ArXiv Domain 2026-08-07

数据来源：ArXiv Domain

LLM Domain Papers

1. Simulator-Grounded Large Language Models for Industrial Causal Reasoning: Tool-Use, Structured Injection, and Plant-Portable Retrieval for Wastewater Treatment Decision Support

Abstract:Wastewater operators need answers grounded in how their plant’s variables interact and how fast effects propagate, not in generic pretraining text, when asking causal questions such as “why is N2O rising?” or “what happens if I cut aeration by 20%?”. We compare three concrete ways to ground a frozen Qwen2.5-32B-Instruct model in an architecturally interpretable wastewater simulator (CCSS-IX): a live simulator oracle (Method 1), structured parameter injection (Method 2), and a Decoupled Recall-Reasoning (DRR) retriever (Method 3). On a 198-question causal benchmark the three reach 99.5%, 79%, and 75.8%, forming a deployment ladder above the strongest retrieval-augmented baseline at 48%. The DRR retriever has 110M parameters and trains per plant in ~17 seconds; after cross-plant transfer to a biologically distinct plant it still reaches 88%, while Method 2’s static table cannot transfer. On a 60-question counterfactual benchmark only Method 3 handles queries about what happens after an intervention: +16.3 pp over Method 2, paired 95% CI [+7.1, +26.4] pp, with 100% on the timescale and operating-regime categories. On the AI2 Reasoning Challenge (ARC) with an OpenBookQA fact corpus, the same selective-retrieval mechanism reaches 79% versus unconstrained Llama-3.1-8B 76% and full-injection 74%, a +3 pp out-of-domain replication that argues against a result specific to wastewater treatment. We provide the first single-simulator comparison of live tool-use, static parameter injection, and learned numerical-parameter retrieval for industrial causal question answering.

中文摘要

摘要：废水处理操作员在提出因果性问题（例如“为什么N2O上升？”或“如果我将曝气减少20%，会发生什么？”）时，需要的是基于其工厂变量如何相互作用及效应传播速度的答案，而不是基于通用预训练文本的答案。我们比较了三种将冻结的 Qwen2.5-32B-Instruct 模型与具有可架构解释性的废水模拟器（CCSS-IX）关联起来的具体方法：实时模拟器工具（方法1）、结构化参数注入（方法2）、以及解耦记忆-推理（DRR）检索器（方法3）。在一个198个问题的因果基准测试中，这三种方法分别达到99.5%、79%和75.8%，形成了一个高于最强检索增强基线（48%）的部署梯队。DRR检索器拥有1.1亿个参数，并且每个工厂训练时间约为17秒；在跨工厂迁移到生物学上不同的工厂后，它仍能达到88%，而方法2的静态表无法迁移。在一个60个问题的反事实基准测试中，只有方法3能处理干预后的查询：比方法2高出16.3个百分点，配对95%置信区间 [+7.1, +26.4] 个百分点，在时间尺度和操作模式类别上达到100%。在AI2推理挑战（ARC）中，使用OpenBookQA事实语料库，同样的选择性检索机制达到79%，而无限制的Llama-3.1-8B为76%，全参数注入为74%，这是一个+3个百分点的领域外复制结果，说明该现象并非废水处理特有。我们首次提供了单模拟器下实时工具使用、静态参数注入以及学习数值参数检索的工业因果问答比较。

LLM Analysis

Q: 这篇论文试图解决什么问题？

这篇论文致力于解决工业过程控制领域中，大语言模型（LLMs）缺乏特定工厂因果动力学 grounding 的问题。具体而言，论文聚焦于污水处理厂的决策支持场景，针对以下核心挑战：

1. 通用 LLM 与工业因果推理之间的鸿沟

现有 LLM 的预训练知识源于互联网文本的统计规律，无法捕捉特定工厂内部的非线性生化路径、变量耦合关系及效应传播时标。当操作员提出因果性问题（例如“ N_2O 为何上升？”或“曝气量削减 20% 将产生何后果？”）时，通用模型可能给出与特定工厂物理机制脱节的回答，而工业场景中的错误回答可能导致监管违规或危险排放事件。

2. 监督微调的“回忆–推理”权衡困境

默认的领域适配方式——在领域语料上进行监督微调（SFT）——存在系统性的权衡：模型参数化记忆工厂事实（如工况时标、耦合权重、阈值规则）的能力提升时，其基于这些事实进行组合推理的能力反而退化。论文指出，当存在架构可解释的工厂模拟器时，在推理时刻将 LLM 与模拟器动态对接，要优于通过 SFT 将模拟器输出吸收为模型参数。

3. 部署约束下的多模态 grounding 策略缺失

工业现场存在差异化的运行约束：有的环境可实时运行模拟器（控制室），有的环境需完全离线（气隙/边缘侧），还有的场景需跨工厂快速迁移。论文指出，业界缺乏在同一基准、同一基础模型、同一模拟器后端下，对以下三种 grounding 模式进行系统性比较的研究：

实时模拟器工具调用（Live Simulator Oracle）
结构化参数注入（Structured Parameter Injection）
解耦回忆–推理检索架构（Decoupled Recall–Reasoning, DRR）

4. 反事实查询的架构性瓶颈

静态知识库（无论是微调后的参数还是注入的静态参数表）无法回答需要干预后参数的“如果…会怎样”类问题，因为干预后的工况、耦合矩阵和时标均不同于基态。论文试图通过引入可动态调用模拟器后端的检索架构，解决传统静态方法在反事实推理上的结构性失效。

总结

论文试图通过构建并比较上述三种 grounding 模式，在保持 LLM 基座冻结的前提下，建立一条从“实时高依赖、高准确率”到“离线轻量、可跨厂迁移”的部署阶梯（deployment ladder），使得水务运营方可根据自身的模拟器可用性、网络隔离要求和延迟约束，选择最匹配的因果决策支持方案。

Q: 有哪些相关研究？

Authors: Gary Simethy, Daniel Ortiz Arroyo, Petar Durdevic

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2608.05151.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05151

Published: 2026-08-07T01:53:26.082Z

2. Mean-Field Dynamics of Chain-of-Thought Reasoning in Large Language Models

Abstract:Large language models (LLMs) with chain-of-thought reasoning have been widely applied in recent years, and theoretical explanations of their behavior may help deepen our understanding and guide model optimization. In this study, we introduce a framework that seeks statistical regularities and theoretical interpretations in LLM reasoning without simplifying the model architecture or making analogies to existing physical systems. We formulate LLM reasoning as a guided discovery process on a clue graph, and derive a one-dimensional ordinary differential equation for the fraction of discovered clues using the mean-field approximation. Experimentally, clue tokens are identified using the normalized surprisal of a student LLM on the outputs of a teacher LLM, and statistical regularities are obtained by averaging over many reasoning chains of thought. Our experiments show that the resulting statistical regularities are reproducible within the same dataset and can be fitted by the solving the proposed theoretical equation.

中文摘要

摘要：近年来，具有链式思维推理能力的大型语言模型（LLMs）得到了广泛应用，对其行为的理论解释可能有助于加深我们的理解并指导模型优化。在本研究中，我们介绍了一个框架，该框架在不简化模型架构或类比现有物理系统的情况下，寻求LLM推理中的统计规律性和理论解释。我们将LLM推理表述为在线索图上的引导发现过程，并使用平均场近似推导出一个描述已发现线索比例的一维常微分方程。在实验中，线索标记是通过在教师LLM的输出上使用学生LLM的归一化惊讶度来识别的，并通过对多个思维链条的平均获得统计规律。我们的实验表明，得到的统计规律在同一数据集中是可重复的，并且可以通过求解所提理论方程进行拟合。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Waiting failed: 30000ms exceeded

Authors: Hao Ai

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2608.05152.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05152

Published: 2026-08-07T01:53:26.082Z

3. Universal Pathologies, Conditional Consequences: A Triple-Robustness Analysis of RAG for Multi-Hop Traceability

Abstract:GraphRAG underperforms vector RAG on citation precision in many reports, but where and why have remained corpus-bound. We present a triple-robustness analysis that holds the retrieval architecture fixed and varies three orthogonal axes embedder (local e5-small -> Azure text-embedding-3-small), corpus (DO-178C typed-edge requirements -> Wikipedia paragraph chains via MuSiQue), and judge (paired GPT-5.4 x GPT-4.1) across 4,440 main-matrix runs, 600 cross-corpus runs, and 1,200 paired faithfulness judgments. (C2a) Over-citation is architecturally universal: GraphRAG emits 11-15 IDs per answer at citation precision 0.12-0.23 and retrieval recall 0.68-0.87 across all three settings. (C2b) Its faithfulness consequence is corpus-conditional: in typed-edge DO-178C, GraphRAG faithfulness collapses 74%->40% across hops; on Wikipedia chains the same pipeline rises 42%->58% because over-cited paragraphs remain topically supporting. (C1) Stratum-conditional winners are corpus-conditional but embedder-robust: vanilla wins 2-hop on DO-178C, GraphRAG wins 2-hop on MuSiQue, identical under either embedder. (C3) Single-judge LLM faithfulness is fragile to retrieval state: same-judge self-kappa across embedders is 0.137 for GPT-5.4 (verdict change on 41% of items). A learned router on dense embeddings alone reaches macro-F1 0.86 on hop classification (C4). We argue triple-robustness is the minimum bar for trustworthy RAG architecture claims.

