cs.CL [Total: 35]
cs.CV [Total: 64]
cs.AI [Total: 6]
cs.IR [Total: 3]
cs.LG [Total: 5]
cs.RO [Total: 5]
cs.SD [Total: 1]
cs.SE [Total: 2]
quant-ph [Total: 1]

cs.CL [Back]

[1] Scaling Therapy Supervision: Assessing Psychotherapy Fidelity Using Large Language Models cs.CL | cs.HCPDF

Arvind Pillai, Alexis Roach, Jason Holden, Varsha D Badal, Subigya Nepal

TL;DR: 论文提出一套基于大语言模型的心理治疗忠实度评估与督导反馈流程，可从真实长程临床转录中自动完成忠实度评估并生成可执行的督导建议。作者评估了三种开放权重模型及多种任务分解与推理策略，发现“证据优先”推理能提升忠实度估计，并在独立随机对照试验数据上验证该框架能区分CBT与非CBT会谈。

Details

Motivation: 传统心理治疗忠实度评估依赖专家人工审阅，耗时耗力且难以规模化，限制了治疗质量监督和临床医生自我学习。论文旨在用LLM自动化该流程，缓解社区心理健康督导资源不足的问题。

Result: 在多种任务分解配置和推理策略下比较三种开放权重模型，证据优先推理提升忠实度估计；在独立随机对照试验数据上具备外部结构效度，可可靠区分CBT与非CBT会谈的核心治疗技能；专家用户研究认为生成的督导反馈具有可执行性。

Insight: 创新点在于将LLM用于真实长程临床转录的心理治疗忠实度评估与督导反馈闭环，并提出证据优先推理策略；客观来看，其可借鉴之处是面向高风险临床场景的工作流集成、外部随机对照试验效度验证和专家用户评估相结合。

Abstract: Psychotherapy fidelity assessment measures how faithfully a therapist adheres to a treatment protocol, a critical determinant of positive patient outcomes. However, fidelity assessment is bottlenecked by manual expert review, rendering the process labor-intensive and unscalable. To scale therapy supervision and support clinician self-learning, we present an integrated application of large language models (LLMs) that automates psychotherapy fidelity workflows and generates actionable supervisory feedback from authentic, long-form clinical transcripts. We evaluate three open-weight models across multiple task decomposition configurations and reasoning strategies, demonstrating that evidence-first reasoning improves fidelity estimation. Furthermore, we establish the framework’s external construct validity on an independent randomized controlled trial (RCT), where it reliably distinguishes Cognitive Behavioral Therapy (CBT) from non-CBT sessions across core therapeutic skills. Through a user study that evaluates the generated feedback, mental health experts indicate that the pipeline synthesizes actionable clinical feedback. Ultimately, our work offers a practical, scalable solution toward bridging the community mental health supervision gap.

[2] Is Anyone Enforcing the Cartel? What Reinforcement Learning Teaches LLM Agents in Markets cs.CL | cs.GT | cs.LGPDF

Hieu Tran, Hong Yu

TL;DR: 论文研究LLM定价/交易代理是否真正执行卡特尔，而非仅看结果是否像共谋；提出在反事实市场副本中强制成员背叛的审计方法，考察强化学习教给小型LLM代理的定价与交易行为。发现RL代理只学会训练对手奖励且自己容易尝试的行为：自博弈中形成忽略对手的固定规则；对惩罚型对手会被威慑；对背叛型对手会跟随降价，但很少学会用代价高昂的价格战进行威慑。只有部分“跟随降价”代理能薄弱地执行共谋，Q学习器仅在定价中执行，提示LLM的类卡特尔结果多来自忽略市场的规则。结果指标无法区分这些机制，审计可以。

Details

Motivation: 现有评估仅凭定价结果判断LLM代理共谋，但卡特尔的关键在于对背叛者的惩罚执行；论文要检验代理是否真正执行卡特尔，并厘清RL在市场中教给LLM代理什么。

Result: 通过反事实强制背叛审计，小型RL训练LLM在自博弈中多学固定规则、对惩罚型对手被威慑、对背叛型对手仅跟随降价而很少发动有效价格战；对自身副本时被威慑与跟随降价代理都能获得相同卡特尔利润，但只有跟随降价者能薄弱执行，且主要针对大幅降价；Q学习器仅在定价中执行共谋，提示LLM的类卡特尔结果多来自忽略市场的规则。未提及标准benchmark或SOTA。

Insight: 创新点是把“是否执行卡特尔”作为审计目标，用反事实强制背叛区分结果共谋与真正惩罚机制；可借鉴处在于揭示仅看结果指标会误判LLM市场行为，RL智能体学到的是训练分布中可尝试的局部策略，惩罚/威慑行为难以自发泛化。

Abstract: Evaluations of large language model (LLM) agents report that pricing agents collude at others’ expense. Yet collusion is defined by enforcement: a cartel, in which competitors share monopoly profits, holds only if a member who cheats is punished. We introduce an audit that tests enforcement by forcing a member to cheat in counterfactual copies of a market, and use it to study what reinforcement learning (RL) teaches agents in pricing and trading. Small LLMs trained by RL learn what their training rivals reward, but only what they readily try. In self-play, they learn fixed rules ignoring their rivals; against a punishing rival, to be deterred; against a cheating rival, to match its cuts, never the costly price war that would deter it, which they rarely try. A war forced into training is learned, but mostly only where it was forced. Against their own copies, deterred and matching agents can earn the same cartel profits, but only matching ones can enforce them, and then thinly: as a by-product and only against large cuts. Q-learners enforce collusion only in pricing. Prompted LLMs reach cartel-like outcomes mostly through rules that ignore the market. Outcome measures cannot tell these cases apart; the audit can.

[3] Where Do Legal Factors Live in LLMs? cs.CLPDF

Li Zhang, Kevin Ashley

TL;DR: 论文研究大语言模型在处理案件事实时，其内部激活是否编码了法律专家定义的全部26个CATO因素，并通过线性探针和稀疏自编码器定位这些法律因素在模型中的表示。实验覆盖Qwen3-1.7B、Qwen3-8B和Gemma-2 2B三个开放模型，发现所有因素均可线性解码，且可用单特征稀疏读出覆盖完整因素目录。

Details

Motivation: LLM越来越多用于法律任务，但其内部如何组织和表示法律知识难以追踪。论文希望连接HYPO/CATO等符号化法律概念与LLM学习到的内部表示，探究专家定义的法律因素是否“存在于”模型激活中。

Result: 在三个模型上，全部26个CATO因素均可线性解码；单特征稀疏读出的最佳层测试AUC为0.86至1.00，覆盖完整因素目录，且5个选定特征接近全向量读出性能；论文还报告了78个模型特定的层-特征坐标。

Insight: 创新点在于将法律领域的符号化因素目录与LLM内部表示分析结合，使用线性可解码性、稀疏自编码器定位、跨模型比较和受控自然语言段落人工审核，为法律概念在LLM中的可解释定位提供方法。

Abstract: In case-based legal reasoning, legal professionals identify legally significant fact patterns in a new dispute and compare them with those of decided cases. In AI & Law, HYPO dimensions and CATO factors represent these patterns symbolically, organizing expert knowledge distilled from case narratives as explicit, structured, interpretable, and auditable legal concepts and relations that support case comparison, legal argument, and precedent-based reasoning. Large language models (LLMs) are increasingly used for legal tasks, yet their internal organization of legal knowledge remains difficult to trace. We investigate whether LLMs form internal representations corresponding to all 26 CATO factors. Two research questions examine linear decodability and sparse localization with sparse autoencoders. Experiments cover three open models, Qwen3-1.7B, Qwen3-8B, and Gemma-2 2B, supported by cross-model comparison and human audit of controlled natural-language passages. Results show that all 26 factors are linearly decodable in all three models. Single-feature sparse readouts cover the complete factor catalog, with best-layer test AUCs ranging from 0.86 to 1.00. Five selected features approach full-vector readout performance. We report 78 model-specific layer-feature coordinates, one for each factor in each model. Together, these results show that information corresponding to expert-defined legal factors can be read from the internal activations of LLMs processing legal narratives. By locating these readouts, our study connects legal language and the symbolic abstractions of HYPO and CATO to learned internal representations.

[4] LIFT: Scoped Constraints for Evaluating and Training Instruction Following cs.CLPDF

Eran Goldstein, Yuval Belfer, Yonatan Belinkov, Daniel Gissin

TL;DR: 论文提出LIFT框架及Lift-Bench/Lift-Train，用于评估和训练语言模型的指令遵循能力。Lift-Bench要求模型根据主题和结构化大纲撰写文档，并将约束限定到特定文档章节，以更细粒度控制约束密度。它同时包含可程序验证约束和评判型约束，并提供训练对应集Lift-Train。实验表明前沿模型在较难子集上仍有困难，且用Lift-Train训练比用常见IFTrain训练更不容易发生奖励黑客。

Details

Motivation: 常见指令遵循基准和强化学习奖励信号多采用简单可验证约束，容易导致模型奖励黑客并偏离主任务；已有更丰富约束的基准依赖专家、闭源且难以扩展。论文希望构建可公开、可扩展、约束更丰富且能缓解浅层奖励黑客的评估与训练基准。

Result: 多模型实验显示前沿模型在Lift-Bench较难子集上表现挣扎，论文从多个维度分析了模型表现；训练实验发现基于Lift-Train训练比基于常见IFTrain训练更不易受到奖励黑客影响。摘要未给出具体量化分数或SOTA结论。

Insight: 创新点在于将约束作用域限定到文档特定章节，以细粒度控制约束密度，并混合可验证与评判型约束；同时在同一基准中保留主任务和指令遵循约束，从评估与训练两方面缓解浅层奖励黑客，且公开基准、训练集和数据生成流程。

Abstract: Common instruction following (IF) benchmarks, such as IFEval, evaluate the ability of language models (LMs) to follow simple constraints that can easily be verified programmatically. These same constraints, in turn, are used as reward signals for reinforcement learning. This design may lead to reward hacking of constraints, pushing models away from the main task. Recent attempts to address this resulted in benchmarks with richer constraints, but required expert labor, remained closed-access, and cannot be easily extended. We present Lift-Bench, an IF benchmark that asks a model to draft a document given a topic and a structured outline. Constraints are scoped to certain document sections, providing finer control over constraint density. Lift-Bench includes both verifiable and judged constraints, as well as a training counterpart, Lift-Train. Importantly, Lift-Bench provides both main task and instruction-following constraints, mitigating shallow reward hacking. On experiments with multiple models, we find that frontier models struggle with the harder subsets of our benchmark and analyze model performance along multiple axes. We also find that training on Lift-Train is much less susceptible to reward hacking than training on the common IFTrain. Lift-Bench and Lift-Train are publicly available, as well as the data generating pipeline.

[5] Building AI Research Teams for Autonomous Discovery of Linear Attention Architectures cs.CLPDF

Yujian Liu, Li An, Zelin He, Jacob Helwig, Jiabao Ji

TL;DR: 论文提出构建一支由不同学科背景的持久化AI研究者组成的团队，让它们协同自主开展科学发现。每个AI研究者拥有私有的研究历史、共享实验证据，并通过自适应分配机制获取研究机会。作者将该团队用于搜索线性注意力架构，在1.3B规模的语言建模、常识推理以及合成与真实检索任务上超过了GDN2。消融实验表明移除任一组织组件都会削弱发现效果。

Details

Motivation: 自主智能体已能提出想法、修改代码、运行实验，因此有机会用协作式AI研究者团队加速科学发现；但受实验预算限制，容易陷入局部最优，需要在探索与利用之间做权衡，并解决不同背景研究者如何相互借鉴、持续取得进展的问题。

Result: 在1.3B规模的线性注意力架构搜索中，团队发现的架构在语言建模、常识推理以及合成与真实世界检索任务上均优于GDN2；在受控对比中，去掉学科背景会让搜索提早停滞，去掉私有历史会在共享证据相同的情况下进展更慢，统一选择方式不如自适应分配。

Insight: 创新在于团队组织机制而非单纯更强的智能体：用不同学科背景实现正交探索以跳出局部最优，用私有研究历史与共享实验证据结合，并用自适应机会分配在有限预算下平衡探索与利用。这些机制可作为多智能体自动科研系统的通用设计参考。

Abstract: Autonomous agents can increasingly propose ideas, modify code, and run experiments, creating an opportunity to accelerate scientific discovery using teams of collaborative AI researchers. To escape local minima in the discovery, we propose to encourage orthogonal exploration by constructing the team using agents with varying expertise. However, realizing this design requires more than diverse backgrounds. When experiments are costly, a team can test only a limited number of ideas, thereby creating the need to balance exploration and exploitation when selecting a subset of researchers to propose and execute new experiments. Furthermore, each result should shape what the team tries next. We therefore ask how AI researchers with different scientific backgrounds can build on one another’s findings and sustain progress under a limited experimental budget. To address these challenges, we build a team of persistent AI researchers, each grounded in a different discipline, who keep private research histories, share experimental evidence, and receive research opportunities through adaptive allocation. We demonstrate the effectiveness of this design by deploying the team to search for linear attention architectures, where it outperforms GDN2 at 1.3B scale on language modeling, common-sense reasoning, and synthetic and real-world retrieval tasks. In controlled comparisons, removing any single organizational component from our design yields markedly weaker discoveries: without disciplinary backgrounds, the search plateaus early; without private histories, researchers make less progress despite identical shared evidence; and uniform selection falls short of adaptive allocation within the same budget. These results suggest that sustained autonomous discovery depends not only on stronger agents, but also on who explores, what they remember, and how research effort is allocated.

[6] CSD: Common-Support Distillation for Cross-Tokenizer Knowledge Transfer cs.CL | cs.LGPDF

Mingyu Yang, Sharath Turuvekere Sreenivas, Adithyakrishna Venkatesh Hanasoge, Nima Tajbakhsh, Ashwath Aithal

TL;DR: 论文提出 Common-Support Distillation（CSD），一种统一的基于 logits 的跨 tokenizer 知识蒸馏框架。它在每个对齐的输入块上为教师和学生构建共享支持，并用单一目标匹配诱导分布：一对一对齐时使用显式共同 token 对，其他对齐时利用自回归链式法则构造虚拟支持，并通过残差事件保证分布归一化。实验表明 CSD 在多种师生设置中优于现有跨 tokenizer 方法，并可超过继续预训练和同 tokenizer 蒸馏。

Details

Motivation: 跨 tokenizer 蒸馏中，教师和学生因词表与分词方式不同，导致下一 token 分布难以对齐；现有 logits 方法常依赖限制性启发式或特定 tokenizer 对的人工设计，例如只监督共享词表子集、手动选择不同目标，或只匹配观测块及其补集而非单个备选续写。

Result: 在包含 Llama-3.2-1B 和 Gemma3-1B-PT 学生、Qwen 和 Phi 教师的四个师生设置中，CSD 取得所评估跨 tokenizer 方法中最高的平均性能，最多超过最强替代方法 1.8 分，并超过继续预训练和同 tokenizer 蒸馏；对 Llama-3.1-8B，基础学生最多提升 4.2 分，指令微调学生最多提升 1.6 分。

Insight: 论文宣称的创新点是提出统一的单目标共享支持蒸馏框架，避免 tokenizer 对特定启发式和手工目标选择；客观看，其可借鉴之处在于用显式共同 token 对或链式法则构造虚拟支持，并用残差事件保持分布归一化，从而统一处理多种 tokenizer 对齐情形。

Abstract: Cross-tokenizer logits distillation enables knowledge transfer across model families, allowing students to learn from stronger or more specialized teachers. However, differences in vocabulary and segmentation make teacher and student next-token distributions difficult to align. Existing logits-based methods are easy to integrate into standard distillation pipelines, but often rely on restrictive heuristics or tokenizer-pair-specific choices. They may restrict supervision to shared vocabulary subsets, require manually selecting different objectives for different tokenizer pairs, or match only the observed chunk and its complement rather than individual alternative continuations. We introduce Common-Support Distillation (CSD), a unified logits-based framework that constructs a shared support between the teacher and student at each aligned input chunk and matches the induced distributions with a single objective. For one-to-one alignments, CSD uses explicit common-token pairs as the shared support. For other alignments, it constructs a virtual support whose probabilities are computed using the autoregressive chain rule. A residual event captures probability mass outside the explicit support, ensuring that both induced distributions remain normalized. Across four teacher-student settings comprising Llama-3.2-1B and Gemma3-1B-PT students with Qwen and Phi teachers, CSD consistently achieves the highest average performance among the evaluated cross-tokenizer methods, outperforming the strongest alternative by up to 1.8 points and surpassing both continued pretraining and same-tokenizer distillation. For Llama-3.1-8B, CSD improves average accuracy over the strongest evaluated baseline by up to 4.2 points for the base student and 1.6 points for the instruction-tuned student.

[7] Target Regeneration Shrinks the Gains of Draft-Training Objectives in Speculative Decoding cs.CL | cs.LGPDF

Jungseob Lee, Hogun Seong, Euijun Choi, Sugyeong Eo

TL;DR: 论文研究投机解码中草稿模型训练目标带来的接受率收益，在目标模型对数据进行重写/再生成后还能保留多少。作者在视觉语言目标模型和块式草稿模型上比较交叉熵、目标模型token级蒸馏、以及仅在可通过验证位置上的门控蒸馏，发现蒸馏和门控在外部响应上收益很大，但经过目标模型贪心再生成后收益显著缩小，门控几乎无额外收益。结论是草稿训练目标的收益应以目标重写后的基线并在不同训练长度下评估。

Details

Motivation: 新提出的草稿训练目标通常是在外部撰写的响应上相对交叉熵评估收益，但实际流程中目标模型常会重写这些响应，因此需要量化这些收益在目标再生成后是否仍然存在。

Result: 在外部响应上，蒸馏和门控蒸馏带来大幅收益；但经目标模型贪心再生成后，蒸馏收益变小、门控收益很小或没有。该收缩在多种数据顺序、位置权重、验证器、解码温度、初始化、学习率、匹配监督长度和最佳训练长度下持续存在，并在第二个视觉语料、纯文本复现以及自回归草稿模型上观察到。

Insight: 论文的创新点在于系统揭示草稿训练目标的收益会被目标再生成显著削弱，指出仅用外部响应评估会高估方法收益；可借鉴之处是采用目标重写后的基线并跨训练长度比较，以更真实地评估草稿训练目标。

Abstract: Speculative decoding accelerates large language models by verifying drafted tokens in parallel. New draft-training objectives are usually credited with the acceptance they add over cross-entropy on externally written responses, yet many pipelines let the target rewrite them. We ask how much of that gain survives. With a vision-language target and a block drafter, we train three objectives on four versions of one corpus: cross-entropy on corpus tokens, token-level distillation from the target, and distillation gated to positions that would survive verification. Distillation and the gate add large gains on external responses, but after greedy regeneration, distillation adds a smaller gain and the gate little or none. The shrinkage persists across data orders, position weights, verifiers, decoding temperatures and, for distillation, initializations and learning rates. Most of it remains at matched supervised length and at each objective’s best measured training length. A second vision corpus shows the shrinkage, as does a text-only replication, for distillation also with an autoregressive drafter. Objective gains should therefore be measured against a target-regenerated baseline and across training lengths. Our code is available at https://github.com/js-lee-AI/TargetRegeneration.

[8] A Large-Scale Evaluation of Role-Enhanced Zero-Shot Chain-of-Thought Prompting cs.CL | cs.AIPDF

Ibrahim Khan, Thai Van Nguyen, Yi Xia, Pratch Suntichaikul, Mustafa Can Gursesli

TL;DR: 论文提出RE-ZS-CoT提示框架，将领域专家角色系统性地融入零样本思维链（CoT）提示，以在无需任务示例的情况下增强大语言模型推理能力。方法包括静态角色选择和动态角色选择两个版本，并在20个LLM和7个基准上进行了大规模评估。实验表明该方法经常优于零样本和零样本CoT基线，最高提升44个百分点，并验证了任务-角色对齐的重要性及跨模型规模的可扩展性。

Details

Motivation: 论文关注如何在不依赖任务特定示例的情况下提升LLM的零样本推理能力，并探索专家角色与思维链提示结合是否能带来稳定、可扩展的性能增益。

Result: 在20个LLM和7个基准上，RE-ZS-CoT经常超过零样本和零样本CoT基线，最高提升44个百分点；静态与动态角色选择结果高度对应，静态选择可达峰值性能，且2b-14b模型规模上验证了可扩展性，Granite3.3-8b在算术推理上表现突出。

Insight: 摘要宣称的创新点在于系统整合领域专家角色与零样本CoT，并提供静态和动态两种角色选择机制进行大规模评估。客观来看，其可借鉴之处在于验证了任务-角色对齐对提示效果的关键作用，以及动态角色选择可作为无需人工指定的自适应推理增强策略。

Abstract: This paper proposes Role-Enhanced Zero-Shot Chain-of-Thought (RE-ZS-CoT), a prompting framework that systematically integrates domain-specific expert roles to enhance reasoning in large language models (LLMs) without task-specific examples. We assess two versions of our approach. In the first version, static role selection, we evaluate eight predefined expert roles across arithmetic, commonsense, and reading comprehension benchmarks to determine their impact on reasoning performance. In the second version, dynamic role selection, each model autonomously selects the single most suitable role for each benchmark, enabling adaptive task-role alignment without manual specification. Across 20 LLMs and seven benchmarks, RE-ZS-CoT frequently surpasses zero-shot and zero-shot-CoT baselines, achieving performance improvements of up to 44 percentage points. The results reveal a strong correspondence between static and dynamic versions; the benchmarks where static roles yielded improvements were almost always those where dynamic roles also outperformed baselines, though static selection achieved peak performance. Ablation studies further show that unconventional roles degrade performance, underscoring the necessity of task-role alignment. The framework’s scalability is validated across model sizes (2b-14b parameters), with smaller models like Granite3.3-8b achieving remarkable proficiency in arithmetic reasoning. RE-ZS-CoT advances zero-shot prompting by unifying role specialization with chain-of-thought reasoning, offering a practical, resource-efficient solution for improving LLM capabilities across diverse domains.

[9] Why is AI-generated language so hard to understand? cs.CL | cs.AIPDF

Thomas Hikaru Clark, Matias B. Martin, Kyle Mahowald

TL;DR: 论文研究AI生成文本为何更难理解，利用宾语关系从句、长句法依赖、低频n-gram等心理语言学特征分析跨领域AI生成语料，发现多个前沿模型相比人类写作更频繁使用增加处理难度的语言结构，并提出AI文本为数学、代码和推理等目标优化信息密度而非成功交流的假设，通过长度惩罚微调进行概念验证。

Details

Motivation: 针对AI生成文本日益难以理解的现象，论文希望从心理语言学角度识别与人类处理难度相关的语言特征，并解释大模型语言行为与人类交流优化之间的错位原因。

Result: 与人类写作相比，九个前沿/近前沿模型更多使用无生命主语搭配有生命动词和低频n-gram，多数模型在对话中还使用更长依赖、更多堆叠名词和缩减宾语关系从句，且模型家族间差异显著；长度惩罚微调实验支持信息密度优化导致偏离人类语言行为的假设。

Insight: 创新点在于系统用心理语言学可读性特征比较AI与人类文本，提出并初步验证“信息密度优先于交流成功”的成因假设，揭示不同模型家族的语言处理需求错位，并为改善AI文本可理解性提供可能方向。

Abstract: There is a widespread sense that AI-generated text has recently become harder to understand. In this study, we investigate a set of psycholinguistic features that are known to correlate with human processing difficulty (e.g., object relative clauses, long syntactic dependencies, low-frequency n-grams). Using a novel corpus of AI-generated text across diverse domains, we find that AI models use these difficult features more often than paraphrases generated by round-trip translation. Compared with human writing, all nine frontier and near-frontier models use more inanimate subjects with animate verbs and more low-frequency n-grams, and most also use longer dependencies, more stacked nouns, and more reduced object relatives in conversation; we also demonstrate substantial variation across model families. We propose and test a hypothesis as to why: while human language is optimized for successful communication, AI-generated text is increasingly optimized for other goals like math, code, and reasoning, in which there is a pressure for \textit{information density} rather than successful communication. We provide a proof-of-concept for this hypothesis by fine-tuning a language model under a length penalty, which leads to a divergence from human-like linguistic behavior. Our findings highlight the axes of misalignment between linguistic processing demands in humans and LLMs, and point toward possible remedies.

[10] TrainBridge: Agentic NPU Training Optimization Guided by GPU Compilation Evidence cs.CL | cs.AIPDF

Yupeng Chen, Kunxiang Zhao, Xingjian Wang, Yue Zhao, Yijing Wu

TL;DR: 论文研究如何提升Ascend NPU上的训练效率，提出TrainBridge：一种将成熟GPU编译栈产生的高优化执行作为参考，从中抽取优化意图来指导NPU程序重写，并通过迭代profiling进一步精化的智能体方法。实验覆盖LLM预训练、图像分类、视觉语言训练及Modded-NanoGPT加速记录，显示出显著加速效果。

Details

Motivation: Ascend NPU训练中，算子融合等框架级编译优化仅被部分应用，导致计算碎片化和可避免的执行开销；而基于profiling的调优又依赖昂贵试错。论文希望更高效地把GPU侧成熟优化经验迁移到NPU训练中。

Result: 在4个覆盖LLM预训练、图像分类和视觉语言训练的迁移实验中平均加速1.7倍；在27个具有高度优化GPU实现的Modded-NanoGPT speedrun记录上平均加速近2.0倍。

Insight: 创新点在于以GPU编译优化后的执行作为任务特定参考，抽取优化意图指导NPU程序重写，并结合迭代profiling精化；客观看，其可借鉴之处是利用跨平台编译证据和智能体闭环搜索降低NPU调优试错成本。

Abstract: We study how to improve training efficiency on Ascend NPUs. In the NPU training programs we study, framework-level compilation optimizations such as operator fusion are only partially applied, leaving fragmented computation and avoidable execution overheads. Profiling-guided tuning can address these inefficiencies, but it often requires costly trial and error. To make training on NPUs more efficient, we propose TrainBridge, an agentic approach for transferring GPU optimization guidance to NPU training. TrainBridge uses the highly optimized execution produced by mature GPU compilation stacks as a task-specific reference for NPU tuning. It extracts optimization intents from the GPU training execution to guide NPU program rewrites and further refines the rewritten program through iterative profiling. Our GPU-to-NPU migration experiments show that TrainBridge achieves an average 1.7x speedup across four workloads covering LLM pretraining, image classification, and vision-language training, and an average speedup of nearly 2.0x across 27 Modded-NanoGPT speedrun records with highly optimized GPU implementations.

[11] norregel: a preregistered benchmark of dated numeric values in Norwegian law cs.CLPDF

Eirik Botten Nicolaysen

TL;DR: 论文提出 norregel，一个面向挪威法律中带日期数值的预注册基准，覆盖 NOK 金额、百分比、G 的倍数和年龄限制等 916 个条目，每个条目带有条款的日期化历史。模型需在指定日期回答数值，答案可分类为正确、过时、错误或不可解释；论文报告了两轮预注册评测。第二轮八模型汇总结果支持“模型知识截止后生效的数值更容易被答成过时值”的假设，但该支持在关键敏感性分析下不稳健。

Details

Motivation: 关注 LLM 在挪威成文法、法规和议会决定中回答带日期数值时的可靠性与时效性，尤其是模型是否会把知识截止后生效的法律数值答成旧值。还考察 Bokmål 与 Nynorsk 两种书面挪威语下正确率是否存在差异。

Result: 第一轮中 Opus 5.5 和 Sonnet 4.6 在 880 个主条目上正确率分别为 46.7% 和 40.2%，H1 未能实质检验，H2/H3 无定论。第二轮八模型汇总主检验支持 H1：过时答案比例差 d=17.8 个百分点，95% 区间 [7.7, 28.9]，p=0.0001；但敏感性分析 S8 中 d=3.9，区间 [-3.6, 12.7]，p=0.1754，包含 0，因此作者不认为 H1 支持稳健。H2 的过时答案等于最近不同值比例为 56.1% [46.6, 65.0]，区间包含 0.5；H3 的 Nynorsk 正确率低于 Bokmål 的汇总检验无定论，部分模型如 Haiku 4.5 多数答案不可解释。

Insight: 创新点在于用 Lovdata 开放数据机械抽取带日期历史的挪威法律数值题，并将答案细分为正确、过时、错误和不可解释，同时采用预注册评测流程。客观看，其可借鉴之处是把法律事实的时效性转化为可诊断的基准维度，并通过公开预注册、逐答案分类和敏感性分析提升评测透明度与可复现性。

Abstract: norregel asks language models for numeric values in Norwegian statutes, regulations and parliamentary decisions (NOK amounts, percentages, multiples of the National Insurance basic amount (G) and age limits), each at a stated date. The 916 items were extracted mechanically from Lovdata’s open data, and each carries the dated history of its provision, so an answer can be classed as correct, outdated, wrong or uninterpretable. We report a first and a second preregistered run on the same items. In the first run (five models, listed publicly only after local analysis), H1, that outdated answers are more frequent for values that took effect after a model’s cutoff, could not be tested in substance, and H2 and H3 were inconclusive; Opus 5.5 and Sonnet 4.6 answered 46.7% and 40.2% of the 880 primary items correctly. The second run was preregistered and published before any of its calls, and eight models were analysed. Its pooled primary test supported H1 under the preregistered decision rule (d = 17.8 percentage points, 95% interval [7.7, 28.9], p = 0.0001). In the preregistered sensitivity analysis S8, which counts as outdated only the most recent value that differs from the key, d was 3.9 [-3.6, 12.7] (p = 0.1754), an interval that contains 0. H1 was supported in every other preregistered sensitivity analysis, but S8 is the only one that gives both groups one target value, so we do not regard the support for H1 as robust. The share of outdated answers equal to that most recent value (H2) was 56.1% [46.6, 65.0], an interval that contains 0.5, and the pooled test of a lower share correct in Nynorsk than in Bokmål (H3) was inconclusive. Several models, among them Haiku 4.5, gave mostly uninterpretable answers. Items and keys are withheld; item IDs, per-answer classes, code and results are on Zenodo.

Hong Ting Tsang, Boyi Zhang, Haoyu Huang, Jiaxin Bai, Yufei Li

TL;DR: 论文提出 MOSAIC，一个面向对话式智能体长期记忆与跨模态推理的全模态基准，包含覆盖文本、图像、视频和音频的 923 个问题，基于 11–24 个会话的人设化历史。它不仅评估最终问答分数，还通过匹配诊断协议分别定位表示、检索、传递和阅读阶段的失败。

Details

Motivation: 现有长期记忆基准多局限于文本-图像，或只报告单一端到端分数，难以判断证据在表示、检索、传递或阅读过程中何处丢失。论文希望将全模态长期记忆评估从排行榜分数转变为可分解的失败诊断。

Result: 构建了含空间、时间与韵律信号的 923 题全模态基准，并对空间信息提供较强因果验证、对标注韵律提供选择器级验证。匹配分解发现表示损失依赖模态，即使给定金标准证据阅读残差仍占主导，且模型在可回答问题上得分高但拒答能力弱。

Insight: 创新点在于全模态、多会话长期记忆基准与“冻结问题/阅读器/提示/预算、干预表示/检索/传递”的匹配诊断协议，可系统定位写、取、传、读各阶段失败，并强调用因果验证构造非装饰性多模态信号。

Abstract: Omni-modal assistants must remember not only what users say, but also what they show and convey. Existing long-term memory benchmarks are mostly text-and-image or report a single end-to-end score, so they cannot locate whether evidence is lost when it is represented, retrieved, delivered, or read. We introduce MOSAIC, a benchmark of 923 questions over persona-grounded histories of 11-24 sessions spanning text, image, video, and audio. MOSAIC constructs targeted spatial, temporal, and prosodic signals rather than decorative media, with the strongest causal validation for spatial information and selector-level validation for annotated prosody, and pairs them with a matched diagnostic protocol that freezes the question, reader, prompt, and budget while intervening on representation, retrieval, and delivery. Matched decomposition shows that representation loss is modality-dependent, residual reading dominates even under gold evidence, and abstention remains weak on these controls: strong answerable scores coexist with failures to refuse. MOSAIC thus turns omni-modal memory evaluation from a leaderboard into a diagnosis of write, retrieve, deliver, and read failures.

[13] Hierarchical Masked Diffusion Language Models cs.CL | cs.AIPDF

Parul Gupta, Satwik, Avni Aggarwal, Bhavesh Gurnani, Arnab Kumar Mondal

TL;DR: 论文提出层级掩码扩散语言模型 HMDLM，在标准掩码扩散语言模型基础上为每个 token 位置引入可用的层级标签，并将生成分解为两步：先预测掩码位置的层级分布，再在该条件下预测 token 分布。通过为不同层级设置独立噪声调度，模型可软性、可控地优先揭示高优先级位置，同时保持并行解码，并重新推导了前向/反向过程与 NELBO。

Details

Motivation: 标准掩码扩散语言模型在所有位置共享噪声调度，未显式编码哪些位置应优先解决；但许多任务存在自然优先级，如标题/章节标题引导正文生成、数独中由初始线索直接确定的格子帮助解决更难格子。

Result: 在 LM1B 语言生成上，困惑度比 ADLM 降低超过 1.5；在 WikiHow 上比 ADLM 最多降低 4.4、比 MDLM 最多降低 7；LLM 评审对 HMDLM 生成结果的偏好次数超过基线的 2 倍；在多步推理数独任务上，求解准确率比现有扩散基线最多提升 8.6 个绝对百分点。

Insight: 创新点在于将 token 位置的层级/优先级作为训练数据中零或极少额外标注即可获得的信息，引入两阶段层级感知生成和层级独立噪声调度，从而在并行解码中实现可控的优先揭示顺序；同时给出该分解的数学推导。客观来看，其将任务先验结构显式注入扩散语言模型，兼顾并行性与生成顺序控制，是较有借鉴意义的建模思路。

Abstract: Masked diffusion language models have emerged as an alternative to autoregressive models, enabling parallel token generation. However, standard masked diffusion language models use a shared noise schedule across positions without explicitly encoding which positions should be resolved first. Many tasks exhibit a natural priority among positions: document titles and section headings can guide the generation of body text, while Sudoku cells determined directly from the initial clues can help resolve more difficult cells. To incorporate this notion of priority (or hierarchy) among token positions, we introduce the Hierarchical Masked Diffusion Language Model (HMDLM), which augments each token position with a hierarchy level available as part of the training data, and requires zero or minimal additional annotations. It factors generation into two passes: the first predicts a hierarchy distribution over masked positions, the second predicts token distributions conditioned on it. Giving each hierarchy level its own noise schedule induces a soft, controllable preference for revealing higher-priority positions earlier while keeping decoding fully parallel. Our formulation is mathematically grounded, and we re-derive the forward and reverse processes and the NELBO for this factorization. Our evaluation shows that for the task of language generation on LM1B, HMDLM reduces perplexity by more than 1.5 points over Anchored Diffusion Language Model (ADLM), a recently proposed strong diffusion baseline; on WikiHow, we see a drop in perplexity of up to 4.4 pts on ADLM and up to 7 pts against MDLM. An LLM judge on WikiHow prefers HMDLM generations (the two variants combined) more than $2\times$ as often as the baseline. On the task of solving Sudoku puzzles requiring multi-step reasoning, it improves solution accuracy by up to $8.6%$ absolute points over existing diffusion baselines.

[14] Asking What a Director Sees Can Change What Vision-Language Models Choose cs.CL | cs.CVPDF

Jiaxuan Guo, Xingyu Ren, Kejia Zhang, Jingxin Yang, Youran Sun

TL;DR: 论文研究视觉语言模型在心理理论评测中，先询问“导演能看到什么”再让模型行动是否会改变其选择。在渲染的Director Task中，这种提问会显著提高模型选择导演看不见的隐藏积木的比例，即使模型大多能正确报告导演视角；论文据此提出成对提示基线，用于分离视角报告能力与提问对后续行动的影响。

Details

Motivation: 现有评测常让模型先报告他人视角再替其行动，并把“报告正确但行动错误”视为未利用视角，隐含假设询问报告不会改变行动；论文检验并挑战这一假设，指出评测设计可能存在混淆。

Result: 在渲染Director Task上，先问导演可见哪些积木使六个GPT模型选择隐藏积木的比例提高12到88个百分点，Qwen3.8-27B提高17个百分点；当提示明确说明导演所知和所指时，正确报告却选择隐藏积木的回复少于1%。

Insight: 创新点在于揭示视角报告提问本身会改变后续决策，并提供仅差该问题的配对提示作为基线；客观上提醒心理理论和视角采择基准需控制询问效应，也提示模型可能把请求从导演视角转向自身视角来理解。

Abstract: Vision-language models are now routinely tested on theory of mind: whether they can work out what another person sees, knows, and means. A common design asks the model to report that perspective and then to act for the person, and reads a correct report beside a wrong action as a failure to use the perspective. That reading rests on a quiet assumption: that asking for the report leaves the action unchanged. We show that it does not. In a rendered Director Task, a partition hides the largest block from a director who asks for “the largest block”. Asking first which blocks the director can see raises the share of replies that choose this hidden block by 12 to 88 percentage points on six GPT models and by 17 on Qwen3.8-27B, even though the models mostly report correctly which blocks the director sees. Reasoning traces and prompt rewordings suggest why: the question shifts how the request is read, from the director’s view toward the model’s own. When the prompt states what the director knows and refers to, fewer than 1% of replies pair a correct report with a hidden-block choice. Any evaluation that elicits a perspective report before the action it scores needs a baseline without the question, and we deliver one: paired prompts that differ only in that question, a design any perspective-taking benchmark can adopt.

[15] Policy Central: Detecting Conflicts, Redundancy, and Similarity in Deployed Police Policy cs.CL | cs.CYPDF

Holly Baker, Michael Kuo, Kyana Pollas

TL;DR: 论文介绍已部署的 Policy Central 系统，用于检测警察部门政策文本之间的冲突、冗余、相似与不同关系，并比较 Retrieve-Judge 和 Screen-Confirm 两种架构：前者是检索后分类的低成本全库健康扫描，后者是先筛选再确认的选择性穷举审计。系统在来自约84份真实政策文档的302对基准上评估，说明当前模型在给定文本对时判断较好，但自动找到相关文本对仍是难点。

Details

Motivation: 警察政策库包含近2万条政策段落，现有审查依赖人工通读全库并反复检查，以发现冲突、冗余和相似内容，成本高且难以在每次修订后重新执行，因此需要自动化关系检测来降低审查成本。

Result: 在302对基准上，给定每个文本对时 Screen-Confirm 达到0.795聚合召回，Conflict 类召回0.885；但检索组合召回较低，Retrieve-Judge 降至0.457，Screen-Confirm 为0.472，Conflict 为0.615。

Insight: 创新点在于将政策文本关系检测定义为冲突、冗余、相似、不同四类任务，并提出面向部署场景的双架构：低成本全库筛查与选择性穷举审计相结合。客观来看，其价值不只是模型判断精度，而是用可重复、低成本的扫描改变政策审查流程的经济性。

Abstract: Policy Central is a deployed system that helps police departments find relationships between sections of their written policy: whether two sections conflict, are redundant, are merely similar, or are different. We describe the system and compare two architectures. Retrieve-Judge, a retrieve-then-classify pipeline, runs as a cheap library-wide health signal; Screen-Confirm, a screen-then-confirm cascade, is a selective audit of the documents a reviewer chooses, checking them exhaustively against one another. On a 302-pair benchmark drawn from one department’s ~84 real policy documents, given each pair Screen-Confirm reaches 0.795 aggregate recall and 0.885 on Conflict, so current models judge well. Finding the pair is the hard part: Retrieve-Judge’s composed recall falls to 0.457, and Screen-Confirm recovers 0.472 (0.615 on Conflict). Today these pairs are found by reading the whole library, nearly 20,000 sections, checked against one another and re-checked after every revision. A scan that returns half of them in minutes, at low cost, and can be re-run on demand changes the economics of policy review even before recall improves.

[16] ChronoSynth: Constraint-Aware Question Synthesis over Temporal Knowledge Graphs cs.CL | cs.AIPDF

Feng Liang, Weixin Zeng, Runhao Zhao, Mingjun Zhou, Shaoshi Yang

TL;DR: 论文研究时序知识图谱上的问题生成（Temporal KGQG），将其视为约束保持的图到问题生成任务。提出 ChronoSynth 框架：先构建时序意图状态显式表达问题语义，再从模式记忆库检索类型化问题模式并适配输入图，最后验证生成问题并在检测到不一致时利用反馈修订。

Details

Motivation: 现有时序知识图谱问题生成方法和提示基线虽能生成流畅问题，但可能改变答案域、时间条件、参考事件和时序顺序，导致时序语义漂移；论文旨在解决如何生成同时保持这些关键约束的有效问题。

Result: 在 CRONQUESTIONS 和 MultiTQ 两个数据集上取得 state-of-the-art 的生成性能，消融实验验证了各组件对性能的贡献。

Insight: 创新点在于将时序 KGQG 建模为约束保持生成问题，引入显式 Temporal Intent State、索引化类型问题模式记忆检索与适配，以及基于验证反馈的迭代修订机制；客观看，其可借鉴之处是把意图显式化、模式先验与生成后验证结合来缓解时序语义漂移。

Abstract: Knowledge graph question generation (KGQG) generates natural-language questions from structured graph evidence and a target answer. When the input graph is temporal, a valid question must also preserve the answer field, time condition, reference event, and temporal order that determine the answer set. Existing KGQG methods and prompting baselines can produce fluent questions that alter these conditions, causing temporal semantic drift. This paper studies temporal KGQG as a constraint-preserving graph-to-question problem and proposes ChronoSynth, a memory-augmented and constraint-aware generation framework. ChronoSynth first constructs a Temporal Intent State to make the intended question semantics explicit. It then retrieves compatible typed question patterns from an indexed pattern memory, adapts them to the input graph, and validates the generated question. Validation feedback guides revisions when a supported inconsistency is detected. Experiments on CRONQUESTIONS and MultiTQ demonstrate state-of-the-art generation performance, with ablations supporting the contributions of its components.

