，同时在合成和真实主体的关节边界渲染中呈现更干净的效果。
View on arXiv Download PDF AI Translation
cs.CV / 13 / 2606.06690

RPC-GS: Gaussian Splatting with native RPC Rendering for Satellite Imagery

RPC-GS：用于卫星图像的原生RPC渲染的高斯溅射
Wagner, Valentin, Bullinger, Sebastian, Bodensteiner, Christoph, Arens, Michael
Abstract
We present RPC-GS, the first Gaussian Splatting framework for satellite imagery that operates natively with Rational Polynomial Camera (RPC) models. The RPC model is the de facto standard for representing the complex imaging geometry of modern pushbroom satellite sensors. To simplify rendering, prior satellite Gaussian Splatting methods replace the RPC model with perspective or affine camera approximations, leading to geometric errors during reconstruction. RPC-GS avoids these approximations by projecting Gaussian means and covariances directly through the RPC model during the splatting process. We embed the RPC model in a chain of carefully selected geo-coordinate transformations representing a mapping from splatting-suitable scene coordinates to image coordinates. To map the Gaussian covariance matrices, we derive a numerically robust Jacobian-based covariance projection for the (partially nonlinear) coordinate transformations. Since RPCs lack an explicit notion of camera depth, we integrate a metric ray-based depth formulation. We benchmark RPC, perspective, and affine camera models in a unified framework, with our native RPC renderer consistently achieving the lowest reconstruction error on leading satellite benchmark datasets, improving mean altitude error over perspective and affine approximations by 29.6% and 63.8% on DFC2019, and by 9.9% and 37.9% on IARPA2016. We release our code to support future research of Gaussian Splatting in the satellite imaging domain.
Chinese Translation
我们提出了RPC-GS，这是第一个针对卫星图像的高斯溅射框架，能够原生地与有理多项式相机（Rational Polynomial Camera, RPC）模型配合使用。RPC模型是现代推扫卫星传感器复杂成像几何的事实标准。为了简化渲染，以往的卫星高斯溅射方法用透视或仿射相机近似替代RPC模型，导致重建过程中出现几何误差。RPC-GS通过在溅射过程中直接通过RPC模型投影高斯均值和协方差，避免了这些近似。我们将RPC模型嵌入一系列精心选择的地理坐标变换中，这些变换表示从适合溅射的场景坐标到图像坐标的映射。为了映射高斯协方差矩阵，我们推导出了一种数值稳健的基于雅可比的协方差投影方法，适用于（部分非线性）坐标变换。由于RPC缺乏明确的相机深度概念，我们整合了一种基于度量射线的深度公式。我们在统一框架中对RPC、透视和仿射相机模型进行了基准测试，结果显示我们的原生RPC渲染器在领先的卫星基准数据集上始终实现了最低的重建误差，在DFC2019数据集上，相比透视和仿射近似，平均高度误差分别改善了29.6%和63.8%，在IARPA2016数据集上改善了9.9%和37.9%。我们发布了我们的代码，以支持未来在卫星成像领域的高斯溅射研究。
View on arXiv Download PDF AI Translation
cs.CV / 14 / 2606.06695

S23DR 2026 Winning Solution

S23DR 2026 胜利解决方案
Skvrna, Jan, Purkrabek, Miroslav, Neumann, Lukas
Abstract
This text presents the winning solution to the S23DR 2026 challenge for structured 3D wireframe reconstruction from sparse SfM, fitted depth, and semantic segmentations. The method treats vertices as a conditional set and denoises 64 vertex tokens with a flow-matching DiT conditioned on Perceiver-style scene tokens. A global pass predicts the coarse structure, a hull-cropped second pass refines it, and a small multi-sample consensus step keeps the stochastic sampler well behaved. The final system ranked first on the private leaderboard, achievingHSS = 0.654.
Chinese Translation
本文介绍了 S23DR 2026 挑战的获胜解决方案，该挑战旨在从稀疏的结构光重建（SfM）、拟合深度和语义分割中进行结构化 3D 线框重建。该方法将顶点视为条件集，并使用基于 Perceiver 风格场景标记的流匹配 DiT 对 64 个顶点标记进行去噪。全局传递预测粗略结构，经过裁剪的第二次传递对其进行细化，而一个小型多样本共识步骤确保随机采样器的行为良好。最终系统在私有排行榜上排名第一，达到 HSS = 0.654。
View on arXiv Download PDF AI Translation
cs.CV / 15 / 2606.06696

MMBU: A Massive Multi-modal Biomedical Understanding Benchmark to Probe the Perception Capabilities of Vision-Language Models

MMBU：一个大规模多模态生物医学理解基准，以探测视觉-语言模型的感知能力
D'Cunha, Ryan, Lozano, Alejandro, Sun, Xiaoxiao, Jarquin, Daniel Vela, Sun, Min Woo, Aklilu, Josiah, Burgess, James, Zhang, Yuhui, Nayebi, Ryan, Avila, Paola, Robayo, Ye, Jin, Hu, Ming, Deng, Zhongying, He, Junjun, Chen, Xin, Yao, Yue, Tibshirani, Robert, Nirschl, Jeffrey J., Yeung-Levy, Serena
Abstract
Vision and language models (VLMs) hold immense promise to transform biomedical imaging workflows, from detecting lesions in chest X-rays to profiling cellular features in microscopy. Realizing this potential, however, requires robust and fine-grained visual perception. Models need to correctly interpret subtle features in images, and they must do so across diverse biomedical modalities, scales, and contexts. Nevertheless, current benchmarks remain limited. To address these gaps, we introduce the Massive Multimodal Biomedical Understanding (MMBU) benchmark. It is the largest biomedical vision and language benchmark to date, covering 35 submodalities with rich structured metadata. It includes both open and closed versions of ungrounded classification, grounded classification, and object detection, enabling systematic evaluation of model performance across biological scales, clinical settings, and imaging modalities. Evaluating 15 open-weight and 2 frontier VLMs, we find that while medical adaptation provides measurable gains for some models, the high accuracy often reported on established benchmarks can mask deficiencies in visual perception and domain generalization.
Chinese Translation
视觉和语言模型（VLMs）在转变生物医学成像工作流程方面具有巨大的潜力，从检测胸部X光片中的病变到在显微镜下分析细胞特征。然而，实现这一潜力需要强大的、细致的视觉感知。模型需要正确解读图像中的微妙特征，并且必须能够在多种生物医学模态、尺度和背景下进行解读。然而，目前的基准测试仍然有限。为了解决这些问题，我们引入了大规模多模态生物医学理解（MMBU）基准。这是迄今为止最大的生物医学视觉和语言基准，涵盖35个子模态，并配有丰富的结构化元数据。它包括无基础分类、有基础分类和目标检测的开放和封闭版本，使得能够系统地评估模型在生物学尺度、临床环境和成像模态下的性能。通过评估15个开放权重和2个前沿VLM，我们发现，尽管医学适应为某些模型提供了可测量的提升，但在已建立的基准上报告的高准确率往往掩盖了视觉感知和领域泛化的不足。
View on arXiv Download PDF AI Translation
cs.CV / 16 / 2606.06709

USU-Corn-WeedDB: A UAV RGB Image Dataset for Multi-Species Weed Detection in Forage Corn

USU-Corn-WeedDB：用于饲料玉米中多种杂草检测的无人机RGB图像数据集
Bhandari, Utsav, Burlakoti, Saroj, Miller, Rhonda, Young, Sierra, Westra, Eric, Etienne, Aaron
Abstract
Weed pressure in forage corn production causes yield losses of up to 31.5%, yet site-specific weed management (SSWM) systems built on UAV imagery and deep learning remain constrained by the scarcity of field-representative training datasets. We present USU-Corn-WeedDB, a publicly available UAV RGB image dataset collected from a commercial forage corn field in Cache Valley, Utah, designed to support multi-class weed detection under both supervised and semi-supervised learning frameworks. RGB imagery was acquired on 27 June 2025 using an Autel EVO II Dual 640T V2 drone at ~10m above ground level, yielding a ground sampling distance of approximately 0.48 cm/pixel. A total of 366 full-resolution images were tiled into 8,800 patches at 640 x 640-pixel resolution. Of these, 800 images were manually annotated for three weed species; common lambsquarters (Chenopodium album), redroot pigweed (Amaranthus retroflexus), and green foxtail (Setaria viridis) comprising 10,539 bounding-box instances, with the remaining 8,000 tiles retained as an unlabeled pool for semi-supervised experiments. This dataset reflects a natural class imbalance where redroot pigweed constitutes 53.86% of annotated instances, which was preserved intentionally to mirror real field conditions. To validate dataset utility, we trained 28 object detection models spanning five architecture families including YOLOv8, YOLOv9, YOLOv10, YOLO11, YOLO26, and RT-DETR under identical conditions without hyperparameter tuning. Test set [email protected] ranged from 0.773 to 0.840, with lightweight models achieving competitive performance relevant to edge-deployed UAV systems. USU-Corn-WeedDB is publicly available at https://doi.org/10.5281/zenodo.20044178.
Chinese Translation
饲料玉米生产中的杂草压力导致高达31.5%的产量损失，但基于无人机图像和深度学习的特定地点杂草管理（SSWM）系统仍受到缺乏具有代表性的现场训练数据集的限制。我们提出了USU-Corn-WeedDB，这是一个公开可用的无人机RGB图像数据集，收集自犹他州Cache Valley的一处商业饲料玉米田，旨在支持在监督和半监督学习框架下的多类别杂草检测。RGB图像于2025年6月27日使用Autel EVO II Dual 640T V2无人机在约10米的高度获取，地面采样距离约为0.48厘米/像素。共获取366幅全分辨率图像，并将其切割为8,800个640 x 640像素的补丁。其中，800幅图像被手动标注了三种杂草物种：常见羊角豆（Chenopodium album）、红根猪草（Amaranthus retroflexus）和绿狐尾草（Setaria viridis），共计10,539个边界框实例，其余8,000个补丁保留为未标记池以供半监督实验使用。该数据集反映了自然的类别不平衡，其中红根猪草占标注实例的53.86%，这一点是故意保留的，以反映真实的田间条件。为了验证数据集的实用性，我们在相同条件下训练了28个对象检测模型，涵盖了包括YOLOv8、YOLOv9、YOLOv10、YOLO11、YOLO26和RT-DETR在内的五个架构系列，且未进行超参数调优。测试集的[email protected]范围为0.773到0.840，轻量级模型在边缘部署的无人机系统中表现出竞争力。USU-Corn-WeedDB可在https://doi.org/10.5281/zenodo.20044178公开获取。
View on arXiv Download PDF AI Translation
cs.CV / 17 / 2606.06714

Anchored, Not Graded: Vision-Language Models Fail at Slant-from-Texture Perception

锚定而非评分：视觉-语言模型在纹理倾斜感知中的失败
Zhang, Qian, Golovanevsky, Michal, Domini, Fulvio, Tompkin, James
Abstract
Human perception of surface slant from texture exhibits systematic, graded biases that emerge reliably in psychophysical experiments. Prior work showed that unsupervised CNNs reproduce several human-like biases, while supervised CNNs do not. Do Vision-Language Models (VLMs) exhibit similar competences? Across multiple VLM families and model scales, zero-shot and in-context prompting both produce distinctive failures: slant is predicted at only a small set of anchors (e.g., 0\degree, $\pm$25\degree, $\pm$45\degree) with little dependence on stimulus field of view, optical slant, or surface curvature. Supervised fine-tuning partially remediates the failure, but residual anchoring persists. While success in high-level vision-language benchmarks might not require sensitivity to low-level geometric cues, we interpret anchoring as a failure at the representation-to-output language interface: Not necessarily an absence of geometric encoding, but a failure to express it in a graded form.
Chinese Translation
人类对纹理表面倾斜的感知表现出系统性的、分级的偏差，这些偏差在心理物理实验中可靠地出现。先前的研究表明，无监督卷积神经网络（CNN）再现了几种类似人类的偏差，而有监督的CNN则没有。视觉-语言模型（VLM）是否表现出类似的能力？在多个VLM家族和模型规模中，零样本和上下文提示均产生了显著的失败：倾斜仅在一小组锚点（例如，0°，±25°，±45°）上进行预测，并且几乎不依赖于刺激的视场、光学倾斜或表面曲率。有监督的微调部分缓解了这一失败，但残余的锚定现象依然存在。尽管在高级视觉-语言基准测试中的成功可能不需要对低级几何线索的敏感性，但我们将锚定视为在表征到输出语言接口上的失败：这并不一定是几何编码的缺失，而是未能以分级形式表达它。
View on arXiv Download PDF AI Translation
cs.CV / 18 / 2606.06760

MedSIGHT: Towards Grounded Visual Comprehension in Medical Large Vision-Language Models

MedSIGHT：迈向医学大规模视觉-语言模型的扎根视觉理解
Chang, Aofei, Huang, Le, Boyd, Alex James, Bhatia, Parminder, Kass-Hout, Taha, Ma, Fenglong, Xiao, Cao
Abstract
Medical large vision-language models (Med-LVLMs) have recently achieved remarkable progress in vision-language comprehension and medical image segmentation. However, existing models still struggle to unify these two capabilities, which is essential for achieving clinically reasoning that connects visual findings with semantic interpretation. We present MedSIGHT, a unified framework that equips Med-LVLMs with structured, pixel-level understanding for grounded visual comprehension. MedSIGHT introduces a novel Region Perceiver module that produces region-centric tokens, encoding spatial information directly into representation space of the language model. We further propose a medical region codebook into the LLM vocabulary, allowing the model to generate discrete region codes as symbolic representations of anatomical and pathological regions. These codes are decoded through the Region Perceiver to reconstruct segmentation mask, achieving end-to-end spatial grounding. Lastly, MedSIGHT combines Region Perceiver, Codebook and LLM using our proposed progressive training strategy to gradually aligns these modules stably. Trained on only 72K multimodal instruction pairs, MedSIGHT achieves state-of-the-art performance across diverse imaging modalities on both medical comprehension and segmentation tasks.
Chinese Translation
医学大规模视觉-语言模型（Med-LVLMs）最近在视觉-语言理解和医学图像分割方面取得了显著进展。然而，现有模型仍然难以统一这两种能力，而这对于实现将视觉发现与语义解释相连接的临床推理至关重要。我们提出了MedSIGHT，这是一种统一框架，赋予Med-LVLMs结构化的像素级理解，以实现扎根视觉理解。MedSIGHT引入了一种新颖的区域感知模块（Region Perceiver），生成以区域为中心的标记，将空间信息直接编码到语言模型的表示空间中。我们进一步在LLM词汇中提出了医学区域词典，允许模型生成离散的区域代码，作为解剖和病理区域的符号表示。这些代码通过区域感知器解码，以重建分割掩膜，实现端到端的空间扎根。最后，MedSIGHT结合区域感知器、词典和LLM，采用我们提出的渐进式训练策略，逐步稳定地对齐这些模块。在仅使用72K多模态指令对进行训练的情况下，MedSIGHT在医学理解和分割任务的多种成像模态上实现了最先进的性能。
View on arXiv Download PDF AI Translation
cs.CV / 19 / 2606.06813

Breaking the Lock-in: Diversifying Text-to-Image Generation via Representation Modulation

打破锁定：通过表征调制实现文本到图像生成的多样化
Kwon, Dahee, Lee, Haeun, Choi, Jaesik
Abstract
Recent text-to-image models built on large-scale Transformer backbones and flow-based objectives deliver strong text-image alignment and high visual quality, yet often produce overly similar samples under a fixed prompt. Existing diversity-enhancement methods alleviate this issue, but typically require expensive sampling or auxiliary optimization, incurring non-trivial overhead. To investigate the root cause of this homogeneity, we examine intermediate Transformer features and observe that the zero-frequency spatial average (DC) component rapidly converges across seeds early in generation, causing early trajectory lock-in that limits downstream variation. Building on this observation, we propose DC Attenuation for diVersity Enhancement (DAVE), a training-free representation-level intervention that selectively attenuates this component in the early regime. DAVE preserves the sampling pipeline with negligible overhead, improving prompt-consistent diversity while maintaining competitive image quality.
Chinese Translation
近期基于大规模Transformer骨干网络和基于流的目标的文本到图像模型在文本与图像的对齐和视觉质量上表现出色，但在固定提示下往往生成过于相似的样本。现有的多样性增强方法缓解了这一问题，但通常需要昂贵的采样或辅助优化，导致不容小觑的开销。为了探究这种同质性的根本原因，我们检查了中间的Transformer特征，并观察到零频率空间平均（DC）成分在生成初期迅速收敛于不同的种子，从而导致早期轨迹锁定，限制了下游的变化。基于这一观察，我们提出了用于多样性增强的DC衰减（DAVE），这是一种无训练的表征级干预方法，选择性地在早期阶段衰减该成分。DAVE在保持采样流程的同时几乎没有开销，提高了与提示一致的多样性，同时保持了竞争性的图像质量。
View on arXiv Download PDF AI Translation
cs.CV / 20 / 2606.06819

VideoSEG-O3: A Multi-turn Reinforcement Learning Framework for Reasoning Video Object Segmentation

VideoSEG-O3：一种用于推理视频目标分割的多轮强化学习框架
Dai, Ming, Yang, Sen, Duan, Boqiang, Tong, Boyuan, Zhuang, Jiedong, Yang, Wankou, Wang, Jingdong
Abstract
Reasoning Video Object Segmentation (RVOS) demands a sophisticated integration of temporal dynamics, spatial details, and linguistic reasoning to achieve precise pixel-level localization. Existing methods are limited to reasoning over fixed initial inputs and lack the capacity to actively acquire further visual evidence, which is often essential for resolving complex references in long or intricate videos. To address this, we propose \textbf{VideoSEG-O3}, the first multi-turn reinforcement learning framework for RVOS that emulates the human \textit{``coarse-to-fine''} cognitive process. It employs a \textit{multi-turn temporal-spatial chain-of-thought} to capture fine-grained details by iteratively pinpointing critical intervals and keyframes. Additionally, to enable the policy to perceive segmentation quality beyond mere text probability of \texttt{[SEG]} during the RL stage, we introduce \textit{SEG-aware logit calibration}, which integrates pixel-wise segmentation feedback directly into the token-level logits. Furthermore, we design a \textit{decoupled thinking trace} to hierarchically decompose the reasoning process into temporal, spatial, and linguistic dimensions, and construct \textbf{VTS-CoT}, a specialized cold-start dataset featuring comprehensive reasoning trajectories. The code and models will be released at https://github.com/Dmmm1997/VideoSEG-O3.
Chinese Translation
推理视频目标分割（RVOS）需要对时间动态、空间细节和语言推理进行复杂的整合，以实现精确的像素级定位。现有方法仅限于对固定初始输入进行推理，缺乏主动获取进一步视觉证据的能力，而这在处理长视频或复杂视频中的复杂引用时往往是必不可少的。为了解决这个问题，我们提出了 extbf{VideoSEG-O3}，这是第一个模拟人类 extit{“粗到细”}认知过程的多轮强化学习框架。它采用 extit{多轮时空思维链}通过迭代确定关键时间段和关键帧来捕捉细粒度细节。此外，为了使策略在强化学习阶段能够感知分割质量，而不仅仅是 exttt{[SEG]}的文本概率，我们引入了 extit{SEG感知对数校准}，将逐像素分割反馈直接整合到标记级对数中。此外，我们设计了 extit{解耦思维轨迹}，将推理过程分解为时间、空间和语言维度，并构建了 extbf{VTS-CoT}，这是一个具有全面推理轨迹的专用冷启动数据集。代码和模型将发布在 https://github.com/Dmmm1997/VideoSEG-O3。
View on arXiv Download PDF AI Translation
cs.CV / 21 / 2606.06828

AdaGRPO: A Capability-Aware Adaptive Enhancement for Flow-based GRPO

AdaGRPO：一种基于能力的自适应增强方法用于流式 GRPO
Bu, Jiazi, Ling, Pengyang, Zhou, Yujie, Wang, Yibin, Zang, Yuhang, Wei, Tianyi, Zhan, Xiaohang, Wang, Jiaqi, Wu, Tong, Pan, Xingang, Lin, Dahua
Abstract
Group Relative Policy Optimization (GRPO) has demonstrated remarkable success in aligning text-to-image (T2I) flow models with human preferences. However, we have identified that the learning loop of current flow-based GRPO is fundamentally decoupled from the learner's current capability, suffering from critical blind spots at both prompt selection and advantage estimation: (i) Existing methods sample prompts randomly, overlooking the substantial impact of data selection on reinforcement learning (RL) efficacy--a factor proven crucial in GRPO for large language models; (ii) They evaluate sample quality solely relying on intra-group statistics, lacking a global perspective to accurately measure true policy improvement. To address these issues, we propose Adaptive GRPO (AdaGRPO), a novel capability-aware RL algorithm tailored for flow models. Specifically, AdaGRPO consists of two principal components: (i) Online Curriculum Filtering Strategy: Dynamically tracks the model's proficiency and adaptively selects prompts that best match its current learning boundary; (ii) Cross-Level Advantage Fusion: Synergistically integrates fine-grained intra-group advantages with macro-level global advantages, providing a comprehensive and unbiased policy evaluation. As a lightweight, plug-and-play module, AdaGRPO can be seamlessly integrated with existing frameworks such as Flow-GRPO, DanceGRPO, and Flow-CPS. Extensive experiments demonstrate that AdaGRPO consistently drives performance gains while significantly stabilizes GRPO training for flow models.
Chinese Translation
群体相对策略优化（GRPO）在将文本到图像（T2I）流模型与人类偏好对齐方面取得了显著成功。然而，我们发现当前基于流的 GRPO 的学习循环与学习者当前能力之间存在根本的脱节，在提示选择和优势估计方面存在关键盲点：（i）现有方法随机采样提示，忽视了数据选择对强化学习（RL）效果的重大影响——这一因素在大语言模型的 GRPO 中被证明至关重要；（ii）它们仅依赖组内统计来评估样本质量，缺乏全球视角以准确衡量真实策略的改进。为了解决这些问题，我们提出了自适应 GRPO（AdaGRPO），这是一种针对流模型的新型基于能力的 RL 算法。具体而言，AdaGRPO 由两个主要组成部分构成：（i）在线课程过滤策略：动态跟踪模型的熟练程度，并自适应选择与其当前学习边界最匹配的提示；（ii）跨层优势融合：协同整合细粒度的组内优势与宏观层面的全球优势，提供全面且无偏的策略评估。作为一个轻量级的即插即用模块，AdaGRPO 可以与现有框架如 Flow-GRPO、DanceGRPO 和 Flow-CPS 无缝集成。大量实验表明，AdaGRPO 一直推动性能提升，同时显著稳定流模型的 GRPO 训练。
View on arXiv Download PDF AI Translation
cs.CV / 22 / 2606.06850

CFRNet: Cycle-Consistent Fixed-Point Training for Real-Time Blind Face Restoration on Consumer Embedded NPUs

CFRNet：用于消费嵌入式神经处理单元的实时盲人脸修复的循环一致性定点训练
Li, Fuchen, Wang, Xinyang, Zhang, Yahui, Chen, Yuhan, Guo, Jiahong, Qin, Zhuohan, Ma, Wenbo
Abstract
Blind face restoration on consumer devices has to balance image quality against speed and memory. Strong methods such as GFPGAN and CodeFormer give good perceptual quality, but they rely on large pretrained generative priors and on operators such as attention, codebook lookup, and style modulation that are hard to compile and quantize on the small neural processing units (NPUs) used in consumer hardware. Small convolutional restorers run fast enough, but they tend to over-smooth and to leave artifacts around the eyes, nose, and mouth. We present CFRNet, a 2.0,M-parameter ResNet-style restorer for on-device use at $256\times256$, the common face-crop size on consumer NPUs. The main idea is Cycle-Consistent Fixed-Point Training (CCFP). Instead of training the network for one pass and then running it several times by hand, we train it to act as a fixed-point operator, so that applying it again to a restored face does not change the face. CCFP uses three training losses, namely progressive multi-cycle supervision, an idempotence loss, and a re-degradation cycle loss, and it adds no cost at inference. To compare fairly under our deployment limits, we retrain all baselines from scratch at the same $256\times256$ resolution. On a 300-image test set, CFRNet reaches the best perceptual score (LPIPS 0.250 at three cycles, which is 31% lower than one cycle) and also the best PSNR and SSIM at two cycles. It runs in about 23,ms per cycle in INT8 on a HiSilicon Hi3402 NPU, while the same baselines cannot be compiled to that chip. The cycle count $k$ acts as a simple quality knob that needs no retraining: PSNR is best at $k\!=\!2$ and LPIPS keeps improving up to $k\!=\!3$. We further show that the same idea works with a plain CNN that is even easier to deploy, and we run the model in real time on an in-car driver-monitoring board.
Chinese Translation
在消费设备上进行盲人脸修复需要在图像质量、速度和内存之间取得平衡。像GFPGAN和CodeFormer这样强大的方法能够提供良好的感知质量，但它们依赖于大型预训练生成先验以及难以在消费硬件的小型神经处理单元（NPU）上编译和量化的操作符，如注意力机制、代码本查找和风格调制。小型卷积修复器运行速度足够快，但往往会过度平滑，并在眼睛、鼻子和嘴巴周围留下伪影。我们提出了CFRNet，一种具有2.0M参数的ResNet风格修复器，适用于在$256 imes256$的设备上使用，这是消费NPU上常见的人脸裁剪尺寸。其主要思想是循环一致性定点训练（CCFP）。我们不再仅仅训练网络进行一次传递，然后手动多次运行，而是训练其作为一个定点操作符，使得对修复后的人脸再次应用该网络不会改变人脸。CCFP使用三种训练损失，即渐进式多周期监督、幂等损失和再退化周期损失，并且在推理时没有额外成本。为了在我们的部署限制下进行公平比较，我们在相同的$256 imes256$分辨率下从头开始重新训练所有基线。在一个包含300张图像的测试集上，CFRNet达到了最佳的感知分数（LPIPS 0.250，经过三次周期，比一次周期低31%），同时在两个周期下也达到了最佳的PSNR和SSIM。它在HiSilicon Hi3402 NPU上以INT8格式每个周期运行约23毫秒，而相同的基线无法编译到该芯片上。周期数$k$作为一个简单的质量调节器，无需重新训练：PSNR在$k eq2$时最佳，而LPIPS在$k eq3$时持续改善。我们进一步展示了相同的思想也适用于一个更易于部署的普通CNN，并且我们在车载驾驶员监控板上实时运行该模型。
View on arXiv Download PDF AI Translation
cs.CV / 23 / 2606.06853

MotionEnhancer: Leveraging Video Diffusion for Motion-Enhanced Vision-Language Models

MotionEnhancer：利用视频扩散提升运动增强的视觉-语言模型
Xu, Yifan, Zhang, Chao, Ma, Ruifei, Gao, Fei, Yang, Zhifei, Qi, Jiaxing, Chen, Zhipeng
Abstract
The new era has witnessed a remarkable capability to extend Vision-Language Models (VLMs) for tackling tasks of video understanding. While current VLMs excel at event- or story-level understanding, their ability to capture fine-grained motion details remains limited, primarily due to their focus on high-level static semantic structures and macro-event logic. In contrast, Video Diffusion Models (VDMs) are adept at modeling dynamic motion patterns, benefiting from large-scale video data and the intrinsic requirement of temporal generation. In this paper, we introduce MotionEnhancer, a novel approach that leverages motion priors distilled from a powerful video diffusion model as auxiliary supervision to enhance the motion understanding capability of a VLM via attention alignment. MotionEnhancer comprises two simple parameter-free modules, Motion-sensitive Head Selection (MHS) and Motion-salient Text Token Identification (MTTI), to directly extract and optimize motion-related attentions from the VDM in a computation-only manner. MotionEnhancer provides a scalable solution for motion understanding without additional training parameters, modifications to existing architectures, or tool calling. Extensive experiments demonstrate that MotionEnhancer can achieve consistent improvements over state-of-the-art VLMs on two motion-level video understanding benchmarks, especially on motion-related metrics.
Chinese Translation
新时代见证了视觉-语言模型（VLMs）在处理视频理解任务方面的显著能力。尽管当前的VLMs在事件或故事层面的理解上表现出色，但它们捕捉细粒度运动细节的能力仍然有限，主要是因为它们专注于高层次的静态语义结构和宏观事件逻辑。相比之下，视频扩散模型（VDMs）擅长建模动态运动模式，受益于大规模视频数据和时间生成的内在要求。在本文中，我们介绍了MotionEnhancer，这是一种新颖的方法，利用从强大的视频扩散模型中提取的运动先验作为辅助监督，通过注意力对齐来增强VLM的运动理解能力。MotionEnhancer包含两个简单的无参数模块：运动敏感头选择（MHS）和运动显著文本标记识别（MTTI），以计算方式直接提取和优化来自VDM的运动相关注意力。MotionEnhancer提供了一种可扩展的运动理解解决方案，无需额外的训练参数、对现有架构的修改或工具调用。大量实验表明，MotionEnhancer在两个运动级别的视频理解基准上能够在最先进的VLMs上实现一致的改进，尤其是在与运动相关的指标上。
View on arXiv Download PDF AI Translation
cs.CV / 24 / 2606.06856

FS-DVS: A Frequency-Selective Dynamic Visual Sensing Paradigm for Enhancing Information Completeness

FS-DVS：一种频率选择性动态视觉感知范式以增强信息完整性
Ji, Feiyu, Yang, Xiaokang, Yuan, Xiaoyun
Abstract
Dynamic vision sensors (DVS) offer exceptional temporal resolution and dynamic range by asynchronously reporting pixel-level intensity changes. However, conventional DVS rely on a per-pixel independent triggering mechanism, ignoring the spatial integration performed by biological retinal ganglion cells (RGCs). Consequently, they lack the contrast sensitivity function (CSF) and its inherent sensitivity to mid-spatial frequencies, which inevitably leads to information incompleteness due to sub-threshold signal loss. To bridge this gap, we propose FS-DVS (Frequency-Selective Dynamic Vision Sensor), a novel paradigm that integrates a learnable spatial filter strictly preceding the event triggering process to mimic the RGC aggregation mechanism. By developing a differentiable event simulation framework, the spatial filter can be optimized end-to-end with downstream tasks. Our study reveals that starting from a delta function, the learned spatial filters spontaneously evolve into center-surround patterns that emphasize mid-frequency components, consistently aligning with human CSF. Beyond achieving substantial performance gains in object detection and action recognition, the consistent convergence to human-like CSF characteristics across different tasks underscores the universality of this mid-frequency selective mechanism. Compared to naively increasing sensor sensitivity or relying on post-processing, our paradigm achieves selective information enhancement with high noise resilience, providing a robust, biologically plausible blueprint for next-generation neuromorphic sensors.
Chinese Translation
动态视觉传感器（DVS）通过异步报告像素级强度变化，提供了卓越的时间分辨率和动态范围。然而，传统的DVS依赖于每个像素独立的触发机制，忽视了生物视网膜神经节细胞（RGC）所执行的空间整合。因此，它们缺乏对比敏感度函数（CSF）及其对中频率的固有敏感性，这不可避免地导致了由于阈下信号丢失而造成的信息不完整。为了解决这一问题，我们提出了FS-DVS（频率选择性动态视觉传感器），这是一种新颖的范式，集成了一个可学习的空间滤波器，严格位于事件触发过程之前，以模拟RGC聚合机制。通过开发一个可微分的事件仿真框架，空间滤波器可以与下游任务进行端到端优化。我们的研究表明，从一个δ函数开始，学习到的空间滤波器自发演变为强调中频成分的中心-周围模式，始终与人类的CSF一致。除了在物体检测和动作识别中实现显著的性能提升外，不同任务中一致收敛到类人CSF特征也凸显了这一中频选择机制的普遍性。与简单地提高传感器灵敏度或依赖后处理相比，我们的范式实现了选择性信息增强，并具有较高的抗噪声能力，为下一代神经形态传感器提供了一个稳健且生物合理的蓝图。
View on arXiv Download PDF AI Translation
cs.CV / 25 / 2606.06864

LRMIL: Efficient Low-Resolution Multiple Instance Learning via High-Resolution Knowledge Distillation for Whole Slide Image Classification

LRMIL：通过高分辨率知识蒸馏实现高效低分辨率多实例学习用于全切片图像分类
Shin, Yonghan, Jeong, Won-Ki
Abstract
Multiple instance learning (MIL) has become a standard paradigm for whole slide image (WSI) analysis in digital pathology, as it enables slide-level prediction without dense annotations. Existing MIL methods typically rely on exhaustive extraction and encoding of high-resolution patches. However, this practice suffers from two critical limitations in real-world clinical settings: it struggles to capture global visual cues at lower magnifications, and incurs substantial computational overhead due to the massive number of high-resolution patches per slide. To address these limitations, we propose an efficient low-resolution multiple instance learning (LRMIL) framework that transfers high-resolution knowledge to low-resolution representations. LRMIL adopts a two-stage distillation strategy. First, patch-level cross-resolution distillation aligns low-resolution patch embeddings with high-resolution representations. Second, slide-level knowledge distillation trains a low-resolution student MIL model under both slide-level supervision and teacher guidance. At inference time, LRMIL operates exclusively on low-resolution patches, substantially reducing data preprocessing and computational cost. Extensive experiments on multiple WSI benchmarks demonstrate that LRMIL consistently outperforms state-of-the-art MIL methods while achieving more efficient inference. These results highlight LRMIL as a practical and scalable solution for WSI analysis in clinical pathology.
Chinese Translation
多实例学习（MIL）已成为数字病理学中全切片图像（WSI）分析的标准范式，因为它能够在没有密集注释的情况下进行切片级预测。现有的MIL方法通常依赖于对高分辨率补丁的全面提取和编码。然而，这种做法在实际临床环境中存在两个关键限制：在较低放大倍数下难以捕捉全局视觉线索，并且由于每个切片中高分辨率补丁的数量庞大，导致显著的计算开销。为了解决这些限制，我们提出了一种高效的低分辨率多实例学习（LRMIL）框架，该框架将高分辨率知识转移到低分辨率表示中。LRMIL采用了两阶段的蒸馏策略。首先，补丁级跨分辨率蒸馏将低分辨率补丁嵌入与高分辨率表示对齐。其次，切片级知识蒸馏在切片级监督和教师指导下训练低分辨率学生MIL模型。在推理时，LRMIL仅在低分辨率补丁上操作，显著减少了数据预处理和计算成本。在多个WSI基准上的广泛实验表明，LRMIL始终优于最先进的MIL方法，同时实现了更高效的推理。这些结果突显了LRMIL作为临床病理学中WSI分析的实用和可扩展解决方案。
View on arXiv Download PDF AI Translation
cs.CV / 26 / 2606.06867

Multi-FRuGaL: Multimodal Flexible Redundancy-aware Decomposed Gated Learning for Cancer Diagnosis and Prognosis

多模态灵活冗余感知分解门控学习用于癌症诊断和预后
Kachole, Sanket, Thakur, Siddhesh, Innani, Shubham, Adap, Sanyukta, You, Suhang, Pitarch-Abaigar, Carla, Bakas, Spyridon
Abstract
Modern medicine relies on heterogeneous data sources spanning radiology, pathology, text reports, and structured clinical information. However, real-world patient data are frequently incomplete, with missing or sparsely acquired modalities, limiting the effectiveness of standard multimodal fusion approaches. To this end, we propose the Multimodal Flexible Redundancy-aware decomposed GAted Learning (Multi-FRuGaL) framework, a decomposition-aware, adaptive gated intermediate-fusion framework that performs modality-level representation learning under missing data. Multi-FRuGaL integrates per-modality encoders with a signal decomposition layer, an input-conditioned gating network, and an information-aware fusion objective to separate redundant from modality-specific complementary signals, selectively upweighting informative modalities and suppressing redundant or noisy inputs, and remaining well-defined even when multiple modalities are absent. We evaluate Multi-FRuGaL on two multimodal head and neck cancer cohorts: the HANCOCK challenge dataset (N = 763) comprising five modalities and two prognostic endpoints (5-year survival and 2-year recurrence), and the HECKTOR challenge dataset (N = 588) comprising three modalities for human papillomavirus (HPV) status classification. Multi-FRuGaL consistently achieves higher mean performance than the evaluated baselines across multiple tasks, improving AUC from 0.601 to 0.8496 for survival, from 0.672 to 0.8102 for recurrence, and achieving 0.975 AUC for HPV prediction on HECKTOR. For survival analysis, it further achieves a concordance index of 0.6814 for overall survival, 0.7421 for recurrence-free survival, and 0.7143 for progression-free survival on HANCOCK, and 0.7203 for recurrence-free survival on HECKTOR. Qualitative analyses further show that Multi-FRuGaL learns discriminative and robust multimodal representations, even under severe missing-modality conditions.
Chinese Translation
现代医学依赖于跨越放射学、病理学、文本报告和结构化临床信息的异构数据源。然而，现实世界中的患者数据常常不完整，缺失或稀疏获取的模态限制了标准多模态融合方法的有效性。为此，我们提出了多模态灵活冗余感知分解门控学习（Multi-FRuGaL）框架，这是一个感知分解的自适应门控中间融合框架，能够在缺失数据的情况下进行模态级表示学习。Multi-FRuGaL将每个模态的编码器与信号分解层、输入条件门控网络和信息感知融合目标相结合，以区分冗余信号和模态特异性互补信号，选择性地增强信息丰富的模态，抑制冗余或噪声输入，即使在多个模态缺失的情况下也保持良好的定义。我们在两个多模态头颈癌队列上评估了Multi-FRuGaL：HANCOCK挑战数据集（N = 763），包括五种模态和两个预后终点（5年生存率和2年复发率），以及HECKTOR挑战数据集（N = 588），包括三种模态用于人乳头瘤病毒（HPV）状态分类。Multi-FRuGaL在多个任务中始终实现了比评估基线更高的平均性能，生存率的AUC从0.601提高到0.8496，复发率的AUC从0.672提高到0.8102，并在HECKTOR上实现了0.975的HPV预测AUC。在生存分析中，它在HANCOCK上进一步实现了总体生存的协调指数为0.6814，复发无生存的协调指数为0.7421，进展无生存的协调指数为0.7143，以及在HECKTOR上实现了复发无生存的协调指数为0.7203。定性分析进一步表明，Multi-FRuGaL即使在严重缺失模态条件下也能学习到具有区分性和鲁棒性的多模态表示。
View on arXiv Download PDF AI Translation
cs.CV / 27 / 2606.06872

EgoPressDiff: Multimodal Video Diffusion for Egocentric UV-Domain Hand-Pressure Estimation

EgoPressDiff：用于自我中心UV域手部压力估计的多模态视频扩散
Zeng, Yuan, Gao, Zilue, Shi, Yujia, Lu, Zongqing, Yang, Wenming, Liao, QingMin
Abstract
Estimating hand-surface contact pressure from an egocentric view is crucial for AR/VR devices, robotic imitation, and ergonomic analysis. Existing methods often discretize pressure signal and process frames independently, leading to quantization errors and temporal inconsistencies. We present \emph{EgoPressDiff}, a conditional video diffusion framework that generates UV-pressure maps from visual input. The core of our approach is a multi-modal conditioning strategy, introducing a PoseNet and a Vertex Encoder to efficiently extract features from hand pose and 3D mesh vertices. These signals, along with depth information, guide the generative process to ensure the pressure fields are physically grounded. To effectively fuse these heterogeneous features, we further propose a Distribution-Calibrated Spatial Layer, which aligns their statistical properties before combination. Evaluated on the EgoPressure ego-view setting, EgoPressDiff achieves state-of-the-art results, improving Volumetric IoU by over 34\% relative to prior baseline, while reducing MAE and maintaining high temporal accuracy. Our project page is at https://egopressdiff.github.io/.
Chinese Translation
从自我中心视角估计手部与表面的接触压力对于增强现实/虚拟现实（AR/VR）设备、机器人模仿和人体工程学分析至关重要。现有方法通常将压力信号离散化，并独立处理帧，导致量化误差和时间不一致性。我们提出了 extit{EgoPressDiff}，一种条件视频扩散框架，能够从视觉输入生成UV压力图。我们方法的核心是多模态条件策略，引入了PoseNet和Vertex Encoder，以高效提取手部姿态和3D网格顶点的特征。这些信号以及深度信息共同指导生成过程，确保压力场在物理上是合理的。为了有效融合这些异构特征，我们进一步提出了分布校准空间层（Distribution-Calibrated Spatial Layer），在组合之前对其统计特性进行对齐。在EgoPressure自我视角设置下进行评估时，EgoPressDiff实现了最先进的结果，相较于之前的基线，体积交并比（Volumetric IoU）提高了34%以上，同时减少了平均绝对误差（MAE）并保持了高时间准确性。我们的项目页面为 https://egopressdiff.github.io/。
View on arXiv Download PDF AI Translation
cs.CV / 28 / 2606.06875

Unified Safe In-context Image Generation in Multimodal Diffusion Transformers via Restricting Unsafe Information Flows

通过限制不安全信息流在多模态扩散变换器中实现统一的安全上下文图像生成
Yang, Xiang, Li, Feifei, Zhang, Mi, Hong, Geng, You, Xiaoyu, Wen, Mi, Yang, Min
Abstract
Diffusion transformers (DiTs) equipped with multimodal attention (MM-Attn) have become a dominant paradigm for image generation. However, preventing the generation of harmful content remains a critical challenge, particularly in image-to-image (I2I) editing tasks. Existing safety mechanisms are primarily designed for text-to-image (T2I) synthesis or U-Net-based architectures, which limits their effectiveness for unified safety mitigation in DiT-based frameworks. To bridge this gap, we propose Unified Visual Safety Regulator (UVR), a training-free safe generation framework that regulates unsafe semantics in generated images. UVR is grounded in an analysis of attention dynamics from the perspective of information flow in MM-Attn. We identify a task-independent start-up stage, during which unsafe semantics in output patches rapidly emerge and can be accurately localized, followed by task-specific semantic amplification and interference stages, where harmful signals are further propagated and entangled with benign content. Based on these observations, UVR mitigates unsafe generation through unified, targeted attention modulation and explicit restriction of harmful information flow over the identified unsafe output patches. Experiments across various concepts show that UVR achieves state-of-the-art safety performance by achieving 91% and 77% erase rate in image synthesis and editing tasks, while preserving visual quality and fidelity with minimal degradation. Code is available at https://github.com/deng12yx/UVR.
Chinese Translation
配备多模态注意力（MM-Attn）的扩散变换器（DiTs）已成为图像生成的主导范式。然而，防止生成有害内容仍然是一个关键挑战，特别是在图像到图像（I2I）编辑任务中。现有的安全机制主要针对文本到图像（T2I）合成或基于U-Net的架构设计，这限制了它们在DiT框架中实现统一安全缓解的有效性。为了解决这一问题，我们提出了统一视觉安全调节器（Unified Visual Safety Regulator, UVR），这是一个无训练的安全生成框架，用于调节生成图像中的不安全语义。UVR基于对MM-Attn中信息流视角下的注意力动态分析而建立。我们识别出一个与任务无关的启动阶段，在此阶段中，输出补丁中的不安全语义迅速出现并可以被准确定位，随后是任务特定的语义放大和干扰阶段，在这些阶段中，有害信号进一步传播并与良性内容交织。基于这些观察，UVR通过统一的、针对性的注意力调制和对识别出不安全输出补丁的有害信息流的明确限制来缓解不安全生成。各种概念的实验表明，UVR在图像合成和编辑任务中实现了91%和77%的消除率，达到了最先进的安全性能，同时在视觉质量和保真度方面保持了最小的降级。代码可在 https://github.com/deng12yx/UVR 获取。
View on arXiv Download PDF AI Translation
cs.CV / 29 / 2606.06885

FreeAnimate: Training-Free Human Image Animation with Preview-Guided Denoising

FreeAnimate：无训练的人像动画生成与预览引导去噪
Zeng, Yuan, Shi, Yujia, Lu, Zongqing, Liao, QingMin
Abstract
Human Image Animation has seen significant advancements, primarily driven by diffusion models. However, existing methods typically demand substantial training data and resources to achieve high-quality results, limiting generalization and accessibility. In this work, we introduce \emph{FreeAnimate}, a training-free framework that leverages the inherent capabilities of image diffusion models to enable temporal consistency, identity preservation, and background stability. Our approach incorporates a novel preview generation strategy that provides temporal and structural priors from generated preview frames, effectively guiding pose alignment and background consistency without training. Additionally, FreeAnimate introduces Inversion-Boosted Attention and Reference-Anchored Self-Attention modules to guarantee temporal consistency and identity preservation. Experimental results demonstrate that FreeAnimate outperforms existing training-free competitors and training-based baseline methods, achieving generation quality comparable to state-of-the-art methods and offering robust generalization across diverse datasets. Our project page is at https://freeani.github.io/.
Chinese Translation
人像动画技术已经取得了显著进展，主要得益于扩散模型。然而，现有方法通常需要大量的训练数据和资源才能实现高质量的结果，这限制了其泛化能力和可及性。在本研究中，我们提出了 extit{FreeAnimate}，一个无训练的框架，利用图像扩散模型的内在能力，实现时间一致性、身份保持和背景稳定。我们的方法结合了一种新颖的预览生成策略，从生成的预览帧中提供时间和结构先验，有效地引导姿态对齐和背景一致性，而无需训练。此外，FreeAnimate引入了反演增强注意力和参考锚定自注意力模块，以确保时间一致性和身份保持。实验结果表明，FreeAnimate在性能上优于现有的无训练竞争者和基于训练的基线方法，生成质量可与最先进的方法相媲美，并在多样化数据集上提供了强大的泛化能力。我们的项目页面位于 https://freeani.github.io/。
View on arXiv Download PDF AI Translation
cs.CV / 30 / 2606.06887

ARAPDiffusion: ARAP Regularization for Diffusion-Based Deformable Shape Space Learning

ARAPDiffusion：基于ARAP正则化的扩散式可变形形状空间学习
Liu, Haibo, Ke, Jinghan, Yang, Haitao, Huang, Xiangru, Pavlakos, Georgios, Huang, Qixing
Abstract
This paper introduces ARAPDiffusion, a latent diffusion model to learn the underlying continuous shape space of a deformation shape collection. The key innovation is in injecting the as-rigid-as-possible (ARAP) deformation model as regularization losses into latent diffusion (LD), releasing the requirement of having abundant 3D training data for learning generative models. In contrast to the standard LD, we show how the ARAP model can be used to improve both the encoder/decoder and the LD model. The training procedure alternates between using the synthetic distribution defined by the LD model to develop a regularization loss that enhances the shape encoder/decoder and using the shape decoder to develop a regularization loss to improve the LD model. We also show the benefit of the LD paradigm in combining a representation-free LD process and an implicit shape decoder that is applicable to unorganized point clouds. The experimental results of unconditional and conditional shape generation demonstrate the advantages of ARAPDiffusion over baseline approaches.
Chinese Translation
本文介绍了ARAPDiffusion，一种潜在扩散模型，用于学习变形形状集合的潜在连续形状空间。关键创新在于将尽可能刚性（as-rigid-as-possible，ARAP）变形模型作为正则化损失注入潜在扩散（latent diffusion，LD）中，从而释放了对丰富3D训练数据的需求，以学习生成模型。与标准的LD相比，我们展示了ARAP模型如何改善编码器/解码器和LD模型。训练过程交替进行，利用LD模型定义的合成分布来开发增强形状编码器/解码器的正则化损失，同时利用形状解码器开发改善LD模型的正则化损失。我们还展示了LD范式在结合无表示的LD过程和适用于无序点云的隐式形状解码器方面的优势。无条件和条件形状生成的实验结果表明，ARAPDiffusion相较于基线方法具有显著优势。
View on arXiv Download PDF AI Translation
cs.CV / 31 / 2606.06890

Diagnosing Visual Ignorance in Vision-Language Models

视觉-语言模型中的视觉无知诊断
Zhou, Runyu, Zhang, Qi, Wang, Qixun, Wang, Yisen
Abstract
Vision-Language Models (VLMs) frequently rely on language priors, producing confident answers that are weakly grounded in visual evidence. While this behavior is widely observed, its internal mechanisms and its impact on benchmark evaluation remain insufficiently understood. In this work, we study language-prior reliance from both mechanistic and behavioral perspectives. Internally, we combine counterfactual layer replacement with supervised layer-wise MLP probing to trace how ground-truth visual semantics and language-prior semantics compete across the language decoder. Our analysis reveals a multi-stage bottleneck: intermediate layers often fail to effectively retrieve visual information, while later layers can further suppress surviving visual signals in favor of text-space biases. Externally, we introduce a progressive visual decay metric based on multi-step Gaussian blurring, which identifies instances whose answers remain invariant even as visual content is increasingly destroyed. Across twelve visual question-answering benchmarks and three representative VLMs, we find that a substantial fraction of examples remain answerable under severe or total visual obfuscation, indicating that current benchmarks can inadvertently reward visual ignorance. These findings demonstrate that language-prior reliance is a systematic routing failure affecting both model internals and benchmark validity. Finally, we outline critical pathways for future research, highlighting the necessity of designing training distributions and evaluation protocols built on structurally isolated or counterfactual data to enforce genuine cross-modal grounding.
Chinese Translation
视觉-语言模型（VLMs）通常依赖于语言先验，产生的自信答案在视觉证据上却缺乏坚实的基础。尽管这种行为被广泛观察到，但其内部机制及其对基准评估的影响仍然理解不足。在本研究中，我们从机制和行为两个角度研究语言先验的依赖性。在内部，我们结合反事实层替换与监督层级多层感知机（MLP）探测，追踪真实视觉语义与语言先验语义在语言解码器中的竞争。我们的分析揭示了一个多阶段瓶颈：中间层往往无法有效检索视觉信息，而后续层则可能进一步抑制幸存的视觉信号，以偏向文本空间的偏见。在外部，我们引入了一种基于多步高斯模糊的渐进视觉衰减度量，识别出即使视觉内容逐渐被破坏，其答案仍然保持不变的实例。在十二个视觉问答基准和三个代表性VLMs中，我们发现相当一部分示例在严重或完全视觉模糊的情况下仍然可以回答，这表明当前基准可能无意中奖励了视觉无知。这些发现表明，语言先验的依赖性是一种系统性的路由失败，影响模型内部和基准的有效性。最后，我们概述了未来研究的关键路径，强调设计基于结构孤立或反事实数据的训练分布和评估协议以强制实施真正的跨模态基础的重要性。
View on arXiv Download PDF AI Translation
cs.CV / 32 / 2606.06891

Stream3D-VLM: Online 3D Spatial Understanding with Incremental Geometry Priors

Stream3D-VLM：基于增量几何先验的在线3D空间理解
Yu, Hanxun, Qu, Xuan, Ke, Lei, Zhang, Boqiang, Wang, Yuxin, Zhu, Jianke, Yu, Dong
Abstract
Despite advances in 3D scene understanding, existing 3D Large Multimodal Models operate in offline settings, requiring complete scene observations or predefined video clips. In this paper, we present an online 3D vision-language model that enables real-time spatial understanding from streaming video. Our approach adopts an autoregressive streaming control modeling based on the LLM's next-token prediction objective to learn when to respond, and employs a lightweight Visual-Spatial Feature Integration (VSFI) module to incrementally inject temporally aligned geometry priors into the visual stream. To alleviate long-context decoding overhead, we propose a plug-and-play Geometry-Adaptive Voxel Compression (GAVC) module for efficient visual token compression. To address the scarcity of streaming 3D-language data, we further develop a scalable data generation pipeline that curates over 1M online spatio-temporal 3D QA pairs and establishes a comprehensive benchmark spanning 29 tasks. Extensive experiments show that our approach significantly outperforms both proprietary and open-source models across online and offline 3D spatial understanding, reasoning, and grounding tasks. The project page is available at https://stream3d-vlm.github.io/
Chinese Translation
尽管3D场景理解已有所进展，现有的3D大型多模态模型仍在离线环境中运行，需依赖完整的场景观测或预定义的视频片段。本文提出了一种在线3D视觉-语言模型，能够从流媒体视频中实现实时空间理解。我们的方法采用基于大语言模型（LLM）下一个标记预测目标的自回归流控制建模，以学习何时响应，并使用轻量级的视觉-空间特征集成（VSFI）模块逐步将时间对齐的几何先验注入视觉流中。为减轻长上下文解码的开销，我们提出了一种即插即用的几何自适应体素压缩（GAVC）模块，以实现高效的视觉标记压缩。为了解决流媒体3D语言数据的稀缺问题，我们进一步开发了一个可扩展的数据生成管道，策划了超过100万个在线时空3D问答对，并建立了涵盖29个任务的综合基准。大量实验表明，我们的方法在在线和离线3D空间理解、推理和定位任务中显著优于专有和开源模型。项目页面可访问：https://stream3d-vlm.github.io/
View on arXiv Download PDF AI Translation
cs.CV / 33 / 2606.06899

Lighting-Aware Representation Learning under Controllable Lighting Variation

可控光照变化下的光照感知表示学习
Zhu, Lizhen, Pochimireddy, Charantej Reddy, Wang, James Z, Wyble, Brad
Abstract
Variations in illumination remain a major challenge for visual representation learning, as they induce substantial appearance changes both across and within environments. While existing approaches typically address this issue through data augmentations that encourage models to become invariant to lighting changes, such strategies do not explicitly model lighting information during learning. Inspired by theories of human vision, we propose a lighting-aware representation learning framework that incorporates illumination variation as an explicit training signal rather than a nuisance factor to be suppressed. Our method extends contrastive learning by introducing an auxiliary objective that captures illumination-dependent variation in rendered scenes, enabling the model to jointly learn representations that preserve semantic consistency while remaining sensitive to lighting-dependent visual structure. We evaluate the proposed model on image classification and object detection tasks across the ImageNet, ExDark, and PASCAL VOC benchmarks. Results demonstrate that the proposed lighting-aware training consistently improves downstream performance over standard contrastive learning baselines, while maintaining the same architecture and training budget. Furthermore, our approach shows promising performance in supervised learning frameworks and under settings involving simpler lighting variation, suggesting broad applicability beyond complex illumination scenarios. These results indicate its potential to enhance model robustness and adaptability in complex visual environments as well as in more conventional image processing tasks.
Chinese Translation
光照变化仍然是视觉表示学习的一个主要挑战，因为它们在环境之间和环境内部引起了显著的外观变化。虽然现有的方法通常通过数据增强来解决这个问题，以鼓励模型对光照变化保持不变，但这些策略并没有在学习过程中显式建模光照信息。受到人类视觉理论的启发，我们提出了一种光照感知表示学习框架，将光照变化作为一个显式的训练信号，而不是一个需要抑制的干扰因素。我们的方法通过引入一个辅助目标来扩展对比学习，该目标捕捉渲染场景中依赖于光照的变化，使模型能够共同学习保持语义一致性的表示，同时对依赖于光照的视觉结构保持敏感。我们在ImageNet、ExDark和PASCAL VOC基准上评估了所提出的模型在图像分类和目标检测任务中的表现。结果表明，所提出的光照感知训练在下游任务中始终优于标准对比学习基线，同时保持相同的架构和训练预算。此外，我们的方法在监督学习框架和涉及更简单光照变化的设置下表现出良好的性能，表明其在复杂光照场景之外的广泛适用性。这些结果表明其在复杂视觉环境以及更传统的图像处理任务中增强模型鲁棒性和适应性的潜力。
View on arXiv Download PDF AI Translation
cs.CV / 34 / 2606.06901