中文摘要

摘要：在许多报告中，GraphRAG 在引用精确度上不如向量 RAG，但它表现不佳的具体位置和原因仍局限于语料库。我们提出了一种三重稳健性分析方法，该方法固定检索架构，变化三个正交轴：嵌入器（local e5-small -> Azure text-embedding-3-small）、语料库（DO-178C 类型化边缘需求 -> 通过 MuSiQue 的维基百科段落链）和评审者（成对的 GPT-5.4 x GPT-4.1），分别进行了 4,440 次主矩阵运行、600 次跨语料库运行，以及 1,200 次成对的忠实性判断。(C2a) 过度引用在架构上是普遍存在的：GraphRAG 在三种设置下每个答案发出 11-15 个 ID，引用精度为 0.12-0.23，检索召回率为 0.68-0.87。(C2b) 它的忠实性后果受语料库条件影响：在类型化边缘 DO-178C 中，GraphRAG 的忠实性在多跳间下降 74%->40%；而在 Wikipedia 链上，相同管道的忠实性上升 42%->58%，因为被过度引用的段落仍然在主题上提供支持。(C1) 分层条件的赢家受语料库影响但嵌入器稳健：在 DO-178C 上 vanilla 赢得 2 跳任务，GraphRAG 赢得 MuSiQue 上的 2 跳任务，无论使用哪种嵌入器结果相同。(C3) 单一评审者的 LLM 忠实性对检索状态敏感：同一评审者在不同嵌入器上的自我 kappa 值为 0.137（41% 项目的判决发生变化）。仅在密集嵌入上训练的路由器在跳数分类上达到宏 F1 为 0.86。(C4) 我们认为，三重稳健性是可信 RAG 架构主张的最低门槛。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Waiting failed: 30000ms exceeded

Authors: Meftun Akarsu, Burak Ozdemir

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2608.05153.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05153

Published: 2026-08-07T01:53:26.082Z

4. RIG-RoPE: Relation- and Instance-Gated Rotary Positional Encoding with Duration-Aware Temporal Coordinates

Abstract:Rotary positional encoding (RoPE) is a core component of modern language models and has been extended to multimodal LLMs through multidimensional variants such as multimodal RoPE (M-RoPE), which split positional channels into temporal, height, and width subspaces. This report identifies two limitations of static multidimensional position assignment in interleaved multimodal contexts. First, height/width rotations may be applied to token pairs whose spatial displacement is not a well-defined geometric object, producing cross-modal and inter-instance spatial interference. Second, temporal coordinates are often treated as equal-step counters, so a text token, an image block, and a video segment can advance the temporal phase by comparable amounts despite different information density. We propose RIG-RoPE, a relation- and instance-gated RoPE mechanism with duration-aware temporal coordinates. RIG-RoPE augments each token with a modality indicator, a visual instance identifier, and a scalar information-duration coordinate. It enables H/W rotations only for query-key pairs from the same visual instance; otherwise the unknown spatial displacement is marginalized rather than set to zero. Temporal rotations use interpolated cumulative block durations: text tokens consume unit duration, images use a dimension-aware logarithmic spatial scale, and videos further apply a logarithmic temporal extension over effective frames. We provide a gauge-invariance argument for avoiding ordinary cross-instance spatial rotation, an impossibility result for static IDs under shared H/W subspaces, and a duration-consistency argument against equal-step multimodal time. RIG-RoPE adds no learned parameters and can be implemented inside tiled attention kernels with constant additional metadata per token. This preliminary report establishes the formulation and validation path without claiming empirical superiority.

中文摘要

摘要：旋转位置编码（RoPE）是现代语言模型的核心组件，并已通过多维变体（如多模态 RoPE（M-RoPE））扩展到多模态大语言模型，这些变体将位置通道拆分为时间、高度和宽度子空间。本报告指出在交错多模态上下文中静态多维位置分配的两个局限性。首先，高度/宽度旋转可能会应用于空间位移不构成明确定义几何对象的令牌对，从而产生跨模态和跨实例的空间干扰。其次，时间坐标通常被视为等步计数器，因此文本令牌、图像块和视频片段可能会以相当的量推进时间相位，尽管信息密度不同。我们提出了 RIG-RoPE，一种带有持续时间感知时间坐标的关系和实例门控 RoPE 机制。RIG-RoPE 为每个令牌增加了模态指示器、视觉实例标识符以及标量信息-持续时间坐标。它仅对来自同一视觉实例的查询-键对启用 H/W 旋转；否则未知的空间位移被边际化处理而不是置零。时间旋转使用插值的累计块持续时间：文本令牌消耗单位持续时间，图像使用维度感知的对数空间尺度，视频进一步在有效帧上应用对数时间延展。我们提供了避免常规跨实例空间旋转的规约不变性论证，对共享 H/W 子空间下静态 ID 的不可能性结果，以及反对等步多模态时间的持续时间一致性论证。RIG-RoPE 不增加任何学习参数，并且可以在平铺注意力内核中实现，每个令牌仅需常量额外元数据。本初步报告建立了公式和验证路径，但未声称在实验上具有优越性。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Waiting failed: 30000ms exceeded

Authors: Donggen Li

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2608.05154.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05154

Published: 2026-08-07T01:53:26.082Z

5. Beyond Sentiment: Comparing Traditional NLP and LLM-Based Multi-Dimensional Analysis for Political News Evaluation

Abstract:Traditional sentiment analysis (SA) models, while effective for polarity classification, provide limited insight into the rhetorical, ideological, and framing dimensions of political discourse — dimensions that are central to research in the social sciences and humanities (SSH). In this paper, we present a comparative study of RoBERTa-based sentiment analysis and an LLM-based multi-dimensional framing analysis platform applied to a corpus of 50 political news articles from 17 international media outlets. The results reveal a critical limitation we term neutral collapse: RoBERTa classifies 70% of articles as neutral, effectively flattening substantively rich political content into an analytically uninformative category. We find that 23% of neutral-classified articles exhibit negative probability scores above 0.30. By contrast, the LLM-based approach captures political bias direction and intensity, sensationalism, emotional appeal, and political framing — yielding multi-dimensional analytical outputs aligned with SSH epistemologies. We argue that for political media analysis, traditional SA alone is insufficient, and that LLM-based multi-dimensional frameworks offer a more epistemologically adequate computational lens for SSH research needs.

中文摘要

摘要：传统的情感分析（SA）模型在极性分类方面虽然有效，但在政治话语的修辞、意识形态和框架维度上的洞察有限——这些维度是社会科学和人文学科（SSH）研究的核心。在本文中，我们展示了一项将基于RoBERTa的情感分析与基于大型语言模型（LLM）的多维度框架分析平台进行比较的研究，该研究应用于来自17个国际媒体的50篇政治新闻文章的语料库。结果揭示了一个关键的限制，我们称之为“中性崩塌”：RoBERTa将70%的文章分类为中性，从而将实质内容丰富的政治内容有效地压平为分析上缺乏信息的类别。我们发现，在被分类为中性的文章中，有23%的文章负向概率分数超过0.30。相比之下，基于LLM的方法能够捕捉政治倾向方向和强度、耸人听闻性、情感诉求以及政治框架——产生与SSH认知论相一致的多维分析输出。我们认为，对于政治媒体分析，单纯的传统情感分析是不够的，而基于LLM的多维框架提供了更为认知论上适宜的计算视角，以满足SSH研究的需求。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Waiting failed: 30000ms exceeded

Authors: Maryam Fooladi, Federico Bottino

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2608.05155.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05155

Published: 2026-08-07T01:53:26.082Z

6. Scaffold-Mediated Post-Training: Co-Evolving Model Parameters and Procedural Scaffold Graphs

Abstract:Post-training of large language models optimizes only parameters, while inference-time procedural scaffolds are typically designed independently of parameter training. This disconnect makes it difficult to automatically acquire and internalize complex strategies. We propose scaffold-mediated post-training: procedural scaffolds are organized into an evolvable graph structure that co-evolves with model parameters through discovery, distillation, and dynamic recompilation. We instantiate this paradigm as Skill Training. On FeatureBench, automatically discovered skills improve the passed rate by 8.1pp, and after progressive distillation the model still achieves a 27.7% passed rate without any external scaffold (distillation retention rate 85.2%, defined as post-distillation / with-skill passed rate), significantly outperforming standard SFT on the same data.

中文摘要

摘要：大型语言模型的后训练只优化参数，而推理时的程序支架通常是独立于参数训练设计的。这种脱节使得自动获取和内化复杂策略变得困难。我们提出了基于支架的后训练：程序支架被组织成一个可进化的图结构，通过发现、蒸馏和动态重新编译与模型参数共同进化。我们将这一范式实例化为技能训练。在 FeatureBench 上，自动发现的技能将通过率提高了 8.1 个百分点，并且经过逐步蒸馏后，模型在没有任何外部支架的情况下仍实现了 27.7% 的通过率（蒸馏保留率 85.2%，定义为蒸馏后通过率 / 有技能通过率），显著优于在相同数据上的标准 SFT。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Waiting failed: 30000ms exceeded

Authors: Fei Ding, Yongkang Zhang, Runhao Liu, Yuhao Liao, Zijian Zeng, Huiming Yang

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2608.05156.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05156

Published: 2026-08-07T01:53:26.082Z

Abstract:Double blind peer review serves as the scientific community primary defense against status and affiliation bias. Its effectiveness rests on the assumption that anonymized manuscripts convey scientific merit without revealing their authors. While authorship can often be recovered using citation networks or stylistic markers, we show that this assumption is increasingly fragile in the presence of large language models (LLMs). Using only titles and abstracts from papers published after model training, we find that LLMs collapse anonymity more efficiently than humans, with belief concentrating onto a small subset of plausible authors drawn from pools of five domain expert candidates. This vulnerability persists even when stylistic and bibliographic cues are excluded, indicating that stable patterns in problem framing and research focus function as latent conceptual signatures of authorship. Together, these findings indicate that double blind review is vulnerable to automated semantic inference, necessitating a revaluation of how anonymity and fairness are maintained in an AI augmented research ecosystem.

中文摘要

摘要：双盲同行评审是科学界抵御地位和隶属偏见的主要防线。其有效性依赖于这样一种假设：匿名稿件能够传达科学价值而不透露作者信息。虽然通常可以通过引用网络或风格特征来追溯作者身份，但我们表明，在大型语言模型（LLM）的存在下，这一假设日益脆弱。仅使用模型训练后发表论文的标题和摘要，我们发现大语言模型在揭示匿名性方面比人类更有效，信念集中于从五名领域专家候选人中选出的少数合理作者。即便在排除风格和书目线索的情况下，这种脆弱性依然存在，这表明问题表述和研究重点中的稳定模式可以作为潜在的作者概念特征。综上所述，这些发现表明双盲评审容易受到自动语义推理的影响，因此需要重新评估在人工智能增强的研究生态系统中如何维护匿名性和公平性。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Bulambo Mwendelwa Gloire, Prasenjit Mitra

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2608.05157.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05157

Published: 2026-08-07T01:53:26.082Z

8. Safe Evolution with Circuit Anchors

Abstract:In biological evolution, unconstrained mutation can lead to catastrophic outcomes: organisms may evolve enhanced capabilities while losing essential functions for survival. Nature’s solution is \textit{developmental constraints}, where core regulatory genes remain anchored while peripheral genes adapt freely. We observe that current self-evolution algorithms for large language models lack analogous constraints. They optimize purely for capability, implicitly assuming safety will be preserved. Our experiments reveal this assumption to be dangerously wrong: models can \textit{misevolve} into powerful yet dangerous entities. Inspired by how Hox genes anchor body structure across $500$ million years of evolution, we propose \textbf{Circuit-Anchored Evolution (CAE)}. Using mechanistic interpretability, we identify a tiny \textit{safety circuit}, comprising less than $2$\% of model features, that causally mediates safety behaviors. We anchor this circuit during evolution, constraining it within a small displacement bound while allowing the remaining features to evolve freely. This mirrors the biological principle of \textit{evolvability with constraint}: preserving what is essential while adapting what is peripheral. Experiments across $3$ model families and two evolution algorithms demonstrate that CAE achieves superior safety preservation with minimal capability loss, substantially outperforming explicit reward-based constraints in both effectiveness and efficiency. Just as developmental constraints prevent biological evolution from producing nonviable organisms, circuit anchoring prevents model evolution from producing capable but dangerous systems.