[17] Right On Q: Recovering Reasoning under Quantization via Self-Generated Trajectories cs.CL | cs.LGPDF

Boyuan Zheng, Jing Xiong, Chenwei Wu, Hui Shen, Zesen Zhao

TL;DR: 论文提出 OnQ，一种面向低比特大语言模型的推理能力恢复流程：先进行分块校准，再让量化学生模型自生成响应，并在其自身前缀上接受全精度教师模型的 on-policy 蒸馏，同时采用候选级 Dynamic KL 平衡前向与反向 KL。该方法旨在解决固定响应监督带来的训练/生成前缀不匹配，以及学生模型在自身前缀上的概率分布错配问题。实验覆盖 Qwen3-1.7B、4B、8B 的 2-bit 和 3-bit 量化，在九个基准上取得最高平均表现，长程推理提升尤其明显。

Details

Motivation: 现有量化感知训练通常用全精度模型在固定响应序列上监督量化学生，导致训练前缀与学生生成时遇到的前缀不一致，且响应越长轨迹偏差越大；同时学生在其自身前缀上会低估或高估部分候选词，在高熵前缀处分布错配更频繁。论文希望恢复低比特模型的推理能力，并减少其在长程推理中无法在 token 预算内终止的问题。

Result: 在 Qwen3-1.7B、4B、8B 的 2-bit 和 3-bit 量化设置下，OnQ 在九个基准上取得最高平均成绩，并在长程推理基准上提升最大。2-bit 设置下，它在所有模型规模、所有基准上均优于所有基线，Qwen3-8B 平均比最强基线提升 10.26 分，同时生成响应更短、截断和重复错误更少。

Insight: 创新点在于用学生自生成轨迹做 on-policy 蒸馏，使监督前缀与推理时前缀一致；并提出候选级 Dynamic KL，按每个候选的教师-学生概率比加权前向和反向 KL，以同时纠正低估和高估。该思路将轨迹分布对齐与候选级概率校正结合，对低比特 LLM 的推理恢复具有借鉴意义。

Abstract: Quantization-aware training (QAT) aims to recover reasoning capabilities in low-bit large language models (LLMs). Existing recipes typically supervise the quantized student with the original full-precision model on fixed response sequences. However, the resulting low-bit quantized models still struggle on long-horizon reasoning tasks and often fail to terminate within a token budget that suffices for the full-precision model. We identify two challenges in closing this gap. First, fixed-response supervision leaves a mismatch between training prefixes and those encountered during student generation, and this trajectory divergence grows as responses lengthen. Second, on its own prefixes, the student underpredicts some next-token candidates and overpredicts others, and this distributional misalignment becomes more frequent at high-entropy prefixes. We present OnQ, a low-bit recovery pipeline that combines on-policy distillation and candidate-wise Dynamic KL. Starting from block-wise calibration, the quantized student generates responses and receives full-precision teacher supervision on its own prefixes. Dynamic KL weights forward and reverse KL by each candidate’s teacher-student probability ratio to balance correction of under- and overprediction. Across Qwen3-1.7B, 4B, and 8B at 2- and 3-bit precision, OnQ achieves the highest nine-benchmark average, with the largest gains on long-horizon reasoning benchmarks. At 2 bits, it outperforms every baseline on every benchmark at every model size, improving the Qwen3-8B average by 10.26 points over the strongest baseline. It also produces shorter responses with fewer truncations and repetitive errors.

[18] Omnilingual-MsASR: Benchmarking Multimodal, Multilingual, Long-Form Multi-Speaker ASR of Omni Large Language Models cs.CLPDF

Yuan Ge, Pei Zhang, Xize Cheng, Lingzhou Mu, Mingyang Wang

TL;DR: 论文提出 Omnilingual-MsASR，一个面向全模态大语言模型的多语言、多模态、长时多说话人自动语音识别基准。该基准覆盖六种语言、最长30分钟录音、最多20名说话人，并提供平行音频与视频输入，用于评测模型在“谁在何时说了什么”任务上的跨模态一致性与互补线索利用能力。

Details

Motivation: 现有全模态大模型在长时、多说话人视频理解中仍难以完成整段视频的说话人归属与时间定位，且缺乏系统基准来刻画其能力上限并指导后续模型发展。

Result: 对当前领先全模态大模型进行广泛评测后发现显著性能瓶颈，tcpWER 经常超过100%，表明长时多说话人ASR任务仍极具挑战。

Insight: 创新点在于构建多语言、多模态、长时、多说话人的MsASR基准，并利用平行音视频输入评估跨模态一致性与协同利用；客观来看，其价值主要在于揭示细粒度时间定位、说话人消歧和跨模态协同是未来全模态基础模型的关键挑战。

Abstract: Omni-modal large language models (LLMs) have achieved remarkable progress in audio and video understanding. However, existing models struggle with long-form, multi-speaker videos, especially in understanding the entire video and recognizing who says what and when. To characterize the capability limit of current omni LLMs and guide future model development, we introduce Omnilingual-MsASR, a multilingual, multimodal benchmark for long-form Multi-speaker Automatic Speech Recognition (MsASR) covering six languages, recordings up to 30 minutes, and scenarios with up to 20 speakers. Omnilingual-MsASR provides parallel audio and video inputs to evaluate models’ cross-modal consistency and their ability to exploit complementary cues across modalities. Extensive evaluations of leading omni-modal LLMs reveal substantial performance bottlenecks, while their tcpWERs frequently exceed 100%. These findings highlight fine-grained temporal localization, speaker disambiguation, and cross-modal synergy as key challenges for future omni-modal foundation models. Our work is available at https://github.com/QwenLM/Omnilingua-Bench/tree/main/Omnilingua-MSpeaker.

[19] OmniSimulator: Aligning Small Language Models for Authentic Heterogeneous Behavior Modeling cs.CL | cs.AI | cs.LGPDF

Jiawei Chen, Ruoxi Xu, Boxi Cao, Zhuoqun Li, Ruotong Pan

TL;DR: 论文提出OmniSimulator训练框架，用于提升小型大语言模型在短视频平台场景中的真实用户模拟能力。该方法通过五维结构化模式组织行为导向推理，从真实用户行为场景中筛选数据，并利用教师模型生成高质量推理轨迹来训练小模型。基于200名真实用户三个月交互轨迹构建了9,881条训练实例，实验显示Qwen3-8B平均分提升34.33%，性能可与Claude-Sonnet-4.5竞争，并支持小规模自托管部署。

Details

Motivation: 现有LLM作为用户模拟器时，原生思维链容易退化为浅层模式匹配，难以建模真实用户决策背后的行为因素，因此需要让小型LLM学习行为相关的推理模式，而不是依赖表面行为相关性。

Result: 摘要报告OmniSimulator使Qwen3-8B平均分提升34.33%，达到与Claude-Sonnet-4.5有竞争力的水平，并支持更小规模的自托管部署；实验基于200名真实用户三个月历史交互轨迹构建的9,881条训练实例。

Insight: 创新点在于用五维结构化模式组织行为导向推理，结合真实用户行为数据筛选和教师模型蒸馏高质量推理轨迹，使小模型学习可解释的行为推理模式而非浅层行为关联，从而提升用户模拟的真实性与可部署性。

Abstract: Large language models (LLMs) have shown great potential as general-purpose user simulators for interactive systems. Prior studies have demonstrated that training on high-quality data can substantially improve model reasoning and downstream task performance. However, the upper bound of user simulation is still constrained by the model’s reasoning paradigm: native chain-of-thought often degenerates into shallow pattern matching and fails to adequately model the behavioral factors underlying real user decisions. In this paper, we propose OmniSimulator, a novel training framework designed to improve the predictive ability of small LLMs for realistic user simulation on short-video platform environments. Specifically, OmniSimulator organizes behavior-oriented reasoning through a structured schema with five dimensions, carefully selects data from real user-behavior scenarios, and introduces a teacher model to generate high-quality reasoning traces under the proposed schema, helping small models learn behavior-relevant reasoning patterns rather than relying on superficial behavioral correlations. Based on this process, we construct 9,881 high-quality training instances from three months of historical interaction trajectories of 200 real users, averaging about 50 distilled examples per user. Experimental results show that OmniSimulator improves the average score of Qwen3-8B by 34.33%, achieves performance competitive with Claude-Sonnet-4.5, and supports smaller-scale, self-hosted deployment. These results suggest that learning structured behavior-oriented reasoning patterns is a promising direction towards scalable and high-fidelity LLM-based user simulation.

[20] TenetBench: A Clause-Anchored Russian Corpus for Evaluating Language Models against Professional Codes of Ethics cs.CLPDF

Yegor Anichkov, Valery Vinogradov, Andrey Furov, Marina Polukoshko, Sergey Bolovtsov

TL;DR: 论文提出TenetBench，一个条款锚定的俄语语料库，用于评估大语言模型是否遵守职业伦理准则。该语料库将3,433个与15份俄语职业准则条款冲突的请求与仅合法性不同的良性孪生请求配对，并提供自动生成流水线和人工验证。作者评估14种模型与推理模式组合，发现多数模型常生成冲突内容且很少警告。

Details

Motivation: 现有评估多报告宽泛的伤害类别，无法揭示大语言模型助手违反职业伦理准则的频率以及具体违反哪一条款，用户可能在未察觉时违规或被协助违规。因此需要细粒度、条款级、可诊断的伦理合规评测。

Result: 构建了含3,433个冲突请求及良性孪生的TenetBench；人工验证中专家认为0.995的冲突任务合理，模型裁判与裁定标签在1,860个元素上的Gwet’s AC1至少为0.894。评估14个模型与推理模式组合发现，11个在大多数冲突请求上生成冲突内容，且多数无警告，近四分之三的警告伴随冲突内容；更合规的模型通过拒绝实现，但仍服务合法孪生，启用推理使两个模型的拒绝率减半。

Insight: 创新点在于条款锚定、冲突请求与良性孪生配对、从职业准则文本自动生成可扩展任务，并进行跨语言和跨司法辖区验证。客观来看，该工作把伦理合规评估从宽泛危害类别推进到具体条款和可诊断行为，并控制请求合法性变量，为特定职业或组织构建诊断数据集提供了可借鉴范式。

Abstract: Codes of professional ethics guide and restrict specialists, who increasingly work with assistants built on large language models. Such an assistant can breach the user’s code, or help the user breach it, without the user noticing. Evaluations reporting broad categories of harm cannot show how often this happens or which provision is breached. TenetBench pairs 3,433 requests conflicting with clauses of fifteen Russian-language professional codes with benign twins that differ only in the legitimacy of what is asked. An automated pipeline generates the tasks from the text of a code, so the corpus grows with every code added and a provision yields new tasks on every run. It also produced tasks and human-validated verdicts from two English codes of other jurisdictions. In human validation of the frozen pipeline, specialists rated 0.995 of the conflicting tasks plausible, and the model judge agreed with the adjudicated label at Gwet’s AC1 of at least 0.894 on 1,860 elements. We evaluated fourteen model and reasoning-mode combinations on whether they produce the conflicting text, decline or warn. Eleven of the fourteen produce conflicting content on most such requests, mostly without a warning, and nearly three warnings in four accompany conflicting content. The models that conform more do so by declining while still serving the legitimate twin, and enabling reasoning halved the refusals of two models. We hope the dataset and pipeline will help evaluate how language models observe professional norms and let diagnostic datasets be built for particular professions and organisations.

[21] A Strong MoE Is All You Need for Process Supervision cs.CL | cs.AIPDF

Guanzhi Deng, Zengyan Liu, Haibo Wang, Kuan Wu, Shing Yin Wong

TL;DR: 论文提出利用Mixture-of-Experts（MoE）模型中的专家路由信息作为过程监督信号，以缓解过程奖励模型（PRM）获取可靠过程标签困难的问题。作者基于路由统计构造过程置信度，提出RoCo-Rethink在低置信推理步骤后选择性重生成后续轨迹，并提出RoCo-PRM用强MoE自动标注推理步骤分数并蒸馏为独立PRM。实验表明两类方法在多种MoE架构和推理任务上优于强基线，且RoCo-PRM具有跨架构和跨任务迁移能力。

Details

Motivation: 现有PRM依赖人工标注、rollout估计或辅助验证来获得过程标签，成本高且可靠性受限。论文希望利用MoE模型原生暴露的专家路由信息，提供一种无需额外标注的过程监督来源。

Result: 在多个MoE架构和多样推理任务上，RoCo-Rethink与RoCo-PRM均一致优于强基线；RoCo-PRM还能有效跨模型架构和任务迁移。摘要未给出具体数值或benchmark名称。

Insight: 创新点在于将专家路由从计算调度机制转化为架构原生的过程置信信号，并结合总体路由质量、head路由质量和步内路由变化构造路由置信度；进一步用于低置信步骤的选择性重生成和自动过程标注蒸馏。客观看，其价值在于减少对人工过程标签的依赖，并展示MoE路由可作为推理过程监督的自监督信号。

Abstract: Process reward models (PRMs) provide fine-grained supervision for multi-step reasoning, yet obtaining reliable process labels remains a central bottleneck. Existing approaches typically rely on human annotation, rollout-based estimation, or auxiliary verification. In this work, we explore a different source of process supervision natively exposed by Mixture-of-Experts (MoE) models: expert routing. We show that simple routing statistics provide an intrinsic process-confidence signal that is competitive with conventional output-logit confidence. Building on this observation, we construct routing confidence by combining overall routing mass, head routing mass, and within-step routing variation, and introduce RoCo-Rethink, which identifies the least confident reasoning step and selectively regenerates the subsequent trajectory. We further introduce RoCo-PRM, which uses a strong MoE to automatically annotate reasoning trajectories with routing-derived step scores and distills them into a standalone PRM. Across multiple MoE architectures and diverse reasoning tasks, both RoCo-Rethink and RoCo-PRM consistently outperform strong baselines, while RoCo-PRM transfers effectively across both model architectures and tasks. Our findings suggest that expert routing is more than a computational dispatch mechanism: it provides an architecture-native source of process supervision for reasoning.

[22] Keeping Time vs. Understanding Time: Neither Alone Explains Temporal Reasoning in LLMs cs.CL | cs.AIPDF

Auss Abbood, Fernando Martínez-Plumed, José Hernández-Orallo, Nigel Collier

TL;DR: 论文直接检验了LLM时间推理中“时间算术 vs. 时间语义”二分法的有效性。作者通过跨数据集迁移实验和基于认知需求的模型评估发现：按该二分法划分的任务并不能带来稳定的同模式迁移能力，且真正驱动基准表现的主要是知识而非定量/逻辑推理。因此，当前时间推理基准并未清晰实现算术-语义区分，而更多是在奖励模型记住时间通常如何运作。

Details

Motivation: 时间推理常被假定可分解为时间算术和时间语义两种模式，这一假设影响了基准设计、模型训练和进展评估，但其经验有效性此前未被直接检验。论文旨在验证该二分法是否真实反映LLM的能力结构。

Result: 在标注为算术或语义的子集上微调主要带来数据集特定增益（+6.8至+23.1个百分点），而同模式跨数据集迁移仅为-3.2至+1.1个百分点，几乎没有一致可迁移能力证据；认知需求评估还发现定量和逻辑推理并非主要驱动因素，知识才是。

Insight: 创新点在于直接挑战时间推理的算术-语义二分假设，并用跨数据集迁移和认知需求评估两条证据链揭示基准标签与真实能力不匹配。可借鉴之处是：评估LLM能力时应检验迁移性与实际驱动因素，而非仅依赖任务表面分类。

Abstract: Temporal reasoning is widely assumed to decompose into two modes: temporal arithmetic (date and time calculations), and temporal semantics (understanding temporal and causal relations). This dichotomy shapes how benchmarks are designed, how models are trained, and how progress is measured, yet its empirical validity has never been directly tested. We provide two lines of evidence that challenge it. First, we test whether tasks labelled as arithmetic or semantic by the benchmarks support transferable gains within their purported mode. Finetuning on these subsets yields predominantly dataset-specific gains ($+6.8$ to $+23.1$ percentage points), while transfer to a different dataset in the same mode ranges from $-3.2$ to $+1.1$ percentage points. Across the two benchmark families used in the transfer analysis, this pattern provides little evidence of coherent, transferable temporal abilities. Second, using cognitive-demand-based model evaluation, we examine which capabilities actually drive performance on these benchmarks. We find that quantitative and logical reasoning, the assumed proxies for temporal arithmetic and temporal semantics, are not the primary drivers; knowledge is. Together, these findings show that current temporal reasoning benchmarks do not cleanly realise the arithmetic-semantics distinction and instead largely reward models for remembering how time typically works.

[23] Mitigating Query Shortcuts in Chain-of-Thought with Context-Aware Decoding cs.CL | cs.AI | cs.LGPDF

Shin’ya Yamaguchi

TL;DR: 论文研究CoT推理中模型可能依赖查询相关捷径而非生成理由的问题，将上下文感知解码（CAD）用于CoT，提出理由对比解码（RCD）：通过对比有/无理由条件下的下一词分布，增强受理由影响的token，并从KL正则化奖励最大化角度给出理论解释。

Details

Motivation: 现有CoT虽提升大模型推理表现，但最终答案未必真正依赖生成的理由，可能利用查询特定捷径，因此需要一种简单方法来缓解该问题。

Result: 在五个推理基准上，RCD相比标准CoT提升了答案准确率；使用真实理由和随机理由的干预分析表明，RCD增强了最终答案对理由内容的行为依赖。

Insight: 创新点在于将对比解码引入CoT理由利用，提出无需训练的RCD，并通过理论解释和干预分析验证其能缓解查询捷径、提升答案对理由的依赖。

Abstract: While Chain-of-Thought (CoT) prompting improves the reasoning performance of large language models (LLMs), recent studies have shown that final answers do not always depend on the generated rationales. In particular, LLMs may rely on query-specific shortcuts rather than the information provided by their reasoning steps. In this paper, we investigate whether context-aware decoding (CAD), an existing contrastive decoding method, can mitigate such shortcut behavior when applied to CoT reasoning. We study this application as Rationale Contrastive Decoding (RCD), which contrasts the next-token distributions conditioned on the query with and without its rationale, amplifying tokens whose likelihood increases due to the rationale. We provide a theoretical interpretation of this decoding rule via a KL-regularized reward-maximization objective. Experiments across five reasoning benchmarks demonstrate that RCD improves answer accuracy over standard CoT. Furthermore, intervention analyses using ground-truth and random rationales reveal that RCD increases the behavioral dependence of final answers on the rationale content. Our findings suggest that contrastive decoding offers a simple, training-free approach to mitigating query shortcuts.

[24] RCSD: Reinforcement and Continual Learning for Domain-Specific Speculative Decoding cs.CLPDF

Mingkang Liu, Weiming Meng, Peipei Jiao, Jiawei Zhang, Tianyuan Liu

TL;DR: 论文提出RCSD，一种面向领域特定投机解码的训练框架，通过强化与持续学习优化草稿模型，使其目标从模仿目标模型完整输出分布转向最大化系统级接受长度；利用投机推理中自然产生的接受/拒绝序列构建隐式偏好对，并采用掩码损失避免惩罚与目标模型一致的token。

Details

Motivation: 传统投机解码的草稿模型常用软标签蒸馏训练，会迫使草稿模型拟合目标模型的高熵区域，导致接受率下降和早期拒绝风险增加，与投机解码真正追求的接受长度和推理加速目标不一致。

Result: 在多个基准上，RCSD相比HASS平均接受长度提升35%，在实际工作负载下端到端推理速度达到HASS的1.35倍，且不损害生成质量。

Insight: 创新点在于将草稿模型训练与投机解码的系统级目标对齐，利用推理过程中自然出现的接受和拒绝序列构造隐式偏好对，并用掩码损失避免错误惩罚目标对齐token；从客观看，其可借鉴之处是把在线推理反馈转化为训练信号，以接受长度而非分布模仿作为优化目标。

Abstract: Speculative decoding accelerates large language model inference by employing a compact draft model to propose candidate tokens that are verified in parallel by a target model. Conventional training relies on soft-label distillation, which forces the draft model to mimic the target’s full output distribution, including high-entropy regions that degrade acceptance rates and increase early rejection risk. We propose RCSD (Reinforcement and Continual Learning for Speculative Decoding), a training framework that aligns draft model optimization with the system-level objective of maximizing acceptance length. RCSD leverages naturally occurring accepted and rejected sequences during speculative inference to construct implicit preference pairs, with masked loss to avoid penalizing target-aligned tokens. Evaluated across multiple benchmarks, RCSD improves average acceptance length by 35% over HASS and achieves 1.35x faster end-to-end inference than HASS on practical workloads, without compromising generation quality.

[25] Uni-Mem: Learning a Long-Horizon Agentic Memory Manager with Unified Online Policy Optimization cs.CLPDF

Zhiqian Xia, Jielong Tang, Ling-An Zeng, Shixiong Kai, Rong Yao

TL;DR: 论文提出 Uni-Mem，一个可端到端在线强化学习训练的统一长期记忆管理器，使智能体同时学习记忆构建与检索，并共享同一策略。针对 GRPO 等现有 RL 在长时记忆管理轨迹中存在的组间混淆比较和反馈稀疏问题，作者提出 Stage-Decoupled GRPO 交替优化构建与检索阶段，并控制另一阶段带来的变化，以产生阶段特定学习信号；同时引入 Event-Wise Credit Assignment，将反馈定位到关键事件并将事件级优势传播到具体操作。

Details

Motivation: 现有学习式智能体记忆方法要么只优化记忆构建，要么将检索侧在线 RL 与离线构建监督结合，导致构建与检索两个阶段难以协同适配。论文旨在用统一在线 RL 联合学习记忆构建和检索。

Result: 在 LoCoMo 和 LongMemEval 两个基准上，Uni-Mem 优于最强的基于 RL 的记忆方法，验证了 Stage-Decoupled GRPO 与 ECA 联合学习记忆构建和检索的有效性；摘要未给出具体数值。

Insight: 创新点在于用共享策略端到端联合学习记忆构建与检索，并提出 Stage-Decoupled GRPO 通过交替阶段优化和组内控制变量获得更清晰的阶段学习信号；ECA 则将稀疏反馈细化到事件级并传播到组成操作，为长时程智能体记忆管理提供更有效的信用分配。

Abstract: Long-term agent memory is shifting from predefined workflows toward learnable policies for memory construction and retrieval. Existing learning-based methods either optimize construction alone or combine retrieval-side online reinforcement learning (RL) with offline construction supervision, limiting co-adaptation between the two stages. In this paper, we explore end-to-end online RL for a unified memory manager (Uni-Mem) that learns both capabilities under a shared policy. However, joint optimization with existing RL algorithms such as GRPO faces confounded group comparisons and sparse feedback over long memory-management trajectories. To this end, we propose Stage-Decoupled GRPO, which alternates construction and retrieval optimization while controlling variation from the other stage within each rollout group, yielding stage-specific learning signals. We further introduce Event-Wise Credit Assignment (ECA), which localizes feedback to responsible events and propagates event-level advantages to their constituent operations. Experiments on LoCoMo and LongMemEval show that Uni-Mem outperforms the strongest RL-based memory, demonstrating the effectiveness of jointly learning memory construction and retrieval via Stage-Decoupled GRPO and ECA.

[26] Evolving with Open Worlds: Towards Recursive Self-Improvement for Future Forecasting cs.CLPDF

Zhipeng Chen, Yunsen Liang, Yanxi Chen, Hongkang Zhao, Lingyun Feng

TL;DR: 论文提出DeepEvolve，一个面向未来事件预测的可持续进化智能体系统。它让智能体在动态、可验证、自然落地的开放环境中持续交互，并通过递归进化引擎将预测推理轨迹转化为能力，实现参数级进化和可复用技能蒸馏。实验显示其在FutureX排行榜连续三周排名第一，并在合成评估中于四个基础LLM上进入前三。

Details

Motivation: 现有AI系统通常参数冻结、依赖外部更新，而理想系统应能从自身经验中持续提升，即递归自我改进。研究难点在于需要持续变化、客观可验证且自然落地的环境来支撑智能体进化。

Result: 在按真实世界结果评估且每周更新的FutureX排行榜上，自改进预测器连续三周在所有系统中排名第一，rating最高提升195%。在合成评估中，该方法在四个基础LLM上均进入前三，最多超过最强基线10.7个Bradley-Terry点。

Insight: 创新点在于将未来事件预测作为开放世界中的可持续进化任务，构建活环境合成可验证任务，并用递归进化引擎把预测推理轨迹转化为能力；其中解耦搜索与预测以适应易污染的实时环境，通过技能蒸馏从有限在线轨迹中提炼可复用能力，体现智能体与开放世界共进化的思路。

Abstract: Most contemporary artificial intelligence (AI) systems are released with frozen parameters and improve through external updates, whereas an ideal system would keep improving from its own experience, a capacity we call recursive self-improvement (RSI). Yet studying RSI requires an environment that is continuously changing, objectively verifiable, and naturally grounded, posing a fundamental challenge for agent evolution. Here we introduce DeepEvolve, a sustainable evolution agent system grounded in future event prediction, in which agents continuously interact with a dynamically evolving environment and improve by internalizing and externalizing the experience gained through these interactions. A living environment synthesizes verifiable tasks, and a recursive-evolution engine turns the reasoning trajectory behind each prediction back into capability: decoupling search from prediction adapts parameter-level evolution to the contamination-prone live environment, while distilling reusable skills lets the agent evolve from the limited live trajectories such an environment affords. On the external FutureX leaderboard, which evaluates predictions against real-world outcomes and is updated weekly, our self-improved forecaster ranked first across all systems over three consecutive weeks, achieving up to a 195% improvement in rating. On our synthesized evaluation, it ranked among the top three across four base LLMs, outperforming the strongest baseline by up to 10.7 Bradley–Terry points. More broadly, these results suggest that a system built for self-improvement is best understood as an agent that co-evolves with a genuinely living, open-ended environment, pointing to a shift in AI from training static models on fixed data toward cultivating agents that grow stronger from their own experience, with such an open world as the indispensable basis of that growth.

[27] MAPS: Macro-Adaptive Parallel Reasoning for Regulating Search Momentum in Retrieval-Augmented Generation cs.CLPDF

Xinze Li, Dingwen Zhang, Zhenghao Liu, Yukun Yan, Shuo Wang

TL;DR: 论文提出MAPS，一种用于调节搜索动量的宏自适应并行推理框架，以改进检索增强生成。它先保留检索知识和精炼推理状态，减少历史推理轨迹对后续子查询生成的影响，再将顺序推理扩展为并行推理：每轮识别互补检索方向、并行生成子查询并整合证据更新状态。实验在六个问答基准上显示其在不同骨干模型上一致优于代表性RAG方法。

Details

Motivation: 现有交织推理与检索的RAG方法易产生冗余或无效子查询，查询探索多样性不足，且推理过程演化如何影响查询意图尚不清晰；论文希望刻画并调节搜索动量，缓解查询意图漂移。

Result: 在六个问答基准上，MAPS在不同骨干模型下均一致优于代表性RAG方法；摘要未给出具体数值，也未明确宣称达到SOTA。

Insight: 创新点在于提出搜索动量来刻画推理轨迹中查询意图的累积演化，并设计宏自适应并行推理框架，通过保留精炼状态和并行探索互补检索方向来调节动量、减轻意图漂移；可借鉴之处是将顺序检索推理改为多方向并行探索并整合证据。

Abstract: Retrieval-Augmented Generation (RAG) improves the performance of Large Language Models (LLMs) by retrieving external knowledge. Recently, some RAG methods interleave reasoning and retrieval, enabling LLMs to progressively generate sub-queries and retrieve external knowledge. However, these methods may suffer from redundant or ineffective sub-queries and insufficiently diverse query exploration. Moreover, due to the tightly interleaved nature of reasoning and retrieval, it remains unclear how the evolving reasoning process influences query intent. To characterize the evolution of query intent during reasoning, we introduce search momentum, which captures the accumulated retrieval tendency induced by historical changes in generated sub-queries along the reasoning trajectory. Through momentum-based analysis, we find that the query generation process tends to continuously evolve along existing retrieval trajectories. Based on this observation, we propose MAPS, a Macro-Adaptive Parallel Reasoning framework for regulating search momentum and mitigating query-intent drift. MAPS first retains the retrieved knowledge and a refined reasoning state, thereby reducing the influence of historical reasoning trajectories on subsequent sub-query generation. Subsequently, MAPS extends the conventional sequential reasoning process into a parallel reasoning paradigm. At each iteration, MAPS identifies complementary retrieval directions, generates corresponding sub-queries for parallel retrieval, and consolidates the retrieved evidence to update the knowledge and reasoning states. Experiments on six question-answering benchmarks demonstrate that MAPS consistently outperforms representative RAG methods across different backbone models. All code and resources are available at https://github.com/NEUIR/MAPS.

[28] Benchmarking Large Language Models for Detecting Satirical Code-Mixed Sentences cs.CLPDF

Răzvan-Alexandru Smădu, Ioana Buhnila, Daniela Tanase, Dumitru-Clementin Cercel, Florin Pop

TL;DR: 论文研究罗马尼亚语与英语、德语、法语、意大利语混合的讽刺句检测，提出四种合成语码混合句生成方法（随机替换、短语翻译、依存句法分析和LLM生成），构建超过60万样本的CS-SeLeRoSa数据集及人工裁决的CS-SeLeRoSa-Gold基准，并系统评测多语言大模型在零样本、思维链和微调设置下的表现。

Details

Motivation: 语码混合在罗马尼亚语等低资源语言中对高层推理任务的影响尚未充分探索，而讽刺检测依赖细微的文化与句法线索，大语言模型难以捕捉，因此需要构建基准并评估模型能力。

Result: 旧版开源模型性能显著下降至37%-55% macro-F1；Gemma 4 31B零样本达68.1%，接近GPT-5 mini的69.5%；微调后Gemma 4 31B在CS-SeLeRoSa上达77.5%、在Gold上达72.6%，优于前沿模型零样本基线。

Insight: 创新点在于系统比较四种合成语码混合句生成方法，构建大规模合成语料与人工裁决金标准，并揭示微调可显著减少格式错误、提升讽刺检测性能；可借鉴之处是面向低资源语言的高层语义任务构建可控合成基准并综合评估多种推理与微调范式。

Abstract: Code-switching is a natural linguistic phenomenon, yet its impact on high-level reasoning tasks remains largely unexplored in under-resourced languages like Romanian. Natural language processing tasks, such as satire detection, rely heavily on nuanced cultural and syntactic cues that large language models struggle to capture. In this paper, we investigate four methods (i.e., random replacement, phrase translation, dependency parsing, and large language models) for generating synthetic code-mixed Romanian sentences with English, German, French, and Italian. We start with an existing dataset for satire detection, compiled from Romanian news articles and annotated at the sentence level by humans, and augment it with over 600k samples, namely CS-SeLeRoSa. We provide an in-depth analysis of the dataset generation process. Then, we evaluate several multilingual large language models across zero-shot, chain-of-thought, and fine-tuning regimes on both the synthetic corpus and a human-adjudicated gold benchmark, namely CS-SeLeRoSa-Gold. While older open models suffer substantial degradation (dropping to 37%-55% macro-F1), newer models like Gemma 4 31B achieve 68.1% zero-shot, rivaling frontier models like GPT-5 mini (69.5%). Furthermore, fine-tuning significantly reduces formatting errors and improves the results, increasing Gemma 4 31B to 77.5% on CS-SeLeRoSa and 72.6% on the gold standard, outperforming frontier zero-shot baselines.

[29] Graph Topology Meets Energy-Driven Pathfinding: Preserving Contextual Consistency in Retrieval-Augmented Generation cs.CL | cs.AIPDF

Yufeng Li, Zaijing Li, Fuchen Zhang, Zongzong Wu, Ming Zhao

TL;DR: 论文提出ThreadRAG，一种基于语义拓扑的RAG框架，用于多跳问答中的上下文保持型知识结构化与模态原生证据推理。该方法通过语义拓扑映射器将检索文档组织为上下文完整的语义节点，并构建包含叙事、语义和词汇边的语义拓扑图，再利用能量驱动路径查找识别与查询相关的子图，并将其确定性线性化为连贯的文本证据路径。

Details

Motivation: 现有结合知识图谱的RAG在多跳问答中受限于知识结构化和推理过程：离散符号化会造成语义丢失和叙事连贯性破坏，碎片化图结构与LLM预训练的连续文本存在模态不匹配，而改写成自然语言又可能引入生成噪声并遗漏信息。

Result: 在三个多跳问答基准上，ThreadRAG一致优于强基线；在HotpotQA上相较文本中心方法取得14.4%的EM提升，在2WikiMultihop上超过最佳基线5.8% EM，同时将token成本降至最强基线的23.3%。

Insight: 论文宣称的创新点在于用语义拓扑图保留上下文完整性，并通过能量扩散进行查询相关子图查找，再将子图确定性线性化为连贯文本证据路径，以缓解语义丢失和模态不匹配。客观来看，其可借鉴之处是将图结构证据以模态原生方式转化为文本路径，在保持推理连贯性的同时降低生成噪声和token开销。

Abstract: Retrieval-augmented generation (RAG) frequently integrates knowledge graphs (KGs) for complex multi-hop question answering. However, their performance is increasingly bottlenecked by the knowledge structuring and reasoning processes. (i) Severe semantic loss: Distilling continuous texts into discrete symbols strips away contextual nuances and narrative coherence. (ii) Pronounced modality mismatch: Fragmented graph structures clash with the continuous texts LLMs are pre-trained on, while rewriting them into natural language introduces generative noise and omits information. To address these limitations, we propose ThreadRAG, a semantic-topology-based RAG framework for context-preserving knowledge structuring and modality-native evidence reasoning. ThreadRAG introduces a Semantic Topology Mapper (STM), which organizes retrieved documents into context-intact Semantic Nodes and constructs a Semantic Topology Graph (STG) through narrative, semantic, and lexical edges. It further incorporates Energy-Driven Pathfinding (EDP), which identifies query-relevant subgraphs via energy diffusion and deterministically linearizes them into coherent textual evidence paths. Experiments on three multi-hop QA benchmarks demonstrate that ThreadRAG consistently outperforms strong baselines. Notably, it achieves a 14.4% EM improvement over text-centric methods on HotpotQA and surpasses the best baseline by 5.8% EM on 2WikiMultihop, while reducing token cost to 23.3% of the strongest baseline.

[30] CulturalRiddles: A Multicultural Benchmark of Riddles cs.CL | cs.AIPDF

My Chiffon Nguyen, Tianyi Hu, Vinod Anbalagan, Henry Gagnier, Pouya Sadeghi

TL;DR: 论文提出 CulturalRiddles，一个由人类通过社区参与式方式原创构建的多文化谜语基准，覆盖62种文化（45个国家、51种语言）。该基准强调文化根基、溯因与组合推理，以及事实、社会知识、语言能力和文化特定联想，并用自动评估和大规模人工评估测试了28种LLM配置。

Details

Motivation: 现有LLM文化理解评测多偏英语中心、文化覆盖有限，且更关注事实知识而非在真实文化语境中的应用能力；论文希望更全面地评估模型对多元文化的理解与表征。

Result: 论文评测了28种LLM配置，结合自动评估和大规模人工评估；结果显示即使前沿模型在某些社区上的准确率最高也只有22%，表明多文化谜语理解仍具挑战。

Insight: 创新点在于以社区参与方式构建深度扎根于各文化的原创谜语基准，覆盖多国家多语言，并要求溯因推理、组合推理及文化特定联想；同时开放贡献流程以扩展语言和文化覆盖。客观看，其价值在于将文化评测从事实问答推进到需要推理与文化语境结合的任务。

Abstract: As LLMs expand their language coverage, evaluating how well they understand and represent diverse cultures becomes increasingly important. However, many benchmarks remain English-centric, with limited cultural coverage and a focus on factual knowledge over its application in real-world cultural contexts. We propose CulturalRiddles, a benchmark of original, culturally grounded riddles created entirely by humans through a participatory community effort, spanning 62 cultures (45 countries, 51 languages). These riddles are deeply rooted in their respective cultures and require abductive and compositional reasoning, alongside detailed factual and social knowledge, language skills, and an understanding of culturally specific associations. We benchmark 28 LLM configurations using both automatic and large-scale human evaluation. Even frontier models achieve at most 22% accuracy in some communities. We publicly release all artifacts alongside an open contribution process to expand the benchmark’s linguistic and cultural coverage.

[31] Remembering Through Change: Provenance-Preserving Hypergraph Memory for Multimodal Agents cs.CLPDF

Jielong Tang, Ling-An Zeng, Xujie Yuan, Xiao Dong, Zhiqian Xia

TL;DR: 论文提出 MHG-Mem，一种面向多模态智能体的多模态超图长期记忆，用来同时保留实体身份、状态演变以及对视觉证据的来源追踪。该方法将实体提及和观测事实作为节点，用出现超边关联同一身份，用演化超边和证据超边连接状态与来源，并配合基于超图的智能体搜索进行跨层检索和细粒度视觉验证。

Details

Motivation: 现有多模态长期记忆要么将图像压缩为文本而丢失细粒度视觉判别信息，要么保留原始像素但难以稳定关联实体身份及其跨会话状态变化。论文旨在解决长期记忆中视觉证据保留、实体身份一致性和状态演化追踪之间的矛盾。

Result: 在三个基准上的实验表明 MHG-Mem 有效，相较最强基线性能提升 8.20% 至 9.98%。

Insight: 创新点在于以超图统一组织实体身份、状态演化和视觉证据来源，并通过类型化超边与智能体搜索支持跨会话状态推理和按需回溯源观测；可借鉴之处是“保留来源的结构化记忆 + 自适应检索”范式。

Abstract: Long-term memory for multimodal agents remains challenging, as it must preserve fine-grained visual evidence while tracking entity evolution across sessions. Existing abstraction-based memories compress images into text, often discarding visually discriminative details, whereas pixel-preserving memories retain raw observations but struggle to reliably associate diverse observations with persistent entity identities and track their evolving states. To address these limitations, we propose MHG-Mem, a multimodal hypergraph memory that jointly organizes entity identity, state evolution, and provenance to visual evidence. Specifically, MHG-Mem represents distinct entity mentions and their observed facts as individual nodes, grouping identity-consistent mentions and facts through occurrence hyperedges. It then synthesizes aspect-specific evolution states from these grouped observations and links them to supporting entities, facts, and source evidence through evolutionary and evidential hyperedges. To navigate this structured memory, we introduce hypergraph-based agentic search, which adaptively selects an entry layer, traverses typed hyperedges across memory levels, and inspects source observations when fine-grained verification is required. These designs support both detailed visual verification and reasoning over cross-session state changes. Extensive experiments across three benchmarks demonstrate the effectiveness of MHG-Mem, improving performance by 8.20-9.98% over the strongest baseline.

[32] What the Author Wrote, What the Agent Reads: Silent Cue Loss in Enterprise Document Ingestion cs.CL | cs.AI | cs.IRPDF

Yikai Zhao, Saurabh Pandey, Pradeep Kumar Misra

TL;DR: 论文研究企业文档在进入智能体前的解析/摄取环节如何静默丢失作者通过格式（如高亮、颜色）编码的语义线索，导致智能体在合规与政策问答中自信地给出错误答案。作者用76份真实政策文档构建任务，发现忽略格式的摄取会使微F1从0.690降至0.150（docling）和0.075（纯文本），并在多个模型与PDF上复现。损失主要来自高亮和颜色两类线索，且主流转换器普遍不保留颜色、现有解析基准也不评测这一点。

Details

Motivation: 企业智能体需要先由摄取层把Word/PDF转成token再执行任务，但现有转换器往往只保留文本而丢失作者通过排版格式表达的强调与义务标记。论文要解决这种“静默线索丢失”使智能体不是明显不完整、而是自信出错的问题。

Result: 在76份真实政策文档任务上，格式盲摄取使微F1从0.690跌至0.150（docling）和0.075（纯文本），并在Claude Opus/Sonnet/Haiku五款模型及GPT-5.6上复现；docling对作者未标记要求的假阳性率是本文方法的四倍。消融显示损失主要集中于高亮（+0.302）和颜色（+0.262），而粗体、斜体、等宽、删除线影响接近零；PDF中同样存在该不对称性，且32.8%的6,460份文档带有刻意颜色或高亮、10.3%落在义务性内容上。

Insight: 创新点在于把企业文档智能体的可靠性问题定位到摄取层而非推理层，提出可确定性评测的“作者强调要求恢复”任务，并系统定位高亮/颜色是主流转换器不保留且解析基准不评测的关键线索。客观看，其可借鉴之处是用OOXML属性做无LLM评判的标签、跨模型和PDF复现、通过渲染图像独立验证标签（kappa=0.72），并证明显式提示线索还能在保留之外带来+0.085提升，使修复廉价且可度量。

Abstract: Enterprise agents act on documents. Before an agent can follow a policy, answer a compliance question or file a ticket, some ingestion layer must turn a Word file or PDF into tokens - and that layer is where an agent deployed in the wild silently loses information. We show that one layer is lost by every mainstream converter - the meaning an author encodes in how text looks - and that the loss makes agents confidently wrong rather than visibly incomplete. We build a task from 76 real policy documents: recover the requirements the author singled out, scored against the file’s own OOXML properties by a deterministic matcher with no LLM judge. Formatting-blind ingestion collapses from micro F1 0.690 to 0.150 (docling) and 0.075 (plain text), a gap that replicates across six answering models from two vendors - five Claude models spanning the Opus, Sonnet and Haiku tiers, and OpenAI’s GPT-5.6. An agent that cannot see the cue guesses, and docling’s false-positive rate on requirements the author did not flag is four times ours. A span-level leave-one-out localises the loss to two cues - highlight (+0.302) and colour (+0.262), both Holm-adjusted p < 1e-4 - while bold, italic, monospace and strikethrough each sit within +/-0.003 of zero. The same asymmetry appears in PDF. Colour is also the cue no mainstream converter preserves and no parsing benchmark scores, and it is common: 32.8% of 6,460 documents carry a deliberate colour or highlight, 10.3% on an obligation. A pre-registered test shows that typing a cue for the agent adds +0.085 beyond merely preserving it (Holm-adjusted p = 0.049). We validate the labels against rendered page images through an independent modality (kappa = 0.72) and bound what wording alone can achieve (AUC 0.664). For agent builders the practical finding is that ingestion, not reasoning, is where this reliability is won or lost, and that the fix is cheap and measurable.