LUCID: Learning Unified Control for Image Deflaring and Exposure Mastery in Nighttime Photography

LUCID：统一控制的学习用于夜间摄影中的图像去眩光和曝光掌控
Yang, Tingyu, Cheng, Yuan, Yuan, Xiaoyun
Abstract
Photography is the art of painting with light, yet nighttime scenes are shaped by competing degradations: intense flares obscure scene structure, while photon-limited regions collapse into noise. Conventional approaches address these factors in isolation, overlooking the fact that these degradations are fundamentally entangled. To bridge this gap, we introduce LUCID, a unified framework that reframes nighttime restoration as a continuous and controllable process rather than a fixed correction. We decompose nighttime restoration into two cooperative components: a flare disentanglement module that lifts the 'curtain' of optical artifacts to provide reliable structural guidance, and a diffusion-driven module that leverages generative priors to reconstruct clean and well-exposed imagery. Crucially, LUCID introduces explicit controllability through a novel four-mode training strategy, enabling users to steer the restoration process via classifier-free guidance (CFG) and allowing selective control over light sources and their associated flare and ghosting artifacts, while also supporting high dynamic range (HDR) reconstruction through continuous exposure control. Extensive experiments demonstrate that LUCID consistently outperforms state-of-the-art methods across diverse real-world nighttime scenarios.
Chinese Translation
摄影是用光绘画的艺术，但夜间场景受到多种退化因素的影响：强烈的眩光遮蔽了场景结构，而光子限制区域则崩溃为噪声。传统方法通常孤立地处理这些因素，忽视了这些退化本质上是相互纠缠的。为了解决这一问题，我们提出了LUCID，一个统一框架，将夜间恢复重新定义为一个连续且可控的过程，而不是固定的修正。我们将夜间恢复分解为两个协作组件：一个眩光解缠模块，揭开光学伪影的“帷幕”，提供可靠的结构指导；一个基于扩散的模块，利用生成先验重建干净且曝光良好的图像。重要的是，LUCID通过一种新颖的四模式训练策略引入了明确的可控性，使用户能够通过无分类器引导（CFG）来引导恢复过程，并允许对光源及其相关的眩光和鬼影伪影进行选择性控制，同时支持通过连续曝光控制进行高动态范围（HDR）重建。大量实验表明，LUCID在各种真实世界夜间场景中始终优于最先进的方法。
View on arXiv Download PDF AI Translation
cs.CV / 35 / 2606.06903

Beyond Skeletons: Learning Animation Directly from Driving Videos with Same2X Training Strategy

超越骨架：通过 Same2X 训练策略直接从驱动视频学习动画
Zeng, Yuan, Shi, Yujia, Yang, Yuhao, Liu, Dongxia, Lu, Zongqing, Yang, Wenming, Liao, Qingmin
Abstract
Human image animation aims to generate a video from a static reference image, guided by pose information extracted from a driving video. Existing approaches often rely on pose estimators to extract intermediate representations, but such signals are prone to errors under occlusion or complex poses. Building on these observations, we present DirectAnimator, a framework that bypasses pose extraction and directly learns from raw driving videos. We introduce a Driving Cue Triplet consisting of pose, face, and location cues that captures motion, expression, and alignment in a semantically rich yet stable form, and we fuse them through a CueFusion DiT block for reliable control during denoising. To make learning dependable when the driving and reference identities differ, we devise a Same2X training strategy that aligns cross-ID features with those learned from same-ID data, regularizing optimization and accelerating convergence. Extensive experiments demonstrate that DirectAnimator attains state-of-the-art visual quality and identity preservation while remaining robust to occlusions and complex articulation, and it does so with fewer computational resources. Our project page is at https://directanimator.github.io/.
Chinese Translation
人像动画旨在从静态参考图像生成视频，依据从驱动视频中提取的姿态信息进行指导。现有方法通常依赖姿态估计器提取中间表示，但在遮挡或复杂姿态下，这些信号容易出现错误。基于这些观察，我们提出了 DirectAnimator，一个绕过姿态提取并直接从原始驱动视频学习的框架。我们引入了一个由姿态、面部和位置线索组成的 Driving Cue Triplet，它以语义丰富且稳定的形式捕捉运动、表情和对齐信息，并通过 CueFusion DiT 模块进行融合，以在去噪过程中实现可靠控制。为了在驱动身份和参考身份不同的情况下使学习变得可靠，我们设计了一种 Same2X 训练策略，该策略将跨身份特征与从同身份数据中学习到的特征对齐，从而规范化优化并加速收敛。大量实验表明，DirectAnimator 在视觉质量和身份保留方面达到了最先进的水平，同时在面对遮挡和复杂关节时保持鲁棒性，并且所需的计算资源更少。我们的项目页面位于 https://directanimator.github.io/。
View on arXiv Download PDF AI Translation
cs.CV / 36 / 2606.06908

polyDAG: Polynomial Acyclicity Constraints for Efficient Continuous Causal Discovery in Visual Semantic Graphs

polyDAG：用于视觉语义图高效连续因果发现的多项式无环约束
Zhang, Wenhao, Ramezani, Ramin, Han, Tao, Hwang, Kai, Guo, Minyi
Abstract
Modern image-analysis pipelines often convert images into structured semantic variables, such as facial attributes, object concepts, and scene descriptors. Learning directed dependencies among these variables can produce interpretable visual semantic graphs, but continuous directed acyclic graph learning is limited by the cost of enforcing acyclicity. We present polyDAG, a polynomial acyclicity framework for efficient continuous causal discovery in visual semantic graphs. polyDAG replaces the matrix-exponential acyclicity constraint with a finite polynomial trace constraint and proves that the new constraint is zero exactly for acyclic graphs. We further derive a geometric-series implementation that avoids the explicit summation loop while preserving the same acyclicity condition. Experiments on synthetic Erdos-Renyi graphs and CelebA facial visual attributes show that polyDAG improves efficiency and structure recovery. Averaged over the revised synthetic protocol with d in {100, 200, 500}, polyDAG reduces mean structural Hamming distance from 318.4 to 285.4 and improves mean F1 score from 0.725 to 0.756. At 100 nodes, the geometric variant runs in 3.44 seconds compared with 5.16 seconds for the exponential baseline, corresponding to a 33.4 percent speedup. Code and data are publicly available at https://github.com/wenhaoz-fengcai/polyDAG.
Chinese Translation
现代图像分析流程通常会将图像转换为结构化的语义变量，例如面部属性（facial attributes）、对象概念（object concepts）和场景描述符（scene descriptors）。学习这些变量之间的有向依赖关系能够构建具有良好可解释性的视觉语义图（Visual Semantic Graphs），然而，连续有向无环图（Directed Acyclic Graph, DAG）学习的效率往往受到无环性约束（acyclicity constraint）计算成本的限制。本文提出 polyDAG，一种面向视觉语义图高效连续因果发现的多项式无环性框架。polyDAG 使用有限阶多项式迹约束（polynomial trace constraint）替代传统的矩阵指数无环性约束（matrix-exponential acyclicity constraint），并证明该新约束当且仅当图为无环图时取零值。进一步地，本文推导出一种基于几何级数（geometric series）的实现方法，在保持相同无环性条件的同时避免了显式求和循环。基于合成 Erdős–Rényi 图以及 CelebA 人脸视觉属性数据集的实验结果表明，polyDAG 在计算效率和图结构恢复能力方面均优于现有方法。在修订后的合成实验协议下，当节点维度 d ∈ {100, 200, 500} 时，polyDAG 将平均结构汉明距离（Structural Hamming Distance, SHD）从 318.4 降低至 285.4，同时将平均 F1 分数从 0.725 提升至 0.756。在 100 个节点的实验设置下，几何级数变体的运行时间为 3.44 秒，而指数基线方法为 5.16 秒，对应获得 33.4% 的速度提升。相关代码与数据已公开发布于：https://github.com/wenhaoz-fengcai/polyDAG。
View on arXiv Download PDF AI Translation
cs.CV / 37 / 2606.06918

DRIFT: From Robustness Gaps to Invariance Manifolds for AI-Generated Image Detection

DRIFT：从鲁棒性差距到AI生成图像检测的不变流形
Ameta, Abhishek, Banerjee, Sayan, Pandith, Shreyas, Harshit, Chatterjee, Ankita, Bankar, Akshay Janardan, Unde, Amit Satish
Abstract
The rapid evolution of generative image models challenges existing AI-generated image detectors, particularly in open-world settings with unseen generators. Recent training-free approaches measure robustness gaps in frozen vision foundation models (VFMs), detecting fakes via perturbation-induced embedding drift. However, these methods rely on fixed invariance geometry inherited from pretraining and lack principled adaptation to the detection task. We instead formulate AI-generated image detection as learning a structured invariance manifold of real images under one-class supervision. Building upon a frozen VFM, we introduce lightweight projection heads that decompose representation space into complementary robust and fragile subspaces. The robust subspace is explicitly trained to suppress variations induced by physically plausible imaging transformations, approximating tangent directions of a real-image manifold, while the fragile subspace retains sensitivity to edit-like perturbations. A structured ordering margin enforces hierarchical separation between physical invariance and edit-induced variability, enabling detection as a margin-violation test relative to the learned manifold. At inference, multi-scale patch-wise drift under both transformation families yields a dual-channel invariance signature and interpretable localization. Extensive experiments demonstrate strong open-world generalization across unseen generators and resolutions, consistently outperforming training-free robustness-based baselines while providing interpretable invariance-violation maps.
Chinese Translation
生成图像模型的快速发展对现有的AI生成图像检测器提出了挑战，特别是在具有未见生成器的开放世界环境中。最近的无训练方法通过测量冻结的视觉基础模型（VFM）中的鲁棒性差距，利用扰动引起的嵌入漂移来检测伪造图像。然而，这些方法依赖于从预训练中继承的固定不变几何结构，缺乏对检测任务的原则性适应。相反，我们将AI生成图像检测形式化为在单类监督下学习真实图像的结构化不变流形。在冻结的VFM基础上，我们引入轻量级投影头，将表示空间分解为互补的鲁棒和脆弱子空间。鲁棒子空间经过明确训练，以抑制由物理上合理的成像变换引起的变化，近似真实图像流形的切向方向，而脆弱子空间则保持对编辑类扰动的敏感性。结构化排序边界强制物理不变性与编辑引起的变异性之间的层次分离，使得检测可以作为相对于学习到的流形的边界违反测试。在推理阶段，针对两种变换族的多尺度块级漂移产生双通道不变性特征和可解释的定位。大量实验表明，在未见生成器和分辨率下，强大的开放世界泛化能力，始终优于基于无训练鲁棒性的方法，同时提供可解释的不变性违反图谱。
View on arXiv Download PDF AI Translation
cs.CV / 38 / 2606.06926

SVHighlights: Towards Extremely Long Sport Video Highlight Detection

SVHighlights：面向极长体育视频精彩片段检测
Lee, Donggyu, Ki, Youngbin, Kang, Jeonghun, Kim, Taehwan
Abstract
While highlight detection for long-form videos is of great practical importance, most existing methods remain limited to short-form content, largely due to the absence of a suitable benchmark. To bridge this gap, we introduce SVHighlights, to the best of our knowledge, the first benchmark for highlight detection in extremely long sports videos, each exceeding one hour in duration, across multiple sports categories. SVHighlights is constructed from pairs of full-length sports videos and their corresponding official highlight videos using a dataset generation pipeline, enabling scalable label generation without conventional per-clip saliency annotation. The benchmark comprises 320 videos with an average duration of 2.00 hours and a total of 640.18 hours, substantially exceeding previous datasets. Existing methods also face fundamental challenges on long videos: models trained on short clips fail to generalize to hour-long content, and their clip-level scoring lacks the broader context needed to identify highlights. To address this and provide a strong baseline, we present TF-SELECTOR, a training-free segment-based approach that divides each video into context-aware segments by merging adjacent shots sharing the same semantic content, and predicts segment-level saliency scores using a large language model with multimodal inputs including visual captions, transcripts, and audio volume. Experiments demonstrate that TF-SELECTOR achieves superior performance across most metrics compared to Video Temporal Grounding (VTG)-tuned baselines, with improvements of +3.12 in HIT@1, +4.06 in HIT@K, and +2.95 in IoU. These results establish SVHighlights as a challenging testbed for long-form highlight detection and demonstrate that a simple segment-based strategy can effectively scale to hour-long videos.
Chinese Translation
尽管长视频的精彩片段检测具有重要的实际意义，但现有大多数方法仍然局限于短视频内容，这在很大程度上是由于缺乏合适的基准。为了解决这一问题，我们介绍了SVHighlights，尽我们所知，这是第一个针对极长体育视频（每个视频时长超过一小时）进行精彩片段检测的基准，涵盖多个体育类别。SVHighlights是通过一套数据集生成管道，从全长体育视频及其对应的官方精彩片段视频对中构建的，使得在没有传统逐片显著性标注的情况下实现可扩展的标签生成。该基准包含320个视频，平均时长为2.00小时，总时长达到640.18小时，显著超过以往的数据集。现有方法在长视频上也面临基本挑战：在短片段上训练的模型无法推广到长达一小时的内容，其片段级评分缺乏识别精彩片段所需的更广泛上下文。为了解决这一问题并提供强有力的基线，我们提出了TF-SELECTOR，这是一种无训练的基于片段的方法，通过合并共享相同语义内容的相邻镜头，将每个视频划分为上下文感知的片段，并使用大型语言模型预测片段级显著性评分，该模型的多模态输入包括视觉字幕、文本记录和音频音量。实验表明，TF-SELECTOR在大多数指标上相较于经过视频时间定位（VTG）调优的基线表现出更优越的性能，HIT@1提高了+3.12，HIT@K提高了+4.06，IoU提高了+2.95。这些结果确立了SVHighlights作为长视频精彩片段检测的挑战性测试平台，并证明了简单的基于片段的策略能够有效扩展到长达一小时的视频。
View on arXiv Download PDF AI Translation
cs.CV / 39 / 2606.06938

When CLIP Sees More, It Fights Back Harder: Multi-View Guided Adaptive Counterattacks for Test-Time Adversarial Robustness

当 CLIP 看到更多时，它反击得更强烈：用于测试时对抗鲁棒性的多视角引导自适应反击
Kim, Sunoh, Um, Daeho
Abstract
Vision-language models such as CLIP have achieved remarkable zero-shot recognition capabilities, yet their robustness against adversarial perturbations remains limited. Test-time counterattack (TTC) was recently proposed to improve CLIP's robustness by perturbing an input image to steer it away from a corrupted state during inference. However, TTC remains fragile under strong attacks because its counterattack relies on a directly corrupted original view and employs a noise-driven hard-gating scheme that cannot adapt to varying corruption severity. To address these limitations, we introduce Multi-view guided Adaptive Counterattack (MAC), which performs counterattacks for multi-view with corruption-aware soft weighting. Specifically, MAC first constructs augmented views of an input image to obtain diverse embeddings. It then performs counterattacks to refine corrupted embeddings of views. Next, MAC adaptively scales the counterattack intensity for each view based on its estimated corruption degree. Finally, the adaptively counterattacked views are aggregated to yield a robust final prediction. Extensive experiments across 20 datasets and diverse attack scenarios demonstrate that MAC substantially improves robustness while preserving high inference speed and memory efficiency with its tuning-free design. Our code is available at https://github.com/sunoh-kim/MAC.
Chinese Translation
视觉-语言模型如 CLIP 已实现显著的零样本识别能力，但其对抗扰动的鲁棒性仍然有限。最近提出的测试时反击（Test-time Counterattack, TTC）旨在通过扰动输入图像，在推理过程中将其引导远离受损状态，从而提高 CLIP 的鲁棒性。然而，TTC 在强攻击下仍然脆弱，因为其反击依赖于直接受损的原始视图，并采用一种无法适应不同损坏严重程度的噪声驱动硬门控方案。为了解决这些局限性，我们提出了多视角引导自适应反击（Multi-view guided Adaptive Counterattack, MAC），该方法通过对多视角进行损坏感知的软加权来执行反击。具体而言，MAC 首先构建输入图像的增强视图，以获得多样化的嵌入。然后，它对视图的受损嵌入进行反击以进行精炼。接下来，MAC 根据每个视图的估计损坏程度自适应地调整反击强度。最后，经过自适应反击的视图被聚合，以产生鲁棒的最终预测。在 20 个数据集和多种攻击场景下的广泛实验表明，MAC 在保持高推理速度和内存效率的同时，显著提高了鲁棒性，且其设计无需调优。我们的代码可在 https://github.com/sunoh-kim/MAC 获取。
View on arXiv Download PDF AI Translation
cs.CV / 40 / 2606.06943

SS-TPT: Stability and Suitability-Guided Test-Time Prompt Tuning for Adversarially Robust Vision-Language Models

SS-TPT：基于稳定性和适用性指导的测试时提示调优用于对抗鲁棒的视觉-语言模型
Kim, Sunoh, Um, Daeho
Abstract
Vision-language models (VLMs) such as CLIP achieve strong zero-shot recognition but remain highly fragile under adversarial perturbations. Recent test-time adaptation defenses improve robustness by leveraging many augmented views, but this leads to impractical slowdown and a clear robustness-throughput trade-off. To address this challenge, we present Stability and Suitability-guided Test-time Prompt Tuning (SS-TPT), evaluating the quality of each augmented view via two complementary scores: (1) stability, measuring prediction invariance to weak augmentations, and (2) suitability, measuring feature-space density among views. These stability and suitability (SS) scores guide both adaptation and inference through an SS-guided consistency loss and an SS-weighted prediction, amplifying trustworthy views while suppressing corrupted ones. Extensive experiments demonstrate that SS-TPT significantly outperforms prior state-of-the-art methods, achieving superior robustness-throughput trade-offs across diverse datasets and varying numbers of views, thereby demonstrating both strong practicality and generality. Our code is available at https://github.com/sunoh-kim/SS-TPT.
Chinese Translation
视觉-语言模型（VLMs）如 CLIP 在零样本识别中表现出色，但在对抗扰动下仍然非常脆弱。近期的测试时适应防御通过利用多种增强视图来提高鲁棒性，但这导致了不切实际的速度下降和明显的鲁棒性-吞吐量权衡。为了解决这一挑战，我们提出了基于稳定性和适用性指导的测试时提示调优（SS-TPT），通过两个互补评分评估每个增强视图的质量：（1）稳定性，测量对弱增强的预测不变性；（2）适用性，测量视图之间的特征空间密度。这些稳定性和适用性（SS）评分通过 SS-guided 一致性损失和 SS 加权预测指导适应和推理，放大可信视图，同时抑制受损视图。大量实验表明，SS-TPT 显著优于之前的最先进方法，在不同数据集和不同数量的视图中实现了更优的鲁棒性-吞吐量权衡，从而展示了强大的实用性和普适性。我们的代码可在 https://github.com/sunoh-kim/SS-TPT 获取。
View on arXiv Download PDF AI Translation
cs.CV / 41 / 2606.06950

When is 3D Worth It? A Resource-Performance Frontier for CNNs and Transformers in Lung CT

3D模型何时值得使用？CNN和Transformer在肺部CT中的资源-性能边界
Hoq, Md Enamul, Hossain, Sharafat, Emmaka, Imraul, Larson-Prior, Linda, Tarbox, Lawrence, Bona, Jonathan, Prior, Donald Johann Jr. and Fred
Abstract
Three-dimensional models are widely assumed preferable for volumetric medical imaging, yet their practical value depends on whether performance gains justify added computational cost and complexity. Rather than proposing a new architecture, we study how input dimensionality (2D, 2.5D, 3D) affects model behavior across convolutional neural networks (CNNs) and Vision Transformers (ViTs) under a fixed training protocol. Using a leakage-free NLST cohort (n = 1,977) with supporting LIDC-IDRI data, we find that the 2.5D CNN offers the most favorable discrimination-stability trade-off in our comparison (ROC-AUC 0.682, 95% CI [0.546, 0.799]) with a stable operating point. In contrast, 3D CNNs show threshold instability, and transformers exhibit degenerate predictions, such as all-positive predictions. Confidence intervals are wide and overlapping, so we present these results as a controlled resource-performance frontier and a failure-mode taxonomy rather than as definitive superiority claims. For class-imbalanced lung cancer screening classification, 2D and 2.5D inputs provide a more reliable trade-off between performance, stability, and computational efficiency than full 3D representations.
Chinese Translation
三维模型在体积医学成像中被广泛认为是更优的选择，然而它们的实际价值取决于性能提升是否足以证明额外的计算成本和复杂性是合理的。我们并未提出新的架构，而是研究输入维度（2D、2.5D、3D）如何在固定训练协议下影响卷积神经网络（CNN）和视觉变换器（ViTs）的模型行为。利用无泄漏的NLST队列（n = 1,977）以及支持的LIDC-IDRI数据，我们发现2.5D CNN在我们的比较中提供了最有利的区分-稳定性权衡（ROC-AUC 0.682，95% CI [0.546, 0.799]），并且具有稳定的操作点。相比之下，3D CNN显示出阈值不稳定性，而变换器则表现出退化的预测，例如全正预测。置信区间较宽且重叠，因此我们将这些结果呈现为受控的资源-性能边界和失败模式分类，而不是作为绝对优越性的声明。在类别不平衡的肺癌筛查分类中，2D和2.5D输入在性能、稳定性和计算效率之间提供了比完整3D表示更可靠的权衡。
View on arXiv Download PDF AI Translation
cs.CV / 42 / 2606.06958

MVSegNet: A Lightweight Boundary-Aware Network for Fetal Lateral Ventricle Segmentation and Atrial Width Estimation in Prenatal Ultrasound

MVSegNet：一种轻量级边界感知网络，用于产前超声中胎儿侧脑室分割和心房宽度估计
Sayem, Arafat Hossain
Abstract
Fetal ventriculomegaly is assessed by measuring the atrial width of the lateral ventricle in prenatal ultrasound. Accurate segmentation is essential for this measurement, but acoustic shadowing, speckle noise, and poor contrast make it difficult. We developed MVSegNet, a lightweight encoder-decoder network combining multi-scale feature extraction and boundary-aware refinement. The model was trained and evaluated on 584 expert-annotated transventricular ultrasound frames using a 70/15/15 split. Performance was compared against six segmentation baselines using overlap, boundary, and measurement metrics. MVSegNet achieved a Dice score of 80.79%, IoU of 68.47%, Hausdorff distance of 4.07 mm, and atrial width mean absolute error of 3.40 mm. The model contains 2.31 million parameters and runs at 165.6 frames per second on an NVIDIA T4 GPU. MVSegNet outperformed all evaluated baselines on boundary and measurement metrics while maintaining low computational cost, supporting its use in automated fetal ultrasound analysis.
Chinese Translation
胎儿脑室扩大症通过在产前超声中测量侧脑室的心房宽度来评估。准确的分割对于这一测量至关重要，但声学阴影、斑点噪声和对比度差使其变得困难。我们开发了MVSegNet，这是一种轻量级的编码器-解码器网络，结合了多尺度特征提取和边界感知细化。该模型在584个专家标注的跨脑室超声帧上进行了训练和评估，采用70/15/15的划分方式。通过重叠、边界和测量指标将性能与六个分割基线进行比较。MVSegNet达到了80.79%的Dice系数、68.47%的IoU、4.07毫米的Hausdorff距离和3.40毫米的心房宽度平均绝对误差。该模型包含231万个参数，并在NVIDIA T4 GPU上以每秒165.6帧的速度运行。MVSegNet在边界和测量指标上超越了所有评估的基线，同时保持了低计算成本，支持其在自动化胎儿超声分析中的应用。
View on arXiv Download PDF AI Translation
cs.CV / 43 / 2606.06966

From Vision to Text: A Compact Multimodal Approach for Robust, Cross-Domain Presentation Attack Detection on ID Cards

从视觉到文本：一种紧凑的多模态方法用于在身份证上进行稳健的跨域呈现攻击检测
Zeng, Qingwen, Tapia, Juan E., Das, Sneha, Busch, Christoph
Abstract
Cross-domain shifts challenge Presentation Attack Detection (PAD) on ID Cards, given the restricted data available due to privacy concerns. This work proposes a compact multimodal model, based on new generative and discriminative blocks, which combines visual and textual data for PAD on genuine and synthetic ID images. While multimodal models exhibit strong generalisation after supervised fine-tuning, they fail in zero-shot settings. Our findings underscore that model capacity and real-world data are essential for reliable PAD, while existing synthetic datasets may not reflect real-world challenges. We argue for a re-evaluation of synthetic data as a benchmark and emphasise the need for more realistic, diverse datasets to advance PAD research.
Chinese Translation
跨域转变对身份证上的呈现攻击检测（PAD）构成挑战，因为由于隐私问题可用的数据受到限制。本研究提出了一种基于新生成和判别模块的紧凑多模态模型，该模型结合了视觉和文本数据，用于对真实和合成身份证图像进行PAD。尽管多模态模型在经过监督微调后展现出强大的泛化能力，但在零样本设置中却表现不佳。我们的研究结果强调了模型容量和真实世界数据对于可靠PAD的重要性，而现有的合成数据集可能无法反映现实世界的挑战。我们主张重新评估合成数据作为基准，并强调需要更真实、多样化的数据集以推动PAD研究的进展。
View on arXiv Download PDF AI Translation
cs.CV / 44 / 2606.06978

CL-CLIP: CLIP-Based Continual Learning Framework with Cost-Volume Category Decoupling for Object Detection

CL-CLIP：基于CLIP的持续学习框架，通过成本-体积类别解耦实现目标检测
Liu, Zihan, Yang, Yuguang, Su, Shengjie, Pang, Jianing, Yang, Linlin, Xie, Chunyu, Zolotykh, Nikolai Yu., Zhang, Baochang
Abstract
Continual Object Detection (COD) requires a detector to acquire new categories over time while preserving previously learned ones. This goal is closely related to open-vocabulary detection, since both settings require reasoning over categories that are not fully covered by the annotations available at the current training stage. Recent CLIP-based open-vocabulary detectors have shown strong zero-shot generalization, and frameworks such as F-ViT demonstrate that vision-language pretraining can provide powerful zero-shot detection ability for unseen categories. However, real-world deployments cannot remain purely zero-shot: once these detectors are continually updated on newly introduced categories, they suffer severe catastrophic forgetting and quickly lose their previously calibrated detection ability. We therefore propose CL-CLIP, a CLIP-based COD framework that equips open-vocabulary detectors with better continual learning ability through cost-volume-guided category decoupling. Specifically, following CAT-Seg, we compute a CLIP image-text similarity cost volume, defined as dense category-wise response maps between visual tokens and class text embeddings. This zero-shot spatial prior decomposes shared region features into class-specific pathways, which are then processed by a Multi-Expert RoI head. Extensive experiments on PASCAL VOC and MS-COCO show that CL-CLIP substantially improves the F-ViT baseline under continual fine-tuning and achieves competitive performance with existing continual object detectors, especially in adapting to newly introduced categories while preserving competitive base-class performance.
Chinese Translation
持续目标检测（COD）要求检测器在保留先前学习的类别的同时，随着时间的推移获取新类别。这个目标与开放词汇检测密切相关，因为这两种设置都需要对当前训练阶段可用的注释未完全覆盖的类别进行推理。最近基于CLIP的开放词汇检测器展示了强大的零样本泛化能力，而像F-ViT这样的框架则表明视觉-语言预训练可以为未见类别提供强大的零样本检测能力。然而，实际应用不能完全依赖零样本：一旦这些检测器在新引入的类别上被持续更新，它们就会遭遇严重的灾难性遗忘，并迅速失去先前校准的检测能力。因此，我们提出了CL-CLIP，一个基于CLIP的COD框架，通过成本-体积引导的类别解耦，赋予开放词汇检测器更好的持续学习能力。具体而言，遵循CAT-Seg，我们计算了一个CLIP图像-文本相似性成本体积，定义为视觉标记与类别文本嵌入之间的密集类别响应图。这个零样本空间先验将共享区域特征分解为类别特定的路径，然后由多专家RoI头进行处理。在PASCAL VOC和MS-COCO上的大量实验表明，CL-CLIP在持续微调下显著提高了F-ViT基线，并在与现有持续目标检测器的竞争性能中表现出色，尤其是在适应新引入的类别的同时保持竞争的基础类别性能。
View on arXiv Download PDF AI Translation
cs.CV / 45 / 2606.06991

Don't Pause: Streaming Video-Language Synchrony for Online Video Understanding

不停顿：在线视频理解的流媒体视频-语言同步
Yang, Zhenyu, Zhang, Kairui, Qian, Shengsheng, Dong, Weiming, Xu, Changsheng
Abstract
Online Video Large Language Models (Video-LLMs) have advanced toward seamless human-AI interaction through frame-by-frame processing and proactive responding. However, a critical challenge remains in streaming scenarios: existing models typically pause video perception while generating responses, breaking real-time video-language synchrony and causing stutters. To address this, we introduce a novel paradigm for online video understanding: Streaming Video-Language Synchrony (SVLS), and present LyraV, a live streaming assistant built upon a hierarchical control framework with two core innovations. First, the Frame-Driven Transition Controller (FDTC), a training-free verification-based finite-state machine, makes high-level semantic decisions on when to continue speaking, start a new response, or stay silent. Second, the Streaming Token Pacer (SToP), a plug-and-play lightweight predictive module, dynamically adapts the language generation rate to match the pace of the visual content. Concretely, LyraV performs \emph{per-frame incremental, sub-budget decoding}: within each frame interval it emits only a small chunk of tokens that fits the real-time budget, so perception is never blocked for a full sentence. Together, these components enable LyraV to seamlessly interleave incoming video frames with generated word tokens, achieving a fine-grained synchrony. Extensive experiments conducted on five online and three offline benchmarks demonstrate that LyraV preserves the backbone's general understanding ability while substantially improving streaming synchrony and narrative fluency, delivering a 98.29\% synchrony with video playback and a real-time processing speed of 3.89 FPS. Interestingly, we observe an empirical capability in LyraV: dynamic reasoning over streaming tokens, enabling continuous interpretation and "thinking" alongside visual input.
Chinese Translation
在线视频大型语言模型（Video-LLMs）通过逐帧处理和主动响应，已向无缝的人机交互迈进。然而，在流媒体场景中仍然存在一个关键挑战：现有模型在生成响应时通常会暂停视频感知，这打破了实时视频-语言同步，导致卡顿。为了解决这个问题，我们提出了一种新的在线视频理解范式：流媒体视频-语言同步（Streaming Video-Language Synchrony, SVLS），并展示了基于分层控制框架构建的实时流媒体助手LyraV，该框架具有两个核心创新。首先，帧驱动过渡控制器（Frame-Driven Transition Controller, FDTC）是一种无需训练的基于验证的有限状态机，它在何时继续说话、开始新响应或保持沉默方面做出高层次的语义决策。其次，流媒体令牌节奏器（Streaming Token Pacer, SToP）是一个即插即用的轻量级预测模块，它动态调整语言生成速率，以匹配视觉内容的节奏。具体而言，LyraV执行 extit{逐帧增量、子预算解码}：在每个帧间隔内，它仅发出适合实时预算的小块令牌，从而确保感知不会因完整句子而被阻塞。这些组件共同使LyraV能够无缝地将传入的视频帧与生成的词令牌交错，实现精细的同步。在五个在线和三个离线基准上进行的大量实验表明，LyraV在保持主干模型一般理解能力的同时，显著提高了流媒体同步性和叙述流畅性，实现了与视频播放的98.29 ext{%}同步率和3.89 FPS的实时处理速度。有趣的是，我们观察到LyraV具有一种经验能力：对流媒体令牌进行动态推理，使其能够在视觉输入的同时进行持续的解释和“思考”。
View on arXiv Download PDF AI Translation
cs.CV / 46 / 2606.07024

GuideCAD: A Lightweight Multimodal Framework for 3D CAD Model Generation via Prefix Embedding

GuideCAD：一种轻量级多模态框架，通过前缀嵌入生成3D CAD模型
Kim, Minseong, Park, Jinyeong, Park, Sungho, Kim, Jibum
Abstract
Multi-modal approaches used for 3D CAD generation require substantial computational resources, necessitating efficient training. To address this, we propose GuideCAD, which leverages semantically rich visual-textual representations having only a small number of trainable parameters to generate 3D CAD models. Specifically, GuideCAD uses a mapping network that converts image embeddings into prefix embeddings, enabling a pretrained large language model to integrate visual and textual information. As a result, a transformer-based decoder predicts the construction sequence using the visual-textual embeddings in order to generate the 3D CAD model. For experimental evaluation, we construct a new dataset, referred to as GuideCAD, which consists of text-image pairs. Each pair includes a text prompt that represents a 3D CAD construction sequence and its corresponding 3D CAD image. Our experimental results show that GuideCAD generates comparably high-quality 3D CAD models while using approximately four times fewer parameters and achieving twice the training efficiency compared to fine-tuning approaches. We have released the source code and dataset for our method at: https://github.com/mskimS2/GuideCAD
Chinese Translation
用于3D CAD生成的多模态方法需要大量计算资源，因此需要高效的训练。为了解决这个问题，我们提出了GuideCAD，它利用语义丰富的视觉-文本表示，仅使用少量可训练参数来生成3D CAD模型。具体而言，GuideCAD使用一个映射网络，将图像嵌入转换为前缀嵌入，从而使预训练的大型语言模型能够整合视觉和文本信息。因此，基于变换器的解码器使用视觉-文本嵌入预测构建顺序，以生成3D CAD模型。为了进行实验评估，我们构建了一个新的数据集，称为GuideCAD，该数据集由文本-图像对组成。每对包括一个表示3D CAD构建顺序的文本提示及其对应的3D CAD图像。我们的实验结果表明，GuideCAD生成的3D CAD模型质量相当高，同时使用的参数数量约为微调方法的四分之一，并且训练效率是其两倍。我们已在以下网址发布了我们方法的源代码和数据集：https://github.com/mskimS2/GuideCAD
View on arXiv Download PDF AI Translation
cs.CV / 47 / 2606.07032

Never Seen Before: Benchmarking Genuine Zero-Shot Composed Image Retrieval with Consistent Video-Sourced Datasets

前所未见：基于一致性视频源数据集的真实零-shot组合图像检索基准评估
Yang, Zhenyu, Du, Zemin, Qian, Shengsheng, Xu, Changsheng
Abstract
Zero-Shot Composed Image Retrieval (ZS-CIR) aims to retrieve a target image based on a query composed of a reference image and a relative caption without training samples. Existing ZS-CIR datasets often suffer from complete irrelevance between reference and target images due to noisy image sources, and do not achieve a true zero-shot scenario as they use public image datasets that models like CLIP have been trained on. To tackle these challenges, we introduce ZeroSight, a novel benchmark for ZS-CIR. It includes a dataset with consistent reference-target pairs sourced from videos, a data construction pipeline, and evaluation methods that consider the ranking of multiple positive and negative target images. We ensure visually and semantically consistent reference-target pairs by extracting frames from a single video and generating relative captions using LLM-assisted methods. To ensure a true zero-shot scenario, we use video data published after March 31, 2022, ensuring it was not included in CLIP's pre-training data. Additionally, we propose a training-free MLLM-driven method, SC4CIR (Symmetric Consistency for CIR), which can effectively identify hard negative targets through 3 symmetric consistency checks. This method is plug-and-play, seamlessly integrating with various CIR methods and significantly improving performance. Our experimental results from 27 methods reveal that current ZS-CIR datasets and evaluation metrics result in inflated retrieval performance, exaggerating the capabilities of CIR methods. Our benchmark and models can be accessed at https://github.com/sotayang/ZeroSight.
Chinese Translation
零-shot组合图像检索（ZS-CIR）旨在基于由参考图像和相关标题组成的查询检索目标图像，而无需训练样本。现有的ZS-CIR数据集往往由于噪声图像源而导致参考图像与目标图像之间完全不相关，并且由于使用了像CLIP这样的模型已训练过的公共图像数据集，未能实现真正的零-shot场景。为了解决这些挑战，我们引入了ZeroSight，一个新的ZS-CIR基准。它包括一个由视频源生成的一致性参考-目标对的数据集、一个数据构建管道，以及考虑多个正负目标图像排名的评估方法。我们通过从单个视频中提取帧并使用LLM辅助方法生成相关标题，确保视觉和语义上一致的参考-目标对。为了确保真正的零-shot场景，我们使用2022年3月31日之后发布的视频数据，确保其未包含在CLIP的预训练数据中。此外，我们提出了一种无训练的MLLM驱动方法SC4CIR（对称一致性用于CIR），该方法可以通过3个对称一致性检查有效识别困难的负目标。该方法即插即用，可以与各种CIR方法无缝集成，并显著提高性能。我们对27种方法的实验结果表明，当前的ZS-CIR数据集和评估指标导致检索性能膨胀，夸大了CIR方法的能力。我们的基准和模型可以在https://github.com/sotayang/ZeroSight访问。
View on arXiv Download PDF AI Translation
cs.CV / 48 / 2606.07034

ForensicConcept: Transferable Forensic Concepts for AIGI Detection

法医概念：可转移的法医概念用于AI生成图像检测
Zhou, Menyanshu, Zhou, Ziyin, Sun, Ke, Luo, Yunpeng, Ji, Jiayi, Sun, Xiaoshuai, Ji, Rongrong
Abstract
AI-generated image detectors achieve high accuracy on in-distribution data but often fail on unseen generators. A key obstacle to understanding this failure is the black-box nature of current detectors: they do not reveal which evidence drives their decisions. We propose ForensicConcept, a framework that extracts explicit forensic concepts from detectors and enables their transfer across backbones. Our method localizes decision-critical patches via Transformer attribution, clusters them into a compact concept codebook, and uses a concept-aligned projection to produce auditable evidence readouts. Motivated by prior studies showing that DINO representations can guide diffusion generation and exhibit concept-level correspondence with diffusion features, we introduce a generation-trace reference based on CleanDIFT diffusion features and quantify backbone-trace alignment via neighborhood-structure consistency (CKNNA). We further propose concept codebook injection to transfer diffusion-derived concepts into target backbones. Experiments on GenImage, GAN-family, and Chameleon benchmarks show consistent improvements over prior methods. We also find that CKNNA alignment predicts transfer effectiveness, providing a principled explanation for why some backbones yield more transferable forensic evidence than others.
Chinese Translation
AI生成的图像检测器在分布内数据上实现了高准确率，但在未见过的生成器上往往失败。理解这一失败的关键障碍在于当前检测器的黑箱特性：它们并未揭示哪些证据驱动了它们的决策。我们提出了法医概念（ForensicConcept），一个从检测器中提取明确法医概念并实现其在不同骨干网络间转移的框架。我们的方法通过Transformer归因定位决策关键补丁，将其聚类成紧凑的概念代码本，并利用概念对齐投影生成可审计的证据读数。受到先前研究的启发，表明DINO表示可以指导扩散生成，并与扩散特征展现概念级别的对应关系，我们引入了基于CleanDIFT扩散特征的生成追踪参考，并通过邻域结构一致性（CKNNA）量化骨干追踪对齐。我们进一步提出概念代码本注入，以将扩散派生的概念转移到目标骨干网络中。在GenImage、GAN家族和Chameleon基准上的实验显示出对先前方法的一致性改进。我们还发现CKNNA对齐能够预测转移效果，为为何某些骨干网络能产生更可转移的法医证据提供了原则性解释。
View on arXiv Download PDF AI Translation
cs.CV / 49 / 2606.07036

STREAM: Stochastic Riemannian Flow Matching with Anisotropic Decoder for Digital Histopathology Image Generation

STREAM：具有各向异性解码器的随机黎曼流匹配用于数字组织病理图像生成
Cho, Won June, Jeong, Daeky, Lim, Hyeongyeol, Yoon, Hongjun
Abstract
Synthetic histopathology image generation addresses critical challenges in computational pathology, including patient privacy and the growing need for large-scale training data for foundation models. Latent diffusion models have dominated the image generation domain, with recent works emphasizing that the choice of latent space is critical to the quality of generated images. Existing state-of-the-art generative models in histopathology use pretrained Vision Foundation Models (VFMs) as conditioning signals, and we observe that this leads to "conditioning collapse," where the conditioning signal dominates the latent space and lowers the quality and diversity of generated samples. Therefore, we instead use pretrained histopathology VFMs as the latent space itself, leveraging their patch-token features that encode rich semantic information. We empirically show that these features are $\ell_2$-normalized and lie on the unit hypersphere $\mathcal{S}^{d-1}$ with strong angular dominance and intrinsic curvature, making them naturally suited for a Riemannian formulation. We therefore present STREAM, the first framework to apply Riemannian flow matching in the pathology domain. STREAM consists of two stages: 1) a bridge-type stochastic perturbation that establishes per-token rectifiability on $\mathcal{S}^{d-1}$ for training a Diffusion Transformer (DiT) in latent space, and 2) a novel anisotropic decoder that allocates robustness to low-energy directions of the velocity-field Jacobian while preserving fidelity along its high-energy directions. Together, STREAM achieves state-of-the-art reconstruction and generation performance on breast and colorectal cancer datasets. The code will be publicly released upon acceptance.
Chinese Translation
合成组织病理图像生成解决了计算病理学中的关键挑战，包括患者隐私和对大规模训练数据的日益需求，以支持基础模型的训练。潜在扩散模型在图像生成领域占据主导地位，近期研究强调潜在空间的选择对生成图像的质量至关重要。现有的组织病理学最先进的生成模型使用预训练的视觉基础模型（Vision Foundation Models, VFMs）作为条件信号，我们观察到这导致了“条件崩溃”（conditioning collapse），其中条件信号主导了潜在空间，降低了生成样本的质量和多样性。因此，我们改为使用预训练的组织病理学VFMs作为潜在空间本身，利用其编码丰富语义信息的补丁-标记特征。我们通过实验证明，这些特征是$ ext{l}_2$-归一化的，并且位于单位超球面$ ext{S}^{d-1}$上，具有强角度主导性和内在曲率，使其自然适合于黎曼公式化。因此，我们提出了STREAM，这是第一个在病理学领域应用黎曼流匹配的框架。STREAM由两个阶段组成：1）一种桥接型随机扰动，在$ ext{S}^{d-1}$上建立每个标记的可整形性，以训练潜在空间中的扩散变换器（Diffusion Transformer, DiT）；2）一种新颖的各向异性解码器，在保持其高能方向的保真度的同时，增强对速度场雅可比低能方向的鲁棒性。综合来看，STREAM在乳腺癌和结直肠癌数据集上实现了最先进的重建和生成性能。代码将在接受后公开发布。
View on arXiv Download PDF AI Translation
cs.CV / 50 / 2606.07053

TrioPose: Native Triple-Stream Diffusion Transformers for Pose-Guided Text-to-Image Generation

TrioPose：基于姿态引导的三重流扩散变换器用于文本到图像生成
Gu, Dian, Yang, Zhengyi
Abstract
Pose-guided text-to-image generation often suffers from limb distortions and feature crosstalk in complex multi-person scenarios. While existing UNet-based adapters struggle with long-range spatial dependencies, emerging Multimodal Diffusion Transformers (MM-DiTs) offer superior global modeling. However, naive signal concatenation in MM-DiTs severely disrupts pre-trained latent distributions. To address this, we propose TrioPose, a native pose-driven framework built upon the SD3.5M architecture. Specifically, we introduce a Triple-Stream Pose-Aware DiT (TSPA-DiT) that treats pose as an independent modality. It employs layer-wise activation and zero-initialized dual-residual injection to smoothly enforce geometric constraints while preserving pre-trained latent stability. To resolve severe multi-instance occlusions, we design a Learnable Relational Bias Mask that categorizes topological connectivity into fine-grained physical states, mapping them into continuous attention soft constraints to effectively decouple inter-instance interference. Furthermore, a Pose-Guided Spatial Loss Weighting strategy modulates the native diffusion objective using heatmap-derived error maps, focusing anatomical supervision strictly on distortion-prone regions. Extensive experiments demonstrate that TrioPose achieves state-of-the-art performance across challenging benchmarks, including Human-Art, CrowdPose, and OCHuman. Notably, it attains an AP of $64.33$ on Human-Art, representing a $30\%$ improvement over prior arts, while setting new standards for visual fidelity and text-image semantic alignment in complex multi-human generation.
Chinese Translation
姿态引导的文本到图像生成在复杂的多人物场景中常常面临肢体扭曲和特征串扰的问题。现有的基于UNet的适配器在处理长距离空间依赖时表现不佳，而新兴的多模态扩散变换器（Multimodal Diffusion Transformers, MM-DiTs）则提供了更优的全局建模能力。然而，MM-DiTs中的简单信号连接严重干扰了预训练潜在分布。为了解决这一问题，我们提出了TrioPose，这是一种基于SD3.5M架构的原生姿态驱动框架。具体而言，我们引入了一种三重流姿态感知扩散变换器（Triple-Stream Pose-Aware DiT, TSPA-DiT），将姿态视为独立的模态。它采用逐层激活和零初始化的双残差注入，平滑地施加几何约束，同时保持预训练潜在的稳定性。为了解决严重的多实例遮挡问题，我们设计了一种可学习的关系偏置掩码，将拓扑连接分类为细粒度的物理状态，并将其映射为连续注意力软约束，以有效解耦实例间的干扰。此外，姿态引导的空间损失加权策略利用热图导出的误差图调节原生扩散目标，严格关注易受扭曲影响的解剖区域。大量实验表明，TrioPose在包括Human-Art、CrowdPose和OCHuman在内的挑战性基准测试中实现了最先进的性能。值得注意的是，它在Human-Art上达到了$64.33$的平均精度（AP），相比之前的研究提高了$30\%$，同时在复杂的多人物生成中设定了视觉保真度和文本-图像语义对齐的新标准。
View on arXiv Download PDF AI Translation
cs.CV / 51 / 2606.07079

AsyncPatch Diffusion: spatially-flexible image generation

异步补丁扩散：空间灵活的图像生成
Papa, Samuele, De Bortoli, Valentin, Couairon, Guillaume, Sýkora, Daniel, Elie, Romuald, Greff, Klaus
Abstract
Standard diffusion models corrupt an entire sample with a single shared noise level, forcing all spatial regions to follow the same denoising trajectory. We introduce AsyncPatch Diffusion, a joint-diffusion framework that assigns distinct noise levels to different input dimensions, such as image pixels, or latent tokens. We show how this asynchronous corruption defines a valid generative process while supporting a richer family of spatially heterogeneous denoising trajectories, and prove the first valid ELBO for this process. We show that a single pretrained model can perform spatially adaptive generation, where different regions are denoised on different schedules. A key challenge is training: naive independent noise-level sampling overemphasizes highly heterogeneous configurations and underrepresents homogeneous noise levels, that are crucial during sampling. We address this with a controlled noise-level sampler that regulates both the average corruption level and its spatial variability. AsyncPatch achieves generation quality comparable to conventional diffusion on ImageNet 256 and LSUN, while being natively suited for inpainting without task-specific fine-tuning. We further introduce input guidance, which uses clean or partially corrupted regions to guide the generation of unknown regions, improving local consistency and texture matching. Finally, we demonstrate adaptive generation strategies including uncertainty-guided acceleration and autoregressive sampling.
Chinese Translation
标准扩散模型以单一共享噪声水平破坏整个样本，迫使所有空间区域遵循相同的去噪轨迹。我们提出了异步补丁扩散（AsyncPatch Diffusion），这是一种联合扩散框架，为不同的输入维度（如图像像素或潜在标记）分配不同的噪声水平。我们展示了这种异步破坏如何定义一个有效的生成过程，同时支持更丰富的空间异质去噪轨迹，并证明了该过程的第一个有效的证据下界（ELBO）。我们表明，单个预训练模型可以执行空间自适应生成，其中不同区域在不同的时间表上进行去噪。一个关键挑战是训练：简单的独立噪声水平采样过度强调高度异质的配置，而对在采样过程中至关重要的均匀噪声水平的代表性不足。我们通过一个受控噪声水平采样器来解决这个问题，该采样器调节平均破坏水平及其空间变异性。异步补丁在ImageNet 256和LSUN上的生成质量可与传统扩散相媲美，同时天生适合于无任务特定微调的图像修复。我们进一步引入输入引导，利用干净或部分损坏的区域来指导未知区域的生成，从而提高局部一致性和纹理匹配。最后，我们展示了包括不确定性引导加速和自回归采样在内的自适应生成策略。
View on arXiv Download PDF AI Translation
cs.CV / 52 / 2606.07086

An Adaptive Data cleaning Framework for Noisy Label Detection

一种自适应数据清洗框架用于噪声标签检测
Fang, Chen-Hsuan, Chen, Wei-Hsinag, Yu, Pin-Hsuan, Wang, Jung-Hua, Pan, Tsung-Wei
Abstract
Deep neural networks (DNNs) excel in computer vision tasks given large annotated datasets. In real-world applications, however, labels are often corrupted by ambiguity, human error, or dynamic environments. Over-parameterized DNNs easily memorize these noisy labels during training, degrading model accuracy and generalization. Existing data-cleaning and sample-selection strategies often rely on manually specified thresholds, prior knowledge of the noise ratio, or a single metric (either learning dynamics or geometric structure), making them unstable in complex data regimes. This paper proposes a self-adaptive data-cleaning framework that integrates local, global, and learning dynamics cues for robust noisy-label detection. Samples are mapped into a unified low-dimensional feature space through a modular feature concatenation paradigm. We provide two instantiations: a 2D metric integrating class-adaptive KNN-based local disagreement with k-means-based global centroid distance, and a 3D multi-metric that additionally incorporates a z-normalized score. Unlike conventional 1D Gaussian Mixture Models applied to a single scalar metric, our framework performs multi-metric clustering on the feature space to adaptively partition samples into clean-dominant and noise-dominant components without requiring manual thresholds or noise priors. Experiments on CIFAR-10, MNIST, and ImageNet-100 with 5% to 40% symmetric label noise show high recall across settings, including near-perfect recall (>=98%) on ImageNet-100 at 40% noise. Subsequent training yields accuracy gains across evaluated settings, especially under severe corruption on ImageNet-100. These findings suggest that multi-metric integration provides a threshold-free, practical, and low-tuning strategy for noisy label detection.
Chinese Translation
深度神经网络（DNN）在计算机视觉任务中表现优异，前提是拥有大量标注数据集。然而，在实际应用中，标签常常受到模糊性、人为错误或动态环境的影响而被污染。过参数化的DNN在训练过程中容易记忆这些噪声标签，从而降低模型的准确性和泛化能力。现有的数据清洗和样本选择策略通常依赖于手动指定的阈值、噪声比的先验知识或单一指标（无论是学习动态还是几何结构），使其在复杂数据环境中不稳定。本文提出了一种自适应数据清洗框架，整合了局部、全局和学习动态线索，以实现稳健的噪声标签检测。样本通过模块化特征拼接范式映射到统一的低维特征空间。我们提供了两种实例化：一种是结合类自适应KNN的局部不一致性与基于k-means的全局中心距离的二维度量，另一种是额外结合z标准化分数的三维多度量。与传统的应用于单一标量度量的1D高斯混合模型不同，我们的框架在特征空间上执行多度量聚类，自适应地将样本划分为干净主导和噪声主导的成分，而无需手动阈值或噪声先验。在CIFAR-10、MNIST和ImageNet-100上进行的实验显示，在5%到40%的对称标签噪声下，各种设置下的召回率均很高，包括在40%噪声下ImageNet-100的近乎完美召回率（>=98%）。后续训练在评估的设置中带来了准确性提升，尤其是在ImageNet-100的严重污染情况下。这些发现表明，多度量整合提供了一种无阈值、实用且低调优的噪声标签检测策略。
View on arXiv Download PDF AI Translation
cs.CV / 53 / 2606.07090

Detecting Temporally Localized Manipulations in Authentic Video Streams

检测真实视频流中的时间局部化操控
Umur, Okan, Güşlü, Ali Emre, Delibasoglu, Ibrahim
Abstract
The rapid advancement of video editing and generative artificial intelligence technologies has made realistic video manipulation increasingly accessible. Although existing datasets have significantly advanced research in deepfake detection, object removal, and video inpainting, they do not adequately model scenarios in which a short manipulated segment is inserted into an otherwise authentic video and the original video continues afterward. In this study, we review representative datasets from the literature, analyze their characteristics, and discuss their limitations with respect to temporally localized realistic manipulation detection. Based on this analysis, we motivate the need for a new dataset specifically designed for authentic videos containing short and highly realistic manipulated intervals. Finally, we evaluate two complementary approaches on our custom-curated test set to establish an initial benchmark for this challenging scenario. The first employs a linear probe on DINOv3 features, assessed under three thresholding strategies. The second leverages DINOv3 features with a consecutive frame similarity-based method to detect temporal manipulation boundaries. Together, these experiments provide an initial benchmark for partially manipulated video detection and highlight the need for content-adaptive thresholding mechanisms. The dataset, code, and supplementary materials are publicly available at https://github.com/OkanUmur/temporally-localized-video-manipulation-detection.
Chinese Translation
视频编辑和生成性人工智能技术的快速发展使得现实视频操控变得愈加容易。尽管现有的数据集在深伪检测、物体移除和视频修复等研究中取得了显著进展，但它们并未充分模拟将短暂操控片段插入到其他真实视频中且原始视频在之后继续播放的场景。在本研究中，我们回顾了文献中的代表性数据集，分析了它们的特征，并讨论了它们在时间局部化真实操控检测方面的局限性。基于这一分析，我们强调了需要一个专门设计用于包含短暂且高度真实操控区间的真实视频的新数据集。最后，我们在自定义的测试集上评估了两种互补的方法，以建立这一挑战性场景的初步基准。第一种方法在 DINOv3 特征上使用线性探测，并在三种阈值策略下进行评估。第二种方法利用 DINOv3 特征结合基于连续帧相似度的方法来检测时间操控边界。这些实验共同为部分操控视频检测提供了初步基准，并强调了内容自适应阈值机制的必要性。数据集、代码和补充材料已公开发布在 https://github.com/OkanUmur/temporally-localized-video-manipulation-detection。
View on arXiv Download PDF AI Translation
cs.CV / 54 / 2606.07100