中文摘要

摘要：在生物进化中，不受限制的突变可能导致灾难性后果：生物体可能进化出增强的能力，同时失去生存所需的关键功能。自然界的解决方案是\textit{发育限制}，核心调控基因保持锚定，而外围基因自由适应。我们观察到当前大型语言模型的自我演化算法缺乏类似的约束。他们纯粹优化能力，隐含地假设安全会得到维护。我们的实验揭示了这一假设极其错误：模型可以\textit{误进化}成强大但危险的实体。受到Hox基因如何在5亿美元进化中锚定身体结构的启发，我们提出了\textbf{回路锚定进化（CAE）}。利用机制解释性，我们识别出一个极小的\textit{安全电路}，其模型特征比例不到2美元/%，它在因果上介导安全行为。我们在演化过程中锚定该回路，将其限制在一个小的位移界限内，同时允许其余特征自由演化。这反映了生物学上的\textit{约束可进化性}原则：保留本质，同时适应边缘。在3美元模型家族和两种进化算法中的实验表明，CAE在安全性保持上实现了更优越的安全性，能力损失极小，在效能和效率上远远优于显式基于奖励的约束。正如发育限制阻止生物进化产生非存活生物一样，回路锚定防止模型进化产生有能力但危险的系统。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Yan Liu, Jie Fu, Tsung-Yi Ho

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2608.05158.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05158

Published: 2026-08-07T01:53:26.082Z

9. SemiAdapt-Instruct: Extensible Instruction Tuning via Latent Domain-Specialised Adapters

Abstract:Instruction-tuned LLMs are deployed into environments where domains evolve, yet extending a fine-tuned model’s capabilities without full retraining remains an unsolved practical challenge. We present SemiAdapt-Instruct, a modular framework that discovers latent instruction domains, trains per-domain LoRA adapters in parallel, and performs parameter-free routing, incorporating new domains via single-adapter training without modifying existing components. SemiAdapt-Instruct outperforms full model fine-tuning across all configurations on both ROUGE-L and LLM-as-a-judge evaluation, while matching single LoRA fine-tuning and delivering extensibility that monolithic approaches cannot provide. We empirically demonstrate this extensibility by showing that updating a single adapter with new domain data outperforms all monolithic baselines. Our study also finds that independent discovery methods converge on the same specialisation-friendly domains. These findings demonstrate that decomposing heterogeneous instruction data into latent domains enables extensible NLP systems where evolving domains require only targeted single-adapter updates, eliminating the need for full model retraining.

中文摘要

摘要：经过指令调优的大语言模型（LLM）被部署到领域不断发展的环境中，但如何在不进行完整再训练的情况下扩展微调模型的能力仍是一个尚未解决的实际挑战。我们提出了SemiAdapt-Instruct，这是一种模块化框架，可发现潜在的指令领域，按领域并行训练LoRA适配器，并执行无参数路由，通过单适配器训练引入新领域而无需修改现有组件。SemiAdapt-Instruct在所有配置下，在ROUGE-L和LLM作为评审指标的评估中均优于完整模型微调，同时与单LoRA微调相当，并提供了单一方法无法实现的可扩展性。我们通过实验证明了这种可扩展性——更新单个适配器以使用新领域数据的性能优于所有整体基线。我们的研究还发现，独立的发现方法都收敛到相同的利于专业化的领域。这些发现表明，将异构指令数据分解为潜在领域能够实现可扩展的NLP系统，在不断发展的领域中，只需针对性地更新单个适配器，而无需进行完整模型再训练。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Josh McGiff, Salma Mekaoui, Robert Shanahan, Nikola S. Nikolov

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2608.05161.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05161

Published: 2026-08-07T01:53:26.082Z

10. PoolBench: A Benchmark for Pooling Strategies in Concept Representation Evaluation for Decoder-Only LLMs

Abstract:Pooling is a consequential but under-examined design choice in decoder-only concept representation work: practitioners must collapse token-level hidden states into a passage-level vector, yet no shared protocol exists for comparing this choice across concepts, models, and tasks. Reported gains are confounded by simultaneous changes in dataset, layer, construction method, and pooling rule, making principled decisions impossible. We introduce PoolBench, a benchmark that isolates pooling as the experimental variable under a fixed evaluation protocol. PoolBench covers 17 concepts, 19 pooling strategies, and 3 open-weight decoder-only models (Llama-3.1-8B, Gemma-2-9B, Mistral-7B), evaluated on a single audited corpus of 37,693 real-text passages. The primary axis is linear separability (D1/AUROC); steered concept prevalence (D2/SCP) and output-level disentanglement (D3) serve as diagnostic axes. The primary finding is decisive: W4_hierarchical reaches a cross-model mean AUROC of 0.7799, while the widely adopted P1_last_token baseline reaches only 0.7640 and is statistically significantly worse (Friedman+Nemenyi, p = 2.0e-36; 77 significant pairs among 18 effective strategies). Rankings are stable across layers (rho = 0.961—0.990). A key negative result: strong detection does not imply strong steering — D2 and D3 are substantially weaker than D1 for most concepts, indicating a fundamental representational limit rather than a pooling failure. On mid-difficulty concepts, W4_hierarchical outperforms P1_last_token by 0.042—0.113 AUROC; construction method choice (DiffMean vs. REPE) has a larger effect (delta AUROC 0.15) than pooling (delta AUROC 0.016), establishing the correct practical hierarchy. We release the corpus, pre-extracted activations, scorer models, steering vectors, and evaluation code as a reusable protocol for pooling research.

中文摘要

摘要：在仅解码器的概念表示工作中，池化是一项重要但研究不足的设计选择：从业者必须将令牌级隐藏状态压缩为段落级向量，但在不同概念、模型和任务之间尚无统一的比较协议。已报告的增益由于数据集、层数、构建方法和池化规则的同时变化而存在混淆，使得做出有原则的决策变得不可能。我们引入了PoolBench，这是一个在固定评估协议下将池化作为实验变量隔离的基准。PoolBench涵盖17个概念、19种池化策略以及3个开源权重的仅解码器模型（Llama-3.1-8B、Gemma-2-9B、Mistral-7B），在一个经过审计的包含37,693篇真实文本段落的语料库上进行评估。主要评估轴是线性可分性（D1/AUROC）；受控概念分布（D2/SCP）和输出级解缠（D3）作为诊断轴。主要发现具有决定性：W4_hierarchical在跨模型平均AUROC达到0.7799，而广泛使用的P1_last_token基线仅为0.7640，并且统计显著更差（Friedman+Nemenyi, p = 2.0e-36；在18种有效策略中有77对显著差异）。排名在各层间稳定（rho = 0.961—0.990）。一个关键的负面结果是：强检测能力并不意味着强引导能力——对于大多数概念，D2和D3明显弱于D1，这表明这是一个基本的表示能力限制，而非池化失败。在中等难度概念上，W4_hierarchical的AUROC比P1_last_token高0.042—0.113；构建方法选择（DiffMean vs. REPE）的影响（AUROC差值0.15）大于池化（AUROC差值0.016），确立了正确的实际优先级。我们发布了语料库、预提取的激活、评分模型、引导向量和评估代码，作为池化研究的可重复使用协议。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Ayushi Agarwal

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2608.05162.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05162

Published: 2026-08-07T01:53:26.082Z

Agent Domain Papers

1. Agentic Nesting: A New Methodology for Existing Enterprise Application Integration and Services

Abstract:Enterprise operations extensively rely on multiple heterogeneous business systems and information applications, which also result in severe data silos and process fragmentation. Enterprises have invested considerable financial and material resources in building these applications, however, effectively leveraging and orchestrating them remains a formidable challenge. Conventional approaches to enterprise application integration, encompassing middleware architectures such as Enterprise Service Bus (ESB), API gateway infrastructures, and Robotic Process Automation (RPA), suffer from inherent limitations like high architectural coupling, escalating operation and maintenance costs, and limited intelligence capabilities. This paper proposes Agentic Nesting, a multi-agent collaboration framework in which existing enterprise applications are encapsulated as autonomous AI agents within a hierarchically nested structure. Rather than flat interconnection, agents are organized into layered stewardship topologies that mirror the compositional complexity of enterprise ecosystems. The framework extracts a digital agent proxy from each legacy application to enable natural-language interaction and autonomous manipulation, coordinates multiple agents through a central orchestrator for task decomposition and dynamic dispatching, and exposes a unified conversational interface for cross-application querying and process orchestration. The main contributions of this paper are the proposition of the “Application-as-Agent” integration paradigm and the “Conversation-as-Integration” interaction philosophy, together with an exploration of the generalization potential of this methodology in scenarios encompassing heterogeneous system coordination, and large-scale data applications.