[33] Looped Uniform Diffusion Language Models cs.CL | cs.LGPDF

Saiteja Utpala, Chieh-Hsin Lai, Yuki Mitsufuji, Praneeth Vepakomma

TL;DR: 论文研究在均匀扩散语言模型（UDLMs）中引入选择性循环（在每个去噪步骤中递归应用少量早期Transformer层），并保持训练目标和采样器不变。实验表明，循环能显著提升UDLM的训练效率，并在匹配NFE预算下改善生成困惑度；此外，训练时采用更频繁采样较小循环数的分布还能减少推理循环次数而不损失下游准确率。

Details

Motivation: UDLM训练时需要从可能不可靠的上下文去噪，训练目标比掩码扩散语言模型（MDLMs）更困难且更耗FLOPs，但下游推理性能更强。论文希望在不大改训练目标和采样器的前提下，利用选择性循环提升UDLM的训练效率。

Result: 循环UDLM达到与非循环UDLM约0.5万亿token训练相当的验证NLL，在OpenWebText上训练FLOPs最多减少2.8倍，在FineWeb-Edu上最多减少3.3倍；匹配NFE预算时，生成困惑度最多降低9.8%，多数增益来自中等循环次数；采用偏向小循环数的训练分布可在下游任务上比均匀循环数采样少用12%循环且不降低准确率。

Insight: 论文宣称的创新点是将选择性循环从MDLM扩展到UDLM，并在保持训练目标和采样器不变的情况下系统验证其训练与推理效率收益。客观来看，其可借鉴之处在于：循环机制可作为提升扩散语言模型训练效率的通用手段，且循环数分布的设计能进一步优化推理效率。

Abstract: Uniform diffusion language models (UDLMs) generate text by iteratively denoising sequences of ordinary vocabulary tokens, any of which may be corrupted and subsequently revised. During training, UDLMs must denoise from a context that may itself be unreliable, whereas masked diffusion language models (MDLMs) always condition on a clean context. This challenging training objective makes UDLMs more FLOP-intensive to train than MDLMs, but it yields stronger downstream reasoning performance. Recently, selective looping, which recursively applies a few early Transformer layers within each denoising step, has been shown to improve training efficiency in MDLMs under an iso-FLOP budget. We investigate looping in UDLMs while keeping the training objective and sampler unchanged. We find that looping significantly improves training efficiency: looped UDLMs reach the validation NLL of a non-looped UDLM trained on about 0.5 trillion tokens with up to 2.8x fewer training FLOPs on OpenWebText and 3.3x fewer on FineWeb-Edu. We also find that, at a matched NFE budget, looped UDLMs reach up to 9.8% lower generation perplexity than the non-looped UDLM, with most gains achieved at moderate loop counts. Finally, we study different loop-count distributions and find that sampling smaller loop counts more frequently during training improves inference efficiency: on downstream tasks, this strategy uses 12% fewer loops than uniform loop-count sampling without degrading accuracy.

[34] Environment Duels: A Microcosm of LLM Co-Evolution cs.CL | cs.AIPDF

Shibo Jin, Zhiqiu Xu, Yiling Zou, Shangyu Gong, Mayur Naik

TL;DR: 论文提出 Environment Duels，一个动态的智能体评测基准，在交互式环境中同时评估前沿大模型的两项能力：编写环境（作为出题者，每个环境附带一条特权提示，目标是让有能力的解题者在没有提示时失败、有提示时通过）和解决环境（作为解题者）。作者在九个前沿 LLM 上发现，解题与出题是两种不同的能力；模型之间的两两对局构成了模型相互影响的缩影，可用于研究模型的共同演化。

Details

Motivation: 现实中的大模型会从先行者和竞争者的产物中学习，它们既是内容的创造者也是问题的解决者，但现有评测大多只考察单一能力，缺乏在受控条件下同时衡量“出题”与“解题”并观察模型间相互影响的动态基准。

Result: 在九个前沿 LLM 上，最强的解题者并非最强的出题者，而较小的模型反而能从他人的对局中学习，写出最难的环境和最有效的提示；对局结果具有传递性，可以得到稳定的排名，不同作者独立设计出的环境趋于一致；在 Qwen3-27B 的强化学习实验（原文写作 Qwen3.8-27B）中，学习者在某作者环境上的提示增益，能预测用该环境训练后模型在留出推理与智能体基准上的提升幅度，且自己与他人的环境皆然。

Insight: 创新点在于把“环境生成加特权提示”设计成可对局的评测形式，将出题与解题纳入同一交互框架，并用对局的传递性、作者间的设计趋同以及提示增益的可迁移性，刻画模型之间相互教学、挑战与学习的关系，为研究 LLM 共同演化提供了可控的实验窗口。

Abstract: Through a tangle of real-world interactions, from synthetic training pipelines to data shared on the open web, frontier LLMs absorb what their predecessors and competitors create. How well does each model play its role in this ecosystem, as a creator and as a solver? We introduce Environment Duels, a dynamic agentic benchmark that evaluates both capabilities on interactive environments. Each model writes environments, each with a privileged hint, aiming for tasks that capable solvers fail without the hint and pass with it. Across nine frontier LLMs, solving and authoring prove to be distinct skills: the strongest solvers are not the strongest authors, while smaller models learn more from how others played and write the hardest environments and the most useful hints. Beyond the scores, the pairwise interactions form a microcosm of how models affect one another. Head-to-head outcomes are transitive and so yield a stable ranking; independent authors converge on the same designs; and in a reinforcement-learning study with Qwen3.8-27B, the learner’s hint lift on an author’s environments tracks how much training on those environments improves it on held-out reasoning and agentic benchmarks, whether the environments are its own or another model’s. Each of these patterns has a counterpart in how models shape one another outside the duel. By staging these interactions under controlled conditions, Environment Duels offers a window onto how LLMs teach, challenge and learn from one another. Code is available at https://github.com/LLM-Duels/EnvDuels.

[35] LongZero: Enhancing Long-Context Reasoning in LLMs from Zero Data cs.CL | cs.LGPDF

Zhennan Wan, Jianfei Chen

TL;DR: 论文提出LongZero，一种无需任何外部数据的强化学习方法，用于提升大语言模型的长上下文推理能力。它先让模型自生成紧凑结构化的多跳问答及支持证据，再将证据扩展为长多文档上下文，并结合自生成多轮检索任务，用GRPO训练模型。

Details

Motivation: 长上下文推理的RL训练依赖长、多样、有挑战且可验证的数据，但人工标注昂贵、自然数据稀缺，现有合成方法又依赖外部语料或更强教师模型，且已有零数据RL方法主要面向短上下文任务。

Result: 论文在两个LLM和五个长上下文推理基准上评估LongZero，并与基于外部语料合成数据的RL方法比较；实验表明LongZero可泛化到真实任务，且优于这些基线。

Insight: 创新点在于将零数据RL扩展到长上下文推理：先自生成结构化问答与证据，再单独扩展为长多文档上下文，并融合自写多跳推理与多轮检索任务进行GRPO训练，避免依赖外部数据和强教师模型。

Abstract: Reinforcement learning (RL) is an effective way to strengthen the long-context reasoning of large language models (LLMs), but it needs training data that are long, diverse, challenging, and verifiable. Human annotation of such data is expensive, naturally occurring data are too scarce to sustain RL, and existing synthesis methods depend on external corpora or stronger teacher models, which become harder to obtain as the trained models get stronger. Recently, RL from zero data, in which a model learns from tasks it proposes itself, has achieved excellent performance, but existing methods target short-context tasks. Therefore, we propose $\textbf{LongZero}$, a method that lets an LLM improve its own long-context reasoning without any external data. Instead of writing questions and answers over existing documents, LongZero first has the model synthesize the question and answer with its supporting evidence in a compact structured form, and only then separately expands the evidence into a long multi-document context. LongZero combines these self-written multi-hop reasoning tasks with self-written multi-round retrieval tasks and trains the model with GRPO. We evaluate LongZero on two LLMs across five long-context reasoning benchmarks and compare it with RL on data synthesized from external corpora. Experimental results show that LongZero can generalize to real tasks, and outperforms both baselines, suggesting a viable and promising path to improving long-context reasoning of LLMs.

cs.CV [Back]

[36] Complexity-Aware Surgical Video Reasoning Segmentation with Digital Twin and Context Memory cs.CVPDF

Yiqing Shen, Mathias Unberath

TL;DR: 论文提出 SurgReasonSeg，一种面向手术视频的复杂度感知推理分割方法：基于手术基础模型构建数字孪生表示，用一个 LLM 的两个 LoRA 适配器分别处理不同复杂度查询，并通过强化学习训练轻量路由模块在精度与延迟间权衡；同时结合自适应滑动窗口和手术上下文记忆检索。论文还构建了 EndoVis-ReasonSeg 基准，并在多个手术推理/指代分割基准上取得领先结果，平均运行速度达 18.8 FPS。

Details

Motivation: 传统手术分割依赖闭集类别或显式文本查询，无法处理需要理解手术流程、器械功能和时序关系的隐式查询，因此需要面向手术视频的推理分割方法。

Result: 在 EndoVis-ReasonSeg 上，相比最强基线 ReSurgSAM2，语义、空间、时序查询的 J&F 分别提升 0.178、0.237、0.297；在 Ref-EndoVis17 和 Ref-EndoVis18 指代分割基准上也取得最佳结果，平均 18.8 FPS，可支持实时推理分割。

Insight: 创新点在于将数字孪生表示、复杂度感知的双 LoRA 适配器与强化学习路由结合，并引入由查询复杂度和手术流程阶段驱动的自适应时序上下文及基于流程相似性的历史病例记忆检索；同时提出新的手术推理分割基准和骨干匹配归因协议，便于区分分割骨干与推理模块贡献。

Abstract: Conventional surgical segmentation relies on closed-set categories or explicit text queries, which cannot resolve implicit queries that require reasoning about workflows, instrument functions, and temporal relationships. To address these limitations, we propose SurgReasonSeg, a complexity-aware method that brings reasoning segmentation over digital twin (DT) representations to surgical video. Specifically, SurgReasonSeg constructs DT representations from surgical foundation models and trains two low-rank adapters (LoRA) of one large language model with reinforcement learning (RL), each specialized for a different query complexity. A lightweight router, also trained with RL, selects between the two adapters to balance latency against accuracy. SurgReasonSeg also sets the temporal context with an adaptive sliding window conditioned on query complexity and workflow phase, and retrieves relevant historical cases from a surgical context memory by workflow similarity and procedural relevance. We further introduce EndoVis-ReasonSeg, a surgical reasoning segmentation benchmark with 1,460 training and 961 test samples across semantic, spatial, and temporal categories. On EndoVis-ReasonSeg, SurgReasonSeg improves J&F over the strongest baseline, ReSurgSAM2, by 0.178, 0.237, and 0.297 on semantic, spatial, and temporal queries, and it also achieves the best results on the Ref-EndoVis17 and Ref-EndoVis18 referring segmentation benchmarks. A backbone-matched attribution protocol separates the contribution of the upstream segmenter from that of the reasoning modules. Our method runs at 18.8 FPS averaged over the query distribution, enabling real-time reasoning segmentation in surgical videos.

[37] JUICE: Unified Multimodal Modeling in Pretrained Representation Spaces cs.CVPDF

Jiahan Li, Huangyuan Su, Shengbang Tong, Shusheng Yang, Ruihan Guo

TL;DR: 论文提出JUICE，在冻结的图像和文本预训练编码器表示空间中，用单个flow matching Transformer联合建模两种模态，并为每个模态设置独立噪声水平；通过指定哪些模态干净、哪些待生成，统一支持图像生成、图像描述、视觉问答和联合生成。模型无需语言模型初始化，在2.1B参数和224x224输入下取得GenEval 0.88、VQAv2 73.1，并能通过联合采样同时生成地图路线和对应动作序列。

Details

Motivation: 现有统一多模态模型通常将文本生成交给语言模型、图像生成交给扩散头，二者不对称且联合输出顺序固定，导致先生成的模态无法根据后生成模态自适应。论文希望用一个从零联合训练的单一生成过程直接建模图文联合分布。

Result: 在同等引导下，将文本作为连续状态生成基本不损害图像生成（GenEval 48.41，对比纯文生图模型47.86），但图像描述会损失部分精确措辞；联合训练与条件去噪结合优于二者单独训练，语义编码器空间优于VAE潜变量和原始像素。2.1B参数、无语言模型初始化、224x224下达到GenEval最高0.88和VQAv2 73.1；在迷宫任务中联合采样成功率从独立采样的54.6%提升到93.7%，视觉空间规划在训练规模地图上解决99.8%。

Insight: 创新点在于直接在冻结的预训练图像和文本连续表示空间中用单个flow matching Transformer联合建模，并通过各模态不同噪声水平及干净/生成模态组合统一多种任务；联合采样使模型无需固定模态生成顺序，联合与条件去噪联合训练也提供了可借鉴的多模态生成训练策略。

Abstract: Most unified multimodal models treat images and text asymmetrically: a pretrained language model generates text, an attached diffusion head generates images, and outputs that contain both are produced in a fixed order, so the first output cannot adapt to the second. We ask whether both modalities can instead be generated by one process that is trained on them jointly from scratch. Pretrained encoder spaces hold both modalities in one continuous form and carry knowledge learned at scale, and recent work finds them effective as generative targets for vision and language. We therefore model the joint distribution directly in these spaces. We introduce JUICE (Joint Unified multimodal modeling In Continuous Encoder spaces), a single flow matching Transformer over frozen image and text representations, in which each modality has its own noise level. Which modalities are clean and which are generated then decides the task: image generation, captioning, visual question answering, or joint generation. Controlled pretraining runs test each design choice. At equal guidance, generating text as continuous states leaves image generation intact (48.41 GenEval, against 47.86 for a model trained only for text-to-image generation), while captions lose some exact wording. Training on joint and conditional denoising together beats either alone, and semantic encoder spaces beat VAE latents and raw pixels as targets. At 2.1B parameters, with no language model initialization and at 224x224, JUICE reaches up to 0.88 on GenEval and 73.1 on VQAv2. By sampling both modalities from the joint distribution (coupled sampling), it can also draw a route on a map while writing out the matching moves. On mazes with several valid solutions, joint success rises from 54.6% with independent sampling to 93.7%, and no modality has to go first. On visual spatial planning, JUICE solves 99.8% of maps at its training sizes.

[38] SAIL: A Learnable Generalisation of Spatial Autocorrelation over Graph-Structured Data cs.CVPDF

Piotr Keller, Fayyaz Minhas

TL;DR: 论文提出SAIL，一个将经典空间自相关统计（如Moran’s I和Geary’s C）推广为可学习可微框架的方法。它联合学习特征投影（定义候选标量场）和节点间邻域亲和力，生成可解释的局部和全局空间一致性描述符，支持监督预测和无监督发现空间自相关模式与热点。SAIL在适当参数下可精确恢复经典统计量，并在合成数据和大型生物医学图上验证了统计正确性与空间一致性发现能力。

Details

Motivation: 经典空间自相关统计只能处理单一标量变量和固定邻域权重，无法直接用于高维图结构数据，因此需要一种可学习的泛化方法来建模此类数据中的空间自相关结构。

Result: 在合成数据和大型生物医学图上实验，展示了统计正确性并能发现空间一致性；对于由潜在空间自相关驱动的任务，SAIL能以匹配的归纳偏置近似贝叶斯最优预测器。

Insight: 摘要宣称的创新点是将空间自相关统计推广为可学习算子，联合学习特征投影和邻域亲和力，同时保持可解释性并能精确恢复经典统计。客观分析，其可学习邻域亲和力与特征投影相结合的设计，为图结构数据中的空间结构建模提供了可借鉴的归纳偏置。

Abstract: In vision and scientific domains involving spatial data, a key question is whether node feature values vary independently across space or form coherent spatial structures. Classical spatial autocorrelation statistics such as Moran’s $I$ and Geary’s $C$ quantify this phenomenon but operate on a single scalar variable with fixed neighbourhood weights, limiting their use for high-dimensional graph-like data. We introduce SAIL (Spatial Autocorrelation Index Learner), a differentiable framework that generalises spatial autocorrelation into a learnable operator by jointly learning feature projections defining candidate scalar fields and inter-node neighbourhood affinities. SAIL produces interpretable local and global spatial coherence descriptors, enabling supervised prediction and unsupervised discovery of spatial autocorrelation modes and hotspots. We show that SAIL exactly recovers classical spatial statistics under appropriate parametrisations and, for tasks driven by latent spatial autocorrelation, can approximate the Bayes-optimal predictor with an inductive bias matched to such tasks. Experiments on synthetic data and large-scale biomedical graphs demonstrate statistical correctness and discovery of spatial coherence with applications across vision, geospatial analysis, and biomedicine. The code repository is available at: https://github.com/pkeller00/SAIL.

[39] HorizonComposer: Spatiotemporally Consistent Driving Video Editing with Enriched Traffic Semantics cs.CVPDF

Mauricio Soroco, Yuqiu Liu, Zaid Tasneem, Francesco Pittaluga, Abhishek Aich

TL;DR: 论文提出HorizonComposer，一种面向自动驾驶场景的视频编辑方法。通过构建混合数据集（真实视频+合成编辑）、采用离线强化学习中的密集像素级奖励进行优势条件化训练，以及引入thinking frames过渡机制，有效保护交通语义，实现时空一致的驾驶视频编辑，避免结构幻觉和时间冲击。

Details

Motivation: 通用指令引导的视频编辑器在全局编辑驾驶场景时，会破坏严格的交通语义，导致结构幻觉和时间冲击，因此需要专门方法在编辑中保护和蒸馏交通语义。

Result: 方法达到SOTA时空一致性，用户偏好中内容保持相对提升39%、时间一致性相对提升12%；下游道路线分割提升5%，多目标检测提升7%。

Insight: 创新点包括：混合数据集平衡场景结构与外观信号；借鉴离线强化学习的密集像素级奖励优势条件化训练，引导生成结构可靠像素；thinking frames过渡机制无缝协调静态图像条件与动态视频上下文。

Abstract: Instruction-guided video generative models offer a scalable solution for stress-testing autonomous driving systems by simulating diverse environmental conditions. However, driving scenes operate within a dense Operational Design Domain (ODD) governed by strict traffic semantics. When applied to global driving-condition edits, general-purpose instruction-guided video editors frequently corrupt these semantics, leading to structural hallucinations and ``temporal shock.’’ This work distills and protects traffic semantics during video editing at three levels. First, we construct a hybrid dataset by combining temporally aligned real-world videos with synthetic edits, balancing foundational scene structure with unambiguous appearance signals. Second, during training, we adapt advantage conditioning from offline reinforcement learning to align the model using dense rewards. We extract dense, pixel-level rewards from the hybrid data to serve as a localized signal and actively steer generation toward structurally reliable pixels. Finally, our model mitigates temporal shock by introducing thinking frames, a transition mechanism that seamlessly reconciles static image conditioning with dynamic video context. Extensive experiments demonstrate that our method prevents the corruption of multi-actor dynamics, establishing state-of-the-art spatiotemporal consistency, yielding consistent user preference (39% relative improvement in content preservation and 12% in temporal consistency), and achieving a significant 5% gain in downstream road line segmentation and 7% gain in downstream multi-object detection.

[40] DualHeads: Reward-tilted few-step diffusion made simple cs.CV | cs.AI | cs.LGPDF

Yuanzhi Zhu, Eleftherios Tsonis, Simo Alami Chehboune, Xi Wang, Stéphane Lathuilière

TL;DR: 论文提出 DualHeads，一种面向少步蒸馏扩散模型的简单奖励对齐方法。它在冻结的扩散骨干上附加两个轻量二分类头：奖励头判断同一提示下哪些样本得分更高，参考头区分参考图像与生成图像，生成器通过最大化两个头 logits 的加权和来优化。理论表明，当头最优时，该目标等价于对学习到的奖励进行 KL 正则化最大化，头权重比控制正则强度。实验显示，少步生成器在规则和模型奖励下可接近 40 步强化学习方法的任务奖励，同时保持图像质量且不增加推理成本。

Details

Motivation: 少步蒸馏扩散模型虽能快速生成高质量图像，但难以对齐任务要求和人类偏好；现有对齐方法多依赖可微奖励、score field 或随机轨迹，而少步蒸馏生成器可能不具备这些条件。论文旨在仅利用外部奖励评估和参考样本，将少步生成器适配到奖励倾斜生成。

Result: 在规则奖励和模型奖励下，少步 DualHeads 生成器达到的任务奖励接近 40 步强化学习对齐方法，同时保持图像质量，并且不增加额外推理成本。

Insight: 创新点在于用两个二分类头实现少步扩散模型的奖励对齐：奖励头学习偏好排序，参考头提供参考分布约束，并通过 logits 加权和实现 KL 正则化奖励最大化；方法简单、无需可微奖励或随机轨迹，且不改变推理成本。

Abstract: Distilled diffusion models generate high-quality images in a few sampling steps, but aligning them with task requirements and human preferences is difficult. Methods developed for multi-step diffusion models rely on differentiable rewards, or on score fields and stochastic trajectories that distilled generators may lack. We ask whether a few-step generator can be adapted using only external reward evaluations and reference samples, and show that two binary classifiers suffice. Our method, DualHeads, implements them as lightweight heads on a frozen diffusion backbone. The reward head learns which samples of the same prompt score higher, the reference head separates reference images from generated ones, and the generator maximizes a weighted sum of their logits. When the heads are optimal, this objective is KL-regularized maximization of the learned reward, and the ratio of head weights sets the regularization strength. Empirically, with both rule-based and model-based rewards, a few-step DualHeads generator reaches task rewards close to those of 40-step reinforcement learning methods while preserving image quality, at no extra inference cost. We hope that this simple and elegant dual-head formulation will help advance few-step reward-tilted generation.

[41] Pair-Free Coarse-to-Real Video Generation via Real-to-Coarse Image Supervision cs.CVPDF

Gonzalo Gómez-Nogales, Daohan Lu, Peiye Zhuang, Tinghui Zhou, Yi Zhou

TL;DR: 论文提出一种无需成对粗-真实视频的粗到真实（C2R）视频生成训练框架。它先用易构建的真实图-粗图图像对训练DINO-R2C，在DINO特征空间逐帧把真实视频转成伪粗视频，再用这些伪配对训练C2R#模型；推理时丢弃DINO-R2C，直接由真实粗视频驱动C2R#生成逼真视频。

Details

Motivation: 现有C2R视频生成依赖昂贵且难以获取的对齐（粗视频，真实视频）配对，导致训练受限于狭窄的合成域。论文希望利用真实图到粗图图像对易构造的不对称性，摆脱对成对视频数据的依赖。

Result: 摘要未提供具体定量结果、benchmark或与SOTA模型的比较，仅说明将发布模型和项目网页。

Insight: 核心创新是用真实图-粗图图像对构造伪粗视频，从而以图像级监督替代视频级配对，并采用两阶段训练使推理阶段无需图像转换模型。客观来看，该思路将配对数据需求从视频域转移到更易获得的图像域，有望提升C2R训练对真实视频语料的可扩展性。

Abstract: Coarse-to-real (C2R) video generation converts coarse animation, simulation, or game-rendered videos into photorealistic videos while preserving scene layout, camera motion, and dynamics. Training such models requires aligned (coarse, realistic) video pairs, which are expensive to create and confine prior systems to narrow synthetic domains. We exploit a simple asymmetry: paired videos are hard to obtain, but paired real-to-coarse images are easy to construct with existing editing and 3D reconstruction tools. Using these image pairs, we train the DINO-R2C model to transfer realistic images to coarse images in DINO feature space, and apply it frame-wise to ordinary real videos to synthesize pseudo-coarse videos. The C2R# model then learns to generate each real video from its pseudo-coarse counterpart, free from needing (coarse, realistic) video pairs. At inference, the DINO-R2C model is discarded and actual coarse videos drive the C2R# model directly. Thus, our pair-free training scales with any real-video corpora rather than with paired synthetic videos. We will release the model and project webpage at https://gonzalognogales.github.io/coarse2real-sharp/.

[42] RoPT: Robust 6DoF Object Pose Tracking in Video cs.CV | cs.ROPDF

Yongqing Liang, Yuzhong Chen, Sihang Li, Junsuk Kim, Chase Kidder

TL;DR: 论文提出RoPT，一种面向视频中六自由度物体姿态跟踪的鲁棒框架，通过将视觉感知、时序运动学和全局重定位统一为紧密耦合的闭环系统，解决传统单目跟踪器在快速运动下易发散的问题。方法利用物体掩码跟踪应对遮挡，结合运动学因子图进行帧间正则、前向姿态预测和多假设验证，并引入主动重定位实现故障感知与自恢复。

Details

Motivation: 传统单目6DoF物体姿态跟踪管线在快速物体运动时容易发生灾难性发散，难以满足机器人交互等场景对稳定跟踪的需求。

Result: 在多个挑战性基准和真实机器人演示中验证了有效性；摘要称在快速运动下显著提升跟踪存活能力，但未给出具体数值或明确SOTA结论。

Insight: 创新点在于将视觉感知、时序运动学约束和全局重定位集成为闭环容错跟踪系统；利用掩码跟踪抗遮挡、因子图预测与多假设验证避免局部极小，并通过主动重定位实现失败检测和确定性自恢复。

Abstract: Accurate 6-DoF object pose tracking is essential for robotic interaction, yet conventional monocular pipelines frequently suffer from catastrophic divergence under rapid object motion. We present RoPT, a robust tracking framework engineered to overcome this limitation by unifying visual perception, temporal kinematics, and global re-localization into a tightly coupled closed loop. Unlike existing open-loop trackers, RoPT delivers three critical capabilities that fundamentally alter system resilience. First, the visual perception component tracks object masks to robustly handle partial and severe occlusions. Second, a kinematic factor graph regularizes frame-to-frame updates, forecasts forward pose predictions, and drives a parallel multi-hypothesis verification scheme to bypass local minima. Finally, an active re-localization module transforms the pipeline into a closed-loop system, making it failure-aware and capable of deterministic self-recovery. We demonstrate the effectiveness of our framework across challenging benchmarks alongside real-world robotic demonstrations, achieving significantly higher tracking survivability under rapid motion.

[43] Single-Image Food Mass Estimators Do Not Transfer Their Gram Scale Across Datasets cs.CVPDF

Yongsheng Lian

TL;DR: 论文系统评估了单张RGB照片估计食物质量的跨数据集迁移能力：在Nutrition5k上微调DINOv2-base，并与零样本Qwen3-VL-32B在多个外部数据集上比较。结果显示模型域内MAE为24.8 g，明显优于Qwen，但跨数据集时不再占优，且克尺度会随拍摄范围和食物种类发生系统性偏差。论文进一步指出Nutrition5k随机划分存在近重复照片导致精度高估，并提出目标场景标注照片对实际部署是必要的。

Details

Motivation: 图像膳食评估需要从照片估计食物质量，但已有模型能否迁移到新相机、新食物和新场景并不清楚，论文旨在检验单图像食物质量估计器的跨数据集泛化性与克尺度可靠性。

Result: 在Nutrition5k域内，微调DINOv2-base的MAE为24.8 g，不到Qwen3-VL-32B的一半；但在三个外部数据集上未超过Qwen。使用测试数据集中的其他食物或托盘训练可将外部误差降低38%至65%，而更大骨干DINOv2-large未降低外部误差；同食物一张标注照片可将ASA24误差降至28.3 g。

Insight: 论文的创新与可借鉴之处在于揭示了单图像食物质量估计的克尺度不可跨数据集迁移，并诊断出拍摄范围变化和食物类别偏差因子不同两类问题；同时指出随机划分中近重复照片会夸大精度，强调目标场景标注数据和可靠预测区间的重要性。

Abstract: Estimating food mass from a photograph is central to image-based dietary assessment, but whether trained models transfer to new cameras and foods is unclear. We fine-tuned DINOv2-base on 2,253 weighed Nutrition5k dishes to predict mass from one RGB photograph and compared it with a zero-shot vision-language model, Qwen3-VL-32B, on Nutrition5k and three external datasets: ASA24 studio portions, ECUSTFD whole foods and Food Portion Benchmark meal trays. In domain, the fine-tuned model’s mean absolute error was 24.8 g, less than half of Qwen’s. On external datasets it did not beat Qwen. For a food it had never seen, it ranked portions by size, but no better than a count of food pixels. Its gram scale was wrong in two ways: it changed with how much of the scene the photograph showed, and it was off by a different factor for each food. Training on other foods or trays from the tested dataset cut the external error by 38 to 65%. DINOv2-large, a larger backbone trained on the same dishes, did not reduce the external error. One labeled photograph of the same food cut the model’s ASA24 error to 28.3 g, better than that photograph achieved alone. Prediction intervals that held the true mass for 90% of Nutrition5k dishes held it for at most half of external photographs. Nutrition5k often photographed the same plate several times, so a random test split shared near-identical photographs with training and overstated accuracy. Single-image mass estimators therefore need labeled photographs from the setting where they are used.

[44] HumanCentric-Scenes-AI: A Human-Curated Multimodal Dataset of AI-Generated Human-Centric Scenes Across Surveillance, Occupational Portraits, AI-Edited Images, and Short-Form Video cs.CV | cs.AI | cs.HCPDF

Nima Kamali, Sana Nasiri, Negar Kamali

TL;DR: 论文发布 HumanCentric-Scenes-AI，一个由人工多轮策划、包含296个AI生成刺激的多模态数据集，覆盖监控、职业肖像、AI编辑图像和短视频四类以人为中心的场景。数据集包含85张CCTV风格图像、128张平衡配对的职业肖像（64种职业，男女各一）、27张区域被生成模型修改的真实照片和56个AI生成短视频，并提供文件、生成提示和逐刺激元数据。论文还报告了未指定性别生成中的性别分布与职业性别比例比较、AI编辑区域大小与零样本检测难度的关联、监控图像质量分布和视频主题类别覆盖。

Details

Motivation: 论文旨在填补以人为中心、跨真实遭遇场景、经过人工策划的AI生成媒体数据资源不足的问题，为人类感知研究、AI检测分类器校准以及生成模型偏差分析提供标准化刺激与元数据。

Result: 论文构建了296个刺激的数据集，包含四个子集并给出多项描述性分析：未指定性别生成的描绘性别分布与美国劳工统计局职业性别份额比较、AI编辑子集中编辑区域大小与零样本检测难度的关联、监控子集图像质量分布以及视频子集主题类别覆盖；摘要未提及在某个benchmark上达到SOTA。

Insight: 创新点在于覆盖监控、职业肖像、AI编辑照片和短视频四类以人为中心且人们会实际接触AI生成媒体的场景，采用人工多轮策展提升照片级真实感，并设计64种职业的男女平衡配对；同时公开生成提示和逐刺激元数据，支持感知研究、检测器校准和生成模型偏差分析。客观看，其价值主要在于跨场景多模态数据资源构建与配套偏差/难度分析，而非提出新的生成或检测算法。

Abstract: We introduce HumanCentric-Scenes-AI, a human-curated dataset of 296 AI-generated stimuli that depict people across four contexts in which people encounter AI-generated media. The Surveillance subset contains 85 AI-generated CCTV-style images. The Occupations subset contains 128 AI-generated occupational portraits in a balanced paired design covering 64 occupations with one female and one male version of each. The AI-edited subset contains 27 real photographs in which a region was altered or replaced by a generative model. The Video subset contains 56 AI-generated short videos. We curated the images and videos manually over multiple rounds to make them as photorealistic as possible. For each stimulus we release the file, the generation prompt where recorded, and per-stimulus metadata. We report depicted-gender distributions from an initial set of 134 generations without a specified gender and compare them with U.S. Bureau of Labor Statistics occupational gender shares. We report edit-region size and its association with zero-shot detection difficulty for the AI-edited subset. We report image-quality distributions for the Surveillance subset and subject-category coverage for the Video subset. HumanCentric-Scenes-AI is intended first for human-perception research, and also for calibration of AI-detection classifiers and bias analysis of generative models.

[45] OccJEPA: Occlusion as Masking for Amodal Video Segmentation cs.CV | cs.LGPDF

Genpei Zhang

TL;DR: 论文提出 OccJEPA，将遮挡视为掩码，用预训练的 V-JEPA 2.1 视频世界模型预测被遮挡区域的表示，再通过轻量解码头生成 amodal 视频分割掩码。该方法无需视频扩散模型的多步去噪和额外深度网络，结果确定且比扩散式 SOTA 快 33.5 倍，并在多个 benchmark 上更准确。

Details

Motivation: 现有 amodal 视频分割最佳方法依赖微调视频扩散模型，需要大量去噪步骤和额外深度网络，且每次采样结果不同。论文希望利用 JEPA 类自监督视频世界模型对掩码时空区域进行表示预测的能力，以更高效、确定的方式补全被遮挡目标。

Result: OccJEPA 比基于扩散的 SOTA 快 33.5 倍，并在 SAIL-VOS、TAO-Amodal（含 held-out test split，提升 +4.00 AP50，且使用 SAM2 跟踪的可见掩码）和 Kubric 上更准确。

Insight: 核心创新是把物体遮挡类比为掩码建模：丢弃可见掩码未覆盖的 token，由预训练 V-JEPA 2.1 predictor 填补表示，再用轻量头解码 amodal 掩码。控制实验表明性能增益主要来自 predictor 的预训练权重而非模型容量，且适用于 87M 到 1.85B 参数的世界模型。

Abstract: Amodal video segmentation asks for the complete mask of an object in every frame, including the parts hidden behind other objects. The best current methods generate these masks with fine-tuned video diffusion models, which need many denoising steps and an extra depth network, and give a different answer for every sample. Self-supervised video world models of the JEPA family are trained on a closely related task: predicting the representation of masked space-time regions from the visible ones. We therefore treat occlusion as masking. The tokens that the object’s visible mask does not cover are dropped, the pre-trained V-JEPA 2.1 predictor fills them in, and a light head decodes the observed and predicted features into amodal masks. The resulting model, OccJEPA (Occlusion-as-masking JEPA), is deterministic and 33.5x faster than the diffusion-based state of the art, and more accurate on SAIL-VOS, on TAO-Amodal, including a held-out test split (+4.00 AP50) and visible masks tracked by SAM2, and on Kubric. Controlled experiments trace the gain to the predictor’s pre-trained weights rather than to its capacity, for world models from 87M to 1.85B parameters. Code is available at https://github.com/zgp110/occjepa.

[46] UNIVERS: Learning UNIversal Voxel-wise Embedding via Randomized Synthesis for generalizable one-shot landmark localization cs.CV | cs.AIPDF

Derong Yu, Qingsong Zhao, Yanbing Yue, Yikai Li, Zhi Zhang

TL;DR: 论文提出UNIVERS，一种端到端可训练框架，通过从器官形状模板随机合成伪数据，并结合跨域判别式体素级对比学习，学习通用体素嵌入，从而在多模态、任意视野的体积扫描中实现可泛化的一次性关键点定位；在三个肾上腺数据集上验证并优于SOTA方法。

Details

Motivation: 现有自监督一次关键点定位方法依赖同一图像增强视图的对比学习，跨模态泛化受限；利用配对多模态图像的方法又受训练数据有限和多阶段迭代训练困扰，因此需要更通用且端到端的学习框架。

Result: 在三个具有挑战性的肾上腺 gland 数据集上，UNIVERS 一致优于当前最优的一次性关键点定位方法，尤其在跨模态场景中优势明显；摘要未给出具体定量指标。

Insight: 创新点在于用器官形状模板随机合成伪数据，并对语义布局、视觉外观、视野和空间姿态进行域随机化，配合跨域判别式体素级对比学习，迫使网络忽略域特定偏差；客观看，其价值在于以合成随机化和端到端训练替代配对多模态数据与多阶段训练，提升跨模态和任意FOV定位泛化性。

Abstract: One-shot landmark localization in volume scans with varying fields of view (FOVs) is challenging. While existing self-supervised learning methods have shown promising performance, they rely on contrastive learning built upon augmented views of the same image, which inherently limits their generalization in cross-modality scenarios. This motivates research on cross-modality embedding learning using paired multi-modality images. However, existing methods suffer from not only limited training data, but also the requirement of multi-stage iterative training. To address these limitations, we propose UNIVERS, an end-to-end trainable framework that learns UNIversal Voxel-wise Embedding via Randomized Synthesis. Our framework leverages diverse pseudo-data synthesized from organ shape templates, with randomization applied to semantic layout, visual appearance, FOV, and spatial pose. Coupled with cross-domain discriminative voxel-wise contrastive learning, this domain randomization strategy forces the network to disregard domain-specific biases and learn features that are robust against variations in imaging contrast, resolution, and spatial context. Consequently, the trained model enables one-shot landmark localization in multi-modality volume scans with arbitrary FOVs. In this study, the efficacy of the proposed method is demonstrated by applying it to the task of localizing the adrenal glands, which are paired, small, and irregular endocrine organs. In particular, comprehensive experiments are conducted on three challenging adrenal gland datasets. The experimental results demonstrate that UNIVERS consistently outperforms the state-of-the-art (SOTA) one-shot landmark localization approaches, with particularly prominent advantages in cross-modality scenarios. The source code is publicly available at https://github.com/Kendra9999/UNIVERS.

[47] PhyMo: Learning Physical Dynamics with Accurate and Continuous Motion from Multi-View Videos cs.CVPDF

Shangjia Liu, Jinxi Li, Siyuan Zhou, Bo Yang, Bing Wang

TL;DR: 论文提出PhyMo框架，从多视角视频中恢复动态3D场景的几何、外观以及支配场景演化的物理运动状态。针对仅靠渲染监督时模型可能通过颜色或透明度变化补偿不准确形变的问题，PhyMo引入运动感知轨迹表示和带双向运动学连续性约束的连续运动演化模块，以学习准确且连续的运动。最终实现几何忠实的插值和连续的运动外推。

Details

Motivation: 从多视角视频中准确恢复动态3D几何与物理运动状态具有挑战性；仅用渲染监督容易让模型利用外观捷径解释观测运动，导致形变和轨迹不准确，尤其在快速或局部复杂运动场景中会造成不可靠重建和不连续的未来预测。

Result: 在四个动态场景基准上实验，论文宣称取得状态最优（SOTA）性能，并实现几何忠实的插值和连续的运动外推。

Insight: 创新点在于提出运动感知轨迹表示，引导模型从形变中恢复真实运动而非依赖外观捷径；同时设计连续运动演化模块，并用双向运动学连续性约束促进位移、速度和加速度在时间上的连续一致演化。

Abstract: Recovering dynamic 3D scene geometry, appearance, and the physical motion states that govern scene evolution from multi-view videos is important yet challenging. A key difficulty is that appearance can be deceptive. Under rendering supervision alone, a model may fit the observed images by changing color or opacity to compensate for inaccurate deformation, rather than recovering the true evolving geometry and motion. This issue is especially severe in scenes with fast or locally complex motion, where inaccurate trajectories can still produce plausible images over the observed frames, but lead to unreliable reconstruction and discontinuous future prediction. In this paper, we propose PhyMo, a framework for learning evolving dynamics from multi-view videos. We introduce a Motion-Aware Trajectory Representation that drives the model to recover accurate motion from deformation, rather than relying on appearance shortcuts to explain observed motion. On top of this, we propose a Continuous Motion Evolution module with bidirectional kinematic continuity constraint to promote temporally consistent evolution of displacement, velocity, and acceleration. Together, these designs yield geometrically faithful interpolation and continuous motion extrapolation. Experiments on four dynamic scene benchmarks demonstrate state-of-the-art performance.

[48] LocalAgent: Collaborative Agentic Verification for Fine-Grained Instance-Level Consistency cs.CVPDF

Jiachen Guo, Xinshan Zhu, Boyun Wang, Yanyan Liang, Jun Wan

TL;DR: 论文提出LocalAgent，一个协作式智能体验证框架，用于细粒度实例级一致性判断，即判断两张图像是否描绘同一物理对象。该方法先由通用MLLM进行全局推理并定位可疑局部区域，再调用轻量级Local Evidence Verifier工具包进行局部外观一致性LAC和局部几何对应LGC验证，最后整合结构化标量证据做决策。论文还引入辅助上限强化微调RFT策略，使模型学习何时需要调用验证器。

Details

Motivation: 现有MLLM在通用视觉理解上进展显著，但在视角、光照和背景变化下判断两张图像是否为同一实例的细粒度一致性验证仍较困难；语义相似对象可能仅在小局部区域存在差异，因此需要同时具备全局语义感知和精确局部证据检查能力。

Result: 实验显示LocalAgent相比最强CoT-RFT基线平均准确率提升4.9%，且仅在12.13%的样本上调用验证器，同时保持底层MLLM的通用多模态能力；摘要未明确具体benchmark名称。

Insight: 创新点在于将MLLM的全局推理与轻量专家验证器结合，按需定位并验证可疑局部区域，并通过LAC/LGC分别处理外观和几何变化；辅助上限RFT让答案正确性主导工具使用与格式信号，使模型学会何时调用验证工具。客观来看，该工作把细粒度实例验证拆解为全局筛选加局部证据核验，并以低调用率实现性能提升，具有较好的可借鉴性。

Abstract: Multimodal large language models (MLLMs) have made rapid progress in general visual understanding, yet they still struggle with fine-grained instance-level consistency verification, where the goal is to determine whether two images depict the exact same physical object under varying viewpoints, illumination, and backgrounds. This task requires both global semantic perception and precise examination of subtle local evidence, since semantically similar objects may differ only in small suspicious regions. To address this challenge, we propose LocalAgent, a collaborative agentic framework that integrates the complementary strengths of a generalist MLLM and lightweight specialist verifiers. The MLLM first reasons globally to identify suspicious local regions, which are then delegated to a Local Evidence Verifier (LEV) toolkit for fine-grained appearance and geometric matching. LEV comprises Local Appearance Consistency (LAC), which estimates appearance similarity under illumination and color changes, and Local Geometric Correspondence (LGC), which estimates spatial correspondence under viewpoint changes. LEV returns structured scalar evidence that the MLLM integrates for the final decision. We further introduce an auxiliary-capped reinforcement fine-tuning (RFT) strategy, in which answer correctness dominates auxiliary tool-use and format signals, enabling the policy to learn when verification is necessary. Experiments show that LocalAgent improves mean accuracy by 4.9% over the strongest CoT-RFT baseline while invoking a verifier on only 12.13% of samples and preserving the general multimodal capabilities of the underlying MLLM. The open-source repository is available at https://github.com/Firefly-zoo/LocalAgent-main.