LARA: Latent Action Representation Alignment for Vision-Language-Action Models

LARA：用于视觉-语言-动作模型的潜在动作表示对齐
Liu, Mengya, Jia, Baoxiong, Huang, Jiangyong, Zhang, Jingze, Huang, Siyuan
Abstract
Visual-language action (VLA) models enable robots to predict actions directly from observations and language instructions, but their performance depends on large-scale, high-quality data and is limited by the scarcity of real-world robot action datasets. To facilitate VLA model learning with abundant unlabeled human videos, Latent Action Models (LAM) learn latent action representations from visual dynamics to provide additional supervision for VLA learning. However, LAM and VLA are typically trained separately, leaving LAM ungrounded during VLA training and VLA models constrained by frozen LAM representations. To address these issues, we propose Latent Action Representation Alignment (LARA), a plug-and-play framework that jointly optimizes LAM and VLA via representation alignment. This enables reciprocal benefits where LAMs learn with action trajectories to avoid spurious visual changes, while VLAs are regularized by forward dynamics learned within LAMs to reduce hallucinations of functionally ineffective trajectories. We demonstrate LARA versatility and effectiveness for pre-training, post-training enhancement of pre-trained VLA models, and LAM refinement, achieving an average of ~10%, ~5%, and ~15% improvement over 3 simulation and 1 meticulously designed real-world robotic manipulation benchmarks.
Chinese Translation
视觉-语言动作（VLA）模型使机器人能够直接从观察和语言指令中预测动作，但其性能依赖于大规模、高质量的数据，并受到真实世界机器人动作数据集稀缺的限制。为了利用丰富的未标记人类视频促进VLA模型的学习，潜在动作模型（LAM）从视觉动态中学习潜在动作表示，以为VLA学习提供额外的监督。然而，LAM和VLA通常是分开训练的，这使得在VLA训练期间LAM未得到有效的基础，而VLA模型又受到冻结的LAM表示的限制。为了解决这些问题，我们提出了潜在动作表示对齐（LARA），这是一个即插即用的框架，通过表示对齐共同优化LAM和VLA。这使得LAM能够与动作轨迹共同学习，以避免虚假的视觉变化，而VLA则通过在LAM中学习的前向动态进行正则化，从而减少功能上无效轨迹的幻觉。我们展示了LARA在预训练、对预训练VLA模型的后训练增强以及LAM精炼方面的多样性和有效性，在3个模拟和1个精心设计的真实世界机器人操作基准测试中，分别实现了约10%、5%和15%的平均提升。
View on arXiv Download PDF AI Translation
cs.CV / 55 / 2606.07102

GP-Adapter: Gaussian Process CLIP-Adapter for Few-Shot Out-of-Distribution Detection

GP-Adapter：用于少样本分布外检测的高斯过程CLIP适配器
Saito, Taisei, Ogata, Koretaka, Hiroi, Takafumi
Abstract
We propose GP-Adapter, a training-free framework that augments CLIP (Contrastive Language-Image Pre-training) with Gaussian Process (GP) uncertainty modeling for few-shot classification and out-of-distribution (OOD) detection. While CLIP achieves strong zero-shot recognition, it yields deterministic similarity scores and offers limited uncertainty information, which is critical under distribution shift and data scarcity. GP-Adapter constructs modality-specific, class-wise one-class GPs on top of frozen CLIP embeddings using an RBF kernel for image features and a linear kernel for text prompts and fuses their predictive statistics to produce a variance-aware confidence score for OOD detection. The method requires no fine-tuning of the CLIP backbone and relies only on a small $K$-shot cache and lightweight hyperparameter selection, with memory cost scaling as $O(CK^2)$ for $C$ classes and $K$ shots. Experiments on ImageNet and multiple OOD benchmarks show that GP-Adapter provides competitive few-shot performance and consistently improves OOD detection when combined with prompt-learning baselines, highlighting the complementarity between GP-based uncertainty modeling and prompt learning. Overall, our results suggest that integrating probabilistic inference with large pre-trained vision-language models can improve reliability in low-data and distribution-shifted settings. Code is available at https://github.com/tms-byte/GP-Adapter
Chinese Translation
我们提出了GP-Adapter，这是一种无训练框架，通过高斯过程（Gaussian Process, GP）不确定性建模增强了CLIP（对比语言-图像预训练，Contrastive Language-Image Pre-training），用于少样本分类和分布外（Out-of-Distribution, OOD）检测。尽管CLIP在零样本识别中表现出色，但它产生的相似性得分是确定性的，并且提供的有限不确定性信息在分布变化和数据稀缺的情况下至关重要。GP-Adapter在冻结的CLIP嵌入上构建了特定于模态的、按类划分的一类GP，使用RBF核处理图像特征，使用线性核处理文本提示，并融合它们的预测统计量以生成一个考虑方差的置信分数用于OOD检测。该方法不需要对CLIP主干进行微调，仅依赖于一个小的$K$-shot缓存和轻量级的超参数选择，内存成本随着$O(CK^2)$的比例增长，其中$C$为类的数量，$K$为样本数。在ImageNet和多个OOD基准上的实验表明，GP-Adapter提供了具有竞争力的少样本性能，并在与提示学习基线结合时持续改善OOD检测，突显了基于GP的不确定性建模与提示学习之间的互补性。总体而言，我们的结果表明，将概率推理与大型预训练视觉-语言模型结合可以提高在低数据和分布变化环境下的可靠性。代码可在 https://github.com/tms-byte/GP-Adapter 获取。
View on arXiv Download PDF AI Translation
cs.CV / 56 / 2606.07115

3DMorph: Single-Image-Guided Local 3D Shape Editing and Morphing

3DMorph：单图引导的局部3D形状编辑与变形
Preintner, Tobias, Deng, Yunfei, Müller, Phillip, Illing, Sebastian, König, Adrian, Bäck, Thomas, Raponi, Elena, van Stein, Niki
Abstract
Despite recent progress in 3D generation, intuitive editing of existing shapes remains limited. Unlike images, which benefit from well-established inpainting tools, general 3D objects such as meshes still lack simple and effective methods for local shape editing. Existing approaches are often global, domain-specific, require complex user interaction, or focus on appearance (color and texture) rather than geometry. We introduce 3DMorph, a training-free framework for single-image-guided local 3D shape editing and morphing. Given an edited image showing a desired shape modification, our method automatically localizes the relevant 3D region and transfers 2D modifications to 3D while preserving unmodified areas. 3DMorph also enables intermediate shape generation between the original and edited objects, facilitating design exploration. To benchmark editing quality, we introduce Delta3D, an image-guided local 3D editing benchmark with paired ground-truth edits. Experimental results show that 3DMorph translates intuitive 2D edits into 3D, outperforming state-of-the-art generative and editing methods.
Chinese Translation
尽管在3D生成方面取得了近期进展，但对现有形状的直观编辑仍然有限。与受益于成熟的修复工具的图像不同，网格等一般3D对象仍然缺乏简单有效的局部形状编辑方法。现有的方法通常是全局的、领域特定的，要求复杂的用户交互，或关注外观（颜色和纹理）而非几何。我们提出了3DMorph，一个无需训练的框架，用于单图引导的局部3D形状编辑与变形。给定一张显示所需形状修改的编辑图像，我们的方法自动定位相关的3D区域，并在保留未修改区域的同时将2D修改转移到3D。3DMorph还支持在原始对象和编辑对象之间生成中间形状，促进设计探索。为了评估编辑质量，我们引入了Delta3D，一个带有配对真实编辑的图像引导局部3D编辑基准。实验结果表明，3DMorph能够将直观的2D编辑转化为3D，优于最先进的生成和编辑方法。
View on arXiv Download PDF AI Translation
cs.CV / 57 / 2606.07117

Native3D: End-to-End 3D Scene Generation via Unified Mesh-Texture Modeling and Semantic Alignment

Native3D：通过统一网格-纹理建模和语义对齐实现端到端的3D场景生成
Liu, Yibo, Zhang, Ziwei, Pang, Haozhou, Li, Menghao, He, Lanshan, Qi, Gan
Abstract
This paper presents Native3D, the first end-to-end 3D scene generation framework that completely bypasses 2D intermediate representations. Traditional approaches typically require adapting 3D representations to the 2D domain to leverage pre-trained diffusion models, which inevitably introduces domain adaptation issues including geometric structural distortion and texture detail degradation. To address these limitations, we design a unified mesh-texture joint representation that simultaneously models both geometric structures and texture features through a Transformer-based scene encoder, effectively maintaining spatial relationships and visual consistency among objects within scenes. We further propose the 3D Representation Alignment Loss (3D REPA Loss), which employs an improved contrastive learning mechanism to align multi-level semantic representations in the latent space, significantly enhancing geometric and textural fidelity. Experimental results demonstrate that Native3D outperforms existing methods in both generation quality and editing flexibility, providing a novel solution for 3D scene editing.
Chinese Translation
本文提出了Native3D，这是第一个完全绕过2D中间表示的端到端3D场景生成框架。传统方法通常需要将3D表示适配到2D领域，以利用预训练的扩散模型，这不可避免地引入了领域适配问题，包括几何结构扭曲和纹理细节退化。为了解决这些限制，我们设计了一种统一的网格-纹理联合表示，通过基于Transformer的场景编码器同时建模几何结构和纹理特征，有效保持场景中物体之间的空间关系和视觉一致性。我们进一步提出了3D表示对齐损失（3D REPA Loss），该损失采用改进的对比学习机制来对齐潜在空间中的多层次语义表示，显著增强了几何和纹理的保真度。实验结果表明，Native3D在生成质量和编辑灵活性方面均优于现有方法，为3D场景编辑提供了一种新颖的解决方案。
View on arXiv Download PDF AI Translation
cs.CV / 58 / 2606.07145

Consistent-Inversion: Reverse Consistency Guidance for Structure-Preserving Visual Editing

一致反演：用于结构保持视觉编辑的反向一致性引导
Lu, Xiaocheng, Guo, Jingcai, Guo, Song
Abstract
Text-guided diffusion models have become effective tools for real-image visual editing, where the edited image must follow a target instruction while preserving editing-irrelevant structure. Most training-free editors rely on inversion: a source image is mapped to a noisy latent trajectory and the terminal latent is reused for target-prompt denoising. This reuse is useful for preservation, but it also couples source reconstruction and target editing. The resulting trajectory mismatch may either damage background/layout details or over-constrain the intended edit. This paper presents Consistent-Inversion, a training-free reverse consistency guidance framework for structure-preserving visual editing. Instead of treating the inverted source latent as a fixed initialization, Consistent-Inversion checks whether an intermediate target trajectory can be reversed toward the source inversion trajectory under the source prompt. To make this check well-defined, we construct an auxiliary target-side noise representation, perform source-guided reverse denoising, and use the resulting reverse consistency discrepancy as a correction signal for selected early target denoising steps. The method does not update model parameters, is compatible with inversion-based editors, and introduces only a small inference overhead when applied sparsely. Experiments on PIE-Bench show that Consistent-Inversion improves background and structural fidelity under a unified SD3.5 protocol while maintaining target-prompt alignment, and compatibility experiments further verify the same correction principle on classical Stable-Diffusion inversion pipelines.
Chinese Translation
文本引导的扩散模型已成为真实图像视觉编辑的有效工具，其中编辑后的图像必须遵循目标指令，同时保持与编辑无关的结构。大多数无训练编辑器依赖于反演：源图像被映射到一个噪声潜在轨迹，终端潜在被重用于目标提示去噪。这种重用有助于结构保持，但也将源重建和目标编辑耦合在一起。由此产生的轨迹不匹配可能会损害背景/布局细节或过度限制预期的编辑。本文提出了一种一致反演（Consistent-Inversion）的方法，这是一种无训练的反向一致性引导框架，用于结构保持的视觉编辑。与将反演的源潜在视为固定初始化不同，一致反演检查在源提示下，是否可以将中间目标轨迹反转回源反演轨迹。为了使这一检查明确定义，我们构建了辅助的目标侧噪声表示，执行源引导的反向去噪，并使用得到的反向一致性差异作为选定早期目标去噪步骤的修正信号。该方法不更新模型参数，与基于反演的编辑器兼容，并在稀疏应用时仅引入少量推理开销。在PIE-Bench上的实验表明，一致反演在统一的SD3.5协议下提高了背景和结构的保真度，同时保持了目标提示的对齐，兼容性实验进一步验证了相同的修正原理在经典的稳定扩散反演管道上的有效性。
View on arXiv Download PDF AI Translation
cs.CV / 59 / 2606.07161

TraRA: Trajectory-level Recognition Aggregation for Video Text Spotting in Urban Surveillance

TraRA：用于城市监控视频文本识别的轨迹级识别聚合
Tran, Duc Tri, Nguyen, Trung Thanh, John, Vijay, Nguyen, Phi Le, Kawanishi, Yasutomo
Abstract
Video Text Spotting (VTS) is essential for urban surveillance and intelligent transportation systems, enabling automated reading of street signs, vehicle markings, and scene text in video streams. However, reliable recognition remains challenging due to dynamic video factors common in surveillance scenarios, including motion blur, occlusion, and scale variation, which degrade frame-level recognition. Existing VTS methods typically perform recognition independently on each frame, leading to inconsistent and inaccurate results across sequences. To address these limitations, we propose TraRA (Trajectory-level Recognition Aggregation for VTS), a plug-and-play method that performs trajectory-level text recognition by leveraging temporal and multimodal consistency. TraRA integrates two key modules: (1) the Temporal Clustering and (2) the Vision-Language Aggregation. The former refines noisy trajectories by grouping temporally and visually coherent text instances, while the latter employs a Low-Rank Adaptation-enhanced Vision-Language model to fuse visual cues with linguistic context across frames. By aggregating information over entire text trajectories, TraRA achieves robust text recognition even under challenging surveillance conditions. Extensive experiments on four public benchmarks, including road and urban scene datasets (RoadText, BOVText, ArTVideo, and ICDAR15), demonstrate that TraRA consistently improves tracking and recognition performance over state-of-the-art VTS methods. The source code is available at https://github.com/trid2912/TraRA.
Chinese Translation
视频文本识别（VTS）对于城市监控和智能交通系统至关重要，使得自动读取视频流中的街道标志、车辆标记和场景文本成为可能。然而，由于监控场景中常见的动态视频因素，如运动模糊、遮挡和尺度变化，可靠的识别仍然面临挑战，这些因素会降低帧级识别的效果。现有的 VTS 方法通常在每一帧上独立执行识别，导致序列间结果不一致且不准确。为了解决这些局限性，我们提出了 TraRA（Trajectory-level Recognition Aggregation for VTS），这是一种即插即用的方法，通过利用时间和多模态一致性来执行轨迹级文本识别。TraRA 集成了两个关键模块：（1）时间聚类和（2）视觉-语言聚合。前者通过对时间上和视觉上连贯的文本实例进行分组来精炼噪声轨迹，而后者则采用低秩适应增强的视觉-语言模型，将视觉线索与跨帧的语言上下文融合。通过对整个文本轨迹的信息进行聚合，TraRA 即使在具有挑战性的监控条件下也能实现稳健的文本识别。在四个公共基准数据集上的广泛实验，包括道路和城市场景数据集（RoadText、BOVText、ArTVideo 和 ICDAR15），证明了 TraRA 在跟踪和识别性能上始终优于最先进的 VTS 方法。源代码可在 https://github.com/trid2912/TraRA 获取。
View on arXiv Download PDF AI Translation
cs.CV / 60 / 2606.07171

When Recovery Matters: The Blind Spot of Surrogate Privacy in MLLM Editing

恢复的重要性：MLLM 编辑中代理隐私的盲点
Xu, Siyuan, Liu, Yibing, Chen, Peilin, LI, Yung-Hui, Wang, Shiqi, Kwong, Sam
Abstract
Multimodal Large Language Models (MLLMs) enable flexible instruction-driven image editing, but privacy risks arise when user images expose diverse and user-specific private content. Canonical privacy protection strategies typically substitute sensitive regions with surrogate content before cloud editing. Yet, the resulting output is often an edited surrogate rather than the desired edited source image, neglecting the local recovery in both design and evaluation scope. To this end, we introduce SPPE (Surrogate-based Privacy-Preserving Editing), the first recovery-oriented benchmark covering 36 fine-grained privacy categories and 65 editing instructions. It defines two complementary tasks: 1) editability assessment, which estimates before cloud interaction whether a surrogate can induce an edit consistent with the original image; and 2) surrogate-to-source edit recovery, which evaluates whether the edited surrogate can be transferred back to the private source with the edit effect preserved. We address each task with a dedicated method: ERMA predicts surrogate editability through instruction-aware multimodal relation modeling, while \method performs cycle-consistent recovery by using the surrogate editing pair as visual edit evidence and the source image as a source-preserving anchor. Experiments on SPPE and InstructPix2Pix show consistent improvements on both tasks. For editability assessment, ERMA improves over the best-performing baselines by 13.9% in SRCC and 12.3% in PLCC. For surrogate-to-source edit recovery, C2E-S2SER outperforms SOER across all 8 source integrity and edit consistency metrics on SPPE.
Chinese Translation
多模态大型语言模型（MLLMs）能够实现灵活的基于指令的图像编辑，但当用户图像暴露出多样化和用户特定的私人内容时，隐私风险随之而来。传统的隐私保护策略通常在云编辑之前用代理内容替代敏感区域。然而，最终输出往往是编辑后的代理图像，而非期望的编辑源图像，这在设计和评估范围内忽视了局部恢复。为此，我们提出了 SPPE（基于代理的隐私保护编辑），这是第一个面向恢复的基准，涵盖 36 个细粒度隐私类别和 65 条编辑指令。它定义了两个互补任务：1）可编辑性评估，估计在云交互之前代理是否能够引发与原始图像一致的编辑；2）代理到源的编辑恢复，评估编辑后的代理是否能够在保留编辑效果的情况下转回到私人源。我们为每个任务提出了专门的方法：ERMA 通过基于指令的多模态关系建模预测代理的可编辑性，而 extit{method} 通过使用代理编辑对作为视觉编辑证据和源图像作为源保持锚点来执行循环一致性恢复。在 SPPE 和 InstructPix2Pix 上的实验表明这两个任务均有一致的改进。对于可编辑性评估，ERMA 在 SRCC 上比表现最好的基线提高了 13.9%，在 PLCC 上提高了 12.3%。对于代理到源的编辑恢复，C2E-S2SER 在 SPPE 上的所有 8 个源完整性和编辑一致性指标上均优于 SOER。
View on arXiv Download PDF AI Translation
cs.CV / 61 / 2606.07172

Textual Supervision Enhances Geospatial Representations in Vision-Language Models

文本监督增强视觉-语言模型中的地理空间表征
Locatelli, Marcelo Sartori, Tonucci, Fernando, Kwon, Jea, Vecchietti, Luiz Felipe, Wijaya, Bryan Nathanael, Low, Cheng Yaw, Almeida, Virgilio, Cha, Meeyoung
Abstract
Geospatial understanding is a critical yet underexplored dimension in the development of machine learning systems for tasks such as image geolocation and spatial reasoning. In this work, we analyze the geospatial representations acquired by three model families: vision-only architectures (e.g., ViT), vision-language models (e.g., CLIP), and large-scale multimodal foundation models (e.g., LLaVA, Qwen, and Gemma). By evaluating across image clusters, including people, landmarks, and everyday objects, grouped based on the degree of localizability, we reveal systematic gaps in spatial accuracy and show that textual supervision enhances the learning of geospatial representations. Our findings suggest the role of language as an effective complementary modality for encoding spatial context and multimodal learning as a key direction for advancing geospatial AI.
Chinese Translation
地理空间理解是机器学习系统在图像地理定位和空间推理等任务中一个关键但尚未充分探索的维度。在本研究中，我们分析了三种模型家族所获得的地理空间表征：仅视觉架构（例如，ViT）、视觉-语言模型（例如，CLIP）以及大规模多模态基础模型（例如，LLaVA、Qwen 和 Gemma）。通过对基于可定位性程度分组的人物、地标和日常物品等图像集群进行评估，我们揭示了空间准确性方面的系统性差距，并表明文本监督增强了地理空间表征的学习。我们的研究结果表明，语言作为编码空间上下文的有效补充模态的作用，以及多模态学习作为推动地理空间人工智能发展的关键方向。
View on arXiv Download PDF AI Translation
cs.CV / 62 / 2606.07175

Seeing Without Exposing: Adaptive Privacy Control for Open-World, Context-Hungry MLLMs

隐私控制的适应性：开放世界中上下文敏感的多模态大语言模型的无暴露观察
Xu, Siyuan, Liu, Yibing, Chen, Peilin, Li, Yung-Hui, Wang, Shiqi, Kwong, Sam
Abstract
Multimodal large language models (MLLMs) have raised new privacy challenges. On the data side, user-provided inputs often include unpredictable sensitive information; while on the downstream task side, model reasoning depends on rich visual context that may itself be privacy-sensitive. Existing privacy protection methods, however, rely on predefined sensitive categories and fixed obfuscation strategies, struggling to tackle such challenges in MLLMs. To address this dilemma, we propose Anchored Privacy Drifting (APD), a training-free method that drifts privacy-sensitive elements toward semantically equivalent alternatives while anchoring contextual cues to the source image. To systematically evaluate this dual objective of privacy protection and contextual preservation, we introduce AdaptShield, a comprehensive benchmark covering 22 privacy categories, which combines conventional privacy metrics with MLLM-based assessments of contextual utility. Extensive experiments show that our method achieves balanced improvements in both privacy sanitization and content retention, with average gains of 10.4% on textual categories and 8.5% under MLLM-based evaluation across four MLLM series, i.e., Qwen2.5, Qwen3, InternVL3, and InternVL3.5.
Chinese Translation
多模态大语言模型（MLLMs）带来了新的隐私挑战。在数据方面，用户提供的输入往往包含不可预测的敏感信息；而在下游任务方面，模型推理依赖于丰富的视觉上下文，这些上下文本身可能也涉及隐私敏感性。然而，现有的隐私保护方法依赖于预定义的敏感类别和固定的模糊化策略，难以应对MLLMs中的这些挑战。为了解决这一困境，我们提出了锚定隐私漂移（Anchored Privacy Drifting, APD），这是一种无训练的方法，通过将隐私敏感元素漂移到语义等价的替代品，同时将上下文线索锚定到源图像上。为了系统地评估隐私保护和上下文保留的双重目标，我们引入了AdaptShield，这是一个涵盖22个隐私类别的综合基准，结合了传统隐私指标和基于MLLM的上下文效用评估。大量实验表明，我们的方法在隐私净化和内容保留方面均取得了平衡的改善，在文本类别上平均提升10.4%，在基于MLLM的评估中平均提升8.5%，涉及四个MLLM系列，即Qwen2.5、Qwen3、InternVL3和InternVL3.5。
View on arXiv Download PDF AI Translation
cs.CV / 63 / 2606.07179

EvoGS: Constructing Continuous-Layered Gaussian Splatting with Evolution Tree for Scalable 3D Streaming

EvoGS：构建具有进化树的连续分层高斯溅射以实现可扩展的3D流媒体
Shi, Yuang, Gasparini, Simone, Morin, Géraldine, Ooi, Wei Tsang
Abstract
Streaming 3D Gaussian Splatting requires highly scalable, progressive representations. Existing progressive methods rely on \textit{discrete layering}, accumulating separate splat sets for each level of detail. This structural independence between layers inherently leads to error accumulation, severe splat redundancy, and uncontrolled quality transitions. We propose EvoGS, the first \textit{continuous-layering} representation. Organized as an Evolution Tree, EvoGS generates finer details via an explicit, wavelet-inspired parent-child refinement. This empowers child nodes to structurally correct ancestral errors, yield inherently sparse and highly compressible inter-layer signals. Extensive experiments show EvoGS eliminates splat redundancy from over 65\% to under 25\%. Compared to state-of-the-art baselines, it reduces transmission payload and GPU VRAM footprint by up to 2.4$\times$ and 5.5$\times$, respectively, and achieves smooth quality transitions optimal for real-time adaptive streaming. Project page: https://yuang-ian.github.io/evogs/
Chinese Translation
流媒体3D高斯溅射需要高度可扩展的渐进表示。现有的渐进方法依赖于 extit{离散分层}，为每个细节层次积累独立的溅射集。这种层之间的结构独立性固有地导致了误差累积、严重的溅射冗余和不受控制的质量过渡。我们提出了EvoGS，这是第一个 extit{连续分层}表示。EvoGS以进化树的形式组织，通过显式的、受小波启发的父子细化生成更精细的细节。这使得子节点能够在结构上纠正祖先的错误，产生固有稀疏且高度可压缩的层间信号。大量实验表明，EvoGS将溅射冗余从超过65\%降低到不足25\\%。与最先进的基线相比，它将传输负载和GPU VRAM占用分别减少了最多2.4$ imes$和5.5$ imes$，并实现了适合实时自适应流媒体的平滑质量过渡。项目页面：https://yuang-ian.github.io/evogs/
View on arXiv Download PDF AI Translation
cs.CV / 64 / 2606.07180

OPTIMUS-Prime: Minimal and Sufficient Concept Explanations for Deep Vision Models

OPTIMUS-Prime：深度视觉模型的最小和充分概念解释
Hoarau, Arthur, Zhu, Chenrui, Nguyen, Vu Linh
Abstract
The growing demand for transparency in automated decision-making has propelled eXplainable Artificial Intelligence (XAI) to the forefront of machine learning research. In computer vision, however, existing explanation methods often prioritize end-user accessibility at the expense of formal guarantees, leaving a critical gap between practical utility and theoretical rigor. In this paper, we address this gap by introducing OPTIMUS, a novel framework for generating concept-based visual explanations for deep classification models. OPTIMUS explanations take the form of visual heatmaps that not only remain interpretable to end users, but are grounded in the well-established theory of prime implicants, providing formal guarantees that have been largely absent from existing saliency-based methods. Specifically, OPTIMUS explanations satisfy two desirable properties: sufficiency, ensuring that the highlighted concepts provably guarantee the classifier's prediction, and minimality, ensuring that no strict subset of those concepts retains this guarantee. Together, these properties yield explanations that are both logically tight and visually coherent. We validate our approach on a visual classification benchmark, demonstrating that OPTIMUS heatmaps naturally and faithfully surface the decision-relevant concepts underlying model predictions.
Chinese Translation
对自动决策透明度日益增长的需求推动了可解释人工智能（XAI）成为机器学习研究的前沿。然而，在计算机视觉领域，现有的解释方法往往优先考虑最终用户的可及性，而牺牲了形式上的保证，导致实用性与理论严谨性之间存在关键差距。本文通过引入OPTIMUS，一个用于生成基于概念的深度分类模型视觉解释的新框架，来解决这一差距。OPTIMUS的解释以视觉热图的形式呈现，这些热图不仅对最终用户可解释，而且基于已建立的主要蕴涵理论，提供了在现有基于显著性的方法中大多缺失的形式保证。具体而言，OPTIMUS的解释满足两个理想属性：充分性，确保突出显示的概念可以证明地保证分类器的预测；以及最小性，确保没有这些概念的严格子集保留这一保证。这两个属性共同产生了逻辑严密且视觉连贯的解释。我们在一个视觉分类基准上验证了我们的方法，证明OPTIMUS热图自然且忠实地揭示了模型预测背后的决策相关概念。
View on arXiv Download PDF AI Translation
cs.CV / 65 / 2606.07185

AdaTok: Self-Budgeting Image Tokenization with Quality-Preserving Dynamic Tokens

AdaTok：具有质量保留的动态自预算图像标记化
Lu, Xiaocheng, Chen, Yuxi, Zhang, Jie, Liu, Jian, Guo, Jingcai, Zhu, Fangqi, Han, Tao, Guo, Song
Abstract
Image tokenizers, from 2D grids to recent 1D sequences, typically encode every image with the same fixed number of tokens. Yet visual complexity is highly heterogeneous, so a uniform budget overspends on simple inputs and underserves complex ones. Existing elastic tokenizers expose variable-length reconstructions, but often leave token length as a deployment-time operating point, a search target, or an external prediction rather than an output of the tokenizer itself. In this work, we ask whether a discrete visual tokenizer can budget itself in one pass. Our central finding is that actionable elasticity requires a representation--allocation co-design: prefixes must remain decodable across budgets, and the tokenizer must learn which prefix each image needs. We propose AdaTok, a self-budgeting discrete 1D tokenizer. AdaTok combines Prioritized Representation Learning, which orders tokens with nested tail masking and resolves budget-dependent semantic shift through Multi-Head LoRA decoder heads, with Adaptive Token Allocation, which trains a lightweight deterministic-group GRPO policy over candidate budgets. Dynamic Pareto Weighting balances fidelity and efficiency during policy training without manual trade-off sweeps. On ImageNet-1K, AdaTok-Full reaches rFID 1.31 at 256 tokens, while AdaTok-Adaptive attains rFID 1.50 using only ~118 tokens on average, outperforming discrete 1D baselines at comparable budgets. In autoregressive image generation, the shorter adaptive representation yields ~2.1x throughput over a fixed 256-token decode, suggesting that visual token count can be learned as a content-conditioned output rather than set as a fixed hyperparameter.
Chinese Translation
图像标记器，从二维网格到最近的一维序列，通常使用相同固定数量的标记对每个图像进行编码。然而，视觉复杂性高度异质，因此统一预算在简单输入上过度支出，而在复杂输入上则服务不足。现有的弹性标记器展示了可变长度的重构，但通常将标记长度视为部署时的操作点、搜索目标或外部预测，而不是标记器自身的输出。在本研究中，我们探讨了离散视觉标记器是否可以在一次传递中自我预算。我们的核心发现是，切实可行的弹性需要表示-分配的共同设计：前缀必须在预算之间保持可解码性，标记器必须学习每个图像所需的前缀。我们提出了AdaTok，一种自预算的离散一维标记器。AdaTok结合了优先表示学习（Prioritized Representation Learning），通过嵌套尾部掩蔽对标记进行排序，并通过多头LoRA解码器头解决预算依赖的语义偏移，以及自适应标记分配（Adaptive Token Allocation），该方法在候选预算上训练轻量级的确定性组GRPO策略。动态帕累托加权在策略训练过程中平衡了保真度和效率，而无需手动权衡。 在ImageNet-1K上，AdaTok-Full在256个标记时达到rFID 1.31，而AdaTok-Adaptive在平均仅使用约118个标记时获得rFID 1.50，超越了在可比预算下的离散一维基线。在自回归图像生成中，较短的自适应表示在固定256标记解码上实现了约2.1倍的吞吐量，表明视觉标记数量可以作为内容条件输出进行学习，而不是设定为固定超参数。
View on arXiv Download PDF AI Translation
cs.CV / 66 / 2606.07222

DualGate-Net: A Prior-Gated Dual-Encoder Framework for Histopathology Cell Detection

DualGate-Net：一种用于组织病理学细胞检测的先验门控双编码器框架
Tabaghsar, Bahman Jafari, Tran, Son, Devaraja, K., Sajjanhar, Atul
Abstract
Cell detection in histopathology images strongly depends on surrounding tissue context, where visually similar cells may belong to different classes under different microenvironments. Recent tissue-aware methods incorporate contextual priors, but often rely on static fusion strategies that may propagate noisy information. In this work, we propose DualGate-Net, a prior-aware dual-encoder framework that combines a ConvNeXtV2-based local encoder and a SegFormer-based global encoder through a learnable prior-gated fusion mechanism. The proposed module adaptively regulates the influence of tissue priors across spatial locations, while an auxiliary foreground reconstruction branch preserves high-frequency cellular structures during training. In addition, auxiliary cellness-guided cues are incorporated to further improve localization robustness. Experiments on the OCELOT benchmark demonstrate consistent improvements, achieving macro F1-scores of 0.7722 on the validation set and 0.7345 on the test set, highlighting the effectiveness of adaptive prior integration for robust histopathology cell detection.
Chinese Translation
在组织病理学图像中，细胞检测强烈依赖于周围组织的上下文，其中在不同微环境下视觉上相似的细胞可能属于不同的类别。近期的组织感知方法结合了上下文先验，但通常依赖于静态融合策略，这可能会传播噪声信息。在本研究中，我们提出了DualGate-Net，这是一种先验感知的双编码器框架，通过可学习的先验门控融合机制结合了基于ConvNeXtV2的局部编码器和基于SegFormer的全局编码器。所提出的模块自适应地调节组织先验在空间位置上的影响，同时辅助的前景重建分支在训练过程中保留高频细胞结构。此外，还结合了辅助的细胞性引导线索，以进一步提高定位的鲁棒性。在OCELOT基准测试上的实验结果显示出一致的改进，在验证集上实现了0.7722的宏F1分数，在测试集上实现了0.7345，突显了自适应先验融合在鲁棒的组织病理学细胞检测中的有效性。
View on arXiv Download PDF AI Translation
cs.CV / 67 / 2606.07233

Does Appearance Help? A Systematic Study of Image-Based Re-Identification in Online 3D Multi-Pedestrian Tracking

外观有帮助吗？在线3D多行人跟踪中基于图像的再识别的系统研究
Borges, Eduardo, Garrote, Luís, Nunes, Urbano J.
Abstract
LiDAR-based 3D Multi-Object Tracking (MOT) typically relies solely on geometric information, which is often insufficient to distinguish between targets during prolonged occlusions or in crowded human-populated environments. While integrating RGB-based Re-Identification (ReID) offers a theoretical solution for preserving identity context, existing approaches often rely on computationally expensive parallel detectors that hinder real-time robot responsiveness. This work presents a systematic study of image-based ReID in online 3D MOT, utilizing a lightweight projection-based framework to decouple geometric and appearance modeling for mobile robots. A comprehensive analysis of feature extraction architectures is conducted, employing lightweight CNNs and Vision Transformers, and evaluating various multi-modal data association strategies to balance computational latency with robust tracking. Experiments on the Pedestrian class of the KITTI dataset reveal that naive linear fusion, of appearance and motion costs, degrades performance due to visual noise. Conversely, a cascaded matching strategy successfully recovers occluded tracks without compromising overall precision, effectively preventing identity switches to maintain human-robot interaction continuity. We show that lightweight architectures can offer an optimal trade-off between the low latency required for safe navigation and the discriminative power needed for social awareness.
Chinese Translation
基于LiDAR的3D多目标跟踪（MOT）通常仅依赖于几何信息，这在长时间遮挡或拥挤的人群环境中往往不足以区分目标。虽然集成基于RGB的再识别（ReID）为保持身份上下文提供了理论解决方案，但现有方法通常依赖于计算成本高昂的并行检测器，这妨碍了实时机器人响应。本文提出了一项关于在线3D MOT中基于图像的ReID的系统研究，利用轻量级投影框架将几何建模与外观建模解耦，以适应移动机器人。我们对特征提取架构进行了全面分析，采用轻量级卷积神经网络（CNN）和视觉变换器（Vision Transformers），并评估了各种多模态数据关联策略，以平衡计算延迟与稳健跟踪。在KITTI数据集的行人类别上的实验表明，简单的线性融合外观和运动成本由于视觉噪声而降低了性能。相反，级联匹配策略成功恢复了被遮挡的轨迹，而不影响整体精度，有效防止了身份切换，以维持人机交互的连续性。我们展示了轻量级架构能够在安全导航所需的低延迟与社会意识所需的区分能力之间提供最佳权衡。
View on arXiv Download PDF AI Translation
cs.CV / 68 / 2606.07249

Reconstructing Multi-Decadal Forest Disturbances: A Spatio-Temporal Transformer Approach

重建多年代森林干扰：一种时空变换器方法
Scheibenreif, Linus, Raichuk, Anton, Neumann, Maxim
Abstract
Accurate monitoring of forest disturbances is essential for understanding carbon dynamics and land management, yet traditional approaches typically rely on pixel-wise analysis of satellite time-series, ignoring spatial context. We present a deep learning framework that maps 38 years (1984-2022) of forest disturbance across the contiguous United States by modeling temporal trajectories and spatial neighborhoods simultaneously. By leveraging a vision transformer architecture, our approach effectively filters noise from weak supervision signals to produce spatially coherent disturbance maps. We perform exhaustive evaluations across multiple satellites (Landsat, Sentinel-1, Sentinel-2) and temporal windows (38 years and the more recent 6 years), validating performance against a novel, manually annotated validation dataset (n=300) and independent fire perimeter dataset (n=706). The results highlight the complexity of the task: while our spatio-temporal model demonstrates high precision (up to 98.2% for +-1 year detection on MTBS and up to 71.3% on the CONUS validation datasets, with F1-scores up to 75.8% and 47.3%, respectively) and effectively reduces spatial artifacts, it exhibits performance trade-offs across different disturbance regimes compared to pixel-wise baselines. Our method offers a promising foundation for consistent forest monitoring.
Chinese Translation
准确监测森林干扰对于理解碳动态和土地管理至关重要，然而传统方法通常依赖于对卫星时间序列的像素级分析，忽视了空间背景。我们提出了一种深度学习框架，通过同时建模时间轨迹和空间邻域，绘制了1984年至2022年间美国大陆地区38年的森林干扰图。通过利用视觉变换器架构，我们的方法有效地过滤了来自弱监督信号的噪声，生成了空间一致的干扰图。我们在多个卫星（Landsat、Sentinel-1、Sentinel-2）和时间窗口（38年及最近的6年）上进行了全面评估，并通过一个新颖的手动标注验证数据集（n=300）和独立的火灾边界数据集（n=706）验证了性能。结果突显了这一任务的复杂性：尽管我们的时空模型展示了高精度（在MTBS上达到98.2%的±1年检测精度，在CONUS验证数据集上达到71.3%，F1分数分别为75.8%和47.3%），并有效减少了空间伪影，但与像素级基线相比，它在不同干扰机制下表现出性能权衡。我们的方法为一致的森林监测提供了一个有希望的基础。
View on arXiv Download PDF AI Translation
cs.CV / 69 / 2606.07280

Geometric-Aware Hypergraph Reasoning for Novel Class Discovery in Point Cloud Segmentation

基于几何感知的超图推理在点云分割中的新类发现
Zhang, Zihao, Wu, Aming, Li, Yang, Han, Yahong, Shen, Jialie
Abstract
Novel class discovery in point cloud segmentation aims to transfer knowledge from known classes to automatically identify and segment unlabeled novel classes in point clouds. Existing methods mainly rely on pairwise associations for class assignment and novel class reasoning, which limits their ability to capture complex relationships among known and novel classes and may lead to inaccurate semantic segmentation. To address this issue, we introduce a hypergraph-based framework that models high-order associations among classes and enables collaborative reasoning from known classes to novel classes beyond traditional pairwise relations. Moreover, existing methods tend to focus on semantic feature extraction while paying insufficient attention to geometric information in point clouds. To better exploit spatial structure, we propose Geometric-Aware Prototypes to enhance the representation of class-level geometric cues. By propagating geometric information through hyperedges, the proposed method improves the understanding of spatial distributions across classes and leads to more accurate segmentation. Experiments on the SemanticKITTI and SemanticPOSS datasets demonstrate the effectiveness and superiority of our method.
Chinese Translation
点云分割中的新类发现旨在将已知类的知识转移，以自动识别和分割点云中的未标记新类。现有方法主要依赖于成对关联进行类分配和新类推理，这限制了它们捕捉已知类与新类之间复杂关系的能力，并可能导致语义分割的不准确。为了解决这个问题，我们引入了一种基于超图的框架，该框架建模类之间的高阶关联，并使已知类与新类之间的协同推理超越传统的成对关系。此外，现有方法往往侧重于语义特征提取，而对点云中的几何信息关注不足。为了更好地利用空间结构，我们提出了几何感知原型（Geometric-Aware Prototypes），以增强类级几何线索的表示。通过通过超边传播几何信息，所提方法改善了对类之间空间分布的理解，并导致更准确的分割。在SemanticKITTI和SemanticPOSS数据集上的实验验证了我们方法的有效性和优越性。
View on arXiv Download PDF AI Translation
cs.CV / 70 / 2606.07288

ExMesh: EXplicit Mesh Reconstruction with Topology Adaptation

ExMesh：具有拓扑适应的显式网格重建
Fan, Chuanjin, Wu, Lifan, Chang, Wenjie, Chang, Hanzhi, Yang, Wenfei, Zhang, Tianzhu
Abstract
Reconstructing surface meshes from multi-view images has remained a core challenge in recent years. Most existing methods, whether implicit or explicit, depend on intermediate representations and post-processing steps like Marching Cubes or TSDF fusion, often resulting in artifacts and fragmented geometry. Directly optimizing explicit meshes is a promising approach. However, it presents two critical challenges. The first is how to adaptively refine mesh topology to capture detail without introducing degenerate faces. The second is how to maintain consistent UV coordinates for high-fidelity texturing as the mesh structure evolves. To overcome these, we propose ExMesh, a novel framework that directly optimizes explicit meshes by integrating differentiable optimization with discrete topology updates. Specifically, we introduce an adaptive vertex splitting and merging strategy, along with real-time UV maintenance, to enable coarse-to-fine optimization while preserving geometric integrity. To our knowledge, ExMesh is the first framework to seamlessly integrate discrete topology operations into a continuous differentiable optimization pipeline. Extensive experiments demonstrate that ExMesh achieves a balance among accuracy, computational efficiency, and mesh conciseness.
Chinese Translation
从多视角图像重建表面网格在近年来仍然是一个核心挑战。大多数现有方法，无论是隐式还是显式，都依赖于中间表示和后处理步骤，如Marching Cubes或TSDF融合，常常导致伪影和几何碎片化。直接优化显式网格是一种有前景的方法。然而，这带来了两个关键挑战。首先是如何自适应地细化网格拓扑以捕捉细节，而不引入退化面。其次是如何在网格结构演变的过程中保持一致的UV坐标，以实现高保真纹理映射。为了解决这些问题，我们提出了ExMesh，一个新颖的框架，通过将可微优化与离散拓扑更新相结合，直接优化显式网格。具体而言，我们引入了一种自适应顶点分裂和合并策略，以及实时UV维护，以支持从粗到细的优化，同时保持几何完整性。据我们所知，ExMesh是第一个将离散拓扑操作无缝集成到连续可微优化管道中的框架。大量实验表明，ExMesh在准确性、计算效率和网格简洁性之间达成了平衡。
View on arXiv Download PDF AI Translation
cs.CV / 71 / 2606.07311

CULTURESCORE: Evaluating Cultural Faithfulness in Video Generation Models

CULTURESCORE：评估视频生成模型中的文化忠实性
Rani, Anku, Dai, Wei, Nayak, Shravan, Maes, Pattie, Kalayeh, Mahdi M., Liang, Paul Pu
Abstract
As video generation models like Veo 3.1 and LTX-2 advance, their ability to accurately represent diverse global cultures remains a critical yet understudied frontier. Current metrics, such as VideoScore, only measure visual quality but offer no mechanism for assessing cultural faithfulness. Consequently, a model that replaces a Namaste with a handshake receives the same score as one that generates the gesture correctly. We propose CultureScore, a compositional evaluation framework that decomposes cultural faithfulness into three granular dimensions: Identity (who is represented), Context (culturally localized background), and Behavior (normative gestures and interactions). We operationalize this framework through an evaluation suite spanning 10 countries, yielding 6,180 generated videos across three state-of-the-art models. Our evaluation reveals that no current model achieves culturally faithful video generation: the best-performing model reaches only 56.8\% overall CultureScore, with Behavior the most challenging dimension, which remains below 52\% across all models. Furthermore, human preference rankings align directionally with CultureScore but are inverted relative to VideoScore; the highest-scoring model on visual quality was ranked last by annotators, underscoring that cultural faithfulness is an essential criterion for equitable video generation.
Chinese Translation
随着视频生成模型如Veo 3.1和LTX-2的进步，它们准确表现多元全球文化的能力仍然是一个关键但未被充分研究的前沿领域。目前的指标，如VideoScore，仅测量视觉质量，但没有评估文化忠实性的机制。因此，一个将Namaste替换为握手的模型与正确生成该手势的模型获得相同的分数。我们提出了CultureScore，一个组成性评估框架，将文化忠实性分解为三个细化维度：身份（谁被代表）、背景（文化本地化背景）和行为（规范性手势和互动）。我们通过一个涵盖10个国家的评估套件来实现这一框架，生成了6,180个视频，涉及三种最先进的模型。我们的评估显示，当前没有模型能够实现文化忠实的视频生成：表现最佳的模型仅达到56.8%的整体CultureScore，而行为维度是最具挑战性的，在所有模型中均低于52%。此外，人类偏好排名在方向上与CultureScore一致，但相对于VideoScore则呈现反向关系；在视觉质量上得分最高的模型在标注者中排名最后，强调了文化忠实性是公平视频生成的重要标准。
View on arXiv Download PDF AI Translation
cs.CV / 72 / 2606.07326

AnchorWorld: Embodied Egocentric World Simulation with View-based Evolution Customization

AnchorWorld：基于视图演化定制的具身自我中心世界模拟
Li, Yu, Xia, Menghan, Liu, Gongye, Wang, Xintao, Zhang, Conglang, Ke, Lei, Lin, Yuxuan, Chu, Ruihang, Wan, Pengfei, Gai, Kun, Yang, Yujiu
Abstract
Despite being a pivotal frontier, interactive world modeling remains underexplored in terms of the versatile controllability required by practical scenarios. To bridge this gap, we present AnchorWorld, a framework that advances egocentric simulation through enhanced interaction integrity and a flexible mechanism for world customization. First, we utilize 3D human motion as the primary interaction modality. To complement the out-of-view or truncated body parts in egocentric views, we introduce an auxiliary training supervision that incorporates exogenous viewpoints decoupled from the agent's first-person sensorium. It allows the model to observe the agent's full-body positioning relative to the environment, facilitating a more robust spatial grounding of human-world interactions. Furthermore, we propose a simple yet effective mechanism for customizing self-evolving worlds. This is achieved by defining anchor views within a unified world coordinate system, coupled with textual descriptions dictating the dynamic evolution of local scenes. Experimental results show that AnchorWorld significantly outperforms state-of-the-art baselines, while ablation studies validate the effectiveness of our key designs. Notably, our customization scheme exhibits promising spatio-temporal geometric consistency and adheres strictly to the prescribed evolutionary dynamics.
Chinese Translation
尽管互动世界建模是一个关键前沿领域，但在实际场景所需的多功能可控性方面仍然未得到充分探索。为了解决这一问题，我们提出了AnchorWorld，一个通过增强交互完整性和灵活的世界定制机制来推进自我中心模拟的框架。首先，我们利用3D人体运动作为主要的交互方式。为了补充自我中心视图中不可见或截断的身体部位，我们引入了一种辅助训练监督，该监督结合了与代理人第一人称感知解耦的外部视点。这使得模型能够观察代理人在环境中的全身定位，从而促进人类与世界交互的更强空间基础。此外，我们提出了一种简单而有效的自我演化世界定制机制。这是通过在统一的世界坐标系统中定义锚视图，并结合文本描述来指引局部场景的动态演变来实现的。实验结果表明，AnchorWorld显著优于最先进的基线，而消融研究验证了我们关键设计的有效性。值得注意的是，我们的定制方案展现出良好的时空几何一致性，并严格遵循规定的演变动态。
View on arXiv Download PDF AI Translation
cs.CV / 73 / 2606.07333

Varifold Moment Invariants for Sustainable and Explainable Contour Feature Extraction

可持续和可解释的轮廓特征提取中的变体矩不变性
Longari, G., Paiva, J. -C. Alvarez, Tumpach, A. B.
Abstract
We introduce Varifold Moments Invariants (VMI) as a unifying framework for many previously introduced Moment Invariants. These invariants are deeply related to other contour features that are invariant under translations and rotations, like Extended Gaussian Image, Elliptic Fourier Descriptors or Shape Distributions. The advantage of the varifold approach to moments consists in being able to combine the geometry of the region, its boundary, and the family of lines tangent to it, in order to create a substantial number of invariant features with high discriminating power and clear geometric meaning. By coupling our VMI feature extraction with the light feature classifiers Random Forest or Multi-Layer-Perceptron, we outperform state-of-the-art approaches based on contours, while decreasing drastically the computational cost to the point of allowing our algorithm to run on light devices. We tested our approach on classification tasks on a large number of widely-used datasets of various types (leaves, objects, cells) and achieved high accuracy with a low number of geometrically interpretable features.
Chinese Translation
我们引入变体矩不变性（Varifold Moments Invariants, VMI）作为一个统一框架，以整合许多先前提出的矩不变性。这些不变性与其他在平移和旋转下不变的轮廓特征密切相关，如扩展高斯图像（Extended Gaussian Image）、椭圆傅里叶描述子（Elliptic Fourier Descriptors）或形状分布（Shape Distributions）。变体方法在矩的优势在于能够结合区域的几何形状、其边界以及与之相切的直线族，从而创造出大量具有高区分能力和明确几何意义的不变特征。通过将我们的VMI特征提取与轻量特征分类器随机森林（Random Forest）或多层感知器（Multi-Layer-Perceptron）相结合，我们在基于轮廓的最先进方法中表现优越，同时大幅降低计算成本，使我们的算法能够在轻量设备上运行。我们在多种类型（叶子、物体、细胞）的广泛使用数据集上进行了分类任务测试，取得了高准确率，并且仅使用了少量几何可解释特征。
View on arXiv Download PDF AI Translation
cs.CV / 74 / 2606.07338

VeriDrive: Verifiable Counterfactual Supervision for Cost-Efficient Vision-Language Planning

VeriDrive：可验证的反事实监督用于成本高效的视觉-语言规划
Zhang, Zikai, Shum, Hubert P. H., Breckon, Toby P.
Abstract
Vision-language driving models increasingly use reasoning supervision to bridge perception, prediction, and planning, but existing driving rationales are often free-form and expensive to generate with frontier models. We present VeriDrive, a framework for constructing planning-oriented, verifiable counterfactual supervision. VeriDrive converts driving reasoning into a structured Perception-Evaluation-Revision chain that grounds key objects in future motion, evaluates alternative ego trajectories with rule-checkable evidence, revises risky intent toward expert behavior, and produces final planning targets. To scale data construction, VeriDrive combines local generation with validator-guided selective correction, escalating only invalid or difficult samples. We build the VeriDrive dataset on nuScenes and train under the Omni-Q protocol. Controlled open-loop experiments show that VeriDrive improves L2, Collision, and Intersection over OmniDrive while reducing logged token usage, generation time, and actual paid LLM/VLM cost. These results show that auditable intermediate fields and structured revision targets can improve vision-language planning supervision under realistic annotation budgets. Code, prompts, and validator scripts are coming soon and will be released after the review process.
Chinese Translation
视觉-语言驾驶模型越来越多地使用推理监督来桥接感知、预测和规划，但现有的驾驶推理往往是自由形式的，并且使用前沿模型生成成本高昂。我们提出了VeriDrive，一个构建面向规划的可验证反事实监督的框架。VeriDrive将驾驶推理转化为一个结构化的感知-评估-修订链，基于未来运动对关键对象进行定位，使用可规则检查的证据评估替代的自我轨迹，修订朝向专家行为的风险意图，并生成最终的规划目标。为了扩展数据构建，VeriDrive结合了局部生成与验证器引导的选择性修正，仅对无效或困难样本进行升级。我们在nuScenes上构建了VeriDrive数据集，并在Omni-Q协议下进行训练。控制的开环实验表明，VeriDrive在L2、碰撞和交叉口方面优于OmniDrive，同时减少了记录的令牌使用、生成时间和实际支付的LLM/VLM成本。这些结果表明，可审计的中间字段和结构化的修订目标可以在现实的标注预算下改善视觉-语言规划监督。代码、提示和验证器脚本即将发布，并将在审稿过程后发布。
View on arXiv Download PDF AI Translation
cs.CV / 75 / 2606.07355

Spatial-Temporal Decoupled Adapter for Micro-gesture Online Recognition

用于微手势在线识别的时空解耦适配器
Shen, Xucheng, Li, Kun, Wang, Fei, Qian, Wei, Jiang, Jin, Guo, Dan
Abstract
Micro-gesture online recognition aims to temporally localize and classify subtle gestures in untrimmed videos. Owing to their extremely short duration, low motion amplitude, and ambiguous visual cues, capturing discriminative spatiotemporal representations remains highly challenging. Existing parameter-efficient adapters typically employ a single branch to model spatial and temporal cues jointly, which may fail to capture the fine-grained patterns of micro-gestures. To address this limitation, we propose a Spatial-Temporal Decoupled Adapter that decomposes video adaptation into independent temporal and spatial branches via lightweight depthwise convolutions. In addition, to address the long-tail distribution problem in the benchmark dataset, we introduce Adaptive Soft Balanced Augmentation, which dynamically allocates augmentation intensity based on class rarity and learning difficulty, without manual thresholds. Our method achieves an F1 score of 0.43808, ranking 1st in Track 2 of the 4th EI-MiGA-IJCAI Challenge.
Chinese Translation
微手势在线识别旨在对未裁剪视频中的细微手势进行时间定位和分类。由于微手势的持续时间极短、运动幅度低以及视觉线索模糊，捕捉具有区分性的时空表示仍然非常具有挑战性。现有的参数高效适配器通常采用单一分支来联合建模空间和时间线索，这可能无法捕捉微手势的细粒度模式。为了解决这一局限性，我们提出了一种时空解耦适配器（Spatial-Temporal Decoupled Adapter），通过轻量级深度卷积将视频适配分解为独立的时间和空间分支。此外，为了解决基准数据集中长尾分布的问题，我们引入了自适应软平衡增强（Adaptive Soft Balanced Augmentation），该方法根据类别稀缺性和学习难度动态分配增强强度，无需手动阈值。我们的方法在第4届EI-MiGA-IJCAI挑战赛的第2赛道中取得了0.43808的F1得分，排名第一。
View on arXiv Download PDF AI Translation
cs.CV / 76 / 2606.07366