中文摘要

摘要：企业运营在很大程度上依赖于多个异构的业务系统和信息应用，这也导致了严重的数据孤岛和流程碎片化。企业在构建这些应用方面投入了大量的财力和物力，但如何有效利用和协调它们仍然是一项艰巨的挑战。传统的企业应用集成方法，包括企业服务总线（ESB）、API网关基础设施和机器人流程自动化（RPA）等中间件架构，存在固有的局限性，如高架构耦合度、不断上升的运维成本以及有限的智能能力。本文提出了“智能代理嵌套”（Agentic Nesting）——一种多智能体协作框架，其中现有的企业应用被封装为位于分层嵌套结构内的自主AI代理。与平面互联不同，这些代理被组织为分层管理拓扑，以镜像企业生态系统的组合复杂性。该框架从每个传统应用中提取数字代理代理，以支持自然语言交互和自主操作，通过中央协调器对多个代理进行任务分解和动态调度，并提供统一的会话接口实现跨应用查询和流程编排。本文的主要贡献包括提出“应用即代理”（Application-as-Agent）集成范式和“会话即集成”（Conversation-as-Integration）交互理念，并探索了该方法在异构系统协调及大规模数据应用场景中的泛化潜力。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Xi Wang, Kun Li, Xianyao Ling, Gang Yin, Liang Zhang, Jiang Wu, Wenbo Lei, Jun Xu, Annie Wang, Fu Zhang, Weizhe Wang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05159.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05159

Published: 2026-08-07T02:03:52.843Z

2. The Ignition Index: Measuring Global Workspace Dynamics in Language Models

Abstract:We introduce the Ignition Index (I), a validated scalar metric that operationalizes Global Workspace Theory’s (GWT) all-or-none ignition prediction in transformer language models. The metric fits a four-parameter sigmoid to per-layer linear probe accuracy as a function of input signal strength, extracting steepness parameter beta-hat: high values indicate abrupt, ignition-like transitions; low values indicate graded build-up. Across 11 models spanning five architecture families, shuffled-label controls demonstrate 9.6-fold selectivity for genuine linguistic structure over spurious probe capacity (p < 0.001, Mann-Whitney U-test). We find: (1) Feedforward transformers exceed SSMs by 89% in aggregate beta-hat (p < 1e-13, Cohen’s d = 0.52), with Mamba exhibiting near-linear profiles consistent with absent global broadcast. (2) Huginn-3.5B exhibits 2.12-fold higher ignition along its iteration axis than its depth axis, demonstrating that recurrent architectures manifest workspace-like transitions along the recurrence dimension. (3) Pythia-410M shows a PELT-detected phase transition at training step 256 (+67%), preceding induction-head formation. (4) Hypotheses linking ignition to model scale and signal strength were not confirmed, suggesting transformer architectures may saturate available ignition mechanisms. The Ignition Index provides the first validated quantitative bridge between GWT’s dynamical predictions and mechanistic interpretability, with 9.6-fold measurement selectivity and architecture-level discriminability not previously characterized in the scaling literature. Code: this https URL

中文摘要

摘要：我们引入了点火指数（Ignition Index, I），这是一个经过验证的标量指标，用于在变压器语言模型中实现全球工作空间理论（GWT）“全有或全无”点火预测的操作化。该指标通过对每一层线性探针准确率与输入信号强度的关系拟合四参数S形曲线，提取斜率参数β̂：高值表示突发的、类似点火的转变；低值表示渐进的构建。在涵盖五类架构的11个模型中，标签打乱的对照实验显示对真实语言结构的选择性是对伪探针能力的9.6倍（p < 0.001, Mann-Whitney U检验）。我们的发现包括：(1) 前馈变压器在总体β̂上比SSM高出89%（p < 1e-13, Cohen’s d = 0.52），Mamba表现出接近线性的曲线，与缺乏全局广播一致。(2) Huginn-3.5B在迭代轴上的点火率比在深度轴上高2.12倍，表明循环架构在循环维度上表现出类似工作空间的转变。(3) Pythia-410M在训练步256显示PELT检测到的相变（+67%），早于归纳头的形成。(4) 将点火与模型规模和信号强度相关的假设未得到验证，表明变压器架构可能已饱和可用的点火机制。点火指数首次提供了GWT动力学预测与机制可解释性之间的经过验证的定量桥梁，具有9.6倍的测量选择性和架构级可分辨性，在之前的规模化文献中尚未表征。代码：https URL

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Saman Rahbar

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05160.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05160

Published: 2026-08-07T02:03:52.843Z

3. Woodpecker Distillation: Weak Models Diagnose Reasoning Bugs in Strong Models

Abstract:Large language models often fail on reasoning tasks despite possessing the capability to solve them. We argue that many such failures arise from localized reasoning bugs in intermediate steps rather than from global incompetence. We show that these bugs are frequently repairable: inserting a short patch generated by a weak probe model after the same strong-model reasoning prefix can redirect the trajectory toward a correct solution. However, this corrective effect is not reliably internalized by directly fine-tuning on weak patches or repaired trajectories, suggesting that the useful signal lies not in the intervention text itself, but in how it reshapes the model’s future reasoning distribution. We therefore propose Woodpecker Distillation, a weak-to-strong training framework that learns from contrastive local interventions. Our method contrasts successful and unsuccessful weak-model patches at the same prefix, constructs a corrective teacher distribution from their induced future token predictions, and distills this signal into the strong model. Experiments on mathematical reasoning benchmarks show that Woodpecker Distillation consistently improves strong-model performance and outperforms direct imitation baselines.

中文摘要

摘要：尽管大型语言模型具备解决推理任务的能力，但它们在推理任务中经常失败。我们认为，许多此类失败源于中间步骤中的局部推理错误，而非整体能力不足。我们展示了这些错误通常是可修复的：在相同强模型的推理前缀之后插入由弱探测模型生成的短修补，可以将推理轨迹引导向正确的解决方案。然而，通过直接对弱补丁或修复后的轨迹进行微调，这种纠正效应并不能可靠地被模型内化，这表明有用的信号不在于干预文本本身，而在于它如何重塑模型未来的推理分布。因此，我们提出了啄木鸟蒸馏，一种弱到强的训练框架，从对比局部干预中学习。我们的方法在相同前缀下对比强弱模型的成功与失败补丁，根据它们引导的未来令牌预测构建纠正性教师分布，并将该信号蒸馏到强模型中。在数学推理基准测试上的实验表明，啄木鸟蒸馏能够持续提升强模型性能，并优于直接模仿的基线方法。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Dayu Wang, Jiaye Yang, Weikang Li, Jiahui Liang, Yang Li, Deguo Xia, Jizhou Huang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05168.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05168

Published: 2026-08-07T02:03:52.843Z

4. From Continuous Predictors to Clinical Thresholds: Early Evidence on Performance Trade-offs of Guideline-Based Categorisation for Ischaemic Stroke Outcome Prediction

Abstract:Machine learning models achieve strong predictive accuracy for 90-day outcome prediction in acute ischaemic stroke, yet clinical adoption is limited by the misalignment of model explanations with clinicians’ reasoning. Motivated by a clinician user study calling for clinical guideline-aligned cut-offs, we ask whether continuous predictors can be replaced by clinically informed categorical encodings without sacrificing performance. On a multi-centre European registry stratified into three treatment cohorts, we compare standard and fully categorised gradient-boosted models, the latter using stroke guideline-aligned, treatment-specific thresholds. The fully categorised models are statistically indistinguishable from their continuous counterparts in two of the treatment cohorts, with a significant drop in predictive accuracy in one cohort. Global feature importance rankings remain consistent, suggesting that discretising continuous predictors into guideline-based categories preserves the core hierarchy of prognostic factors across all treatment groups. Guideline-based categorisation is thus a viable design choice for stroke-outcome models.

中文摘要

摘要：机器学习模型在急性缺血性脑卒中90天预后预测中取得了很高的预测准确性，但临床应用受到模型解释与临床医生推理不一致的限制。受一项临床用户研究的启发，该研究呼吁采用与临床指南一致的阈值，我们探讨了是否可以在不牺牲性能的情况下，用临床知情的分类编码替代连续预测变量。在一个多中心欧洲登记数据库中，我们将患者按三种治疗方式分层，比较标准和完全分类的梯度提升模型，后者使用与卒中指南一致的、针对特定治疗的阈值。在两个治疗队列中，完全分类模型在统计学上与其连续变量对应模型无显著差异，而在一个队列中预测准确性显著下降。全局特征重要性排名保持一致，表明将连续预测变量离散化为基于指南的类别能够保留所有治疗组的预后因素核心层级。因此，基于指南的分类是卒中预后模型的一种可行设计选择。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Esra Zihni, Katryna Cisek, Hamzah Ziadeh, Hendrik Knoche, Robert Mikulik, John D. Kelleher

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05203.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05203

Published: 2026-08-07T02:03:52.843Z

5. SkillTrace: Multi-Trace Provenance Auditing for LLM-Agent Skill Reuse

Abstract:LLM-agent ecosystems are rapidly growing around reusable skills: mixed-modality packages of metadata, natural-language instructions, code, tools, references, and operational workflows. As skills become marketplace artifacts, auditing their reuse is no longer the same problem as ordinary code clone detection. Existing detectors target single-modality source code or whole-package similarity, yet skill reuse evidence is distributed across authored text, implementation fragments, and operational structure. As a result, they can miss reuse that preserves only one part of a skill. We present SKILLTRACE, a multi-trace provenance auditing framework for LLM-agent skill reuse. SKILLTRACE extracts three provenance traces: Expression, Implementation, and Operational. It represents the Operational Trace as a Skill Operational Graph (SOG) that captures activation, procedure, and resource-flow structure. An LLM assists only the Operational-trace extraction, once at ingestion; at audit time SKILLTRACE compares cached traces deterministically, calibrates each trace against same-function strict negatives, and reports which trace supports a reuse decision. On SKILLTRACE-BENCH, with 820 transformed reuse positives over 100 marketplace anchors and 751 negative controls, SKILLTRACE achieves AUROC 0.938 and F1 0.898. A 36,446-skill wild audit further shows that trace-attributed evidence surfaces actionable reuse review queues beyond repository-level baselines.