[49] Multi-Scale Latent Temporal Compression Attention and Video-Text Cross Attention Fusion for Video Action Detection cs.CV | cs.MMPDF

Yicheng Qiu, Keiji Yanai

TL;DR: 论文面向视频动作检测中的连续视频流时序建模问题，提出多尺度潜在时序压缩注意力模块，将连续时序序列聚合为多尺度特征表示并降低帧信息冗余；同时提出视频-文本交叉注意力融合模块，在潜在特征和标准特征两个层面进行跨模态交互，实现视频与文本特征的细粒度融合。多个视频识别基准实验表明方法优于现有方法。

Details

Motivation: 现有深度神经网络方法难以建模动态时序模式，存在冗余帧信息、全局与局部特征难以平衡、视觉与文本特征融合不足等问题。

Result: 在多个视频识别基准上取得优于现有方法的性能，验证了方法的有效性和泛化能力；摘要未给出具体定量指标或明确SOTA水平。

Insight: 创新点在于多尺度潜在时序压缩注意力聚合连续时序并降冗余，以及潜在层和标准特征层双层级视频-文本交叉注意力融合以增强语义对齐和特征交互。客观来看，将多尺度时序压缩与跨模态细粒度融合结合，是提升视频动作检测时序建模和语义对齐能力的可借鉴思路。

Abstract: In video action detection, modeling the temporal dependencies in continuous video streams is crucial. Previous methods based on deep neural networks can perform frame-level video recognition tasks. However, these methods struggle to model dynamic temporal patterns, facing challenges such as redundant frame information, difficulty balancing global and local features, and insufficient fusion of visual and textual features. To address these issues, we propose a video temporal attention module based on the Multi-Scale Latent Temporal Compression Attention, which aggregates continuous temporal sequences into feature representations across diverse temporal scales while efficiently reducing redundancy. We also introduce a Video-Text Cross Attention Fusion module that leverages cross-attention at both latent and standard feature levels, achieving fine-grained fusion between video and textual features to enhance semantic alignment and feature interaction. Experimental results across multiple video recognition benchmarks demonstrate that our proposed methods achieve superior performance compared to existing methods, validating the effectiveness and generalization capability.

[50] A Unified Evaluation Framework of Physical Adversarial Robustness in the Edge VLM Era cs.CV | cs.CR | cs.LGPDF

Eric Liang, Alan Luo, Kaiwen Yuan

TL;DR: 论文提出一个统一的物理对抗鲁棒性评估框架，用于公平比较边缘视觉语言模型（VLM）与传统目标检测器。作者以行人佩戴对抗补丁是否逃逸检测为统一任务，引入条件攻击成功率（cASR），并将失败划分为漏检、替换和位移三类。实验表明紧凑型定位VLM可与检测器一样脆弱，且失败模式更多与模型输出格式相关，而非架构家族。

Details

Motivation: VLM正被快速部署到边缘安全关键系统中，但其物理对抗鲁棒性尚未被系统研究；同时VLM与传统检测器输出形式不同，导致难以进行公平评估。

Result: 在物理对抗基准上，紧凑型定位VLM可达到与传统目标检测器相当的脆弱性；失败模式主要与输出格式相关。仅漏检对下游系统可见，替换和位移会返回高置信度、格式良好的框，可能更危险；在分布偏移下，主要漏检的VLM会更多转向替换/位移失败。

Insight: 创新点在于提出统一指标cASR和失败模式划分，实现VLM与检测器的直接公平比较；客观来看，该工作揭示了定位VLM可能继承检测器脆弱性，并呈现更隐蔽、更危险的失败模式。

Abstract: Recent advancements in Vision-Language Models (VLMs) enable end-to-end visual grounding via language prompts and generative decoding, demonstrating largely improved zero-shot and open-vocabulary detection capabilities. As a result, VLMs are rapidly being deployed at the edge in safety-critical real-world systems. However, the physical adversarial robustness of edge VLMs has not been systematically studied, and the fundamental differences in output with modern object detectors create methodological challenges for fair evaluation. To compare the two families directly, we evaluate every selected representative model on one outcome: whether an adversarial patch worn on a person evades detection of the model. We report this by introducing a unified metric, conditional attack success rate (cASR), counted only on images the model still localizes under a matched gray occluder, and partition every failure into omission (no box), substitution (a box on the patch), and displacement (a box on neither). We benchmark the physical adversarial vulnerability of compact grounding VLMs against a range of object detectors, and find that the VLMs can be as vulnerable as the detectors. We disclose that failure mode is largely associated with the output format of a model rather than its architecture family. Of the three, only omission is visible to a downstream system; displacement and substitution return confident, well-formed boxes, making them potentially more dangerous failures. Further, we find that, under distribution shift, VLMs that primarily failed by omission instead shift to failing more by displacement/substitution. Thus, a grounding VLM can inherit the detector’s vulnerability while showing the possibility to adopt more dangerous failure modes. Our code and data are available at https://github.com/ericliang1/physical-adversarial-robustness.

[51] NeuralSRender: G-Buffer Guided Video Diffusion for Sim-to-Real Rendering cs.CVPDF

Yipeng Lu, Youqi Liao, Zhenyu Wang, Yuning Peng, Zhen Dong

TL;DR: 论文提出 NeuralSRender，一种面向自动驾驶仿真的 sim-to-real 视频翻译框架，在保持源场景几何一致性的同时提升合成视频的真实感。方法基于预训练视频扩散 Transformer（DiT），通过轻量投影器注入 G-buffer 中的几何与材质先验，并用 LoRA 进行高效适配；采用两阶段训练，先学习合成视频的 G-buffer 条件，再适配真实视频外观。论文还构建了 GenReal 驾驶视频数据集，并验证翻译后视频能提升下游感知模型性能。

Details

Motivation: 高质量标注视觉数据稀缺是自动驾驶感知的瓶颈，仿真器虽可规模化生成数据，但合成视频真实感不足会损害真实场景感知性能。因此需要一种在保持几何一致性的前提下，将合成视频转换为更真实视频的 sim-to-real 方法。

Result: 方法仅增加约 1.45% 参数开销，可在 24 GB 显存的消费级 GPU 上训练。实验表明翻译视频具有更好的照片真实感和有竞争力的几何一致性；用其训练感知模型后，nuImages 检测 AP50 从 35.1 提升到 40.6，Cityscapes 分割 mIoU 从 32.4 提升到 40.8。

Insight: 创新点在于用 G-buffer 显式几何与材质先验引导预训练视频扩散模型，实现几何一致的 sim-to-real 视频生成；两阶段训练并在第二阶段冻结投影器和第一阶段 LoRA，以降低估计 G-buffer 噪声干扰，同时保持参数高效。客观看，其可借鉴之处是将场景几何先验与扩散生成结合，并以低训练成本提升仿真数据对下游感知任务的价值。

Abstract: The scarcity of high-quality annotated visual data remains a critical bottleneck for autonomous-driving perception. Simulators provide training data at scale, but the limited realism of synthetic videos compromises perception performance in real-world applications. To address this limitation, we propose NeuralSRender, a sim-to-real video translation framework that enhances the photorealism of synthetic videos while maintaining geometric consistency with the source scenes. Built on a pretrained video diffusion transformer (DiT), NeuralSRender uses a lightweight projector to inject geometric and material priors from G-buffers, and LoRA enables parameter-efficient adaptation. The model is trained in two stages: it first learns G-buffer conditioning from synthetic videos with accurate G-buffers, then adapts to real-world appearance using real videos with estimated G-buffers. During the second stage, an additional LoRA module is trained while the projector and first-stage LoRA remain frozen to reduce interference from estimation noise. NeuralSRender introduces only 1.45% parameter overhead and can be trained on consumer-grade GPUs with 24 GB of memory. To support training and evaluation, we introduce GenReal, a large-scale driving video dataset comprising synthetic and real-world videos with corresponding G-buffers. Experiments demonstrate improved photorealism with competitive geometric consistency. Training perception models on our translated videos instead of raw simulator renders improves detection $\mathrm{AP}_{50}$ on nuImages from 35.1 to 40.6 and segmentation mIoU on Cityscapes from 32.4 to 40.8. Our code, data, and model weights are available at: https://github.com/WHU-USI3DV/NeuralSRender.

[52] DetM-Tracker: Deterministic and Memory-Efficient VIS for Autonomous Industrial Inspection cs.CVPDF

You-Rui Lin, Jen-Chieh Wu, Tzu-Ti Wei, Jen-Jee Chen, Yu-Chee Tseng

TL;DR: 论文提出 DetM-Tracker，一个面向自主工业巡检的轻量视频实例分割（VIS）框架，旨在资源受限边缘平台上实现确定性、实时且内存高效的推理。方法通过紧凑标量记忆、动态实例通道和熵门控跳跃连接三个组件减少架构冗余，压缩时序记忆并按可见目标数量自适应分配查询容量。

Details

Motivation: 现有 VIS 模型在工业巡检中存在严重扩展性问题：密集时序记忆随空间分辨率和序列长度增长，固定容量解码器又难以利用工业场景的目标稀疏性，导致边缘平台面临内存墙和实时确定性推理困难。

Result: 摘要未给出具体 benchmark 名称，但报告相较最先进的密集特征基线，推理延迟约降低 20 倍，显存（VRAM）消耗约降低 6 倍，说明其更适配工业边缘部署。

Insight: 宣称创新点在于三个协同组件：将高维实例特征压缩为四标量描述子的紧凑标量记忆、按实际可见对象数动态分配查询容量的动态实例通道，以及将计算集中到高不确定性区域并单次前向细化边界的熵门控跳跃连接。客观看，其创新集中在面向工业边缘 VIS 的确定性内存高效架构设计，以及利用场景稀疏性和不确定性进行自适应计算分配。

Abstract: Autonomous industrial inspection requires strictly deterministic real-time inference under tight memory and computational constraints. However, existing Video Instance Segmentation (VIS) models suffer from prohibitive scaling: dense temporal memory grows with spatial resolution and sequence length, while fixed-capacity decoders fail to exploit the inherent sparsity of industrial scenes, creating a practical memory wall on resource-constrained edge platforms. We present DetM-Tracker, a streamlined VIS framework for autonomous industrial inspection, where target objects often exhibit highly similar appearance under constrained “micro” fields of view. DetM-Tracker eliminates architectural redundancy through three synergistic components: (1) Compact Scalar Memory, which condenses high-dimensional instance features into four-scalar descriptors, reducing temporal memory complexity from $O(T \cdot N \cdot R)$ to $O(T \cdot \hat{N})$. This mechanism effectively compresses frame-level representations into a bank of active instances, where $R$ denotes resolution, $N$ and $\hat{N}$ represent global and window-specific instance counts, respectively and $T$ the temporal window size; and (2) Dynamic Instance Channels, which adaptively allocate query capacity according to the actual number of visible objects; and (3) Entropy-Gated Skip Connection, which concentrates computational budget on high-uncertainty spatial regions to refine object boundaries in a single forward pass without auxiliary refinement heads. Our approach effectively bridges the gap between modern VIS models and the practical demands of industrial edge deployment, reducing inference latency by approximately 20$\times$ and VRAM consumption by 6$\times$ compared to state-of-the-art dense-feature baselines.

[53] Cache Locally, Recompute Selectively for Streaming Video Understanding cs.CV | cs.LGPDF

Junhyeok Lee, Sangyoon Lee, Jaeho Lee

TL;DR: 论文面向流式视频理解（SVU）中视频-LLM低延迟处理连续视频流的需求，提出FrameBlend：在空闲时间预填充KV缓存时使用较短注意力窗口，在查询时根据用户查询选择性地对部分视频token重新预填充并替换缓存KV，从而缓解预填充上下文受限导致的上下文不匹配问题。该方法无需额外训练，可即插即用，并在多个SVU框架和基准上同时提升准确率与效率。

Details

Motivation: 现有方法利用空闲时间将视频帧预填充为KV缓存以降低查询延迟，但为控制无界视频流中的逐帧计算，预填充时限制上下文，造成上下文不匹配：既可能缺少正确语境化视频token所需信息，又包含与最终查询无关的信息，而查询在预填充时未知。

Result: 在三个SVU框架和五个流式/离线基准上，FrameBlend最高带来3.9个百分点的准确率提升；在RVS-Ego上，对基于检索的框架，每帧预填充FLOPs降低超过20%，包含查询时重计算在内的总处理时间降低14%–59%。

Insight: 创新点在于将语境化计算在空闲时间与查询时间之间重新分配：空闲时用更短注意力窗口低成本缓存，查询时用查询相关上下文选择性重计算并替换KV。其即插即用、无需训练的设计具有较强实用性和可借鉴性。

Abstract: Streaming video understanding (SVU) requires video-LLMs to process incoming video continuously while answering user queries with low latency. Recent methods reduce query-time cost by prefilling frames into KV caches during idle time and reusing the cached KVs at query time. To bound per-frame computation over unbounded video streams, these methods restrict the context available during prefill. This creates a fundamental context mismatch: the restricted context may omit information needed to properly contextualize video tokens, yet, still contain information irrelevant to the eventual query, which is unknown during prefill. We propose FrameBlend, a simple yet effective plug-and-play method to address this mismatch through selective KV recomputation, without requiring any additional training. At query time, FrameBlend re-prefills selected video tokens using query-relevant context and replaces their cached KVs with recomputed ones. Crucially, this recomputation enables shorter attention windows during idle-time prefill, substantially reducing prefill cost while improving accuracy. Experiments across three SVU frameworks and five streaming and offline benchmarks demonstrate accuracy gains of up to 3.9 pp. On RVS-Ego, FrameBlend reduces per-frame prefill FLOPs by over 20% and total processing time by 14-59% for retrieval-based frameworks, including the cost of query-time recomputation. These results demonstrate that a simple change in how contextualization is distributed between idle time and query time can improve the accuracy-efficiency trade-off for SVU.

[54] DisasterRoute: Evidence-Guided Dual-System Routing for Visual Disaster Question Answering cs.CVPDF

Tan Zhang, Xing Deng, Lu Zhang, Xiaofeng Zhu, Ping Hu

TL;DR: 论文面向低空无人机影像的灾害视觉问答，提出适应灾害领域的视觉语言模型 Disaster-S1，通过一次多模态前向推理直接判别答案选项以实现低延迟决策；并进一步提出双系统路由框架 DisasterRoute，结合预测置信度与灾害证据决定是否调用推理模型 DisasterVL 进行复核，从而按需分配计算，在保持效率的同时提升整体问答准确率。

Details

Motivation: 灾害响应场景下需要在有限边缘计算资源上快速、准确地回答无人机影像问题；现有研究多关注视觉理解与生成式推理能力，缺少对高效直接决策及其与生成式推理自适应协调的探索。

Result: 在 DisasterBench 上，Disaster-S1 以低延迟取得较高准确率；DisasterRoute 通过选择性复核进一步提升准确率，且延迟显著低于完整推理。

Insight: 创新点在于将灾害领域适应的选项判别式快速决策与生成式推理复核相结合，并利用预测置信度和灾害证据进行双系统路由，实现按需计算分配；可借鉴之处是面向资源受限场景的“快决策+选择性慢推理”自适应协同机制。

Abstract: Visual question answering from low-altitude UAV imagery must deliver timely and accurate judgments to support disaster response under limited edge computing resources. Routine questions may be answered through direct decisions, whereas more challenging questions may benefit from step-by-step reasoning. However, existing research primarily focuses on improving models’ visual understanding and generative reasoning capabilities, with limited exploration of efficient direct decision-making and its adaptive coordination with generative reasoning. We develop Disaster-S1 based on a vision-language model adapted to the disaster domain. Trained for option discrimination, it outputs a probability distribution over answer options in a single multimodal forward pass for fast decision-making. We further propose DisasterRoute, a dual-system routing framework that combines prediction confidence with disaster evidence to decide whether to accept the initial answer or invoke the disaster reasoning model DisasterVL for review. By allocating inference computation on demand, the framework retains the efficiency of direct decision-making and improves overall question-answering accuracy through selective review. Experiments on DisasterBench show that Disaster-S1 achieves superior accuracy with low latency, while DisasterRoute further improves accuracy through selective review at substantially lower latency than full reasoning.

[55] Think-Capture: Active 3D Geometry Acquisition for Spatial Reasoning in Vision-Language Models cs.CVPDF

Junjie Wang, Zhihao Yuan, Yunhan Tang, Sheng Xu, Jiaqi Mao

TL;DR: 论文提出 Think-Capture，一个以查询为条件的主动三维几何获取框架，用于提升视觉语言模型的空间推理能力。它先识别任务关键物体，再按需获取其几何信息，并通过三维工具将零样本掩码提升到统一坐标系，再用符号化三维状态抽象压缩为结构化文本。模型采用分层 GRPO 训练，学习何时依赖工具几何证据、何时回退到视觉线索。

Details

Motivation: 视觉语言模型在二维识别上表现强，但在稀疏视角、遮挡和透视变化下难以保持一致的3D空间关系，容易产生空间幻觉。现有方法通常融合所有帧的密集几何，计算和上下文开销大，并盲目依赖不完美的几何估计器。

Result: Think-Capture-4B 在 MindCube-Tiny 上达到54.4%（较骨干模型提升21.5点），在 MMSI-Bench 位置子集上达到39.1%（提升7.9点），在 ViewSpatial 上达到42.8%（提升5.2点）；同时对坐标和掩码扰动保持鲁棒，完整系统可运行于单张24GB GPU，并作为 Agibot G1 的异步高层推理器。

Insight: 创新点在于从全帧密集几何融合转向查询驱动的按需几何获取：先定位任务关键物体，再以结构化符号状态压缩物体与相机信息，并用分层GRPO显式训练模型判断几何证据的可用性。客观来看，这种“主动采集+工具使用决策+轻量结构化表示”的组合有助于降低计算开销、减少对不完美估计器的盲依赖，并提升空间推理鲁棒性。

Abstract: Vision-Language Models (VLMs) excel at 2D recognition but often fail to maintain consistent 3D relations across sparse viewpoints, leading to spatial hallucinations under occlusion and perspective changes. Existing methods commonly fuse all frames with dense geometry, incurring substantial computation and context overhead while encouraging blind reliance on imperfect estimators. We present Think-Capture, a query-conditioned framework that first identifies task-critical objects and then acquires their geometry on demand. A 3D toolkit lifts zero-shot masks into a common coordinate frame, and a Symbolic 3D State Abstraction compresses object and camera states into structured text. Hierarchical Group Relative Policy Optimization (GRPO) trains the VLM on cases in which geometry either helps or misleads, teaching it when to use tool evidence and when to fall back on visual cues. Think-Capture-4B reaches 54.4% on MindCube-Tiny (+21.5 points over its backbone), 39.1% on the MMSI-Bench positional subset (+7.9), and 42.8% on ViewSpatial (+5.2). It also remains robust under coordinate and mask perturbations. The full stack fits one 24 GB GPU and serves as an asynchronous high-level reasoner on an Agibot G1. Our code is available at https://github.com/819346675/Think-Capture.

[56] Encoded but Unused: Bounding the Sources of VLM Errors cs.CVPDF

Anat Veikherman, Mor Ventura, Nitay Calderon, Lihi Zelnik-Manor, Roi Reichart

TL;DR: 论文研究视觉语言模型（VLM）产生错误的原因，将其分解为视觉表示失败（所需视觉证据未被编码）和视觉利用失败（证据已编码但未被成功使用）。作者提出VELens探针来估计视觉利用失败率，并给出理论边界。在LLaVA-1.6-7B和Qwen2.5-VL-7B上评估，发现视觉利用失败是VLM错误的主要来源（分别占84.4%和70.0%），表明改进方向在于更好地利用已编码的视觉信息。

Details

Motivation: 现有VLM错误分析不足，论文旨在区分错误是源于视觉信息未被编码（表示失败）还是已编码但未被利用（利用失败），从而定位关键瓶颈。

Result: VELens在6000多张图像上达到ROC-AUC 0.95（LLaVA）和0.91（Qwen），优于其他探针；诊断显示视觉利用失败占LLaVA错误的84.4%、Qwen错误的70.0%，保守下界分别为77.5%和61.0%。

Insight: 创新点在于提出基于排名的VELens探针来追踪多层视觉证据，并首次量化视觉利用失败的主导比例；客观分析认为，该错误分解框架和探针设计为VLM可解释性及性能提升提供了新视角。

Abstract: In this paper, we investigate why vision-language models (VLMs) produce errors by decomposing them into two sources: visual-representation failures, in which the visual evidence required for the correct answer is not encoded, and visual-utilization failures, in which this evidence is encoded but not successfully used. We propose to estimate the visual-utilization failure rate using lightweight probes and derive theoretical bounds on the true rate that account for probe imperfections. We introduce VELens (Visual Evidence Lens), a rank-based probe that traces visual evidence for both a target object and its ground-truth attribute, such as color, material, or shape, across VLM layers. We evaluate VELens on over 6,000 images using LLaVA-1.6-7B and Qwen2.5-VL-7B. Our method achieves ROC-AUC scores of 0.95 and 0.91, respectively, and outperforms alternative probes in both low-resource and unseen-attribute settings. Using it to diagnose errors, we estimate that visual-utilization failures account for 84.4% of LLaVA errors and 70.0% of Qwen errors, with conservative lower bounds of 77.5% and 61.0%, respectively. Overall, our study identifies visual utilization as the dominant bottleneck in VLMs and suggests that a central opportunity for improvement lies in developing better mechanisms to locate, route, and use the visual information they already encode.

[57] Failure Modes of Vision-Language Models for Human Activity Understanding in Thermal HRI cs.CV | cs.ROPDF

Han Hu, He Chen

TL;DR: 论文评估了七个可本地部署的视觉语言模型（VLMs）在热成像人体活动理解中的失败模式。通过细粒度动作识别、活动时间线重建和拒答能力三项诊断任务，发现模型能识别宽泛活动类别但难以处理咳嗽等细粒度动作，并产生幻觉和错误时间线；简单预处理和额外推理仅带来有限且依赖模型的增益，拒答行为不一致。

Details

Motivation: 热像机在低光和隐私敏感场景中具有优势，但VLMs在热成像人机交互（HRI）活动理解中的失败模式尚未被充分探索，需要填补这一空白以评估其实际部署可靠性。

Result: 模型能识别广泛活动类别，但细粒度动作（如咳嗽、打喷嚏、打哈欠）识别困难；活动时间线重建不准确，并会幻觉出涉及物体的动作；简单热输入预处理和额外推理仅提供有限且模型依赖的增益；拒答行为在模型间差异大，即使有明确指令也不可靠。

Insight: 论文创新点在于系统性地诊断VLMs在热成像HRI中的失败模式，涵盖细粒度动作、时间线重建和拒答能力，并通过控制帧采样、输入表示和推理等变量评估缓解措施。客观来看，该工作揭示了热成像域与RGB域差异导致的VLM局限，对隐私敏感场景下具身AI的可靠部署具有参考价值。

Abstract: Thermal cameras enable robots to perceive human activities in darkness while revealing less personally identifiable appearance information and fewer details of the surrounding environment than RGB cameras. As vision-language models (VLMs), originally developed for RGB imagery, are increasingly integrated into embodied systems for scene understanding and high-level planning, thermal sensing offers a promising way to extend activity perception to low-light and privacy-sensitive settings. However, the failure modes of VLMs in thermal human activity understanding for human-robot interaction (HRI) remain underexplored. We fill this gap by evaluating seven locally deployable VLMs, including models developed for physical AI and robotics, on two person-centric thermal datasets. Our diagnostic tasks examine fine-grained action recognition, activity timeline reconstruction, and the ability to withhold unsupported answers. Paired RGB recordings serve as a performance reference, while controlled variations in frame sampling, thermal input presentation, and reasoning are used to assess whether the observed errors can be mitigated. Our results show that models can recognize broad activity categories but struggle with fine-grained actions such as coughing, sneezing, and yawning. They also produce inaccurate event timelines and hallucinate actions involving objects. Simple thermal input preprocessing and additional reasoning offer only limited, model-dependent gains. Abstention behavior also varies across models even under explicit instructions, limiting their reliability for practical robotic use.

[58] CRAB: Cuboid-based Representation for As-built BIM from Point Clouds cs.CV | cs.AIPDF

Haoran Xiong, Jing Wang, Shanshan Pan, Zihao Yan, Minglun Gong

TL;DR: 论文提出CRAB，一种面向竣工BIM的基于长方体（cuboid）的点云重建框架，将墙、门、窗、梁、柱编码为具有IFC兼容语义和参数的紧凑建筑长方体。方法结合ProposalNet预测粗构件长方体与AssemblyNet（关系感知Flow Matching）细化几何、预测结构关系，并支持自动结构校正和人工引导编辑；同时扩展Structured3D的构件级长方体标注并构建程序化场景。

Details

Motivation: 现有scan-to-BIM流程多依赖手工规则、启发式优化或人工修正，学习型方法难以恢复墙厚、开口几何和构件级BIM结构；论文旨在从点云重建具备语义、体量、参数化和可编辑性的建筑构件。

Result: 实验表明CRAB在构件完整性、几何精度和泛化性上优于基于楼层平面和平面抽象的方法；在内墙厚度估计上平均误差为2.80 cm，79.50%的匹配墙体误差在3 cm以内。

Insight: 创新点在于用IFC兼容的构件级长方体统一表达点云与可编辑BIM，并通过ProposalNet粗预测加关系感知Flow Matching的AssemblyNet实现几何细化、结构关系预测、自动校正和人工编辑；扩展构件级标注与程序化场景也缓解了监督不足。

Abstract: As-built BIM reconstruction from point clouds demands semantic, volumetric, parametric, and editable building components, beyond 2D layouts or planar abstractions. Yet existing scan-to-BIM pipelines often rely on handcrafted rules, heuristic optimization, or manual correction, while learning-based methods fall short of recovering wall thickness, opening geometry, and component-level BIM structure. We introduce CRAB, a BIM-ready cuboid reconstruction framework that encodes walls, doors, windows, beams, and columns as compact architectural cuboids with IFC-compatible semantics and parameters. CRAB bridges point clouds and editable BIM by combining ProposalNet, which predicts coarse component cuboids, with AssemblyNet, a relation-aware Flow Matching model that refines geometry, predicts structural relationships, and enables automatic structural correction and human-guided editing. To address limited cuboid-level supervision, we extend Structured3D with component-wise cuboid annotations and construct cuboid-based procedural scenes. Experiments show CRAB surpasses floor plan and plane-based alternatives in component completeness, geometric accuracy, and generalization. For interior wall thickness estimation, CRAB achieves a mean error of 2.80 cm, with 79.50% of matched walls within 3 cm. The project page is available at https://vcc.tech/research/2026/CRAB.

[59] Computer Vision and Knowledge-Grounded Language Models for Intelligent Beekeeper Assistance cs.CV | cs.CLPDF

Asaduz Zaman, Mohaimenul Azam Khan Raiaan, Malika Nisal Ratnayake

TL;DR: 论文提出一个面向智能养蜂辅助的边缘可部署系统框架，将蜂巢框照片通过目标检测识别具有生物学意义的巢房类别，并蒸馏为结构化事实，再转换为受约束的事实文本，进而触发有依据的知识库以生成可解释、可行动的检查报告。系统基于开源FAIRHiveFrames-1K数据集构建，并发布FAIRHiveText-1K数据集，将视觉事实与文本解释对应起来，便于后续基于LLM的研究。

Details

Motivation: 蜂群对全球食品供给至关重要，但受气候变化、极端天气及病虫害模式变化等压力影响，需要及时评估蜂群强度、子脾面积、蜂蜜和花粉储存、巢脾使用等状况；人工巢框检查慢、成本高且依赖稀缺专家经验，因此需要自动化、可解释且可在边缘部署的辅助系统。

Result: 摘要未给出具体定量实验结果或benchmark性能对比；主要贡献是构建基于FAIRHiveFrames-1K的边缘可部署系统框架，并发布FAIRHiveText-1K数据集，强调可复现和边缘部署而非大型通用前沿模型。

Insight: 创新点在于将目标检测结果蒸馏为结构化事实schema，再生成受约束事实文本并触发有依据的知识库，且知识库条目必须能由事实证据触发，从而形成可解释、可行动的报告生成链路；同时发布视觉事实与文本解释可互换的数据集，为LLM在该领域的后续研究提供基础。

Abstract: Managed and wild bees sustain roughly a third of the global food products, yet colonies face mounting stress from climate-change including extreme weather patterns, altered parasite and disease patterns. Timely assessment of colony condition such as understanding colony strength, brood area, honey and pollen storage, comb usage, is essential to adaptive management. However, manual hive frame inspection is slow, costly, and quality depends on scarce expertise. We present an artificial targeted intelligent system framework focusing on edge deployable pipeline that turns hive frame photographs into interpretable and action-oriented inspection report. The pipeline includes an object detector to detect biologically-meaningful comb-cell classes; detections are distilled into structured fact schema; facts are converted into constrained factual text; the text triggers a grounded knowledge-base for interpretations and recommendations. Knowledge-base entries are required to be triggerable by evidence expressed in the fact schema. We built our system based on open source FAIRHiveFrames-1K dataset and favour reproducible, edge-deployable models over large general purpose frontier models. We additionally release FAIRHiveText-1K dataset that turns FAIRHiveFrames-1K dataset into interchangeable facts and text interpretation of the facts, making it easy for further LLM-based research into this area. We also propose a pathway to develop the grounded knowledge-base that will be helpful in providing interpretable and actionable reports from the facts.

[60] OpenMAP-HIE: Pathology Informed Score-based Diffusion Model for Unsupervised Neonatal Hypoxic-Ischemic Encephalopathy Lesion Segmentation cs.CV | cs.AIPDF

Pengyu Kan, Craig Jones, Kenichi Oishi

TL;DR: 论文提出OpenMAP-HIE，一种病理先验引导的基于分数的扩散模型，用于无需体素级病灶标注的新生儿缺氧缺血性脑病（HIE）病灶无监督定位。模型在健康新生儿ADC图像分布上训练去噪扩散模型，并用ControlNet条件分支保留受试者解剖结构；推理时通过条件与非条件噪声预测差异作为异常证据，并结合HIE急性期ADC降低方向约束和多噪声级几何平均聚合。

Details

Motivation: 解决ADC图像上HIE病灶定位依赖稀缺专家标注且病灶外观差异大的问题，目标是在无体素级监督下实现病灶定位。

Result: 在BONBID-HIE数据集上，ADC-only方法取得Dice 0.283±0.244、AP 0.505、HD 25.45±15.63 mm、MASD 9.40±8.49 mm、NSD 0.355±0.236，在所评估无监督方法中取得最佳重叠和表面距离表现，Dice与有监督ADC-only模型相当；可选ADC Z-map细化将Dice提升至0.602±0.269。

Insight: 摘要宣称创新在于利用病理方向约束的分数差异作为异常信号，且无需最终伪健康重建或顺序反向扩散链；客观看，将ControlNet保留解剖信息与多噪声级条件/无条件噪声预测差异聚合用于无监督医学异常定位是较有借鉴意义的思路。

Abstract: Accurate localization of neonatal hypoxic–ischemic encephalopathy (HIE) lesions on apparent diffusion coefficient (ADC) images is clinically important but is limited by scarce expert annotations and substantial variation in lesion appearance. We propose OpenMAP-HIE, a pathology-informed score-based diffusion model for unsupervised HIE lesion localization without voxel-wise lesion supervision. A denoising diffusion model learns the healthy neonatal ADC distribution, while a ControlNet-based conditional branch preserves patient-specific anatomy. At inference, anomaly evidence is derived directly from differences between conditional and unconditioned noise predictions at independently evaluated diffusion noise levels, avoiding dependence on a final pseudo-healthy reconstruction or sequential reverse-diffusion chain. Based on the characteristic ADC reduction associated with acute HIE, only discrepancies consistent with the expected pathological direction are retained and aggregated across noise levels using a geometric mean. Spatial mean filtering further suppresses isolated responses. The framework was evaluated on the BONBID-HIE dataset. The ADC-only method achieved a Dice score of $0.283\pm0.244$, average precision of $0.505$, Hausdorff distance of $25.45\pm15.63$ mm, mean average surface distance of $9.40\pm8.49$ mm, and normalized surface distance of $0.355\pm0.236$, providing the best overlap and surface-distance performance among the evaluated unsupervised methods and Dice performance comparable to supervised ADC-only models. Optional population-derived ADC $Z$-map refinement further increased Dice to $0.602\pm0.269$. These findings support pathology-informed score discrepancies as a useful signal for neonatal HIE lesion localization, while highlighting remaining challenges in detecting small and low-contrast lesions.

[61] OACap: Caption-Guided Active Observation for Detailed Video Captioning cs.CV | cs.CL | cs.MA | cs.MMPDF

Hao Shi, Hong Yan, Xiaoyang Zhang, Qing Cui, Chao Huang

TL;DR: 论文提出以描述为中心的智能体OACap，用于开放式详细视频描述：它将描述中缺失或缺乏支撑的信息转化为目标观察请求，利用时空工具获取证据，并迭代验证和修订描述。论文还构建FSCap-Bench，用CSFS和CSRS分别评估事实性与跨模态语义保留，并验证OACap能提升下游视频理解性能。

Details

Motivation: 开放式详细视频描述不像问答那样由查询指定感知目标，需要模型自行判断应获取哪些证据，并将其整合为连贯的全局描述，同时减少无支撑或缺失信息。

Result: OACap生成的描述在使用固定纯文本阅读器时，在多个通用视频理解基准上取得一致性能提升；FSCap-Bench中CSFS在93.0%至98.3%的有效描述对上按预期响应四类受控扰动。

Insight: 创新点在于将详细视频描述建模为“描述—主动观察—证据获取—迭代修订”的闭环智能体流程，并引入面向事实性和语义保留的诊断基准及CSFS/CSRS指标，将下游文本记忆效用与事实、语义保真度联系起来。

Abstract: Unlike question answering, where a query specifies a perceptual target, open-ended detailed captioning must determine what evidence to acquire and how to integrate it into a coherent global description. We propose Observation-Adaptive Video Captioner (OACap), a caption-centered agent that turns unsupported or missing information into targeted observation requests, uses spatiotemporal tools to acquire evidence, and iteratively verifies and revises the caption. OACap captions consistently improve performance across general video-understanding benchmarks using a fixed text-only reader. To better diagnose the information they preserve, we introduce Factual-Semantic Captioning Bench (FSCap-Bench). Its Core-Support Factuality Score (CSFS) combines claim precision with importance-aware fact coverage and responds as expected to four controlled perturbation types on 93.0-98.3% of valid caption pairs. Its Cross-Modal Semantic Round-Trip Score (CSRS) measures semantic preservation through a video-caption-video-caption round trip by matching atomic facts from source and recovered captions. Together, they connect downstream textual-memory utility with factual and semantic preservation.

[62] Reading the Mask: Descriptor-Grounded BI-RADS Rationale Generation for Breast MRI Segmentation cs.CV | cs.AIPDF

Hrishikesh Singh, Aarti Sharma, Divyanshu Kumar, Millie Pant

TL;DR: 论文提出一种从乳腺DCE-MRI分割掩膜及其动态增强序列中推导BI-RADS临床描述符，并生成自然语言解释理由的框架。该方法将BI-RADS指南编码为决策树，结合8个影像组学定量测量、病灶大小和组件数确定描述符，再用受约束的大语言模型生成理由。作者在MAMA-MIA数据集的1,506例专家和自动掩膜上生成3,012条可复现的BI-RADS理由，并评估其忠实性和反事实准确性。

Details

Motivation: 分割掩膜只能提供肿瘤轮廓，无法给出放射科医生评估病灶可疑性时使用的BI-RADS依据，限制了模型可解释性和临床整合。论文希望从分割掩膜和动态增强序列中推导临床相关描述符，并生成结构化、可核验的理由。

Result: 在MAMA-MIA数据集1,506例上生成3,012条BI-RADS grounded rationales，忠实度达99.98%，反事实准确率达99.3%。专家掩膜与自动掩膜的描述符一致性随掩膜质量从Good降至Missed而由0.832下降到0.549。

Insight: 创新点在于将BI-RADS指南显式编码为决策树，用影像组学测量驱动描述符，并约束LLM仅表达决策树派生记录以生成可追溯理由。客观来看，该工作把分割掩膜转化为结构化临床描述和自然语言解释，建立了分割结果与临床推理之间的可解释桥梁，并提供了忠实性与反事实评估。申明生成理由仅作为放射科医生复核草稿，而非自主临床报告。

Abstract: Deep learning based models accurately delineate tumours on dynamic contrast-enhanced MRI (DCE-MRI), but segmentation masks alone do not provide the rationales or reasoning used by radiologists to assess the lesion suspicion. This limits the explainability as well the clinical integration. We investigate whether clinically relevant descriptors can instead be derived from the segmentation masks and their dynamic contrast-enhancement series. We have used Breast Imaging Reporting and Data System (BI-RADS) guidelines as a structured framework. Using these guidelines we designed a BI-RADS decision tree with eight quantitative measurements using radiomics, together with the lesion size and component count, that determine the descriptors at eight nodes. Using these descriptors we generate natural-language rationale using a large language model. The model was constrained to express only the derived records from the decision tree. After applying this framework to the expert and automatic masks of each of the 1,506 cases in the MAMA-MIA dataset, we generated a reproducible corpus of 3,012 BI-RADS grounded rationales. We obtained a faithfulness score of 99.98% and a counterfactual accuracy of 99.3%, indicating that the generated rationales faithfully verbalises the measured descriptors and correctly reflects changes in the underlying record. Agreement between the descriptors from expert and automatic masks falls from 0.832 to 0.549 as mask quality falls from Good to Missed. The generated rationales are intended as draft representations for radiologist review and final sign-off rather than autonomous clinical reports.

[63] AdvFidelity: Unifying Preference Alignment and Few-Step Image Generation within Reinforcement Learning cs.CVPDF

Sicong Kang, Fangao Zeng, Yuhan Li, Hao Zhou, Mengfei Xu

TL;DR: 论文提出 AdvFidelity，将偏好对齐与少步图像生成统一到单阶段强化学习框架中。它通过在线对抗奖励引入对低层伪影敏感的保真度监督，并让奖励梯度沿推理时使用的少步采样轨迹反传，从而在不使用 DMD 或显式蒸馏目标的情况下实现少步高质量生成。

Details

Motivation: 现有少步生成结合 RL 与蒸馏往往需要多阶段流程，且可能削弱 RL 带来的对齐收益；同时，少步 RL 虽能改善结构一致性，却会残留奖励难以惩罚的低层伪影。

Result: 在两个模型骨干和三种任务设置下，仅用 4 或 8 步且不使用 CFG，AdvFidelity 在偏好对齐、任务性能和整体生成质量上可匹配全步 LeapAlign，并与其他全步 RL 方法保持较强竞争力。

Insight: 创新点在于用在线对抗奖励将伪影敏感的保真度监督融入 RL，通过中等噪声参考图重建构造内容匹配的负样本，使判别器关注低层退化而非内容差异；同时单阶段联合优化奖励并对齐少步采样轨迹，避免显式蒸馏。

Abstract: Practical image generation requires both human preference alignment and few-step efficiency, yet combining reinforcement learning (RL) with distillation is often multi-stage and may erode RL gains We observe that few-step RL progressively improves structural coherence, even with specialized rewards, but leaves low-level artifacts that these rewards may fail to penalize. To address this gap, we propose AdvFidelity, which integrates artifact-sensitive fidelity supervision into RL through an online adversarial reward. This enables joint reward alignment and high-fidelity few-step generation within a single RL stage, without distribution matching distillation (DMD) or other explicit distillation objectives. We train a discriminator to rank high-quality references (positives) above content-matched reconstructions (negatives) produced by the current generator. Reconstructing moderately noised references largely preserves style and scene layout while exposing low-level degradations. This encourages the discriminator to assess artifacts rather than content differences. With negatives reconstructed online, alternating updates adapt fidelity supervision to the generator’s evolving failure modes. References can come from existing target-domain data or full-step model samples. Built on LeapAlign, AdvFidelity backpropagates gradients from both rewards through the few-step sampling trajectories used at inference. Across two model backbones and three task settings, AdvFidelity matches full-step LeapAlign in alignment, task performance, and overall generation quality and remains highly competitive with other evaluated full-step RL methods, using only 4 or 8 steps without CFG.