Dash2Sim: Closed-Loop Driving Simulation from in-the-wild Dashcam Videos

Dash2Sim：基于真实环境行车记录仪视频的闭环驾驶仿真
Ghosh, Anurag, Pittaluga, Francesco, Vuong, Khiem, Chen, Angela, Alvarez-Padilla, Juan, Chandraker, Manmohan, Narasimhan, Srinivasa
Abstract
Self-driving simulations typically rely on data collected in a small number of cities or on hand-authored synthetic scenarios. Dashcam videos cover a far broader range of locations and situations, including rare or long-tailed scenarios. They are considered less usable for simulation because it is difficult to recover accurate 4D scenes from monocular in-the-wild videos. Work zones are one such class of long-tailed situations that dashcams capture. We present Dash2Sim, a framework that turns in-the-wild monocular dashcam videos into metric, geo-referenced 4D driving logs compatible with existing simulators, and verifies eachone against an independently maintained map without annotations. We apply Dash2Sim to a large video corpus to create the ROADWork4D benchmark dataset, which spans 4,244 scenes with 2.7M 3D objects across 17 cities. On a verified subset ROADWork4D-CL (2,201 scenes), we study privileged closed-loop planners and find that work zone scenarios are difficult: while rule-based and hybrid planners generalize better than learning-based ones, all fall short, failing to make the lane changes that temporary work zone channels require. Beyond planning, dense depth recovered by Dash2Sim improves novel-view synthesis quality by up to 19% on perceptual metrics, suggesting its potential to provide rich conditioning for closed-loop sensor simulation from monocular videos.
Chinese Translation
自动驾驶仿真通常依赖于在少数城市收集的数据或手工编写的合成场景。行车记录仪视频覆盖了更广泛的地点和情况，包括稀有或长尾场景。由于从单目真实环境视频中恢复准确的四维场景较为困难，因此它们被认为不太适合用于仿真。施工区域就是行车记录仪捕捉到的一类长尾情况。我们提出了Dash2Sim，一个将真实环境单目行车记录仪视频转换为与现有仿真器兼容的度量、地理参考四维驾驶日志的框架，并在没有注释的情况下将每个日志与独立维护的地图进行验证。我们将Dash2Sim应用于一个大型视频语料库，创建了ROADWork4D基准数据集，该数据集涵盖了17个城市的4,244个场景和270万个三维对象。在经过验证的子集ROADWork4D-CL（2,201个场景）上，我们研究了特权闭环规划器，发现施工区域场景较为困难：虽然基于规则和混合规划器的泛化能力优于基于学习的规划器，但所有规划器都未能满足临时施工区域通道所需的车道变更要求。除了规划，Dash2Sim恢复的密集深度在感知指标上提高了新视图合成质量，提升幅度可达19%，这表明其在从单目视频中为闭环传感器仿真提供丰富条件方面的潜力。
View on arXiv Download PDF AI Translation
cs.CV / 77 / 2606.07368

Mitosis Detection in the Wild: Multi-Tumor and Context-Aware Generalization in the MIDOG 2025 Challenge

野外的有丝分裂检测：MIDOG 2025挑战中的多肿瘤和情境感知泛化
Aubreville, Marc, Ammeling, Jonas, Banerjee, Sweta, Weiss, Viktoria, Donovan, Taryn A., Klopfleisch, Robert, Lv, Jiaqi, Raza, Shan E Ahmed, Bourgade, Raphaël, Walter, Thomas, Topuz, Yasemin, Varlı, Songül, Collins-Fekete, Charles-Antoine, Shen, Zhuoyan, Kelam, Navya Sri, Singhal, Nitin, Marzahl, Christian, Napora, Brian, Xu, Tengyou, Gu, Hongyan, Vento, Mario, Percannella, Gennaro, Ropiak, Norbert, Wasiak, Izabela, Xiao, Jie, Liu, Shaojun, Choe, Seungho, Khademi, April, Walia, Vidushi, Kotte, Sujatha, Broad, Andrew, Wright, Alex, Balezo, Guillaume, Nasir, Esha Sadia, Jahanifar, Mostafa, Yamagishi, Yosuke, Hanaoka, Shouhei, Sarno, Mattia, Tortorella, Francesco, Meng, Biwen, Liu, Jingxin, Krauss, Sara, Hieber, Daniel, Ramchandani, Lavish, Das, Dev Kumar, Ochi, Mieko, Bae, Yuan, Giedziun, Piotr, Maniewski, Mateusz, Saipradeep, Vangala Govindakrishnan, Sivadasan, Naveen, Benito-Del-Valle, Leire, Galdran, Adrian, Atey, Kaustubh, Jha, Sameer Anand, Dukre, Adinath, Razzak, Imran, Lafarge, Maxime W., Koelzer, Viktor H., Porsche, Nils, Stathonikos, Nikolas, Veta, Mitko, Hirling, Dominik, Iván, Zsanett Zsófia, Horvath, Peter, Breininger, Katharina, Bertram, Christof A.
Abstract
Automated mitosis detection is a well-established task in computational pathology. While previous benchmarks focused on scanner-induced domain shift, clinical "real-world" application requires models to be robust across the vast variance to be expected in the histological landscape. The MItosis DOmain Generalization (MIDOG) 2025 challenge was designed to evaluate algorithmic performance across unprecedented biological and contextual diversity. We curated a test dataset of 365 cases, encompassing 12 distinct human, canine and feline tumor types, digitized across multiple scanning platforms. Moving beyond hand-selected hotspots, the challenge required detection also in random tissue areas (representative of the whole slide detection situation) and challenging areas (areas rich in hard negatives). In the second track, we introduced the classification of atypical mitotic figures (AMFs). There were 18 teams submitting to the detection track, with F1 scores ranging up to 0.740. In the AMF detection track, we had 21 submissions with balanced accuracy values up to 0.908. Our analysis reveals that while most models perform reliably in traditional hotspots, significant performance degradation occurs in challenging ROIs, where false positive rates tripled. Furthermore, performance varied significantly across the 12 tumor types, highlighting "blind spots" in current state-of-the-art architectures when encountering rare or highly pleomorphic malignancies. Moreover, we evaluated the effectiveness of ensembling and found a mean increases of 1.5 and 1.3 percentage points in F1 score and balanced accuracy, respectively. In contrast, TTA showed no relevant improvement. MIDOG 2025 demonstrates that "in the wild" mitosis detection remains a significant hurdle. The transition from hotspot-only evaluation to a multi-contextual framework provides a more realistic proxy for clinical reliability.
Chinese Translation
自动化有丝分裂检测是计算病理学中一个成熟的任务。尽管以往的基准测试集中于扫描仪引起的领域转移，但临床“真实世界”应用要求模型在组织学景观中预期的广泛变异中保持稳健。MItosis DOmain Generalization (MIDOG) 2025挑战旨在评估算法在前所未有的生物和情境多样性下的表现。我们策划了一个包含365个病例的测试数据集，涵盖12种不同的人类、犬类和猫类肿瘤类型，且在多个扫描平台上进行了数字化。挑战不仅要求在手动选择的热点区域进行检测，还要求在随机组织区域（代表整个切片检测情况）和具有挑战性的区域（富含难以检测的负样本的区域）进行检测。在第二个赛道中，我们引入了对非典型有丝分裂细胞（AMFs）的分类。共有18个团队提交了检测赛道的作品，F1分数最高达到0.740。在AMF检测赛道中，我们收到了21份提交，平衡准确率最高达到0.908。我们的分析显示，尽管大多数模型在传统热点区域表现可靠，但在具有挑战性的感兴趣区域（ROIs）中，性能显著下降，假阳性率增加了三倍。此外，不同肿瘤类型的性能差异显著，突显了当前最先进架构在遇到稀有或高度多形性恶性肿瘤时的“盲点”。此外，我们评估了集成方法的有效性，发现F1分数和平衡准确率分别平均提高了1.5和1.3个百分点。相比之下，TTA未显示出相关的改善。MIDOG 2025表明，“野外”有丝分裂检测仍然是一个重大挑战。从仅评估热点区域到多情境框架的过渡，为临床可靠性提供了更现实的代理。
View on arXiv Download PDF AI Translation
cs.CV / 78 / 2606.07394

Mind the Gap: Disentangling Performance Bottlenecks in Video Instance Segmentation

关注差距：解开视频实例分割中的性能瓶颈
Hamdi, Danial, Ayar, Fardin, Javanmardi, Mahdi
Abstract
In Video Instance Segmentation (VIS), classification, segmentation, and tracking objectives are jointly evaluated, but their individual contributions to performance loss remain opaque. We introduce a diagnostic framework that formulates identity and class assignment as an Integer Linear Program (ILP), yielding a model-agnostic oracle that hierarchically isolates each error source. Applied to seven VIS methods spanning online and offline paradigms across YouTube-VIS 2019/2021 and a diagnostic subset of OVIS, our analysis reveals a consistent picture. Tracking instability is a critical bottleneck for online methods, with gaps exceeding 20 AP under heavy occlusion, and grows sharply with video length and instance density. While semantic classification contributes meaningfully on standard benchmarks, its impact becomes negligible where tracking fails most. Although stronger backbones substantially lift default scores, they leave AP tracking gaps largely intact, confirming that temporal fragility is algorithmic rather than purely representational. To complement the oracle, we introduce TrackLens, a visual tool that translates gap magnitude into observable, query-level failure modes. Together, these tools provide a systematic foundation for targeting VIS's core challenge: robust long-term temporal association.
Chinese Translation
在视频实例分割（Video Instance Segmentation, VIS）中，分类、分割和跟踪目标是共同评估的，但它们对性能损失的个体贡献仍然不明确。我们提出了一种诊断框架，将身份和类别分配公式化为整数线性规划（Integer Linear Program, ILP），从而产生一个模型无关的预言机，分层隔离每个错误来源。应用于跨越 YouTube-VIS 2019/2021 的七种 VIS 方法以及 OVIS 的诊断子集，我们的分析揭示了一致的情况。跟踪不稳定性是在线方法的一个关键瓶颈，在严重遮挡下，差距超过 20 AP，并且随着视频长度和实例密度的增加而急剧增长。虽然语义分类在标准基准上有显著贡献，但在跟踪失败最严重的情况下，其影响变得微不足道。尽管更强的主干网络显著提高了默认得分，但它们在 AP 跟踪差距上基本保持不变，确认了时间脆弱性是算法性而非纯粹的表征性。为了补充预言机，我们引入了 TrackLens，这是一种视觉工具，将差距大小转化为可观察的查询级失败模式。这些工具共同为解决 VIS 的核心挑战——稳健的长期时间关联提供了系统基础。
View on arXiv Download PDF AI Translation
cs.CV / 79 / 2606.07401

RealDocBench: A Benchmark for Field-Level QA and Layout Understanding on Real-World Regulated Documents

RealDocBench：针对真实世界受监管文档的字段级问答和布局理解基准
Joshi, Ameya, Kim, Joon, Eggert, Gus, Bajor, Joseph, Hao, Cindy, Reyhan, Jing, Byatnal, Kushal, Badgio, Eli
Abstract
Document parsing systems are increasingly deployed in high-stakes, regulated workflows such as mortgage underwriting, financial reporting, supply-chain logistics, and clinical records. Yet most public benchmarks evaluate parsers on clean academic layouts or synthetic prose, and report a single OCR or markdown-level similarity score. Such documents and metrics correlate poorly with what downstream agents actually need: the correct value for a specific field on a messy real-world page. We introduce RealDocBench, a two-track benchmark built from real regulated documents. The QA track contains 1,356 field-level questions over 581 documents spanning four domains, where each question is paired with a typed gold_dict of key-to-value answers and parsers are scored on both per-field and strict per-question accuracy. The layout track contains 1,500 human-verified page images annotated with COCO-style bounding boxes under a nine-class public taxonomy, scored with a Hungarian matcher that includes adjacency-aware split/merge recovery. We evaluate eighteen systems, spanning commercial parsing APIs, general-purpose VLMs, and open-source OCR models, under a uniform extraction-and-scoring protocol, and report accuracy alongside per-page cost and cache-busted latency. RealDocBench exposes a wide performance spread that single-number benchmarks hide, a persistently hard medical sub-domain, and sharp cost/latency trade-offs across operating points. We release the datasets, parser adapters, and evaluation harness to support reproducible, field-level comparison of document parsing systems.
Chinese Translation
文档解析系统越来越多地应用于高风险的受监管工作流程中，例如抵押贷款承保、财务报告、供应链物流和临床记录。然而，大多数公共基准仅在干净的学术布局或合成文本上评估解析器，并报告单一的OCR或Markdown级相似度分数。这些文档和指标与下游代理实际需要的内容相关性较差：即在杂乱的真实页面上获取特定字段的正确值。我们引入了RealDocBench，这是一个基于真实受监管文档构建的双轨基准。问答轨道包含1,356个字段级问题，涵盖581个文档，涉及四个领域，每个问题都配有一个类型化的gold_dict（键值对答案），解析器的评分基于每个字段和严格的每个问题的准确性。布局轨道包含1,500个经过人工验证的页面图像，这些图像根据九类公共分类法用COCO风格的边界框进行标注，并使用包含邻接感知的匈牙利匹配器进行评分，以实现拆分/合并恢复。我们在统一的提取和评分协议下评估了十八个系统，涵盖商业解析API、通用VLM和开源OCR模型，并报告准确性以及每页成本和缓存失效延迟。RealDocBench揭示了单一数字基准隐藏的广泛性能差异、持续困难的医学子领域，以及在不同操作点之间的显著成本/延迟权衡。我们发布了数据集、解析器适配器和评估工具，以支持可重复的字段级文档解析系统比较。
View on arXiv Download PDF AI Translation
cs.CV / 80 / 2606.07419

DisPOSE: Projected Polystochastic Diffusion for Self-Supervised Multi-View 3D Human Pose Estimation

DisPOSE：用于自监督多视角三维人体姿态估计的投影多随机扩散
Wang, Tony Danjun, Birdal, Tolga, Navab, Nassir
Abstract
Recovering 3D human poses for multiple individuals from different camera views is a fundamental bottleneck for analyzing interacting behaviors. Existing self-supervised approaches leverage synthetic catalogues of 3D poses; however, this leads to poor generalization in real-world scenarios due to distribution shifts. We therefore introduce DisPOSE, a self-supervised framework that approximates the inherently discrete multi-view person-assignment problem as a generative diffusion process over the space of polystochastic tensors. By employing differentiable Sinkhorn projections during denoising, our model learns to guide solutions toward valid and feasible assignments based on 2D image priors. The complete 3D skeletons of localized individuals are then regressed using a Hypergraph-Convolutional Decoder that explicitly models relational structures and articulated joints across multiple views. The proposed approach outperforms current state-of-the-art self-supervised methods on standard datasets and demonstrates strong performance on a newly proposed benchmark featuring highly occluded scenes from surgical operating rooms. Our diffusion-based localization demonstrates high label efficiency, retaining 99% of its performance with only 10% of the pseudo-labels. Notably, disentangling the assignment and root regression components while maintaining differentiability makes DisPOSE nearly agnostic to different camera arrangements.
Chinese Translation
从不同摄像机视角恢复多个个体的三维姿态是分析交互行为的一个基本瓶颈。现有的自监督方法利用合成的三维姿态目录；然而，由于分布转移，这导致在现实场景中的泛化能力较差。因此，我们提出了DisPOSE，一个自监督框架，它将固有离散的多视角人物分配问题近似为在多随机张量空间上的生成扩散过程。通过在去噪过程中采用可微的Sinkhorn投影，我们的模型学习根据二维图像先验引导解决方案朝向有效且可行的分配。然后，使用超图卷积解码器（Hypergraph-Convolutional Decoder）回归局部个体的完整三维骨架，该解码器明确建模多个视角下的关系结构和关节运动。所提出的方法在标准数据集上超越了当前最先进的自监督方法，并在一个新提出的基准上表现出强大的性能，该基准包含来自外科手术室的高度遮挡场景。我们的基于扩散的定位展示了高标签效率，仅使用10%的伪标签便保留了99%的性能。值得注意的是，在保持可微性的同时，解耦分配和根回归组件使DisPOSE几乎对不同的摄像机排列不敏感。
View on arXiv Download PDF AI Translation
cs.CV / 81 / 2606.07431

OpenGlass: Open-Source Smart Glasses for On-Device Event-Based Gesture Recognition

OpenGlass：用于设备端事件驱动手势识别的开源智能眼镜
Bonazzi, Pietro, Moosmann, Julian, Celik, Ahmet, Mayer, Philipp, Magno, Michele
Abstract
Smart eyewear enables unobtrusive, context-aware interaction through multimodal sensors and on-device intelligence, but is severely limited by power, memory, and compute constraints in a compact form factor. Open-hardware platforms supporting event-based vision and embedded ML at this scale are rare. This work introduces an open-source smart glasses platform for rapid prototyping of novel sensors and algorithms. Its modular design uses a flexible FPC interposer to support both event-based and frame-based cameras without full PCB redesign. A hardware-software co-designed power management system combines a configurable PMIC with event-driven wake-up via an nRF5340 coordinator, keeping the GAP9 RISC-V SoC powered down between inferences. The prototype achieves up to 11.8 hours of continuous on-device ML from a 200 mAh battery. As a demonstration, an egocentric hand gesture recognition pipeline was evaluated on the LynX dataset using polarity-separated event histograms from a Prophesee GENX320 camera. R(2+1)D achieved the best cross-subject accuracy of 83.94\% (macro F1 = 0.781) under leave-two-subjects-out validation, with 33.9 ms end-to-end latency on the GAP9. Temporal augmentation and removal of ambiguous classes provided the largest gains (+8.9 pp). All hardware designs, firmware, and models are released open source.
Chinese Translation
智能眼镜通过多模态传感器和设备端智能实现了不引人注目的上下文感知交互，但在紧凑的形态下受到功耗、内存和计算能力的严重限制。支持此规模事件驱动视觉和嵌入式机器学习的开源硬件平台非常稀缺。本研究介绍了一种开源智能眼镜平台，用于快速原型开发新型传感器和算法。其模块化设计采用灵活的FPC中介板，支持事件驱动和帧驱动相机，而无需进行全面的PCB重新设计。硬件-软件协同设计的电源管理系统结合了可配置的PMIC和通过nRF5340协调器的事件驱动唤醒，使GAP9 RISC-V SoC在推理之间保持断电状态。该原型在200 mAh电池下实现了长达11.8小时的连续设备端机器学习。作为演示，基于Prophesee GENX320相机的极性分离事件直方图在LynX数据集上评估了一个自我中心的手势识别管道。在留出两个受试者的验证下，R(2+1)D模型达到了最佳的跨受试者准确率83.94%（宏F1 = 0.781），在GAP9上的端到端延迟为33.9毫秒。时间增强和模糊类别的移除提供了最大的收益（+8.9个百分点）。所有硬件设计、固件和模型均已开源发布。
View on arXiv Download PDF AI Translation
cs.CV / 82 / 2606.07433

Watch, Remember, Reason: Human-View Video Understanding with MLLMs

观看、记忆、推理：基于人类视角的视频理解与多模态大语言模型（MLLMs）
Meng, Jiahao, Tan, Yue, Xu, Qi, Gao, Kuan, Liu, Weisong, Li, Yanwei, Li, Jason, Kong, Lingdong, Wang, Haochen, Zhou, Qianyu, Zhang, Jiangning, Cheng, Guangliang, Tong, Yunhai, Qi, Lu, Yang, Minghsuan
Abstract
Video understanding is being rapidly transformed by multimodal large language models (MLLMs), as research moves from short clips to long, multimodal, and knowledge-intensive video scenarios. These scenarios require models to handle sparse evidence, long-range dependencies, multimodal alignment, and reliable inference under limited computational budgets. This work presents a human-view perspective on LLM-based video understanding, organized around three functional abilities: watching, remembering, and reasoning. Rather than treating video tasks as isolated benchmarks, this view provides a unified structure for analyzing how video MLLMs acquire evidence, preserve context, and produce grounded outputs. We introduce a formulation that characterizes video understanding systems by their perceptual representations, memory states, reasoning traces, and final predictions. Based on this formulation, we identify challenges in spatio-temporal perception, efficient long-video processing, memory modeling, streaming understanding, and faithful reasoning. Representative methods are organized by their roles in video MLLM systems. Watching covers fine-grained, comprehensive, audio-visual, and efficient perception. Remembering includes offline and streaming memory, while reasoning covers text-only reasoning and thinking with videos. We further examine application domains such as egocentric, sports, instructional, medical, and narrative videos, and cover training datasets and evaluation benchmarks across task types, supervision formats, modalities, and capability dimensions. Finally, we outline open problems and future directions for scalable, memory-aware, and evidence-grounded video intelligence. Related works will be continuously traced at https://github.com/marinero4972/Awesome-HumanView-VideoUnderstanding.
Chinese Translation
视频理解正迅速受到多模态大语言模型（MLLMs）的变革，研究从短视频片段转向长时间、多模态和知识密集型的视频场景。这些场景要求模型处理稀疏证据、长距离依赖、多模态对齐，以及在有限计算预算下进行可靠推理。本研究从人类视角出发，探讨基于LLM的视频理解，围绕观看、记忆和推理三种功能能力进行组织。与其将视频任务视为孤立的基准测试，这一视角提供了一个统一的框架，用于分析视频MLLM如何获取证据、保持上下文并生成有根据的输出。我们引入了一种表述，描述视频理解系统的感知表示、记忆状态、推理轨迹和最终预测。基于这一表述，我们识别了在时空感知、长视频高效处理、记忆建模、流媒体理解和忠实推理方面的挑战。代表性方法根据其在视频MLLM系统中的角色进行组织。观看涵盖细粒度、全面的音视频高效感知。记忆包括离线和流媒体记忆，而推理则涵盖仅基于文本的推理和视频思维。我们进一步考察了自我中心、体育、教学、医疗和叙事视频等应用领域，并涵盖了跨任务类型、监督格式、模态和能力维度的训练数据集和评估基准。最后，我们概述了可扩展、关注记忆和基于证据的视频智能的开放问题和未来方向。相关工作将持续追踪，网址为 https://github.com/marinero4972/Awesome-HumanView-VideoUnderstanding。
View on arXiv Download PDF AI Translation
cs.CV / 83 / 2606.07435

The Lipreading Gap: Do VSR Models Perceive Visual Speech Like Human Lipreaders?

唇读差距：视觉语音识别模型是否像人类唇读者一样感知视觉语音？
Jain, Rishabh, Harte, Naomi
Abstract
Visual speech recognition (VSR) models now surpass human lipreaders on benchmarks, but do such gains establish human-like visual speech perception? To explore this, we compare three VSR systems with human baselines on the MaFI word-level lipreading dataset using word, character, phoneme, and viseme-level metrics. Although models achieve higher overall accuracy, they succeed and fail on different words than humans. A text-only n-gram baseline given only a few initial phonemes rivals human lipreading. VSR word-level errors are consistently better explained by training word frequency than by the visual informativeness of words. Viseme accuracies, confusion matrices and human-model correlations further show that models gain most on visemes humans find hardest, and show much weaker dependence on visual clarity. Our work demonstrates that VSR systems rely primarily on language cues from training data rather than visual perception, failing to bind visual features into meaningful words.
Chinese Translation
视觉语音识别（VSR）模型在基准测试中现已超越人类唇读者，但这种进步是否建立了类人视觉语音感知？为探讨这一问题，我们在MaFI单词级唇读数据集上比较了三种VSR系统与人类基线，使用了单词、字符、音素和视觉音素（viseme）级别的指标。尽管模型整体准确率更高，但它们在成功和失败的单词上与人类存在差异。仅给出少量初始音素的文本仅n-gram基线与人类唇读相当。VSR单词级错误的解释更依赖于训练单词频率，而非单词的视觉信息量。视觉音素的准确率、混淆矩阵和人类与模型的相关性进一步表明，模型在视觉音素上获得的进步主要集中在那些人类认为最困难的视觉音素上，并且对视觉清晰度的依赖明显较弱。我们的研究表明，VSR系统主要依赖于训练数据中的语言线索，而非视觉感知，未能将视觉特征结合成有意义的单词。
View on arXiv Download PDF AI Translation
cs.CV / 84 / 2606.07436

Skill-3D: Evolving Scene-Aware Skills for Agentic 3D Spatial Reasoning

Skill-3D：为代理3D空间推理演化场景感知技能
Li, Haoyuan, Hu, Zhengdong, Wang, Jun, Fan, Hehe, Yang, Yi
Abstract
This paper explores agentic 3D spatial understanding, i.e., MLLM agents performing 3D reasoning through tool use. Existing methods often misuse tools and exhibit biased tool preferences under 3D scenarios, leaving the agentic paradigm with only marginal gains over non-agentic strategies. We reveal that 3D spatial reasoning tasks are heterogeneous across scenes, while these agents apply a uniform tool-use strategy to all scenes rather than selecting tools according to the specific scene and task. To address this, we propose Skill-3D, a framework that learns self-evolving scene-aware skills. Specifically, Skill-3D identifies the task scene and records the agent's tool-use trajectory into a Scene Memory, where successful trajectories from similar scenes are aggregated and distilled into a reusable scene-aware skill, with failed ones attached to the skill as lessons. During training, once a similar scene recurs, the corresponding skill is injected to guide the agent, producing new trajectories whose successes and failures further refine the skill, forming a loop in which the memory and the skill library co-evolve. Experiments show that Skill-3D substantially improves tool utilization in 3D spatial reasoning (from 39% to 78% on VSI-Bench), driving the agent toward correct and sufficient tool use. For instance, it improves Gemini-3-Flash by 67% on MMSI-Bench. Furthermore, we conduct agentic post-training over skill-guided trajectories, which boosts Qwen3-VL-8B by 43% on VSI-Bench.
Chinese Translation
本文探讨了代理3D空间理解，即MLLM代理通过工具使用进行3D推理。现有方法在3D场景中往往错误使用工具，并表现出偏向特定工具的偏见，使得代理范式相较于非代理策略仅有微小的提升。我们揭示了3D空间推理任务在不同场景中是异质的，而这些代理对所有场景应用统一的工具使用策略，而不是根据特定场景和任务选择工具。为了解决这一问题，我们提出了Skill-3D，一个学习自我演化场景感知技能的框架。具体而言，Skill-3D识别任务场景并将代理的工具使用轨迹记录到场景记忆中，其中来自相似场景的成功轨迹被聚合并提炼成可重用的场景感知技能，而失败的轨迹则作为经验教训附加到技能上。在训练过程中，一旦相似场景再次出现，相应的技能将被注入以指导代理，产生新的轨迹，其成功与失败进一步完善该技能，形成记忆与技能库共同演化的循环。实验表明，Skill-3D显著提高了3D空间推理中的工具利用率（在VSI-Bench上从39%提升至78%），引导代理朝向正确且充分的工具使用。例如，它在MMSI-Bench上将Gemini-3-Flash的表现提升了67%。此外，我们在技能引导的轨迹上进行代理后训练，使Qwen3-VL-8B在VSI-Bench上提升了43%。
View on arXiv Download PDF AI Translation
cs.CV / 85 / 2606.07451

TEVI: Text-Conditioned Editing of Visual Representations via Sparse Autoencoders for Improved Vision-Language Alignment

TEVI：通过稀疏自编码器进行文本条件的视觉表示编辑，以改善视觉-语言对齐
Mahajan, Sweta, Rao, Sukrut, Xie, Jiahao, Koller, Alexander, Schiele, Bernt
Abstract
Vision-language models such as CLIP are highly useful for diverse tasks due to their shared image-text embedding space. Despite this, the image and text embeddings are often poorly aligned, affecting downstream performance. Recent work has shown that this can be attributed to an information imbalance: images contain more information than their captions describe. In this work, we propose TEVI, a framework that uses captions as a signal for what to retain from image embeddings. Specifically, we use sparse autoencoders to disentangle image embeddings and train a masking module to selectively reconstruct the embedding based on a given caption. In a controlled setup with synthetic captions, we show that TEVI is effective at preserving caption-described attributes while discarding others. By applying TEVI to CLIP models trained on natural images, we further achieve improved retrieval performance across coarse-grained short-caption (MS COCO, Flickr) and fine-grained long-caption (IIW, DOCCI) benchmarks, with stronger gains on richer captions, and improved robustness on the RoCOCO benchmark.
Chinese Translation
视觉-语言模型如CLIP因其共享的图像-文本嵌入空间而在多种任务中具有很高的实用性。尽管如此，图像和文本嵌入往往对齐不佳，影响下游性能。近期研究表明，这可以归因于信息不平衡：图像包含的信息量往往超过其描述的字幕。在本研究中，我们提出了TEVI，一个框架，利用字幕作为信号来决定从图像嵌入中保留哪些信息。具体而言，我们使用稀疏自编码器来解耦图像嵌入，并训练一个掩码模块，根据给定的字幕选择性地重构嵌入。在一个使用合成字幕的受控设置中，我们展示了TEVI在保留字幕描述的属性的同时丢弃其他属性方面的有效性。通过将TEVI应用于在自然图像上训练的CLIP模型，我们进一步在粗粒度短字幕（MS COCO, Flickr）和细粒度长字幕（IIW, DOCCI）基准测试中实现了更好的检索性能，对丰富字幕的提升更为显著，并在RoCOCO基准测试中表现出更强的鲁棒性。
View on arXiv Download PDF AI Translation
cs.CV / 86 / 2606.07498

Implicit Data Synthesis for Contrastive Unsupervised Data Augmentation

用于对比无监督数据增强的隐式数据合成
Kage, Patrick, Hedges, Trevor, Siddharth, N., Andreadis, Pavlos
Abstract
Scientific observations generate large quantities of unlabeled data which is laborious to hand-label, making unsupervised learning techniques valuable for processing datasets. Among these approaches, contrastive learning provides a convenient mechanism for extracting structural representations from unannotated datasets. For natural imagery, the general approach is to use a variety of data-space augmentation methods in order to generate synthetic samples; however, for scientific observations data-space perturbations can fundamentally alter the underlying data. Our proposed method is to generate contrastive samples by perturbing the network weights rather than the underlying data, thus more closely preserving the structure of the data. We demonstrate this technique using a SimCLR-based pipeline applied over radar observations of meteors, and show performance gains under matched protocols.
Chinese Translation
科学观察生成大量未标记的数据，这些数据的手动标注工作繁重，因此无监督学习技术在处理数据集时显得尤为重要。在这些方法中，对比学习提供了一种方便的机制，用于从未标注的数据集中提取结构化表示。对于自然图像，通常的方法是使用多种数据空间增强方法来生成合成样本；然而，对于科学观察，数据空间的扰动可能会根本改变基础数据。我们提出的方法是通过扰动网络权重而非基础数据来生成对比样本，从而更好地保留数据的结构。我们使用基于SimCLR的流程在流星的雷达观测数据上演示了这一技术，并在匹配协议下展示了性能提升。
View on arXiv Download PDF AI Translation
cs.CV / 87 / 2606.07503

Differences in Detection: Explainability Where it Matters

检测差异：在重要性上进行可解释性分析
Theodoridis, Johannes, Maucher, Johannes, Schilling, Andreas
Abstract
We propose Differences in Detection (DnD), an intuitive method to compare two object detection models. Based on the same matching algorithm, it complements the standard metrics of mean Average Precision ($mAP$) and TIDE error analysis with the ability to compare two models directly. More specifically, we calculate the intersection of ground truth labels that are recognized by both models, followed by the corresponding difference sets and the complement set of ground truth labels that are missed by both models. The resulting comparison is more direct and intuitive than a comparison of independent summary statistics. It reveals individual and shared mistakes and becomes particularly interesting when combined with error types. In this case, the differences in detection errors can be analyzed naturally in a standard confusion matrix. While valuable in itself, we believe that one of the best applications of DnD is to guide explainability methods such as ODAM towards metric-relevant examples, grounded in structured subsets. The code for our method is available here: https://github.com/JohannesTheo/differences-in-detection
Chinese Translation
我们提出了检测差异（Differences in Detection, DnD），这是一种直观的方法，用于比较两个目标检测模型。基于相同的匹配算法，它补充了标准的平均精度均值（mean Average Precision, $mAP$）和TIDE错误分析的能力，使得可以直接比较两个模型。更具体地说，我们计算两个模型都识别的真实标签的交集，随后得到相应的差异集和两个模型都遗漏的真实标签的补集。与独立的汇总统计比较相比，得到的比较更加直接和直观。它揭示了个体和共享的错误，并在与错误类型结合时变得特别有趣。在这种情况下，检测错误的差异可以在标准混淆矩阵中自然分析。虽然DnD本身具有价值，但我们认为其最佳应用之一是指导可解释性方法（如ODAM）朝向与度量相关的示例，这些示例基于结构化子集。我们的方法的代码可在此获取：https://github.com/JohannesTheo/differences-in-detection
View on arXiv Download PDF AI Translation
cs.CV / 88 / 2606.07508

Streaming Video Generation with Streaming Force Control

基于流媒体控制的流媒体视频生成
Wang, Hanhui, Xie, Yiming, Feng, Haiwen, Lv, Zhaoyang, Wang, Shenlong, Jiang, Huaizu
Abstract
We introduce StreamForce, a streaming video generation framework that enables physically grounded control through continuous force inputs. Unlike prior video models that train separate models for different force types, assume fixed forces, or rely on non-causal processing, StreamForce is a causal and unified model that responds instantly and coherently to both local and global, time-varying forces. To achieve this, we design a unified force representation as a control signal and develop a distillation pipeline for force-controllable video generation. Our model combines autoregressive efficiency with force responsiveness, sustaining stable photometric and dynamic realism. StreamForce runs at up to 16.6 FPS on a single GPU, achieving state-of-the-art performance in both force adherence and motion realism. Project website: https://neu-vi.github.io/StreamForce/
Chinese Translation
我们介绍了StreamForce，这是一种流媒体视频生成框架，通过连续的力输入实现物理基础的控制。与之前为不同力类型训练单独模型、假设固定力或依赖非因果处理的视频模型不同，StreamForce是一个因果且统一的模型，能够即时且连贯地响应局部和全局的时间变化力。为此，我们设计了一个统一的力表示作为控制信号，并开发了一个用于力可控视频生成的蒸馏管道。我们的模型结合了自回归效率和力响应性，保持了稳定的光度和动态真实感。StreamForce在单个GPU上运行速度高达16.6 FPS，在力遵循性和运动真实感方面实现了最先进的性能。项目网站：https://neu-vi.github.io/StreamForce/
View on arXiv Download PDF AI Translation
cs.CV / 89 / 2606.07512

MemDreamer: Decoupling Perception and Reasoning for Long Video Understanding via Hierarchical Graph Memory and Agentic Retrieval Mechanism

MemDreamer：通过分层图记忆和自主检索机制解耦感知与推理以实现长视频理解
Chen, Cong, Gan, Guo, Ji, Kaixiang, Zhang, ChaoYang, Yang, Zhen, Yao, Guangming, Chen, Hao, Chen, Jingdong, Yuan, Yi, Shen, Chunhua
Abstract
Current Vision-Language Models struggle with hours-long videos because processing full-length visual sequences induces prohibitive token explosion and attention dilution. To overcome this, we introduce MemDreamer to decouple perception and reasoning, shifting long-video understanding into an agentic exploration process. As a plug-and-play framework, it incrementally streams videos to construct a Hierarchical Graph Memory, a top-down three-tier architecture for semantic abstraction, anchored by a foundational graph capturing spatiotemporal and causal relations. During inference, the reasoning model employs agentic tool-augmented retrieval, navigating hierarchies, searching nodes, and traversing logical edges via an Observation-Reason-Action loop. Experiments show MemDreamer achieves SOTA results across four mainstream benchmarks, narrowing the gap with human experts to only 3.7 points. It constrains the reasoning context window to merely 2% of full-context ingestion while delivering a 12.5 point absolute accuracy gain. Furthermore, statistical analysis uncovers a strong positive linear correlation between an VLM's performance on logic reasoning and long-video understanding benchmarks, establishing agentic capability scaling as a new paradigm for multimodal comprehension.
Chinese Translation
当前的视觉-语言模型在处理数小时长的视频时面临困难，因为处理完整的视觉序列会导致不可承受的令牌爆炸和注意力稀释。为了解决这个问题，我们提出了MemDreamer，旨在解耦感知与推理，将长视频理解转变为一种自主探索过程。作为一个即插即用的框架，它逐步流式传输视频，以构建分层图记忆（Hierarchical Graph Memory），这是一个自上而下的三层架构，用于语义抽象，基于一个基础图来捕捉时空和因果关系。在推理过程中，推理模型采用自主工具增强检索，导航层级，搜索节点，并通过观察-推理-行动（Observation-Reason-Action）循环遍历逻辑边缘。实验表明，MemDreamer在四个主流基准上达到了最先进的结果，与人类专家的差距仅为3.7分。它将推理上下文窗口限制在仅占完整上下文摄取的2%，同时实现了12.5分的绝对准确性提升。此外，统计分析揭示了视觉-语言模型在逻辑推理和长视频理解基准上的表现之间存在强正线性相关性，确立了自主能力扩展作为多模态理解的新范式。
View on arXiv Download PDF AI Translation
cs.CV / 90 / 2606.07514

UniSHARP: Universal Sharp Monocular View Synthesis

UniSHARP：通用锐利单目视图合成
Song, Meixi, Zhang, Dizhe, Ren, Hao, Zhang, Ruiyang, Du, Bo, Yang, Ming-Hsuan, Qi, Lu
Abstract
In this work, we focus on extending SHARP, the popular photorealistic view synthesis method, for universal monocular rendering across a continuum of camera systems, from conventional perspective cameras to wide-field-of-view, fisheye and omnidirectional panoramic settings. To overcome the pinhole-specific assumptions of SHARP, our key idea is to align various images in a unified omnidirectional latent space. Thus, we propose UniSHARP, which performs implicit alignment in both feature and Gaussian spaces. Specifically, Gaussian primitives are arranged along rays and radial distances in a ray-based universal representation, while 2D semantic and 3D spatial features extracted from UniK3D-inspired encoders are jointly decoded to generate the complete Gaussian cloud. To comprehensively evaluate our method, we construct a benchmark covering diverse imaging systems across various scenes. The benchmark is further stratified by field of view (FoV) to enable fine-grained assessment of the universal monocular rendering task. Extensive experiments on the proposed benchmark demonstrate the effectiveness of UniSHARP, outperforming alternative methods by a large margin. The project page can be found at: https://insta360-research-team.github.io/Unisharp-website/
Chinese Translation
在本研究中，我们专注于扩展SHARP这一流行的真实感视图合成方法，以实现跨越从传统透视相机到广角、鱼眼和全景设置的通用单目渲染。为了克服SHARP的针孔特定假设，我们的关键思想是将各种图像对齐到统一的全向潜在空间。因此，我们提出了UniSHARP，该方法在特征空间和高斯空间中执行隐式对齐。具体而言，高斯原语沿光线和径向距离在基于光线的通用表示中排列，而从受UniK3D启发的编码器中提取的二维语义和三维空间特征则共同解码以生成完整的高斯云。为了全面评估我们的方法，我们构建了一个基准，涵盖了各种场景中的多样成像系统。该基准进一步按视场（FoV）进行分层，以便对通用单目渲染任务进行细致评估。在所提出的基准上进行的广泛实验表明，UniSHARP的有效性，显著优于其他方法。项目页面可访问： https://insta360-research-team.github.io/Unisharp-website/
View on arXiv Download PDF AI Translation
人工智能 (Artificial Intelligence)
34
cs.AI / 1 / 2606.06514

Detecting and Mitigating Bias by Treating Fairness as a Symmetry Operation

通过将公平性视为对称操作来检测和缓解偏见
Singh, Nishit
Abstract
Machine learning systems deployed in high stakes socioeconomic settings routinely display bias. We formalize bias as a symmetry breaking operation: a classifier is fair if its outputs remain invariant under the counterfactual operation of switching a sensitive attribute, with merit features held fixed. We implement loss based regularization as a symmetry restoring mechanism and evaluate the framework on four synthetic datasets with varying levels of noise, correlation, and bias. The framework achieves upwards of 90\% violation reduction, with accuracy costs around 5\%. This framework does not require causal graph knowledge, is computationally lightweight, and generalizes to any sensitive attribute definable as a bit-flip, making it suitable for contexts where local sources of discrimination remain absent from mainstream benchmarks.
Chinese Translation
在高风险社会经济环境中部署的机器学习系统通常表现出偏见。我们将偏见形式化为一种对称破缺操作：如果分类器的输出在反事实操作下保持不变（即切换敏感属性，同时保持优点特征不变），则该分类器是公平的。我们实现了基于损失的正则化作为一种恢复对称的机制，并在四个具有不同噪声、相关性和偏见水平的合成数据集上评估该框架。该框架实现了超过90%的违规减少，准确度损失约为5%。该框架不需要因果图知识，计算轻量，并且可以推广到任何可定义为位翻转的敏感属性，使其适用于主流基准中缺乏地方歧视来源的上下文。
View on arXiv Download PDF AI Translation
cs.AI / 2 / 2606.06518

DiBS: Diffusion-Informed Branch Selection

DiBS：扩散信息驱动的分支选择
Liu, Bo, Xie, Yuan, Gao, Yuan, Luo, Xiaolong, Ye, Peng, Chen, Tao, Han, Fujun
Abstract
Sudoku is a representative constraint satisfaction problem that requires global structural reasoning under strict discrete constraints. The existing works of solving Sudoku mainly focus on two dominant approaches, i.e., traditional heuristic and deep learning solver. However, they suffer from two complementary limitations: learning-based solvers lack hard correctness guarantees, while complete symbolic solvers are still prone to long-tail search. To address these shortcomings, we propose a novel diffusion model-guided approach, termed as DiBS, for the branch selection search process. Specifically, DiBS keeps the symbolic solver complete and uses the diffusion model as a branch-ordering guide. The core method is ranking candidate values under the current partial assignment and lightweight consistency signal. Furthermore, we provide an in-depth theoretical proof to reveal how it works and why it works. Experiments on the challenging Royle 17-clue Sudoku benchmark show that our DiBS substantially reduces search cost relative to strong heuristic baselines, especially in nodes, backtracks, and long-tail percentiles. Besides, these results confirm that learned global guidance is effective on hard instances where branch-order mistakes are most expensive. All codes are available at https://github.com/shanxierdan/DiBS.
Chinese Translation
数独是一种典型的约束满足问题，要求在严格的离散约束下进行全局结构推理。现有的数独求解方法主要集中在两种主流方法上，即传统启发式方法和深度学习求解器。然而，它们存在两个互补的局限性：基于学习的求解器缺乏严格的正确性保证，而完全符号求解器仍然容易遭遇长尾搜索问题。为了解决这些缺点，我们提出了一种新颖的扩散模型引导的方法，称为DiBS，用于分支选择搜索过程。具体而言，DiBS保持符号求解器的完整性，并使用扩散模型作为分支排序的指导。核心方法是在当前部分赋值和轻量级一致性信号下对候选值进行排名。此外，我们提供了深入的理论证明，以揭示其工作原理及有效性。在具有挑战性的Royle 17线索数独基准测试中的实验表明，我们的DiBS相较于强启发式基线显著降低了搜索成本，特别是在节点、回溯和长尾百分位数方面。此外，这些结果确认了学习的全局指导在分支排序错误代价最高的困难实例上是有效的。所有代码可在 https://github.com/shanxierdan/DiBS 获取。
View on arXiv Download PDF AI Translation
cs.AI / 3 / 2606.06519

SafeGene: Reusable Adapters for Transferable Safety Alignment

SafeGene：可重用的适配器用于可转移的安全对齐
Wang, Yanghan, Kou, Zhiqiang, Feng, Fu, Wang, Jing, Geng, Xin
Abstract
Open-weight LLMs are increasingly fine-tuned into customized assistants, but downstream fine-tuning can weaken safety alignment and make models more vulnerable to malicious prompts, even when the training data is not intentionally harmful. This creates a recurring safety recovery problem as target models are repeatedly updated with new task data or user interactions. We propose SafeGene, a reusable safety-adapter module designed for cross-task reuse within each architecture-compatible model family. Rather than treating safety recovery as a model-specific repair step, SafeGene treats safety capability as an independent, reusable adapter representation decoupled from task-specific updates. This representation is obtained from aligned--degraded model discrepancies, refined into task-transferable safety vectors through data-aware layer selection, and expressed in each downstream task-adapted model via few-shot layer-wise coefficient recalibration. Experiments across multiple model families, downstream tasks, and safety judges show that SafeGene-enhanced models reduce harmful response rates while maintaining downstream performance, outperforming representative safe adaptation methods in safety--utility trade-off.
Chinese Translation
开放权重的大型语言模型（LLMs）越来越多地被微调为定制助手，但下游微调可能会削弱安全对齐，使模型在面对恶意提示时更加脆弱，即使训练数据并非故意有害。这导致了一个反复出现的安全恢复问题，因为目标模型在不断更新新的任务数据或用户交互。我们提出了SafeGene，一种可重用的安全适配器模块，旨在为每个架构兼容的模型家族提供跨任务重用。SafeGene并不将安全恢复视为特定模型的修复步骤，而是将安全能力视为一种独立的、可重用的适配器表示，与任务特定的更新解耦。该表示是通过对齐的退化模型差异获得的，经过数据感知的层选择精炼为任务可转移的安全向量，并通过少量样本的层级系数重新校准在每个下游任务适应的模型中表达。针对多个模型家族、下游任务和安全评估者的实验表明，增强了SafeGene的模型在保持下游性能的同时减少了有害响应率，在安全与效用的权衡中优于代表性的安全适应方法。
View on arXiv Download PDF AI Translation
cs.AI / 4 / 2606.06523

Lean4Agent: Formal Modeling and Verification for Agent Workflow and Trajectory

Lean4Agent：代理工作流和轨迹的形式建模与验证
Wang, Ruida, Huang, Jerry, Wang, Pengcheng, Liu, Xuanqing, Kong, Luyang, Zhang, Tong
Abstract
Equipping Large Language Models (LLMs) to execute reliable multi-step workflows has become a central challenge in artificial intelligence. Despite recent advances in LLMs' agentic capabilities, most agent systems still lack formal methods for specifying, verifying, and debugging their workflow and execution trajectories. This challenge mirrors a long-standing problem in mathematics, where the ambiguity of natural languages (NLs) motivates the development of formal languages (FLs). Inspired by this paradigm, we propose **Lean4Agent**, to the best of our knowledge, the first framework that uses Lean4, a dependent-type FL to model and verify agent behavior. **Lean4Agent** launches **FormalAgentLib**, an extensible Lean4 library for formally modeling and verifying agent workflows' semantic consistency under explicit assumptions, and enabling localization of execution-time failures revealed by trajectories. Building on **FormalAgentLib**, we further develop **LeanEvolve**, which applies results in **FormalAgentLib** to revise workflows to enhance its capability. Extensive experiments on a hard problem subset of SWE-Bench-Verified and a subset of ELAIP-Bench across 5 leading LLMs indicate that the verification-passing workflows outperform the failing ones by an average of **11.94%**, and **LeanEvolve** further improves SWE performance by **7.47%** on average. Furthermore, **Lean4Agent** establishes a foundation for a new field of using expressive dependent-type FL to formally model and verify agent behavior.
Chinese Translation
为大型语言模型（LLMs）赋能以执行可靠的多步骤工作流已成为人工智能领域的一项核心挑战。尽管LLMs在代理能力方面取得了近期进展，但大多数代理系统仍然缺乏用于指定、验证和调试其工作流及执行轨迹的形式方法。这一挑战反映了数学中长期存在的问题，其中自然语言（NLs）的模糊性促使了形式语言（FLs）的发展。受到这一范式的启发，我们提出了**Lean4Agent**，据我们所知，这是第一个使用Lean4（依赖类型形式语言）来建模和验证代理行为的框架。**Lean4Agent**推出了**FormalAgentLib**，这是一个可扩展的Lean4库，用于在明确假设下形式化建模和验证代理工作流的语义一致性，并能够定位由轨迹揭示的执行时故障。在**FormalAgentLib**的基础上，我们进一步开发了**LeanEvolve**，该工具应用**FormalAgentLib**中的结果来修订工作流，以增强其能力。在SWE-Bench-Verified的一个困难问题子集和5个领先LLMs的ELAIP-Bench子集上进行的广泛实验表明，验证通过的工作流平均优于失败的工作流**11.94%**，而**LeanEvolve**进一步平均提高了SWE性能**7.47%**。此外，**Lean4Agent**为使用表达性依赖类型形式语言形式化建模和验证代理行为的新领域奠定了基础。
View on arXiv Download PDF AI Translation
cs.AI / 5 / 2606.06526

CrowdMath: A Dataset of Crowdsourced Mathematical Research Discussions

CrowdMath：一个众包数学研究讨论的数据集
Muckatira, Sherin, Geneson, Jesse, Gerovitch, Slava, Etingof, Pavel, Gronas, Mikhail, Rumshisky, Anna
Abstract
Large language models have made substantial progress on mathematical reasoning, but existing benchmarks typically evaluate well-specified problems with final answers, step-by-step solutions, or complete proofs. They do not capture collaborative open-problem solving: a setting in which participants propose partial arguments, identify gaps or errors in prior steps, repair flawed reasoning, and gradually synthesize incremental contributions into a proof. We introduce CrowdMath, a dataset of 164 expert-annotated progress chains from the MIT PRIMES--Art of Problem Solving (AoPS) CrowdMath program (2016-2025), a collaborative research initiative whose discussions have led to peer-reviewed publications. Each chain traces a multi-participant forum discussion from an open-problem statement to a completed proof. Posts are labeled by their functional roles in the evolving solution process, including partial progress, proof completion, erroneous reasoning, and error identification. We define evaluation tasks and benchmark six frontier models. Models achieve 83-88% accuracy on next-post prediction, suggesting that they can follow the local flow of mathematical discussion. However, they struggle to identify the functional significance of individual contributions with the best model achieving only 0.42 macro-F1 on post-role classification. CrowdMath exposes a gap between solving well-specified mathematical problems and understanding collaborative mathematical progress as it unfolds.
Chinese Translation
大型语言模型在数学推理方面取得了显著进展，但现有基准通常评估明确的问题，要求给出最终答案、逐步解决方案或完整证明。它们未能捕捉到协作开放问题解决的过程：在这一过程中，参与者提出部分论证，识别先前步骤中的空白或错误，修正有缺陷的推理，并逐步将增量贡献综合成一个证明。我们介绍了CrowdMath，这是一个包含164个专家注释的进展链的数据集，来源于麻省理工学院PRIMES--问题解决艺术（AoPS）CrowdMath项目（2016-2025），这是一个协作研究倡议，其讨论已导致同行评审的出版物。每个进展链追踪了从开放问题陈述到完成证明的多参与者论坛讨论。帖子根据其在不断演变的解决过程中的功能角色进行标记，包括部分进展、证明完成、错误推理和错误识别。我们定义了评估任务并基准测试了六个前沿模型。模型在下一帖预测上达到了83-88%的准确率，表明它们能够跟随数学讨论的局部流动。然而，它们在识别个别贡献的功能重要性方面表现不佳，最佳模型在帖子角色分类上仅达到0.42的宏观F1值。CrowdMath揭示了在解决明确数学问题与理解协作数学进展之间的差距。
View on arXiv Download PDF AI Translation
cs.AI / 6 / 2606.06529

Attack Selection in Agentic AI Control Evaluations Meaningfully Decreases Safety

代理人工智能控制评估中的攻击选择显著降低安全性
Ge-Wang, Catherine, Crosse, Tyler, Hadad IV, Benjamin, Schaeffer, Joachim, Potham, Ram, Tracy, Tyler
Abstract
An attacker that strategically chooses when to attack is much harder to catch than one that attacks indiscriminately. AI control is a safety framework for deploying capable but untrusted AI agents under the oversight of a weaker, trusted monitor and a limited human audit budget. Control evaluations stress-test these protocols by pitting a red-team attack policy against the blue-team monitor, but current evaluations typically assume attackers that do not strategically select when to attack. We study this capability, attack selection, in agentic settings by decomposing attack decisions into a start policy, which decides when an attacker should attack, and a stop policy, which decides when an attacker should abort an ongoing attack. Across two agentic settings, BashArena and LinuxArena, both policies substantially lower measured empirical safety without changing the underlying attack capability. At a 1% audit budget, our start policy reduces safety by 20pp on both BashArena and LinuxArena, and our stop policy reduces safety by 20pp on BashArena and 28pp on LinuxArena. These reductions should be interpreted as upper bounds on the effect of attack selection. Existing control evaluations may therefore yield overly optimistic safety estimates against selective attackers. We recommend that future evaluations, system cards, and safety cases elicit attack selection to produce more realistic safety estimates.
Chinese Translation
一个战略性选择攻击时机的攻击者比一个无差别攻击的攻击者更难被捕捉。人工智能控制是一种安全框架，用于在一个较弱的、可信的监控者的监督下部署能力强但不可信的人工智能代理，并且有限的人类审计预算。控制评估通过将红队攻击策略与蓝队监控者对抗来对这些协议进行压力测试，但当前的评估通常假设攻击者不会战略性地选择攻击时机。我们通过将攻击决策分解为开始策略（决定攻击者何时应进行攻击）和停止策略（决定攻击者何时应中止正在进行的攻击）来研究这一能力，即攻击选择。在两个代理环境中，BashArena 和 LinuxArena，这两种策略显著降低了测量的经验安全性，而没有改变基础的攻击能力。在 1% 的审计预算下，我们的开始策略在 BashArena 和 LinuxArena 上分别降低了 20 个百分点的安全性，而我们的停止策略在 BashArena 上降低了 20 个百分点，在 LinuxArena 上降低了 28 个百分点。这些降低应被解释为攻击选择影响的上限。因此，现有的控制评估可能会对选择性攻击者产生过于乐观的安全估计。我们建议未来的评估、系统卡和安全案例应引入攻击选择，以产生更现实的安全估计。
View on arXiv Download PDF AI Translation
cs.AI / 7 / 2606.06531