中文摘要

摘要：LLM代理生态系统正在围绕可重复使用的技能快速发展：这些技能是元数据、自然语言指令、代码、工具、参考资料和操作工作流的混合模态包。随着技能成为市场中的商品，对其重复使用进行审计不再是普通代码克隆检测的问题。现有的检测器主要面向单模态源代码或整个包的相似性，而技能重复使用的证据分布在创作文本、实现片段和操作结构中。因此，它们可能会遗漏仅保留技能某一部分的重复使用。我们提出了SKILLTRACE，一种用于LLM代理技能复用的多追踪溯源审计框架。SKILLTRACE提取三种溯源追踪：表达、实现和操作。它将操作追踪表示为技能操作图（SOG），捕捉激活、流程和资源流结构。LLM仅在数据摄取时协助操作追踪的提取；在审计时，SKILLTRACE对缓存的追踪进行确定性比较，将每个追踪与同功能的严格阴性样本校准，并报告哪种追踪支持复用决策。在SKILLTRACE-BENCH上，针对100个市场锚点的820个转换复用正例和751个负控样本，SKILLTRACE实现了AUROC 0.938和F1 0.898。对36,446个技能的野外审计进一步表明，基于追踪属性的证据能在仓库级基准之外提出可操作的复用审查队列。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Jialuo Chen, Minghe Wang, Lingqi Jiang, Jianan Ma, Xinhao Deng, Xiaohu Du, Ruixiao Lin, Yunhao Feng, Linkang Du, Jingyi Wang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05204.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05204

Published: 2026-08-07T02:03:52.843Z

6. Abstract Event Causal Rules: Induction and Application

Abstract:Event-centric intelligent analytical systems heavily depend on explicit causal event knowledge for risk early warning, decision-making support and narrative comprehension. Nevertheless, existing instance-level causal pairs suffer severe generalization deficits on low-frequency long-tail and unseen event combinations. To address this limitation, this work proposes Abstract Event Causal Rule (AECR), a novel relation-level causal abstraction paradigm that transforms concrete cause-effect pairs into generalized abstract causal logic while retaining their intrinsic causal relationships. We design a multi-agent Concrete-to-Abstract Causal Induction (CACI) system coupled with similarity-constrained clustering to distill trustworthy AECRs from noisy raw causal data, based on which two complete AECR knowledge bases are built. To validate the practical utility of abstract causal knowledge, we propose an Abstract Rule-Guided Causal Attention Encoder (AR-GCAE), which injects the retrieved AECRs into the causality Graph Event Prediction (CGEP) benchmark task via rule-guided attention layers and gated representation fusion. Quantitative experimental results reveal that applying AECRs substantially strengthens the generalization capacity of event causal reasoning and brings consistent performance improvements to event prediction, with the most prominent gains observed on rare and unseen event samples.

中文摘要

摘要：以事件为中心的智能分析系统在风险预警、决策支持和叙事理解中高度依赖明确的因果事件知识。然而，现有的实例级因果对在低频长尾事件和未见事件组合上存在严重的泛化不足。为了解决这一限制，本研究提出了抽象事件因果规则（AECR），这是一种新颖的关系级因果抽象范式，它将具体的因果对转化为通用的抽象因果逻辑，同时保留其内在的因果关系。我们设计了一个多智能体的具体到抽象因果归纳（CACI）系统，并辅以相似性约束的聚类，从嘈杂的原始因果数据中提炼出可信的AECR，在此基础上构建了两个完整的AECR知识库。为了验证抽象因果知识的实际效用，我们提出了一种抽象规则指导的因果注意编码器（AR-GCAE），通过规则引导的注意力层和门控表示融合，将检索到的AECR注入因果图事件预测（CGEP）基准任务。定量实验结果显示，应用AECR显著增强了事件因果推理的泛化能力，并为事件预测带来了持续的性能提升，其中稀有和未见事件样本的增益最为显著。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Ziwei Zheng, Peiqiong Chen, Bang Wang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05205.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05205

Published: 2026-08-07T02:03:52.843Z

7. Otter: A Time-Aware, History-Conditioned Human Chess AI

Abstract:Otter is a 15.3M-parameter human chess AI that predicts human move selection by modeling play as a time-aware, sequential process rather than treating each position in isolation. It combines two conditioning signals: (1) a move history encoder that conditions predictions on the last 20 moves, capturing opening preferences, positional drift, and intra-game behavioral tendencies; and (2) a time control module that modulates predictions based on clock pressure. Otter is trained on 6.1 billion positions from 117 million Lichess rapid games over 30 days on a single T4 GPU. Otter achieves 55.23% top-1 and 90.95% top-5 move-prediction accuracy, surpassing the prior state-of-the-art human chess model, Maia 2, with far fewer parameters and less training data. Across 11 Elo brackets (<1100 to >=2000), accuracy peaks at 57.38% in the 1900-1999 bracket. These results show that modeling chess as a time-aware, sequential activity yields more human-accurate move prediction than position-only approaches, using a smaller model. Code, trained models, and complete training logs are publicly released.

中文摘要

摘要：Otter 是一个拥有 1530 万参数的人类国际象棋 AI，通过将对局建模为时间感知的序列过程来预测人类的走法选择，而不是将每一个局面孤立处理。它结合了两个条件信号：(1) 走法历史编码器，根据最近 20 步走法对预测进行条件化，以捕捉开局偏好、局势漂移和对局内的行为倾向；(2) 时间控制模块，根据计时压力调节预测。Otter 在 30 天内使用单个 T4 GPU 在来自 1.17 亿局 Lichess 快棋的 61 亿个局面上进行训练。Otter 的顶步预测准确率（top-1）为 55.23%，前五步预测准确率（top-5）为 90.95%，在参数更少、训练数据更少的情况下，超越了先前最先进的人类国际象棋模型 Maia 2。在 11 个 Elo 等级组（<1100 到 >=2000）中，准确率在 1900-1999 等级组达到峰值 57.38%。这些结果表明，将国际象棋建模为时间感知的序列活动，比仅以局面为基础的方法更能准确预测人类走法，同时使用更小的模型。代码、训练好的模型和完整训练日志已公开发布。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Tarun Kumar S

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05206.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05206

Published: 2026-08-07T02:03:52.843Z

8. SearchAuditor: Auditing and Attributing Failures in Long-Horizon Search Agents

Abstract:Deep search agents tackle challenging questions through long-horizon web interactions, a process that is both complex and fragile: small reasoning errors may propagate through long, noisy trajectories into fluent but incorrect answers. Diagnosing such failures is difficult, requiring the manual inspection of extremely long execution traces, which could be beyond human capacity. We therefore introduce SearchAuditBench, a benchmark that evaluates whether LLM auditors can localize, attribute, and repair these failures, thereby reducing the human burden. SearchAuditBench comprises 1,243 failed trajectories, averaging 73.1 messages and 65.1K tokens, collected from eight open-weight models on five deep-search benchmarks, each expert-annotated with the critical error step, a search-specific root cause, and a reference repair with grading rubrics. We further propose SearchAuditor, a multi-perspective auditing framework that effectively localizes, attributes, and repairs search-agent failures through evidence-grounded adjudication. Experimental results show that even the strongest baseline, when powered by a frontier model like GPT-5.5, attains only a 26.6% end-to-end pass rate. In contrast, our SearchAuditor consistently outperforms all baselines across different frontier models, achieving an end-to-end pass rate of 32.3%, and resuming failed runs with its repairs enables agents to better recover from errors.

中文摘要

摘要：深度搜索代理通过长航程的网络交互来解决具有挑战性的问题，这一过程既复杂又脆弱：小的推理错误可能会通过冗长且嘈杂的轨迹传播，最终生成流畅但错误的答案。诊断此类失败非常困难，需要手动检查极长的执行轨迹，这可能超出人类能力。因此，我们提出了 SearchAuditBench，一个评估大型语言模型审计器能否定位、归因并修复这些失败的基准，从而减轻人类负担。SearchAuditBench 包含 1,243 条失败轨迹，平均 73.1 条消息和 65.1K 个令牌，汇集自五个深度搜索基准上的八个开源模型，每条由专家标注关键错误步骤、搜索特定根本原因及参考修复方案附带评分标准。我们进一步提出了 SearchAuditor，一种多视角审计框架，通过基于证据的裁决有效地定位、归因并修复搜索代理的失败。实验结果显示，即便是最强的基线模型在使用前沿模型如 GPT-5.5 时，也仅能达到 26.6% 的端到端通过率。相比之下，我们的 SearchAuditor 在不同前沿模型上始终优于所有基线，实现了 32.3% 的端到端通过率，并通过其修复恢复失败运行，使代理能够更好地从错误中恢复。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Zhixiang Liang, Yifei Liu, Yidan Huang, Haozhe Zhao, Beichen Huang, Jiaqi Wang, Nan Duan, Qiong Cao

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05212.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05212

Published: 2026-08-07T02:03:52.843Z

9. PD-GS: Phoneme-Driven 3DGS for Audio-Driven Talking Heads

Abstract:3D Gaussian Splatting (3DGS) enables fast, photorealistic talking-head rendering, yet accurate lip articulation remains elusive: mouth motion is often over-smoothed and may violate hard articulatory constraints such as bilabial closures, producing the notorious ``leaky mouth’’ artifact. A key difficulty is that brief, discrete articulatory events are inferred from a continuous acoustic embedding under a regression objective, which biases predictions toward averaged mouth configurations. While modern self-supervised speech encoders provide rich prosodic and phonetic cues, they do not provide an explicit, frame-aligned linguistic target that reliably disambiguates closure-level events. We propose \textbf{Phoneme-Driven Gaussian Splatting (PD-GS)}, which augments a 3DGS talker with time-aligned phoneme tokens obtained from an automatic ASR and forced-alignment pipeline. Our core component, the \textbf{Linguistic Fusion Module (LFM)}, adaptively fuses continuous audio context with discrete phoneme embeddings through a learned gate, allowing the model to preserve smooth audio-driven dynamics while strengthening phoneme guidance on articulation-critical segments. PD-GS is trained purely from monocular video using image reconstruction and lip landmark supervision. On HDTF, PD-GS achieves the best lip geometry among the compared baselines (LMD 2.66) and qualitatively reduces closure violations in challenging phoneme sequences, yielding more linguistically faithful neural avatars.