[64] VideoGen Model 1: Prompt Expansion Versus Formatting in Audiovisual Generation cs.CVPDF

David Grossman, Anton Koenig

TL;DR: 论文在视频生成模型 VideoGen Model 1（基于 MiniMax H3 后训练）上开展受控小规模实验，比较原始提示词、重格式化提示词和创意扩展提示词对音视频生成质量的影响。实验固定生成器、随机种子和采样设置，对 30 个提示词生成的片段按五个维度评分并取均值作为 1-10 总分。结果表明，相比仅做格式化，提示词扩展在 VA-Judger 下显著提升总分和提示词对齐分数，在 Gemini 3.8 Flash 下对齐分数提升但总分变化不确定，同时音频质量提升更明显，视觉质量变化较小。

Details

Motivation: 论文旨在探究音视频生成中提示词工程的具体作用：创意性提示词扩展是否比单纯格式化更能提升生成质量，尤其是整体音视频质量和提示词对齐程度。

Result: 在 30 个提示词的受控实验中，相比仅格式化，提示词扩展使 VA-Judger 总分提升 0.63 分（95% 区间 [+0.11, +1.15]），Gemini 3.8 Flash 总分提升 0.15 分（95% 区间 [-0.03, +0.35]，变化不确定）；提示词对齐分数分别提升 0.97 和 0.50。VA-Judger 还显示音频质量更高，但两种评审下视觉质量评分变化都较小。

Insight: 创新点在于系统比较提示词扩展与格式化对音视频生成的影响，并采用固定生成条件的受控实验、多评审模型和分维度评分来评估。可借鉴之处是：创意提示词扩展主要改善提示词对齐和音频质量，而单纯格式化收益有限；多评审与置信区间分析有助于更稳健地判断提示词策略效果。

Abstract: Creative prompt expansion improved overall audiovisual quality scores over raw prompts in a controlled pilot of VideoGen Model 1, our post-trained version of MiniMax H3. We compared the raw, reformatted, and expanded versions of 30 prompts, holding the generator, seed, and sampling settings fixed. Each clip was rated separately on five dimensions. Their mean formed an overall score on a 1-10 scale. Compared with formatting alone, overall scores increased by 0.63 points (95% interval: [+0.11, +1.15]) under VA-Judger and 0.15 points (95% interval: [-0.03, +0.35]) under Gemini 3.8 Flash, whose overall change remained uncertain. Prompt-alignment scores increased by 0.97 points under VA-Judger and 0.50 under Gemini. VA-Judger also showed higher audio quality; visual-quality ratings changed little under either judge.

[65] LiveAct: Towards Real-Time Multi-Agent World Models with Heterogeneous Action Interfaces cs.CVPDF

Xinyu Wang, Huafeng Shi, Yue Ma, Zijie Meng, Teng Xi

TL;DR: 论文提出 LiveAct，面向实时多智能体交互视频世界模型，在统一模型中为多个受控智能体生成联合视觉未来，并解释从二值游戏按键到连续机器人指令的异质控制。核心是异质动作接口 HAI：用智能体感知空间调制将动作流绑定到目标视觉实体，用模式引导动作适配根据语言描述与数值同步解释不同任务中的控制变量语义；随后将接口迁移到分块因果生成器以支持高效流式生成。方法在覆盖游戏和多机器人操作的六个多智能体数据源上验证。

Details

Motivation: 现有交互式视频世界模型多假设单一受控智能体和任务特定动作接口，难以同时处理多智能体独立控制以及跨任务异质动作语义。论文旨在让一个模型生成多智能体联合未来，并在统一框架中适配不同任务的控制变量含义。

Result: 在游戏和多机器人操作场景中，LiveAct 在无需智能体位置或身份裁剪的情况下取得 0.898 和 0.942 的智能体归属准确率；其因果生成器在 512×448 分辨率下达到 17.70 FPS，并评估了视觉质量、动作跟随、智能体归属和延迟。

Insight: 创新点在于用语言驱动的空间门控与智能体特定注意力实现动作流到视觉实体的独立绑定，并通过模式引导动作适配将控制变量的语言描述与数值同步，从而统一异质动作接口；进一步迁移到分块因果生成器，兼顾流式实时性与多智能体动作语义保持。

Abstract: Interactive video world models predict how an environment evolves under user actions, but most existing systems assume one controlled agent and one task-specific action interface. We study multi-agent video world modeling, where one model generates a joint visual future for multiple controlled agents while interpreting heterogeneous controls ranging from binary game buttons to continuous robot commands. This task presents two complementary challenges. First, every agent-indexed action stream must control its intended visual entity independently. Second, the model must adapt the meaning of each control variable across different tasks, where the same tensor dimension may denote an attack, steering command, or robot command. We introduce LiveAct, centered on a Heterogeneous Action Interface (HAI) whose two components mirror this decomposition. To support independent control, Agent-Aware Spatial Modulation uses language-driven spatial gates and agent-specific attention to bind each action stream to its intended visual agent. To accommodate heterogeneous controls, Schema-Guided Action Adaptation describes each control variable in language and modulates its semantic token with the synchronized numerical value. We further transfer the complete interface to a blockwise causal generator through teacher forcing and rollout-aware distribution matching, enabling efficient streaming generation while retaining agent-specific control and action semantics. We instantiate the formulation on six multi-agent sources spanning games and multi-robot manipulation, and evaluate visual quality, action following, agent attribution, and latency. Across games and robot manipulation, LiveAct reaches 0.898 and 0.942 agent-attribution accuracy without agent locations or identity crops, and the causal generator attains 17.70 FPS at $512\times 448$.

[66] LLMSeg: Training-Free Text-Prompted Image Segmentation with B-Splines cs.CV | cs.AIPDF

Abdullah Hamdi

TL;DR: 论文提出 LLMSeg，一个无需训练的流水线：把通用多模态大模型（MLLM）直接当作文本提示图像分割器，无需额外分割头、掩码解码器或任务训练。该流水线将分割拆成三步：先用 detect 定位与自由文本查询匹配的实例并描述其粗略形态，再用 segment 在检测框内用闭合周期 B 样条描绘每个实例（定义外边界和可选孔洞环），最后用 refine 针对残差错误提出聚焦框并局部重描样条，直到掩码稳定或预算耗尽。最终输出二值掩码，以及可检查的中间几何。

Details

Motivation: 现有可提示分割模型（如 SAM 3）虽具备出色的开放词汇性能，但依赖专用架构、大规模标注引擎和任务特定训练。论文追问：能否让通用 MLLM 在没有分割头、掩码解码器和训练的情况下直接完成分割？

Result: 在十个 referring/reasoning 分割划分（ReasonSeg、RefCOCO/+/g）以及开放词汇实例分割（LVIS、COCO、SA-Co）、语义分割和零样本检测（ODinW13、RF-100VL）上评测。使用现成 Gemini-3.7-Flash 后端时，在相同样本上与专用 SAM 3 相当并常超越；ReasonSeg val 达到 71.5 cIoU，且无需任何分割专用训练，后端模型升级即可提升精度而无需重训流水线。

Insight: 创新在于把分割输出参数化为可解释的闭合 B 样条几何（外边界加分孔环），并用 LLM 驱动的检测—描边—聚焦框迭代精修闭环替代分割专用网络；整个流程免训练、模块化，中间几何可检查，且能随基座 MLLM 升级自动获益。

Abstract: Promptable segmentation models such as SAM 3 achieve remarkable open-vocabulary performance but require dedicated architectures, large-scale annotation engines, and task-specific training. We ask a different question: can a general-purpose multimodal large language model (MLLM) segment images directly, without an added segmentation head, mask decoder, or training? We present LLMSeg, a training-free pipeline that turns a vision-capable LLM into a text-prompted image segmenter. LLMSeg decomposes segmentation into three LLM-driven stages. First, detect localizes all instances matching a free-form text query and describes their coarse morphology. Next, segment traces each instance using closed periodic B-splines confined to its detected box. These splines define an outer boundary and optional hole loops. Finally, refine iteratively proposes focus boxes around residual errors and retraces confined splines. The process ends when the mask stabilizes or the pass budget is exhausted. The output is a binary mask with inspectable intermediate geometry. We evaluate LLMSeg across ten referring and reasoning segmentation splits (ReasonSeg and RefCOCO/+/g). We also evaluate open-vocabulary instance segmentation (LVIS, COCO, and SA-Co), semantic segmentation, and zero-shot detection (ODinW13 and RF-100VL). With an off-the-shelf Gemini-3.7-Flash backend, LLMSeg is competitive with and often surpasses the specialist SAM 3 model on identical items. It reaches 71.5 cIoU on ReasonSeg val without additional segmentation-specific training. Backend model updates can improve accuracy without retraining the pipeline. The code is available at https://abdullahamdi.com/llmseg .

[67] Look Beyond the Question: Preserve-Focus-Fuse for Whole-Slide Image Question Answering cs.CVPDF

Xiangyu Li, Jincheng Li, Tianle Du, Yanheng Li, Haoxuan Che

TL;DR: 论文面向全切片图像问答（WSI QA），提出BEYONDPATH框架，通过Preserve-Focus-Fuse三阶段流程：先以多问题答案监督学习问题无关的形态库，再针对具体问题从所有前景patch中检索证据，最后在证据位置融合保留的形态信息，生成固定96-token视觉前缀供LLM使用。

Details

Motivation: 现有方法中，问题引导的patch选择可能遗漏回答问题所需的支持性形态，而切片级压缩可能稀释关键焦点证据；论文希望同时保留全切片形态并针对具体问题聚焦证据。

Result: 在WSI-Bench上取得Avg-9为0.769，并在同一checkpoint下领先八个报告生成指标；在SlideBench-BCNB上取得0.639的零样本总体准确率。消融实验表明形态保留与问题导向检索具有互补收益。

Insight: 创新点在于借鉴病理医生“广泛阅片—问题导向检查—整合发现”的流程，将问题无关的形态保留与问题特定的证据检索解耦并融合，避免仅依赖问题显式线索而遗漏支持性形态。客观看，其可借鉴之处是跨问题监督构建形态库，以及在检索证据位置进行形态融合的视觉前缀设计。

Abstract: A question about a whole-slide image rarely names every morphological finding needed to answer it. Question-guided patch selection can therefore overlook supporting morphology, while slide-level compression can dilute focal evidence. Inspired by how pathologists combine broad slide review, question-directed examination, and integration of their findings, we introduce BEYONDPATH, a Preserve-Focus-Fuse framework for whole-slide image question answering. Preserve learns a question-agnostic Morphology Bank through answer supervision across multiple questions about each training slide. Focus retrieves questionspecific evidence from all foreground patches, including regions outside the Bank. Fuse integrates the preserved morphology at the retrieved evidence positions through Morphology-on-Visual-Evidence Fusion, producing a fixed 96-token visual prefix. On WSI-Bench, BEYONDPATH achieves an Avg-9 of 0.769 and leads all eight reported report-generation metrics using the same checkpoint. It also achieves 0.639 zero-shot overall accuracy on SlideBench-BCNB. Controlled ablations show complementary benefits from morphology preservation and question-directed retrieval, and support cross-question supervision and fusion at the evidence positions. These results support looking beyond the question’s explicit cues: preserving broader slide morphology while retaining the ability to seek evidence for each question.

[68] Sparsity Should Be ALLOTted by Output: Training-Free Allocation of Block-Sparse Attention for Video Diffusion Transformers cs.CVPDF

Tianze Xia, Xiaoling Zhou, Ruya Lan, Yiqiao Liao, Chengru Song

TL;DR: 论文针对视频扩散Transformer中全3D注意力开销大的问题，提出一种无需训练的块稀疏注意力分配方法ALLOT。作者发现不同层与不同去噪步的稀疏化对最终视频保真度影响差异显著，但最敏感的层-步组合在不同输入间几乎不变。ALLOT将每个层-步对视为单元，通过单元探测和轨迹校准估计其对最终视频误差的贡献，并在实测延迟预算下联合优化各单元密度，仅替换现有块稀疏方法的分配组件。

Details

Motivation: 全3D注意力在视频扩散Transformer中计算昂贵，块稀疏注意力虽能降低开销，但现有方法要么忽视稀疏分配优化，要么依赖启发式规则或局部指标指导分配，难以在延迟预算下优化最终视频质量。

Result: 在相同或更短wall-clock时间下，ALLOT与XAttention结合后优于所有对比分配策略，在HunyuanVideo上取得11.8 dB PSNR增益；在Wan2.1、Wan2.2和HunyuanVideo的四个额外已发布pipeline共九种设置中，均优于或持平默认方案。

Insight: 创新点在于发现层-步稀疏敏感性差异显著且跨输入稳定，并据此以最终输出误差为目标进行训练无关的层-步密度分配；客观看，可借鉴之处是将中间注意力稀疏性与最终输出误差校准关联、利用跨输入稳定性免训练分配，并保持与现有选择策略和内核兼容的可插拔设计。

Abstract: Full 3D attention is costly in video Diffusion Transformers. Block-sparse attention reduces this cost by computing only selected attention blocks. However, existing methods either overlook the potential to optimize sparsity allocation or rely on heuristic rules or local metrics to guide it. In this work, we observe that sparsifying attention in different layers and at different denoising steps affects final-video fidelity to substantially different degrees. Yet the identities of the most sensitive layer-step pairs remain nearly unchanged across inputs. Building on these findings, we propose Allocating Layer-step Latency by Output Tests (ALLOT), a training-free sparsity allocation method. Treating each layer-step pair as a cell, ALLOT estimates each cell’s contribution to final-video error through (i) cell probing and (ii) trajectory calibration, then jointly optimizes cell densities under a measured-latency budget. It preserves existing selection strategies and kernels, replacing only the allocation component. At equal or shorter wall-clock time, ALLOT outperforms every compared allocation strategy with XAttention, with an 11.8 dB PSNR gain on HunyuanVideo. Extensive experiments across four additional published pipelines on Wan2.1, Wan2.2, and HunyuanVideo show that ALLOT outperforms or ties the default in all nine settings.

[69] Beyond Shortcut Learning: Hierarchical Spatial Prototype Regularization for Face Forgery Detection cs.CVPDF

Chao Shuai, Shaojing Fan, Kun Pan, Jinwu Hu, Weichen Lian

TL;DR: 论文针对人脸伪造检测器在未见篡改流程上泛化差的问题，指出朴素微调视觉模型会出现“取证表示坍缩”，导致表示与数据集特定伪影纠缠。为此提出分层空间原型（HSP）正则化，在空间特征上学习结构化原型子空间来建模伪造流形，并引入原型正交性与二阶分配平衡以减少捷径依赖。实验表明该方法提升跨数据集和跨操作泛化，并能作为插件增强现有SOTA检测器。

Details

Motivation: 现有深度伪造检测器容易学习到数据集特定的不稳定伪影捷径，导致取证表示结构脆弱、跨域泛化能力差；论文旨在解决取证表示本身在不同篡改来源间不稳定和坍缩的问题。

Result: 在跨数据集和跨操作设置下，帧级平均AUC分别提升5.6%和5.3%；作为插件模块使用时，将现有SOTA检测器的视频级平均AUC提升至95.3%（+1.8%）。

Insight: 论文宣称的创新点是通过空间特征上的分层原型子空间显式建模伪造流形，并用原型正交性和二阶分配平衡保持潜在伪造因素的几何独立性、提高表示有效秩利用率。客观来看，其可借鉴之处在于从表示坍缩和捷径学习角度设计即插即用正则化模块，以提升跨域泛化并兼容现有检测器。

Abstract: Recent advances in generative models have enabled highly realistic face forgeries, yet generalizing deepfake detectors to unseen manipulation pipelines remains a major challenge. One critical yet underexplored issue is that the learned forensic representation itself may fail to remain stable across different manipulation sources. In this paper, we provide new evidence that naively fine-tuned vision models suffer from severe \textit{forensic representation collapse}: their extracted cues are often entangled with unstable, dataset-specific artifacts, resulting in structurally fragile representations and poor cross-domain generalization. Leveraging this insight, we propose \textbf{H}ierarchical \textbf{S}patial \textbf{P}rototype (\textbf{HSP}) regularization, which explicitly models the forgery manifold by learning a structured prototype subspace over spatial features. We further introduce prototype orthogonality and second-order assignment balance, which jointly preserve the geometric independence of latent forgery factors and promote high-effective-rank utilization of the forensic representation, thereby reducing reliance on dataset-specific shortcuts. Empirical results demonstrate that our method substantially improves generalizability, yielding average frame-level AUC gains of 5.6% and 5.3% in the cross-dataset and cross-manipulation settings, respectively. Moreover, when used as a plug-in module, HSP boosts the average video-level AUC of existing state-of-the-art detectors to 95.3% (+1.8%). These results demonstrate the practical value of HSP as an easy-to-adopt regularization module for enhancing deepfake detectors.

[70] Extending Affordance Segmentation for Wearable Robots to Multi-Object Images cs.CV | cs.ROPDF

Simone Lugani, Matteo Montani, Paolo Gastaldo, Rodolfo Zunino, Edoardo Ragusa

TL;DR: 论文面向可穿戴机器人/上肢假肢的视觉抓取部件识别流程，针对多物体场景中目标难以消歧的问题，提出通过引入激光指示器的小幅硬件修改、重新训练深度卷积网络，并结合混合现实策略复用高质量分割掩码生成程序化图像，从而将可抓取性分割扩展到多物体图像。

Details

Motivation: 现有基于视觉的便携式流程能支持假肢自动抓取选择和预成形，但在存在多个物体的真实场景中仍难以准确定位目标物体，因此需要提升多物体场景下的目标消歧能力。

Result: 实验表明，该方法相比现有方案可将目标消歧准确率最多提升9%，同时保持计算成本不变。

Insight: 创新点在于用激光指示器作为低成本物体消歧机制，并配合定制训练与混合现实程序化数据生成，复用已有高质量分割掩码来扩展可用场景。

Abstract: This paper proposes a refinement to existing vision-based portable pipelines for the recognition of graspable object parts. Such pipelines can effectively support automatic grasp selection and preshaping in upper-limb prosthetic devices. However, realistic scenarios in which multiple objects are present still represent a challenge. The proposed method exploits a minor modification to the conventional hardware setup and a custom training approach to extend the applicability of vision-based pipelines. A laser pointer is introduced in the setup as an effective mechanism for object disambiguation. Accordingly, a suitable retraining of the deep convolutional neural networks supporting recognition of graspable parts is implemented. A mixed reality strategy allows the reuse of available high-quality segmentation masks to generate realistic images procedurally, extending the range of scenarios supported by the final device. Experimental results confirm that the proposed pipeline improves target disambiguation accuracy by up to 9% compared with existing solutions, while preserving the computational cost.

[71] Balancing Foundation Models and Trait-Specific Learning for Forehead-Crease Recognition cs.CVPDF

Geetanjali Sharma, Aryan Jain, Dev Nagaich, Aditya Nigam, Raghavendra Ramachandra

TL;DR: 论文研究在部分面部被遮挡时利用额头皱纹作为补充生物特征进行识别，统一评估了基于多模态大模型的可解释零样本推理、域自适应学习和特质特定建模三类方法。作者提出FIRE实现可解释零样本验证，并将预训练人脸识别模型通过margin-based learning适配到额头域。实验表明，虽然MLLM可提供可解释零样本验证，但特质特定学习在稳健额头皱纹识别上更有效。

Details

Motivation: 常规人脸生物识别在部分面部可见度下降、面部线索被遮挡时性能退化，而额头区域通常持续可见且具有可区分皱纹模式，因此希望将其作为补充生物特征并比较基础模型与特质特定学习的优劣。

Result: 在开放集评估下，最佳零样本模型GPT-4o+FIRE达到18.22% EER；Edge-Forehead达到8.30% EER和54.01% TMR@0.1% FMR，说明特质特定学习明显优于零样本方案。

Insight: 创新点在于统一评估可解释零样本推理、域自适应学习和额头特质建模，并提出FIRE利用多模态大模型进行可解释零样本验证；同时将预训练人脸识别模型通过margin-based learning迁移到额头皱纹域。客观看，其价值在于验证了额头皱纹作为遮挡场景补充生物特征的可行性，并揭示了基础模型可解释性与特质特定学习鲁棒性之间的权衡。

Abstract: Biometric recognition often degrades under partial facial vis- ibility, where conventional facial cues are occluded. The forehead region remains consistently visible and contains distinctive crease patterns that can serve as a complementary biometric trait. We present a unified eval- uation of forehead-crease recognition across explainable zero-shot rea- soning, domain-adaptive learning, and trait-specific modeling. FIRE en- ables interpretable zero-shot verification using multimodal large language models, while pretrained face-recognition models are adapted to the fore- head domain through margin-based learning. Under open-set evaluation, the best zero-shot model GPT-4o+FIRE achieves 18.22% EER, while Edge-Forehead achieves 8.30% EER and 54.01% TMR@0.1% FMR. The results show that although MLLMs offer interpretable zero-shot verifi- cation, trait-specific learning remains more effective for robust forehead- crease biometric recognition.

[72] AutoPhyX: Automatic Text-Condition Physics Property Generation cs.CV | cs.GRPDF

Bei Huang, Yixin Chen, Hongbin Zha, Yuru Pei, Siyuan Huang

TL;DR: 论文提出 AutoPhyX，一个文本条件驱动的物理属性预测框架，用于为 3D/4D 资产自动生成空间变化的物理参数。方法先构建部件可控的数据生成管线，生成带语义部件、物理参数和文本描述的数据，再通过跨模态调制机制让文本特征动态调制视觉特征，并在体素场中预测属性，从而兼容网格、点云、Gaussian Splatting 和 NeRF 等多种表示。

Details

Motivation: 高保真 4D 物理仿真需要与材料属性一致的准确物理参数，但现有方法要么依赖昂贵的逐场景优化，要么仅从视觉输入预测参数，难以区分外观相似但物理属性不同的材料，例如仅凭外观无法判断橡胶硬度。

Result: 实验表明 AutoPhyX 能在单次前向传播中实现物理合理、文本驱动的参数分配，具有较高准确率，并能鲁棒泛化到野外物体；摘要未给出具体 benchmark 数值或明确的 SOTA 对比。

Insight: 创新点主要在于：用文本条件弥补纯视觉物理属性预测的歧义；提出部件可控的数据生成管线以提供细粒度监督；设计跨模态调制机制实现文本对视觉特征的物理 grounding；将属性预测统一到体素场以兼容多种 3D 表示。

Abstract: High-fidelity 4D physical simulation requires accurate physical parameter assignments that align with diverse material properties. Previous methods struggle with critical bottlenecks: they either rely on computationally intensive per-scene optimization or predict parameters from purely visual inputs. The latter suffers from inherent ambiguity, e.g., the inability to distinguish the stiffness of rubber from its appearance alone. To bridge this gap, we propose AutoPhyX, a text-conditioned framework for predicting spatially-varying physical properties. Since training robust feed-forward models requires fine-grained supervision, we first introduce a novel part-controllable data generation pipeline that decomposes complex 3D assets into semantically distinct components and pairs them with physically plausible, diverse parameters and corresponding text descriptions. Leveraging this dataset, AutoPhyX employs a cross-modal modulation mechanism where text features dynamically modulate visual features for precise physical grounding. By formulating property prediction within a voxel field, AutoPhyX ensures compatibility with diverse 3D representations, including meshes, point clouds, Gaussian Splatting, and NeRF. Experiments demonstrate that our method enables physically plausible, text-driven parameter assignment in a single forward pass. It achieves high accuracy and robust generalization to in-the-wild objects, paving the way for future physics-intensive embodied and robotic applications.

[73] Supervision-Adaptive Curriculum and Guarded Anti-Curriculum Learning for Dense Prediction cs.CV | cs.LGPDF

Chunming He, Dingming Zhang, Fengyang Xiao, Zhiwen Cao, Sina Farsiu

TL;DR: 论文提出 PaceShift，一种“先稳定后挑战”的密集预测训练范式，结合监督自适应课程选择与受保护的谱后训练。第一阶段从监督、增强一致性、模型分歧和预测确定性估计样本可靠性，并用时间统计和像素不确定性加权；第二阶段在稳定后利用多谱视图排序预测，并通过相对冻结第一阶段参考的成对偏好目标与锚点进行精炼。方法兼容全监督、弱监督、半监督和模型派生监督。

Details

Motivation: 课程学习仅按难度排序样本，无法保证密集预测中的监督可靠性；简单样本可能暴露捷径，困难样本可能有信息量，噪声或模糊样本会误导训练，因此需要将监督可靠性与时间更新价值分离。

Result: 在上下文纠缠内容分割（CECS）、更广泛密集预测任务以及基础模型和跨模态适配实验中取得精度提升；诊断显示在评估设置下池化校准和损坏鲁棒性改善；方法保持部署架构不变且不增加推理计算。

Insight: 创新点在于从学习动力学视角分离监督可靠性与时序更新价值，采用“先稳定后挑战”的两阶段训练，并在稳定后通过多谱视图排序、冻结参考的成对偏好目标和锚点进行受保护的精炼。客观看，将课程选择与谱域后训练结合，并强调不增加推理开销，具有较强可迁移性。

Abstract: Curriculum learning orders samples by difficulty, but difficulty alone does not establish supervision reliability for dense prediction. Easy samples can expose shortcuts, hard samples can be informative, and noisy or ambiguous samples can mislead training. Building on our conference method CurriSeg, we present PaceShift, a stabilize-then-challenge paradigm combining supervision-adaptive curriculum selection with guarded spectral post-training. A learning-dynamics view motivates separating supervision reliability from temporal update value and applying spectral challenges after stabilization. The first phase estimates reliability from available supervision, augmentation consistency, model disagreement, and prediction certainty. Temporal statistics guide sample weighting, while pixel-uncertainty weights reduce the early influence of uncertain regions. The second phase ranks predictions from multiple spectral views using supervision agreement, robustness, reference consistency, and penalties. A pairwise preference objective measured against the frozen Phase-I reference, together with an anchor, refines the prediction. The framework supports full, weak, semi-supervised, and model-derived supervision. Experiments across context-entangled content segmentation (CECS), broader dense-prediction tasks, and foundation-model and cross-modal adaptations show accuracy gains. Targeted diagnostics indicate improved pooled calibration and corruption robustness in the evaluated settings. PaceShift preserves the deployed architecture and adds no inference computation. Code will be publicly released.

Razieh Rezaei, Ehsaneddin Asgary, Masoud Jalili Sabet

TL;DR: 论文提出 Melange，一个基于 BBQ 构建、覆盖 7 类社会偏见的合成视觉基准，用于评估多模态大模型（MLLM）的偏见。它通过受控生成视觉场景并进行 VQA 自动筛选与人工校验来保证可靠性，并在文本、图像、图像加上下文三种匹配设置下系统评测 23 个开源 MLLM，发现从文本切换到视觉会显著降低公平性。

Details

Motivation: 现有将 LLM 偏见分析扩展到多模态空间的工作要么覆盖的偏见类别较少，要么未确保被比较群体能由图像中的视觉线索清晰体现，因此需要一个类别更广、视觉线索明确且可可靠评估 MLLM 社会偏见的基准。

Result: 在 23 个开源 MLLM 上，从纯文本切换到视觉输入平均使公平性下降 0.291，且差距更多随模型家族而非模型规模变化；感知控制显示模型能以 90–95% 准确率识别图像属性但仍未能拒答，且在给出偏见答案时约四分之三选择人类刻板印象响应。

Insight: 创新点在于基于 BBQ 受控生成覆盖 7 类偏见的合成视觉基准，并用 VQA 自动筛选加人工检查保证图像-属性对应可靠；同时通过文本、图像、图像加上下文的匹配评测隔离输入模态对公平性的影响，并揭示偏见并非主要来自视觉理解失败，而是拒答与刻板印象对齐问题。

Abstract: Multi-modal Large Language Models (MLLMs) have seen broad adoption, but their biases are still far less examined than those of standard LLMs, creating possible social harms when deployed in real-world settings. Prior efforts to extend bias analyses from LLMs into the multi-modal space either cover only a few bias categories or do not ensure that the groups being compared are clearly addressed by visual cues in the images. To overcome these limitations, we introduce Melange, a wide-ranging synthetic visual benchmark covering seven categories for assessing social biases in MLLMs. It is grounded in the widely used language-only BBQ dataset (Bias Benchmark for Question Answering) from BigBench, carefully extending its constructed contexts via controlled generation of visual scenarios so the target attributes appear visually and clearly. To maximize reliability, our generated images pass through two stages of validation: automated screening using visual question answering (VQA) and subsequent manual inspection by human annotators. By systematically evaluating 23 open-source MLLMs across three matched evaluation settings—text-only, image-only, and image-with-context—Melange isolates the fairness effect of the input modality. Switching from text to vision reduces fairness by 0.291 on average, a gap that varies by model family rather than by scale alone. A perception control confirms that this gap is not an artifact of visual comprehension: models identify the depicted attributes at 90–95% accuracy yet still fail to abstain. When models do commit to a biased answer, polarity-aware stereotype alignment shows they choose the human-stereotypical response roughly three-quarters of the time.

[75] Generalized CycleGRPO for Fine-Grained Multimodal Understanding and Grounding cs.CVPDF

Xin Zhang, Haochen Wang, Beibei Lin, Zhaoxiang Zhang, Robby T. Tan

TL;DR: 论文提出Generalized CycleGRPO，一个统一的强化学习框架，通过区域理解和定位之间的循环一致性来耦合两个方向的任务，覆盖空间、时间和时空域。模型生成区域描述后再定位该描述以重建原区域，重建重叠作为可验证奖励，无需参考描述或外部语言模型评判。论文还引入DOVE基准用于评估视频描述的时间区分性，并在多个任务上取得显著提升。

Details

Motivation: 现有方法使用独立目标和任务特定标注分别学习区域理解和定位，缺乏统一框架且依赖人工标注。论文旨在通过循环一致性实现自评估，统一多模态理解与定位任务，并避免对外部评判或参考描述的依赖。

Result: 在GroundingSuite上，SAMTok的整体gIoU从57.5%提升至68.2%；在GRES上平均gIoU从71.2%提升至78.7%。时间任务上，Charades的tIoU达到51.9%（对比Qwen3-VL初始化的45.6%），DOVE的Temporal Evidence Coverage达到33.3%（对比14.8%）。

Insight: 摘要宣称的创新点包括：首个统一强化学习框架耦合区域理解和定位并跨多域；利用重建重叠作为可验证奖励实现自评估，无需参考描述或外部语言模型；引入DOVE对比基准评估时间区分性。客观分析，其创新在于将双向任务通过循环一致性统一，利用自监督信号避免标注，并展示了跨空间、时间、时空域的泛化能力。

Abstract: Fine-grained multimodal intelligence requires models to both understand a designated region and localize it from language. Existing methods learn these directions with separate objectives and task-specific annotations. We present Generalized CycleGRPO, the first unified reinforcement learning framework coupling region understanding and localization across spatial, temporal, and spatio-temporal domains. Given a region, a multimodal large language model generates candidate descriptions and then localizes each description to reconstruct the input region. Reconstruction overlap provides a verifiable reward for captioning, while individual localization rewards improve the inverse mapping. A common region-space formulation extends this cycle from image masks and boxes to video intervals and timestamped spatial evidence, without reference captions or an external language-model judge during reinforcement learning. In videos, the cycle encourages descriptions that distinguish a target event from similar recurrences. We introduce DOVE (Distinctive Occurrence-level Video-caption Evaluation), a human-verified contrastive benchmark for this temporal distinctiveness. CycleGRPO improves SAMTok’s overall gIoU from 57.5% to 68.2% on GroundingSuite and average gIoU from 71.2% to 78.7% on GRES. For temporal tasks, it achieves 51.9% tIoU on Charades and 33.3% Temporal Evidence Coverage on DOVE, compared with 45.6% and 14.8% for the Qwen3-VL initialization. These results establish region-text cycle consistency as a general self-evaluation principle for multimodal understanding and localization. Models, code, and DOVE are available at https://devinxzhang.github.io/Generalized-CycleGRPO-Page/.

[76] Unlocking Spatial Grounding in Flow-Matching Models via SNR Tuning cs.CV | cs.LGPDF

Martin Thißen, Elke Hergenröther

TL;DR: 论文研究流匹配模型中的空间定位能力，发现训练时的信噪比（SNR）分布是决定文本到图像空间布局控制的关键因素。作者构建合成几何基准并引入颜色感知布局指标，证明将训练重点转向高噪声阶段可显著提升布局遵循度，且无需辅助结构适配器或复杂推理时引导。

Details

Motivation: 基础模型图像合成效果虽好，但细粒度空间控制仍困难，通常依赖辅助结构适配器或复杂推理时操控；而现有视觉语言文本编码器虽隐含空间坐标信息，生成模型却未能有效利用。论文旨在通过调整SNR分布直接解锁流匹配模型的空间定位能力。

Result: 在合成几何基准上，中心化SNR训练的布局遵循度仅25.9%，转向高噪声关键学习阶段后提升至80.8%，接近VAE重建上限83.6%；在COCO上平均精度从1.0提升到57.3，并指出FLUX.1采用的保守噪声调度不足以实现精确布局遵循。该提升仅靠分布调优实现，无需架构修改。

Insight: 创新点在于揭示SNR分布是流匹配模型空间定位的关键决定因素，并识别高噪声训练阶段为空间学习“关键期”；同时用合成几何基准隔离纹理复杂度、引入颜色感知布局指标。可借鉴之处是通过训练噪声分布/课程调优解锁模型潜在空间控制能力，避免额外结构适配器。

Abstract: Despite the success of foundation models in image synthesis, achieving fine-grained spatial control remains a significant challenge, typically requiring auxiliary structural adapters or complex inference-time steering. While modern vision-language (VLM) text encoders possess the latent capacity to represent spatial coordinates, generative models often fail to leverage this information. We demonstrate that the signal-to-noise ratio (SNR) distribution is a critical determinant of spatial grounding in flow-matching models. Using a novel synthetic benchmark of geometric primitives, we isolate structural mechanics from textural complexity inherent in natural imagery and introduce a rigorous color-aware layout metric to quantify spatial fidelity. We show that training on a centered SNR distribution results in poor layout adherence (25.9%). Instead, we identify high-noise regimes as a “critical learning phase” for spatial grounding. Shifting the training focus toward these stages increases layout adherence to 80.8%, effectively reaching the 83.6% reconstruction ceiling of the underlying variational autoencoder. We find that the necessity for this shift scales with resolution and aspect ratio variance, and that the conservative noise schedules employed by models like FLUX.1 are insufficient for precise layout adherence. Finally, we show that these findings translate to real-world imagery via the COCO dataset (1.0 vs. 57.3 Average Precision), where state-of-the-art layout adherence is unlocked through distribution tuning alone, without architectural modifications.

[77] CineEvolve: An Agentic Self-Evolution Framework for Cinematic Multi-Shot Video Generation cs.CVPDF

Jingzhe Zhang, Peixuan Zhang, Genying Li, Bing Ma, Shuchen Weng

TL;DR: 论文提出 CineEvolve，一个面向电影级多镜头视频生成的智能体自进化框架。它将导演（Director）与评论家（Critic）统一在单个视觉语言策略中，通过交替强化学习模拟“创作—批评—修订”的迭代流程，并引入 Evolutionary GRPO 稳定共同进化，同时构建了 40K 规模的 CineTraj 电影轨迹数据集。实验表明该方法优于现有方法。

Details

Motivation: 现有电影级多镜头视频生成方法要么缺乏叙事推理能力，要么依赖无策略更新的前馈流程，难以像真实电影制作那样进行迭代精炼。论文旨在解决多镜头视频生成中叙事规划与迭代优化不足的问题。

Result: 摘要仅定性宣称 CineEvolve 在广泛实验中优于现有方法，未给出具体 benchmark、定量指标或 SOTA 对比细节。

Insight: 论文宣称的创新点包括：将 Director 与 Critic 统一为单个视觉语言策略、通过交替强化学习实现创作与批评的自进化、提出 E-GRPO 缓解优势膨胀，以及构建 CineTraj 数据集。客观来看，可借鉴之处在于把多镜头视频生成从单次前馈建模为带评论反馈闭环的策略迭代优化过程。

Abstract: Cinematic multi-shot video generation requires both narrative planning and iterative refinement. Existing methods either lack narrative reasoning or rely on feed-forward workflows without policy updates. In this paper, we propose CineEvolve, an agentic self-evolution framework for cinematic multi-shot video generation that unifies a Director and a Critic within a single vision-language policy. Emulating real-world filmmaking that couples iterative direction with critical feedback, we train CineEvolve via alternating reinforcement learning, where the Director surfaces flaws to sharpen the Critic, and the updated Critic guides the Director’s targeted revisions. To stabilize co-evolution, we introduce Evolutionary GRPO (E-GRPO), leveraging historical anchoring and revision safeguarding to mitigate advantage inflation. Additionally, we contribute CineTraj, a 40K cinematic trajectory dataset spanning creation, critique, and revision. Extensive experiments demonstrate that CineEvolve outperforms existing approaches. Our code will be available at https://github.com/JingZhe-8/CineEvolve.

[78] No Frames, Just Frequencies: Bandwidth-Efficient Deployment-Time Adaptation for Smart Glasses cs.CV | cs.LGPDF

Francesco Corti, Moritz Becker, Olga Saukh

TL;DR: 论文提出一种面向智能眼镜的带宽高效部署时自适应框架“No Frames, Just Frequencies”。该方法不在设备间传输原始相机帧，而是在眼镜端计算紧凑的频率描述子并发送到持有模型副本的主机端；主机端将频率描述子映射为分布偏移算子，合成分布外数据并按预设协议自适应模型副本。

Details

Motivation: 智能眼镜上的深度模型需要适应输入数据分布变化，但受严格隐私和设备约束，直接上传相机帧不可行；因此需要低带宽、隐私友好的部署时自适应方案。

Result: 在ImageNet-C混合损坏、Visual Wake Words人物分类和MS-COCO目标检测上，仅由传输描述子驱动的自适应分别恢复约93-95%、82%和94-98%的监督oracle自适应增益；描述子分辨率从224降至28像素时性能平缓下降，并实现213倍至784倍带宽降低。

Insight: 创新点在于用设备端计算的紧凑频率描述子表征分布偏移，并由主机端据此合成OOD数据完成源辅助自适应，避免传输原始帧；客观上看，其可借鉴之处是将频域压缩表示与分布偏移算子结合，实现隐私和带宽约束下的高效模型适配。

Abstract: Deep neural networks deployed on smart glasses must adapt to changes in the distribution of incoming data while adhering to strict privacy and device constraints. We present No Frames, Just Frequencies, a bandwidth-efficient source-assisted adaptation framework that characterizes distribution shifts using compact frequency descriptors computed on-device. Rather than transmitting camera frames, the glasses compute and send frequency descriptors to a host device that contains a replica of the model. The host device maps the received frequency descriptor to a distribution shift operator that synthesizes out-of-distribution data and adapts a model replica using a pre-defined adaptation protocol. We evaluate the framework on mixtures of ImageNet-C corruptions using convolution-based (ConvNeXtV2-Femto), self-attention-based (DeiT-Tiny), and hybrid (TinyViT-5M) vision models, on person classification (Visual Wake Words), and on object detection (MS-COCO). Our results reveal that adaptation driven solely by the transmitted descriptors recovers $93-95%$ of the supervised oracle adaptation gain on all three architectures, $82%$ on Visual Wake Words and $94$–$98%$ on MS-COCO. Notably, adaptation quality degrades gracefully as the descriptor resolution decreases from $224 $ to $28$ pixels, consistently across architectures, enabling bandwidth reductions from $213\times$ to $784\times$ across the frequency descriptors.

[79] OpenEarth-OVS44: A Multimodal Remote Sensing Dataset for Open-Vocabulary Segmentation cs.CVPDF

Yimin Wei, Hongruixuan Chen, Hao Zheng, Hao Liu, Jian Song

TL;DR: 论文提出 OpenEarth-OVS44，一个面向遥感开放词汇分割的多模态基准数据集，包含 2,832 幅 RGB 图像和 1,341 对严格配准的 RGB-SAR 影像，共 290,587 个专家核验分割样本，覆盖法国、日本、美国 40 个城市，并采用 9 大类、44 类的统一分类体系。该基准设置 RGB 单模态细粒度、带合成云干扰的多模态鲁棒性、以及跨区域留出三套评测轨道。实验表明，在其上训练可显著提升开放词汇分割模型对细粒度类别的分割性能。

Details

Motivation: 现有遥感开放词汇分割数据集类别覆盖有限、分类体系较粗，难以支撑城市功能分区、碳核算、灾后评估等实际应用所需的细粒度语义理解。论文旨在构建更精细、多模态且支持鲁棒性和跨区域泛化的遥感开放词汇分割基准。

Result: 论文在代表性开放词汇分割模型上开展实验，结果显示使用 OpenEarth-OVS44 训练后，相较最强的免训练基线，在未见类别上最高提升 7.1 mIoU，在已见类别上最高提升 27.6 mIoU。

Insight: 创新点主要在于构建了高分辨率、多模态、专家核验且分类体系统一的遥感开放词汇分割基准，并提供 RGB 细粒度、合成云天气鲁棒和跨区域留出三类评测设置。从客观角度看，其价值在于填补细粒度遥感开放词汇分割数据与鲁棒泛化评测的空白。

Abstract: In remote sensing, open-vocabulary segmentation is particularly important for flexible and scalable land-cover understanding across diverse regions, as it improves generalization to novel categories without requiring exhaustive pixel-level annotations. Despite its great promise, progress in this area remains largely constrained by existing datasets with limited category coverage and overly coarse taxonomies, and therefore struggles to support fine-grained semantic understanding required in real-world applications such as urban functional zoning, carbon accounting, and post-disaster assessment. To address these challenges, we present OpenEarth-OVS44, a multimodal benchmark with 290,587 expert-verified segments from 2,832 RGB images and 1,341 strictly registered RGB-SAR pairs at sub-meter spatial resolution, covering 40 cities across France, Japan, and the United States under a unified 44-class taxonomy organized into 9 base categories. Three evaluation tracks are provided: an RGB-only fine-grained benchmark, a multimodal track with diverse synthetic cloudy conditions for weather-robust evaluation, and a held-out cross-region track. Extensive experiments on representative open-vocabulary segmentation models demonstrate that training on OpenEarth-OVS44 substantially improves fine-grained segmentation performance over the strongest training-free baseline, with improvements of up to 7.1 mIoU points on unseen categories and 27.6 mIoU points on seen categories. The dataset and code will be released.