CARVE-Q: Quantum-Proposed, Classically Certified Interactive Driving Repair

CARVE-Q：量子提议、经典认证的交互驾驶修复
Wang, Yifan
Abstract
The critical question after a correct driving veto is not only whether a maneuver is unsafe, but whether the blocked interaction admits a lawful, auditable, and responsibility-bounded repair. Prediction and game-theoretic planners can suggest plausible cooperation, yet they do not return a proof that the repair respects hard rules, right-of-way, cost allocation, and ego fallback. We introduce CARVE, Certified Affordable Repair of Vetoed maneuvers via Envelopes, a certificate architecture for prediction-free interactive repair. Given a vetoed maneuver, CARVE constructs a finite repair lattice and emits a structured certificate recording the binding rule, selected joint repair, right-of-way-scaled cooperation envelope, responsibility-weighted cost split, and ego-only fallback. This certificate view reveals the algorithmic bottleneck: multi-owner repair induces a product lattice $M = \prod_j |\mathcal{A}_j|$. We therefore introduce CARVE-Q, a verifier-shielded quantum-AI search layer that applies quantum minimum finding only to this black-box lattice while leaving all safety authority classical. In the conservative verifier-oracle model, exact classical minimum finding requires $\Theta(M)$ queries in the worst case, whereas Durr-Hoyer/Grover minimum finding uses $O(\sqrt{M})$ oracle queries with high probability. We prove verifier-shielded certificate soundness, priority non-elicitation, black-box query separation, and finite-precision reversible-oracle constructibility. We then demonstrate state-vector minimum finding on CARVE repair oracles up to 65,536 assignments and validate certificate preservation on Lanelet2-grounded INTERACTION replay with 100% right-of-way respect, 100% blame consistency, and zero priority false positives. The result is a trust-bounded quantum-AI pattern for certified autonomy: quantum proposes; CARVE certifies.
Chinese Translation
在正确的驾驶否决之后，关键问题不仅是某个操作是否不安全，还包括被阻止的交互是否允许合法、可审计且责任有限的修复。预测和博弈论规划者可以建议合理的合作，但它们并未提供修复遵循严格规则、优先权、成本分配和自我回退的证明。我们引入了CARVE（通过包络线的否决操作的认证可负担修复），这是一个用于无预测交互修复的证书架构。给定一个被否决的操作，CARVE构建一个有限的修复格，并发出一个结构化证书，记录约束规则、选定的联合修复、按优先权缩放的合作包络、责任加权的成本分配，以及仅限自我的回退。该证书视图揭示了算法瓶颈：多所有者修复导致产品格 $M = ext{prod}_j | ext{A}_j|$。因此，我们引入了CARVE-Q，一个验证器保护的量子人工智能搜索层，仅对这个黑箱格应用量子最小值查找，同时保持所有安全权威为经典。在保守的验证器-Oracle模型中，最坏情况下，精确的经典最小值查找需要 $ heta(M)$ 次查询，而Durr-Hoyer/Grover最小值查找在高概率下使用 $O( ext{sqrt}(M))$ 次Oracle查询。我们证明了验证器保护的证书有效性、优先权非引导性、黑箱查询分离和有限精度可逆Oracle构造性。然后，我们在CARVE修复Oracle上演示了状态向量最小值查找，最多支持65,536个赋值，并验证了在Lanelet2基础的INTERACTION重放中证书的保留，确保100%的优先权尊重、100%的责任一致性和零优先权误报。结果是一个受信任的量子人工智能模式，用于认证自主：量子提议；CARVE认证。
View on arXiv Download PDF AI Translation
cs.AI / 8 / 2606.06533

Position: Don't Just "Fix it in Post": A Science of AI Must Study Training Dynamics

立场：不要仅仅在后期“修复”：人工智能的科学必须研究训练动态
Biderman, Stella, Khan, Mohammad Aflah, Mireshghallah, Niloofar, Arnett, Catherine, Barez, Fazl, Saphra, Naomi
Abstract
What would it mean to have a scientific understanding of AI? Models are not static objects: they are snapshots of time-evolving processes shaped by data, objectives, architectures, and optimization dynamics. Yet much of AI research treats models as fixed artifacts, analyzing behaviors after training rather than asking why they emerge. This position paper argues that a science of AI must move beyond post-hoc fixes and study the training dynamics that produce model behavior. Such a science should support progressively stronger forms of understanding: predicting outcomes from early training signals, intervening when trajectories go wrong, and ultimately designing training procedures that more reliably produce desired properties. Scaling laws have made prediction routine for loss; the challenge is extending this success to capabilities, biases, robustness, and safety-relevant behaviors. We articulate requirements for such theories grounded in the history and philosophy of science, examine progress in mechanistic interpretability, fairness, memorization, and simplicity bias, and identify concrete open problems.
Chinese Translation
拥有对人工智能的科学理解意味着什么？模型并不是静态的对象：它们是由数据、目标、架构和优化动态塑造的时间演变过程的快照。然而，许多人工智能研究将模型视为固定的工件，分析训练后的行为，而不是探讨这些行为为何会出现。本文立场论文主张，人工智能的科学必须超越事后修复，研究产生模型行为的训练动态。这种科学应支持逐步增强的理解形式：从早期训练信号预测结果，在轨迹出现问题时进行干预，最终设计出更可靠地产生期望属性的训练程序。缩放法则使得损失的预测变得常规；挑战在于将这一成功扩展到能力、偏见、鲁棒性和安全相关行为。我们阐明了基于科学历史和哲学的理论要求，考察了在机制可解释性、公平性、记忆化和简单性偏见方面的进展，并识别出具体的开放问题。
View on arXiv Download PDF AI Translation
cs.AI / 9 / 2606.06641

Accelerated Fourier SAT (AFSAT): Fully Realising a GPU-based Symmetric Pseudo-Boolean SAT Solver

加速傅里叶SAT（AFSAT）：全面实现基于GPU的对称伪布尔SAT求解器
Christopher, Cody J, Gretton, Charles
Abstract
We present Accelerated Fourier SAT (AFSAT), a GPU-accelerated solver for pseudo-Boolean satisfiability based on continuous local search (CLS). AFSAT realises the proof-of-concept approach, FastFourierSAT, into a fully-engineered solver supporting any heterogeneous mixture of symmetric constraint types and lengths within a single problem instance. Using the JAX compiler, AFSAT leverages pure function composition, automatic vectorisation, automatic differentiation, and just-in-time (JIT) compilation to perform massively parallel CLS across batches of candidate assignments. We demonstrate substantially improved numerical stability, runtime performance, and memory efficiency over the proof-of-concept. We achieve this by way of identifying and addressing various limitations that arise from memory latency and floating-point representation, as well as leveraging automatic parallelisation and compact representations. The inherent representational and stability limitations of floating point are partially addressed by a tailored discrete Fourier transform implementation. We achieve near-linear throughput when scaling to multiple accelerators via JAX array sharding.
Chinese Translation
我们提出了加速傅里叶SAT（AFSAT），这是一种基于连续局部搜索（CLS）的伪布尔可满足性问题的GPU加速求解器。AFSAT将概念验证方法FastFourierSAT转化为一个全面工程化的求解器，支持在单个问题实例中任意异构混合的对称约束类型和长度。通过使用JAX编译器，AFSAT利用纯函数组合、自动向量化、自动微分和即时编译（JIT）来在候选赋值的批次中执行大规模并行的CLS。我们展示了在数值稳定性、运行时性能和内存效率方面相较于概念验证有了显著改善。我们通过识别和解决由内存延迟和浮点表示引起的各种限制，以及利用自动并行化和紧凑表示来实现这一点。针对浮点数固有的表示和稳定性限制，我们采用了定制的离散傅里叶变换实现。通过JAX数组分片，我们在扩展到多个加速器时实现了近线性的吞吐量。
View on arXiv Download PDF AI Translation
cs.AI / 10 / 2606.06656

A Study of Parallel Continuous Local Search

并行连续局部搜索研究
Christopher, Cody J, Gretton, Charles
Abstract
We study parallel Continuous Local Search (CLS) as a solution approach for Boolean satisfiability problems with symmetric pseudo-Boolean (PB) constraints. Here, the $n$-variable PB-satisfiability problem is relaxed to a continuous optimisation problem with a differentiable objective function on an $n$-dimensional hypercube. For satisfiable instances, the global minimisers of this optimisation problem correspond to satisfying assignments of the SAT problem at hand. We present several novel findings via empirical experiments: (i) redundant constraints can inhibit rather than accelerate convergence; (ii) CLS shows promise as a sub-solver in hybridised settings, quickly completing partial assignments; and (iii) local search rapidly converges to a stable distribution of solution quality (i.e., degree of satisfaction), due to saddle-dense objectives where additional solver steps yield diminishing returns. Our findings inform practical uses of CLS for SAT on modern accelerator hardware.
Chinese Translation
我们研究了并行连续局部搜索（Continuous Local Search, CLS）作为解决具有对称伪布尔（pseudo-Boolean, PB）约束的布尔可满足性问题的一种方法。在这里，$n$-变量的PB可满足性问题被放宽为一个在$n$维超立方体上的具有可微目标函数的连续优化问题。对于可满足的实例，这个优化问题的全局最小值对应于当前SAT问题的可满足赋值。通过实证实验，我们提出了几个新发现：（i）冗余约束可能抑制而不是加速收敛；（ii）CLS在混合设置中作为子求解器展现出良好的前景，能够快速完成部分赋值；（iii）由于鞍点稠密的目标，局部搜索迅速收敛到解决方案质量的稳定分布（即满足度），在这种情况下，额外的求解步骤带来的收益递减。我们的发现为CLS在现代加速器硬件上用于SAT的实际应用提供了指导。
View on arXiv Download PDF AI Translation
cs.AI / 11 / 2606.06660

AEGIS: A Backup Reflex for Physical AI

AEGIS：物理人工智能的备份反应机制
Chen, Josef
Abstract
Long-horizon robot manipulation tends to fail gradually: one bad step degrades the state, and the policy spirals into a basin from which it cannot recover. The failure is often visible before it happens. We introduce AEGIS (Activation-probe Early-warning, Gated Inference Switching), a selective escalation method that uses a lightweight probe on a weak policy's frozen activations to detect high-risk steps while there is still time to act. When the probe flags a step, control switches to a stronger separate policy, but only for the steps that need it. On LIBERO-Spatial, AEGIS recovers 10.1% of the trajectories the weak policy alone loses, versus 4.6% for budget-matched blind escalation and 5.1% for a random-trigger placebo. These gains are significant under one-sided exact paired McNemar tests with Holm-Bonferroni adjustment over three pre-registered contrasts: +5.4pp over blind escalation, p=8.5e-6; +5.0pp over random triggering, p=1.0e-4; paired-trajectory bootstrap CIs exclude zero. AEGIS activates the stronger policy on only 38% of steps, so the lever is timing rather than compute. The probe clears its precondition with an early-window AUROC of 0.764, 95% CI [0.70, 0.84], read from the weak-policy path over the first 30% of trajectory steps before any handoff. We pre-register the full analysis plan, including a conditional recovered-task-rate estimand and explicit kill criteria, and confirm the result on 700 common-random-number episodes per arm, with nA-fail=646.
Chinese Translation
长时间范围的机器人操作往往会逐渐失败：一步错误的操作会降低状态，导致策略陷入无法恢复的低谷。失败通常在发生之前就可以被察觉。我们提出了AEGIS（Activation-probe Early-warning, Gated Inference Switching），这是一种选择性升级方法，利用轻量级探测器对弱策略的冻结激活进行监测，以便在仍有时间采取行动时检测高风险步骤。当探测器标记出一个步骤时，控制权切换到一个更强的独立策略，但仅限于需要的步骤。在LIBERO-Spatial上，AEGIS恢复了弱策略单独损失的10.1%的轨迹，而预算匹配的盲目升级仅为4.6%，随机触发的安慰剂为5.1%。在经过Holm-Bonferroni调整的单侧精确配对McNemar检验下，这些增益具有显著性：相较于盲目升级增加了5.4个百分点，p=8.5e-6；相较于随机触发增加了5.0个百分点，p=1.0e-4；配对轨迹自助法置信区间不包含零。AEGIS仅在38%的步骤上激活更强的策略，因此其杠杆作用在于时机而非计算。探测器在早期窗口的AUROC为0.764，95%置信区间为[0.70, 0.84]，该值是从弱策略路径的前30%轨迹步骤中读取的，且在任何交接之前。我们预注册了完整的分析计划，包括条件恢复任务率的估计量和明确的终止标准，并在每个实验组的700个共同随机数实验中确认了结果，其中nA-fail=646。
View on arXiv Download PDF AI Translation
cs.AI / 12 / 2606.06735

A Geometric Account of Activation Steering through Angle-Norm Decomposition

通过角度-范数分解的激活引导的几何解释
Aparin, Georgii, Gaintseva, Tatiana
Abstract
Linear activation steering has gained popularity as a simple and empirically effective way to control language model behavior. More recently, spherical steering paradigms have been proposed to address limitations of additive interventions, often motivated by the assumption that hidden-state norm does not carry concept-relevant information. In this work, we revisit this assumption through a controlled empirical study designed to disentangle the roles of angular and radial components. We show that steering methods differ mainly in how they couple two geometric effects: changing a token's angular alignment with a concept direction and changing its hidden-state norm. Across seven language models, we find that concepts are represented primarily in angular structure, supporting the motivation for spherical methods, but that norm remains important for the stability and downstream effects of steering. Our results explain why interventions with similar concept-level effects can behave differently, and suggest that activation steering should be parameterized by interpretable angular and radial components of the intervention, rather than by a single additive coefficient that entangles these two effects.
Chinese Translation
线性激活引导因其简单且在经验上有效的方式控制语言模型行为而受到广泛关注。最近，提出了球面引导范式，以解决加性干预的局限性，这通常是基于隐藏状态范数不携带与概念相关信息的假设。在本研究中，我们通过一项控制的实证研究重新审视这一假设，旨在理清角度和径向成分的作用。我们表明，引导方法主要在于它们如何耦合两种几何效应：改变一个标记与概念方向的角度对齐以及改变其隐藏状态范数。在七个语言模型中，我们发现概念主要以角度结构的形式表示，这支持了球面方法的动机，但范数在引导的稳定性和下游效应中仍然重要。我们的结果解释了为何具有相似概念层面效应的干预可以表现出不同的行为，并建议激活引导应通过可解释的干预的角度和径向成分进行参数化，而不是通过一个将这两种效应纠缠在一起的单一加性系数。
View on arXiv Download PDF AI Translation
cs.AI / 13 / 2606.06741

OpenSkill: Open-World Self-Evolution for LLM Agents

OpenSkill：开放世界自我进化的 LLM 代理
Yan, Zhiling, Song, Dingjie, Zhang, Hanrong, Liang, Wei, Zhang, Yuxuan, Dai, Yutong, He, Lifang, Yu, Philip S., Xu, Ran, Li, Xiang, Sun, Lichao
Abstract
Self-evolving agents requires adaptation after deployment, but existing approaches assume a usable learning loop, such as curated skills, successful trajectories, or verifier signals. Real open-world deployments may provide none of these, offering only a task prompt. In this work, we study open-world self-evolution, where an agent must build both its skills and its own verification signals from scratch, using open-world resources but no target-task supervision. We propose OpenSkill, a framework that bootstraps this loop: it acquires grounded knowledge and verification anchors from documentation, repositories, and the web, synthesizes them into transferable skills, and refines those skills against self-built virtual tasks grounded in the anchors rather than in target answers. The open world thus supplies both the knowledge to be learned and a supervision-independent practice environment, with target-task supervision reserved for final evaluation. Across three benchmarks and two target agents, OpenSkill attains the best automated pass rate while satisfying the no-supervision constraint. Analysis shows its skills transfer across models without model-specific adaptation, and its self-built verifier aligns with ground-truth outcomes despite never accessing them.
Chinese Translation
自我进化的代理在部署后需要适应，但现有方法假设存在可用的学习循环，例如策划的技能、成功的轨迹或验证信号。真实的开放世界部署可能没有这些，仅提供任务提示。在本研究中，我们探讨开放世界自我进化，其中代理必须从零开始构建其技能和验证信号，利用开放世界资源但不依赖于目标任务的监督。我们提出了 OpenSkill，一个启动这一循环的框架：它从文档、代码库和网络中获取基础知识和验证锚点，将其合成可转移的技能，并针对基于锚点而非目标答案的自建虚拟任务对这些技能进行精炼。因此，开放世界同时提供了需要学习的知识和一个独立于监督的实践环境，目标任务的监督仅保留用于最终评估。在三个基准和两个目标代理上，OpenSkill 在满足无监督约束的同时达到了最佳的自动通过率。分析表明，其技能在模型间转移而无需特定于模型的适应，并且其自建的验证器与真实结果一致，尽管从未访问过这些结果。
View on arXiv Download PDF AI Translation
cs.AI / 14 / 2606.06787

AdMem: Advanced Memory for Task-solving Agents

AdMem：任务解决代理的高级记忆
Wang, Runzhe, Lu, Huilin, Liu, Shengjie, Dong, Li, Zhu, Jason
Abstract
Large Language Models (LLMs) show promise as tool-using agents but remain limited in long-horizon tasks that require remembering, organizing, and reusing knowledge. Prior memory approaches aim to resolve the situation, but mainly focus on storing factual information. Recent work on procedural memory improves task reuse, yet often reduces to replaying past successes without addressing failure cases or online scalability. We introduce a unified and automatic memory framework that integrates semantic, episodic, and procedural memory in a bi-level design combining short-term and long-term stores. A multi-agent architecture with actor, memory, and critic agents enables automatic memory generation, reward annotation, and adaptive retrieval. Long-term memory is managed through reward-based evaluation, merging, and pruning, ensuring scalability and continual improvement. Experiments across various environments show that our approach improves robustness and success on long multi-turn tasks compared to existing baselines. This work highlights the importance of comprehensive, adaptive memory for advancing LLM-based agents.
Chinese Translation
大型语言模型（LLMs）作为工具使用代理展现出潜力，但在需要记忆、组织和重用知识的长期任务中仍然存在局限性。以往的记忆方法旨在解决这一问题，但主要集中于存储事实信息。近期关于程序性记忆的研究改善了任务重用，然而通常仅限于重放过去的成功，而未能解决失败案例或在线可扩展性的问题。我们提出了一种统一的自动记忆框架，该框架在双层设计中整合了语义记忆、情节记忆和程序性记忆，结合了短期和长期存储。一个多代理架构，包含执行者、记忆和评估代理，能够实现自动记忆生成、奖励注释和自适应检索。长期记忆通过基于奖励的评估、合并和修剪进行管理，确保了可扩展性和持续改进。在各种环境下的实验表明，我们的方法在长期多轮任务中相比现有基线提高了鲁棒性和成功率。本研究强调了全面、自适应记忆在推动基于LLM的代理发展中的重要性。
View on arXiv Download PDF AI Translation
cs.AI / 15 / 2606.06869

Evidence-Based Intelligent Diagnostic and Therapeutic Visualization System with Large Language Models: Multi-Turn Interaction and Multimodal Treatment Plan Generation

基于证据的智能诊断与治疗可视化系统：大型语言模型的多轮交互与多模态治疗方案生成
Wang, Yunhan, Wang, Yuda, Tu, Zhiying, Song, Mingqiang, Song, Li, Li, Kun, Chu, Dianhui, Zhang, Bolin
Abstract
Aim: Existing AI-assisted traditional Chinese medicine diagnostic tools suffer from opaque reasoning processes, passive interaction, and limited treatment plan presentation. This study proposes a knowledge-enhanced visual diagnostic system to improve the transparency and interpretability of syndrome differentiation and treatment. Methods: The system is built upon a Neo4j knowledge graph comprising 241 syndromes, 1,263 symptoms, and 2,485 relations. It incorporates a four-stage symptom matching pipeline (exact, semantic, fuzzy, and large language model verification), an information gain-driven proactive questioning strategy optimized with genetic algorithms, and a multimodal treatment presentation integrating artificial intelligence-generated illustrations, three-dimensional meridian-acupoint models, and evidence-based literature. Results: Knowledge graph constraints reduced non-standard outputs by 32%. Case studies validated the effectiveness of the interactive workflow across patient self-assessment, clinician-assisted diagnosis, and traditional Chinese medicine education. Automated paired-comparison evaluation across 30 cases further demonstrated significant improvements in diagnostic trust (Cohen's d = 1.82, p < 0.001), reduced cognitive load (improvements in four of five dimensions), and higher credibility of evidence-based references (4.21 vs. 2.95). Conclusions: The proposed system enhances the transparency of traditional Chinese medicine diagnostic reasoning and the interpretability of treatment plans through knowledge graph-driven visualization and multimodal interaction, offering a practical solution for trustworthy artificial intelligence-assisted traditional Chinese medicine applications.
Chinese Translation
目的：现有的人工智能辅助传统中医诊断工具存在推理过程不透明、交互被动以及治疗方案展示有限等问题。本研究提出了一种知识增强的可视化诊断系统，以提高症状辨识和治疗的透明度和可解释性。方法：该系统基于一个包含241种症状、1263种症状和2485种关系的Neo4j知识图谱构建。它包含一个四阶段的症状匹配流程（精确匹配、语义匹配、模糊匹配和大型语言模型验证）、一种基于信息增益的主动提问策略，并通过遗传算法进行优化，以及一种多模态治疗展示，整合了人工智能生成的插图、三维经络-腧穴模型和基于证据的文献。结果：知识图谱约束将非标准输出减少了32%。案例研究验证了该交互工作流程在患者自我评估、临床医生辅助诊断和传统中医教育中的有效性。对30个案例的自动配对比较评估进一步显示，诊断信任显著提升（Cohen's d = 1.82, p < 0.001），认知负荷降低（五个维度中有四个维度改善），以及基于证据的参考文献的可信度更高（4.21对比2.95）。结论：所提出的系统通过知识图谱驱动的可视化和多模态交互，增强了传统中医诊断推理的透明度和治疗方案的可解释性，为可信赖的人工智能辅助传统中医应用提供了切实可行的解决方案。
View on arXiv Download PDF AI Translation
cs.AI / 16 / 2606.06893

Workflow-to-Skill: Skill Creation via Routing-Workflow-Semantics-Attachments Decomposition

工作流到技能：通过路由-工作流-语义-附加分解进行技能创建
Zhang, Yuyang, Han, Xinyuan, Jiang, Xudong, Wang, Run
Abstract
Large language model agents increasingly rely on Skills to encode procedural knowledge, yet high-quality Skills remain costly to hand-write. This paper studies automatic Skill construction from heterogeneous interaction evidence, including demonstrations, agent trajectories, tool traces, and execution logs. We argue that trace-to-skill construction is not simple summarization tasks, because traces are fragmented, redundant, and may miss rare but safety-critical behaviors. To address this, we introduce RWSA, a workflow-oriented intermediate representation that decomposes Skills into Workflow structure, execution Semantics, and runtime Attachments, capturing task decomposition, control flow, verification, safety, rollback, and state management. Building on RWSA, we propose W2S, a framework that segments traces, induces local Skill drafts, aligns shared structures, reconciles branches, and compresses redundancy while preserving evidence and confidence annotations. Experiments on 70 Skills show that W2S improves behavioral replay consistency by 10.5% over summarization- and prompting-based baselines, highlighting the need to treat traces as executable runtime specifications rather than compressible text.
Chinese Translation
大型语言模型代理越来越依赖技能来编码程序知识，但高质量的技能仍然需要手动编写，成本高昂。本文研究了从异构交互证据中自动构建技能，包括演示、代理轨迹、工具痕迹和执行日志。我们认为，痕迹到技能的构建并不是简单的总结任务，因为痕迹是碎片化的、冗余的，并且可能遗漏稀有但安全关键的行为。为了解决这个问题，我们引入了RWSA（路由-工作流-语义-附加），这是一种面向工作流的中间表示，能够将技能分解为工作流结构、执行语义和运行时附加，捕捉任务分解、控制流、验证、安全性、回滚和状态管理。在RWSA的基础上，我们提出了W2S（Workflow-to-Skill），这是一个框架，能够对痕迹进行分段，诱导局部技能草稿，对齐共享结构，调和分支，并在保留证据和置信度注释的同时压缩冗余。在70个技能上的实验表明，W2S在行为重放一致性上比基于总结和提示的基线提高了10.5%，突显了将痕迹视为可执行运行时规范而非可压缩文本的必要性。
View on arXiv Download PDF AI Translation
cs.AI / 17 / 2606.06923

Declarative Skills for AI Agents in Knowledge-Grounded Tool-Use Workflows

知识基础工具使用工作流程中人工智能代理的声明性技能
Lim, M. Danish, Sharudin, I. Danial Bin, Chen, Wen Han, Lim, Cedric, Wynter, Laura
Abstract
We study orchestration mechanisms for tool-using AI agents in realistic customer-service workflows over an unstructured knowledge base. We argue that declarative agents -- AI agents equipped with natural-language skill files appended to the system prompt -- are an effective orchestration paradigm. Concretely, we compare (i) a DeclarativeAgent that reads three domain-specific skill files at inference time and decides its own control flow, (ii) an ImperativeAgent based on a programmatic state machine with explicit phases, and (iii) an unscaffolded baseline agent modeled after the $\tau$-Knowledge benchmark agent. Our ImperativeAgent is motivated by externalised-control inference as in Recursive Language Models and graph-based orchestration frameworks. We formalise the three agents as policy classes within a decentralised partially-observable Markov decision process and analyse their information-theoretic and structural properties; we then test the predicted differences empirically on five language models and two retrieval regimes. Our results show that retrieval quality is a dominant bottleneck for AI agents: when evidence is incomplete or skewed, all agents degrade substantially, and skill files cannot recover lost performance. Under high-quality retrieval, however, declarative skills consistently improve accuracy on procedural tasks and reduce orchestration errors, while the imperative state machine's brittleness does not reliably improve task success or compliance.
Chinese Translation
我们研究了在非结构化知识库中，工具使用人工智能代理在现实客户服务工作流程中的协调机制。我们认为，声明性代理——即配备有附加自然语言技能文件的人工智能代理——是一种有效的协调范式。具体而言，我们比较了（i）一个在推理时读取三个特定领域技能文件并自行决定控制流程的DeclarativeAgent，（ii）一个基于具有明确阶段的程序状态机的ImperativeAgent，以及（iii）一个以$ au$-Knowledge基准代理为模型的无结构基线代理。我们的ImperativeAgent受到外部控制推理的启发，类似于递归语言模型和基于图的协调框架。我们将这三种代理形式化为去中心化的部分可观测马尔可夫决策过程中的策略类，并分析它们的信息论和结构特性；随后，我们在五种语言模型和两种检索机制上进行了实证测试。我们的结果表明，检索质量是人工智能代理的主要瓶颈：当证据不完整或偏斜时，所有代理的性能都会显著下降，技能文件无法恢复丢失的性能。然而，在高质量检索的情况下，声明性技能在程序性任务上始终提高准确性并减少协调错误，而命令状态机的脆弱性并不能可靠地提高任务成功率或合规性。
View on arXiv Download PDF AI Translation
cs.AI / 18 / 2606.06941

Quantum-Inspired Trace-Augmented Evidence Selection for Reasoning over Structured Hypothesis Spaces

量子启发的迹线增强证据选择用于结构化假设空间的推理
Wynter, Laura, Sahoo, Nirvik, Griffin, Paul
Abstract
Large language models (LLMs) now solve a wide range of expert-level exams at or above human level, yet remain brittle on specialised, evidence-intensive domains such as law. On these tasks, errors arise not only from gaps in world knowledge but also from subtle distinctions between pieces of evidence and inconsistent use of supporting evidence. The most common aggregator over sampled chain-of-thought (CoT) traces, majority vote, returns the most popular answer regardless of whether its evidence is actually strongest. We propose to treat the selection of CoT reasoning fragments into a set of evidence as an explicit combinatorial optimisation problem, allowing well-supported but minority hypotheses to override noisy majorities, and to evaluate the approach on legal-reasoning benchmarks that are particularly sensitive to evidence quality. We introduce EP-HUBO (Evidence Pool Higher-Order Binary Optimisation), which generates multiple CoT traces with a small local model, parses fragments into per-hypothesis evidence pools, solves a higher-order unconstrained binary optimisation per pool with quality-derived weights (relevance, specificity, distinctiveness), and delegates a single adjudication call per question to a frontier model. We evaluate EP-HUBO on two evidence-intensive legal benchmarks using both simulated annealing on classical hardware and the Dirac-3 photonic entropy-quantum machine from Quantum Computing Inc. HUBO-style optimisation gives a principled way to aggregate reasoning fragments while preserving minority-but-correct hypotheses, and is most valuable in low-contamination domains where frontier models have not already absorbed the benchmark material.
Chinese Translation
大型语言模型（LLMs）现在能够以或超过人类水平解决广泛的专家级考试，但在法律等专业且证据密集的领域仍然表现脆弱。在这些任务中，错误不仅源于世界知识的缺口，还源于证据之间的细微差别和支持证据的不一致使用。最常见的对采样的思维链（CoT）迹线进行聚合的方法是多数投票，这种方法返回最受欢迎的答案，而不考虑其证据是否实际上是最强的。我们建议将CoT推理片段选择为一组证据视为一个显式的组合优化问题，从而允许支持良好的但处于少数的假设覆盖噪声较大的多数，并在对证据质量特别敏感的法律推理基准上评估该方法。我们引入了EP-HUBO（证据池高阶二元优化），它使用小型局部模型生成多个CoT迹线，将片段解析为每个假设的证据池，针对每个池解决一个高阶无约束二元优化问题，并根据质量派生的权重（相关性、特异性、独特性）进行优化，最后将每个问题的单一裁决调用委托给前沿模型。我们在两个证据密集的法律基准上评估EP-HUBO，使用经典硬件上的模拟退火和Quantum Computing Inc.的Dirac-3光子熵量子机器。HUBO风格的优化提供了一种原则性的方法来聚合推理片段，同时保留少数但正确的假设，并在前沿模型尚未吸收基准材料的低污染领域中最为有价值。
View on arXiv Download PDF AI Translation
cs.AI / 19 / 2606.06972

Accounting for Context: Shaping Moral Credences for Value Alignment

考虑背景：塑造价值对齐的道德信念
Szabo, Jazon, Modgil, Sanjay
Abstract
Ensuring that agent behaviours are aligned with human moral values inevitably raises the problem of how to account for the plurality of moral perspectives that societies -- and even individuals -- typically adopt. Work on moral uncertainty proposes mechanisms to fairly and democratically aggregate evaluations of actions across different moral theories. However, this paper argues that one needs to account for contextual factors when aggregating moral evaluations. For example, consequentialist perspectives assume an ability to accurately determine how an agent's actions change the world; an assumption that often does not hold in real world settings. We, therefore, formalise agent decision making under moral uncertainty, while also accounting for these kinds of contextual factors. We thereby show that a seemingly commonsensical property -- the weak Pareto principle -- is violated. We argue that this apparent problem is, in fact, a variation of Simpson's paradox, and hence reveals the limitations of aggregation mechanisms that ignore the impact of contextual factors.
Chinese Translation
确保代理行为与人类道德价值观相一致不可避免地引发了如何考虑社会乃至个体通常采用的多元道德视角的问题。关于道德不确定性的研究提出了在不同道德理论之间公平和民主地聚合行为评估的机制。然而，本文认为在聚合道德评估时需要考虑上下文因素。例如，结果主义视角假设能够准确判断代理的行为如何改变世界；这一假设在现实世界中往往并不成立。因此，我们在道德不确定性下形式化代理决策，同时考虑这些上下文因素。由此，我们展示了一个看似常识性的属性——弱帕累托原则——被违反。我们认为，这一明显的问题实际上是辛普森悖论的一种变体，因此揭示了忽视上下文因素影响的聚合机制的局限性。
View on arXiv Download PDF AI Translation
cs.AI / 20 / 2606.06976

Exploring Agentic Tool-Calling Decisions via Uncertainty-Aligned Reinforcement Learning

通过不确定性对齐的强化学习探索工具调用决策的能动性
Zhou, Yijin, Zeng, Linqian, Lu, Xiaoya, Xie, Wenyuan, Liu, Dongrui, Yan, Junchi, Shao, Jing
Abstract
Large language model (LLM)-based agents often make suboptimal tool-use decisions, including unsupported tool invocation and hallucinated direct responses, which may accumulate errors throughout multi-step interactions. Existing approaches mainly improve these behaviors through inference-time correction or coarse-grained reward signals based on decision outcomes and structured checklists, leaving the uncertainty characteristics of agent decisions underexplored. We observe that decision-oriented reinforcement learning tends to weaken the uncertainty separation between correct and incorrect actions, resulting in overconfident mistakes and weaker exploration signals. Therefore, we propose TRUST, which incorporates uncertainty quantification into reward design as a repulsive force for maintaining uncertainty separation, and labels lightweight key-turn annotations for unified post-training of multi-turn trajectories. Experimental results across diverse tool-use benchmarks show that TRUST consistently enhances both decision quality and agent performance while maintaining more reliable uncertainty estimates during optimization.
Chinese Translation
基于大型语言模型（LLM）的智能体通常会做出次优的工具使用决策，包括不支持的工具调用和虚构的直接响应，这可能在多步骤交互中累积错误。现有的方法主要通过推理时的修正或基于决策结果和结构化清单的粗粒度奖励信号来改善这些行为，导致智能体决策中的不确定性特征未得到充分探索。我们观察到，面向决策的强化学习往往削弱了正确与错误行为之间的不确定性分离，导致过于自信的错误和较弱的探索信号。因此，我们提出了TRUST，它将不确定性量化纳入奖励设计，作为维持不确定性分离的排斥力，并为多轮轨迹的统一后训练标记轻量级关键转折注释。在多种工具使用基准上的实验结果表明，TRUST在优化过程中始终提高了决策质量和智能体性能，同时保持了更可靠的不确定性估计。
View on arXiv Download PDF AI Translation
cs.AI / 21 / 2606.07000

Teaching the Way, Not the Answer: Privileged Tutoring Distillation for Multimodal Policy Optimization

教授方法，而非答案：多模态策略优化的特权辅导蒸馏
Xiang, Shizhe, An, Ke, Yu, Wenlong, Liu, Yue, Luan, Jian, Fu, Pei, Wang, Qilong
Abstract
Recent post-training methods, particularly Reinforcement Learning with Verifiable Rewards (RLVR), have significantly enhanced the reasoning ability of Large Vision-Language Models (LVLMs). However, the sparse nature of verifiable rewards provides little token-level supervision for failed rollouts, often leading to inefficient exploration in complex multimodal reasoning tasks. Although policy distillation can offer dense guidance, external teacher based methods introduce substantial computational overhead, while answer conditioned tuning methods may expose answer-level information and induce shortcut-like generation behavior. To address these limitations, we propose PTD-PO, a Privileged Tutoring Distillation Policy Optimization framework for RLVR that provides dense guidance without exposing the answer to the student policy. Specifically, PTD-PO constructs structured privileged hints from spatial attention guidance and intermediate textual reasoning steps, and uses them through in-context learning to produce step-wise token-distribution supervision. The student is still optimized under the original answer-free context, and its failed rollouts are aligned with the hint-augmented reference model at the token-distribution level. To further stabilize distillation under the distribution shift between guided and unguided contexts, we introduce a Top-K Jensen-Shannon divergence objective that focuses alignment on informative token probabilities while reducing memory overhead. Experiments on LVLMs ranging from 2B to 8B parameters show that PTD-PO consistently outperforms RLVR and distillation baselines, mitigates entropy collapse, and improves complex multimodal reasoning performance.
Chinese Translation
近期的后训练方法，特别是可验证奖励的强化学习（Reinforcement Learning with Verifiable Rewards, RLVR），显著增强了大型视觉语言模型（Large Vision-Language Models, LVLMs）的推理能力。然而，可验证奖励的稀疏特性为失败的回合提供的标记级监督有限，常常导致在复杂的多模态推理任务中探索效率低下。尽管策略蒸馏可以提供密集指导，但基于外部教师的方法引入了显著的计算开销，而基于答案条件的调优方法可能暴露答案级信息并诱导类似捷径的生成行为。为了解决这些局限性，我们提出了PTD-PO，一个用于RLVR的特权辅导蒸馏策略优化框架，它提供密集指导而不向学生策略暴露答案。具体而言，PTD-PO从空间注意力指导和中间文本推理步骤构建结构化的特权提示，并通过上下文学习使用这些提示生成逐步的标记分布监督。学生仍在原始无答案的上下文中进行优化，其失败的回合在标记分布级别与增强提示的参考模型对齐。为了进一步稳定在引导和非引导上下文之间的分布转移下的蒸馏，我们引入了一个Top-K Jensen-Shannon散度目标，专注于信息性标记概率的对齐，同时减少内存开销。对参数范围从2B到8B的LVLM进行的实验表明，PTD-PO始终优于RLVR和蒸馏基线，减轻了熵崩溃，并提高了复杂多模态推理性能。
View on arXiv Download PDF AI Translation
cs.AI / 22 / 2606.07017

The Sim-to-Real Gap of Foundation Model Agents: A Unified MDP Perspective

基础模型代理的模拟与现实差距：统一的马尔可夫决策过程视角
Liu, Xiaoou, Chen, Tiejin, Li, Weibo, Hu, Xiyang, Wei, Hua
Abstract
Foundation model agents are increasingly deployed for real-world decision-making, but suffer from the sim-to-real gap. While robotics and classical control have mature frameworks to address this gap, the foundation model community is treating agent robustness as an entirely novel phenomenon. Our paper proposes formalizing the foundation model agent evaluation and training gap as a classical sim-to-real problem structured entirely around the four elements of a Markov Decision Process, including Observation, Action, Transition, and Reward. In this paper, we set a comprehensive research agenda that translates classical discrepancies into the foundation model domain and advocates for adopting established solutions like domain randomization. We provide concrete examples, such as a multilingual tool calling to demonstrate how severe observation space gaps lead to operationally invalid actions despite correct semantic intent. Ultimately, this agenda aims to drive a paradigm shift, yielding a unified vocabulary and standardized stress test benchmarks to foster a new generation of highly trustworthy agents for reliable real-world applications.
Chinese Translation
基础模型代理越来越多地被用于现实世界的决策制定，但面临模拟与现实之间的差距。尽管机器人技术和经典控制领域已经建立了成熟的框架来解决这一差距，基础模型社区却将代理的鲁棒性视为一个全新的现象。本文提出将基础模型代理的评估和训练差距形式化为一个经典的模拟与现实问题，该问题完全围绕马尔可夫决策过程的四个要素构建，包括观察、行动、转移和奖励。我们在本文中设定了一个全面的研究议程，将经典差异转化为基础模型领域，并倡导采用已建立的解决方案，如领域随机化。我们提供了具体的例子，例如多语言工具调用，展示了如何严重的观察空间差距导致尽管语义意图正确，但仍产生操作上无效的行动。最终，这一议程旨在推动范式转变，产生统一的词汇和标准化的压力测试基准，以促进新一代高度可信的代理，为可靠的现实世界应用奠定基础。
View on arXiv Download PDF AI Translation
cs.AI / 23 / 2606.07027

StainFlow: Entity-Stain Tracking and Evidence Linking for Process Rewards in GUI Agents

StainFlow：用于GUI代理的实体污点跟踪和证据链接的过程奖励
Hao, Haojie, Hao, Longkun, Lou, Yihang, Bai, Yan, Li, Zhenyang, Yang, Zhichao, Huang, Dongshuo, Lin, Hongyu, Hong, Lanqing, Wang, Jiakai, Liu, Xianglong
Abstract
Reinforcement Learning (RL) has become a promising approach for improving GUI Agents in long-horizon, stochastic digital environments, but trajectory-level success feedback is too sparse to provide reliable credit assignment for intermediate exploration steps. To mitigate this issue, recent studies introduce Process Reward Models (PRMs), which provide finer-grained training feedback through global milestone verification or local step-level evaluation. However, these methods still suffer from two level-specific limitations: global milestone decomposition is subjective and singular, making it difficult to accommodate the multiple valid execution paths in real GUI tasks, while fixed local judging windows may miss long-range key evidence or dilute the decision signal with irrelevant frames. Inspired by stain-tracing mechanisms in network flow analysis, we propose StainFlow, an entity-stain-flow process reward model for GUI Agents. To reduce the subjectivity of global partitioning, we introduce the Global Entity Stain Tracking module, which extracts visually verifiable task entities and tracks how their stain concentrations and states evolve along the trajectory, allowing task phases to be objectively separated by changes in the entity evidence flow. To improve the accuracy of local verification, we introduce the Local Stain Evidence Linking module. Centered on the triggering entities of each candidate key node, it retrieves relevant steps based on their stain concentrations and state changes, and dynamically constructs high-density evidence windows for verifying true key nodes. Extensive experiments on AndroidWorld and OGRBench show that StainFlow relatively improves online RL success by 3.2% and trajectory completion judgment accuracy by 1.8%.
Chinese Translation
强化学习（RL）已成为改善GUI代理在长时间跨度、随机数字环境中的一种有前景的方法，但轨迹级成功反馈过于稀疏，无法为中间探索步骤提供可靠的信用分配。为了解决这一问题，最近的研究引入了过程奖励模型（PRMs），通过全局里程碑验证或局部步骤级评估提供更细粒度的训练反馈。然而，这些方法仍然存在两个特定于级别的局限性：全局里程碑分解是主观和单一的，难以适应真实GUI任务中的多条有效执行路径，而固定的局部判断窗口可能会错过长距离的关键证据或用无关帧稀释决策信号。受到网络流分析中污点追踪机制的启发，我们提出了StainFlow，一种用于GUI代理的实体污点流过程奖励模型。为了减少全局划分的主观性，我们引入了全局实体污点跟踪模块，该模块提取可视验证的任务实体，并跟踪其污点浓度和状态在轨迹中的演变，从而通过实体证据流的变化客观地分离任务阶段。为了提高局部验证的准确性，我们引入了局部污点证据链接模块。该模块以每个候选关键节点的触发实体为中心，根据其污点浓度和状态变化检索相关步骤，并动态构建高密度证据窗口以验证真实的关键节点。在AndroidWorld和OGRBench上的大量实验表明，StainFlow在线RL成功率相对提高了3.2%，轨迹完成判断准确率提高了1.8%。
View on arXiv Download PDF AI Translation
cs.AI / 24 / 2606.07033

Hierarchical Semantic-Constrained Heterogeneous Graph for Audio-Visual Event Localization

层次语义约束异构图用于音视频事件定位
Yang, Zhe, Zhang, Ruyi, Chen, Hongtao, Li, Wenrui, Man, Hengyu, Zuo, Wangmeng, Fan, Xiaopeng
Abstract
Open-vocabulary audio-visual event localization (OV-AVEL) jointly models audio-visual cues to recognize and temporally localize events, including categories unseen during training. Existing methods primarily learn joint audio-visual representations in Euclidean space, but still face two significant challenges. First, the lack of supervision signals for unseen categories makes it difficult to maintain audio-visual consistency across multiple temporal scales. Second, the lack of hierarchical constraints between segment- and video-level semantics prevents the model from establishing semantic consistency across different levels. To address these challenges, we propose a hierarchical semantic constrained heterogeneous graph (HSCHG) for audio-visual event localization framework. We first construct a heterogeneous hierarchical graph in Euclidean space, which includes audio and visual segment nodes and their corresponding video-level nodes. We use multi-directional temporal edges to capture complete temporal information within each modality. Simultaneously, we employ a dual-threshold filtering gated fusion strategy, introducing cross-modal information only when the alignment confidence is high. Furthermore, we introduce bidirectional semantic constraints between segment- and video-level representations to achieve semantic consistency across different levels. Based on this, we map the multi-level audio-visual representations and text prototypes uniformly into hyperbolic space. We use a hierarchical entailment regularization loss to characterize the hierarchical relationships between videos and segments. Extensive experimental results show that our method outperforms existing methods on the OV-AVEL benchmark. Ablation studies further validate the effectiveness of our method.
Chinese Translation
开放词汇音视频事件定位（OV-AVEL）联合建模音视频线索，以识别和时间定位事件，包括训练期间未见的类别。现有方法主要在欧几里得空间中学习联合音视频表示，但仍面临两个重大挑战。首先，未见类别缺乏监督信号，使得在多个时间尺度上保持音视频一致性变得困难。其次，段级和视频级语义之间缺乏层次约束，阻碍了模型在不同层次之间建立语义一致性。为了解决这些挑战，我们提出了一种层次语义约束异构图（HSCHG）用于音视频事件定位框架。我们首先在欧几里得空间中构建一个异构层次图，其中包括音频和视觉段节点及其对应的视频级节点。我们使用多方向时间边捕捉每种模态内的完整时间信息。同时，我们采用双阈值过滤门控融合策略，仅在对齐置信度高时引入跨模态信息。此外，我们在段级和视频级表示之间引入双向语义约束，以实现不同层次之间的语义一致性。在此基础上，我们将多层次音视频表示和文本原型统一映射到双曲空间。我们使用层次蕴含正则化损失来表征视频和段之间的层次关系。大量实验结果表明，我们的方法在OV-AVEL基准上优于现有方法。消融研究进一步验证了我们方法的有效性。
View on arXiv Download PDF AI Translation
cs.AI / 25 / 2606.07047

Front-to-Attractors: Modifying the Front-to-Front Heuristic in Bidirectional Search

前向吸引子：在双向搜索中修改前向前启发式
Zou, Alvin, Saleem, Muhammad Suhail, Likhachev, Maxim
Abstract
Heuristics play a central role in the performance of bidirectional search algorithms, which commonly rely on two main classes. Front-to-end (F2E) heuristics estimate the distance from a state s to the target of the search (the goal for forward search or the start for backward search). In contrast, front-to-front (F2F) heuristics estimate the distance from s to the opposite search frontier using a pairwise function h(s, s'), where s' ranges over frontier states. Although F2F heuristics are typically more informative and therefore reduce the number of node expansions, their reliance on extensive pairwise evaluations incurs substantial computational overhead. To address this limitation, we introduce a new heuristic class, front-to-attractors (F2A), that preserves much of the informativeness of F2F while dramatically reducing its computational cost. Rather than evaluating distances to all states on the opposite frontier, F2A estimates the distance from s to a small, dynamically maintained set of attractors in the opposite search direction. These attractors serve as a surrogate for the full frontier, enabling rich heuristic guidance at a fraction of the computational expense while maintaining the optimality guarantees offered by F2F. We evaluate F2A across multiple domains and show that it reduces the number of pairwise evaluations by up to 11.2x compared to F2F, while achieving 4.8x fewer node expansions than F2E on average.
Chinese Translation
启发式在双向搜索算法的性能中发挥着核心作用，这些算法通常依赖于两类主要的启发式方法。前向到终点（F2E）启发式估计状态 s 到搜索目标的距离（对于前向搜索是目标，对于后向搜索是起点）。相对而言，前向到前向（F2F）启发式则使用成对函数 h(s, s') 估计从 s 到对立搜索前沿的距离，其中 s' 遍历前沿状态。尽管 F2F 启发式通常更具信息性，从而减少节点扩展的数量，但其对广泛成对评估的依赖会带来显著的计算开销。为了解决这一限制，我们引入了一种新的启发式类别——前向吸引子（F2A），它在显著降低计算成本的同时保留了 F2F 的大部分信息性。F2A 不再评估对立前沿上所有状态的距离，而是估计从 s 到对立搜索方向上一个小的、动态维护的吸引子集合的距离。这些吸引子作为完整前沿的替代，能够在保持 F2F 提供的最优性保证的同时，以较低的计算开销提供丰富的启发式指导。我们在多个领域评估了 F2A，并显示其与 F2F 相比，成对评估数量减少了多达 11.2 倍，而平均节点扩展数量比 F2E 少了 4.8 倍。
View on arXiv Download PDF AI Translation
cs.AI / 26 / 2606.07108

DyCon: Dynamic Reasoning Control via Evolving Difficulty Modeling

DyCon：通过演变难度建模实现动态推理控制
Tu, Tengyao, Li, Yulin, Zhen, Hui-Ling, Qin, Libo, Wei, Zhoujun, Piao, Jinghua, Tian, Zhuotao, Li, Yong, Zhang, Min
Abstract
Recent advances in Large Reasoning Models (LRMs) demonstrate remarkable performance improvements by iteratively reflecting, exploring, and executing complex tasks, yet suffer from inefficiencies due to redundant reasoning, known as "overthinking". Existing methods to mitigate this issue either rely on static difficulty estimates or require task-specific training, and thus fail to adapt to the dynamic complexity during reasoning. In this work, we empirically show that the problem difficulty evolves dynamically throughout the reasoning process and is linearly encoded in the LRM's step-level embeddings. Building on this insight, we propose DyCon, a training-free framework that leverages latent step-level representations to explicitly model the evolving task difficulty, enabling the dynamic control of reasoning depth to mitigate the overthinking issue. Extensive experiments conducted on four models ranging from 4B to 32B, and across twelve benchmarks in math reasoning, general question answering, and coding tasks demonstrate that DyCon significantly enhances reasoning efficiency by reducing redundant steps without sacrificing accuracy or generalization. Project page and code are available at https://github.com/yu-lin-li/DyCon.
Chinese Translation
近期大型推理模型（Large Reasoning Models, LRM）的进展显示，通过迭代反思、探索和执行复杂任务，性能得到了显著提升，但由于冗余推理导致的低效问题，即“过度思考”，仍然存在。现有的方法要么依赖于静态的难度估计，要么需要特定任务的训练，因此无法适应推理过程中的动态复杂性。在本研究中，我们通过实证研究表明，问题的难度在推理过程中是动态演变的，并且在线性编码的LRM步级嵌入中得以体现。基于这一见解，我们提出了DyCon，一个无训练框架，利用潜在的步级表示显式建模演变的任务难度，从而实现推理深度的动态控制，以减轻过度思考的问题。在四个模型（参数范围从4B到32B）和十二个基准测试（包括数学推理、一般问答和编码任务）上进行的广泛实验表明，DyCon通过减少冗余步骤显著提高了推理效率，而不牺牲准确性或泛化能力。项目页面和代码可在 https://github.com/yu-lin-li/DyCon 获取。
View on arXiv Download PDF AI Translation
cs.AI / 27 / 2606.07113

Beyond Post-hoc Explanation: Toward Glassbox AI via Probabilistic Mediation

超越事后解释：通过概率中介迈向透明AI
Leonelli, Manuele
Abstract
Large language models are rapidly becoming infrastructural components in high-stakes institutional settings, including public administration, legal reasoning, and healthcare, where opacity is not merely inconvenient but institutionally and legally untenable. Existing approaches to explainability are predominantly post-hoc, offering unstable, non-contestable accounts that have no formal relationship to the reasoning process that produced the output. We argue that the problem is not the absence of explanation but the absence of structured reasoning in the first place. This paper makes the case for a fundamentally different architecture, which we call the Glassbox Framework, in which Bayesian networks serve as transparent, ante-hoc mediation layers for generative models. Bayesian networks encode domain knowledge, causal assumptions, and probabilistic dependencies before inference occurs, enabling auditable reasoning traces, uncertainty quantification, and contestable outputs. We characterise the architecture of this framework and ground it in a benefit eligibility scenario, identifying the foundational challenges spanning semantic alignment, dynamic model construction, probabilistic grounding, and human governance that must be solved to realise it at scale. By shifting from post-hoc explanation to ante-hoc probabilistic mediation, this work outlines a principled path toward AI systems that are not only powerful but fundamentally accountable.
Chinese Translation
大型语言模型正在迅速成为高风险机构环境中的基础设施组成部分，包括公共管理、法律推理和医疗保健，在这些领域中，不透明性不仅是不便的，而且在制度和法律上都是不可接受的。现有的可解释性方法主要是事后解释，提供不稳定、不可争议的说明，这些说明与产生输出的推理过程没有正式关系。我们认为问题不在于缺乏解释，而在于根本缺乏结构化推理。本文提出了一种根本不同的架构，我们称之为玻璃盒框架（Glassbox Framework），在该框架中，贝叶斯网络（Bayesian networks）作为透明的事前中介层用于生成模型。贝叶斯网络在推理发生之前编码领域知识、因果假设和概率依赖关系，从而实现可审计的推理轨迹、不确定性量化和可争议的输出。我们描述了该框架的架构，并将其应用于福利资格场景，识别出实现该框架所需解决的基础性挑战，包括语义对齐、动态模型构建、概率基础和人类治理等。通过从事后解释转向事前概率中介，本研究勾勒出了一条原则性路径，朝着不仅强大而且根本上负责任的AI系统迈进。
View on arXiv Download PDF AI Translation
cs.AI / 28 / 2606.07157

Think Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models

快速思考：估计前沿人工智能模型的无链思维任务完成时间范围
Gould, Dewi, Ward, Francis Rhys, Woodruff, Anders Cairns, Arike, Rauno, Hills, Josh, Serrano, Alex, Caspary, Ida, Brown, Jason Ross, Jiao, Jo J., Leask, Patrick, Stone, Twm, Potham, Ram, Stan, Ionut Gabriel, Mayne, Harry, Hellsten, Simeon, Biswas, Shubhorup, Azarbal, Ariana, Anderson, William L., Najt, Elle, Greenblatt, Ryan, Stastny, Julian
Abstract
Many efforts to ensure frontier AI models are safe rely on monitoring their chain-of-thought (CoT) reasoning. If models become able to perform sufficiently complex reasoning internally, without explicit thinking tokens, this would undermine such oversight. We measure how well frontier models reason without CoT across a suite of over 30,000 questions spanning 43 benchmarks in domains including math, coding, puzzles, causality, theory-of-mind, and strategic reasoning. To compare models against humans, we estimate the $50\%$-task-completion time horizon (TH): the human time required for tasks a model completes with $50\%$ success rate. We complement this with a $50\%$ reasoning token horizon: the minimum number of o3-mini reasoning tokens needed for tasks a model solves with $50\%$ success rate. We find that the no-CoT $50\%$ TH of frontier models has been doubling roughly every year over the past six years, with GPT-5.5's TH reaching over 3 minutes and reasoning token horizon exceeding 1,500 tokens. Our median estimates predict that frontier no-CoT THs could exceed 7 minutes by 2028, and 25 minutes by 2030, though these projections carry substantial uncertainty. We recommend frontier developers track this explicitly.
Chinese Translation
许多确保前沿人工智能模型安全的努力依赖于监测其链思维（CoT）推理。如果模型能够在内部执行足够复杂的推理，而不需要显式的思维标记，这将削弱这种监督。我们测量了前沿模型在没有链思维的情况下的推理能力，涵盖了超过30,000个问题，涉及43个基准，领域包括数学、编码、谜题、因果关系、心智理论和战略推理。为了将模型与人类进行比较，我们估计了$50\%$任务完成时间范围（TH）：人类完成模型以$50\\%$成功率完成的任务所需的时间。我们还补充了$50\\%$推理标记范围：模型以$50\\%$成功率解决任务所需的最小o3-mini推理标记数量。我们发现，前沿模型的无链思维$50\\%$ TH在过去六年中大约每年翻一番，GPT-5.5的TH超过3分钟，推理标记范围超过1,500个标记。我们的中位数估计预测，到2028年，前沿无链思维TH可能超过7分钟，到2030年可能超过25分钟，尽管这些预测存在相当大的不确定性。我们建议前沿开发者对此进行明确跟踪。
View on arXiv Download PDF AI Translation
cs.AI / 29 / 2606.07253