中文摘要

摘要：三维高斯点渲染（3D Gaussian Splatting, 3DGS）能够实现快速、逼真的人物头像渲染，但准确的唇部发音仍难以实现：口部动作常被过度平滑，并且可能违反诸如双唇闭合等严格的发音约束，从而产生著名的“漏嘴”现象。一个关键难点在于，短暂的离散发音事件需要通过回归目标从连续的音频嵌入中推断，而这会导致预测偏向于平均的口型配置。虽然现代自监督语音编码器提供了丰富的韵律和音素信息，但它们并未提供明确的、帧对齐的语言目标，以可靠地区分闭合级事件。我们提出了\textbf{基于音素的高斯点渲染（Phoneme-Driven Gaussian Splatting, PD-GS）}，该方法通过自动语音识别和强制对齐流程获得时间对齐的音素标记，从而增强3DGS人物说话模型。我们的核心组件\textbf{语言融合模块（Linguistic Fusion Module, LFM）}通过学习的门控机制，自适应地将连续音频上下文与离散音素嵌入融合，使模型在保持平滑的音频驱动动态的同时，加强在发音关键片段对音素的指导。PD-GS完全通过单目视频进行训练，使用图像重建和唇部关键点监督。在HDTF数据集上，PD-GS在比较基线模型中获得了最佳的唇部几何表现（LMD 2.66），并在具有挑战性的音素序列中定性减少了闭合违规情况，从而生成了更符合语言真实的神经化身。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Ao Fu, Yi Zhou

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05218.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05218

Published: 2026-08-07T02:03:52.843Z

10. When Privileged Guidance Misaligns: State-Matched Routing and Contextualized Self-Distillation for Multi-Turn Agents

Abstract:Privileged on-policy distillation provides dense supervision for multi-turn agents by allowing a synchronized teacher to re-score the student’s response at every turn with access to training-only references, such as successful trajectories. In interactive environments, however, the student’s preceding actions continually change the execution state. As the student takes different actions or completes subgoals in a different order, its rollout may reach states not covered by the reference, making the reference an unreliable source of guidance for the state actually reached. Applying privileged distillation indiscriminately therefore creates state—reference mismatch. This mismatch motivates a central objective: providing privileged reference guidance that remains compatible with the student’s current execution state. We introduce State-Matched Routing and Contextualized Self-Distillation (SMRC-SD), which explicitly determines when and how a privileged trajectory should guide an on-policy student. At each turn, SMRC-SD verifies whether the student’s current execution state matches a supported state along the reference trajectory. Distillation is applied only at matched states, filtering out turns for which the reference lacks locally compatible guidance. For each matched state, SMRC-SD further constructs state-conditioned teacher context from the successful trajectory, grounding supervision in the state actually reached. Across ALFWorld and WebShop, SMRC-SD consistently outperforms unconditional successful full-path distillation. With Qwen3-1.7B, it improves task success from $0.746$ to $0.865$ on ALFWorld and from $0.574$ to $0.693$ on WebShop. Controlled routing and context ablations support both selecting locally supported turns and constructing state-compatible teacher context as contributors to these gains. Code is available at this https URL.

中文摘要

摘要：特权的在策略蒸馏通过允许同步教师在每一步对学生的响应重新评分并使用仅训练时可用的参考（如成功轨迹）为多轮代理提供密集监督。然而，在交互环境中，学生之前的动作会不断改变执行状态。随着学生采取不同动作或以不同顺序完成子目标，其展开可能到达参考中未覆盖的状态，使得参考对于实际到达的状态来说成为不可靠的指导来源。因此，不加选择地应用特权蒸馏会造成状态与参考的不匹配。这种不匹配促使一个核心目标：提供与学生当前执行状态兼容的特权参考指导。我们提出了状态匹配路由和情境化自蒸馏（SMRC-SD），它明确确定特权轨迹应何时以及如何指导在策略学生。在每一步，SMRC-SD会验证学生当前执行状态是否与参考轨迹中的支持状态匹配。蒸馏仅在匹配状态下应用，过滤掉参考缺乏局部兼容指导的步骤。对于每个匹配状态，SMRC-SD进一步从成功轨迹中构建状态条件化的教师上下文，将监督基础建立在实际到达的状态上。在ALFWorld和WebShop中，SMRC-SD始终优于无条件的成功全路径蒸馏。使用Qwen3-1.7B，在ALFWorld上的任务成功率从$0.746$提升至$0.865$，在WebShop上从$0.574$提升至$0.693$。受控路由和上下文消融实验表明，选择局部支持的步骤和构建状态兼容的教师上下文均为这些提升的贡献因素。代码可在此https URL获取。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Junzhuo Liu, Weiwei Li, Jun Ling, Peng Wang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05219.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05219

Published: 2026-08-07T02:03:52.843Z

Evaluation Domain Papers

1. Agentic Nesting: A New Methodology for Existing Enterprise Application Integration and Services

中文摘要

摘要：企业运营在很大程度上依赖于多个异构的业务系统和信息应用，这也导致了严重的数据孤岛和流程碎片化。企业在构建这些应用方面投入了大量的财力和物力，但如何有效利用和协调它们仍然是一项艰巨的挑战。传统的企业应用集成方法，包括企业服务总线（ESB）、API网关基础设施和机器人流程自动化（RPA）等中间件架构，存在固有的局限性，如高架构耦合度、不断上升的运维成本以及有限的智能能力。本文提出了“智能代理嵌套”方案，这是一种多智能体协作框架，将现有的企业应用封装为层级嵌套结构中的自主AI代理。代理不是以平面互联方式组织，而是按层级管理拓扑组织，以反映企业生态系统的组合复杂性。该框架从每个遗留应用中提取数字代理代理，以实现自然语言交互和自主操作，通过中央协调器对多个代理进行任务分解和动态调度，并提供统一的会话接口，用于跨应用查询和流程编排。本文的主要贡献是提出了“应用即代理（Application-as-Agent）”的集成范式和“会话即集成（Conversation-as-Integration）”的交互理念，同时探索了该方法在异构系统协调和大规模数据应用场景中的通用化潜力。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Waiting failed: 30000ms exceeded

Authors: Xi Wang, Kun Li, Xianyao Ling, Gang Yin, Liang Zhang, Jiang Wu, Wenbo Lei, Jun Xu, Annie Wang, Fu Zhang, Weizhe Wang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05159.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05159

Published: 2026-08-07T02:09:58.830Z

2. The Ignition Index: Measuring Global Workspace Dynamics in Language Models

中文摘要

摘要：我们引入了点火指数（Ignition Index, I），这是一个经过验证的标量指标，用于在变压器语言模型中实现全球工作空间理论（GWT）“全有或全无”点火预测的操作化。该指标通过对每一层线性探针准确率与输入信号强度的关系拟合四参数S形曲线，提取斜率参数β̂：高值表示突发的、类似点火的转变；低值表示渐进的构建。在涵盖五类架构的11个模型中，标签打乱的对照实验显示对真实语言结构的选择性是对虚假探针能力的9.6倍（p < 0.001, Mann-Whitney U检验）。我们的发现包括：(1) 前馈变压器在总体β̂上比SSM高出89%（p < 1e-13, Cohen’s d = 0.52），Mamba表现出接近线性的曲线，与缺乏全局广播一致。(2) Huginn-3.5B在迭代轴上的点火率比在深度轴上高2.12倍，表明循环架构在循环维度上表现出类似工作空间的转变。(3) Pythia-410M在训练步256显示PELT检测到的相变（+67%），早于归纳头的形成。(4) 将点火与模型规模和信号强度相关的假设未得到验证，表明变压器架构可能已饱和可用的点火机制。点火指数首次提供了GWT动力学预测与机制可解释性之间的经过验证的定量桥梁，具有9.6倍的测量选择性和架构级的可区分性，这是以往扩展文献中未曾描述的。代码地址：此 https URL

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Saman Rahbar

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05160.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05160

Published: 2026-08-07T02:09:58.830Z

3. Woodpecker Distillation: Weak Models Diagnose Reasoning Bugs in Strong Models

中文摘要

摘要：尽管大型语言模型具备解决推理任务的能力，但它们在推理任务中经常失败。我们认为，许多此类失败源于中间步骤中的局部推理错误，而非整体能力不足。我们展示了这些错误通常是可修复的：在相同强模型的推理前缀之后插入由弱探测模型生成的短修补，可以将推理轨迹引导向正确的解决方案。然而，通过直接对弱补丁或修复后的轨迹进行微调，这种纠正效应并不能可靠地被模型内化，这表明有用的信号不在于干预文本本身，而在于它如何重塑模型未来的推理分布。因此，我们提出了啄木鸟蒸馏，一种从弱到强的训练框架，通过对比局部干预进行学习。我们的方法在相同前缀对成功与失败的弱模型补丁进行对比，从它们所引起的未来标记预测中构建纠正教师分布，并将此信号蒸馏至强模型。在数学推理基准上的实验表明，啄木鸟蒸馏能够持续提升强模型的性能，并优于直接模仿基线。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Dayu Wang, Jiaye Yang, Weikang Li, Jiahui Liang, Yang Li, Deguo Xia, Jizhou Huang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05168.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05168

Published: 2026-08-07T02:09:58.830Z

4. From Continuous Predictors to Clinical Thresholds: Early Evidence on Performance Trade-offs of Guideline-Based Categorisation for Ischaemic Stroke Outcome Prediction

中文摘要

摘要：机器学习模型在急性缺血性脑卒中90天预后预测中取得了很高的预测准确性，但临床应用受到模型解释与临床医生推理不一致的限制。受一项临床用户研究的启发，该研究呼吁采用与临床指南一致的阈值，我们探讨是否可以用临床知情的分类编码替代连续预测因子而不损失性能。在一个按三种治疗队列分层的多中心欧洲登记资料中，我们比较了标准与完全分类的梯度提升模型，后者使用与卒中指南一致的、针对特定治疗的阈值。结果显示，在两个治疗队列中，完全分类模型在统计上与其连续模型无显著差异，但在一个队列中预测准确性显著下降。全局特征重要性排名保持一致，这表明将连续预测因子离散化为基于指南的类别可以保留所有治疗组预后因素的核心层次结构。因此，基于指南的分类是卒中预后模型的可行设计选择。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Esra Zihni, Katryna Cisek, Hamzah Ziadeh, Hendrik Knoche, Robert Mikulik, John D. Kelleher

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05203.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05203

Published: 2026-08-07T02:09:58.830Z

5. SkillTrace: Multi-Trace Provenance Auditing for LLM-Agent Skill Reuse

中文摘要

摘要：LLM代理生态系统正在围绕可重复使用的技能快速发展：这些技能是元数据、自然语言指令、代码、工具、参考资料和操作工作流的混合模态包。随着技能成为市场中的商品，对其重复使用进行审计不再是普通代码克隆检测的问题。现有的检测器主要面向单模态源代码或整个包的相似性，而技能重复使用的证据分布在创作文本、实现片段和操作结构中。因此，它们可能会遗漏仅保留技能某一部分的重复使用。我们提出了SKILLTRACE，一种用于LLM代理技能重复使用的多追踪溯源审计框架。SKILLTRACE提取三种溯源追踪：表达、实现和操作。它将操作追踪表示为技能操作图（SOG），捕捉激活、流程和资源流结构。LLM仅在数据摄取时协助操作追踪的提取；在审计时，SKILLTRACE对缓存的追踪进行确定性比较，将每个追踪与同功能的严格负样本进行校准，并报告哪种追踪支持重复使用决策。在SKILLTRACE-BENCH上，在100个市场基准和751个负控中测试了820个转换后的重复使用正样本，SKILLTRACE实现了AUROC 0.938和F1 0.898。对36,446个技能的野外审计进一步表明，基于追踪属性的证据能在仓库级基准之外提出可操作的重复使用审查队列。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Jialuo Chen, Minghe Wang, Lingqi Jiang, Jianan Ma, Xinhao Deng, Xiaohu Du, Ruixiao Lin, Yunhao Feng, Linkang Du, Jingyi Wang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05204.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05204