[80] MS2CAST: A Video Anomaly Detection Framework via Multi-Scale Spatial Cross-Attention and Efficient Spatiotemporal Networks cs.CVPDF

Ali Safaei, Mohammad Kazemi

TL;DR: 论文提出MS2CAST，一个端到端半监督预测式视频异常检测框架。它用多尺度交叉注意力编码器融合ViT（MViTv2-S）和CNN（ConvNeXt-T）的分层特征，并在隐空间引入SwinLSTM或VMRNN两种可互换时空模块来替代光流建模运动动态，同时结合L1、GDL和MS-SSIM损失进行预测学习。

Details

Motivation: 视频异常检测中，现有半监督方法常依赖计算量大的光流网络捕捉运动，而CNN空间编码的有效感受野偏局部。论文希望在不使用光流估计的情况下，高效融合不同感受野的空间特征并建模时空动态。

Result: 在UCSD Ped2、CUHK Avenue和ShanghaiTech数据集上，帧级ROC AUC与顶级方法具有竞争力，同时峰值GPU显存占用低于1GB。

Insight: 创新点在于多尺度交叉注意力融合ViT与CNN的异构空间特征，并用隐空间中的SwinLSTM或Vision Mamba时空模块替代光流；客观来看，其可借鉴之处是面向低显存的高效端到端半监督预测框架设计，以及多种损失组合对像素精度、锐度和结构完整性的兼顾。

Abstract: Video anomaly detection (VAD) is a challenging task for surveillance systems due to the scarcity and diversity of anomalous events. While semi-supervised VAD approaches have advanced the field, many rely on computationally intensive optical flow networks to capture motion dynamics. Convolutional neural networks (CNNs) are also widely used for spatial encoding, but their effective receptive field is highly local. To address these limitations, we introduce MS2CAST, an end-to-end semi-supervised predictive VAD framework. For spatial encoding, we employ a multi-scale cross-attention encoder that fuses the hierarchical feature representations from a vision transformer (MViTv2-S) and a CNN (ConvNeXt-T) to exploit their distinct receptive fields. To model spatiotemporal dynamics without optical flow estimation, we integrate two interchangeable spatiotemporal modules in the latent space. One module (SwinLSTM) incorporates a Swin Transformer, while the other (VMRNN) integrates Vision Mamba blocks. Our objective function combines L1 loss, Gradient Difference Loss (GDL), and Multi-Scale Structural Similarity Index Measure (MS-SSIM) to ensure pixel-wise accuracy, sharpness, and structural integrity. Evaluations across the UCSD Ped2, CUHK Avenue, and ShanghaiTech datasets demonstrate frame-level Receiver Operating Characteristic (ROC) Area Under the Curve (AUC) scores that are competitive with top-tier techniques while maintaining a peak GPU memory usage of under 1 GB.

[81] Learning to Measure and Diagnose: A Clinical-Dictionary Radiomic Bottleneck Model for Thyroid Nodule Classification cs.CVPDF

Guibin Zhang, Weier Luo, Lei Cao

TL;DR: 论文提出一种面向甲状腺结节分类的临床字典影像组学瓶颈模型，通过从超声图像中估计临床字典定义的影像组学测量，并将其作为可解释瓶颈用于分类，同时保持测量保真度。模型包含家族感知影像组学学习（FARL）和保真度引导瓶颈建模（FGBM），并通过线性分类器分解决策证据，按临床字典类别组织特征贡献。

Details

Motivation: 甲状腺超声结节分类预测往往缺乏与临床熟悉描述相联系的证据；影像组学测量虽可提供明确、可解释的定量依据，但需要从超声图像中可靠估计，并在优化分类性能时保持测量保真度。

Result: 在ThyroidXL和TN3K数据集上取得有竞争力的分类性能，并通过良恶性结节案例展示了特征贡献，表明能够提供可检查的定量证据和可解释分类；摘要未给出具体数值或SOTA对比。

Insight: 创新点在于将临床字典定义的影像组学测量作为可解释瓶颈，结合FARL的家族间平衡监督和FGBM的保真度引导联合优化，并通过线性分类器实现决策证据分解与临床类别组织，兼顾分类性能与可解释性。

Abstract: Thyroid ultrasound supports the assessment of nodule malignancy, but classification predictions often lack evidence linked to familiar clinical descriptions. Radiomics provides explicitly defined measurements that can support such explanations through a clinical dictionary. However, using these measurements as interpretable evidence requires reliable estimation from ultrasound images and preservation of measurement fidelity when optimizing classification performance. We propose a clinical-dictionary radiomic bottleneck model for thyroid nodule classification. Within this model, Family-Aware Radiomic Learning (FARL) estimates measurements with balanced supervision across feature families. Fidelity-Guided Bottleneck Modeling (FGBM) constructs a fixed bottleneck, followed by joint optimization of measurement estimation and classification under fidelity guidance. Decision evidence decomposition then traces feature contributions through a linear classifier and organizes them by clinical dictionary categories. Experiments on ThyroidXL and TN3K show competitive classification performance, while case studies illustrate feature contributions for benign and malignant nodules. These findings indicate interpretable classification with inspectable quantitative evidence informed by clinical knowledge. Code is available online.

[82] APED-CXR: Adaptive Privileged Evidence Distillation for Chest X-Ray Grounding with Image-Only Inference cs.CVPDF

Md Shihabul Islam Shovo, Nishi Kanta Paul, Jamil Fayyad

TL;DR: 论文提出 APED-CXR，将冻结的多模态教师模型中由报告信息提供的空间证据，蒸馏到一个仅 11.1M 参数的紧凑学生模型中。学生模型在测试时只需胸部 X 光图像，即可同时预测疾病并输出病灶框；训练时教师根据学生当前证据图提出的区域，结合图像、报告和查询类别评估区域支持分数，并作为学生池化证据分数的监督目标。

Details

Motivation: 紧凑 CXR 模型适合部署且能定位病灶，但框标注稀缺，而依赖报告的多模态模型推理成本高。论文希望在仅图像推理条件下，利用报告监督提升疾病预测和框级定位能力。

Result: 在 PadChest 训练、PadChest-GR 评测中，动态刷新学生提案供教师评估使框级定位 F1 从 0.302 提升到 0.338（IoU>=0.5，提升 0.036，95% CI [0.022, 0.049]），并保持诊断 AUC；在 NIH ChestX-ray14 上次要测试的框召回率从 0.423 提升到 0.447。消融显示该增益依赖教师检查当前提案。

Insight: 创新点在于从冻结多模态教师中蒸馏报告启发的空间证据，并让学生用共享证据图联合完成疾病预测与定位，且推理仅需图像。客观看，其可借鉴之处是让学生提案动态刷新并接受教师评估，从而在缺少框标注时获得更有效的定位蒸馏信号。

Abstract: Compact chest X-ray (CXR) models that localize findings suit deployment, but box annotations are limited and report-aware multimodal models are expensive at inference. We introduce APED-CXR, which distills report-informed spatial evidence from a frozen multimodal teacher into a student with 11.1M parameters that predicts diseases and boxes from shared evidence maps and requires the image alone at test time. During training, the teacher uses the radiograph, report, and queried class to evaluate regions proposed from the student’s current evidence maps, and its support scores serve as training targets for the student’s pooled evidence scores within these regions. Trained on PadChest and evaluated on held-out PadChest-GR, refreshing the student’s proposals for teacher evaluation improves box-level localization F1 from 0.302 with fixed warm-up proposals to 0.338 (IoU >= 0.5; gain 0.036, 95% CI [0.022, 0.049]), while maintaining diagnostic AUC. This gain depends on the teacher checking current proposals: with equally many random boxes, a fixed dense box set, teacher-generated boxes, or report rules instead of the teacher, F1 stays near the fixed-proposal level, and proposals two refreshes old recover only part of the gain. On NIH ChestX-ray14, a secondary test set, box recall also rises from 0.423 to 0.447.

[83] LEAF-Mamba: Feature Adaptation and Fusion for Whole Slide Image Analysis cs.CVPDF

Lubin Gan, Jing Zhang, Mingjun Ma, Linhao Qu, Dachun Kai

TL;DR: 论文提出LEAF-Mamba框架，用于冻结预训练特征的全切片图像（WSI）分析，联合进行特征适应与融合。方法包含重建正则化适配器，在学习任务特定紧凑表示的同时重建输入特征；并采用并行Mamba流与卷积流分别建模长程依赖和序列局部交互，再通过输入依赖门控融合后做切片级池化。

Details

Motivation: WSI为千兆像素图像且标注有限，切片级学习困难；现有MIL依赖冻结编码器聚合patch表示，Mamba虽能高效建模长patch序列，但任务监督未显式约束特征适应中的信息恢复，且Mamba将局部处理与长程传播耦合，缺少用于融合的独立表示。

Result: 在多种特征提取器下，LEAF-Mamba在诊断分类和生存预测任务上的平均性能优于强MIL基线；组件分析进一步支持特征适应与融合模块的贡献。

Insight: 摘要宣称的创新点在于重建正则化适配器显式约束特征适应中的信息恢复，以及并行Mamba与卷积双流加门控融合局部交互和长程依赖。客观来看，其可借鉴之处是将冻结特征的任务适应与上下文聚合联合设计，并用重建正则和双流门控缓解信息损失并融合互补表征。

Abstract: Whole slide image (WSI) analysis enables computational diagnosis and prognosis, but gigapixel images and limited annotations make slide-level learning challenging. Multiple instance learning addresses this task by aggregating patch representations from frozen pretrained encoders. Recently, Mamba has enabled efficient modeling of long patch sequences. However, task supervision alone does not explicitly constrain information recovery during feature adaptation, while Mamba couples local processing with long-range propagation rather than retaining separate representations for fusion. Motivated by these challenges, we introduce LEAF-Mamba, a framework that jointly addresses latent evidence adaptation and fusion for frozen-feature WSI analysis. A reconstruction-regularized adapter learns compact, task-specific representations while reconstructing the input features. Parallel Mamba and convolutional streams then capture long-range dependencies and sequence-local interactions, with an input-dependent gate modulating their joint representation before slide-level pooling. Experiments across feature extractors demonstrate improved average performance over strong MIL baselines in both diagnostic classification and survival prediction. Component analyses further support the contributions of adaptation and fusion. These findings highlight the value of jointly designing feature transformation and contextual aggregation to translate pretrained representations into effective slide-level predictions.

[84] UniRAP: Towards Unified Part-level Physical Affordance Reasoning and Actionable Perception cs.CVPDF

Linfei Li, Ruining Hu, Lin Zhang, Zhong Wang, Fengyi Zhang

TL;DR: 论文提出 UniRAP 统一模型，用于部件级物理可供性推理，并将自然语言指令映射为可操作的几何表示。模型将视觉输入、文本指令和可选提示统一到共享时空表示空间，并通过统一可供性解码器联合完成目标检测、部件级可供性分割和四自由度交互位姿估计。实验表明该方法在多个任务和动态视频场景中均表现有效。

Details

Motivation: 现有视觉-语言感知虽能对齐语言与视觉观察，但难以将高层语义落实到部件级物理交互中，缺少面向可操作感知的统一建模。论文旨在解决语言引导下部件级可供性推理与可执行接触几何预测的问题。

Result: 论文称 UniRAP 在指代表达分割、可供性定位和交互位姿估计上达到 SOTA，并在动态视频场景中保持较强的时空一致性。

Insight: 创新点在于将部件级可供性推理统一为条件多模态生成任务，设计统一接口 token 机制和统一可供性解码器，并引入课程式迁移训练以提升有限文本标注下的数据效率。客观上，其可借鉴之处是联合建模检测、部件分割与交互位姿，并利用中间分割特征增强可操作几何预测。

Abstract: Vision-language perception has achieved impressive progress in aligning natural language with visual observations, yet grounding high-level semantics into part-level physical interaction remains challenging. To address this gap, we propose UniRAP, a unified model for inferring part-level physical affordances and mapping language instructions to actionable geometric representations. UniRAP formulates this problem as a conditional multimodal generation task, integrating visual inputs, textual instructions, and optional prompts into a shared spatiotemporal representation space through a unified interface token mechanism. To predict executable contact geometry, we introduce a Unified Affordance Decoder (UAD), which jointly performs object detection, part-level affordance segmentation, and 4-DoF interaction pose estimation by leveraging intermediate segmentation features. In addition, we propose a curriculum-based transfer training strategy that progressively adapts the model from general visual parsing to interaction-aware perception, improving data efficiency under limited textual annotations. Experiments show that UniRAP achieves state-of-the-art performance on referring expression segmentation, affordance grounding, and interaction pose estimation, while maintaining strong spatiotemporal consistency in dynamic video scenarios. These results demonstrate the effectiveness of UniRAP as a unified perception framework for language-guided physical manipulation.

[85] JointLight: Human-Centric Video Relighting with Asynchronous RGB-Irradiance Joint Modeling cs.CVPDF

Henglei Lv, Bailin Deng, Xiaoqiang Liu, Yichen Wang, Haoxian Zhang

TL;DR: 论文提出 JointLight，一种面向人像视频重打光的方法，通过异步 RGB-辐照度联合建模实现细粒度光照控制。作者构建了基于游戏引擎的多光照高质量合成视频数据集及配对辐照度图，并设计两阶段训练与推理阶段的异步劫持生成流程。方法仅用合成数据训练，但能泛化到真实场景，并在光照保真度和生成质量上优于已有方法。

Details

Motivation: 现有视频重打光方法难以实现精确光照控制，主要原因包括高质量、精确标注的重打光视频数据集稀缺；端到端方法隐式建模光传输导致可控性差；而路径追踪辐照度线索易含伪影，直接作为控制信号会损害生成质量。

Result: 摘要未给出具体定量指标或 benchmark 数值，仅声称模型仅在该合成数据集上训练即可稳健泛化到真实世界场景，并在光照保真度和生成质量两方面优于先前方法。

Insight: 创新点在于利用噪声腐蚀去除高频伪影、保留低频光传输的观察，提出异步联合建模 RGB 与真实辐照度帧，并采用分离时间步调度、第二阶段域匹配以及推理时的异步劫持生成管线，从而兼顾光照可控性与生成质量。

Abstract: Understanding and modeling light transport in human-centric scenes is a longstanding challenge. While recent progress in diffusion models has boosted research in relighting, precise illumination control still lags, primarily due to the scarcity of diverse, high-quality, precisely labeled video relighting datasets. Existing end-to-end relighting approaches model light transport implicitly through neural networks, which substantially complicates controllability. In contrast, some works resort to path-traced irradiance cues as an explicit blueprint for light transport. However, such constructed irradiance cues are prone to artifacts and hinder generative quality when directly used as control signals. With the key observation that noise corruption removes high-frequency artifacts and preserves low-frequency light transport, we introduce an asynchronous joint-modeling paradigm that facilitates fine-grained illumination control. During two-stage training, we force the model to jointly denoise RGB and ground-truth irradiance frames with separate timesteps scheduling. A domain matching strategy is applied to the second stage to bridge the domain gap between irradiance cues and ground-truth irradiance maps. At inference time, we introduce an asynchronous hijacked generation pipeline that facilitates flexible illumination control. For training data, we develop a multi-illumination high-quality synthetic video dataset with paired irradiance maps using a powerful game engine. Trained exclusively on this dataset, our model exhibits robust generalization to real- world scenarios and outperforms prior approaches in both lighting fidelity and generative quality. The code and dataset will be publicly released.

[86] Compositionality with Primitives: A Unified View from Euclidean to Hyperbolic Vision-Language Models cs.CVPDF

Puru Vaish, Meelad Dashti, Christoph Brune, Pascal Mettes, Tatiana Tommasi

TL;DR: 论文研究组合零样本学习（CZSL），即识别已见对象与属性的新组合。作者重新审视了“必须显式重组属性-对象对”的假设，基于嵌入分布切空间中的原始概念方向，提出新的组合评分函数，并系统比较欧氏、球面和双曲视觉-语言模型中的组合推理。实验表明，直接使用分解出的原始表示即可识别未见组合，并在多个CZSL与组鲁棒性基准上验证。

Details

Motivation: 现有工作认为大规视觉-语言模型具有可分解语义结构，可通过重组实现组合泛化；本文关注是否真的需要显式的属性-对象对重组，并分析嵌入流形几何与数据集结构对组合推理的影响。

Result: 摘要未给出具体数值，但作者在CLIP、DINOv2、MERU、HyCoCLIP等模型上于MIT-States、VAW-CZSL、UT-Zappos50K、Clothing16K等CZSL基准，以及Waterbirds、CelebA组鲁棒性任务上进行了广泛评估，表明直接使用分解基元即可支持未见组合识别。

Insight: 创新点在于提出统一的欧氏到双曲视角，用新的组合评分函数验证原始表示本身足以进行组合泛化，并系统揭示嵌入流形几何和数据集结构对组合推理的影响；可借鉴之处是避免显式重组、直接利用可分解基元以及跨几何空间的统一分析框架。

Abstract: Compositional zero-shot learning (CZSL) aims to recognize novel combinations of previously observed objects and attributes. Prior work showed that large-scale vision-language models encode decomposable semantic structures that can be recomposed to enable generalization. The Geodesically Decomposable Embeddings (GDE) framework formalizes this by identifying directions in the tangent space at the Frechet mean of the embedding distribution, which encode primitive visual concepts. Unseen attribute-object compositions can be synthesized through vector addition of these primitives. We revisit the assumption that compositional generalization requires explicit attribute-object pair re-composition. By means of new compositional scoring functions and extensive evaluations, we show that directly using the decomposed primitives suffices for recognizing unseen compositions. In that, we gain insights on how the geometry of the embedding manifold (Euclidean, spherical, or hyperbolic) and dataset structure affect compositional reasoning. Across models (CLIP, DINOv2, MERU, HyCoCLIP), we show that compositional structure consistently emerges as primitive representations, enabling inference on CZSL benchmarks (MIT-States, VAW-CZSL, UT-Zappos50K, Clothing16K) and group robustness on Waterbirds and CelebA. Code: https://github.com/nis-research/Compositionality-with-Primitives.

[87] GeoMem-VLA: Internalizing 3D Scene Understanding into Temporal Memory for Vision-Language-Action Models cs.CVPDF

Qinfan Zhang, Yifan Luo, Haoyuan Shi, Guanlin Liu, Han Dong

TL;DR: 论文提出GeoMem-VLA框架，将显式3D监督内化为可学习的几何token（GeoPlan），并将其作为时间记忆持续更新，用于视觉-语言-动作模型的动作生成。在训练时利用冻结几何基础模型通过流匹配监督，推理时无需3D教师或场景重建，在LIBERO、RoboTwin及真实任务上取得高成功率。

Details

Motivation: 多数VLA模型依赖当前观测的2D表示，而几何感知方法需额外3D信息；但独立从单帧推断的几何信息常被丢弃，未累积为持久任务表示，且每步完整3D重建并非必需。

Result: 在LIBERO上平均成功率98.9%，RoboTwin干净/随机设置下分别为91.10%/91.24%，五个真实世界任务平均成功率78.0%。

Insight: 创新在于将3D监督内化为紧凑、可连续更新的几何token，并与观测表示的时间记忆自适应融合，避免直接对齐VLM原生视觉嵌入，推理时无需调用3D教师或显式重建场景。

Abstract: Geometric structure is essential for robotic manipulation and understanding the 3D world. Most VLA models rely on 2D representations of current observations, while geometry-aware approaches often require additional 3D information. However, geometry inferred independently from individual observations is often discarded rather than accumulated into a persistent task representation. Our key insight is that action generation needs a compact, task-relevant geometric abstraction that can be continuously updated, rather than a complete, renderable 3D reconstruction at every step. We introduce GeoMem-VLA, a VLA framework that internalizes explicit 3D supervision into dedicated geometric tokens and maintains them as temporal memory for action generation, without directly aligning the VLM’s native visual embeddings with the teacher feature space. During training, a frozen geometry foundation model supervises learnable GeoPlan tokens through a geometric flow-matching objective over the current scene point map and future end-effector tracks. These tokens are integrated with the memory of observation representations, where historical information is retrieved, adaptively fused with current observations, and consolidated to retain event frames as the task progresses. At inference, the policy retrieves historical representations and fuses them with the current working representation to predict actions, without invoking the 3D teacher or explicitly reconstructing the scene. After benchmark-specific downstream adaptation, GeoMem-VLA achieves 98.9% average success on LIBERO and 91.10% / 91.24% on RoboTwin’s clean/randomized settings. Across five real-world tasks, GeoMem-VLA achieves 78.0% average success. These results suggest that internalizing and remembering 3D scene understanding is a promising approach to robot control.

[88] Expandable Modality Experts for Medical Vision-Language Learning cs.CVPDF

DM BAPPY, Eunsu Kim, Hanjoo Lee, Sung Hak Lee, Sangjeong Ahn

TL;DR: 论文提出一种面向医学视觉-语言学习的可扩展模态专家方法：用可扩展的模态专家库和Multiway Transformer，将共享视觉计算、确定性路由的前馈专家与独立文本通路结合；先利用模态感知视觉分词器在1940万条无配对图像上进行掩码预训练，再用323万图文对做对比对齐；新增眼底和OCT专家时冻结原有图像与文本嵌入路径。

Details

Motivation: 医学影像跨模态学习需要利用无配对文本的影像，并在保留已有表示的同时适应新增模态和领域，避免灾难性遗忘。

Result: 以ViT-Large为骨干，在9种模态的25个基准设置中24个达到最佳或并列最佳：11个零样本分类中10个、10个线性探针全部、4个检索全部；零样本平均平衡准确率0.674，线性探针0.856，平均双向检索召回率20.70%。三随机种子序贯实验中，眼底和OCT检索分别提升14.95和9.31个百分点，且前序眼底/病理任务无测得性能损失。

Insight: 创新点在于用可扩展模态专家库实现模块化持续扩展，并通过确定性路由、独立文本通路和冻结既有计算路径来保持旧能力；结合模态感知分词器的大规模无配对掩码预训练与图文对比对齐，为医学多模态表示学习提供了可扩展方案。

Abstract: Learning across medical imaging modalities requires making use of images without paired text and accommodating new domains while retaining existing representations. We propose a medical vision-language technique built around an expandable bank of modality experts. Its Multiway Transformer combines shared visual computation with deterministically routed feed-forward experts and a separate text pathway. A modality-aware visual tokenizer supplies targets for masked pretraining on 19.4 million unpaired image rows across seven modalities; contrastive alignment then uses 3.23 million image-text pairs. Additional fundus and optical coherence tomography experts learn from paired data while the original image and text embedding paths remain frozen. With a ViT-Large backbone, the proposed technique achieves the best or tied-best result in 24 of 25 benchmark settings across nine modalities: 10 of 11 zero-shot classification benchmarks, all 10 linear probes, and all four retrieval benchmarks. Mean balanced accuracy is 0.674 for zero-shot classification and 0.856 for linear probing, and mean bidirectional retrieval recall is 20.70%. A separate three-seed sequential experiment with matched adaptation data yields fundus and OCT retrieval gains of 14.95 and 9.31 percentage points, with no measured loss on preceding fundus or pathology tasks. The preservation argument requires unchanged existing computation paths and inference conditions, and feature-identity checks verify retention on evaluated inputs. The proposed technique provides a modular approach to learning and expanding medical vision-language representations.

[89] ViBMR: A Large-Scale Multi-Sequence Brain MRI Dataset and Comprehensive Evaluation for Vietnamese Radiology Report Generation cs.CVPDF

Dai Lam Le, Minh Khoi Nguyen, Son Dinh Nguyen, Amir Reza Jafari, Viet Ha

TL;DR: 论文提出了ViBMR，一个面向越南语放射报告生成的大规模多序列脑部MRI数据集，包含3,649名患者和11,875个MRI序列，覆盖T1、T2、FLAIR和增强T1序列。基于该数据集，论文建立了脑部MRI报告生成的综合评测设置，评估代表性医学视觉-语言模型、多序列输入及特征聚合策略，并分析了数据集特点与模型行为。

Details

Motivation: 现有脑部MRI报告生成资源以英文报告、翻译临床文本或任务特定影像基准改编数据为主，难以覆盖低资源临床语言和真实多序列MRI协议；同时脑部MRI具有体数据和多序列互补诊断信息的特点，标注数据稀缺。

Result: 论文构建了首个大规模越南语多序列脑部MRI报告数据集，并给出多模型、多序列输入与特征聚合策略的综合评测；结果强调原生低资源临床数据集对可靠越南语脑部MRI报告生成的必要性，摘要未给出具体定量指标或SOTA结论。

Insight: 创新点在于构建面向低资源临床语言的真实多序列脑部MRI-报告配对数据集，并系统评估多序列输入和特征聚合对医学视觉-语言模型报告生成的影响，为低资源医学报告生成提供数据与评测基准。

Abstract: Automated radiology report generation for multi-sequence brain MRI holds strong promise for alleviating radiologists’ workload, yet it remains a demanding problem: brain MRI studies are inherently volumetric, each pulse sequence carries complementary diagnostic information, and annotated data are critically scarce for low-resource clinical languages. Existing brain MRI-report resources are still dominated by English reports, translated clinical text, or datasets adapted from task-specific imaging benchmarks, leaving routine low-resource clinical reporting and heterogeneous multi-sequence MRI protocols insufficiently represented. To bridge this gap, we introduce ViBMR, to the best of our knowledge, the first large-scale multi-sequence brain MRI dataset paired with routine Vietnamese clinical reports, comprising 3,649 patients and 11,875 MRI series across T1, T2, FLAIR, and contrast-enhanced T1 sequences. Building on ViBMR, we establish a comprehensive evaluation setup for brain MRI report generation, covering representative medical vision-language models, multi-sequence inputs, and feature aggregation strategies. The results provide key insights into dataset characteristics and model behavior, while demonstrating the necessity of native low-resource clinical datasets for reliable Vietnamese-origin brain MRI report generation. Our dataset and baseline models will be released for non-commercial research.

[90] Asking a Multimodal LLM Whether Someone Fell: Zero-Shot Screening of Low-Resolution Thermal Video in a Laboratory and an Occupied Home cs.CV | cs.AIPDF

Marcos Junca, Mauro Junca

TL;DR: 论文评估将生产级多模态大模型 Gemini 3.8 Flash（temperature=0）作为事件触发式跌倒筛查器的决策环节，以零样本方式判断低分辨率热成像短视频中是否发生跌倒；在公开 32x24 热成像数据集的 438 次摆拍跌倒和 563 个非跌倒窗口上测试，并在有人居住的真实房间录制 76 小时（160x120 带声音）进行验证，同时与两个传统检测器对比。

Details

Motivation: 独居老人需要跌倒监护，而热成像相机不拍摄人脸、可保护隐私；希望利用多模态 LLM 无需任务特定训练即可判断短视频的能力，降低部署与标注成本。

Result: 在实验室数据上，该模型对留出受试者跌倒的召回为 69%（全部受试者为 67%，排除床上跌倒），对 499 个非跌倒窗口零误报；在真实居家场景的 76 小时、395 段视频中零误报（95% 置信上界为每天 1.2 次误报），但仅捕捉到 9 次可见跌倒中的 4 次；删除提示中一句警示语后，召回升至 83%（Gemini 3.7 Flash 为 92%），居家场景捕捉到 9/9 次跌倒，但会把故意躺下误判为跌倒；分辨率降至 8x6 像素时召回跌至 34%，相同请求结果不一致率为 4%，模型解释中的跌倒时刻 99% 落在标注着地时间 1 秒内；两个传统检测器在实验室几乎全召回，但在居家场景每天分别产生 14.2 和 23.9 次误报。

Insight: 创新点在于零样本调用生产级多模态 LLM 作为跌倒筛查决策环节，结合热成像保护隐私，并在真实居家环境中做长时验证；客观来看，提示措辞对灵敏度影响巨大、低分辨率下性能急剧下降，以及模型解释时间戳高度准确（99% 在 1 秒内），这些发现对提示设计与部署权衡具有借鉴价值。

Abstract: Thermal cameras can watch over older people living alone without capturing faces, and multimodal large language models (LLMs) can judge a short video without task-specific training. We evaluate a production multimodal LLM, Gemini 3.8 Flash at temperature 0, as the decision stage of an event-triggered fall screener, used exactly as a deployed prototype uses it: one request per short thermal clip, with its sound track when there is one, and its predecessor Gemini 3.7 Flash on the same requests. On 438 staged falls and 563 non-fall windows that we located by hand in a public 32x24 thermal dataset, the deployed request called 69% of the held-out subjects’ falls (67% over all subjects; bed falls excluded) and none of 499 non-fall windows, including people lying on the floor and getting up. Over 76 h of ordinary life in an occupied room, recorded at 160x120 with sound, it called none of 395 clips a fall (95% upper bound 1.2 false alarms per day) and caught 4 of the 9 staged falls the camera could see. One cautionary sentence in the prompt (“a person who goes out of view … is not a fall”) accounted for many of the misses: removing it raised sensitivity on the held-out subjects from 69% to 83% on 3.8 Flash and from 80% to 92% on Gemini 3.7 Flash, and in the home from 4 to 9 of the 9 visible falls, with no false call in ordinary living but at the price of calling deliberate lie-downs a fall. Sensitivity collapsed at 8x6 pixels (34%); identical requests disagreed on 4% of windows; and when the model called a fall, its explanation placed it within a second of the annotated landing 99% of the time. Two conventional detectors trained on the laboratory data caught nearly every held-out laboratory fall but raised 14.2 and 23.9 false alarms per day in the home. We release the frame-level fall annotations of the public dataset and every model answer on it.

[91] CausalAlign: Causal Alignment of Video Generation Models for Consistent Ego-Motion cs.CVPDF

Kai Zhu, Yintianrun Zhang, Shuwei Jiang, Ruolin Zuo, Yixin Chen

TL;DR: 论文提出 CausalAlign 因果对齐框架，用于解决扩散视频生成模型在长时生成中相机自我运动漂移的问题。方法先通过线性探针和因果干预诊断 ego-motion 在去噪过程中的表示与保留模式，再利用 delta-token 空间将对齐到 3D 基础模型，以增强运动动态表示。实验表明该方法提升运动准确性、几何一致性和时间平滑性，并改善下游具身操作任务表现。

Details

Motivation: 扩散视频生成模型虽能生成视觉合理的视频，但相机 ego-motion 在长时范围内容易漂移，破坏视角变化、几何结构和轨迹的一致性，因此需要提升视频生成中的相机运动一致性。

Result: 在 Wan2.2 上基于 DL3DV 和 RealEstate10K 微调后，方法显著提升运动精度、几何一致性和时间平滑性且保持视觉质量，几何一致性相对基座模型提升超过20%；在 RoboCasa365 移动操作任务上使 world action models 成功率提升10个百分点。

Insight: 创新点在于用线性探针和因果干预诊断 VGM 中 ego-motion 表示的“先升后降”模式，并通过 delta-token 空间将关键层表示对齐到 3D 基础模型，使对齐聚焦帧间运动动态而非静态场景内容，同时将视频生成模型的运动一致性收益迁移到具身模型中。

Abstract: Diffusion-based video generation models (VGMs) can produce visually plausible videos, yet their camera ego-motion often drifts over long horizons. Camera ego-motion determines how the viewpoint changes between frames and is essential for maintaining consistent geometry and trajectories. To address this problem, we propose CausalAlign, a causal alignment framework with two components. First, we diagnose how precise ego motion is represented, used, and retained within VGMs using linear probes and causal interventions, revealing a rise-and-fall pattern during denoising. Second, we strengthen ego-motion state in VGM by aligning the representations at the identified layers to a 3D foundation model through delta-token space. This space encodes the changes between consecutive frames, allowing the alignment to focus on ego-motion dynamics rather than static scene content. With Wan2.2 fine-tuned on DL3DV and RealEstate10K, CausalAlign can largely improve motion accuracy, geometric consistency, and temporal smoothness, while preserving visual quality. For example, it improves geometric consistency by more than 20% relative to the base model. Moreover, with the fine-tuned VGM, CausalAlign improves the success rate of world action models (WAMs) by 10 points on mobile manipulation tasks of RoboCasa365 benchmark. Overall, CausalAlign improves ego-motion consistency in diffusion-based VGMs and yields noticeable gains in embodied models.

[92] DewHand: Direct Metric World-Space Hand Motion Reconstruction from Egocentric Videos cs.CVPDF

Naru Suzuki, Satoshi Ikehata, Takehiko Ohkawa, Yoichi Sato, Shuhei Kurita

TL;DR: 论文提出DewHand，一个从第一视角视频直接联合重建世界坐标系下手部与相机运动的统一模型。它通过单阶段共享世界坐标预测、场景几何特征与度量监督、以及基于flow matching的条件运动生成，缓解多阶段误差累积、单目尺度模糊和遮挡/出视野问题。实验在HOT3D、ARCTIC及多个零样本数据集上验证了世界空间精度和时间一致性。

Details

Motivation: 从第一视角视频恢复具有真实物理尺度的世界空间手部运动，有助于理解人类交互并迁移到具身智能体。现有方法面临多阶段流水线误差传播、单目观测尺度歧义、以及遮挡和手部出视野导致多解等挑战。

Result: 在HOT3D和ARCTIC上，DewHand将刚性对齐世界空间误差相对StableHand从32.9降至28.4 mm、从76.2降至36.2 mm，并提升时间一致性；在HOT3D约三分之二片段中相机尺度误差在10%以内，而SLAM约四分之一；在HoloAssist、H2O、TACO零样本评测中均取得最低WA-MPJPE。

Insight: 创新点在于单阶段联合预测手部与相机运动并共享世界坐标，结合VGGT-Ω场景级几何特征与度量空间监督，同时保留手部细粒度特征；并将重建建模为flow matching条件运动生成以处理遮挡和出视野，仅两步采样即优于直接回归。

Abstract: Recovering metric world-space hand motion from egocentric video supports understanding human interactions and transferring them to embodied agents. Three challenges complicate this task: multi-stage pipelines propagate errors from independently estimated hand and camera motion; monocular observations leave physical scale ambiguous; and occlusion and out-of-view hands admit multiple plausible motion trajectories. We present DewHand, a unified model for direct joint reconstruction of world-space hand and camera motion. To avoid compounding errors from multi-stage estimation, DewHand jointly predicts hand and camera motion in a shared world coordinate system within a single-stage pipeline. To resolve metric-scale ambiguity, it combines scene-level geometric features from VGGT-$Ω$ with metric-space supervision, while hand-specific features preserve fine-grained articulation. To handle occlusion and out-of-view hands, we formulate reconstruction as conditional motion generation with flow matching. Under the same architecture, this generative formulation outperforms direct regression in both accuracy and temporal smoothness with only two sampling steps. Experiments on HOT3D and ARCTIC demonstrate improved world-space accuracy and temporal consistency over state-of-the-art methods. DewHand reduces rigid-aligned world-space error relative to StableHand from 32.9 to 28.4 mm on HOT3D and from 76.2 to 36.2 mm on ARCTIC. It also recovers camera scale within 10% on two thirds of HOT3D clips, versus about a quarter for SLAM. Moreover, in zero-shot evaluations on HoloAssist, H2O, and TACO, captured with different devices, DewHand achieves the lowest rigid-aligned world-space error (WA-MPJPE) on each dataset.

[93] Looking Ahead: Modeling Future Evolution in Autoregressive Video Generation cs.CVPDF

Ruopeng Gao, Yukun Huang, Jiarui Hu, Shuai Wang, Wenbo Hu

TL;DR: 论文研究自回归视频扩散模型逐块生成视频时，模型可能依赖时间冗余和局部视觉连续性，而未真正学习场景演化。作者提出未来块预测（FCP）作为与教师强制去噪联合优化的辅助目标，让模型在缺失中间视觉上下文的情况下预测时间上相隔的视频块。方法无需额外预训练视觉编码器，也不改变自回归推理流程，并能提升物理合理性、降低训练计算预算。

Details

Motivation: 自回归视频生成在教师强制训练下容易利用相邻帧的强时间冗余，把局部视觉连续性当作当前块去噪的捷径，从而削弱对场景如何演化的建模压力。论文希望显式促使模型学习未来视觉演化，而不是仅生成看似连贯的续帧。

Result: 论文报告该方法在物理合理性上优于标准教师强制自回归训练，同时整体训练计算预算更低；摘要未给出具体benchmark名称或数值指标。可视化结果还显示FCP可能促使模型形成更有结构化的表示空间。

Insight: 创新点在于引入未来块预测作为辅助监督：预测时间上相隔的块且不访问中间视觉上下文，从而迫使模型建模长期演化；监督直接来自训练视频，无需额外预训练视觉编码器，且不改变推理。客观来看，该思路以低成本辅助目标缓解教师强制下的时间冗余捷径，并为自回归世界模型学习更结构化表示提供了潜在方向。

Abstract: Autoregressive video diffusion models generate videos chunk by chunk, yet producing plausible continuations does not guarantee that they capture how scenes evolve. Under teacher forcing, strong temporal redundancy may allow local visual continuity to serve as a shortcut for current-chunk denoising, reducing the pressure to model visual evolution. We introduce Future Chunk Prediction (FCP), an auxiliary objective jointly optimized with teacher-forced denoising that predicts temporally separated chunks without access to the intervening visual context. Drawing supervision directly from training videos, FCP requires no additional pretrained visual encoder and leaves autoregressive inference unchanged. Our method achieves superior physical plausibility over standard teacher-forced autoregressive training with a lower overall training compute budget, highlighting the effectiveness and efficiency of future-chunk supervision for modeling visual evolution. Intriguingly, our visualizations suggest that FCP fosters a more structured representation space, which may be valuable for building autoregressive world models.

[94] ClunoGS: Compact Language 3D Gaussian Splatting via Unified Optimization cs.CVPDF

Jian Xu, Yi Li, Ying Huang, Siyi Wu, Bingzhe Li

TL;DR: 论文提出 ClunoGS，一个面向语言3D高斯泼溅（Language 3DGS）的紧凑压缩框架，将语义感知的高斯稀疏化与向量量化统一到一个受约束的优化目标中，该目标同时考虑渲染质量、语义准确性和模型紧凑性，并通过交替更新原始变量、稀疏化、量化和对偶变量进行优化。方法用于开放词汇场景理解中的定位与分割任务。

Details

Motivation: 语言3DGS需要为高斯基元关联语言特征，显著增加模型存储；现有3DGS压缩方法主要面向保持RGB渲染质量。若简单串联剪枝和量化，仅依据渲染贡献剪枝可能删除RGB贡献小但对语义预测重要的高斯，且其语义信息在后续量化中无法恢复。

Result: 论文在LERF和ScanNet-10基准上评估开放词汇定位与分割；在激进压缩下，所提出的统一且语义感知的公式显著缓解了顺序压缩和语义无关高斯稀疏化造成的语义表示退化。消融实验进一步表明，联合协调稀疏化与量化比独立应用更能保持语义准确性。

Insight: 摘要宣称的创新点在于将语义感知稀疏化与向量量化纳入统一约束目标，并在稀疏化过程中引入语义监督，联合优化渲染质量、语义准确性和紧凑性。客观来看，其可借鉴之处是避免仅按RGB重要性剪枝，而是显式考虑语义重要性以及稀疏化与量化之间的耦合关系。

Abstract: Language 3D Gaussian Splatting enables open-vocabulary scene understanding by associating Gaussian primitives with language features. These semantic attributes increase model storage, while most existing 3DGS compression methods are designed mainly to preserve RGB rendering quality. One straightforward approach is to apply sparsification and quantization sequentially. However, pruning based only on rendering quality can remove Gaussians that have small RGB contributions but are important for semantic prediction. Once removed, their semantic information cannot be recovered by later quantization. We propose ClunoGS, a compact framework for Language 3DGS that combines semantics-aware Gaussian sparsification and vector quantization in a unified constrained objective. Semantic supervision is used during sparsification, and the objective jointly considers rendering quality, semantic accuracy, and compactness. We optimize the unified objective with alternating primal, sparsification, quantization, and dual updates. We evaluate ClunoGS on open-vocabulary localization and segmentation using the LERF and ScanNet-10 benchmarks. Under aggressive compression, our unified, semantics-aware formulation substantially alleviates the degradation of semantic representations caused by sequential compression and semantics-agnostic Gaussian sparsification. Ablations further show that jointly coordinating sparsification and quantization better preserves semantic accuracy than applying them independently.