TOPSIS-RAD: Ranking According to Desires

TOPSIS-RAD：基于愿望的排序
Costa, Leonardo Fernandes, Costa, Helder Gomes, Lima, Diogo, Rodrigues, Brunno
Abstract
Traditional TOPSIS derives its reference points -- the Positive Ideal Solution ($PIS$) and Negative Ideal Solution ($NIS$) -- from the observed alternative set, making rankings susceptible to misalignment with decision-maker (DM) requirements, sensitivity to outlier performances, and rank reversal. This paper proposes TOPSIS-RAD, which addresses these issues by incorporating two arrays of DM-defined reference levels. Vetoed Performance Levels ($VPL$) exclude non-viable alternatives before normalisation, preventing them from distorting the ranking frontiers. Desired Performance Levels ($DPL$) cap performances at the DM's desired level before normalisation, anchoring the $PIS$ in explicit aspirations rather than dataset extremes. Three toy examples demonstrate each mechanism: $VPL$ reshapes normalisation boundaries by removing a non-viable alternative; fixed $DPL$ frontiers stabilise rankings by limiting the influence of performances well above the desired level. The method preserves the familiar distance-based structure of TOPSIS while grounding the ranking in stable, DM-specified boundaries. Limitations and future research directions are also discussed.
Chinese Translation
传统的TOPSIS方法从观察到的备选集导出其参考点——正理想解（Positive Ideal Solution, $PIS$）和负理想解（Negative Ideal Solution, $NIS$），这使得排名容易与决策者（Decision Maker, DM）的要求不一致，对异常表现敏感，并可能导致排名逆转。本文提出了TOPSIS-RAD，通过引入两个由DM定义的参考水平数组来解决这些问题。否决性能水平（Vetoed Performance Levels, $VPL$）在归一化之前排除不可行的备选方案，防止其扭曲排名边界。期望性能水平（Desired Performance Levels, $DPL$）在归一化之前将表现限制在DM所期望的水平，将$PIS$锚定在明确的期望上，而非数据集的极端值。三个简单示例展示了每种机制：$VPL$通过去除一个不可行的备选方案来重塑归一化边界；固定的$DPL$边界通过限制远高于期望水平的表现的影响来稳定排名。该方法保留了TOPSIS熟悉的基于距离的结构，同时将排名建立在稳定的、由DM指定的边界上。本文还讨论了局限性和未来研究方向。
View on arXiv Download PDF AI Translation
cs.AI / 30 / 2606.07299

DuMate-DeepResearch: An Auditable Multi-Agent System with Recursive Search and Rubric-Grounded Reasoning

DuMate-DeepResearch：一种可审计的多智能体系统，具有递归搜索和基于评分标准的推理
Yan, Lingyong, Xu, Can, Zhao, Yukun, Li, Wenxuan, Chen, Qingyang, Wu, Jiulong, Song, Wenli, Li, Xiangnan, Shi, Weixian, Chen, Yiqun, Ma, Xuchen, Li, Yuchen, Zhao, Jiashu, Wang, Shuaiqiang, Wu, Jianmin, Yin, Dawei
Abstract
Deep Research (DR) has emerged as a new agentic paradigm to tackle complex, open-ended research tasks, demanding systems that can iteratively frame problems, acquire evidence, verify sources, and synthesize long-form reports. In practice, however, current DR systems are constrained by four interrelated limitations: long-horizon planning over an underspecified scope, the bottleneck of decomposing and scheduling such tasks within a single agent, hallucination risk in long-form synthesis, and limited process auditability. This technical report presents DuMate-DeepResearch, a multi-agent DR framework built on the Qianfan Agent Foundry. The framework decouples the Agent Core, which handles task understanding, planning, and scheduling, from an extensible Tool Ecosystem for retrieval, evidence acquisition, and report rendering, making every intermediate decision and tool invocation explicitly traceable. Building on this infrastructure, DuMate-DeepResearch further introduces three mechanisms: (i) a graph-based dynamic planning strategy expands the research roadmap coarse-to-fine and continuously revises it through reflection, re-planning, backtracking, and parallel branching; (ii) a recursive two-level execution design delegates each complex search sub-task to an inner Search Agent that runs its own planning loop, isolating noisy retrieval and stabilizing long-horizon execution; (iii) a rubric-based test-time optimization mechanism dynamically generates task-specific quality criteria and uses them as live reasoning scaffolds for evidence-grounded synthesis and adaptive stopping. Across two deep research benchmarks, DuMate-DeepResearch establishes new state-of-the-art results: the best overall score (58.03%) on DeepResearch Bench, and the best overall score (61.95%) on DeepResearch Bench II while ranking first in information recall and analysis.
Chinese Translation
深度研究（Deep Research, DR）作为一种新的智能体范式，旨在应对复杂的开放式研究任务，这些任务需要系统能够迭代性地框定问题、获取证据、验证来源并综合长篇报告。然而，当前的DR系统在实践中受到四个相互关联的限制：在不明确范围内进行长期规划的困难、在单一智能体内分解和调度此类任务的瓶颈、长篇综合中的幻觉风险以及有限的过程审计能力。本技术报告介绍了DuMate-DeepResearch，一个基于Qianfan Agent Foundry构建的多智能体DR框架。该框架将处理任务理解、规划和调度的Agent Core与用于检索、证据获取和报告渲染的可扩展工具生态系统解耦，使每个中间决策和工具调用都可以明确追溯。在此基础上，DuMate-DeepResearch进一步引入了三种机制：（i）基于图的动态规划策略粗到细地扩展研究路线图，并通过反思、重新规划、回溯和并行分支不断修订；（ii）递归的两级执行设计将每个复杂搜索子任务委派给一个内部搜索智能体，该智能体运行自己的规划循环，从而隔离噪声检索并稳定长期执行；（iii）基于评分标准的测试时优化机制动态生成特定任务的质量标准，并将其作为证据基础综合和自适应停止的实时推理支架。在两个深度研究基准上，DuMate-DeepResearch建立了新的最先进结果：在DeepResearch Bench上获得最佳总体得分（58.03%），在DeepResearch Bench II上获得最佳总体得分（61.95%），并在信息召回和分析方面排名第一。
View on arXiv Download PDF AI Translation
cs.AI / 31 / 2606.07308

Off-Policy Evaluation with Strategic Agents via Local Disclosure

通过局部披露进行战略代理的离线评估
Vo, Kiet Q. H., Reddy, Abbavaram Gowtham, Rodemann, Julian, Chau, Siu Lun, Muandet, Krikamol
Abstract
We study off-policy evaluation (OPE) under strategic behavior where decision subjects (or agents) respond to a decision maker's policy by strategically modifying their covariates. Such behavior induces a policy-dependent covariate shift, breaking the standard assumption in existing methods that covariates are exogenous to the policy. Related work addresses this challenge by imposing strong assumptions such as repeated interactions or full knowledge of agents' response behavior, substantially limiting its applicability to OPE. In contrast, we consider a one-shot OPE setting where the decision maker has only partial knowledge of the agents' response behavior. Our key insight is that disclosing local information through post-hoc explanations reveals agents' pre-strategic covariates prior to adaptation, mitigating the information loss induced by strategic behavior. Leveraging this structure, we estimate a statistical model for the agents' responses and construct a doubly robust estimator for policy value. By assuming that the agents' cost sensitivity follows a conditional log-normal distribution, we establish consistency of the proposed estimator and validate our approach empirically. More broadly, our results highlight how interaction design can mitigate information asymmetry by revealing otherwise hidden structure in agents' strategic responses.
Chinese Translation
我们研究了在战略行为下的离线评估（OPE），其中决策主体（或代理）通过战略性地修改其协变量来响应决策者的政策。这种行为导致了政策依赖的协变量偏移，打破了现有方法中协变量与政策之间外生性的标准假设。相关研究通过施加强假设（如重复互动或对代理响应行为的完全了解）来应对这一挑战，这在很大程度上限制了其在OPE中的适用性。相反，我们考虑了一种单次OPE设置，其中决策者对代理的响应行为只有部分了解。我们的关键见解是，通过事后解释披露局部信息可以揭示代理在适应之前的前战略协变量，从而减轻战略行为引起的信息损失。利用这一结构，我们估计了代理响应的统计模型，并构建了一个双重稳健的政策价值估计器。通过假设代理的成本敏感性遵循条件对数正态分布，我们建立了所提估计器的一致性，并在实证上验证了我们的方法。更广泛地说，我们的结果强调了互动设计如何通过揭示代理战略响应中本来隐藏的结构来减轻信息不对称。
View on arXiv Download PDF AI Translation
cs.AI / 32 / 2606.07392

Online Pandora's Box for Contextual LLM Cascading

在线情境LLM级联的潘多拉盒子
Belloni, Alexandre, Chen, Yan, Wei, Yehua
Abstract
Motivated by Large Language Model (LLM) cascading, we propose an online contextual Pandora's Box model for adaptively querying and selecting LLM APIs. In each period, a decision-maker observes a request context and faces a two-phase decision problem. In the query phase, the decision-maker sequentially queries APIs, where each query reveals a generated output and the decision-maker incurs an (output-dependent) cost. In the selection phase, the decision-maker selects one of the generated outputs to deploy and observes only the downstream reward of the deployed output. This output-mediated feedback structure differs from classical online contextual Pandora's Box models, in which opening a box directly reveals its reward. Rather than estimating the full conditional output and cost distributions of each API, we directly model the reservation index and develop a learning approach for the query phase. Specifically, we impose a parametric structure on the contextual reservation index functions induced by the classical Weitzman's policy. Our policy combines generalized method of moments (GMM) type estimation of these reservation indices with UCB-style confidence bounds for both these indices and the shared output-level reward evaluator. Under regularity conditions, we prove that the resulting policy achieves dimension-dependent $\widetilde O(\sqrt T)$ cumulative regret over a horizon of $T$ periods.
Chinese Translation
受到大型语言模型（LLM）级联的启发，我们提出了一种在线情境潘多拉盒子模型，用于自适应查询和选择LLM API。在每个阶段，决策者观察请求上下文，并面临一个两阶段的决策问题。在查询阶段，决策者依次查询API，每次查询都会揭示一个生成的输出，并且决策者会产生一个（依赖于输出的）成本。在选择阶段，决策者从生成的输出中选择一个进行部署，并仅观察所部署输出的下游奖励。这种输出介导的反馈结构与经典的在线情境潘多拉盒子模型不同，后者打开一个盒子会直接揭示其奖励。我们并不直接估计每个API的完整条件输出和成本分布，而是直接对保留指数进行建模，并为查询阶段开发了一种学习方法。具体而言，我们对由经典Weitzman政策引导的情境保留指数函数施加了参数结构。我们的政策将这些保留指数的广义矩方法（GMM）类型估计与UCB风格的置信界结合在一起，适用于这些指数和共享输出级奖励评估器。在常规条件下，我们证明了所得到的政策在$T$个周期的时间范围内实现了与维度相关的$ ilde O( ext{sqrt}(T))$累积遗憾。
View on arXiv Download PDF AI Translation
cs.AI / 33 / 2606.07462

Act As a Real Researcher: A Suite of Benchmarks Evaluating Frontier LLMs and Agentic Harnesses in Research Lifecycle

作为真实研究者的行动：评估前沿大型语言模型及其在研究生命周期中的代理应用的一系列基准测试
Wang, Jiayu, Lv, Weijiang, Fu, Bowen, Fu, Jing, Song, Jiayi, Zhang, Lingyu, Xue, Lanxuan, Chen, Luodi, Xin, Zepeng, Li, Kaiyu, Cao, Xiangyong
Abstract
As foundation models advance and agent scaffolding becomes increasingly sophisticated, agents have demonstrated remarkable proficiency in complex, long-horizon coding tasks and even autonomous experiment execution. Despite their evolution from research assistants into autonomous research agents, these systems still exhibit significant limitations in field sensitivity, research ethics, and nuanced scientific judgment. Consequently, frontier agents remain unable to fully replace human researchers. To bridge this gap, we conceptualize the AARR (Act As a Real Researcher) benchmark series. Unlike existing benchmarks that primarily assess macro-level execution capabilities, AARR focuses on whether agents can emulate the professionalism, thoroughness, and nuanced reasoning that characterize human researchers in granular research scenarios. In this work, we propose AARRI-Bench (Act As a Real Research Intern), the first benchmark in this series. We conduct extensive experiments across frontier models and agentic systems, revealing that even the best-performing configuration (Mini-SWE-Agent with Claude Opus 4.7) achieves only 68.3\% success rate, frequently overlooking subtle yet critical details that are obvious to real human researchers. Our results indicate that developing researcher-like AI requires further exploration of research behavior, rather than merely complex scaffolding. Our data is released at https://github.com/AARR-bench/AARRI-bench.
Chinese Translation
随着基础模型的进步和代理框架的日益复杂化，代理在复杂的长期编码任务甚至自主实验执行方面展现出了显著的能力。尽管这些系统已从研究助手演变为自主研究代理，但它们在领域敏感性、研究伦理和细致科学判断方面仍存在显著局限。因此，前沿代理仍无法完全替代人类研究者。为了弥补这一差距，我们构思了AARR（Act As a Real Researcher）基准系列。与现有主要评估宏观执行能力的基准不同，AARR关注代理是否能够在细致的研究场景中模拟人类研究者所特有的专业性、严谨性和细致推理。在本研究中，我们提出了AARRI-Bench（Act As a Real Research Intern），该系列中的第一个基准。我们在前沿模型和代理系统上进行了广泛实验，结果显示，即使是表现最佳的配置（Mini-SWE-Agent与Claude Opus 4.7）也仅达到了68.3%的成功率，常常忽视那些对真实人类研究者而言显而易见的微妙但关键的细节。我们的结果表明，开发类研究者的人工智能需要进一步探索研究行为，而不仅仅是复杂的框架。我们的数据已发布在 https://github.com/AARR-bench/AARRI-bench。
View on arXiv Download PDF AI Translation
cs.AI / 34 / 2606.07489

How AI Agents Reshape Knowledge Work: Autonomy, Efficiency, and Scope

人工智能代理如何重塑知识工作：自主性、效率与范围
Yang, Jeremy, Zyskowski, Kate, Yonack, Noah, Ma, Jerry
Abstract
Frontier AI systems are bridging the gap between intelligence and utility by shifting from conversational assistants to autonomous agents that execute tasks end to end. Using production data from Perplexity's Search and Computer products, we study this transition by examining how AI agents accelerate and reshape knowledge work. Three key empirical findings emerge. First, using sessions with near-identical initial query pairs as natural experiments for the same underlying task attempted with both products, Computer performs 26 minutes of autonomous work per user session, versus 33 seconds for Search. Computer automates task decomposition and execution that Search users might otherwise manually orchestrate and implement. As a result, Computer shifts follow-up query distribution toward higher-order work such as verification and extension. Autonomy also increases execution quality, with per-query dissatisfaction rates 55% lower on Computer than on Search. Second, due to its autonomy advantage, Computer reduces completion time from 269 to 36 minutes on matched tasks, lowering estimated time and cost by 87% and 94%, respectively, compared to humans equipped with Search alone. Third, Computer changes the scope of work that users attempt: Computer queries more often cross occupational boundaries, require higher-order cognition, draw on broader expertise, take the form of composite tasks that bundle interdependent subtasks into a single query, and unlock work activities that are essentially absent from Search usage among the same users. Together, the evidence indicates that AI agents accelerate workflows, enhance output quality, reduce costs, and expand the breadth and depth of automated work.
Chinese Translation
前沿人工智能系统通过从对话助手转变为能够端到端执行任务的自主代理，弥合了智能与实用之间的差距。我们利用Perplexity的搜索和计算产品的生产数据，研究这一转变，考察人工智能代理如何加速和重塑知识工作。研究得出了三个关键的实证发现。首先，使用几乎相同初始查询对的会话作为自然实验，比较两种产品在相同基础任务上的表现，计算机（Computer）在每个用户会话中执行了26分钟的自主工作，而搜索（Search）仅为33秒。计算机自动化了任务的分解和执行，而搜索用户可能会手动协调和实施这些任务。因此，计算机将后续查询的分布转向更高层次的工作，如验证和扩展。自主性还提高了执行质量，计算机的每查询不满率比搜索低55%。其次，由于其自主性优势，计算机在匹配任务上将完成时间从269分钟缩短至36分钟，分别降低了估计的时间和成本87%和94%，相比之下，单靠搜索的人工用户则无法实现这一点。第三，计算机改变了用户尝试的工作范围：计算机查询更频繁地跨越职业边界，要求更高层次的认知，依赖更广泛的专业知识，呈现为将相互依赖的子任务捆绑成单一查询的复合任务，并解锁在同一用户的搜索使用中基本不存在的工作活动。综合来看，证据表明人工智能代理加速了工作流程，提高了输出质量，降低了成本，并扩展了自动化工作的广度和深度。
View on arXiv Download PDF AI Translation
计算语言学 (Computation and Language)
62
cs.CL / 1 / 2606.06586

Improving Cross-Lingual Factual Recall via Consistency-Driven Reinforcement Learning

通过一致性驱动的强化学习提高跨语言事实回忆
von Rad, Jonathan, Arts, Louis, Burgess, George, Kolokytha, Eleftheria, O'Donnell, Harry, Doumpas, Ektor Oikonomidis, Sanchez, Eduardo, Lu, Yao, Stenetorp, Pontus
Abstract
Large language models (LLMs) trained predominantly on English data encode substantial world knowledge, yet often fail to express it reliably in other languages, a phenomenon known as cross-lingual factual inconsistency. To study and address this, we introduce PolyFact, a large-scale parallel multilingual factual QA dataset containing 100K Wikidata-grounded facts across 12 typologically diverse languages. Using PolyFact, we compare light continual pretraining (CPT), supervised fine-tuning (SFT), and reinforcement learning via Group Relative Policy Optimization (GRPO) for improving cross-lingual factual recall in Qwen-2.5-7B and OLMo-2-1124-7B. We find that GRPO consistently outperforms SFT, improving both cross-lingual consistency and generalization to unseen languages, while CPT on parallel data yields limited additional gains. Mechanistic analyses further show that GRPO reorganizes multilingual routing by reducing language specialization in MLP layers and attention heads, thereby promoting more shared cross-lingual representations. We release our code, models, and dataset.
Chinese Translation
大型语言模型（LLMs）主要在英语数据上训练，编码了大量的世界知识，但在其他语言中往往无法可靠地表达这些知识，这种现象被称为跨语言事实不一致。为了研究和解决这一问题，我们引入了PolyFact，一个大规模的平行多语言事实问答数据集，包含12种类型多样的语言中的10万条基于Wikidata的事实。利用PolyFact，我们比较了轻量持续预训练（CPT）、监督微调（SFT）和通过群体相对策略优化（GRPO）的强化学习，以提高Qwen-2.5-7B和OLMo-2-1124-7B中的跨语言事实回忆。我们发现GRPO在性能上始终优于SFT，改善了跨语言一致性和对未见语言的泛化能力，而在平行数据上进行CPT的额外收益有限。机制分析进一步表明，GRPO通过减少多语言路由中的语言专业化，重新组织了多语言路由，促进了更共享的跨语言表示。我们将发布我们的代码、模型和数据集。
View on arXiv Download PDF AI Translation
cs.CL / 2 / 2606.06614

Re-Centering Humans in LLM Personalization

重新聚焦人类在大型语言模型个性化中的角色
Zhang, Lechen, Liu, Jiarui, August, Tal
Abstract
Despite growing interest, most evaluations of large language models' (LLMs') personalization abilities have relied on synthetic data. It remains unclear how well current personalization systems work for real users. In this paper, we study the gap in LLM personalization performance when using synthetic versus human data. We collect human conversations (550 conversations) and judgments across three stages of personalization: extracting user attributes from conversations (5,949 judgments), pairing relevant attributes with new prompts (11,919), and incorporating relevant attributes into a personalized response (1,101). Incorporating human data reveals system limitations at each stage. Models struggle to extract attributes from human conversations, disagree with human judgments on relevant attributes, and generate personalized responses that humans judge no better than generic responses (though that LLM judges widely rate as better). We introduce two lightweight training-based interventions that shift automated personalization evaluation closer to human data in our first two stages. However, in our third stage we find that learned reward models achieve only modest correlation with human ratings, suggesting that human-aligned personalization quality judgments are difficult to model directly. Our collected data provides a foundation for studying how models should extract, select, and incorporate user information in ways that humans find useful.
Chinese Translation
尽管对大型语言模型（LLMs）个性化能力的兴趣日益增加，但大多数评估仍依赖于合成数据。目前尚不清楚现有的个性化系统在真实用户中的表现如何。本文研究了使用合成数据与人类数据时LLM个性化性能的差距。我们收集了人类对话（550个对话）和在个性化的三个阶段中的判断：从对话中提取用户属性（5,949个判断）、将相关属性与新提示配对（11,919个判断），以及将相关属性纳入个性化响应（1,101个判断）。纳入人类数据揭示了系统在每个阶段的局限性。模型在从人类对话中提取属性时表现不佳，与人类对相关属性的判断存在分歧，并生成的个性化响应在评估中与通用响应无显著差异（尽管LLM评估普遍认为其表现更好）。我们提出了两种轻量级的基于训练的干预措施，使自动化个性化评估在前两个阶段更接近人类数据。然而，在第三阶段，我们发现学习到的奖励模型与人类评分的相关性仅为适中，这表明人类对个性化质量判断的对齐较难直接建模。我们收集的数据为研究模型应如何提取、选择和纳入用户信息，以便人类认为有用提供了基础。
View on arXiv Download PDF AI Translation
cs.CL / 3 / 2606.06622

UnpredictaBench: A Benchmark for Evaluating Distributional Randomness in LLMs

UnpredictaBench：评估大型语言模型中分布随机性的基准
Abaskohi, Amirhossein, Dabiriaghdam, Amirhossein, Luo, Liang, Wen, Ellie Dingqiao, Wang, Lele, Carenini, Giuseppe, West, Peter
Abstract
We introduce UnpredictaBench, an evaluation that tests the ability of large language models (LLMs) to capture true underlying distributions. As LLMs are increasingly used as substitutes for other entities (e.g., for humans in economic simulations), the tendency of many models to collapse towards a single plausible answer means a failure to capture the unpredictability of real systems. Recent work on improving output diversity is insufficient for this setting: simulation requires samples that are calibrated to a target distribution, not merely varied outputs. UnpredictaBench isolates a simplified but fundamental version of this problem: sampling outcomes from individual target distributions, including canonical statistical distributions, distributions induced by stochastic programs, and natural-language scenarios that describe random processes. We introduce 448 such problems together with KS@N, a general-purpose evaluation metric that quantifies how well a model outputs approximate black-box target distributions via the Kolmogorov-Smirnov statistical test. This is the rate at which we fail to reject model samples of size N against ground-truth samples, with larger N indicating greater difficulty. Tested across open and proprietary models, we find a large spread in distributional capabilities. For instance, when models generate samples of size 100 (KS@100, our standard metric), scores range from near 0 to over 20%. No model is able to achieve over 40% at KS@100, showing significant headroom in distributional sampling as a capability. Although adding reasoning can somewhat increase scores, we find no immediate solution for this issue. UnpredictaBench shows that even simple distributional simulation remains challenging, making it a necessary first step toward using LLMs as stand-ins for complex systems.
Chinese Translation
我们介绍了UnpredictaBench，这是一种评估大型语言模型（LLMs）捕捉真实潜在分布能力的工具。随着LLMs越来越多地被用作其他实体（例如，在经济模拟中替代人类），许多模型倾向于收敛到一个单一的合理答案，这意味着未能捕捉真实系统的不可预测性。近期关于提高输出多样性的研究在这种情况下是不够的：模拟需要与目标分布相匹配的样本，而不仅仅是多样化的输出。UnpredictaBench孤立出这个问题的一个简化但基本的版本：从各个目标分布中采样结果，包括经典统计分布、由随机程序引发的分布，以及描述随机过程的自然语言场景。我们引入了448个此类问题，以及KS@N，一个通用评估指标，用于量化模型通过Kolmogorov-Smirnov统计检验输出近似黑箱目标分布的能力。这是我们在样本大小为N的情况下未能拒绝模型样本与真实样本的比率，较大的N表示更大的难度。在开放和专有模型的测试中，我们发现分布能力差异很大。例如，当模型生成大小为100的样本时（KS@100，我们的标准指标），得分范围从接近0到超过20%。没有模型能够在KS@100上达到超过40%的得分，显示出在分布采样能力方面仍有显著的提升空间。尽管增加推理可以在一定程度上提高得分，但我们没有找到该问题的直接解决方案。UnpredictaBench表明，即使是简单的分布模拟仍然具有挑战性，这使其成为将LLMs作为复杂系统替代品的必要第一步。
View on arXiv Download PDF AI Translation
cs.CL / 4 / 2606.06635

How Language Models Fail: Token-Level Signatures of Committed and Persistent Reasoning Failures

语言模型失败的表现：承诺性和持续性推理失败的标记级特征
Thoria, Tanvi, Jafari, Kiana, Schlichting, Marc R., Kochenderfer, Mykel J.
Abstract
Failures in language model reasoning emerge through distinct processes that leave identifiable signatures in the reasoning trace. We characterize these failures using token-level uncertainty signals, finding they arise through two empirically distinguishable processes. The first is committed failure, in which a model locks onto an incorrect reasoning path early in its trace. A central diagnostic signature is the commitment point, beyond which considering additional tokens hurt rather than help failure detection. In the second, persistent uncertainty, uncertainty instead accumulates throughout, and the full trace is needed to best distinguish failing from successful completions. These signatures reproduce across 23 model-dataset configurations, with the framework's falsifiable predictions holding in 20 of 23 cases, well above chance across both failure modes. Finally, we demonstrate our failure mode framework has direct implications for self-consistency, identifying when uncertainty signals complement it and when it can be selectively skipped. These results offer a foundation for understanding when LLM reasoning failures become detectable and for adapting detection strategies accordingly.
Chinese Translation
语言模型推理中的失败通过不同的过程显现，这些过程在推理轨迹中留下可识别的特征。我们使用标记级不确定性信号来表征这些失败，发现它们通过两种经验上可区分的过程产生。第一种是承诺性失败，在这种情况下，模型在推理轨迹的早期锁定于错误的推理路径。一个核心的诊断特征是承诺点，超过此点后，考虑额外的标记反而会妨碍而非帮助失败检测。第二种是持续性不确定性，在这种情况下，不确定性在整个过程中累积，并且需要完整的轨迹才能最佳区分失败与成功的完成。这些特征在23种模型-数据集配置中得以重现，框架的可证伪预测在23个案例中的20个中成立，远高于偶然水平，适用于两种失败模式。最后，我们展示了我们的失败模式框架对自一致性具有直接影响，识别出何时不确定性信号与其相辅相成，以及何时可以选择性跳过。这些结果为理解何时语言模型推理失败变得可检测以及相应调整检测策略提供了基础。
View on arXiv Download PDF AI Translation
cs.CL / 5 / 2606.06646

CAF-Gen: A Multi-Agent System for Enriching Argumentation Structures

CAF-Gen：一个用于丰富论证结构的多代理系统
Bąba, Jakub, Chudziak, Jarosław
Abstract
Formalizing complex reasoning from natural text is one of the central challenges in computational linguistics. It requires systems to understand not just keywords but also the context and complex reasoning embedded in a text. Current Argument Mining (AM) techniques identify basic claims and premises, yet they often struggle to capture the richer structural information required by advanced schemas such as the Carneades Argumentation Framework (CAF), which incorporates features such as premise types, proof standards, and argument schemes. We address this limitation by introducing CAF-Gen, an automated multi-agent framework designed to enrich shallow argument structures into CAF-compliant argument models. By employing an iterative Creator-Reviewer pipeline, a creator agent's output is validated by a critical agent to ensure structural integrity. This multi-agent collaboration is crucial for mitigating the structural instability typical of single-pass generative models. Our experiments demonstrate that the iterative feedback loop improves the quality of the resulting data and achieves strong alignment with the original annotations, while producing structurally richer models. Our findings show that the multi-agent system can overcome the limitations of single-pass generation, providing a robust methodology for the automated modeling of formal argumentation.
Chinese Translation
将自然文本中的复杂推理形式化是计算语言学中的一个核心挑战。这要求系统不仅理解关键词，还要理解文本中嵌入的上下文和复杂推理。目前的论证挖掘（Argument Mining, AM）技术识别基本的主张和前提，但通常难以捕捉高级框架（如Carneades Argumentation Framework, CAF）所需的更丰富的结构信息，该框架包含前提类型、证明标准和论证方案等特征。我们通过引入CAF-Gen来解决这一局限性，CAF-Gen是一个自动化的多代理框架，旨在将浅层论证结构丰富为符合CAF的论证模型。通过采用迭代的创作者-审阅者管道，创作者代理的输出由一个关键代理进行验证，以确保结构的完整性。这种多代理协作对于减轻单遍生成模型典型的结构不稳定性至关重要。我们的实验表明，迭代反馈循环提高了结果数据的质量，并与原始注释实现了强对齐，同时生成了结构更丰富的模型。我们的研究结果表明，多代理系统能够克服单遍生成的局限性，为自动化形式论证建模提供了一种稳健的方法。
View on arXiv Download PDF AI Translation
cs.CL / 6 / 2606.06667

The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignment

泛化的搭载假设：解释和缓解新出现的错位
Zhao, Jiachen, Wu, Zhengxuan, Arora, Aryaman, Sun, Yiyou, Bau, David, Shi, Weiyan
Abstract
The mechanisms behind LLMs' broad over-generalization beyond training examples remain unclear. Emergent misalignment (EM) offers a striking case study: finetuning on narrow tasks induces broad misalignment to semantically-unrelated test domains. In this work, we propose the Piggyback Hypothesis: the chat-template tokens can piggyback the finetuned behaviour onto out-of-domain queries. We validate this hypothesis by showing that subtle perturbations to the prefix (tokens preceding all user queries), or patching the prefix representations with those from the unfinetuned model, can restore alignment without changing the user query. Building on this finding, we propose Token-Regularized Finetuning (TReFT), which regularizes specific token representations during training to mitigate EM. Across different models and multiple EM-inducing datasets, TReFT reduces EM while preserving in-domain learning. On Llama-3.1-8B finetuned on the legal domain, TReFT achieves 33.5% more EM reduction than data interleaving with a retain set of aligned examples. We further show that TReFT extends to other narrow-finetuning settings, including abstention, tool use, and refusal (off-topic generalization is reduced by 54.3% on average), supporting the Piggyback Hypothesis. Broadly, our work highlights that LLMs may learn and generalize in unintended ways and suggests a path toward more constrained finetuning. It also calls for further study of how shared input features can piggyback model behavior across domains.
Chinese Translation
大型语言模型（LLMs）在训练示例之外的广泛过度泛化机制仍不清楚。新出现的错位（Emergent Misalignment, EM）提供了一个引人注目的案例研究：在狭窄任务上的微调会导致与语义无关的测试领域的广泛错位。在本研究中，我们提出了搭载假设（Piggyback Hypothesis）：聊天模板令牌可以将微调行为搭载到域外查询上。我们通过展示对前缀（所有用户查询之前的令牌）进行微小扰动，或用未微调模型的前缀表示进行修补，可以在不改变用户查询的情况下恢复对齐，从而验证了这一假设。基于这一发现，我们提出了令牌正则化微调（Token-Regularized Finetuning, TReFT），该方法在训练过程中对特定令牌表示进行正则化，以缓解EM。在不同模型和多个诱发EM的数据集上，TReFT在保持领域内学习的同时减少了EM。在针对法律领域微调的Llama-3.1-8B模型中，TReFT实现了比数据交错与保留对齐示例集的EM减少高出33.5%。我们进一步展示了TReFT可以扩展到其他狭窄微调设置，包括弃权、工具使用和拒绝（平均减少54.3%的离题泛化），支持搭载假设。总体而言，我们的工作强调了LLMs可能以意想不到的方式学习和泛化，并建议了一条更受限的微调路径。同时也呼吁进一步研究共享输入特征如何在不同领域间搭载模型行为。
View on arXiv Download PDF AI Translation
cs.CL / 7 / 2606.06674

What Do People Actually Want From AI? Mapping Preference Plurality

人们实际上希望从人工智能中获得什么？偏好多样性的映射
Coelho, Julia Sepúlveda, Hale, Scott A.
Abstract
Large Language Models (LLMs) are often fine-tuned through Reinforcement Learning from Human Feedback (RLHF) to align with people's preferences and values. However, this method has known limitations: it aggregates conflicting preferences, often relies on unrepresentative samples, and uses only binary comparisons. Analysing 1,500 open-ended responses from the PRISM dataset across 75 countries, we examine what people actually want from AI systems and reveal concrete failures of current methods. We find that different people want different things: most values are requested by fewer than a quarter of respondents, with truthfulness the sole exception at 49%. Furthermore, the same words hide divergent meanings: when people describe what they mean by "truthfulness", they reveal distinct, potentially incompatible, epistemological bases, as some ask for sourced claims, some for expert opinions, and some even ask for unpopular views. Certain capabilities, namely how human-like a model behaves, and some features, like AI guardrails, are outright controversial, with some desiring them and others rejecting them. We additionally find that people often use contextual distinctions (what AI should do "by default" versus "if requested") that binary comparisons cannot capture. These findings expose fundamental problems in current alignment practices. When 49% request truthfulness but define it differently, this is unlikely to be captured by a single reward model. The persistence of high hallucination rates in well-funded models, despite users' clear demands for accuracy, suggests that current methods fail to identify actual preferences. This paper sheds light on the situated, contested, imperfect signals that are currently being flattened into universal preference models, a practice others have characterised as epistemic violence.
Chinese Translation
大型语言模型（LLMs）通常通过人类反馈的强化学习（RLHF）进行微调，以与人们的偏好和价值观保持一致。然而，这种方法存在已知的局限性：它聚合了相互冲突的偏好，往往依赖于不具代表性的样本，并且仅使用二元比较。通过分析来自75个国家的PRISM数据集中1500个开放式回应，我们考察了人们实际上希望从人工智能系统中获得什么，并揭示了当前方法的具体失败。我们发现，不同的人希望获得不同的东西：大多数价值观的请求者不到四分之一，唯有“真实”这一项例外，达到49%。此外，相同的词语隐藏着不同的含义：当人们描述他们所理解的“真实”时，他们揭示了不同的、可能不兼容的认识论基础，因为一些人要求有来源的主张，一些人要求专家意见，甚至一些人要求不受欢迎的观点。某些能力，特别是模型的类人行为，以及一些特征，如人工智能的保护措施，都是明显有争议的，有些人渴望这些特征，而另一些人则拒绝它们。我们还发现，人们经常使用上下文区分（人工智能应该“默认”执行的任务与“在请求时”执行的任务），而二元比较无法捕捉这些区分。这些发现揭示了当前对齐实践中的根本问题。当49%的人请求“真实”但对其定义不同，这不太可能通过单一的奖励模型来捕捉。尽管用户明确要求准确性，但在资金充足的模型中高幻觉率的持续存在表明，当前方法未能识别实际偏好。本文阐明了当前被简化为普遍偏好模型的情境化、争议性和不完善的信号，这种做法被他人描述为认识论暴力。
View on arXiv Download PDF AI Translation
cs.CL / 8 / 2606.06679

HKJudge: A Legal Discourse-Annotated Corpus for Interpreting What Courts Find, How They Reason, and What They Rule

HKJudge：一个法律话语注释语料库，用于解读法院的发现、推理和裁决
Xuan, Xi, Zhang, Wenxin, Zhou, Yufei, Sin, King-kui, Kit, Chunyu
Abstract
Court judgments are central to legal practice and jurisprudence, yet discourse analysis of Hong Kong judgments has received limited attention, owing largely to the absence of expert-annotated corpora. We introduce the Hong Kong Judgment Discourse Dataset (HKJudge), the first sentence-level expert-annotated legal discourse corpus. HKJudge includes criminal judgments across all five levels of HK's court hierarchy, comprising $\sim$290k sentences and $\sim$6.5 million tokens, fully annotated by legal linguistics experts. We design a two-tier discourse schema that captures what facts a court finds, how it reasons, and what it rules. At the sentence level, each sentence is assigned one of 26 rhetorical roles. At the span level, sentences are further annotated with three sentencing elements (charge, imprisonment term, fine). Ten legal linguistics annotators produced the annotations with an inter-annotator agreement of $\kappa = 0.8$. We formulate two tasks on HKJudge, termed rhetorical role classification and legal element extraction, and provide the first benchmark evaluation of four BERT-based models, two open-source LLMs under zero-shot and fine-tuning settings, and four commercial LLMs on both tasks. Our work demonstrates the value of sentence-level discourse annotation for modeling the structure of HK judgments and provides a rich data foundation for future work on legal judgment prediction. The HKJudge dataset and code are available at https://github.com/xuanxixi/HKJudge.
Chinese Translation
法院判决是法律实践和法理学的核心，但对香港判决的语篇分析关注有限，这在很大程度上是由于缺乏专家注释的语料库。我们介绍了香港判决话语数据集（HKJudge），这是第一个句子级别的专家注释法律话语语料库。HKJudge包括香港法院层级五个级别的刑事判决，包含约29万句和约650万个词元，全部由法律语言学专家进行注释。我们设计了一个两层次的语篇架构，捕捉法院发现的事实、推理方式和裁决内容。在句子级别，每个句子被分配一个26种修辞角色之一。在跨度级别，句子进一步注释了三个判决要素（指控、监禁期限、罚款）。十位法律语言学注释员进行了注释，注释者间一致性为κ = 0.8。我们在HKJudge上制定了两个任务，分别称为修辞角色分类和法律要素提取，并提供了四个基于BERT的模型、两个在零样本和微调设置下的开源大型语言模型（LLM）以及四个商业LLM在这两个任务上的首次基准评估。我们的工作展示了句子级别话语注释在建模香港判决结构中的价值，并为未来的法律判决预测研究提供了丰富的数据基础。HKJudge数据集和代码可在https://github.com/xuanxixi/HKJudge获取。
View on arXiv Download PDF AI Translation
cs.CL / 9 / 2606.06708

Signal-Driven Observation for Long-Horizon Web Agents

基于信号驱动的长时间跨度网页代理观察
Gaur, Shubham, Lane, Ian
Abstract
Web agents operating over long horizons ingest raw DOM and accessibility trees -- routinely tens of thousands of tokens -- at every action step, causing progressive context degradation that erodes reasoning well before tasks complete. We argue that this coupling of observation frequency to action frequency is an architectural mistake. Drawing on the insight from Recursive Language Models that querying a document outperforms reading it wholesale, we propose Signal-Driven Observation (SDO): a dedicated sub-call reads the full DOM but returns only task-relevant elements and their selectors, and is re-invoked only when a lightweight signal detector fires -- triggered by URL transitions, newly visible interactive elements, action failures, or exogenous browser events. We outline the open problems SDO introduces and call on the community to treat observation compression as a core architectural decision in web agent design.
Chinese Translation
在长时间跨度内操作的网页代理在每个动作步骤中都会处理原始的DOM和可访问性树——通常是数万个标记——这导致逐渐的上下文退化，使得推理在任务完成之前就受到影响。我们认为，将观察频率与动作频率耦合是一种架构上的错误。借鉴递归语言模型的见解，即查询文档的效果优于整体读取文档，我们提出了基于信号驱动的观察（Signal-Driven Observation, SDO）：一个专用的子调用读取完整的DOM，但仅返回与任务相关的元素及其选择器，并且仅在轻量级信号检测器触发时重新调用——信号触发的条件包括URL的变化、新出现的可交互元素、动作失败或外部浏览器事件。我们概述了SDO所引入的开放问题，并呼吁社区将观察压缩视为网页代理设计中的核心架构决策。
View on arXiv Download PDF AI Translation
cs.CL / 10 / 2606.06712

Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation

通过在线策略蒸馏实现数据高效的自回归到扩散语言模型
Su, Xingyu, Helwig, Jacob, Parashar, Shubham, Chagi, Atharv, Jotsna, Lakshmi, Zhi, Degui, Caverlee, James, Kalathil, Dileep, Ji, Shuiwang
Abstract
We study the transformation of autoregressive models (ARLMs) into diffusion language models (DLMs). Rather than pretraining from scratch, prior work replaces the causal attention in ARLMs with bidirectional attention and then trains the resulting model using a DLM objective. However, these approaches incur two distribution shifts. First, transitioning from a next-token prediction objective to a DLM objective can discard knowledge acquired by the ARLM during training. Second, standard DLMs suffer from a train-inference mismatch, as the training loss is defined on randomly masked sequences rather than the trajectories encountered at inference produced by confidence-based decoding. To address both challenges, we introduce an On-Policy Diffusion Language Model (OPDLM) in which On-Policy Distillation (OPD) is employed for ARLM-to-DLM transformation. Specifically, OPDLM is trained via self-OPD, where the student, an ARLM with bidirectional attention, generates its own trajectories, and the teacher, the original frozen ARLM, distills its knowledge by providing target logits on these trajectories. By training directly in an on-policy manner, OPDLM eliminates the train-inference mismatch in DLMs, while distillation from the original model enhances knowledge retention from the ARLM. Empirical results demonstrate that OPDLM requires 15x to 7,000x fewer training tokens with strong performance across a wide variety of tasks. OPDLM avoids the prohibitive cost of DLM pretraining and positions DLM transformation as a form of ARLM post-training.
Chinese Translation
我们研究了自回归模型（ARLMs）向扩散语言模型（DLMs）的转化。与从头开始预训练不同，先前的工作将ARLMs中的因果注意力替换为双向注意力，然后使用DLM目标训练得到的模型。然而，这些方法会产生两个分布转变。首先，从下一个标记预测目标转变为DLM目标可能会丢弃ARLM在训练过程中获得的知识。其次，标准DLM在训练和推理之间存在不匹配，因为训练损失是在随机掩蔽序列上定义的，而不是在推理过程中通过基于置信度的解码生成的轨迹上。为了解决这两个挑战，我们引入了一种在线策略扩散语言模型（OPDLM），在其中采用在线策略蒸馏（OPD）进行ARLM到DLM的转化。具体而言，OPDLM通过自我OPD进行训练，其中学生模型是一个具有双向注意力的ARLM，它生成自己的轨迹，而教师模型是原始的冻结ARLM，通过在这些轨迹上提供目标logits来蒸馏其知识。通过直接以在线策略的方式进行训练，OPDLM消除了DLM中的训练-推理不匹配，同时从原始模型的蒸馏增强了ARLM的知识保留。实证结果表明，OPDLM在各种任务中表现强劲，所需的训练标记减少了15倍到7000倍。OPDLM避免了DLM预训练的高昂成本，并将DLM转化定位为ARLM后训练的一种形式。
View on arXiv Download PDF AI Translation
cs.CL / 11 / 2606.06715

Does Topic Sentiment Cause Perceived Ideology? Comparing Human and LLM Annotations in Political News Articles

主题情感是否导致感知意识形态？比较政治新闻文章中人类与大型语言模型（LLM）的标注
Chatterjee, Upasana
Abstract
We ask whether topic sentiment has a causal effect on perceived political ideology, and whether the answer depends on who assigns the ideology label. Using articles from AllSides, paired with shared sentiment annotations from Llama-3.3-70b-versatile, we compare ideology labels from expert human annotators, GPT-4o-mini (baseline and finetuned), and Llama-3.3-70B. We apply Double Machine Learning (DML) and community-level mediation analysis across all four annotation paradigms. Human annotations yield no significant causal effects at the community level. Fine-tuned GPT-4o-mini achieves the highest classification accuracy (F1=72.48) and is the only annotator paradigm that produces significant community-level treatment effects and significant natural direct effects (NDEs) in mediation. We interpret this as evidence of shortcut learning: fine-tuning on ideology-labeled data causes the model to internalise a spurious sentiment--ideology coupling not operative in human judgment for this task. This coupling is structurally invisible to F1-based evaluation, with implications for the use of LLM annotations as silver labels and as proxies for human judgment in downstream causal analyses.
Chinese Translation
我们探讨主题情感是否对感知政治意识形态产生因果影响，以及这一答案是否取决于谁赋予意识形态标签。我们使用来自 AllSides 的文章，并配以 Llama-3.3-70b-versatile 的共享情感标注，比较专家人类标注者、GPT-4o-mini（基线和微调版）以及 Llama-3.3-70B 的意识形态标签。我们在所有四种标注范式中应用双重机器学习（Double Machine Learning, DML）和社区层面的中介分析。人类标注在社区层面没有显著的因果效应。微调后的 GPT-4o-mini 达到最高的分类准确率（F1=72.48），并且是唯一产生显著社区层面处理效应和显著自然直接效应（Natural Direct Effects, NDEs）的标注范式。我们将此解释为快捷学习的证据：在意识形态标注数据上进行微调使模型内化了一种在此任务中不适用于人类判断的虚假情感-意识形态耦合。这种耦合在基于 F1 的评估中是结构上不可见的，这对将 LLM 标注作为银标签和在下游因果分析中作为人类判断代理的使用具有重要影响。
View on arXiv Download PDF AI Translation
cs.CL / 12 / 2606.06738

Modular Monolingual Adaptation using Pretrained Language Models

基于预训练语言模型的模块化单语适应
Kumar, Nalin, Dušek, Ondřej
Abstract
Building monolingual language models (LMs) for low-resource languages typically relies on adapting pretrained language models (PLMs) by finetuning the whole model on the target language. This approach is widely favored over training from scratch, as it enables effective knowledge transfer. Additionally, prior work has shown that using a language-specific tokenizer can enhance the adaptability. In this work, we hypothesize that full model tuning is often unnecessary and propose a more modular approach. Specifically, we replace the tokens, freeze the corresponding embeddings, and tune the rest of the model. We use Scottish Gaelic, Irish, and Quechua for our experiments, with Quechua being a very low-resource language (8.5k training instances). Evaluation on natural language understanding (NLU) tasks -- mask filling, NER, and POS -- shows that our proposed approach improves performance when adapting models to low-resource languages. Additionally, we provide a comprehensive analysis of the effectiveness of training strategies, the choice of pretrained embeddings, and models.
Chinese Translation
为低资源语言构建单语语言模型（LMs）通常依赖于通过在目标语言上微调整个模型来适应预训练语言模型（PLMs）。这种方法因其能够有效地进行知识转移而被广泛青睐于从头训练。此外，先前的研究表明，使用特定语言的分词器可以增强适应性。在本研究中，我们假设全模型微调通常是不必要的，并提出了一种更模块化的方法。具体而言，我们替换了标记，冻结了相应的嵌入，并微调了模型的其余部分。我们使用苏格兰盖尔语、爱尔兰语和克丘亚语进行实验，其中克丘亚语是一种非常低资源的语言（8.5k训练实例）。在自然语言理解（NLU）任务——掩码填充、命名实体识别（NER）和词性标注（POS）上的评估表明，我们提出的方法在将模型适应低资源语言时提高了性能。此外，我们提供了对训练策略、预训练嵌入选择和模型有效性的全面分析。
View on arXiv Download PDF AI Translation
cs.CL / 13 / 2606.06745

When to Think Deeply: Inhibitory Deliberation for LLM Reasoning

何时深入思考：大型语言模型推理的抑制性深思
He, Zhixuan, Feng, Yue
Abstract
Reasoning Large Language Models can improve problem-solving performance through deliberative inference, but invoking slow reasoning for every input is computationally expensive and often unnecessary. We propose IDPR, a framework for response-conditioned inhibitory deliberation. IDPR first generates a concise intuitive answer and then uses an inhibition controller to decide whether that specific response should be released or suppressed in favor of slow reasoning. Unlike input-only routers, the inhibition controller conditions on the fast answer and fast-side evidence, including confidence, logit margin, parseability, and generation cost. We train the controller from paired fast-slow outcomes and select the inhibition threshold on a held-out validation set under an accuracy-first slow-call budget. On a held-out 5,000-example mathematical reasoning test set, IDPR invokes slow reasoning on only 8.20% of examples and improves accuracy from 47.90% to 48.92%. Under the same slow-call budget, random routing decreases accuracy to 46.76%, while the strongest confidence-based baseline reaches 48.22%. IDPR also achieves the highest corrective precision, showing that response-conditioned inhibition better identifies fast answers that benefit from slow reasoning.
Chinese Translation
推理大型语言模型可以通过深思熟虑的推理提高问题解决的表现，但对每个输入都调用缓慢的推理在计算上是昂贵的，并且通常是不必要的。我们提出了 IDPR（响应条件抑制性深思框架），该框架首先生成一个简洁的直观答案，然后使用抑制控制器来决定是否应该发布该特定响应，或抑制该响应以支持缓慢推理。与仅基于输入的路由器不同，抑制控制器基于快速答案和快速侧证据（包括置信度、对数边际、可解析性和生成成本）进行条件判断。我们从配对的快速-慢速结果中训练控制器，并在一个保留的验证集上选择抑制阈值，遵循以准确性为首的慢调用预算。在一个保留的 5,000 个示例的数学推理测试集中，IDPR 仅在 8.20% 的示例上调用缓慢推理，并将准确率从 47.90% 提高到 48.92%。在相同的慢调用预算下，随机路由将准确率降低至 46.76%，而最强的基于置信度的基线达到了 48.22%。IDPR 还实现了最高的纠正精度，表明响应条件抑制更好地识别出受益于慢推理的快速答案。
View on arXiv Download PDF AI Translation
cs.CL / 14 / 2606.06748

Evidence Graph Consistency in Retrieval-Augmented Generation: A Model-Dependent Analysis of Hallucination Detection

检索增强生成中的证据图一致性：基于模型的幻觉检测分析
Shen, Jianru
Abstract
Retrieval-Augmented Generation (RAG) reduces but does not eliminate hallucination in large language models. Existing detection methods rely on flat similarity between generated answers and retrieved passages, ignoring structural relationships among evidence pieces and answer claims. We propose Evidence Graph Consistency (EGC), a framework that constructs a local evidence graph per response and computes five structural consistency measures as hallucination indicators. Evaluated on the full question answering split of RAGTruth across six LLMs (5,767 responses), EGC reveals a consistent model-family split: graph consistency features show the expected diagnostic direction for hallucinations in Llama-2 models but exhibit systematic reversal in GPT-4, GPT-3.5, and Mistral-7B. This reversal suggests qualitatively different hallucination patterns across model families and indicates that embedding-based graph consistency cannot serve as a model-independent hallucination detection signal.
Chinese Translation
检索增强生成（RAG）在一定程度上减少了大型语言模型中的幻觉现象，但并未完全消除。现有的检测方法依赖于生成答案与检索段落之间的平面相似性，忽视了证据片段与答案主张之间的结构关系。我们提出了证据图一致性（EGC）框架，该框架为每个响应构建一个局部证据图，并计算五个结构一致性度量作为幻觉指示器。在六个大型语言模型（LLM）上对RAGTruth的完整问答分割进行评估（共5,767个响应），EGC揭示了一个一致的模型家族分裂：图一致性特征在Llama-2模型中显示出预期的幻觉诊断方向，但在GPT-4、GPT-3.5和Mistral-7B中则表现出系统性的反转。这种反转表明不同模型家族之间存在质 qualitatively不同的幻觉模式，并表明基于嵌入的图一致性不能作为模型独立的幻觉检测信号。
View on arXiv Download PDF AI Translation
cs.CL / 15 / 2606.06755

PromptPrint: Behavioral Biometrics Through Natural Language Prompting in LLMs

PromptPrint：通过自然语言提示在大型语言模型中实现行为生物识别
Patel, Shaiv, Narayan, Kartik, Patel, Vishal
Abstract
Authorship attribution research has traditionally focused on long-form, expressive texts; however, interactions with large language models (LLMs) are typically brief and task-driven prompts. This raises a fundamental question: do such prompts contain a stable, author-identifiable, and distinctive signal? We introduce PromptPrint, a systematic study of prompt-based identity, the hypothesis that a user's habitual vocabulary, syntax, and discourse patterns form a learnable behavioral biometric. Using 20,680 real prompts from 1,034 users, we establish three key findings. First, lexical representations significantly outperform semantic encoders, supporting the "lexical stability hypothesis": identity is primarily encoded in surface-level word choice rather than abstract intent. Second, stylometric features exhibit a "uniqueness-consistency paradox": users are highly distinctive across the population, yet behaviorally inconsistent across contexts. Third, adversarial analysis reveals a clear vulnerability spectrum: identity signals are robust to minor lexical perturbations but degrade substantially under semantic paraphrasing. Overall, our results demonstrate strong identification performance at scale, establishing prompt-based identity as a viable behavioral biometric. This work introduces a new perspective on user modeling in LLM interactions, with important implications for security and privacy. Data and code will be released upon the acceptance of our work.
Chinese Translation
作者归属研究传统上集中于长篇、富有表现力的文本；然而，与大型语言模型（LLMs）的交互通常是简短且以任务为驱动的提示。这引发了一个根本性的问题：这些提示是否包含稳定的、可识别作者的独特信号？我们提出了PromptPrint，这是对基于提示的身份的系统研究，假设用户的习惯词汇、语法和话语模式形成了一种可学习的行为生物识别特征。通过分析来自1,034名用户的20,680个真实提示，我们得出了三个关键发现。首先，词汇表示显著优于语义编码器，支持“词汇稳定性假设”：身份主要编码在表层的词汇选择中，而非抽象意图。其次，风格特征表现出“独特性-一致性悖论”：用户在群体中高度独特，但在不同情境下行为不一致。第三，对抗性分析揭示了明确的脆弱性谱：身份信号对小的词汇扰动具有鲁棒性，但在语义改述下显著降低。总体而言，我们的结果在大规模上展示了强大的识别性能，确立了基于提示的身份作为一种可行的行为生物识别特征。这项工作为大型语言模型交互中的用户建模提供了新的视角，对安全性和隐私具有重要意义。数据和代码将在我们工作被接受后发布。
View on arXiv Download PDF AI Translation
cs.CL / 16 / 2606.06758

A Four-Condition Diagnostic Protocol for Evidence Utilization in Long-Context and Retrieval-Augmented Language Models