Published: 2026-08-07T02:09:58.830Z

6. Abstract Event Causal Rules: Induction and Application

中文摘要

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Ziwei Zheng, Peiqiong Chen, Bang Wang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05205.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05205

Published: 2026-08-07T02:09:58.830Z

7. Otter: A Time-Aware, History-Conditioned Human Chess AI

中文摘要

摘要：Otter 是一个拥有 1530 万参数的人类国际象棋 AI，通过将对局建模为时间感知的序列过程来预测人类的走法选择，而不是将每一个局面孤立处理。它结合了两个条件信号：(1) 走法历史编码器，根据最近 20 步走法对预测进行条件化，以捕捉开局偏好、局势漂移和对局内的行为倾向；(2) 时间控制模块，根据计时压力调节预测。Otter 在 30 天内使用单个 T4 GPU 对来自 1.17 亿局 Lichess 快棋的 61 亿个局面进行训练。Otter 实现了 55.23% 的 top-1 和 90.95% 的 top-5 走法预测准确率，超越了先前最先进的人类棋类模型 Maia 2，同时使用的参数更少、训练数据更少。在 11 个 Elo 等级组（<1100 到 >=2000）中，准确率在 1900-1999 等级组达到最高 57.38%。这些结果表明，将国际象棋建模为时间感知、序列性活动比仅基于局面的方法可以实现更符合人类的走法预测，同时使用规模更小的模型。代码、训练好的模型以及完整的训练日志已公开发布。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Tarun Kumar S

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05206.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05206

Published: 2026-08-07T02:09:58.830Z

8. SearchAuditor: Auditing and Attributing Failures in Long-Horizon Search Agents

中文摘要

摘要：深度搜索代理通过长航程的网络交互来解决具有挑战性的问题，这一过程既复杂又脆弱：小的推理错误可能会通过冗长且嘈杂的轨迹传播，最终生成流畅但错误的答案。诊断此类失败非常困难，需要手动检查极长的执行轨迹，这可能超出人类能力。因此，我们提出了 SearchAuditBench，一个评估大型语言模型审计器能否定位、归因并修复这些失败的基准，从而减轻人类负担。SearchAuditBench 包含 1,243 条失败轨迹，平均 73.1 条消息和 65.1K 个 token，采集自五个深度搜索基准上的八个开放权重模型，每条由专家标注关键错误步骤、搜索特定根本原因及参考修复方案附带评分标准。我们进一步提出了 SearchAuditor，一种多视角审计框架，通过基于证据的裁定，有效地定位、归因并修复搜索代理的失败。实验结果表明，即使是最强的基线，在像 GPT-5.5 这样的前沿模型驱动下，其端到端通过率也仅为 26.6%。相比之下，我们的 SearchAuditor 在不同前沿模型上始终优于所有基线，端到端通过率达到 32.3%，并且通过其修复恢复失败的运行，使代理能够更好地从错误中恢复。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Zhixiang Liang, Yifei Liu, Yidan Huang, Haozhe Zhao, Beichen Huang, Jiaqi Wang, Nan Duan, Qiong Cao

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05212.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05212

Published: 2026-08-07T02:09:58.830Z

9. PD-GS: Phoneme-Driven 3DGS for Audio-Driven Talking Heads

中文摘要

摘要：三维高斯点渲染（3D Gaussian Splatting, 3DGS）能够实现快速、逼真的人物头像渲染，但准确的唇部动作仍难以实现：口部运动通常被过度平滑处理，并且可能违反诸如双唇闭合等严格的发音约束，从而产生著名的“漏嘴”现象。一个关键难点在于，短暂的、离散的发音事件需要从连续的声学嵌入中在回归目标下推断，这会使预测结果偏向平均的口型配置。虽然现代自监督语音编码器提供了丰富的韵律和语音学线索，但它们并不会提供可靠区分闭合类事件的明确、帧对齐的语言目标。我们提出 extbf{音素驱动高斯点渲染（Phoneme-Driven Gaussian Splatting, PD-GS）}，通过自动语音识别（ASR）和强制对齐管道获取的时间对齐音素标记来增强3DGS人物生成器。我们的核心组件 extbf{语言融合模块（Linguistic Fusion Module, LFM）}通过学习到的门控，将连续的音频上下文与离散音素嵌入自适应融合，使模型在保持平滑音频驱动动态的同时，在发音关键片段上强化音素指导。PD-GS完全基于单目视频进行训练，使用图像重建和唇部关键点监督。在HDTF数据集上，PD-GS在比较基线中获得最佳唇部几何效果（LMD 2.66），并在具有挑战性的音素序列中定性减少闭合违规现象，从而生成更符合语言规律的神经虚拟形象。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Ao Fu, Yi Zhou

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05218.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05218

Published: 2026-08-07T02:09:58.830Z

10. When Privileged Guidance Misaligns: State-Matched Routing and Contextualized Self-Distillation for Multi-Turn Agents

中文摘要

摘要：特权的在策略蒸馏通过允许同步教师在每一步对学生的响应重新评分并使用仅限训练的参考数据（如成功轨迹），为多轮代理提供密集监督。然而，在交互式环境中，学生之前的动作会不断改变执行状态。随着学生采取不同动作或以不同顺序完成子目标，其展开可能到达参考中未覆盖的状态，使得参考对于实际到达的状态来说成为不可靠的指导来源。因此，不加选择地应用特权蒸馏会造成状态与参考的不匹配。这种不匹配促使一个核心目标：提供与学生当前执行状态兼容的特权参考指导。我们提出了状态匹配路由和情境化自蒸馏（SMRC-SD），它明确确定特权轨迹应何时以及如何指导在策略学生。在每一步，SMRC-SD会验证学生当前执行状态是否与参考轨迹中的支持状态匹配。蒸馏仅在匹配状态下应用，过滤掉参考缺乏局部兼容指导的步骤。对于每个匹配状态，SMRC-SD进一步从成功轨迹中构建状态条件化的教师上下文，将监督基础建立在实际到达的状态上。在 ALFWorld 和 WebShop 上，SMRC-SD 一贯优于无条件的成功全路径蒸馏。使用 Qwen3-1.7B，在 ALFWorld 上任务成功率从 $0.746$ 提升至 $0.865$，在 WebShop 上从 $0.574$ 提升至 $0.693$。受控路由和上下文消融实验表明，选择局部支持的步骤和构建状态兼容的教师上下文都是这些提升的关键因素。代码可在此 https URL 获取。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Junzhuo Liu, Weiwei Li, Jun Ling, Peng Wang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2608.05219.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05219

Published: 2026-08-07T02:09:58.830Z

VLM Domain Papers

1. MapTCL: Temporal Consistency Learning via Bidirectional Alignment for Vectorized HD Map Construction

Abstract:Constructing reliable online HD maps remains challenging in dynamic urban environments due to moving objects and occlusions. While recent works employ feature-level temporal fusion to address this, they rely solely on per-frame ground truth supervision. Consequently, they lack an explicit objective to directly penalize the geometric noise and temporal jitter between consecutive online HD maps. To address this, we propose MapTCL, an auxiliary training strategy that formulates temporal consistency loss between current and past frames via bidirectional alignment. Specifically, Bidirectional Vector Consistency Learning (BVCL) models the geometric and semantic discrepancies between associated past and current vector instances as an auxiliary loss. We also employ Raster map Consistency Learning (RCL) as an additional loss to stabilize dense BEV features. By jointly training with these dual losses, MapTCL improves the temporal stability of generated HD maps. Extensive experiments on two standard benchmarks demonstrate the effectiveness of our approach. As a versatile plug-and-play module, MapTCL consistently enhances existing baseline models, achieving gains of +3.7 mAP & +2.8 C-mAP on nuScenes and +3.1 mAP & +2.5 C-mAP on Argoverse 2 without additional inference overhead.

中文摘要

摘要：在动态城市环境中，由于移动物体和遮挡，构建可靠的在线高清地图仍然具有挑战性。尽管近期的工作采用特征级时间融合来解决这一问题，但它们仅依赖每帧的真实值监督。因此，它们缺乏一个明确的目标来直接惩罚连续在线高清地图之间的几何噪声和时间抖动。为了解决这个问题，我们提出了 MapTCL，一种辅助训练策略，通过双向对齐在当前帧和过去帧之间构建时间一致性损失。具体来说，双向向量一致性学习（BVCL）将相关的过去和当前向量实例之间的几何和语义差异建模为辅助损失。我们还采用栅格地图一致性学习（RCL）作为附加损失，以稳定密集的俯视图特征。通过联合训练这两种损失，MapTCL 提升了生成高清地图的时间稳定性。在两个标准基准上的大量实验表明了我们方法的有效性。作为一个通用的即插即用模块，MapTCL 持续增强现有的基线模型，在 nuScenes 数据集上取得 +3.7 mAP 和 +2.8 C-mAP 的提升，在 Argoverse 2 数据集上取得 +3.1 mAP 和 +2.5 C-mAP 的提升，且不增加额外的推理开销。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Hyeonseo Kim, Juyeb Shin, Hyeonjun Jeong, Hiwon Shin, Dongsuk Kum

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2608.05209.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05209

Published: 2026-08-07T02:17:08.233Z

2. Innocent Panels, Hateful Stories: Evaluating and Detecting Hateful Intent in Multi-Turn Visual Story Generation

Abstract:Picture books and comics have long been used to disseminate hateful narratives because they are easily understood even by children, as exemplified by the notorious Nazi propaganda picture book \emph{Der Giftpilz}. Recently, frontier text-to-image (T2I) systems such as Gemini and GPT-Image have enabled conversational generation with consistent characters and scenes across turns, making hateful visual stories, namely ordered image groups that collectively convey hateful narratives, cheap and scalable to produce. Although prior work has studied hateful content generation by T2I systems, it focuses on individual images, leaving group-level hateful meaning largely unexplored. We aim to address the gap. Concretely, we introduce \texttt{HatefulStoryPrompts}, comprising 330 multi-turn configurations from 55 hateful stories across two languages and three visual styles, and evaluate five frontier models over 4,950 attempts. Every model completes over 80\% of the stories, with the strongest reaching 99.0\%. We further evaluate existing moderation systems on \texttt{HatefulVisualStory}, a human-labeled dataset of 969 hateful image sets and 990 benign controls, and find that they frequently miss group-level hateful meaning: dedicated safety models achieve at most 34.9\% recall, while a strong vision-language model reaches 67.5\%. Finally, we propose complementary proactive and post-generation defenses. An interaction-aware monitor achieves 97.3\% recall for prompt-only sessions and 92.6\% when the user supplies the first image, while post-generation methods jointly analyzing completed image groups reach 80.2\%. Our work shows that, as image generation evolves from isolated outputs to coherent visual narratives, safety must evolve accordingly, from per-image moderation to stateful reasoning over interactions and image relationships.