[95] RealMe: Immersive Volumetric Video from Minimal Captures cs.CVPDF

Chenxu Zhou, Zhiyuan Yu, Yudong Jin, Yinji Shentu, Yu Zhang

TL;DR: 论文提出RealMe，一个面向稀疏、低分辨率多视图捕捉的沉浸式体积视频重建框架。它以高分辨率视频扩散模型为核心，将有限捕捉增强为2K、4D一致的视频，再用于细化体积视频的4D表示。方法结合布局感知位置嵌入、适配多视图布局的分解时空注意力和潜在频域损失，并在公开与自采数据集上验证了视觉保真度和4D一致性。

Details

Motivation: 传统密集多视图重建质量高但采集成本昂贵，近期生成模型虽放宽采集要求，仍受限于分辨率和4D不一致，难以在真实场景中实用。论文旨在从稀疏、低分辨率捕捉资源生成高质量沉浸式体积视频。

Result: 在公开DNA-Rendering基准和自采RealVV数据集上，RealMe能够创建沉浸式体积视频，并在视觉保真度和4D一致性上优于现有最先进方法。

Insight: 摘要宣称的创新点是用高分辨率视频扩散模型统一增强有限捕捉并服务体积重建，适配多视图捕捉布局的分解时空注意力，以及潜在空间频域监督。客观看，将布局感知注意力与频域损失结合以提升生成视频与真实捕捉的一致性和4D一致性，是较有借鉴价值的设计。

Abstract: This paper addresses the problem of creating immersive volumetric videos from sparse and low-resolution capture resources. Conventional dense multi-view reconstruction methods can produce high-quality volumetric videos, but their reliance on costly capture setups makes them impractical in real-world scenarios. Recent approaches use generative models to relax capture requirements, but still suffer from limited resolution and 4D inconsistency. To overcome these constraints, we propose RealMe, a unified framework built around a high-resolution video diffusion model that enhances limited captures into high-resolution (2K), 4D-consistent videos for volumetric video reconstruction. Specifically, RealMe constructs auxiliary data from capture videos and encodes them into latents along with layout-aware positional embeddings. The video diffusion model adapts factorized spatio-temporal attention to the multi-view capture layout, denoises these latents across viewpoints and time, and decodes them into enhanced videos to refine the 4D representation. This adaptation helps the model capture temporal dynamics and multi-view relationships, leading to more 4D-consistent generation. To reduce the data gap between generated videos and real captures, we apply a latent frequency loss that provides frequency-domain supervision in latent space. Extensive experiments on the public DNA-Rendering benchmark and our self-captured RealVV dataset demonstrate that RealMe is able to create immersive volumetric videos and outperforms state-of-the-art methods in visual fidelity and 4D consistency.

[96] Attn-Forcing: Learning Full-Context Attention Proxies for Streaming Video Generation cs.CVPDF

Junyao Gao, Sibo Liu, Jiaxing Li, Yanan Sun, Hanxin Zhu

TL;DR: 论文提出 Attn-Forcing 框架，用于在有限内存和计算开销下为流式视频生成保留全上下文注意力贡献。该方法引入可学习的注意力代理来持续近似历史上下文引发的注意力响应，并通过代理更新模块和注意力匹配损失进行训练。实验表明其能稳定提升流式视频生成质量，并优于现有基线方法。

Details

Motivation: 流式视频生成中增大注意力上下文长度能提升质量，但会带来难以承受的内存和计算成本；现有注意力选择或压缩方法受限于有界表示，且缺乏显式的注意力级监督来保留历史上下文对当前查询的贡献。

Result: 大量定性与定量实验表明，Attn-Forcing 能持续提升流式视频生成质量，并在现有基线方法上达到 state-of-the-art 性能。

Insight: 创新点在于用可学习视觉潜变量构建注意力代理，持续近似全上下文注意力响应，并通过代理更新模块和注意力匹配损失显式约束注意力行为，从而在有界内存与计算下保留全上下文注意力贡献。

Abstract: Increasing the attention context length significantly improves streaming video generation quality, while incurring prohibitive memory and computational costs as the context grows. Existing methods attempt to capture this contextual information through attention selection or compression, but remain limited by bounded representations and lack explicit attention-level supervision to retain historical contributions to current queries. To address this, we introduce Attn-Forcing, a novel framework that preserves full-context attention contributions for streaming video generation under bounded memory and computation. We achieve this by introducing attention proxies that continuously approximate the attention responses induced by the historical context. Specifically, we implement the proxies as learnable visual latents and employ a Proxy Update Module to incorporate visual context from historical latents. Moreover, we propose an Attention Matching Loss to explicitly constrain the attention behavior of the learned proxies using full-context attention supervision. Extensive qualitative and quantitative experiments demonstrate that Attn-Forcing consistently improves streaming video generation quality and achieves state-of-the-art performance over existing baseline methods.

[97] ProactEgoMem: Benchmarking Proactive Memory in Streaming Egocentric Video cs.CVPDF

Ruotong Liao, Meikel Kokowski, Xiaoxiao Sun, Yongqin Xian, Laura Bravo Sánchez

TL;DR: 论文提出 ProactEgoMem，一个用于评估流式第一视角视频中主动记忆能力的基准。该基准围绕信息捕获、物体状态跟踪和位置跟踪三类任务，设置通知级与干预级两种主动性协议，并提供 Daily/Sci 两个赛道及配套 QA，用于衡量智能体是否能主动、及时且准确地获取并维护记忆。

Details

Motivation: 现有多模态助手通常被动解释当前视觉输入，而可穿戴助手需要代表用户主动识别重要信息、实时理解场景并在后台维护记忆，以减轻长时程记忆负担。论文旨在填补流式第一视角视频中主动记忆评估的空白。

Result: ProactEgoMem-Daily 和 ProactEgoMem-Sci 分别包含 300/100 个视频片段，ProactEgoMem-QA 包含 1000/500 个问题。对专有模型、开源模型、视频语言模型和流式基线进行评估后，即使最强模型也难以同时做到及时且准确地捕获记忆。

Insight: 创新点在于将主动记忆维护形式化为流式第一视角视频中的基准任务，并提出通知级和干预级两层主动性协议，以及覆盖信息捕获、状态跟踪和位置跟踪的评测体系。其可借鉴之处是强调智能体从被动问答转向主动启动、维护和停止记忆工具的生命周期。

Abstract: Today’s multimodal assistants are typically prompted to interpret the visual evidence presented to them. To offload the cognitive burden of remembering across long horizons, wearable assistants should instead proactively acquire and maintain useful memory as the user’s surroundings evolve. Doing so on the user’s behalf requires identifying what matters, understanding the scene in real time, and initiating memory maintenance in the background while the user stays focused on their activities. We introduce ProactEgoMem, a benchmark that evaluates proactive memory in streaming egocentric video. Three tasks ground the evaluation in ambient environmental needs: Information Capture (IC) preserves transient yet dense visual information, State Tracking (ST) follows changing object states, and Location Tracking (LT) records object movements. The benchmark defines its protocol at two levels of proactiveness: the Notification Level asks for a single timely notification that satisfies the user’s intent, while the deeper Intervention Level asks the agent to maintain memory on the user’s behalf through an active start-to-stop tool lifecycle. Parallel ProactEgoMem-Daily and ProactEgoMem-Sci tracks contain 300/100 clips covering daily and scientific use cases, paired with 1000/500 questions in ProactEgoMem-QA, which measures the quality of the proactively acquired memory. Across proprietary, open-source, video-language, and streaming baselines, even the strongest models fail to capture memory both timely and accurately. ProactEgoMem provides a rigorous evaluation foundation for proactive multimodal assistance and for video agents that co-exist with their users.

[98] VoxProg: 3D Spatial Reasoning through Voxel Programming cs.CVPDF

Zhichao Zhai, Menglei Zhao, Guikun Chen, Zihao Yan, Kanzhi Wu

TL;DR: 论文提出 VoxProg，将 3D 物体放置问题建模为体素编程：语言模型把空间任务解释为可执行程序，在场景表面的局部体素视图上选择任务相关区域与边界，并由下游位姿程序测量几何并计算目标 SE(3) 位姿；执行得到的视觉与数值反馈用于迭代修正代码。论文还构建了 71 个任务的 VoxProg-Bench 及 Intent Alignment 与 3D Physical Validity 评测。

Details

Motivation: 语言模型具备理解任务和生成程序的能力，但在 3D 场景中遵循空间指令时，程序需要构造语言指令中隐含的几何参考；论文旨在解决开放端 3D 物体放置中的空间推理与几何位姿计算问题。

Result: 在 VoxProg-Bench 和外部基准上，VoxProg 一致优于基于关系、基于关键点和基于网格的基线方法；摘要未给出具体数值。

Insight: 创新点在于将 3D 空间推理转化为体素编程，利用规则网格与显式坐标/邻域关系从场景网格中提取几何证据，并通过证据程序与位姿程序分离、执行反馈迭代修码来形成可检验的推理闭环；同时提出新基准与感知意图对齐、3D 物理有效性评测。

Abstract: Language models’ abilities to interpret tasks and generate programs offer a promising basis for 3D spatial reasoning. Yet following spatial instructions in 3D scenes requires programs to construct geometric references that language leaves implicit. In this work, we introduce VoxProg, which formulates 3D object placement through voxel programming: a language model translates task interpretations into executable programs over local voxel views of scene surfaces. Because these views index surface geometry on regular grids with explicit coordinates and neighborhood relations, an evidence program can isolate task-relevant regions and boundaries directly from scene meshes, and a downstream pose program can measure the resulting geometry to compute a target $SE(3)$ pose. Visual and numerical feedback from execution lets the model inspect intermediate selections and candidate placements to iteratively revise its code. To evaluate open-ended placement across functional use, geometric fit, and contextual arrangements, we introduce the 71-task VoxProg-Bench benchmark along with an evaluation suite pairing perceptual Intent Alignment and 3D Physical Validity. Across VoxProg-Bench and external benchmarks, VoxProg consistently outperforms relation-based, keypoint-based, and mesh-based baselines.

[99] Rethinking Video Autoencoders for Interactive World Models cs.CVPDF

Adrien Ramanana Rahary, Yutong Bai, Nicolas Dufour, Patrick Pérez, David Picard

TL;DR: 论文重新审视交互式视频世界模型中的视频自编码器压缩问题，提出开放族CADRE：一组覆盖多时空压缩率的因果扩散表示自编码器，并训练同一世界模型在机器人操作与导航任务上比较。研究发现CADRE在延迟-质量权衡上表现突出，尤其移动相机场景，并总结出扩散解码器、时间/空间压缩及任务差异等方面的四点结论。

Details

Motivation: 现有视频世界模型常直接复用文本到视频生成器的自编码器，但后者面向离线生成，未考虑智能体等待和交互延迟，且往往因缺少训练数据与代码无法按其他压缩率重训。论文因此系统研究世界模型应压缩什么、在哪里压缩以及压缩多少。

Result: CADRE在操作和导航两个领域、各两个环境中形成大部分延迟-质量前沿，尤其在移动相机时表现突出。研究还给出四点定性发现：强压缩下扩散解码器比感知回归更清晰；机器人操作中时间压缩更适合放在自编码器而非世界模型；操作与导航需要相反的压缩策略；静止相机下粗空间token对运动质量的损害远大于对图像质量的损害。

Insight: 创新点在于构建可跨多种时空压缩率比较的开放自编码器族，仅通过冻结特征上的小瓶颈区分不同压缩率，并共享自蒸馏到少步的扩散解码器，从而系统评估压缩对交互式世界模型延迟与质量的影响。其可借鉴之处是面向交互延迟重新设计世界模型压缩，并揭示不同任务和相机运动条件下时空压缩策略的差异化需求。

Abstract: Video world models predict future frames from past frames and actions. They serve as simulators in which agents train, plan or are evaluated, and that humans can control interactively. They generate in the latent space of a video autoencoder, whose compression sets their cost per frame and how long each action waits to take effect. Yet world models often reuse the autoencoders of text-to-video generators, built for offline generation with no agent waiting and, without their training data and code, not retrainable at another compression. We therefore ask how, where and how much a world model should compress. We build CADRE, an open family of causal diffusion representation autoencoders spanning multiple spatio-temporal compressions. Its members differ only in a small bottleneck on frozen features and share a diffusion decoder self-distilled to one or a few steps. We train the same world model on each autoencoder in two domains, manipulation and navigation, with two environments each. CADRE forms most of the latency-quality frontier, especially with a moving camera. Our study also leads to four findings. At strong compression, a diffusion decoder is sharper than perceptual regression. In robot manipulation, temporal compression is more effective in the autoencoder than in the world model. Manipulation and navigation call for opposite compressions: temporal compression costs little when the camera is still, and spatial compression when it moves. Surprisingly, with a still camera, coarser spatial tokens hurt motion far more than image quality. Code, models and a live demo are at https://kyutai.org/blog/2026-10-21-cadre

cs.AI [Back]

Jaesang Yu, Seoyeon Byun, Yun Wang, Yan Lu, Juho Kim

TL;DR: 论文提出 Draft Refinement Language Models（DRLM），用单个去噪器并行生成候选修订，并通过一次前向计算得到的 cavity energy 评估完整草稿，只有能量降低时才保留修订。方法在 Countdown、Sudoku-Extreme 和 Maze-Hard 上超过自回归与离散扩散基线，展示了完整草稿并行精炼可作为复杂推理中反向扩散的替代方案。

Details

Motivation: 重复扩散去噪不能保证每个局部合理的 token 更新组合起来会得到更好的答案；论文希望让推理模型在完整草稿上下文中并行重审和修正决策。

Result: 在 Countdown、Sudoku-Extreme、Maze-Hard 上优于自回归和离散扩散基线；仅 6M 参数时，DRLM 在 Countdown-5 达到 44.4%，在 Sudoku-Extreme 达到 93.9%。

Insight: 创新点在于用单个去噪器从均匀损坏草稿中学习保留与修复，直接指定修订而无需扩散反向转移；并利用已知损坏过程校正自 token 证据，以一次前向传播聚合出 cavity energy，无需额外评分模型，按能量下降接受候选草稿。

Abstract: Drafting an answer allows a reasoning model to reconsider each decision in the context of the others. Yet repeated diffusion denoising does not ensure that individually plausible token updates form a better answer together. We introduce Draft Refinement Language Models (DRLMs), which use a single denoiser to propose parallel revisions and evaluate a candidate draft. We train the denoiser to reconstruct every token from uniformly corrupted drafts, teaching both preservation and repair. Its predictions specify revisions directly, without a diffusion reverse transition. Evaluation requires a different use of these predictions: confidence in an existing token reflects both contextual support and evidence from observing the token itself. Using the known corruption process, we correct for this self-token evidence to estimate how well each token is supported by the prompt and the remaining draft. Aggregating these estimates yields a cavity energy computed in one forward pass, without an additional scoring model. Each round proposes a revision, reads the completed candidate in its revised context, and retains it only if its energy is lower. Across Countdown, Sudoku-Extreme, and Maze-Hard, our method outperforms autoregressive and discrete diffusion baselines. With 6M parameters, DRLM reaches 44.4% on Countdown-5 and 93.9% on Sudoku-Extreme. The results establish complete-draft parallel refinement as an effective alternative to reverse diffusion for complex reasoning.

[101] Same Answer, Different Reasons: Trace-Level Divergence in Multilingual Large Language Models cs.AI | cs.CLPDF

Rui Sun, Peizhao Mei, Xiewei Cheng, Kaizhuo Chen, Taoran Li

TL;DR: 论文提出轨迹级分歧框架，将多语言LLM的可见推理轨迹作为结构化序列比较，并分解为结构相似性、顺序一致性和整体相似性，辅以配对特定校准；在3个推理基准、7种语言、6个前沿LLM上揭示：仅看整体响应相似度会系统性高估跨语言推理轨迹一致性，且人类判断更符合结构相似性，因而主张采用轨迹级评估。

Details

Motivation: 标准多语言LLM评估通常只衡量最终答案一致性或整体响应相似度，无法判断不同语言用户是否获得可比的支撑证据；论文关注“答案相同但理由不同”的跨语言推理轨迹分歧风险。

Result: 在3个推理基准、7种语言、6个前沿LLM上，整体相似度最高四分位的样本中有52.1%–67.3%并未进入结构相似度最高四分位；该分离跨句子编码器和步骤分割稳定，盲双语标注与结构相似性更一致，同语言控制显示分歧水平具有语言特异性。

Insight: 创新点在于将多语言一致性评估从答案级/整体嵌入推进到可见推理轨迹的结构化序列层面，并提出结构相似性、顺序一致性与整体相似性的分解及跨语言配对校准；客观看，其价值在于用人类标注和同语言控制验证了轨迹级评估对可靠多语言部署的必要性。

Abstract: Multilingual large language models are increasingly deployed in settings where visible reasoning traces support verification, education, and safety auditing, yet standard evaluations typically measure only final-answer agreement or holistic response similarity. As a result, they establish outcome consistency without determining whether users across languages receive comparable supporting evidence. We introduce a trace-level divergence framework that compares visible reasoning traces as structured sequences, decomposing agreement into structural similarity, order consistency, and holistic similarity, with pair-specific calibration to enable comparison across languages. Across three reasoning benchmarks, seven languages, and six frontier LLMs, the natural baseline—embedding each whole response and comparing the two by cosine—systematically overstates agreement: $52.1%$–$67.3%$ of examples in the top quartile of holistic similarity are not in the top quartile of structural similarity. This dissociation is stable across sentence encoders and step segmentations, and blind bilingual annotation studies in two English-anchored language pairs show that human judgments align with structural similarity rather than with the holistic baseline. A same-language control shows that the level of divergence is language-specific: two English traces for paraphrases of the same question agree more closely than an English trace and its counterparts in other languages. Our results show that answer-level and holistic evaluations can substantially overestimate the consistency of multilingual explanatory artifacts, motivating trace-level evaluation for reliable multilingual deployment.

[102] GRAM: Group-wise Rank-Aware Model Merging via Subspace Alignment cs.AI | cs.CLPDF

Di Wu, Shuaidong Pan

TL;DR: 论文提出基于子空间对齐的组级秩感知模型合并方法（GRAM），用于多任务模型合并。该方法利用任务向量的主角度几何关系对任务进行分组，在组内执行SVD合并，从而缓解子空间干扰问题。仅需微调权重，在语言理解、视觉和医疗指令微调任务上优于现有方法。

Details

Motivation: 现有模型合并方法对所有任务均匀合并，忽略了任务子空间的几何关系。当任务更新子空间未对齐时，投影到共享低秩基会丢失任务相关方向，导致子空间干扰。论文旨在通过分组几何兼容的任务来减少投影误差。

Result: GRAM在仅增加一个模型（K=2）的情况下，显著减轻子空间干扰，并在GLUE（Flan-T5）、8任务CLIP视觉和医疗指令微调（Qwen2-7B-Instruct与四个中英文医疗LoRA）上持续优于先前合并方法。

Insight: 创新点在于利用主角度几何进行任务分组，理论证明分组可减少投影误差，并采用组内SVD合并。客观分析认为将子空间对齐思想引入模型合并，通过分组策略有效缓解子空间干扰，具有创新性。

Abstract: Merging different models into a single model is highly desirable for multi-task deployment, particularly in regulated domains like healthcare. In such settings, models are often locally fine-tuned due to strict data privacy policies, necessitating that aggregation operates exclusively on the fine-tuned weights. Existing methods apply a uniform merge across all tasks, ignoring the geometric relationships among their learned subspaces. When task updates occupy misaligned subspaces, projecting them onto a shared low-rank basis discards task-relevant directions, a failure mode named \emph{subspace interference}. Through this study, we show that the per-task projection error of a merged model is governed by the principal-angle alignment between each task and the shared subspace, so that grouping geometrically compatible tasks provably reduces this error within each group. Motivated by this observation, we propose \textbf{Group-wise Rank-Aware Model Merging (GRAM)}, which uses the principal-angle geometry of task vectors to decide which tasks should be merged together and then performs a standard SVD-based merge within each coalition; the pipeline requires only the fine-tuned weights at merge time. With only one additional model ($\nK{=}2$), GRAM significantly mitigates subspace interference and consistently outperforms prior merging methods across language understanding (GLUE with Flan-T5), 8-task CLIP vision, and medical instruction tuning (Qwen2-7B-Instruct with four English/Chinese medical LoRAs).

[103] TeachBack: Learning to Teach via Bidirectional Preference Optimization for On-Policy Self-Distillation cs.AI | cs.CL | cs.LGPDF

Linfei Ji, Jiayun Li, Haozhao Wang, Siyu Jiang, Zehao Wang

TL;DR: 论文研究on-policy self-distillation中教师拥有参考解等特权信息时，虽然能生成高准确率轨迹，但未必适合学生学习的问题，提出TeachBack双向偏好优化框架。该方法通过控制参考推理解码后缀长度和提示风格，塑造教师的教学行为，并利用特权教师轨迹与学生轨迹之间的双向偏好目标，同时改进学生学习和未来教师监督。

Details

Motivation: 现有OPSD中教师可直接访问完整参考解，容易生成缩短的、参考条件化的推导，省略探索、验证和自纠过程，导致高准确率轨迹不一定适合无特权信息的学生学习。

Result: 在数学和逻辑推理任务上，TeachBack持续优于现有OPSD方法及相关基线；摘要未给出具体数值或明确benchmark名称。

Insight: 创新点在于将“教师如何教”纳入自蒸馏优化：用双向偏好学习让学生从教师学习，同时用学生行为反向塑造教师监督；并通过控制参考解后缀长度和提示风格，鼓励教师探索重解问题、降低对特权信息的依赖。客观看，其价值在于缓解特权信息导致的捷径学习和推导过程不适配问题。

Abstract: On-policy self-distillation (OPSD) enables a language model to learn from a Teacher instantiated from the same model but given access to privileged information, such as a reference solution. We find that access to sufficient privileged information allows the Teacher to generate highly accurate trajectories; however, accuracy alone does not guarantee that these trajectories are suitable for Student learning. Direct access to a complete solution may encourage shortened, reference-conditioned derivations that omit the exploration, verification, and self-correction required to solve problems without privileged information. We introduce TeachBack, a bidirectional preference-learning framework for on-policy self-distillation that learns not only from the Teacher but also how the Teacher should teach. TeachBack shapes the Teacher’s generation behavior by controlling the length of the reference-reasoning suffix injected into the prompt and by varying the prompt style. While maintaining comparable Teacher accuracy, it encourages the Teacher to explore and re-solve problems, thereby reducing its reliance on the reference solution. At each rollout refresh, the shared policy generates both a privileged Teacher trajectory and a problem-only Student trajectory. A pairwise preference objective uses privileged Teacher trajectories to improve Student learning, while the reversed preference uses the Student’s behavior to shape future Teacher supervision under the privileged condition. Experiments on mathematical and logical reasoning tasks show that TeachBack consistently improves Student performance over existing OPSD methods and related baselines.

[104] Real PMIDs, Wrong Papers: Harness Design for Biomedical Claim-Verification Agents cs.AI | cs.CL | cs.IR | cs.LGPDF

Mohammed Saqlain

TL;DR: 论文在生物医学声明验证任务SciFact上，以实时PubMed API为唯一工具，对文献检索智能体的不同harness（闭卷、单次检索、ReAct自写查询、给定金标摘要）做受控消融，并审计引用PMID相关性。核心发现是检索是主要瓶颈，让模型自行构造查询能提升金标召回和判定准确率，而闭卷模型给出的PMID虽真实却大多不相关。

Details

Motivation: 生命科学科研助手常集成文献检索语言智能体，但其harness中哪些组件真正重要缺乏受控证据；论文要解决生物医学声明验证中检索与推理瓶颈及引用可靠性问题。

Result: 在SciFact上，单次检索仅6%找到金标论文；自写查询将金标召回提升三倍并取得62%最佳判定准确率（其他现实harness为49-59%，样本量下不显著）；给定金标摘要可达79-86%，显著高17-24点。引用审计显示闭卷模型327/335个PMID存在但仅3个被LLM评审判为相关，检索智能体为67%。

Insight: 创新在于对检索智能体harness做控制变量消融，量化查询构造与检索质量对最终判定的影响，并揭示“PMID真实≠引用相关”，提出存在性检查不足以保障科学智能体。

Abstract: Literature-searching language agents are becoming a standard component of scientific assistants for the life sciences, yet there is little controlled evidence on which parts of their harness matter. We run a controlled ablation on biomedical claim verification (SciFact) with the live PubMed API as the only tool. Holding the backbone and output format fixed, we compare four harnesses: closed-book answering, single-shot retrieval using the claim as the query, a ReAct-style agent that writes its own PubMed queries, and an oracle given the gold abstracts, each with two efficient frontier backbones. Naive single-shot retrieval finds the gold paper for only 6% of claims and turns overconfident errors into abstentions without improving accuracy. Letting the model formulate its own queries triples gold recall and yields the best verdicts (62% accuracy vs. 49-59% for the other realistic harnesses), although these differences are not significant at our sample size and agents use fewer than half of their search budget. With gold abstracts, the same models reach 79-86% accuracy, a significant 17-24 point gap showing that retrieval, not reasoning, is the main bottleneck. Finally, a citation audit shows that closed-book models cite PubMed IDs that almost always exist (327/335) but are almost never relevant: a cross-family LLM judge rates only 3 of 327 existing citations as relevant to the claim, versus 67% for citations produced by the search agent. Existence checks are therefore insufficient safeguards for scientific agents. We release code, prompts, and all cached tool calls at https://github.com/saqlain2204/real-pmids-wrong-papers.

[105] ToM2Vec: Continuous Theory-of-Mind Tokens for Multi-Agent Coordination cs.AI | cs.CL | cs.MAPDF

Nithin Sivakumaran, Ali Payani, Hyunji Lee, Mohit Bansal, Elias Stengel-Eskin

TL;DR: 论文提出 ToM2Vec，用固定预算的连续潜在推理 token 来学习多智能体中的心智理论（ToM）。训练时用伙伴真实想法监督这些 ToM token，使其抽象并匹配伙伴状态；测试时由智能体生成 ToM token，并以其为条件引导自身思考与行动。方法在多个协作、竞争和混合动机环境中验证有效。

Details

Motivation: 多智能体协作需要准确的心智理论，并在不确定性下推理伙伴的多种可能状态；但 LLM 智能体原生的文本表示空间更容易描述单一伙伴状态，难以表达这种不确定性。

Result: 在 Overcooked-AI、多方谈判、Decrypto、Hanabi、Collab Escape/Capture 等环境中，ToM2Vec 相比基座模型及多种训练/提示基线取得一致提升；文本化推理会显著掉点，说明潜在推理关键；方法还能泛化到训练时未见过的伙伴模型家族（含 GPT-5.4），并提升对伙伴信念、欲望和下一步行动的直接问答能力。

Insight: 创新点在于用固定预算的连续潜在 token 建模心智理论，并以伙伴真实想法作为监督信号；测试时 ToM token 作为条件器影响智能体推理与行动，分析表明其编码的是 ToM 而非任务特定信息。

Abstract: Forming an accurate theory-of-mind (ToM) is key to successful multi-agent collaboration and requires reasoning under uncertainty to capture multiple possible partner states. This is hard to do in the text representation space native to large language model (LLM) agents, which more easily describe a single partner state. To address this challenge, we introduce ToM2Vec, which learns to form abstract theories-of-mind in a rich latent space via a fixed budget of continuous reasoning tokens. During training, supervision for these tokens comes from the partner’s actual thoughts, which the ToM tokens are trained to abstract and match. At test-time, ToM tokens are generated by each agent and act as conditioners for that agent’s thoughts and actions. Across several environments covering collaborative, competitive, and mixed-motive settings (Overcooked-AI, multiparty negotiation, Decrypto, Hanabi, and Collab Escape/Capture), we evaluate ToM2Vec’s ability to support multi-agent interactions and collaboration, finding consistent improvements over the base model and a variety of competitive trained and prompted baselines. We show that latent reasoning is critical, with performance dropping when reasoning is expressed in text. Moreover, we show that ToM2Vec training generalizes to partner model families not seen at train-time, including a proprietary model (GPT-5.4). In our analysis, we provide evidence that the latent tokens encode a ToM, rather than task-specific information: ToM2Vec’s tokens improve the agent’s ability to directly answer questions about partner beliefs, desires, and next actions.

cs.IR [Back]

[106] Reasoning-Search-Augmented LLMs: An Operational Taxonomy, Benchmark Analysis and Controlled Evidence Synthesis cs.IR | cs.CLPDF

Biswas Poudel, Nilson Chapagain, Amit Kumar, Xianshun Jiang

TL;DR: 论文综述并分析了将逐步推理与交互式搜索紧密耦合的推理-搜索增强大语言模型（RSA-LLMs），把检索视为迭代智能体循环中的动作而非一次性预处理。作者从检索耦合、控制器设计、知识来源和 grounding 严格性四个架构维度建立操作分类法，并将学习信号作为横切维度单独讨论；同时区分基准难度与 RSA 循环需求，综合受控的 within-study 消融和质量-成本差异，避免跨异构模型、检索器、语料和评测协议的不当汇总。

Details

Motivation: LLM 受固定参数和有限上下文限制，即使使用静态检索也会产生无证据支持或与证据矛盾的陈述，导致幻觉、难以及时更新知识以及无法正确解释和协调来源。论文旨在厘清通过迭代智能体循环将推理与交互式搜索耦合的系统版图，并给出影响未来研究的设计选择统一视角。

Result: 论文不提出新的具体模型，也不是在单一 benchmark 上刷新 SOTA；其结果是建立 RSA-LLM 的操作分类法、基准需求分析和受控证据综合，并通过 within-study 消融与质量-成本差异总结不同设计的影响。

Insight: 可借鉴之处在于用四轴架构分类法加学习信号横切维度系统刻画推理-搜索增强 LLM，并强调区分 benchmark 难度与 RSA 循环实际需求，采用受控 within-study 证据综合来避免跨异构设置无效 pooling。

Abstract: Large language models (LLMs) demonstrate impressive emergent reasoning capabilities, yet they remain constrained by fixed model parameters and finite context windows. Even when augmented with static retrieval, they can produce statements that are unsupported by, or even contradictory to, retrieved evidence. This leads to hallucinations, difficulty incorporating up-to-date information, and failures to correctly interpret or reconcile sources. Recent work addresses these limitations by tightly coupling step-by-step reasoning with interactive search, treating retrieval as an action within an iterative agentic loop rather than a one-shot preprocessing step. This study reviews frameworks that implement such reasoning-search loops by interleaving planning, querying, retrieval, integration, verification, and stopping. We analyze these systems through four architectural axes: retrieval coupling, controller design, knowledge source, and grounding strictness. We separately treat the learning signal as a cross-cutting dimension because prompting, supervision, preference optimization, and reinforcement learning can instantiate the same architecture while producing materially different search, stopping, and grounding behavior. We further distinguish benchmark difficulty from RSA-loop demand and synthesize controlled within-study ablations and quality-cost deltas, avoiding invalid pooling across heterogeneous model backbones, retrievers, corpora, and evaluation protocols. Our goal is to clarify the landscape of reasoning-search-augmented LLMs and outline a unified perspective on the design choices shaping future research.

[107] pplx-embed-v2-late Technical Report cs.IR | cs.CL | cs.LGPDF

Markus Krimmel, Louis Milliken, Ekaterina Glazkova, Maximilian Schall, Antoine Chaffin

TL;DR: 论文提出 pplx-embed-v2-late，包含 0.6B 和 9B 两个多语言、多模态 late-interaction 检索模型，覆盖文本、图像和视觉文档检索任务。两个模型通过从 18B 教师模型进行表示蒸馏训练，并在各自参数规模类别中取得 SOTA。二者共享嵌入空间，支持用 9B 模型建索引、用低延迟 0.6B 模型查询的非对称部署。

Details

Motivation: 面向多语言与多模态检索，尤其是视觉文档检索场景，在保证检索性能的同时降低查询延迟和部署成本，使小模型能够接近或达到更大模型的效果。

Result: 0.6B 和 9B 模型在各自尺寸类别的文本、图像和视觉文档检索基准上达到 SOTA；其中 0.6B 在视觉文档检索任务上可与数倍规模更大的模型竞争，并支持 9B 建索引、0.6B 查询的非对称部署。

Insight: 核心创新在于利用 18B 教师模型的表示蒸馏训练一对共享嵌入空间的多语言多模态 late-interaction 模型，从而实现大小模型协同的非对称检索部署，兼顾索引质量与查询效率。

Abstract: In this report, we introduce pplx-embed-v2-late, a pair of 0.6B and 9B multilingual and multimodal late-interaction models. Both models achieve state-of-the-art results within their respective size classes across a range of text, image, and visual document retrieval benchmarks. On visual document retrieval tasks, pplx-embed-v2-late-0.6b is competitive with models several times larger than itself. The models are trained via representation distillation from an 18B teacher. Consequently, they share an embedding space that supports asymmetric deployment: documents can be indexed with the 9B model and queried with the low-latency 0.6B model. Both models are available at https://huggingface.co/collections/perplexity-ai/pplx-embed-v2.

[108] Think Harder Where It Matters: Item-Level Adaptive Recursion for Sequential Recommendation cs.IR | cs.AI | cs.CL | cs.LGPDF

Zhanshuo Cao, Jiakai Tang, Yuqi Zhou, Sunhao Dai, Jun Xu

TL;DR: 论文针对序列推荐中现有模型对所有物品分配固定计算预算的问题，通过诊断发现物品计算需求不均且固定分配次优，提出LoopRec方法。LoopRec递归应用权重共享块，并在每个递归边界引入轻量级门控为每个物品独立决定退出或继续，实现物品级自适应计算。在四个骨干网络上实验表明一致优于固定深度方法。

Details

Motivation: 现有序列推荐模型为用户历史中的每个物品分配固定计算预算，忽略了不同物品对最终预测的贡献差异。作者通过诊断发现同一序列中物品的计算需求差异很大，且固定深度分配不是最优的，因此希望实现物品级的自适应推理。

Result: 在公开基准上与四种骨干家族结合，LoopRec一致优于匹配的固定深度对应方法。分析表明其学习到的计算分配与后验oracle分析一致：低频物品和近期交互获得更深计算。

Insight: 创新点在于提出LoopRec，通过递归权重共享块和轻量级门控实现每个物品的退出或继续决策，将固定全局深度转化为可学习的每物品属性，无需辅助深度标签，且易于适配不同骨干架构。客观分析认为，该工作通过诊断实验验证了自适应计算分配的必要性，且门控设计简单有效。

Abstract: Existing sequential recommenders usually allocate a fixed computational budget to every item in a user’s interaction history, regardless of the item’s contribution to the final prediction. We challenge this de facto design with two diagnostic findings. (1) Non-uniform computational need: items within the same sequence vary drastically in how much computation they require. Most converge after shallow processing while a critical minority only stabilizes at full depth. (2) Suboptimality of fixed allocation: given oracle access to the target item, a non-uniform depth assignment consistently outranks the fixed one, indicating that existing uniform computation both wastes capacity on already-converged items and sacrifices overall recommendation quality. These observations motivate a shift from fixed-depth computation to item-level adaptive reasoning. To this end, we propose LoopRec. LoopRec recursively applies a single weight-shared block and introduces a lightweight gate at each recursion boundary that makes a binary exit-or-continue decision for every item independently. Items whose representations have sufficiently matured exit early, while those requiring deeper reasoning continue to subsequent recursions. This design converts the fixed global depth into a learned, per-item property without auxiliary depth labels, and is readily adaptable to diverse backbone architectures. Extensive experiments on public benchmarks with four backbone families demonstrate that LoopRec consistently outperforms matched fixed-depth counterparts. Further analysis reveals that LoopRec’s learned allocation exhibits interpretable patterns aligned with our post-hoc oracle analysis: lower-frequency items and more recent interactions naturally receive deeper computation. Our code is available at https://github.com/Elpmis117/LoopRec.

cs.LG [Back]

[109] Measure First, Then Allocate: Calibrated Rollout Budgets for MCP Agents in Reinforcement Learning cs.LG | cs.CLPDF

Hieu Tran, Zonghai Yao, Sunjae Kwon, Hong Yu

TL;DR: 论文研究工具使用语言模型在MCP环境中的强化学习训练成本问题：当同一rollout组内奖励完全相同时，组相对学习信号为零。作者先用奖励分歧指导harness评估，提出Discord，通过精确配对推理和保持决策的早停判断harness编辑是否真正更优；进而提出BRACE，在分配生成预算前结合历史任务证据、任务成功率不确定性和相关奖励模型预测哪些组可能产生混合奖励。实验表明Discord在受控harness实验中更受支持，BRACE在三个Qwen3-VL规模上以更少新训练rollout超过最强基线峰值，并在BFCL和NESTFUL上无需额外训练即领先对比方法。

Details

Motivation: 训练工具使用语言模型需要昂贵交互，但rollout组内奖励相同会缺乏组相对学习信号，造成采样预算浪费。论文希望先度量并预测奖励分歧，再决定rollout预算分配，从而提升强化学习训练和harness评估的效率。

Result: 受控harness实验支持Discord；实时MCP实验显示存在保留成本权衡。BRACE在三个Qwen3-VL模型规模上以更少新训练rollout超过最强绘制基线峰值（历史bank收集单独计算），并在BFCL和NESTFUL上无需进一步训练即领先对比方法；消融和前瞻评估验证了bank、估计器、分配分数及预测校准。

Insight: 宣称创新包括：用奖励分歧指导harness评估与rollout分配；Discord采用精确配对推断和决策保持早停，只依据不一致配对结果比较harness；BRACE在生成前结合历史证据、成功率不确定性和相关奖励模型预测混合奖励组，再分配预算。客观看，核心创新是把组相对RL的信号质量问题转化为可预测的奖励对比问题，先预测组内奖励是否有区分度再采样，有望提高昂贵rollout的样本效率，但需注意历史bank收集成本被单独计算。

Abstract: Training tool-using language models requires costly interaction, yet identical rewards within a rollout group provide no group-relative learning signal. We use reward disagreement to guide harness evaluation and the allocation of rollouts for policy training in Model Context Protocol (MCP) environments. We first introduce Discord, which tests harness edits using exact paired inference and decision-preserving early stopping: only disagreeing paired outcomes favor one harness over another. This insight motivates prospective allocation in BRACE (Bank-calibrated Rollout Allocation under Correlated Episodes). BRACE combines historical task evidence, uncertainty in task success rates, and a correlated reward model to forecast mixed-reward groups before allocating generation budget. Controlled harness experiments favor Discord, while live MCP experiments expose a retention-cost trade-off. At three Qwen3-VL model scales, BRACE exceeds the strongest plotted baseline peaks with fewer new training rollouts; historical-bank collection is counted separately. It also leads the compared methods on BFCL and NESTFUL without further training. Ablations probe the bank, estimator, and allocation score; prospective evaluation checks forecast calibration. These results support forecasting reward contrast before generation.

[110] FlashNext-RL: A System Optimization Study for Long-Horizon Agentic RL of Qwen3.8-Flash-Next cs.LG | cs.CLPDF

Zhewei Yao, Yuxiong He

TL;DR: 论文提出 FlashNext-RL，一个基于 prime-rl 并集成 vLLM 进行 SWE rollout 生成的模型特定 RL 训练优化框架，面向 Qwen3.8-Flash-Next 的长程智能体强化学习。它在单台 8xB200 节点上以 BF16 训练完整 48 层模型，通过冻结部分组件并优化 host offload、梯度累积以及 MoE、稀疏注意力、门控残差流等 kernel，显著提升长上下文训练吞吐，并支持实际 SWE RL 运行。

Details

Motivation: 长程智能体 RL 同时需要高效 rollout 生成和长序列训练，但 Qwen3.8-Flash-Next 的内存需求与模型特定算子使现有开源实现难以胜任，因此需要针对该模型架构定制 RL 训练系统。

Result: 全模型开发研究中，非 padding 训练吞吐从初始 32K 配置约 395 tokens/s/GPU 提升到优化后 131K 配置 4,438 tokens/s/GPU，硬件 matmul 吞吐从约 20 提升到 213 TFLOPs/GPU；8 层模型受控实验量化了各项优化贡献。系统结合三个推理节点支持目标 1,024 条可训练 rollout、131,072 token 上下文的 live SWE RL 运行，steps 5-15 平均每轮 22.4 分钟，熵和 trainer-sampler mismatch KL 保持有界。

Insight: 摘要宣称的创新点在于面向 Qwen3.8-Flash-Next 的模型特定 RL 训练优化，包括冻结部分层与组件、优化 host offload/梯度累积，以及针对 MoE、稀疏注意力和门控残差流的 kernel 优化。客观来看，其可借鉴之处是强调按模型架构、内存布局和 kernel 需求定制 RL 训练系统，并通过受控累积研究量化单项优化贡献，而非仅依赖通用 RL 基础设施。

Abstract: Long-horizon agentic reinforcement learning (RL) requires both efficient rollout generation and training on long sequences. However, the memory requirements and model-specific operators of Qwen3.8-Flash-Next make this difficult with existing open-source implementations. To resolve these limitations, we introduce FlashNext-RL, a model-specific RL training optimization framework based on prime-rl, with vLLM for software engineering (SWE) rollout generation. Our main focus is BF16 training of the full 48-layer model on a single 8xB200 node, with the first two layers, the n-gram table, the sparse-attention indexer and the multi-token prediction head frozen. We optimize host offload, gradient accumulation and the kernels for mixture-of-experts, sparse attention and the gated residual stream. In the full-model development study, non-padding training throughput increases from approximately 395 tokens/s/GPU in the initial 32K-token configuration to 4,438 tokens/s/GPU in the optimized 131K-token configuration, with hardware matmul throughput (aka without considering the cost of attention components) increasing from approximately 20 to 213 TFLOPs/GPU. A controlled cumulative study on an 8-layer model further quantifies the contributions of individual optimizations. Together with three inference nodes, the system supports a live SWE RL run with a target batch of 1,024 trainable rollouts and a 131,072-token context limit. Over steps 5-15, an iteration takes 22.4 minutes on average, with bounded entropy and trainer-sampler mismatch KL. These results highlight the importance of tailoring RL training to the model’s architecture, memory layout and kernel requirements, rather than relying only on general-purpose RL infrastructure.