用于长上下文和检索增强语言模型的四条件证据利用诊断协议
Xia, Haizhou
Abstract
Final-answer accuracy, retrieval recall, and citation overlap do not by themselves identify whether a long-context or retrieval-augmented language model used the evidence it was given. A model can answer from parametric memory, fail despite receiving the right passages, or cite evidence without converting it into the requested answer. This paper proposes a matched four-condition evidence-availability protocol--no evidence, full context, retrieved evidence, and oracle-evidence reference--for diagnosing evidence utilization under fixed examples, prompts, score fields, retrieval settings, and validity checks. ONCU is used as a protocol-bound estimator of recovered oracle-reference evidence advantage and is computed only for denominator-valid groups; denominator-free answer, evidence, retrieval, and failure-audit metrics are reported separately. The empirical study evaluates five local open-weight models from the Qwen, Gemma, Llama, and Mistral families across Controlled-ONCU-safe16K, HotpotQA-ONCU, and 2WikiMultiHopQA-ONCU, with 18,000 ONCU-compatible predictions. The main finding is a task-dependent bottleneck split: controlled synthetic settings primarily expose full-context utilization failures, whereas the tested realistic multi-hop settings primarily expose retrieval-chain coverage failures in denominator-free answer and evidence metrics, with ONCU supporting the same direction on oracle-improving groups. The contribution is a diagnostic protocol for separating no-evidence answerability, oracle-evidence recoverability, full-context utilization, and retrieval-conditioned utilization, rather than a single-score leaderboard for long-context or retrieval-augmented systems.
Chinese Translation
最终答案的准确性、检索召回率和引用重叠率本身并不能确定长上下文或检索增强语言模型是否使用了所提供的证据。一个模型可能从参数记忆中回答问题，尽管接收了正确的段落但仍然失败，或者在没有将证据转化为请求的答案的情况下引用证据。本文提出了一种匹配的四条件证据可用性协议——无证据、完整上下文、检索证据和oracle证据参考——用于在固定示例、提示、评分字段、检索设置和有效性检查下诊断证据利用情况。ONCU被用作恢复的oracle参考证据优势的协议绑定估计量，仅对有效分母组进行计算；不依赖分母的答案、证据、检索和失败审计指标则单独报告。实证研究评估了来自Qwen、Gemma、Llama和Mistral系列的五个本地开放权重模型，涵盖Controlled-ONCU-safe16K、HotpotQA-ONCU和2WikiMultiHopQA-ONCU，共有18,000个ONCU兼容的预测。主要发现是一个任务依赖的瓶颈分裂：受控合成设置主要暴露完整上下文利用的失败，而测试的现实多跳设置主要暴露在不依赖分母的答案和证据指标中检索链覆盖的失败，ONCU在改善oracle的组中支持相同的方向。贡献在于提供了一种诊断协议，用于区分无证据可答性、oracle证据可恢复性、完整上下文利用和检索条件利用，而不是针对长上下文或检索增强系统的单一评分排行榜。
View on arXiv Download PDF AI Translation
cs.CL / 17 / 2606.06781

When Better Codebooks Are Not Enough: Predictive Performance and Behavioral Reliability in LLM Political Event Coding

当更好的编码本不足以满足需求：大型语言模型在政治事件编码中的预测性能与行为可靠性
He, Zixian, Murugesan, Bharath Raahul, Brandt, Patrick, Hu, Yibo
Abstract
High accuracy does not necessarily make an LLM a faithful coder. This issue matters because many social-science studies rely on expert-written codebooks to turn text into structured data. We study this problem in political event coding, a challenging source-target relation classification task beyond ordinary sentence-level classification, where models must determine what one actor did to another using detailed coding rules. We test whether expert codebooks become more effective when operationalized into LLM-friendly forms with clearer definitions, examples, retrieved context, and rules for difficult cases. We then evaluate behavioral reliability under controlled changes to label names, codebook order, and label-definition mappings. Clearer codebooks substantially improve classification performance, especially for fine-grained event classification. However, these predictive gains do not fully translate into behavioral reliability. Models may produce valid labels and recover definitions while still failing behavioral reliability tests under controlled codebook changes. These findings suggest that codebook-guided LLM systems should be evaluated not only by accuracy, but also by whether they preserve the coding logic that makes coded outputs meaningful for social-science research.
Chinese Translation
高准确性并不一定意味着大型语言模型（LLM）是一个忠实的编码者。这个问题很重要，因为许多社会科学研究依赖专家编写的编码本将文本转化为结构化数据。我们研究了这一问题，聚焦于政治事件编码，这是一个超越普通句子级分类的挑战性源-目标关系分类任务，其中模型必须根据详细的编码规则确定一个行为者对另一个行为者的行为。我们测试了当专家编码本被转化为适合LLM的形式，并提供更清晰的定义、示例、检索的上下文和困难案例的规则时，是否会变得更加有效。随后，我们在控制标签名称、编码本顺序和标签定义映射的变化下评估行为可靠性。更清晰的编码本显著提高了分类性能，尤其是在细粒度事件分类方面。然而，这些预测性提升并未完全转化为行为可靠性。在控制编码本变化的情况下，模型可能会生成有效的标签并恢复定义，但仍然未能通过行为可靠性测试。这些发现表明，基于编码本的LLM系统应不仅通过准确性进行评估，还应考虑它们是否保持了使编码输出对社会科学研究有意义的编码逻辑。
View on arXiv Download PDF AI Translation
cs.CL / 18 / 2606.06788

Explain Like I'm 5 or Whatever I Choose: Evaluating the Interactive Potential of Language Model Responses

像我五岁一样解释，或者我选择的任何方式：评估语言模型响应的互动潜力
Panigrahi, Indu, August, Tal
Abstract
Evaluations of large language models (LLMs) in scientific information seeking tasks have become increasingly use-centric, such as conducting live or multi-turn evaluations with real users. These evaluations still assume a single, static chat interface, but as models are integrated into new interfaces, evaluations must shift to incorporate interface-specific criteria. We propose a new evaluation framework based on a formative study with $16$ participants that tests models' ability to generate multiple responses to one query that differ along an interpretable axis of language (language complexity), inspired by direct manipulation interfaces from human-centered design literature. We evaluate GPT-5.1, GPT-5 mini, Claude Sonnet 4.5 + Thinking, and DeepSeek-V3.1 by generating 5 responses at different levels of language complexity for $98$ scientific queries. While models vary complexity across responses, most changes remain inconsistent, with the best performing model (Claude Sonnet 4.5) only shifting reliable complexity measures in the correct direction $46\%$ of the time. Our findings hold with increased sample size and alternative complexity levels.
Chinese Translation
在科学信息检索任务中，对大型语言模型（LLMs）的评估越来越以用户为中心，例如与真实用户进行实时或多轮评估。这些评估仍然假设一个单一的、静态的聊天界面，但随着模型被整合到新的界面中，评估必须转向纳入特定于界面的标准。我们提出了一种新的评估框架，基于对$16$名参与者的形成性研究，测试模型生成多个响应的能力，这些响应在一个可解释的语言轴（语言复杂性）上有所不同，灵感来源于以人为本设计文献中的直接操作界面。我们通过生成$98$个科学查询在不同语言复杂性水平下的5个响应，评估了GPT-5.1、GPT-5 mini、Claude Sonnet 4.5 + Thinking和DeepSeek-V3.1。尽管模型在响应之间的复杂性有所变化，但大多数变化仍然不一致，表现最佳的模型（Claude Sonnet 4.5）仅在正确方向上可靠地改变复杂性度量的频率为$46\%$。我们的发现随着样本量的增加和替代复杂性水平的变化而保持一致。
View on arXiv Download PDF AI Translation
cs.CL / 19 / 2606.06794

TA-RAG: Tone-Aware Retrieval-Augmented Generation for Peer-Support Health Communication

TA-RAG：面向语气的检索增强生成模型用于同伴支持健康沟通
Kang, Yong-Bin, McCosker, Anthony
Abstract
Retrieval-augmented generation (RAG) successfully grounds large language model (LLM) outputs in trusted documents, but factual grounding alone is insufficient for sensitive peer-support health communication. In domains such as HIV peer support, responses must also be accessible, stigma-free, empathetic, and tailored to the recipient. This paper presents TA-RAG, a lightweight, prompt-based tone-aware RAG framework that embeds explicit tone control into a RAG pipeline without requiring model fine-tuning. We operationalise tone across four core components: stigma-free rewriting, readability adjustment, recipient adaptation, and empathy rephrasing. We evaluate TA-RAG through component-level tests using questions derived from HIV Online Learning Australia (HOLA), UNAIDS terminology guidance, readability metrics, peer-support standards from National Association of People with HIV Australia (NAPWHA), and a public empathy dataset. Results show that the TA-RAG's components improve their targeted communication quality while preserving key content. These findings emphasise that prompt-based tone control is a potential direction for making RAG outputs suitable for sensitive peer-support health communication.
Chinese Translation
检索增强生成（RAG）成功地将大型语言模型（LLM）的输出与可信文档相结合，但仅靠事实基础对于敏感的同伴支持健康沟通而言是不够的。在如HIV同伴支持等领域，回应还必须易于理解、无污名、富有同情心，并且针对接收者进行定制。本文提出了TA-RAG，一个轻量级的基于提示的面向语气的RAG框架，它在RAG流程中嵌入了明确的语气控制，而无需对模型进行微调。我们通过四个核心组件来实现语气的操作化：无污名重写、可读性调整、接收者适应和同情心重述。我们通过使用来自HIV在线学习澳大利亚（HOLA）、联合国艾滋病规划署（UNAIDS）术语指导、可读性指标、澳大利亚HIV患者国家协会（NAPWHA）的同伴支持标准以及一个公共同情心数据集的问题进行组件级测试来评估TA-RAG。结果表明，TA-RAG的各个组件在保持关键内容的同时，提高了其目标沟通质量。这些发现强调，基于提示的语气控制是使RAG输出适合敏感同伴支持健康沟通的一个潜在方向。
View on arXiv Download PDF AI Translation
cs.CL / 20 / 2606.06797

Korean Culture into LLM Alignment: Toward Cultural Coherence

将韩国文化融入大型语言模型对齐：迈向文化一致性
Jung, MinJae, Kim, Minwoo
Abstract
Cultural-aspect work on large language models is dominated by a negative target: which outputs to suppress. We argue that a constructive counterpart is also needed, a working definition of what a culturally coherent response is rather than only what it must avoid, and instantiate it for Korean. We design an alignment-data pipeline around a prompt-based LLM seed generator that expands a Korean harm taxonomy, with a Korean-culturally-adapted safe-response policy at its centre: a per-category guideline grounded in Korean legal frameworks, social norms, and interpretive conventions, against which three frontier models each produce a candidate response. DPO fine-tuning on the resulting triplets improves the Korean cultural safe rate across six open-weight LLMs while causing no large degradation on Korean general-capability benchmarks, and qualitative outputs show fine-tuned models naming Korean statutes and institutional procedures and, where appropriate, supplying constructive Korean-context information alongside refusal.
Chinese Translation
关于大型语言模型的文化方面的研究主要集中在一个消极目标上：即应抑制哪些输出。我们认为，还需要一个建设性的对应概念，即什么是文化上连贯的回应的工作定义，而不仅仅是它必须避免的内容，并且我们将其具体化为韩国文化。我们围绕一个基于提示的LLM种子生成器设计了一个对齐数据管道，该生成器扩展了韩国的危害分类法，以一个适应韩国文化的安全响应政策为中心：这是一个基于韩国法律框架、社会规范和解释惯例的每类指导方针，三种前沿模型根据这些指导方针各自生成候选响应。对生成的三元组进行DPO微调，提高了六个开放权重LLM在韩国文化安全率的表现，同时对韩国通用能力基准没有造成显著下降，定性输出显示微调后的模型能够引用韩国法规和机构程序，并在适当的情况下，提供建设性的韩国背景信息与拒绝响应并存。
View on arXiv Download PDF AI Translation
cs.CL / 21 / 2606.06812

Quantifying Media Representation Dynamics Across 25 Years of News Reporting on Policing-related Deaths

量化媒体在25年新闻报道中对与警务相关死亡的表现动态
Samir, Farhan, Dhillon, Jappun, Ravikumar, Meghna, Ahmed, Syed Ishtiaque, Shwartz, Vered
Abstract
We perform the largest known computational analysis of Canadian news narratives about police-involved deaths, spanning 4,000 articles from the last quarter-century. We develop a novel computational model, PerspectiveGap, grounded in prior sociological work on media representation of policing. We find that reporting on police-involved deaths on average features perspectives from state bureaucrats at a rate nearly three times as much as perspectives from other members of the public, including relatives, community members, eyewitnesses, lawyers representing the family, or civil liberties groups. A considerable fraction of articles contain no points of view from civilian actors, though civilian representation has increased in recent years. Qualitatively, we find that state bureaucrats' accounts of these deaths tend to be clinical and procedural, while civilian discourse carries considerably more emotional valence. The PerspectiveGap framework developed here can be contextualized to other jurisdictions, offering a scalable approach for analyzing how media systems construct narratives around policing and accountability.
Chinese Translation
我们对加拿大新闻叙事中关于警察涉入死亡事件进行了已知规模最大的计算分析，涵盖了过去25年间的4000篇文章。我们开发了一种新颖的计算模型，PerspectiveGap，基于先前关于媒体对警务表现的社会学研究。我们的研究发现，关于警察涉入死亡的报道中，来自国家官僚的视角平均出现的频率几乎是其他公众成员（包括亲属、社区成员、目击者、代表家庭的律师或公民自由团体）视角的三倍。相当一部分文章没有来自平民参与者的观点，尽管近年来平民的表现有所增加。从定性分析来看，我们发现国家官僚对这些死亡事件的叙述往往显得临床和程序化，而平民的论述则带有更多的情感色彩。这里开发的PerspectiveGap框架可以在其他司法管辖区进行背景化，为分析媒体系统如何构建关于警务和问责的叙事提供了一种可扩展的方法。
View on arXiv Download PDF AI Translation
cs.CL / 22 / 2606.06825

Progress-SQL: Improving Reinforcement Learning for Text-to-SQL via Progressive Rewards

Progress-SQL：通过渐进奖励提升文本到SQL的强化学习
Zhang, Shihao, Wang, Xiaoman, Liu, Yuan, Lan, Yunshi, Qian, Weining
Abstract
Reinforcement learning has recently shown promise in improving large language models for Text-to-SQL generation, yet existing methods typically optimize one-shot rewards defined over a single SQL state. Such rewards provide limited guidance for iterative SQL correction and are insufficient to capture the improvement of multi-turn SQL refinement. In this paper, we propose Progress-SQL, a multi-turn reinforcement learning framework with progressive rewards for Text-to-SQL. Our approach introduces an Oracle-guided Diagnostic Tree (ODT), which abstracts SQL queries into clause-level structural profiles and produces diagnostic feedback for next-turn refinement. To provide dense and robust reward signals, we combine ODT-based structural alignment with lexical alignment and define a progressive reward that measures the improvement from the initial SQL to the final SQL. We further incorporate a progression latency reward that favors earlier correctness and an execution status reward that encourages recovery from the invalid SQL. Experiments on BIRD, Spider, and Spider robustness variants demonstrate that our method consistently improves Text-to-SQL performance across both primary and robustness evaluations.
Chinese Translation
强化学习最近在提升大型语言模型用于文本到SQL生成方面显示出了潜力，但现有方法通常优化定义在单一SQL状态上的一次性奖励。这类奖励对迭代SQL修正提供的指导有限，且不足以捕捉多轮SQL细化的改进。本文提出了Progress-SQL，一种具有渐进奖励的多轮强化学习框架，用于文本到SQL。我们的方法引入了一个由Oracle指导的诊断树（Oracle-guided Diagnostic Tree, ODT），将SQL查询抽象为子句级结构特征，并为下一轮的细化提供诊断反馈。为了提供密集且稳健的奖励信号，我们结合了基于ODT的结构对齐与词汇对齐，并定义了一种渐进奖励，衡量从初始SQL到最终SQL的改进。此外，我们还结合了一种进展延迟奖励，偏向于早期的正确性，以及一种执行状态奖励，鼓励从无效SQL中恢复。对BIRD、Spider及其稳健性变体的实验表明，我们的方法在主要评估和稳健性评估中均持续提升了文本到SQL的性能。
View on arXiv Download PDF AI Translation
cs.CL / 23 / 2606.06834

The Dark Regulome: Disentangling Predictability from Regulation in Genomic Foundation Models

黑暗调控组：从基因组基础模型中理清可预测性与调控之间的关系
Baranwal, Chahat, Baranwal, Aadtya, Tandon, Lakshya Nitin
Abstract
High-grade gliomas integrate into neural circuits through functional synapses with neurons, raising the question of which noncoding elements shape synaptogenic gene expression in tumor cells. The regulatory program written across the dark genome, what we call the $\textit{dark regulome}$, is the natural substrate to probe, and sequence foundation models offer a zero-shot route through in-silico mutagenesis (ISM); yet likelihood-based scoring is tautologically coupled to local sequence predictability, leaving the regulatory interpretation underdetermined. Across three architecturally distinct foundation models (Caduceus-Ph, HyenaDNA, Enformer) and 30,448 dark genome elements at 92 glioma-relevant loci, we introduce a residualization-and-permutation diagnostic that separates predictability-driven from regulation-driven RIS variance. A sharp 10kb proximal-regulatory horizon survives every control we apply, but the LM-derived element-class hierarchy does not: a six-feature linear baseline matches Caduceus top-decile membership at AUC $= 0.985$. Cross-architecture decomposition cleanly separates a sequence-predictability layer (the two language models co-rank long well-predicted transposable elements) from a regulatory-output layer (Enformer alone retains residual cCRE-discriminative signal), with literally zero overlap between the two top-100 lists. Conservation, brain cis-eQTL, and STRING-PPI cross-checks then anchor what biology survives: top-100 elements across all three models are $3.3\times$ enriched per model for matching brain eQTLs ($p_\mathrm{emp} < 5\times 10^{-3}$), while a tempting transposable-element regulatory layer and a striking NRXN1+NLGN1 protein-pair convergence both fail proper permutation tests once those tests are constructed. We deliver the diagnostic as a general methodological tool for any ISM-based regulatory study.
Chinese Translation
高等级胶质瘤通过与神经元的功能性突触融入神经回路，这引发了一个问题：哪些非编码元素塑造了肿瘤细胞中的突触生成基因表达。我们称之为 $ extit{dark regulome}$ 的黑暗基因组中的调控程序是探测的自然基础，而序列基础模型通过计算机突变（in-silico mutagenesis, ISM）提供了一种零样本的途径；然而，基于似然的评分与局部序列可预测性是同义的，导致调控的解释不确定。在三个结构上不同的基础模型（Caduceus-Ph、HyenaDNA、Enformer）和92个与胶质瘤相关的位点上的30,448个黑暗基因组元素中，我们引入了一种残差化和置换诊断，区分了由可预测性驱动的与由调控驱动的RIS方差。一个明确的10kb近端调控视野在我们应用的每一个控制下都存活下来，但基于语言模型的元素类别层次结构却没有：一个六特征线性基线在AUC $= 0.985$时匹配Caduceus的前十分之一成员。跨架构分解干净地将序列可预测性层（这两个语言模型共同排名长的高可预测性转座子）与调控输出层（仅Enformer保留残余的cCRE区分信号）分开，且两者的前100列表之间几乎没有重叠。保护性、脑内cis-eQTL和STRING-PPI的交叉检查则锚定了存活下来的生物学：在所有三个模型中，前100个元素在每个模型中对于匹配脑eQTL的富集度为$3.3 imes$（$p_ ext{emp} < 5 imes 10^{-3}$），而一个诱人的转座子调控层和一个显著的NRXN1+NLGN1蛋白对的汇聚在构建适当的置换测试后均未能通过测试。我们将该诊断作为任何基于ISM的调控研究的一种通用方法工具交付。
View on arXiv Download PDF AI Translation
cs.CL / 24 / 2606.06835

Translate-R1: Cost-Aware Translation Tool Use via Reinforcement Learning

Translate-R1：基于强化学习的成本感知翻译工具使用
Jayarao, Pratik, Dwivedi, Chaitanya, Gupta, Himanshu, Varshney, Neeraj, Devraj, Adithya M, Vadera, Meet, Nigam, Priyanka, Yin, Bing
Abstract
The performance gap across languages in LLMs is well documented, and closing it natively requires pretraining or fine-tuning on corpora that, for most languages, do not exist. Translation offers an alternative: converting an input into the model's dominant language unlocks its full capabilities at once. Applying translation to every input, however, is wasteful for languages the model already handles, while leaving the choice to the model fails in the opposite way, as LLMs are overconfident and skip the tool even when they cannot understand the input. Prior work resolves this with language-specific rules, domain heuristics, language identifiers, or external routers, each requiring manual engineering. We instead learn a single policy that decides when to translate from reward alone, developing language- and domain-adaptive introspection that assesses its own comprehension and invokes translation only when it cannot solve a task natively. Using data built by our answer-preserving translation pipeline, we continue RL on the post-trained Qwen3-4B across 22 languages in 3 resource tiers (High, Low, XLow) and 5 domains, and introduce confidence-gated GSPO for cost-sensitive tool use. The gated policy lifts reward over the baseline by +4.6 on High, +23.5 on Low, and +17.5 on XLow. Against an unconstrained policy that almost always translates, it preserves full reward at 63% of the cost and is Pareto-optimal across 87% of the cost-sensitivity range. Additionally, to simulate behavior on a completely unseen language, we create 2 synthetic languages, where our gated policy improves +18.7 over the overconfident baseline that underutilizes the tool even on these incomprehensible inputs. The policy transfers zero-shot to 9 held-out languages, and we analyze how tool use emerges over training, per language and per domain.
Chinese Translation
大型语言模型（LLMs）在不同语言之间的性能差距已得到充分记录，而要在本地弥补这一差距，通常需要在大多数语言上进行预训练或微调，但这些语料库往往并不存在。翻译提供了一种替代方案：将输入转换为模型的主导语言可以立即释放其全部能力。然而，对于模型已经能够处理的语言，应用翻译于每个输入是浪费的，而将选择权留给模型则会出现相反的问题，因为LLMs往往过于自信，即使无法理解输入也会跳过工具。以往的研究通过语言特定规则、领域启发式、语言识别器或外部路由器来解决这个问题，但这些方法都需要手动工程。我们则学习了一种单一策略，仅根据奖励决定何时进行翻译，开发出语言和领域自适应的内省机制，评估自身的理解能力，仅在无法本地解决任务时调用翻译。利用我们保留答案的翻译管道构建的数据，我们在22种语言、3个资源层级（高、中、低）和5个领域上继续对后训练的Qwen3-4B进行强化学习，并引入基于置信度的GSPO以实现成本敏感的工具使用。该门控策略在高资源层级上将奖励提高了+4.6，在低资源层级上提高了+23.5，在极低资源层级上提高了+17.5。与几乎总是进行翻译的无约束策略相比，它在63%的成本下保留了全部奖励，并在87%的成本敏感范围内是帕累托最优的。此外，为了模拟在完全未见语言上的行为，我们创建了2种合成语言，其中我们的门控策略在这些无法理解的输入上相较于过于自信的基线提高了+18.7，该基线在这些输入上未能充分利用工具。该策略在9种保留语言上实现了零样本迁移，我们分析了工具使用如何在训练过程中按语言和领域逐步出现。
View on arXiv Download PDF AI Translation
cs.CL / 25 / 2606.06840

Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces

先特征化再提炼：大输出空间中的机制推理
Roy, Debjyoti Saha, Wallace, Byron C., Aslam, Javed A.
Abstract
Modern reasoning models offer surprisingly strong zero-shot performance on challenging multi-label tasks that require selecting a small set of relevant options from hundreds of thousands to millions of candidate labels. We investigate how they achieve this mechanistically. We characterize reasoning as a two-phase process: A broad "shortlisting" of candidates followed by fine-grained reasoning over the resulting set. We provide evidence across a range of datasets that these steps can be isolated and are complementary. Using this characterization, we develop a mechanistic distillation strategy that consistently outperforms standard distillation.
Chinese Translation
现代推理模型在需要从数十万到数百万个候选标签中选择一小组相关选项的挑战性多标签任务上，展现出令人惊讶的零样本性能。我们探讨它们是如何在机制上实现这一点的。我们将推理特征化为一个两阶段的过程：首先是对候选者的广泛“初选”，然后是对结果集的细致推理。我们在一系列数据集上提供证据，表明这些步骤可以被孤立并且是互补的。基于这种特征化，我们开发了一种机制提炼策略，始终优于标准提炼。
View on arXiv Download PDF AI Translation
cs.CL / 26 / 2606.06842

CRAFT: A Unified Counterfactual Reasoning Framework for Tabular Question Answering and Fact Verification

CRAFT：一个统一的反事实推理框架用于表格问答和事实验证
Pan, Chenshuo, Zhao, Yu, Zhang, Jie, Pan, Changzai, Wu, Zhenhe, Liang, Jiayi, Mao, Yujie, Song, Shuangyong, Li, Yongxiang, He, Zhongjiang
Abstract
Table reasoning remains challenging for large language models (LLMs), particularly in tasks that require multi-step inference over long and structured tables. Existing approaches predominantly rely on single-direction reasoning, which limits their ability to explore alternative hypotheses across tasks. In this work, we propose CRAFT, a unified Counterfactual Reasoning Framework that reformulates Tabular question answering and fact verification into a general bidirectional verification process. Our method explicitly constructs both declarative statements and their counterfactual variants. Evidence is then extracted from reasoning along both the original and counterfactual paths, and integrated via a weighted mechanism to arrive at the final answer. Experimental results show that our approach consistently surpasses representative baselines on table reasoning datasets such as WikiTQ and TabFact, achieving especially large improvements on complex question answering. Our framework also significantly mitigates performance gaps between different backbone LLMs. This indicates that counterfactual reasoning effectively overcomes the limitations of single-direction inference, guiding LLMs toward more discerning reasoning and establishing a more principled paradigm for structured reasoning tasks. Our code will be made publicly available upon acceptance.
Chinese Translation
表格推理对于大型语言模型（LLMs）仍然具有挑战性，特别是在需要对长且结构化的表格进行多步推理的任务中。现有的方法主要依赖单向推理，这限制了它们在任务中探索替代假设的能力。在本研究中，我们提出了CRAFT，一个统一的反事实推理框架，将表格问答和事实验证重新构造成一个通用的双向验证过程。我们的方法明确构建了陈述和其反事实变体。然后，从原始路径和反事实路径的推理中提取证据，并通过加权机制整合以得出最终答案。实验结果表明，我们的方法在表格推理数据集（如WikiTQ和TabFact）上始终超越代表性基线，在复杂问答上取得了特别大的改善。我们的框架还显著减小了不同主干LLMs之间的性能差距。这表明反事实推理有效克服了单向推理的局限，引导LLMs朝向更具辨别力的推理，并为结构化推理任务建立了一个更有原则的范式。我们的代码将在接受后公开。
View on arXiv Download PDF AI Translation
cs.CL / 27 / 2606.06857

Interpreting Brain Responses to Language with Sparse Features from Language Models

利用语言模型的稀疏特征解读大脑对语言的反应
Lepori, Michael A., Kay, Kendrick, Tuckute, Greta
Abstract
A central goal of cognitive neuroscience is to characterize the features that are represented by human language cortex. Artificial language models (LMs) have emerged as a powerful tool to address this challenge, but studies relating biological and artificial representations are often criticized as relating one black box to another. The present work introduces Augmented Sparse Encoding Models, an encoding framework that replaces dense LM hidden states with hierarchically-organized sparse autoencoder (SAE) features, while explicitly including surprisal as a predictor. Using this approach, we (i) produce interpretations of neural responses and (ii) test whether model-brain alignment reflects primary or idiosyncratic variation in LM representations. Using a high-field 7T fMRI dataset of eight participants listening to 200 linguistically diverse sentences, we first validate our modeling framework by recovering previous interpretations of voxel populations tuned to processing difficulty and meaning abstractness. We then interpret a previously-uncharacterized (but reliable) voxel population and find that it is tuned to people-related content. Next, we show that the fronto-temporal human language network is predicted by a common set of features across its constituent regions, but find that frontal regions are relatively well-explained by surprisal alone, even in the absence of LM-based features. Finally, we show that brain responses during language processing are not merely predictable from an arbitrary set of LM features. Rather, brain responses are best explained by the features that tend to capture the most general information encoded in LM representations, suggesting a nontrivial correspondence between brain and LM language representation.
Chinese Translation
认知神经科学的一个核心目标是表征人类语言皮层所代表的特征。人工语言模型（LMs）已成为应对这一挑战的强大工具，但将生物和人工表征联系起来的研究常常被批评为将一个黑箱与另一个黑箱相联系。本研究介绍了增强稀疏编码模型（Augmented Sparse Encoding Models），该编码框架用分层组织的稀疏自编码器（SAE）特征替代了密集的LM隐藏状态，同时明确将惊讶度作为预测因子。通过这种方法，我们（i）生成神经反应的解释，并（ii）测试模型与大脑的对齐是否反映了LM表征中的主要或特异性变异。使用高场7T fMRI数据集，八名参与者聆听200个语言多样性的句子，我们首先通过恢复先前对调谐于处理难度和意义抽象性的体素群体的解释来验证我们的建模框架。然后，我们解释了一个先前未被表征（但可靠）的体素群体，并发现它调谐于与人相关的内容。接下来，我们展示了额颞人类语言网络是由其组成区域的共同特征集所预测的，但发现额叶区域相对较好地被惊讶度单独解释，即使在缺乏基于LM的特征的情况下。最后，我们表明，在语言处理过程中，大脑反应并不仅仅是可以从任意一组LM特征中预测的。相反，大脑反应最好是通过那些倾向于捕捉LM表征中编码的最一般信息的特征来解释，这表明大脑与LM语言表征之间存在非平凡的对应关系。
View on arXiv Download PDF AI Translation
cs.CL / 28 / 2606.06865

Are Large Language Models Suitable for Graph Computation? Progress and Prospects

大型语言模型适合图计算吗？进展与展望
Zhang, Yuting, Han, Yi, Wang, Kai, Ni, Wei, Bonifati, Angela, Zhang, Wenjie
Abstract
Large language models (LLMs) have been increasingly explored for graph computation, where tasks require reasoning over structured relationships and algorithmic operations. Yet, it remains unclear when LLMs can reliably support such computation and how they should be incorporated into graph-solving pipelines. Existing surveys at the intersection of LLMs and graphs primarily focus on graph learning, text-attributed graphs, or graph-language modeling. To bridge this gap, we provide a comprehensive review of LLMs for graph computation through a role-based taxonomy. Specifically, we identify two major paradigms: i) LLMs as executors, where models directly solve graph tasks from graph descriptions and instructions; and ii) LLMs as planners, where models formulate problems, decompose reasoning steps, and invoke external tools or agents for execution. Based on this taxonomy, we analyze the strengths and limitations of current methods. Our review indicates that LLMs are promising for simple, small-scale tasks, but remain unreliable for large-scale and exactness-demanding tasks. Finally, we summarize available datasets and suggest four future directions.
Chinese Translation
大型语言模型（LLMs）在图计算领域的探索日益增多，该领域的任务需要对结构化关系和算法操作进行推理。然而，目前尚不清楚LLMs何时能够可靠地支持此类计算，以及它们应如何融入图解决方案流程。现有的关于LLMs与图的交叉研究主要集中在图学习、文本属性图或图语言建模。为了填补这一空白，我们通过基于角色的分类法对LLMs在图计算中的应用进行了全面回顾。具体而言，我们识别出两种主要范式：i）LLMs作为执行者，模型直接根据图描述和指令解决图任务；ii）LLMs作为规划者，模型制定问题，分解推理步骤，并调用外部工具或代理进行执行。基于这一分类法，我们分析了当前方法的优缺点。我们的回顾表明，LLMs在简单的小规模任务中表现出色，但在大规模和对精确性要求高的任务中仍然不可靠。最后，我们总结了现有的数据集，并提出了四个未来研究方向。
View on arXiv Download PDF AI Translation
cs.CL / 29 / 2606.06879

An Expanded Synthetic Conversation Dataset for Multi-Turn Smishing Detection

用于多轮短信钓鱼检测的扩展合成对话数据集
Lochstampfor, Carl, Roy, Ayan
Abstract
Our prior work introduced COVA, a synthetically generated multi-turn conversational smishing dataset of 3,201 labeled conversations, establishing baseline detection benchmarks across eight models. While XGBoost with TF-IDF features achieved the best performance, with 72.5\% accuracy and 0.691 macro F1, transformer models underperformed, which was attributed to input truncation and insufficient training data. We present COVA-X, an expanded dataset of 10,985 conversations spanning eight elder-targeted scam categories, produced by an improved generation pipeline addressing contamination, label mismatch, stage-direction bleed, and prompt-design failures from the first iteration. Retraining all classifiers on the expanded dataset yields the central finding of this work: Longformer now surpasses XGBoost on all evaluation metrics, achieving 79.71\% accuracy and 0.7786 macro F1 compared with 78.43\% and 0.7563 for XGBoost. This directly confirms that transformer models require larger conversational corpora to realize their contextual advantages. We additionally document a quality life-cycle including a 12.7$\times$ improvement in label correction rate, from 49.8\% to 3.9\%, an architectural intervention reducing virtual-kidnapping artifact rates from 67.1\% to 46.5\%, and a per-scam-type outcome analysis showing that scam categories modulate results in mechanism-consistent ways. A pre/post-cleanup sensitivity analysis confirms that dataset refinement recovers genuine label-relevant signal across all three classifier architectures.
Chinese Translation
我们之前的工作介绍了COVA，这是一个合成生成的多轮对话短信钓鱼数据集，包含3,201个标记对话，为八个模型建立了基线检测基准。尽管使用TF-IDF特征的XGBoost模型表现最佳，达到了72.5\%的准确率和0.691的宏F1值，但变换器模型的表现不佳，这归因于输入截断和训练数据不足。我们提出了COVA-X，这是一个扩展的数据集，包含10,985个对话，涵盖八个针对老年人的诈骗类别，采用改进的生成流程，解决了第一轮中的污染、标签不匹配、阶段指导干扰和提示设计失败等问题。对所有分类器在扩展数据集上重新训练得出了本工作的核心发现：Longformer在所有评估指标上现在超越了XGBoost，准确率达到79.71\%，宏F1值为0.7786，而XGBoost的准确率为78.43\\%，宏F1值为0.7563。这直接证实了变换器模型需要更大的对话语料库来实现其上下文优势。我们还记录了一个质量生命周期，包括标签修正率的12.7倍改善，从49.8\\%降至3.9\\%，以及一种架构干预将虚拟绑架伪影率从67.1\\%降低到46.5\\%，并且每种诈骗类型的结果分析显示诈骗类别以机制一致的方式调节结果。前后清理的敏感性分析确认，数据集的精炼在所有三种分类器架构中恢复了真实的标签相关信号。
View on arXiv Download PDF AI Translation
cs.CL / 30 / 2606.06906

EASE-TTT: Evidence-Aligned Selective Test-Time Training for Long-Context Question Answering

EASE-TTT：证据对齐的选择性测试时间训练用于长上下文问答
Yuan, Xiaopeng, Wang, Zebin, Wang, Suwen, Yang, Zongxin, Wang, Haohan, Dong, Yushun
Abstract
Long-context question answering (QA) remains challenging for smaller language models even when answer-bearing evidence is already present in the input. Existing within-context retrieval methods localize and expose candidate evidence chunks for the question, but they stop at input-level evidence exposure rather than adapting the query-side attention parameters that control how the model allocates attention over full-context positions. In contrast, lightweight test-time adaptation methods, such as query-only test-time training (qTTT), leave evidence localization unresolved because their generic span-level self-supervised objectives do not identify which context positions support the current answer. In this paper, we propose Evidence-Aligned SElective Test-Time Training (EASE-TTT), a within-context retrieval-augmented test-time training framework that converts selected evidence chunks into a soft attention supervision target over their token positions. Instead of replacing the full context with retrieved chunks, EASE-TTT uses the resulting attention target to guide query-side adaptation, with the adapted model generating the final answer from the original full context. Experiments on six LongBench QA tasks and three small decoder-only language models show that EASE-TTT achieves the strongest macro-average performance among full-context inference, retrieval-only baselines, and qTTT, supporting evidence-aligned test-time adaptation in long-context QA.
Chinese Translation
长上下文问答（QA）对于较小的语言模型仍然具有挑战性，即使在输入中已经存在答案相关的证据。现有的上下文内检索方法定位并暴露了问题的候选证据块，但它们仅停留在输入级别的证据暴露，而未能调整控制模型在全上下文位置分配注意力的查询侧注意力参数。相比之下，轻量级的测试时间适应方法，如仅查询的测试时间训练（qTTT），未能解决证据定位问题，因为它们的通用跨度级自监督目标未能识别哪些上下文位置支持当前答案。在本文中，我们提出了证据对齐的选择性测试时间训练（EASE-TTT），这是一种增强上下文内检索的测试时间训练框架，旨在将选定的证据块转换为其标记位置上的软注意力监督目标。EASE-TTT并不通过检索块替换完整上下文，而是利用生成的注意力目标来指导查询侧的适应，使得适应后的模型从原始完整上下文中生成最终答案。在六个LongBench QA任务和三个小型解码器语言模型上的实验表明，EASE-TTT在全上下文推理、仅检索基线和qTTT中实现了最强的宏平均性能，支持长上下文问答中的证据对齐测试时间适应。
View on arXiv Download PDF AI Translation
cs.CL / 31 / 2606.06915

ThinkBooster: A Unified Framework for Seamless Test-Time Scaling of LLM Reasoning

ThinkBooster：无缝测试时间计算扩展的大型语言模型推理统一框架
Smirnov, Vladislav, Nguyen, Chieu, Senichev, Sergey, Ta, Minh Ngoc, Fadeeva, Ekaterina, Vazhentsev, Artem, Galimzianova, Daria, Rozanov, Nikolai, Mazanov, Viktor, Ni, Jingwei, Wu, Tianyi, Kiselev, Igor, Sachan, Mrinmaya, Gurevych, Iryna, Nakov, Preslav, Baldwin, Timothy, Shelmanov, Artem
Abstract
Test-time compute (TTC) scaling has emerged as a powerful paradigm for improving large language model (LLM) reasoning by allocating additional compute during inference, e.g., via multi-sample generation and verifier-based reranking. Existing TTC scaling strategies and reasoning scorers remain fragmented, evaluated under inconsistent protocols, and are rarely analyzed through the lens of quality-cost trade-offs. We introduce ThinkBooster, a unified framework for seamless test-time compute scaling of LLM reasoning, which consists of (i) a modular Python library implementing state-of-the-art TTC scaling strategy and scorer families, (ii) a benchmark that jointly evaluates performance and computational efficiency, and (iii) a deployable OpenAI-compatible proxy service that enables drop-in integration of adaptive reasoning into real-world applications. We further provide a demo visual debugger for inspecting the reasoning trajectories, intermediate selection decisions, and alternative reasoning paths. Empirical results on mathematical and coding tasks reveal the performance-compute trade-offs of TTC scaling strategies and scoring methods and demonstrate that ThinkBooster provides practical gains in real-world tasks. The code is available online under an MIT license.
Chinese Translation
测试时间计算（TTC）扩展已成为一种强大的范式，通过在推理过程中分配额外的计算资源来提升大型语言模型（LLM）的推理能力，例如，通过多样本生成和基于验证者的重新排序。现有的TTC扩展策略和推理评分器仍然存在碎片化的问题，评估标准不一致，且很少从质量-成本权衡的角度进行分析。我们提出了ThinkBooster，这是一个无缝的LLM推理测试时间计算扩展统一框架，包含（i）一个模块化的Python库，实施最先进的TTC扩展策略和评分器系列，（ii）一个共同评估性能和计算效率的基准，以及（iii）一个可部署的与OpenAI兼容的代理服务，支持自适应推理在现实应用中的无缝集成。我们还提供了一个演示可视化调试器，用于检查推理轨迹、中间选择决策和替代推理路径。在数学和编码任务上的实证结果揭示了TTC扩展策略和评分方法的性能-计算权衡，并展示了ThinkBooster在现实任务中提供的实际收益。代码已在线发布，采用MIT许可证。
View on arXiv Download PDF AI Translation
cs.CL / 32 / 2606.06942

Didact: A Cross-Domain Capability Discovery System for Defence

Didact：一个跨领域的能力发现系统用于国防
Bodhankar, Aarya, Joshi, Aditya, Doan, Bao Gia, Marchant, Thomas, Leslie, Oscar, Salim, Flora
Abstract
Policymakers in defence and defence-aligned sectors must monitor rapidly evolving research alongside sector priorities relevant to operational and strategic needs. In practice, these sources are fragmented across heterogeneous formats, disjoint repositories, and siloed update streams, making capability discovery slow and difficult to audit. We present Didact, a prototype that integrates publicly available defence reports and policy documents from Australia with a purpose-built knowledge graph derived from Australian research publications. Didact provides natural language conversations for policy-oriented workflows, and leverages a composite retrieval-augmented generation (RAG) pipeline. A key feature of Didact is an interactive Evidence Rail that visualises retrieved evidence and source relationships. Our evaluation of the output quality and runtime of Didact highlights its utility. While Didact has been co-developed as an academia-industry project for the Australian context, it is adaptable to other domains where knowledge is similarly fragmented. A demonstration video is available here:
Chinese Translation
国防及相关领域的决策者必须监测快速发展的研究成果，以及与操作和战略需求相关的行业优先事项。在实践中，这些信息源分散在异构格式、孤立的存储库和独立的更新流中，使得能力发现变得缓慢且难以审计。我们提出了Didact，一个原型系统，它整合了来自澳大利亚的公开国防报告和政策文件，以及基于澳大利亚研究出版物构建的专用知识图谱。Didact为政策导向的工作流程提供自然语言对话，并利用复合检索增强生成（RAG）管道。Didact的一个关键特性是交互式证据轨道，它可视化检索到的证据及其来源关系。我们对Didact的输出质量和运行时间的评估突显了其实用性。虽然Didact作为一个学术与产业合作项目在澳大利亚背景下共同开发，但它可以适应其他知识同样分散的领域。演示视频可在此观看：
View on arXiv Download PDF AI Translation
cs.CL / 33 / 2606.06946

Auditing Training Data in Domain-adapted LLMs: LoRA-MINT

审计领域适应的大型语言模型中的训练数据：LoRA-MINT
Mancera, Gonzalo, DeAlcala, Daniel, Morales, Aythami, Fierrez, Julian, Tolosana, Ruben, Jurado, Francisco
Abstract
We present LoRA-MINT, a new methodology for Membership Inference Test (MINT) applied to recent Large Language Models (LLMs) fine-tuned for specific Natural Language Processing (NLP) tasks through Low-Rank Adaptation (LoRA). The primary goal is to assess whether individual samples were part of the training data of these adapted models, providing a useful auditing tool for the management of intellectual property and sensitive data. Our analysis explores the relationship between model perplexity and membership status, providing a systematic framework for estimating data exposure in fine-tuned LLMs. We conducted experiments on four models and three benchmark datasets, obtaining precision values in determining if given data were used for training ranging from 0.77 to 0.92, which outperform state-of-the-art baselines and demonstrate the robustness and generality of the proposed method. In general, our findings underscore the potential of LoRA-MINT as an effective and scalable framework for auditing LLMs, improving transparency, and fostering the ethical and responsible deployment of AI and NLP technologies. For the sake of concreteness and current relevance, our discussion and experiments are centered on LoRAadjusted LLMs, but note that most of the presented methodology is easily applicable for auditing training data given any other technique for adapting LLMs or, more generally, any other domain-adapted AI models.
Chinese Translation
我们提出了LoRA-MINT，这是一种新的成员推断测试（Membership Inference Test, MINT）方法，应用于最近通过低秩适应（Low-Rank Adaptation, LoRA）微调的特定自然语言处理（Natural Language Processing, NLP）任务的大型语言模型（Large Language Models, LLMs）。其主要目标是评估个别样本是否属于这些适应模型的训练数据，为知识产权和敏感数据的管理提供有用的审计工具。我们的分析探讨了模型困惑度与成员身份状态之间的关系，提供了一个系统框架来估计微调LLMs中的数据暴露。我们在四个模型和三个基准数据集上进行了实验，获得了在判断给定数据是否用于训练方面的精确度值，范围从0.77到0.92，超越了最先进的基线，展示了所提方法的鲁棒性和普适性。总体而言，我们的研究结果强调了LoRA-MINT作为审计LLMs的有效且可扩展框架的潜力，提高了透明度，并促进了人工智能和自然语言处理技术的伦理和负责任的部署。为了确保具体性和当前相关性，我们的讨论和实验集中在LoRA调整的LLMs上，但请注意，所提出的方法大部分也可以轻松应用于审计任何其他适应LLMs或更一般的任何其他领域适应的人工智能模型的训练数据。
View on arXiv Download PDF AI Translation
cs.CL / 34 / 2606.06959

OpenHalDet: A Unified Benchmark for Hallucination Detection across Diverse Generation Scenarios

OpenHalDet：一个统一的基准，用于在多样化生成场景中进行幻觉检测
Li, Xinyi, Fang, Zhen, Deng, Yongxin, Luo, Jinyuan, Ma, Hongnan, Oh, Changdae, Shi, Zijing, Ye, Shanshan, Wang, Hanchen, Chen, Shu-Lin, Luo, Yadan, Yang, Mengyue, Du, Sean, Li, Sharon, Chen, Ling
Abstract
Hallucination detection is essential for the reliable deployment of large language models (LLMs). However, existing evaluations face two core challenges: inconsistent inference configuration and evaluation, and limited coverage of downstream domains and tasks. Consequently, reported detector performance is often difficult to compare, reproduce, and generalize beyond specific experimental settings. We introduce OpenHalDet, a unified benchmark for hallucination detection across diverse generation scenarios. OpenHalDet standardizes the evaluation pipeline, from prompt construction and response generation to truthfulness annotation, detector scoring, and metric computation. It supports heterogeneous detector families under different access settings, including black-box methods that use only generated outputs, gray-box methods that rely on probability-based signals, and white-box methods that exploit internal model signals. By bringing diverse tasks, models, and detectors into a shared framework, OpenHalDet enables controlled comparison and provides a systematic view of how different detection paradigms behave in LLM applications. We release OpenHalDet as an open and extensible codebase to facilitate reproducible evaluation and future development of hallucination detection methods. The code and datasets are available at https://github.com/Nellie179/Hallucination-Detection.
Chinese Translation
幻觉检测对于大型语言模型（LLMs）的可靠部署至关重要。然而，现有的评估面临两个核心挑战：推理配置和评估的不一致性，以及下游领域和任务的覆盖范围有限。因此，报告的检测器性能往往难以比较、重现，并且难以在特定实验设置之外进行推广。我们介绍了OpenHalDet，这是一个用于在多样化生成场景中进行幻觉检测的统一基准。OpenHalDet标准化了评估流程，从提示构建、响应生成到真实性标注、检测器评分和指标计算。它支持在不同访问设置下的异构检测器家族，包括仅使用生成输出的黑箱方法、依赖于基于概率信号的灰箱方法，以及利用内部模型信号的白箱方法。通过将多样化的任务、模型和检测器纳入共享框架，OpenHalDet实现了受控比较，并提供了不同检测范式在LLM应用中的行为的系统视角。我们将OpenHalDet作为一个开放且可扩展的代码库发布，以促进可重现的评估和未来幻觉检测方法的发展。代码和数据集可在 https://github.com/Nellie179/Hallucination-Detection 获取。
View on arXiv Download PDF AI Translation
cs.CL / 35 / 2606.06960

Tree-of-Experience: A Structured Experience-Management Solution for Self-Evolving Agents under Low-Repetition and Implicit-Reward Environments

经验树：一种针对低重复和隐性奖励环境下自我进化代理的结构化经验管理解决方案
Deng, Zihao, Zhu, Yining, Wang, Leiming, Lu, Jingfei, Wang, Junbo, Ran, Chuncheng, Yang, Yu, Yang, Dixuan, Shen, Jikun
Abstract
Experience-based self-evolution is crucial for LLM agents, but existing benchmarks often assume explicit goals, stable task patterns, and clear feedback. We study a more challenging setting: low-repetition tasks with implicit rewards, where past experience is difficult to reuse and feedback is delayed, noisy, and outcome-level. We introduce \textsc{FinEvolveBench}, a temporally controlled benchmark for financial sentiment prediction that links daily news-driven predictions to future excess returns. We further propose Tree-of-Experience (ToE), a structured experience-management method that organizes, retrieves, validates, and updates agent experience. Experiments show that general-purpose experience mechanisms do not consistently outperform no-experience baselines, while ToE achieves stronger overall performance. These results highlight the importance of structured experience management for self-evolving agents in implicit-reward environments.
Chinese Translation
基于经验的自我进化对于大型语言模型（LLM）代理至关重要，但现有基准通常假设明确的目标、稳定的任务模式和清晰的反馈。我们研究了一种更具挑战性的环境：低重复任务和隐性奖励，其中过去的经验难以重用，反馈延迟、噪声且为结果级别。我们引入了 extsc{FinEvolveBench}，一个时间控制的金融情感预测基准，将每日新闻驱动的预测与未来超额收益联系起来。我们进一步提出了经验树（Tree-of-Experience, ToE），一种结构化的经验管理方法，能够组织、检索、验证和更新代理的经验。实验表明，通用的经验机制并未始终优于无经验基线，而 ToE 实现了更强的整体性能。这些结果强调了在隐性奖励环境中，结构化经验管理对于自我进化代理的重要性。
View on arXiv Download PDF AI Translation
cs.CL / 36 / 2606.06985

Contrastive Training with LLM-generated Near-Misses for Robust Code-Switching Speech Recognition

基于LLM生成近似错误的对比训练用于鲁棒的代码切换语音识别
Nguyen, Tung X., Truong, Hieu Minh, Nguyen, Giang-Son, Vo, Nhu, Buntine, Wray, Le, Dung D.
Abstract
Code-switching (CS), the alternation between multiple languages within a single utterance, remains challenging for Automatic Speech Recognition (ASR). To address this issue, we propose a Point-of-Interest (POI)-aware contrastive training framework that improves recognition at CS-critical regions. We first identify CS spans by adopting POI detection method from literature, then construct acoustically plausible near-miss hypotheses by perturbing POIs in ASR N-best outputs and expanding candidates with a large language model. Hard but plausible negatives are retained through filtering with acoustic, phonemic, and textual constraints. Finally, we fine-tune Whisper-small with LoRA using a POI-weighted cross-entropy anchor objective together with a multi-negative contrastive ranking loss. Experiments on CS-FLEURS (cmn-eng) and ViMedCSS (vie-eng) show consistent reductions of over 2% in both general and CS-aware error rates compared to standard LoRA fine-tuning.
Chinese Translation
代码切换（CS），即在单一话语中交替使用多种语言，仍然是自动语音识别（ASR）面临的挑战。为了解决这一问题，我们提出了一种关注兴趣点（POI）的对比训练框架，旨在改善在CS关键区域的识别效果。我们首先通过采用文献中的POI检测方法识别CS区间，然后通过扰动ASR N-best输出中的POI并利用大型语言模型扩展候选项，构建声学上合理的近似错误假设。通过声学、音素和文本约束进行过滤，保留了困难但合理的负样本。最后，我们使用POI加权的交叉熵锚目标和多负对比排名损失，结合LoRA对Whisper-small进行微调。在CS-FLEURS（cmn-eng）和ViMedCSS（vie-eng）上的实验表明，与标准的LoRA微调相比，通用和CS感知错误率均一致降低超过2%。
View on arXiv Download PDF AI Translation
cs.CL / 37 / 2606.06994

Principles of Concept Representation in Sentence Encoders

句子编码器中的概念表示原理
Mohr, Isabelle, Dujany, John, Souquet, Jonathan, Freitas, Andre
Abstract
What makes a sentence encoder produce good concept representations? We approach this through the lens of representational compositionality: an encoder supports a concept family only when its latent space admits a low-distortion realization of the corresponding semantic operator. This framing predicts both where current encoders succeed and where they are structurally mismatched to their supervision. Through a controlled ablation over encoder conditions trained on 3.3 million synonym and definition pairs from WordNet and Wiktionary, evaluated on three decontaminated splits and a modifier-labeled noun-phrase benchmark, we identify four principles. Fine-tuning recalibrates the latent geometry rather than expanding it (P1). Semantic signal concentrates in the final transformer layer before concept-specific training begins, making cross-layer pooling redundant (P2). Hard negatives improve discrimination and stress-test robustness without improving retrieval ranking, showing that calibration and ranking are independently addressable (P3). Finally, the effectiveness of supervision depends on the composition type of the target concept. Extensional training helps intersective and subsective families while degrading relational and intensional ones, exposing a structural limitation of current training paradigms (P4). We release two new evaluation datasets: a DBpedia semantic-gap benchmark and a modifier-labeled NP paraphrase suite.
Chinese Translation
是什么使得句子编码器能够生成良好的概念表示？我们通过表征组合性的视角来探讨这一问题：当编码器的潜在空间能够低失真地实现相应的语义运算符时，它才支持一个概念家族。这一框架预测了当前编码器成功的地方以及与其监督结构不匹配的地方。通过对在来自WordNet和Wiktionary的330万同义词和定义对上训练的编码器条件进行控制性消融，并在三个去污染的拆分和一个修饰符标记的名词短语基准上进行评估，我们识别出了四个原理。微调重新校准潜在几何结构，而不是扩展它（P1）。在概念特定训练开始之前，语义信号集中在最终的变换器层，使得跨层池化变得多余（P2）。困难负样本提高了区分度并对鲁棒性进行了压力测试，而并未改善检索排名，显示出校准和排名是可以独立解决的（P3）。最后，监督的有效性取决于目标概念的组合类型。扩展训练有助于交集和子集家族，而对关系和意向家族则有负面影响，暴露了当前训练范式的结构限制（P4）。我们发布了两个新的评估数据集：一个DBpedia语义差距基准和一个修饰符标记的名词短语释义套件。
View on arXiv Download PDF AI Translation
cs.CL / 38 / 2606.07020

MADE: Beyond Scoring via a Multilingual Agentic Diagnosing Engine for Fine-Grained Evaluation Insights