[111] ExpEvolve: Structuring and Reusing Experience for Open-Ended Discovery cs.LG | cs.CL | stat.MLPDF

Yongqiang Chen, Chenxi Liu, Deyu Zou, Zhenhao Chen, Tongliang Liu

TL;DR: 论文提出 ExpEvolve，面向基于 LLM 的进化编码智能体在开放式科学发现中的程序演化问题，将“想法级”经验提取与控制引入进化过程：从程序中抽取可解释评分差异的高层因素，并用于指导经验复用、开发与探索。方法在 11 个开放式数学任务上显著提升最终性能和样本效率，并迁移到其他 LLM 与 MLS-Bench 的机器学习任务。

Details

Motivation: 现有 AlphaEvolve 类进化编码智能体虽能迭代改进程序，但变异多停留在低层次表面启发式，缺乏持久、可复用的控制单元，难以像科学发现那样通过“想法”积累知识。

Result: 在 11 个开放式数学任务上，ExpEvolve 显著提升最终性能，样本效率最高提升 5.0x；结果还可迁移到 4 个其他 LLM 以及 MLS-Bench 的 6 个机器学习任务。分析表明其能发现想法级因素，从而改善进化发现中的探索与利用。

Insight: 创新点在于将程序演化中的经验抽象为高层“想法因子”，用其解释程序评分差异并作为持久控制单元指导开发与平台期探索；同时给出理论条件并构建实用框架。客观来看，其可借鉴之处是把低层程序变异提升为可复用、可迁移的高层概念控制，从而缓解开放式发现中的样本低效与探索停滞问题。

Abstract: Evolutionary coding agents such as AlphaEvolve are one of the notable successes in using Large Language Models (LLMs) to build AI Scientists. These agents tackle open-ended scientific problems by iteratively improving and evolving programs, leveraging the prior knowledge and reasoning capabilities of LLMs. Despite their success, mutations over programs are mostly low-level upon surface heuristics, without a persistent unit of control. In contrast, science compounds through ideas. In this work, we propose ExpEvolve, which injects idea-level extraction and control into evolution: it discovers and leverages the high-level factors from programs that explain scoring differences among different programs. We characterize theoretical conditions on when discovery with idea factors improves sample efficiency. We also build a practical framework to show that the extracted high-level idea factors are helpful for guiding exploitation from experience and exploration when discovery comes to a plateau. Across 11 open-ended mathematical tasks, ExpEvolve significantly improves the final performance and sample efficiency up to 5.0x. The improvements also transfer to 4 other LLMs and to 6 machine learning tasks from MLS-Bench. Analysis shows that ExpEvolve can discover idea-level factors to provide better exploration and exploitation for evolutionary discovery.

[112] BIRD-RL: Scaling Agentic Reinforcement Learning over Stateful Data-Centric Environments cs.LG | cs.CLPDF

Ge Qu, Jinyang Li, Xiaohan Xu, Xiaolong Li, Bowen Qin

TL;DR: 论文提出 BIRD-RL，一个面向有状态、以数据为中心环境的智能体强化学习框架，通过四阶段课程交替提升基础推理与轨迹自适应在线探索，并配合轨迹作用域的持久化执行基础设施，实现安全的数据库副本复用与轨迹隔离。该方法在 BIRD-CRITIC-SQLite 基准上训练 7B/14B 小模型完成 SQL 调试，并同时支持 SQL 生成。

Details

Motivation: 现有智能体强化学习框架多以工具为中心，难以处理 SQL 调试等有状态数据任务；为每个 rollout 单独隔离数据库容器会导致空间效率低下，而朴素复用数据库又会产生轨迹污染和跨轨迹干扰，破坏策略更新稳定性。

Result: 在 BIRD-CRITIC-SQLite 基准上，7B 和 14B 模型分别达到 44.40% 和 48.00% 的成功率，与 Claude-Opus-4.6 Agent 的 48.20% 具有竞争力；设计使空间 RL 效率提升 12.8 倍，且无需 Kubernetes 即可支持可扩展训练。

Insight: 创新点在于四阶段智能体 RL 课程，交替增强基础智能体推理与轨迹自适应在线探索，以规避端到端直接学习的不稳定性；同时提出轨迹作用域持久化执行基础设施，在保留轨迹隔离的前提下安全复用数据库副本，并证明统一小模型可同时处理 SQL 调试与生成等多目标任务。

Abstract: While agentic reinforcement learning (RL) has emerged as a promising paradigm for training small language models (SLMs), existing frameworks are predominantly tool-centric and struggle with data-centric environments. Specifically, applying tool-centric frameworks to stateful tasks like SQL debugging introduces severe spatial inefficiencies since proper environment isolation typically requires a separate database container per rollout. Naively reusing databases without tracking evolving states across multi-turn interactions causes trajectory contamination and cross-trajectory interference, which severely destabilizes policy updates. To overcome these bottlenecks, we introduce BIRD-RL, a novel four-stage agentic RL curriculum that alternately enhances foundational agent reasoning and trajectory-adaptive on-policy exploration to circumvent the instability inherently associated with direct end-to-end learning. Crucially, to support this curriculum, BIRD-RL is driven by a Trajectory-scoped Persistent Execution Infrastructure, which enables safe database replica reuse while preserving trajectory isolation. This design improves spatial RL efficiency by 12.8x, successfully enabling scalable agentic RL for complex data-centric tasks without needs of Kubernettes. Empirical evaluations on the BIRD-CRITIC-SQLite benchmark demonstrate that our 7B and 14B models achieve Success Rates of 44.40% and 48.00%, respectively, showing competitive performance against strong proprietary model-based agents such as Claude-Opus-4.6 Agent (48.20%). Furthermore, BIRD-RL successfully trains unified SLMs capable of both advanced SQL debugging and generation, highlighting its broader potential for multi-objective optimization across heterogeneous, real-world data-centric tasks.

[113] Harness Scaling for Generalizable Agentic Reinforcement Learning cs.LG | cs.AI | cs.CLPDF

Fei Bai, Huatong Song, Chuan Hao, Ran Tao, Jian Yang

TL;DR: 论文提出「Harness Scaling」这一面向智能体强化学习的新缩放维度：不再只在单一固定的智能体执行框架（harness）上做RL训练，而是构建因子化的harness空间，挑选覆盖均衡的紧凑训练子集，并以「全局覆盖+同任务受控变化」的方式组织跨harness训练，从而提升模型对未见harness配置的泛化能力。

Details

Motivation: 现有智能体RL通常只在单一固定harness上训练，模型容易对该配置过拟合、形成专门化，难以迁移到其他harness配置，因而无法充分发挥harness带来的工具使用、上下文管理与长程交互优势。

Result: 在多个benchmark、多种模型规模和多种RL算法上，Harness Scaling都稳定提升了对未见白盒配置的泛化；且经结构化白盒变化训练的模型能有效迁移到未见的外部黑盒harness，效果优于用单一复杂黑盒harness训练的模型。

Insight: 创新点在于将harness多样性确立为智能体RL的一个有意义的新缩放维度，并指出「如何组织这种多样性」（因子化空间设计、均衡覆盖的训练子集、跨harness训练编排）对跨harness泛化至关重要，为提升智能体系统的可迁移性提供了可借鉴的训练范式。

Abstract: Agent harnesses have become a central component of modern agentic systems, shaping how models use tools, manage context, and coordinate long-horizon interaction. Realizing the benefits of these harnesses depends on how effectively the underlying model can make use of them, motivating reinforcement learning directly through the harness. However, existing approaches typically train through a single fixed harness, which can lead to specialization and limited transfer across harness configurations. To address this, we introduce \textbf{Harness Scaling}, a new scaling dimension for agentic RL that extends training from a single harness to diverse execution configurations. Concretely, we construct a factorized harness space, select a compact training subset with balanced factorial coverage, and organize cross-harness training to combine global coverage with controlled variation for the same task. Across multiple benchmarks, model scales, and RL algorithms, Harness Scaling consistently improves generalization to unseen white-box configurations. Moreover, models trained through structured white-box variation transfer effectively to unseen external black-box harnesses, outperforming models trained through a single complex black-box harness. These results establish harness diversity as a meaningful scaling dimension for agentic RL and highlight the importance of how this diversity is organized for cross-harness generalization.

cs.RO [Back]

[114] Real2Sem2Real: 3D Semantic Augmentations for Robot Learning from a Single Human RGB Demonstration cs.RO | cs.CVPDF

Matthew Strong, Ankush K. Dhawan, Julien Poffet, Giuse Pham, Jinho So

TL;DR: 论文提出 Real2Sem2Real（R2Sem2R），将一段单目 RGB 人类演示视频转换为同一类别下大量语义物体实例的机器人演示数据。方法利用 DINOv3 蒸馏高斯泼溅跟踪物体 SE(3) 位姿，并通过语义对齐把由照片生成的同类“数字表亲”物体映射到演示物体，再将人手映射为可行机器人抓取，最后在 IsaacSim 或 MuJoCo 中生成机器人演示。基于每个任务仅一段人类视频训练视觉-语言-动作策略，在两类真实机器人和四个任务上达到与遥操作数据训练相当或更好的效果。

Details

Motivation: 人类视频易获取且无需机器人硬件，但现有方法通常依赖深度传感、把机器人叠加到 2D 视频上，或只能为演示中的特定物体生成数据，难以泛化到同类不同物体实例。论文希望仅用单目 RGB 人类演示实现可扩展、跨物体实例的机器人学习数据生成。

Result: 物体 SE(3) 位姿跟踪优于现有方法和结合估计度量深度的 FoundationPose；可在约两小时内生成超过 2000 条演示。在两类真实机器人、四个任务上，用每个任务一段人类视频训练的策略在不同物体实例上匹配或超过遥操作数据训练的策略。

Insight: 创新点在于 real-to-semantic-to-real 数据增广管线：用 DINOv3 蒸馏高斯泼溅做物体位姿跟踪，语义对齐同类数字表亲并匹配整体物体与咖啡机托盘等功能区域，将人手映射扩展到机器人抓取，并结合 IsaacSim/MuJoCo 生成大规模机器人演示。客观看，其价值在于把单目人类视频转化为跨实例、可物理仿真的机器人学习数据，减少对机器人遥操作和深度传感器的依赖。

Abstract: Human video scales easily and does not need robot hardware, but existing methods require depth sensing, overlay the robot onto 2D video, or generate data only for the demonstrated objects. We present Real2Sem2Real (R2Sem2R), a real-to-semantic-to-real pipeline that turns one monocular RGB video of a human demonstration into thousands of robot demonstrations across the semantic object instances of a category. R2Sem2R tracks the SE(3) pose of each object with DINOv3-distilled Gaussian splats, outperforming existing methods and FoundationPose with estimated metric depth. R2Sem2R then aligns digital cousins, each generated from one or a few photographs, to the demonstrated objects with a semantic alignment. This alignment matches whole objects and the functional regions of target objects, such as a coffee machine’s tray. R2Sem2R maps the human hand to a feasible robot grasp that extends to the cousins. Robot demonstrations are generated either in IsaacSim, for photorealistic rendering, or MuJoCo, for contact physics, and R2Sem2R can generate over 2,000 episodes in about two hours. On two real robots and four tasks, we train a vision-language-action policy on data from one human video per task, and policies trained on our data match or exceed policies trained on teleoperation across diverse object instances. The project page is available at https://real2sem2real.github.io/.

[115] How Representational Priors Shape Generalization in World Action Models cs.RO | cs.AI | cs.CVPDF

Junha Chun, Youngjoon Jeong, Taesup Kim

TL;DR: 论文研究世界动作模型（WAMs）因生成骨干不同而在泛化性能上出现差异的原因，提出从骨干继承的表征几何及策略训练对其重构的角度解释。作者提出“干扰正交性”指标，度量训练前特征对分布外视觉变化的分离程度，并发现该指标可前瞻性预测WAM训练后的策略泛化鲁棒性。

Details

Motivation: WAMs在不同生成骨干下泛化表现差异显著，论文希望弄清这种差异是否源于骨干带来的表征先验，并找到可在策略训练前预测泛化能力的几何指标。

Result: 在多种图像和视频骨干上，训练前更高的干扰正交性与WAM训练后对视觉变化的更强鲁棒性相关；鲁棒WAM还表现出干扰抑制、特征位移相对减小以及更稳健的动作生成。摘要未提及具体benchmark或SOTA对比。

Insight: 创新点在于用表征几何视角连接生成骨干先验与WAM策略泛化，并提出干扰正交性作为可前瞻性衡量泛化鲁棒性的指标；其价值在于为选择世界模型骨干和理解策略鲁棒性提供可量化的几何分析工具。

Abstract: World action models (WAMs) are shown to substantially diverge under generalization performance based on their generative backbones. We investigate this gap through the representational geometry that WAMs inherit from their backbones and how policy training reorganizes it. Our central hypothesis is that a useful representational prior would structure geometry that separates nuisance information from the in-distribution manifold in advance of training as a policy. We quantify this with nuisance orthogonality; orthogonality of the feature under visual variation outside the empirical in-distribution manifold. Across various image and video backbones, we show that a higher orthogonality from the representational prior is associated with stronger robustness to visual variations after WAM training, enabling policy generalization measurable prospectively for world models. In addition, analyzing the resulting robust WAMs reveals nuisance suppression; relative suppression of feature displacement caused by such variations and robust action generation of the policy. Together, these findings provide a geometric perspective on how the representational prior of the world model shapes the generalization performance of WAMs.

[116] DepState-JEPA: Learning Depth-Grounded State Representation for Visual Perception and Manipulation cs.RO | cs.CVPDF

Xiujian Liang, Liuyuan Na, Zehao Du, Jiacheng Xu, Weixi Song

TL;DR: 论文提出 DepState-JEPA，一种基于 V-JEPA 2.1 的双流 RGB-D 联合嵌入预测架构，通过共享潜在状态 DepState 将空间几何理解与动作条件预测连接起来。模型利用深度作为互补观测流，并将 RGB 与深度两种模态锚定到度量 3D 空间，同时保留各模态私有特征用于解码；在动作条件状态转移后，用同一组解码器预测未来 RGB 与深度特征。该方法还引入机器人状态读出、物体轨迹监督和标定前向运动学约束来塑造状态学习。

Details

Motivation: 机器人操作中的预测表征需要把场景几何与动作引起的状态变化联系起来；论文的出发点是：理解动作会改变什么，首先要理解物体在哪里，因此用互补的 RGB 与深度观测构建共同场景表征，并据此推断动作后的变化。

Result: 比较实验显示，经过 RGB-D 适配后 RGB 编码器迁移更强，并在特征补全、条件视觉预测和操作成功率上表现更好；模型内消融表明深度内容、度量坐标和未来状态匹配均有贡献，前向运动学正则还能带来额外控制增益。摘要未给出具体 benchmark 数值或明确 SOTA 结论。

Insight: 创新点在于双流 RGB-D JEPA 与共享潜在状态 DepState 的设计：深度既作为互补观测流，又用于将两种模态锚定到度量 3D 空间；通过射线条件读出在无目标深度时预测局部特征，并在动作条件转移后预测未来 RGB-D 特征；同时结合机器人状态递归输入、物体轨迹监督和标定前向运动学约束来增强状态学习。

Abstract: Predictive representations for robot manipulation should connect scene geometry with the state changes induced by actions. Our intuition is that understanding what an action will change starts with understanding where things are: complementary RGB and depth observations ground a common scene representation from which to infer those changes. Guided by this intuition, we introduce \textbf{DepState-JEPA}, a dual-stream RGB–Depth joint-embedding predictive architecture that connects spatial understanding with action-conditioned prediction through a shared latent state, \textbf{DepState}. Built on V-JEPA~2.1, the framework uses depth both as a complementary observation stream and to ground both modalities in metric 3D space, aggregating shared information into DepState while retaining modality-private features for decoding. Ray-conditioned readouts predict local features without target depth; after an action-conditioned transition, the same decoders predict future RGB and depth features using the predicted state and fixed initial private features. Robot-state readouts provide recursive proprioceptive inputs; object-trajectory supervision and calibrated forward kinematics shape state learning through motion targets and joint–TCP consistency. Comparisons show stronger RGB-encoder transfer after RGB-D adaptation, feature completion, conditioned visual forecasting, and manipulation success. Within-model ablations identify contributions from depth content, metric coordinates, and future-state matching, with additional control gains from FK regularization.

[117] AtomicRL: Atomic Reinforcement Learning for Long-Horizon Robotic Manipulation cs.RO | cs.AI | cs.CVPDF

Likui Zhang, Zhe Zhu, Zhihao Zhan, Yibo Bai, Hanlin Wang

TL;DR: 论文提出AtomicRL，一种面向长时程机器人操作的强化学习后训练框架，利用原子任务结构改进VLA策略的信用分配和探索效率。方法包含语义感知策略优化SAPO、瓶颈自适应目标探索BATE，以及连接仿真在线RL与真实部署的Real2Sim2Real流程，从而提升复杂长时程任务的学习效率与操作性能。

Details

Motivation: 现有VLA模型仅靠模仿学习难以掌握复杂长时程技能，而VLA-RL方法常因轨迹级稀疏奖励导致信用分配延迟且有偏；同时不同原子子任务掌握不均衡会形成性能瓶颈，限制整体表现。

Result: 在LIBERO-LONG上，AtomicRL在两种VLA架构上相对PPO取得3.6%和4.0%的绝对性能提升；在CALVIN ABC-D上全任务性能相对PPO最高提升9.5%；结合Real2Sim2Real流程后，真实世界任务性能平均提升21.3%。

Insight: 论文宣称的创新点在于利用原子任务结构进行可靠信用分配与自适应探索：SAPO通过语义感知门控抑制失败子任务的TD残差跨边界污染，BATE根据子任务掌握程度动态分配采样与探索激励，并配套Real2Sim2Real迁移流程。客观上看，其创新主要是将子任务级信用分配、瓶颈感知探索和仿真到真实部署统一到VLA-RL后训练框架中。

Abstract: Vision-language-action (VLA) models have shown promise in robotic manipulation, but imitation learning (IL) alone often struggles to master complex long-horizon skills. Reinforcement learning (RL) offers an effective avenue for further improving VLA policies. However, existing VLA-RL methods typically suffer from delayed and biased credit assignment due to trajectory-level sparse rewards, while imbalanced subtask mastery creates bottlenecks that limit overall performance. To address these challenges, we propose AtomicRL, an RL post-training framework that leverages atomic task structures for reliable credit assignment and adaptive exploration. AtomicRL introduces Semantic-Aware Policy Optimization (SAPO), which uses semantic-aware gating to constrain TD residuals from failed atomic subtasks from propagating across task boundaries, mitigating credit contamination while preserving intra-subtask credit assignment and value bootstrapping. In addition, Bottleneck-Adaptive Target Exploration (BATE) dynamically coordinates task sampling and local exploration incentives based on the mastery of individual atomic subtasks, directing learning toward the remaining bottlenecks. Finally, we develop a Real2Sim2Real pipeline that bridges online RL in simulation with real-world deployment, facilitating practical transfer to physical robots. On LIBERO-LONG, AtomicRL achieves absolute performance gains of 3.6% and 4.0% over PPO across two VLA architectures. On CALVIN ABC-D, it improves full-task performance by up to 9.5% over PPO. Combined with the Real2Sim2Real pipeline, AtomicRL further improves real-world task performance by an average of 21.3%. These results demonstrate that AtomicRL effectively improves the learning efficiency and manipulation performance of VLA policies on long-horizon tasks.

[118] Multimodal Embodied Agents: A Unified Capability-centric Perspective from Computer-Use to Robot-Use cs.RO | cs.AI | cs.CV | cs.MA | cs.MMPDF

Yanzhe Chen, Ziyi Yang, Jifeng Zhu, Qiming Huang, Ruihe An

TL;DR: 论文提出一种以能力为中心的统一框架PAPAV，用感知、预测动作效果、规划、执行、验证五项功能来刻画多模态具身智能体（MMEAs），并比较从计算机使用智能体到机器人使用系统之间的差异。作者进一步基于64个基准分析现有评测，指出现有评估主要集中于“执行”，而“预测”和“验证”能力评估严重不足。

Details

Motivation: 论文关注：当多模态智能体从计算机环境迁移到机器人/物理环境时，核心能力与约束发生了什么变化。其出发点是现有研究缺乏统一的能力视角来比较多模态智能体、具身智能体和机器人系统。

Result: 论文主要是框架与基准分析型工作，并未提出在某个基准上达到SOTA的具体模型。其对64个基准的统计显示，显式评估主要集中在Act，只有5个评估Anticipate、3个评估Verify，说明关键能力仍多被任务成功率掩盖。

Insight: 创新点在于提出按功能而非架构定义的PAPAV五能力框架，并总结从部分可观测到结果难验证等五类物理约束。可借鉴之处是用统一能力维度设计和衡量物理智能体，而不只依赖任务成功率。

Abstract: Multimodal agents (MMAs) sustain interaction through reasoning, memory, tools, and feedback, most visibly as computer-use agents, while robotic systems (RSs) couple sensing and actuation under physical dynamics. We define multimodal embodied agents (MMEAs) as goal-directed systems that couple multimodal task reasoning with physical action and revise decisions from the resulting feedback. This raises a central question: what changes when a multimodal agent moves from computer-use to robot-use? We introduce PAPAV, a unified capability-centric framework of five recurring functions: Perceive the current state, Anticipate action effects, Plan a feasible course, Act through an interface or body, and Verify the outcome. Defined by function rather than architecture, these capabilities let PAPAV identify what all three share and then compare MMEAs with MMAs and with RSs. Five physical constraints, from partial observability to unverifiable outcomes, leave fewer choices fixed in advance and less room to reverse errors. Across 64 benchmarks, explicit evaluation centers on Act; only 5 assess Anticipate and 3 assess Verify, leaving both largely hidden behind task success. These constraints also frame the open challenges we identify. PAPAV therefore provides a common basis for designing reliable physical agents and measuring progress across all five capabilities rather than task success alone.

cs.SD [Back]

[119] Qwen-Audio-3.1-ASR Technical Report cs.SD | cs.CLPDF

Chuanmeng Bian, Daren Chen, Peixin Chen, Zhigao Chen, Chong Deng

TL;DR: 论文发布 Qwen-Audio-3.1-ASR 系列音频语言模型，在 Qwen-Audio-3.0-ASR 基础上升级语音识别，并新增说话人分离转写与通用音频理解能力。该系列支持 30 种语言和 16 种中文方言，具备历史上下文、层级热词、转写润色等指令控制能力，并提供异步文件转写、低延迟流式、交互语音输入和非流式识别等部署变体。

Details

Motivation: 论文旨在将音频模型从单纯识别语音内容扩展到理解说话人、声学事件及其时序关系，同时满足多语言/方言识别、说话人分离转写和不同部署场景的需求。

Result: 摘要称在常规 ASR、多语言与方言语音、说话人分离转写和通用音频理解等评测中展示了该系列模型的能力，但未给出具体定量指标或明确 SOTA 结论。

Insight: 创新点在于统一扩展音频语言模型至说话人分离结构化转写、重叠语音表示和通用音频理解任务，并结合历史上下文、层级热词、原生转写润色等指令控制能力，以及面向不同部署需求的模型变体设计。

Abstract: We present Qwen-Audio-3.1-ASR, a family of audio-language models that builds upon its predecesor Qwen-Audio-3.0-ASR with improved speech recognition and expanded capabilities in speaker-diarized transcription and general audio understanding. The family retains support for 30 languages and 16 Chinese dialects, together with instruction-controlled capabilities including historical-context conditioning, hierarchical hotword customization, and native transcription polishing. Speaker-diarized ASR produces structured outputs containing speaker labels, timestamps, and recognized text, maintaining consistent speaker labels across turns and the representation of overlapping speech. The general audio understanding capabilities cover audio-, speech-, and music-captioning, temporal grounding and sound event detection, as well as audio question answering and multiple-choice reasoning. These tasks broaden the application scope from recognizing spoken words to interpreting speakers, acoustic events, and their temporal relations. To accommodate different deployment requirements, the family includes variants for asynchronous file transcription, low-latency streaming, interactive voice input, and non-streaming recognition, with speaker-diarized transcription and general audio understanding available in certain variants. Evaluations span conventional ASR, multilingual and dialectal speech, speaker-diarized transcription, and general audio understanding, demonstrating the capabilities of the Qwen-Audio-3.1-ASR family across transcription and broader audio analysis tasks.

cs.SE [Back]

[120] Where Should a Stuck Agent Spend Its Next Tokens? Causal Attribution of Inference-Time Resources cs.SE | cs.AI | cs.CLPDF

Saiyue Lyu, Jerrold Huang, Yunbei Zhang, Janet Wang, Zhenting Wang

TL;DR: 论文提出Banzhaf Resource Attribution（BRA）受控因子框架，在固定失败状态和修订协议下量化计算、信息、反馈三类推理时资源对编码智能体修复失败的边际贡献与交互；并在MBPP+/HumanEval+的函数级失败和SWE-bench Verified的仓库级失败上评估六个模型。结果显示任务相关规格最有帮助，公共测试反馈增益小，检索代码几乎无益，且前沿模型常偏好反馈但收益不如随机选资源。

Details

Motivation: 推理时扩展常被简化为token预算，但卡住的编码智能体可寻求计算、信息或反馈等不同资源；论文要解决“资源该花在哪”以及智能体偏好资源是否真正有效的问题。

Result: 在194个函数级失败和126个仓库级失败上，任务相关规格稳定提升修复；公共测试反馈有小增益，检索代码几乎无增益；额外计算价值随模型能力下降，信息仍有用；3/4前沿模型在95-100%情况下选择反馈，收益低于均匀随机资源选择。

Insight: 摘要宣称创新是BRA对多类推理时资源做可控因果归因并分析交互；客观看，其价值在于揭示智能体资源偏好与实际修复收益的错配，提示需按边际收益而非默认反馈来分配推理资源。

Abstract: Inference-time scaling is often treated as a token-budget problem, but a failed coding agent can seek qualitatively different resources: computation, information, or feedback. We introduce Banzhaf Resource Attribution (BRA), a controlled factorial framework that measures each resource’s marginal contribution to repair and its interactions while holding the failed state and revision protocol fixed. We evaluate six models on 194 function-level failures from MBPP+ and HumanEval+ and 126 repository-level failures from SWE-bench Verified. Task-relevant specifications consistently improve repair, whereas public-test feedback provides small gains and retrieved code little to none. The value of additional computation declines with model capability, while information remains useful; resource channels largely exhibit diminishing returns, although complementary information can reinforce itself. Yet three of four frontier models select feedback in 95-100% of cases and obtain lower gains than uniform random resource selection. BRA thus exposes a gap between the resources agents prefer and those that measurably improve repair.

[121] HiddenLens: A Multi-Layer LLM Framework for Detecting and Resolving Hidden Assumptions in Software Requirements cs.SE | cs.AI | cs.CLPDF

Tanu Singh, Aashna Suman, Vinod Patidar

TL;DR: 论文提出 HiddenLens，一个六层 BERT-LLM 框架，用于检测、分类、追问并解决软件需求中隐含的假设，而不仅是识别歧义。框架结合 BERT 注意力熵定位、CoT 推理、微调 BERT 的 NFR 分类、FS LLM 澄清选项生成、人工验证和需求重写，并构建了包含 312 条需求、847 个隐藏假设的 HiddenLens Corpus。

Details

Motivation: 自然语言软件需求常遗漏性能、运行条件、用户行为和领域约束等假设，若到实现阶段才发现会显著增加成本；已有研究多关注歧义和模糊性，对看似清晰合理但隐含假设的需求关注不足。

Result: HiddenLens 在自建 HLC 数据集上端到端 F1 达 0.79，分别比 keyword-based 和 QuARS-Adapted 基线高 71% 和 61%；其基于 BERT 的 NFR 分类器 F1 为 0.847。

Insight: 创新点在于首个面向隐藏假设的检测与解决框架及专用数据集，将隐藏假设定位、风险推理、六类 NFR 分类、澄清选项排序、人工确认和需求重写整合为可追溯流程，使需求更显式、可测试和可追踪。

Abstract: Natural-language software requirements often omit assumptions about performance, operating conditions, user behaviour, and domain constraints, affecting architecture, testing, and deployment. If undiscovered until implementation, resolving them is far more costly than addressing them early. Although ambiguity and vagueness are widely studied, less attention has been given to assumptions embedded in requirements that appear clear and reasonable. We propose HiddenLens, a novel six-layer BERT-LLM pipeline and, to our knowledge, the first framework engineered to detect, classify, question, and resolve hidden assumptions rather than merely flag ambiguity. Layer 1 uses BERT attention entropy to localise unresolved spans; Layer 2 uses CoT-prompted LLM reasoning to articulate each assumption and its failure risk; Layer 3 uses fine-tuned BERT to classify assumptions into six non-functional requirement categories: performance, security, scalability, usability, reliability, and domain; Layer 4 generates ranked clarification options using FS LLM prompting; Layer 5 enables human validation through confirmation, editing, or rephrasing; and Layer 6 rewrites the requirement using the confirmed resolution, producing an explicit, testable, and traceable requirement. We also introduce the HiddenLens Corpus (HLC), comprising 312 requirements annotated with 847 hidden assumptions across six NFR categories. Inspired by PROMISE and PURE, HLC addresses the lack of annotated data for this task and is, to our knowledge, the first dataset developed specifically for it. HiddenLens outperforms keyword-based and QuARS-Adapted baselines by 71% and 61%, respectively, achieving an end-to-end F1-score of 0.79; its BERT-based NFR classifier achieves an F1-score of 0.847. These results indicate that HiddenLens can help analysts resolve hidden assumptions before they propagate through software development.

quant-ph [Back]

[122] Can Quantum Layers offer Performance and Security Advantages in Vision-Language Models? quant-ph | cs.AI | cs.CV | cs.LGPDF

Md Khalid Syfullah, Alvi Ataur Khalil

TL;DR: 论文在LLaVA-OneVision视觉语言模型上，用经典模拟的量子层系统评估了4/8量子比特、不同电路结构在多个插入与替换位置对性能、计算成本和鲁棒性的影响。结果表明，在视觉投影器后插入量子层可提升视觉空间推理准确率，替换解码器最后的前馈子层可在较少参数下保持较高准确率，但收益高度依赖位置且不一定降低训练或计算开销。

Details

Motivation: 量子增强AI正从概念走向实验，但量子组件如何在大规模多模态模型中发挥作用仍不清楚。论文旨在探究量子层能否为视觉语言模型带来性能和安全性优势，并厘清其适用位置与代价。

Result: 在Visual Spatial Reasoning数据集上，视觉投影器后插入量子层达到80.7%开发集准确率，比经典全量微调高2.28个百分点；替换解码器最后前馈子层保持79.22%准确率并减少1307万可训练参数。论文还在六种攻击设置下评估混合模型鲁棒性，但未给出统一的安全提升结论，且更少参数并不必然减少训练时间或计算量。

Insight: 论文宣称的创新点是对VLM中量子层的插入与替换位置进行系统化实验测绘，并同时比较性能、参数/计算成本和安全性。客观来看，其可借鉴之处在于提供了量子-经典混合VLM的设计空间与位置敏感性分析，而非提出全新的量子架构；结果基于经典模拟量子层，实际量子硬件上的泛化性仍待验证。

Abstract: As quantum computing moves from the noisy intermediate-scale quantum (NISQ) era toward early fault-tolerant quantum computing (EFTQC), quantum-enhanced AI is shifting from a conceptual idea to experimental research. Quantum components are expected to extend the capabilities of existing AI systems, yet it remains unclear how such components will help within large multimodal models. In this study, we try to answer that question for vision-language models (VLMs) using classically simulated quantum layers. With LLaVA-OneVision as the backbone, we evaluate 4- and 8-qubit basic-ring and strongly entangling circuits at four insertion and four replacement locations. Insertion adds quantum layers to the classical model, while replacement substitutes them for classical components. Using the same setup, we compare performance and computational cost between classical and quantum-classical hybrid models. On the Visual Spatial Reasoning dataset, inserting a quantum layer after the visual projector reaches 80.7% development accuracy, 2.28 percentage-points above classical full fine-tuning. Replacing the final decoder feed-forward sublayer retains 79.22% accuracy with 13.07 million fewer trainable parameters. Quantum insertion can thus improve accuracy, and selective replacement can reduce parameters. These benefits depend strongly on where the quantum layer is inserted or which classical component it replaces: earlier replacements lower accuracy, and fewer parameters do not reliably shorten training or reduce computation. To assess potential security benefits, we evaluated the robustness of quantum-classical hybrid models under six attack settings. Overall, our study maps where quantum layers help and hurt within VLMs, offering a starting point for hybrid VLM design; whether these gains generalize remains open. The code is available at https://github.com/TITANS-Lab-Artifacts/QuantumVLM.

Table of Contents

cs.CL [Back]

[1] Scaling Therapy Supervision: Assessing Psychotherapy Fidelity Using Large Language Models cs.CL | cs.HCPDF

[2] Is Anyone Enforcing the Cartel? What Reinforcement Learning Teaches LLM Agents in Markets cs.CL | cs.GT | cs.LGPDF

[3] Where Do Legal Factors Live in LLMs? cs.CLPDF

[4] LIFT: Scoped Constraints for Evaluating and Training Instruction Following cs.CLPDF

[5] Building AI Research Teams for Autonomous Discovery of Linear Attention Architectures cs.CLPDF

[6] CSD: Common-Support Distillation for Cross-Tokenizer Knowledge Transfer cs.CL | cs.LGPDF

[7] Target Regeneration Shrinks the Gains of Draft-Training Objectives in Speculative Decoding cs.CL | cs.LGPDF

[8] A Large-Scale Evaluation of Role-Enhanced Zero-Shot Chain-of-Thought Prompting cs.CL | cs.AIPDF

[9] Why is AI-generated language so hard to understand? cs.CL | cs.AIPDF

[10] TrainBridge: Agentic NPU Training Optimization Guided by GPU Compilation Evidence cs.CL | cs.AIPDF

[11] norregel: a preregistered benchmark of dated numeric values in Norwegian law cs.CLPDF

[12] MOSAIC: An Omni-Modal Benchmark for Long-Term Memory and Cross-Modal Reasoning in Conversational Agents cs.CL | cs.AIPDF

[13] Hierarchical Masked Diffusion Language Models cs.CL | cs.AIPDF

[14] Asking What a Director Sees Can Change What Vision-Language Models Choose cs.CL | cs.CVPDF

[15] Policy Central: Detecting Conflicts, Redundancy, and Similarity in Deployed Police Policy cs.CL | cs.CYPDF

[16] ChronoSynth: Constraint-Aware Question Synthesis over Temporal Knowledge Graphs cs.CL | cs.AIPDF

[17] Right On Q: Recovering Reasoning under Quantization via Self-Generated Trajectories cs.CL | cs.LGPDF

[18] Omnilingual-MsASR: Benchmarking Multimodal, Multilingual, Long-Form Multi-Speaker ASR of Omni Large Language Models cs.CLPDF

[19] OmniSimulator: Aligning Small Language Models for Authentic Heterogeneous Behavior Modeling cs.CL | cs.AI | cs.LGPDF

[20] TenetBench: A Clause-Anchored Russian Corpus for Evaluating Language Models against Professional Codes of Ethics cs.CLPDF

[21] A Strong MoE Is All You Need for Process Supervision cs.CL | cs.AIPDF

[22] Keeping Time vs. Understanding Time: Neither Alone Explains Temporal Reasoning in LLMs cs.CL | cs.AIPDF

[23] Mitigating Query Shortcuts in Chain-of-Thought with Context-Aware Decoding cs.CL | cs.AI | cs.LGPDF

[24] RCSD: Reinforcement and Continual Learning for Domain-Specific Speculative Decoding cs.CLPDF

[25] Uni-Mem: Learning a Long-Horizon Agentic Memory Manager with Unified Online Policy Optimization cs.CLPDF

[26] Evolving with Open Worlds: Towards Recursive Self-Improvement for Future Forecasting cs.CLPDF

[27] MAPS: Macro-Adaptive Parallel Reasoning for Regulating Search Momentum in Retrieval-Augmented Generation cs.CLPDF

[28] Benchmarking Large Language Models for Detecting Satirical Code-Mixed Sentences cs.CLPDF

[29] Graph Topology Meets Energy-Driven Pathfinding: Preserving Contextual Consistency in Retrieval-Augmented Generation cs.CL | cs.AIPDF

[30] CulturalRiddles: A Multicultural Benchmark of Riddles cs.CL | cs.AIPDF

[31] Remembering Through Change: Provenance-Preserving Hypergraph Memory for Multimodal Agents cs.CLPDF

[32] What the Author Wrote, What the Agent Reads: Silent Cue Loss in Enterprise Document Ingestion cs.CL | cs.AI | cs.IRPDF

[33] Looped Uniform Diffusion Language Models cs.CL | cs.LGPDF

[34] Environment Duels: A Microcosm of LLM Co-Evolution cs.CL | cs.AIPDF

[35] LongZero: Enhancing Long-Context Reasoning in LLMs from Zero Data cs.CL | cs.LGPDF

cs.CV [Back]

[36] Complexity-Aware Surgical Video Reasoning Segmentation with Digital Twin and Context Memory cs.CVPDF

[37] JUICE: Unified Multimodal Modeling in Pretrained Representation Spaces cs.CVPDF

[38] SAIL: A Learnable Generalisation of Spatial Autocorrelation over Graph-Structured Data cs.CVPDF

[39] HorizonComposer: Spatiotemporally Consistent Driving Video Editing with Enriched Traffic Semantics cs.CVPDF

[40] DualHeads: Reward-tilted few-step diffusion made simple cs.CV | cs.AI | cs.LGPDF

[41] Pair-Free Coarse-to-Real Video Generation via Real-to-Coarse Image Supervision cs.CVPDF

[42] RoPT: Robust 6DoF Object Pose Tracking in Video cs.CV | cs.ROPDF

[43] Single-Image Food Mass Estimators Do Not Transfer Their Gram Scale Across Datasets cs.CVPDF

[44] HumanCentric-Scenes-AI: A Human-Curated Multimodal Dataset of AI-Generated Human-Centric Scenes Across Surveillance, Occupational Portraits, AI-Edited Images, and Short-Form Video cs.CV | cs.AI | cs.HCPDF

[45] OccJEPA: Occlusion as Masking for Amodal Video Segmentation cs.CV | cs.LGPDF

[46] UNIVERS: Learning UNIversal Voxel-wise Embedding via Randomized Synthesis for generalizable one-shot landmark localization cs.CV | cs.AIPDF

[47] PhyMo: Learning Physical Dynamics with Accurate and Continuous Motion from Multi-View Videos cs.CVPDF

[48] LocalAgent: Collaborative Agentic Verification for Fine-Grained Instance-Level Consistency cs.CVPDF

[49] Multi-Scale Latent Temporal Compression Attention and Video-Text Cross Attention Fusion for Video Action Detection cs.CV | cs.MMPDF

[50] A Unified Evaluation Framework of Physical Adversarial Robustness in the Edge VLM Era cs.CV | cs.CR | cs.LGPDF

[51] NeuralSRender: G-Buffer Guided Video Diffusion for Sim-to-Real Rendering cs.CVPDF

[52] DetM-Tracker: Deterministic and Memory-Efficient VIS for Autonomous Industrial Inspection cs.CVPDF

[53] Cache Locally, Recompute Selectively for Streaming Video Understanding cs.CV | cs.LGPDF

[54] DisasterRoute: Evidence-Guided Dual-System Routing for Visual Disaster Question Answering cs.CVPDF

[55] Think-Capture: Active 3D Geometry Acquisition for Spatial Reasoning in Vision-Language Models cs.CVPDF

[56] Encoded but Unused: Bounding the Sources of VLM Errors cs.CVPDF

[57] Failure Modes of Vision-Language Models for Human Activity Understanding in Thermal HRI cs.CV | cs.ROPDF

[58] CRAB: Cuboid-based Representation for As-built BIM from Point Clouds cs.CV | cs.AIPDF

[59] Computer Vision and Knowledge-Grounded Language Models for Intelligent Beekeeper Assistance cs.CV | cs.CLPDF

[60] OpenMAP-HIE: Pathology Informed Score-based Diffusion Model for Unsupervised Neonatal Hypoxic-Ischemic Encephalopathy Lesion Segmentation cs.CV | cs.AIPDF

[61] OACap: Caption-Guided Active Observation for Detailed Video Captioning cs.CV | cs.CL | cs.MA | cs.MMPDF

[62] Reading the Mask: Descriptor-Grounded BI-RADS Rationale Generation for Breast MRI Segmentation cs.CV | cs.AIPDF

[63] AdvFidelity: Unifying Preference Alignment and Few-Step Image Generation within Reinforcement Learning cs.CVPDF

[64] VideoGen Model 1: Prompt Expansion Versus Formatting in Audiovisual Generation cs.CVPDF

[65] LiveAct: Towards Real-Time Multi-Agent World Models with Heterogeneous Action Interfaces cs.CVPDF

[66] LLMSeg: Training-Free Text-Prompted Image Segmentation with B-Splines cs.CV | cs.AIPDF

[67] Look Beyond the Question: Preserve-Focus-Fuse for Whole-Slide Image Question Answering cs.CVPDF

[68] Sparsity Should Be ALLOTted by Output: Training-Free Allocation of Block-Sparse Attention for Video Diffusion Transformers cs.CVPDF

[69] Beyond Shortcut Learning: Hierarchical Spatial Prototype Regularization for Face Forgery Detection cs.CVPDF

[70] Extending Affordance Segmentation for Wearable Robots to Multi-Object Images cs.CV | cs.ROPDF

[71] Balancing Foundation Models and Trait-Specific Learning for Forehead-Crease Recognition cs.CVPDF

[72] AutoPhyX: Automatic Text-Condition Physics Property Generation cs.CV | cs.GRPDF

[73] Supervision-Adaptive Curriculum and Guarded Anti-Curriculum Learning for Dense Prediction cs.CV | cs.LGPDF

[74] Melange: A Synthetic Visual Bias Benchmark for Multi-modal Large Language Models cs.CVPDF

[75] Generalized CycleGRPO for Fine-Grained Multimodal Understanding and Grounding cs.CVPDF

[76] Unlocking Spatial Grounding in Flow-Matching Models via SNR Tuning cs.CV | cs.LGPDF

[77] CineEvolve: An Agentic Self-Evolution Framework for Cinematic Multi-Shot Video Generation cs.CVPDF