MADE：通过多语言智能诊断引擎实现更细致的评估洞察
Liu, Yilun, Zhang, Miao, Tao, Shimin, He, Minggui, Zhao, Chunguang, Liu, Chenxin, Zhang, Li, Liu, Chen, Qian, Cheng, Deng, Liqun, Meng, Xiaojun, Wei, Daimeng
Abstract
Multilingual and multicultural benchmarks now cover dozens of languages and model families, but the resulting score landscapes remain metric-rich and insight-poor, necessitating fine-grained multilingual post-evaluation diagnosis. However, single LLMs and open-ended agents are easily swamped by the long, noisy diagnostic input, and no reusable taxonomy exists for it. To address this, we propose MADE, a Multilingual Agentic Diagnosing Engine that decomposes post-evaluation analysis into planning, aggregate analysis, instance-level case inspection, multilingual and cultural reflection, and grounded report synthesis. MADE is paired with an expert-led 54-query and 15-language diagnostic set, evaluated on top of a large-scale multilingual evaluation substrate (33 model families, 11 benchmarks, 26 languages, 34 cultures, 8.66M evaluation records). Experiments show that MADE outperforms the strongest shared baseline by 47% in diagnosis report quality and is preferred by human multilingual experts in 87.9% of pairwise comparisons. Applied with multilingual experts, MADE further surfaces four actionable findings on deployment, iteration, and cross-cultural pitfalls, turning benchmark score tables into model-selection and remediation guidance.
Chinese Translation
多语言和多文化基准现在涵盖数十种语言和模型家族，但由此产生的评分景观虽然指标丰富，却缺乏深入的洞察，迫切需要细致的多语言后评估诊断。然而，单一的大型语言模型（LLM）和开放式代理容易被冗长且嘈杂的诊断输入淹没，且目前没有可重用的分类法来处理这些输入。为了解决这一问题，我们提出了MADE，一个多语言智能诊断引擎，它将后评估分析分解为规划、汇总分析、实例级案例检查、多语言和文化反思以及基于事实的报告综合。MADE配备了一个由专家主导的54个查询和15种语言的诊断集，并在一个大规模多语言评估基础上进行评估（涵盖33个模型家族、11个基准、26种语言、34种文化和8.66百万条评估记录）。实验表明，MADE在诊断报告质量上比最强共享基线高出47%，并在87.9%的成对比较中被人类多语言专家所偏好。在与多语言专家的应用中，MADE进一步提出了关于部署、迭代和跨文化陷阱的四个可操作发现，将基准评分表转化为模型选择和补救指导。
View on arXiv Download PDF AI Translation
cs.CL / 39 / 2606.07040

Beyond Rubrics: Exploration-Guided Evaluation Skills for Reward Modeling

超越评分标准：面向探索的奖励建模评估技能
Yue, Xing, Wu, Linjuan, Zhang, Daoxin, Shen, Yongliang, Lu, Weiming
Abstract
Open-ended reward modeling requires judges that can follow subtle, domain-specific preferences when verifiable answers are unavailable. Existing rubric-based methods often address this by generating criteria online for each query, but the extra generation step can add inference overhead and produce rigid or misaligned guidance. We introduce Eval-Skill, an exploration-guided method that synthesizes reusable evaluation skills for reward modeling and reframes reward guidance as context evolution rather than parameter training or per-query rubric generation. Using only 100 cases per domain for skill evolution, Eval-Skill synthesizes reusable domain-level evaluation skills through two progressive stages, workflow generation followed by principle generation, with exploration and selection interleaved across both stages. Once generated, a skill is directly injected into the judge context. Across multiple RM benchmarks, Eval-Skill consistently improves diverse judge backbones; on RewardBench 2, it yields significant gains over vanilla judging for each main backbone (+13.44% for Qwen3-8B, and 18.51% for DeepSeek-V4-Flash). Further analyses of evolution-time scaling, generalizability, and transferability show that compact evaluation skills offer an efficient new paradigm for LLM-based evaluation. Code is available at https://github.com/xing-stellus-yue/Eval-Skill.
Chinese Translation
开放式奖励建模需要能够在不可验证答案缺失时，遵循微妙的领域特定偏好的评审者。现有的基于评分标准的方法通常通过为每个查询在线生成标准来解决这个问题，但额外的生成步骤可能会增加推理开销，并产生僵化或不一致的指导。我们提出了Eval-Skill，这是一种面向探索的方法，合成可重用的奖励建模评估技能，并将奖励指导重新框架为上下文演变，而不是参数训练或每查询评分标准生成。仅使用每个领域的100个案例进行技能演变，Eval-Skill通过两个渐进阶段合成可重用的领域级评估技能，首先是工作流生成，其次是原则生成，同时在两个阶段之间交替进行探索和选择。一旦生成，技能将直接注入评审者上下文。在多个RM基准测试中，Eval-Skill始终改善多样化的评审者基础模型；在RewardBench 2上，它在每个主要基础模型上相较于原始评审方法实现了显著提升（Qwen3-8B提升13.44%，DeepSeek-V4-Flash提升18.51%）。对演变时间缩放、可推广性和可转移性的进一步分析表明，紧凑的评估技能为基于LLM的评估提供了一种高效的新范式。代码可在https://github.com/xing-stellus-yue/Eval-Skill获取。
View on arXiv Download PDF AI Translation
cs.CL / 40 / 2606.07054

TRACE: Trajectory Reasoning through Adaptive Cross-Step Evidence Aggregation for LLM Agents

TRACE：通过自适应跨步骤证据聚合进行轨迹推理的LLM代理
Mittapalli, Vijitha, Dani, Shreyaa Jayant, Pilli, Satya Srujana, Ansu, Snigdha, Teymoorianfard, Mohammadreza, Dernoncourt, Franck, Chen, Hongjie, Wang, Yu, Rossi, Ryan A., Ahmed, Nesreen K.
Abstract
Autonomous LLM agents can pursue hidden malicious objectives through sequences of individually benign actions, making sabotage difficult to detect using standard trajectory-level monitoring. Existing approaches either evaluate complete trajectories in a single pass or partition them into independently scored windows, limiting their ability to connect evidence across temporally distant actions. We propose TRACE, a monitoring framework for long-horizon LLM agent trajectories. TRACE operates through a TIJ (Triage-Inspect-Judge) loop that identifies high-signal regions, performs targeted inspection while maintaining accumulated evidence across reasoning steps, and synthesizes a trajectory-level verdict. We evaluate TRACE on ten task domains from SHADE-Arena against state-of-the-art baselines. TRACE achieves an aggregate F1 of 0.713 and recall of 0.844, with the largest gains on tasks requiring long-range evidence linking.
Chinese Translation
自主的LLM代理可以通过一系列单独无害的行动追求隐藏的恶意目标，使得通过标准轨迹级监控检测破坏行为变得困难。现有方法要么在单次处理过程中评估完整轨迹，要么将其划分为独立评分的窗口，从而限制了它们在时间上遥远的行动之间连接证据的能力。我们提出了TRACE，这是一个用于长时间跨度LLM代理轨迹的监控框架。TRACE通过TIJ（分类-检查-判断）循环运作，识别高信号区域，进行有针对性的检查，同时在推理步骤中保持累积证据，并综合出轨迹级的裁决。我们在SHADE-Arena的十个任务领域上评估了TRACE，结果与最先进的基线进行比较。TRACE的综合F1得分为0.713，召回率为0.844，在需要长距离证据链接的任务上取得了最大的提升。
View on arXiv Download PDF AI Translation
cs.CL / 41 / 2606.07066

Modeling semantic association in self-paced reading with language model embeddings

利用语言模型嵌入对自我节奏阅读中的语义关联进行建模
Østergaard, Sara Møller, Enevoldsen, Kenneth, Alishahi, Afra, Nicenboim, Bruno
Abstract
Semantic association between a word and its context has been identified as an important component of reading comprehension, even when word predictability is accounted for. Recent research has highlighted the potential of language model ( LM) embeddings to quantify semantic association. Yet, embedding-based semantic association have been operationalized in a myriad of ways. In this study, we use embeddings from LMs to estimate semantic association on a corpus of joint electroencephalography (EEG) and self-paced reading of natural, Dutch texts. Semantic association is calculated in ten different implementations that vary the embedding model and context lengths. The effects of semantic association across the different implementations on the N400 and self-paced reading times are examined using Bayesian hierarchical models and Bayes factor. The results show that the choice of embedding model can alter the estimated effect of semantic association on both the N400 and self-paced reading times. Furthermore, the results demonstrate a promising potential of sentence embeddings for capturing semantic association, as only implementations relying on sentence embeddings indicate reliable results of semantic association beyond word predictability on both neural and behavioral measures. Together, these findings highlight the importance of methodological choices in quantifying semantic association.
Chinese Translation
词汇与其上下文之间的语义关联被认为是阅读理解的重要组成部分，即使在考虑词汇可预测性的情况下也是如此。近期研究强调了语言模型（LM）嵌入在量化语义关联方面的潜力。然而，基于嵌入的语义关联在操作化时采用了多种方式。在本研究中，我们使用语言模型的嵌入来估计自然荷兰文本的联合脑电图（EEG）和自我节奏阅读语料库中的语义关联。语义关联通过十种不同的实现方式进行计算，这些实现方式在嵌入模型和上下文长度上有所不同。我们使用贝叶斯层次模型和贝叶斯因子来检验不同实现方式下语义关联对N400和自我节奏阅读时间的影响。结果表明，嵌入模型的选择可以改变语义关联对N400和自我节奏阅读时间的估计效果。此外，结果还展示了句子嵌入在捕捉语义关联方面的良好潜力，因为仅依赖句子嵌入的实现方式在神经和行为测量上都表明了超越词汇可预测性的可靠语义关联结果。综合来看，这些发现突显了在量化语义关联时方法选择的重要性。
View on arXiv Download PDF AI Translation
cs.CL / 42 / 2606.07069

mmPISA-bench: Do LLMs Reason Equally Well Across 43 Languages?

mmPISA-bench：大型语言模型在43种语言中的推理能力是否相同？
Sapenov, Yerzhan, Savelka, Jaromir
Abstract
We introduce mmPISA-bench, a compact high-quality multilingual reasoning benchmark derived from the OECD Programme for International Student Assessment (PISA). The benchmark consists of 25 multiple-choice questions that require reasoning in order to be answered correctly. Each question is provided in official human translations to 43 languages and complemented with machine-translated counterparts (i.e., 2,150 data points in total). We evaluate two mainstream proprietary LLMs across languages, reasoning effort levels, and translation types in terms of their ability to answer the questions correctly. Our results show that modern LLMs can reason effectively across all evaluated languages, achieve accuracy comparable to human test-takers, with some performance variations across covered languages. We further find that machine-translated questions do not degrade accuracy relative to official human translations which suggests that high-quality machine translation (synthetic data) might often be adequate for large-scale multilingual reasoning evaluations where official translations are not available. Finally, we analyze token usage and related inference cost and find that LLMs usage in some languages is simultaneously more expensive and less accurate.
Chinese Translation
我们介绍了mmPISA-bench，这是一个基于经济合作与发展组织（OECD）国际学生评估项目（PISA）衍生的紧凑型高质量多语言推理基准。该基准包含25个多项选择题，这些问题需要推理才能正确回答。每个问题都提供了官方的人类翻译，涵盖43种语言，并配以机器翻译的对应版本（即总共2,150个数据点）。我们评估了两种主流的专有大型语言模型（LLMs）在不同语言、推理努力水平和翻译类型下正确回答问题的能力。我们的结果表明，现代大型语言模型能够在所有评估的语言中有效推理，其准确性可与人类考生相媲美，但在不同语言之间存在一定的性能差异。我们进一步发现，机器翻译的问题在准确性上并不逊色于官方的人类翻译，这表明高质量的机器翻译（合成数据）在官方翻译不可用的情况下，可能足以用于大规模多语言推理评估。最后，我们分析了标记使用情况及相关推理成本，发现某些语言中大型语言模型的使用既更昂贵又准确性较低。
View on arXiv Download PDF AI Translation
cs.CL / 43 / 2606.07098

SigmaScale: LLM Compression with SVD-based Low-Rank Decomposition and Learned Scaling Matrices

SigmaScale：基于SVD的低秩分解与学习缩放矩阵的LLM压缩
Lavrinovics, Ernests, Letizia, Marco, Janco, Roy, Segal, Shai, Bjerva, Johannes, Pierini, Maurizio
Abstract
We present SigmaScale, a method for learning auxiliary scaling matrices $S$ to aid truncated Singular Value Decomposition (SVD) based Large Language Model (LLM) compression. Instead of deriving scaling matrices analytically, SigmaScale optimizes two sets of vectors that define diagonal row and column scaling transformations under an activation-aware compression loss. We show that learned scaling lowers the effective intrinsic rank of weight matrices, as reflected by reductions in effective-rank entropy, and that this reduction is strongly correlated with compression loss. Experiments on Llama 3.1 8B Instruct and Qwen3-8B show that SigmaScale is competitive with closely related state-of-the-art SVD-based compression methods across perplexity and zero-shot benchmarks. By using learned activation-aware transformations, SigmaScale explores a more flexible route to low-rank LLM compression by adapting to the structure of individual model weights. The advantage observed in specific tasks makes our approach a valid option for applications requiring a reduced LLM-inference computing cost.
Chinese Translation
我们提出了SigmaScale，一种学习辅助缩放矩阵$S$的方法，以帮助基于截断奇异值分解（SVD）的大型语言模型（LLM）压缩。SigmaScale并不是通过解析推导缩放矩阵，而是优化两组向量，这些向量在激活感知压缩损失下定义了对角行和列缩放变换。我们展示了学习到的缩放降低了权重矩阵的有效内在秩，这一点通过有效秩熵的减少得以体现，并且这种减少与压缩损失之间存在强相关性。在Llama 3.1 8B Instruct和Qwen3-8B上的实验表明，SigmaScale在困惑度和零-shot基准测试中与密切相关的最先进的基于SVD的压缩方法具有竞争力。通过使用学习到的激活感知变换，SigmaScale探索了一条更灵活的低秩LLM压缩路径，适应个别模型权重的结构。在特定任务中观察到的优势使我们的方法成为需要降低LLM推理计算成本的应用的有效选择。
View on arXiv Download PDF AI Translation
cs.CL / 44 / 2606.07103

Style or Content? Evaluating Style Classifiers with Controlled Content Overlap

风格还是内容？使用控制内容重叠评估风格分类器
Liu, Zhuo, Du, Haozheng, Xu, Xiangxiang, He, Hangfeng
Abstract
Style classifiers can use content cues that correlate with style labels in naturally collected data, yet we lack a systematic way to measure this reliance. We study this problem with a controlled content overlap setup built on parallel Bible translations. Specifically, we define the overlap parameter $\alpha$ as the normalized residual of mutual information between content identity and style label, so that it measures how much content is shared across style classes: from no shared content ($\alpha=0$) to fully shared content ($\alpha=1$). Cross-overlap evaluation of RoBERTa-based classifiers shows that low-overlap models degrade when content cues are removed, while high-overlap models transfer more robustly. A cross-style content retrieval probe further shows that content becomes less recoverable as $\alpha$ increases, with training dynamics showing this removal occurs gradually. Together, these results suggest that controlled overlap provides a simple diagnostic for separating style learning from content shortcuts.
Chinese Translation
风格分类器可以利用与风格标签相关的内容线索，这些线索来自自然收集的数据，但我们缺乏系统的方法来衡量这种依赖关系。我们通过基于平行圣经翻译的控制内容重叠设置来研究这个问题。具体而言，我们将重叠参数 $eta$ 定义为内容身份与风格标签之间互信息的标准化残差，从而测量不同风格类别之间共享的内容量：从没有共享内容（$eta=0$）到完全共享内容（$eta=1$）。对基于 RoBERTa 的分类器进行的交叉重叠评估表明，当移除内容线索时，低重叠模型的性能下降，而高重叠模型则表现出更强的迁移能力。跨风格内容检索探测进一步表明，随着 $eta$ 的增加，内容的可恢复性降低，训练动态显示这种移除是逐渐发生的。综合来看，这些结果表明，控制重叠提供了一种简单的诊断方法，用于区分风格学习与内容捷径。
View on arXiv Download PDF AI Translation
cs.CL / 45 / 2606.07123

Learning Perspectivist Social Meaning via Demographic-Conditioned Fusion Embeddings

通过人口条件融合嵌入学习视角主义社会意义
Curry, Amanda Cercas, La Cava, Lucio, Aiello, Luca Maria, Morales, Gianmarco De Francisci
Abstract
Social meaning in language is inherently perspectival, varying across annotator backgrounds, demographics, and ideological positions. However, most NLP systems collapse this variation into a single ground-truth label, ignoring the diversity of interpretations. In this work, we model social dimensions along a perspectivist spectrum, capturing how interpretations vary across demographic groups on a dataset consisting of 28k human annotations. We benchmark multiple modeling paradigms, including zero-shot, few-shot, and fine-tuned approaches, and propose fusion embeddings that integrate textual and demographic representations. Our fusion models yield consistent and statistically significant improvements over text-only baselines across all fusion strategies (+5.9-6.5% relative macro PR-AUC), with shuffle ablations confirming that demographic profiles carry genuine predictive signal rather than spurious correlations.
Chinese Translation
语言中的社会意义本质上是视角性的，因注释者的背景、人口统计特征和意识形态立场而异。然而，大多数自然语言处理（NLP）系统将这种变化简化为单一的真实标签，忽视了解释的多样性。在本研究中，我们沿着视角主义光谱对社会维度进行建模，捕捉不同人口群体在一个包含28,000条人类注释的数据集上的解释差异。我们基准测试了多种建模范式，包括零样本、少样本和微调方法，并提出了融合嵌入，整合文本和人口统计表示。我们的融合模型在所有融合策略上相较于仅使用文本的基线模型均实现了一致且统计显著的改进（相对宏观PR-AUC提高5.9-6.5%），通过洗牌消融实验确认人口特征携带真实的预测信号，而非虚假的相关性。
View on arXiv Download PDF AI Translation
cs.CL / 46 / 2606.07130

Explicit Evidence Grounding via Structured Inline Citation Generation

通过结构化内联引用生成的显式证据基础
Yeginbergen, Anar, Wührl, Amelie, Rogers, Anna, Agerri, Rodrigo
Abstract
As AI systems become more widely adopted, the demand for factual and faithful generation grows. Properly attributing information through citations becomes, therefore, crucial. This work introduces FullCite, a framework that, in contrast to most previous works, generates structured inline citations linking each claim to both its source document and supporting evidence. FullCite proposes three strategies to inline citation generation: prompt-based generation, constrained decoding over a citation grammar, and posthoc span alignment. Using three question answering benchmarks, namely, ASQA, BioASQ, and ExpertQA, we assess citation quality and faithfulness along three dimensions: document-level correctness, evidence span identification, and claim-citation faithfulness. Our evaluation shows that while LLMs are generally effective at identifying relevant documents, they struggle to identify the precise supporting spans within them. This gap suggests that achieving faithful attributed QA will require research to place greater emphasis on precise evidence span identification.
Chinese Translation
随着人工智能系统的广泛应用，对事实和真实生成的需求日益增长。因此，通过引用正确归属信息变得至关重要。本研究介绍了FullCite，一个框架，与大多数先前的工作不同，它生成结构化的内联引用，将每个主张链接到其源文档和支持证据。FullCite提出了三种内联引用生成策略：基于提示的生成、对引用语法的约束解码和事后跨度对齐。通过三个问答基准，即ASQA、BioASQ和ExpertQA，我们从三个维度评估引用的质量和真实性：文档级正确性、证据跨度识别和主张-引用的真实性。我们的评估表明，尽管大型语言模型（LLMs）在识别相关文档方面通常有效，但它们在识别文档内精确的支持跨度方面存在困难。这一差距表明，实现真实归属的问答需要更多的研究，强调精确的证据跨度识别。
View on arXiv Download PDF AI Translation
cs.CL / 47 / 2606.07167

UrduMMLU: A Massive Multitask Benchmark for Urdu Language Understanding

UrduMMLU：乌尔都语理解的大规模多任务基准
Tabassum, Ahmer, Ahmad, Sarfraz, Iqbal, Hasan, Aijaz, Owais, Ahsan, Momina, Nakov, Preslav
Abstract
Meaningful multilingual evaluation must test models in the target language and educational context. Urdu, spoken by more than 230 million people, lacks a broad MMLU-style benchmark built from native educational sources. We introduce UrduMMLU, a benchmark of 26,431 Urdu MCQs across 26 subjects and five domains, collected from native Urdu MCQ banks and public examination PDFs. Unlike translation-based resources, UrduMMLU covers both standard academic subjects and Urdu- and region-specific content. We label the exam-derived portion through dual human annotation with strict consensus filtering. We evaluate 30 LLMs under English and Urdu prompts, yielding 60 zero-shot evaluations, and further evaluate four open-source LLMs under multiple few-shot settings across both prompt languages. Gemini-3.5-Flash performs best, reaching 90.20% and 90.34% accuracy, while no other model exceeds 85%. The strongest open-source model trails by 7.79 and 8.92 points, and many models lose 25 to 40 points on Urdu-centered Humanities subjects compared with STEM. Few-shot prompting yields only modest gains. UrduMMLU shows that Urdu knowledge remains uneven in current LLMs, especially for regionally grounded content.
Chinese Translation
有意义的多语言评估必须在目标语言和教育背景下测试模型。乌尔都语是超过2.3亿人使用的语言，但缺乏基于本土教育资源构建的广泛MMLU风格基准。我们推出了UrduMMLU，这是一个包含26,431个乌尔都语选择题（MCQs）的基准，涵盖26个学科和五个领域，数据来源于本土乌尔都语选择题库和公共考试PDF。与基于翻译的资源不同，UrduMMLU同时涵盖标准学术科目和乌尔都语及区域特定内容。我们通过双重人工标注和严格共识过滤对考试衍生部分进行标注。我们在英语和乌尔都语提示下评估了30个大型语言模型（LLMs），共进行60次零样本评估，并在两种提示语言下进一步评估了四个开源LLMs在多种少样本设置下的表现。Gemini-3.5-Flash表现最佳，分别达到90.20%和90.34%的准确率，而其他模型均未超过85%。最强的开源模型落后7.79和8.92个百分点，许多模型在以乌尔都语为中心的人文学科上相比于STEM领域损失了25到40分。少样本提示仅带来了适度的提升。UrduMMLU显示当前LLMs在乌尔都语知识方面仍然不均衡，尤其是在区域性内容方面。
View on arXiv Download PDF AI Translation
cs.CL / 48 / 2606.07183

Geometry of Semantic Space: Comparative Study of Discrete and Continuous Models

语义空间的几何：离散模型与连续模型的比较研究
Bounias, Gabriel, Ploux, Sabine
Abstract
This work examines the semantic geometry underlying NLP models. We compare supervised vector embeddings, such as CamemBERT, with lexical co-occurrence graphs that encode semantic relations more directly. While transformer-based embeddings achieve strong performance, their induced geometries often display unsatisfactory distributions. In contrast, graph-based models reveal a clearer and more human-readable organization of meaning. We have implemented a methodology that allows us to perform a comparative analysis either based on the structure of the graphs or based on the topology of the embeddings induced by these two approaches. The results of the comparison -- applied to the French "Great National Debate" corpus a collection of citizen contributions to the public debate -- show a similar local topology but a very different overall structure and topology. Theses findings suggest complementary perspectives between deep supervised models and graph-based models, considering a new pathway to guide neural architectures toward more stable and interpretable convergence with graphs structures.
Chinese Translation
本研究探讨了自然语言处理模型背后的语义几何。我们比较了监督向量嵌入（如 CamemBERT）与更直接编码语义关系的词汇共现图。尽管基于变换器的嵌入在性能上表现强劲，但其诱导的几何结构往往显示出不令人满意的分布。相比之下，基于图的模型展现了更清晰且更易于人类理解的意义组织。我们实施了一种方法论，使我们能够根据图的结构或这两种方法诱导的嵌入的拓扑进行比较分析。比较结果——应用于法国“伟大的全国辩论”语料库，该语料库收集了公民对公共辩论的贡献——显示出相似的局部拓扑，但整体结构和拓扑却截然不同。这些发现表明深度监督模型与基于图的模型之间存在互补的视角，并考虑了一条新的路径，以引导神经架构朝向与图结构更稳定且可解释的收敛。
View on arXiv Download PDF AI Translation
cs.CL / 49 / 2606.07190

From Correctness to Utility: Gain-Based Prefix Evaluation for LLM Reasoning

从正确性到效用：基于增益的前缀评估用于大语言模型推理
Zhou, Yuhang, Cao, Yixin, Ye, Guangnan
Abstract
Reasoning prefixes shape the future trajectory of LLM problem solving, yet existing process reward models usually evaluate them through local step correctness. We argue that correctness is a useful but indirect proxy for the effect we ultimately care about: whether a prefix increases the probability of successful completion. We define this effect as prefix gain, the solve-rate improvement induced by conditioning lightweight student model group on a prefix, and use it to train a Prefix Utility Model (PUM) with a simple pairwise ranking objective. PUM learns outcome-grounded prefix utility and can score both complete trajectories and partial reasoning prefixes. Across Best-of-$N$ selection, beam search, and reinforcement learning on mathematical reasoning, PUM provides a strong prefix-level supervision signal, especially when candidate pools are large, search budgets increase, or rule-based rewards are sparse. We release all data, models, and code at https://zhiqix.github.io/pum-project-page.
Chinese Translation
推理前缀塑造了大语言模型（LLM）问题解决的未来轨迹，但现有的过程奖励模型通常通过局部步骤的正确性来评估它们。我们认为，正确性是一个有用但间接的代理，最终我们关心的是：一个前缀是否增加了成功完成的概率。我们将这种效果定义为前缀增益，即通过在前缀上对轻量级学生模型组进行条件化所引起的解题率提升，并利用它来训练一个前缀效用模型（Prefix Utility Model, PUM），其目标是简单的成对排名。PUM 学习基于结果的前缀效用，并能够对完整轨迹和部分推理前缀进行评分。在最佳的 $N$ 选择、束搜索和数学推理的强化学习中，PUM 提供了强有力的前缀级监督信号，尤其是在候选池较大、搜索预算增加或基于规则的奖励稀疏的情况下。我们将在 https://zhiqix.github.io/pum-project-page 发布所有数据、模型和代码。
View on arXiv Download PDF AI Translation
cs.CL / 50 / 2606.07219

Adversarial Creation and Detection of AI-Generated Social Bot Content

对抗性生成与检测AI生成社交机器人内容
Trokhymovych, Mykola, Baeza-Yates, Ricardo, Flammini, Alessandro, Saez-Trumper, Diego, Menczer, Filippo
Abstract
The convergence of large language models and social bots allows malicious actors to manipulate the information ecosystem by generating human-like content at scale. Existing models for detecting AI-generated content often fail in the wild, primarily due to the lack of ground-truth data. We address this gap through an adversarial methodology that models the impersonation of real social media users by malicious actors. Using this methodology, we curate a multilingual, cross-platform dataset of paired human and AI-generated messages. Training on such adversarial data yields accurate detection of AI-generated text. Our approach significantly outperforms existing models for content-based bot detection in real-world, out-of-distribution data.
Chinese Translation
大型语言模型与社交机器人之间的融合使恶意行为者能够通过大规模生成类人内容来操控信息生态系统。现有的AI生成内容检测模型在实际应用中往往失效，主要原因在于缺乏真实的标注数据。我们通过一种对抗性方法来填补这一空白，该方法模拟恶意行为者对真实社交媒体用户的冒充。利用该方法，我们策划了一个多语言、跨平台的配对人类与AI生成消息的数据集。在这样的对抗性数据上进行训练能够准确检测AI生成的文本。我们的方法在现实世界的分布外数据中显著优于现有的基于内容的机器人检测模型。
View on arXiv Download PDF AI Translation
cs.CL / 51 / 2606.07237

When Large Language Models Fail in Healthcare: Evaluating Sensitivity to Prompt Variations

大型语言模型在医疗领域的失败：评估对提示变动的敏感性
Alkaeed, Mahdi
Abstract
Large Language Models (LLMs) are increasingly used in healthcare for tasks such as clinical question answering, diagnosis support, and report summarization. Despite their promise, these models remain highly sensitive to subtle prompt perturbations, both lexical and syntactic, posing serious risks in safety-critical clinical applications. In this study, we conduct a systematic sensitivity analysis to evaluate the robustness of both general-purpose (e.g., GPT-3.5, Llama3) and medical-specific LLMs (e.g., ClinicalBERT, BioLlama3, BioBERT) using the MedMCQA benchmark. We categorize perturbations into natural and adversarial types and examine their effect on model consistency, accuracy, and reliability in clinical reasoning tasks. Our findings reveal that medical LLMs are not intrinsically safe. Even minor variations in phrasing can alter clinical advice, and targeted adversarial prompts can provoke harmful outputs. In high-stakes settings like healthcare, such unpredictability is unacceptable-models that change diagnoses due to reworded inputs or hallucinate medications when slightly rephrased cannot be reliably trusted by clinicians. While models tend to show resilience to simple lexical substitutions or paraphrasing, they often break down under syntactic reordering or misleading contextual cues. This fragility is evident across both general-purpose and domain-specific LLMs. Notably, adversarial manipulations can lead to clinically dangerous outputs, such as recommending incorrect dosages or omitting critical findings.
Chinese Translation
大型语言模型（LLMs）在医疗领域的应用日益增多，涉及临床问答、诊断支持和报告摘要等任务。尽管这些模型前景广阔，但它们对细微的提示扰动（无论是词汇还是句法）仍然高度敏感，这在安全关键的临床应用中带来了严重风险。在本研究中，我们进行了一项系统的敏感性分析，以评估通用型（例如，GPT-3.5、Llama3）和医学特定型LLMs（例如，ClinicalBERT、BioLlama3、BioBERT）的鲁棒性，使用MedMCQA基准进行测试。我们将扰动分为自然型和对抗型，并考察其对模型一致性、准确性和临床推理任务可靠性的影响。我们的研究结果显示，医学LLMs并非本质安全。即使是措辞上的微小变化也可能改变临床建议，而针对性的对抗提示可能引发有害输出。在医疗等高风险环境中，这种不可预测性是不可接受的——由于措辞变化而改变诊断或在稍作改述时产生虚构药物的模型，无法被临床医生可靠信任。虽然模型在简单的词汇替换或意译方面表现出一定的韧性，但在句法重组或误导性上下文提示下，它们往往会崩溃。这种脆弱性在通用型和领域特定的LLMs中都很明显。值得注意的是，对抗性操控可能导致临床上危险的输出，例如推荐错误的剂量或遗漏重要发现。
View on arXiv Download PDF AI Translation
cs.CL / 52 / 2606.07240

KIT's Submission to Cross-Lingual Voice Cloning in IWSLT 2026

KIT在IWSLT 2026跨语言语音克隆中的提交
Akti, Seymanur, Waibel, Alexander
Abstract
Cross-lingual voice cloning aims to generate speech in a target language while preserving speaker identity from a source-language reference. This task is central to speech translation and is the focus of the IWSLT 2026 Cross-Lingual Voice Cloning track. A key challenge is maintaining intelligibility and naturalness in the presence of accent variation and domain-specific vocabulary. We build on a multilingual text-to-speech model, FishAudio-S2-Pro, and introduce language tag prompting to improve language control and reduce accent leakage. We further apply reinforcement learning (RL) fine-tuning for task adaptation and observe improvements in intelligibility. Finally, we propose a reference-conditioned lexical matching method that improves pronunciation of domain-specific terms when lexical overlap is present. Results show that language prompting provides the largest gains, while lexical matching yields consistent improvements on matched subsets.
Chinese Translation
跨语言语音克隆旨在在目标语言中生成语音，同时保留来自源语言参考的说话者身份。该任务是语音翻译的核心，且是IWSLT 2026跨语言语音克隆赛道的重点。一个关键挑战是在口音变化和领域特定词汇的情况下保持可懂性和自然性。我们基于多语言文本到语音模型FishAudio-S2-Pro，并引入语言标签提示以改善语言控制并减少口音泄漏。我们进一步应用强化学习（Reinforcement Learning, RL）进行任务适应的微调，并观察到可懂性有所改善。最后，我们提出了一种参考条件下的词汇匹配方法，当存在词汇重叠时，该方法改善了领域特定术语的发音。结果表明，语言提示提供了最大的增益，而词汇匹配在匹配子集上产生了一致的改善。
View on arXiv Download PDF AI Translation
cs.CL / 53 / 2606.07300

Phun-Bench: Evaluating LLMs on Phonological Understanding in Chinese

Phun-Bench：评估大型语言模型在中文音韵理解上的表现
Yue, Xing, Shen, Yongliang, Lu, Weiming
Abstract
Language is a vehicle for thought, intricately tied to sounds, symbols, and meaning. However, most large language model (LLM) research focuses on meaning (semantics) and symbols (spelling) while largely overlooking sounds. Existing benchmarks on LLMs' phonological abilities are either solvable through rote memorization or intertwined with other abilities, making them inadequate to measure LLMs' genuine ability in phonological understanding. Here, we present Phun-Bench, a purpose-built Chinese benchmark with diverse tasks and settings across three dimensions (Homophony, Rhyme, and Phonetic Similarity), designed to systematically evaluate LLMs' phonological understanding. Our results show that while LLMs excel at recalling correct pronunciations, they generally struggle to leverage phonological knowledge in the flexible and intuitive way that human speakers do. Moreover, through detailed analyses, we propose a hypothesis regarding the underlying mechanism of LLMs' phonological understanding and "perception", highlighting an underexplored frontier for future research.
Chinese Translation
语言是思想的载体，紧密关联于声音、符号和意义。然而，大多数大型语言模型（LLM）研究集中于意义（语义）和符号（拼写），而在很大程度上忽视了声音。现有的关于LLM音韵能力的基准测试要么可以通过死记硬背解决，要么与其他能力交织在一起，这使得它们无法充分衡量LLM在音韵理解方面的真实能力。在此，我们提出了Phun-Bench，这是一个专门构建的中文基准，涵盖了三个维度（同音、韵律和语音相似性）下的多样化任务和设置，旨在系统地评估LLM的音韵理解能力。我们的结果表明，尽管LLM在回忆正确发音方面表现出色，但它们通常难以像人类说话者那样灵活和直观地利用音韵知识。此外，通过详细分析，我们提出了关于LLM音韵理解和“感知”潜在机制的假设，强调了未来研究中一个尚未充分探索的前沿领域。
View on arXiv Download PDF AI Translation
cs.CL / 54 / 2606.07313

SV-Detect: AI-generated Text Detection with Steering Vectors

SV-Detect：基于引导向量的人工智能生成文本检测
Vishnyakov, Mikhail, Gaintseva, Tatiana
Abstract
Detecting machine-generated text is especially difficult under distribution shift, such as transfer across domains, source models, and editing attacks. We propose a fake-text detector based on steering vectors extracted from the hidden representations of a frozen language model. At each layer, we construct a direction that separates human-written from machine-generated text, and represent each input by its layer-wise alignment with these directions. A lightweight classifier trained on these projection features yields the final detection score. Our method achieves strong performance both in-distribution and under distribution shift, including across domains, source models, and machine-editing transformations such as polishing and rewriting. Interpretation analyses show that the learned directions align with recognizable stylistic cues while capturing substantial additional signal beyond surface features. These results position fake-text detection as a representation-space probing problem and show that steering vectors provide a simple and effective solution.
Chinese Translation
在分布转移的情况下，例如跨领域、源模型和编辑攻击，检测机器生成的文本尤其困难。我们提出了一种基于从冻结语言模型的隐藏表示中提取的引导向量的假文本检测器。在每一层中，我们构建一个方向，以区分人类撰写的文本和机器生成的文本，并通过这些方向的逐层对齐来表示每个输入。一个在这些投影特征上训练的轻量级分类器生成最终的检测得分。我们的方法在分布内和分布转移下均表现出强大的性能，包括跨领域、源模型以及机器编辑变换（如润色和重写）。解释分析表明，学习到的方向与可识别的风格线索对齐，同时捕捉到超出表面特征的实质性额外信号。这些结果将假文本检测定位为一个表示空间探测问题，并表明引导向量提供了一种简单而有效的解决方案。
View on arXiv Download PDF AI Translation
cs.CL / 55 / 2606.07342

LLM-Guided Evolution for Medical Decision Pipelines

基于大型语言模型的医疗决策管道演化
Sviridov, Ivan, Oskin, Artem, Panin, Ivan, Bespalov, Iaroslav, Dylov, Dmitry, Oseledets, Ivan, Nesterov, Aleksandr
Abstract
Adapting large language models (LLMs) to clinical workflows often requires costly fine-tuning or manual prompt and pipeline engineering. We study LLM-guided MAP-Elites evolution as an inference-time alternative for discovering medical decision strategies and provide an implementation repository at https://github.com/univanxx/llm_guided_evo_medical. We formulate urgency triage, interactive consultation, and medical image classification as evolutionary searches over executable artifacts optimized by task-specific fitness functions. Across all three settings, evolution improves over manually designed baselines under practical constraints. In triage, evolved programs increase Semigran accuracy from $77.3\%$ to $87.1\%$ and emergency recall from $0.60$ to $0.97$, while improving safety-weighted held-out MIMIC-ESI performance. In interactive consultation, evolved policies improve the accuracy--cost frontier across Llama-3, Qwen-3.5, and Gemma-4 and transfer to held-out iCRAFTMD. In PneumoniaMNIST, prompt-only evolution improves frozen MedGemma VLMs while preserving strict JSON outputs. Qualitative analysis shows that the gains come from interpretable program-level mechanisms, calibrated triage boundaries, targeted evidence acquisition, selective commitment, and finding-oriented visual decision rules, rather than superficial prompt rewording alone.
Chinese Translation
将大型语言模型（LLMs）适应于临床工作流程通常需要昂贵的微调或手动提示和管道工程。我们研究了基于LLM的MAP-Elites演化，作为一种在推理时发现医疗决策策略的替代方案，并提供了实现库，地址为 https://github.com/univanxx/llm_guided_evo_medical。我们将紧急分诊、互动咨询和医学图像分类形式化为在可执行工件上进行的演化搜索，这些工件由特定任务的适应度函数优化。在所有三个设置中，演化在实际约束下优于手动设计的基线。在分诊中，演化程序将Semigran准确率从77.3%提高到87.1%，紧急召回率从0.60提高到0.97，同时改善了安全加权的MIMIC-ESI表现。在互动咨询中，演化策略提高了Llama-3、Qwen-3.5和Gemma-4的准确性-成本边界，并转移到保留的iCRAFTMD。在PneumoniaMNIST中，仅通过提示的演化改善了冻结的MedGemma VLM，同时保持严格的JSON输出。定性分析表明，这些提升来自可解释的程序级机制、校准的分诊边界、针对性的证据获取、选择性承诺以及面向发现的视觉决策规则，而不仅仅是表面的提示重述。
View on arXiv Download PDF AI Translation
cs.CL / 56 / 2606.07402

M$^3$Exam: Benchmarking Multimodal Memory for Realistic User-Agent Interactions

M$^3$Exam：基于真实用户-代理交互的多模态记忆基准测试
Huang, Zhengjun, Liu, Wenxuan, Tian, Zhoujin, Chen, Wei, Chen, Junle, Wu, Yuqian, Zhang, Fangyuan, Guo, Qintian, Zhou, Xiaofang
Abstract
Language agents are increasingly deployed over accumulating multimodal information, yet existing benchmarks assume a human-human form with sparse visuals and straightforward content, evaluating neither reasoning over authentic multimodal file interaction nor the interpretation of concealed user information. We therefore introduce M$^3$Exam, a query-centric multimodal conversational memory benchmark built on realistic user-agent interaction, with multi-dimensional evaluation spanning cross-modal grounding and implicit information inference. Benchmarking MLLMs and memory systems reveals persistent gaps in cross-modal grounding, cross session reasoning, and the efficiency cost of accumulating multimodal context. We further propose M$^3$Proctor, a multimodal memory method that detects query modality bias and consumes raw visual sources only on demand, improving accuracy by 13% while cutting index-construction time and retrieved tokens by over 70%.
Chinese Translation
语言代理越来越多地在不断积累的多模态信息上部署，然而现有基准测试假设为人际交互形式，视觉信息稀疏且内容简单，既未评估真实多模态文件交互中的推理，也未考虑隐含用户信息的解读。因此，我们引入了 M$^3$Exam，这是一个以查询为中心的多模态对话记忆基准，基于真实的用户-代理交互，具有跨模态基础和隐含信息推理的多维评估。对 MLLMs 和记忆系统的基准测试揭示了跨模态基础、跨会话推理以及积累多模态上下文的效率成本之间的持续差距。我们进一步提出了 M$^3$Proctor，这是一种多模态记忆方法，能够检测查询模态偏差，并仅在需要时使用原始视觉源，从而提高了 13% 的准确性，同时将索引构建时间和检索的标记数减少了超过 70%。
View on arXiv Download PDF AI Translation
cs.CL / 57 / 2606.07422

The Masked Advantage: Uncovering Local-Language Access to Cultural Knowledge in LLMs

掩蔽的优势：揭示大型语言模型中对文化知识的地方语言访问
Zhang, Yang, Fei, Xiao, Mohamed, Amr, Carneiro, Sarah Almeida, Konomi, Mersin, Geng, Mingmeng, Asaad, Ahmed, Shang, Guokan, Vazirgiannis, Michalis
Abstract
Large language models are increasingly used to answer culturally grounded questions across languages, yet it remains unclear whether local cultural knowledge is better accessed through English or the local language. Existing evaluations face two key limitations: many rely on parallel template-based questions that may not reflect how cultural knowledge naturally appears, and raw accuracy conflates general language proficiency with language-conditioned knowledge access. We address these issues with a controlled framework built on real-world cultural questions collected from regional benchmarks and local sources. By crossing question type (culture-agnostic vs. culture-specific) with query language (English vs. local language), and estimating ability with a shared 1PL item response theory model, we separate proficiency from localized knowledge access. Across 13 locales and roughly 80 models, we find a consistent English advantage on culture-agnostic questions, indicating stronger English proficiency. However, after accounting for this proficiency gap, local languages show a positive knowledge-access advantage in nearly all locale-model settings. This advantage is often masked in raw accuracy but becomes more visible for frontier, regionally aligned, or language-adapted models. Our results suggest that weaker local-language performance does not necessarily imply weaker cultural knowledge; rather, local cultural knowledge may be more accessible through the local language but hidden by limited language proficiency.
Chinese Translation
大型语言模型越来越多地用于回答跨语言的文化相关问题，但仍不清楚是通过英语还是地方语言更能有效获取地方文化知识。现有评估面临两个主要限制：许多评估依赖于平行模板问题，这可能无法反映文化知识自然出现的方式，而原始准确性将一般语言能力与语言条件下的知识获取混为一谈。我们通过一个控制框架来解决这些问题，该框架基于从区域基准和地方来源收集的真实文化问题。通过将问题类型（文化无关与文化特定）与查询语言（英语与地方语言）交叉，并使用共享的1PL项目反应理论模型来估计能力，我们将语言能力与地方知识获取分开。在13个地区和大约80个模型中，我们发现，在文化无关问题上，英语表现出一致的优势，表明其更强的英语能力。然而，在考虑到这种能力差距后，地方语言在几乎所有地区-模型设置中显示出积极的知识获取优势。这种优势在原始准确性中往往被掩盖，但对于前沿、区域对齐或语言适应的模型来说，变得更加明显。我们的结果表明，较弱的地方语言表现并不一定意味着较弱的文化知识；相反，地方文化知识可能通过地方语言更易获取，但由于语言能力有限而被隐藏。
View on arXiv Download PDF AI Translation
cs.CL / 58 / 2606.07441

Sycophantic Praise: Evaluating Excessive Praise in Language Models

阿谀奉承的赞美：评估语言模型中的过度赞美
Vennemeyer, Daniel, Duong, Phan Anh, Ye, Meryl, Huang, Ruihong, Jiang, Tianyu
Abstract
Sycophancy in language models is typically studied as excessive agreement or validation, while explicit praise and flattery have received comparatively little attention. We argue that sycophantic praise is a distinct alignment problem that cannot be reliably measured using current methods. We introduce a parameterized framework that measures whether praise is excessive relative to contribution quality and expected user ability. We show that our framework substantially outperforms generic LLM judges in agreement with human annotations, and that sycophantic praise occurs far more frequently in social and interpretive domains than in objective reasoning settings. Together, these findings position praise calibration as a distinct alignment challenge.
Chinese Translation
语言模型中的阿谀奉承通常被研究为过度的同意或验证，而明确的赞美和恭维则相对较少受到关注。我们认为，阿谀奉承的赞美是一个独特的对齐问题，无法通过当前的方法可靠测量。我们引入了一个参数化框架，用于衡量赞美是否相对于贡献质量和预期用户能力而言过度。我们展示了我们的框架在与人类注释的一致性方面显著优于通用的LLM评估者，并且阿谀奉承的赞美在社交和解释性领域的发生频率远高于在客观推理环境中。综上所述，这些发现将赞美校准定位为一个独特的对齐挑战。
View on arXiv Download PDF AI Translation
cs.CL / 59 / 2606.07479

Supervision versus Demonstration-Based In-Context Learning for Multiword Expression Classification

监督学习与基于示范的上下文学习在多词表达分类中的比较
Karakaş, Sercan, Şimşek, Yusuf
Abstract
Turkish idiomatic light verb constructions (LVCs) are challenging for multiword expression processing because they often share the same surface form as fully literal verb-object combinations while functioning as a single, partially idiomatic predicate. We frame Turkish LVC detection as a binary classification task (literal meaning vs. idiomatic meaning) and evaluate on a manually created controlled set (N=147) with matched negatives: out-of-domain random sentences and in-domain literal controls (NLVC), alongside LVC positives. We compare a supervised Turkish encoder baseline (BERTurk with a classifier head) to three instruction-tuned LLMs from different families under zero-shot, one-shot, and few-shot prompting, and analyze how demonstrations shift error profiles. In zero-shot, LLMs perform well on negatives but show very low LVC recall. One-shot prompting sharply improves LVC detection but can induce strong, model-specific biases, leading models to overpredict or underpredict LVCs. A richer few-shot prompt improves calibration and yields robust overall performance for GPT-OSS-20B and Qwen 2.5-14B. Overall, the results highlight substantial prompt sensitivity in Turkish metalinguistic classification: the supervised baseline remains competitive, while prompted LLMs can match or exceed it on LVCs with carefully constructed demonstrations.
Chinese Translation
土耳其语的习惯性轻动词构造（LVCs）在多词表达处理上具有挑战性，因为它们常常与完全字面意义的动词-宾语组合共享相同的表面形式，同时作为一个部分习惯性的谓词发挥作用。我们将土耳其语LVC的检测框架设定为一个二分类任务（字面意义与习惯意义），并在一个手动创建的控制集（N=147）上进行评估，该控制集包含匹配的负例：领域外的随机句子和领域内的字面控制句（NLVC），以及LVC的正例。我们将一个监督的土耳其语编码器基线（带分类头的BERTurk）与来自不同家族的三种经过指令调优的LLM进行比较，采用零-shot、one-shot和few-shot提示，并分析示范如何改变错误特征。在零-shot情况下，LLM在负例上的表现良好，但LVC的召回率非常低。one-shot提示显著提高了LVC的检测，但可能会引入强烈的模型特定偏差，导致模型过度预测或不足预测LVC。更丰富的few-shot提示改善了校准，并为GPT-OSS-20B和Qwen 2.5-14B提供了稳健的整体性能。总体而言，结果突显了土耳其语元语言分类中的显著提示敏感性：监督基线保持竞争力，而经过精心构建的示范的提示LLM可以在LVC上匹配或超越它。
View on arXiv Download PDF AI Translation
cs.CL / 60 / 2606.07502

Your UnEmbedding Matrix is Secretly a Feature Lens for Text Embeddings

你的反嵌入矩阵实际上是文本嵌入的特征透镜
Wu, Songhao, Chen, Zhongxin, Liu, Yuxuan, Cui, Heng, Li, Cong, Yan, Rui
Abstract
Large language models exhibit impressive zero-shot capabilities across a wide range of downstream tasks. However, they struggle to function as off-the-shelf embedding models, leading to suboptimal performance on massive text embedding benchmarks. In this paper, we identify a potential cause underlying this deficiency. Our motivation stems from an unexpected observation: text embeddings tend to align with frequent but uninformative tokens when projected onto the vocabulary space. We argue that this excessive expression of high-frequency tokens suppresses the model's ability to capture nuanced semantics. To address this, we introduce EmbedFilter, a simple linear transformation designed to refine text embeddings derived from LLMs directly. Specifically, we uncover that the unembedding matrix within LLMs encodes a latent space that is actively writing these frequent tokens into embedding space. By filtering out this subspace, EmbedFilter suppress the influence of high-frequency tokens, thereby enhancing semantic representations. As a compelling byproduct, this enables an inherent dimensionality reduction, lowering index storage and speedup retrieval while fully preserving the refined embedding quality. Our experiments across multiple LLM backbones demonstrate that LLMs equipped with EmbedFilter achieve superior zero-shot downstream performance even with significantly reduced embedding dimensions. We hope our findings provide deeper insights into the mechanisms of LLM-based representations and inspire more principled designs to improve text embeddings training. Our code is available at https://github.com/CentreChen/EmbFilter.
Chinese Translation
大型语言模型在广泛的下游任务中展现出令人印象深刻的零样本能力。然而，它们作为现成的嵌入模型的表现却不尽如人意，导致在大规模文本嵌入基准测试中的性能不佳。本文识别了导致这一缺陷的潜在原因。我们的动机源于一个意外的观察：文本嵌入在投影到词汇空间时，往往与频繁但无信息量的标记对齐。我们认为，这种对高频标记的过度表达抑制了模型捕捉细微语义的能力。为了解决这个问题，我们引入了 EmbedFilter，这是一种旨在直接优化来自大型语言模型（LLMs）的文本嵌入的简单线性变换。具体而言，我们发现 LLM 中的反嵌入矩阵编码了一个潜在空间，该空间积极地将这些频繁的标记写入嵌入空间。通过过滤掉这个子空间，EmbedFilter 抑制了高频标记的影响，从而增强了语义表示。作为一个引人注目的副产品，这实现了内在的维度减少，降低了索引存储并加速了检索，同时完全保留了优化后的嵌入质量。我们在多个 LLM 主干上的实验表明，配备 EmbedFilter 的 LLM 即使在显著减少嵌入维度的情况下，也能实现更优越的零样本下游性能。我们希望我们的发现能为 LLM 基础表示的机制提供更深入的见解，并激励更有原则的设计来改善文本嵌入的训练。我们的代码可在 https://github.com/CentreChen/EmbFilter 获取。
View on arXiv Download PDF AI Translation
cs.CL / 61 / 2606.07513

Agentopia: Long-Term Life Simulation and Learning in Agent Societies

Agentopia：代理社会中的长期生活模拟与学习
Wang, Xintao, Zheng, Sirui, Wu, Hongqiu, Li, Weiyuan, Huang, Jen-tse, Zhu, Minghao, Zu, Can, Deng, Qi, Wang, Jiawei, He, Qianyu, Wang, Heng, Wu, Xiaojian, Tao, Yunzhe
Abstract
Humans learn from social life. Simulating this process with LLM-powered agents represents a promising research direction, raising a natural question: whether LLMs can learn from such simulated social experience to better understand and replicate human behavior. However, prior agent society simulations typically operate at the scale of days, limiting the depth of social interactions and long-term growth. In this paper, we study long-term life simulation and LLM learning in agent societies, with two goals: (1) investigating social behaviors that emerge from life-long simulation, and (2) developing anthropomorphic capabilities in LLMs, particularly intelligence in social life, through years of simulated social experience. Specifically, we present Agentopia, a comprehensive framework for long-term life simulation in multi-agent societies, where 100 agents autonomously pursue personal growth, develop social relationships, and fulfill their needs and goals over 10 simulated years. We define life reward to mirror human well-being, and leverage this reward to train LLMs via rejection sampling. Extensive experiments show that agents exhibit rich emergent social behaviors. Furthermore, life reward training effectively enhances the underlying LLM, which leads to improved agent well-being in simulation, and generalizes to downstream role-playing benchmarks with +15.6% improvement.
Chinese Translation
人类从社会生活中学习。利用大型语言模型（LLM）驱动的代理模拟这一过程代表了一个有前景的研究方向，提出了一个自然的问题：LLM是否能够从这种模拟的社会经验中学习，以更好地理解和复制人类行为。然而，以往的代理社会模拟通常在天数的尺度上运行，限制了社会互动的深度和长期成长。在本文中，我们研究了代理社会中的长期生活模拟和LLM学习，设定了两个目标：（1）研究从终身模拟中涌现的社会行为；（2）通过多年的模拟社会经验，发展LLM的人性化能力，特别是在社会生活中的智能。具体而言，我们提出了Agentopia，这是一个用于多代理社会中长期生活模拟的综合框架，其中100个代理自主追求个人成长，发展社会关系，并在10年的模拟中满足他们的需求和目标。我们定义了生活奖励以反映人类的福祉，并利用这一奖励通过拒绝采样来训练LLM。大量实验表明，代理表现出丰富的涌现社会行为。此外，生活奖励训练有效增强了基础LLM，这导致了模拟中代理福祉的改善，并在下游角色扮演基准测试中实现了+15.6%的提升。
View on arXiv Download PDF AI Translation
cs.CL / 62 / 2606.07515

How reliable are LLMs when it comes to playing dice?

大型语言模型在掷骰子时的可靠性如何？
Avena, Luca, Bet, Gianmarco, Busoni, Bernardo
Abstract
We investigate the probabilistic reasoning capabilities of large language models through a controlled benchmarking study on discrete probability problems. We constructed two datasets, respectively a set of standard exercises and a set of counterintuitive exercises, designed to trigger heuristic reasoning, and evaluated 8 state-of-the-art models, each tested with and without Chain-of-Thought prompting. Models achieve an average accuracy of 0.96 on standard problems but only 0.59 on counterintuitive ones. We further provide empirical evidence of token bias: performance drops by over 20% when canonical formulations are replaced by disguised variants. Embedding misleading suggestions in the prompt reduces performance by up to 34%, with no model proving immune. Taken together, the reported findings suggest that current LLMs are not yet genuine probabilistic reasoners, despite their success in advanced mathematical problems.
Chinese Translation
我们通过对离散概率问题的控制基准研究，探讨大型语言模型的概率推理能力。我们构建了两个数据集，分别是一组标准练习和一组反直觉练习，旨在引发启发式推理，并评估了8个最先进的模型，每个模型在有无链式思维（Chain-of-Thought）提示的情况下进行测试。模型在标准问题上的平均准确率为0.96，但在反直觉问题上的准确率仅为0.59。我们进一步提供了标记偏差的实证证据：当规范表述被伪装变体替代时，性能下降超过20%。在提示中嵌入误导性建议会使性能降低多达34%，且没有模型对此表现出免疫。综合来看，报告的发现表明，尽管当前大型语言模型在高级数学问题上取得了成功，但它们尚未成为真正的概率推理者。
View on arXiv Download PDF AI Translation
arXiv Reader
Automatically generated on 2026-07-16 10:14:14
Powered by OpenAI GPT & arXiv