中文摘要

摘要：图画书和漫画长期以来被用来传播仇恨叙事，因为即使是儿童也能够轻松理解，例如臭名昭著的纳粹宣传图画书《Der Giftpilz》。最近，前沿的文本到图像（T2I）系统，如 Gemini 和 GPT-Image，已经实现了跨轮次具有一致角色和场景的对话生成，使仇恨视觉故事——即能整体传达仇恨叙事的有序图像组——的制作变得便宜且可扩展。尽管已有研究探讨了 T2I 系统的仇恨内容生成，但主要集中在单张图像上，对群体级别的仇恨意义探讨较少。我们旨在填补这一空白。具体来说，我们引入了 \texttt{HatefulStoryPrompts}，包含来自两种语言和三种视觉风格的 55 个仇恨故事的 330 个多轮配置，并在 4,950 次尝试中评估了五款前沿模型。每个模型完成故事的比例都超过 80%，最强的一款达到 99.0%。我们还评估了现有的审核系统在 \texttt{HatefulVisualStory} 数据集上的表现，该数据集包含 969 组人工标注的仇恨图像和 990 个良性对照，结果发现它们经常漏掉群体级别的仇恨意义：专用安全模型的召回率最高为 34.9%，而强大的视觉-语言模型可达 67.5%。最后，我们提出了互补的主动与生成后防御措施。交互感知监控在仅有提示的会话中召回率达到 97.3%，当用户提供首张图像时为 92.6%，而针对已完成图像组进行联合分析的生成后方法召回率可达 80.2%。我们的工作表明，随着图像生成从孤立输出发展为连贯的视觉叙事，安全防护也必须相应演进，从单图像审核转向对交互和图像关系的有状态推理。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Ye Leng, Junjie Chu, Yiting Qu, Mingjie Li, Yun Shen, Yang Zhang

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2608.05210.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05210

Published: 2026-08-07T02:17:08.233Z

3. StyleComposer: Training-Free Multi-Reference Style Composition

Abstract:The style of a painting is not monolithic: color, texture, and structure may come from different sources. Existing reference-guided methods transfer them as one style signal, leaving each attribute’s source and strength outside the user’s control. We ask where in a diffusion model one attribute can change while the others hold, and find that no single representation isolates all three. The proposed StyleComposer therefore routes each style attribute through the representation where it separates best and coordinates the routes over denoising time. Without training or inversion, it satisfies three references and the prompt jointly more closely than prior methods, and exposes one strength slider per attribute. Project page: this https URL

中文摘要

摘要：一幅画的风格不是单一的：颜色、纹理和结构可能来自不同的来源。现有的参考引导方法将它们作为一个整体风格信号进行迁移，使每个属性的来源和强度无法由用户控制。我们探讨在扩散模型中哪里可以改变一个属性而保持其他属性不变，并发现没有单一表示能够隔离所有三者。因此，提出的 StyleComposer 将每个风格属性通过最佳分离的表示进行路由，并在去噪时间上协调这些路径。无需训练或反演，它比以往的方法更紧密地满足三个参考和提示的联合要求，并为每个属性提供一个强度调节滑块。项目页面：这个 https URL

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Sanghyeok Lee, Jihye Kang, Namhyuk Ahn

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2608.05213.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05213

Published: 2026-08-07T02:17:08.233Z

4. NeuroAdaptTrainer: A Fiji/ImageJ Plugin for YOLO-Based Neuron Segmentation, InteractiveCorrection and Transfer Learning

Abstract:Neuron counting and segmentation in microscopy images of neuronal cultures is a routine and time-consuming task in neuroscience research, traditionally performed through manual inspection or semi-automatic tools. We present NeuroAdaptTrainer, an open-source Fiji/ImageJ plugin that integrates a YOLO instance-segmentation model directly into the microscopist’s workflow. The plugin allows a user to run automatic neuron detection on a single image or a batch of images, manually correct the resulting detections from within Fiji, and use those corrections to adapt the model to new imaging conditions via transfer learning. A built-in external validation module allows the base and adapted models to be compared quantitatively on a held-out annotated set. NeuroAdaptTrainer lowers the barrier for non-specialist users to benefit from deep-learning-based segmentation while keeping expert supervision at the center of the workflow.

中文摘要

摘要：在神经科学研究中，对神经元培养的显微图像进行神经元计数和分割是一项常规且耗时的任务，传统上通过人工检查或半自动工具完成。我们介绍了 NeuroAdaptTrainer，这是一个开源的 Fiji/ImageJ 插件，将 YOLO 实例分割模型直接集成到显微镜操作人员的工作流程中。该插件允许用户在单张图像或一批图像上运行自动神经元检测，在 Fiji 中手动修正生成的检测结果，并利用这些修正通过迁移学习将模型适配到新的成像条件。内置的外部验证模块允许在保留的标注集上对基础模型和适配模型进行定量比较。NeuroAdaptTrainer 降低了非专业用户利用基于深度学习的分割技术的门槛，同时保持专家监督作为流程的核心。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Daniela Eraso-Casas, Gerard Villarroya-Pique, Esther Serrano-Pertierra, M. Teresa Fernández-Sánchez, Antonello Novellie, Angel Rio-Alvarez, Víctor M. González

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2608.05226.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05226

Published: 2026-08-07T02:17:08.233Z

5. In-Context Forcing: Uncovering Context Effects in Autoregressive Video Diffusion

Abstract:Current few-step autoregressive video diffusion models depend on previous fully denoised clean frames as context for all denoising steps of the current frame. However, these clean frames leak excessive local details, which causes the model to take shortcuts, resulting in compromised temporal semantics and dynamics. Inspired by the perspective of diffusion as masking, we explore the impact of noisy contexts on few-step autoregressive generation. Yet, simply applying contexts with the same noise levels provides insufficient guidance, leading to poor temporal consistency. To resolve this dilemma, we introduce In-Context Forcing, a progressive autoregressive paradigm that utilizes contexts with decreasing noise levels. By applying less masking to distant frames and more masking to adjacent ones, this approach provides adaptive guidance, effectively ensuring both robust temporal consistency and high inter-frame dynamics. Furthermore, by decoupling the strict dependence on previous clean frames, our paradigm enables cross-frame parallel denoising, achieving substantial inference acceleration without sacrificing performance. Extensive experiments on VBench demonstrate that our method significantly outperforms state-of-the-art approaches in both visual fidelity and inference speed.

中文摘要

摘要：当前的少步自回归视频扩散模型依赖于之前完全去噪的干净帧作为当前帧所有去噪步骤的上下文。然而，这些干净帧泄露了过多的局部细节，导致模型采取捷径，从而损害了时间语义和动态。受到将扩散视为掩码的视角启发，我们探索了噪声上下文对少步自回归生成的影响。然而，简单地应用具有相同噪声水平的上下文提供的指导不足，导致时间一致性差。为了解决这一困境，我们引入了上下文强制（In-Context Forcing），这是一种渐进的自回归范式，利用噪声水平逐渐降低的上下文。通过对远处帧应用较少的掩码，对相邻帧应用更多掩码，这种方法提供了自适应指导，有效地保证了稳健的时间一致性和高帧间动态。此外，通过解除对之前干净帧的严格依赖，我们的范式实现了跨帧并行去噪，在不牺牲性能的前提下显著加快了推理速度。在VBench上的大量实验表明，我们的方法在视觉保真度和推理速度上均显著优于现有最先进的方法。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Lingxiao Yang, Liu Liu, Moran Li, Han Feng, Wenjian Cao, Jiangning Zhang, Ye Shi

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2608.05237.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05237

Published: 2026-08-07T02:17:08.233Z

6. Grad-CAM for Vision Transformers: A Systematic Taxonomy and Audit of Methodological Ambiguity in Explainable AI

Abstract:Gradient-weighted Class Activation Mapping (Grad-CAM) is widely used to visualize model decisions, but it was originally formulated for convolutional neural networks, where spatial feature maps and channel dimensions have clear architectural meanings. Vision Transformers (ViTs) do not provide the same structure, instead representing images through tokens, attention, residual streams, and multimodal interactions. This paper presents a systematic taxonomy and literature audit of how Grad-CAM and related methods are adapted, justified, and reported for ViT-based architectures. From an initial search of more than 550 papers, we identify 175 papers that apply Grad-CAM or Grad-CAM-adjacent methods to ViTs. We find that most papers do not provide a full mathematical or implementation-level account of how Grad-CAM is adapted to transformer representations. To characterize this gap, we introduce a descriptive taxonomy of ViT Grad-CAM adaptations that makes explicit the feature locations, gradient targets, spatial reconstruction steps, and aggregation choices that are often left implicit. This taxonomy is not intended to prescribe a single correct adaptation, but to clarify the range of methodological choices being made. The study shows that Grad-CAM on ViTs is often treated as a trivial extension of CNN-based Grad-CAM, despite requiring nontrivial choices that affect rigor, reproducibility, and interpretation.

中文摘要

摘要：梯度加权类激活映射（Grad-CAM）被广泛用于可视化模型决策，但它最初是为卷积神经网络（CNN）制定的，在CNN中，空间特征图和通道维度具有明确的架构意义。视觉变换器（ViTs）并不提供相同的结构，而是通过令牌、注意力、残差流和多模态交互来表示图像。本文对Grad-CAM及相关方法在基于ViT的架构上的适应、合理性和报告方式进行了系统的分类和文献审查。通过对550多篇论文的初步检索，我们确定了175篇将Grad-CAM或与Grad-CAM相关的方法应用于ViT的论文。我们发现，大多数论文没有提供Grad-CAM如何适应变换器表示的完整数学或实现级别说明。为描述这一空白，我们提出了一种ViT Grad-CAM适应的描述性分类法，明确指出特征位置、梯度目标、空间重建步骤和聚合选择，这些通常是隐性的。该分类法并不旨在规定单一正确的适应方法，而是为了阐明方法选择的范围。研究表明，ViTs上的Grad-CAM通常被视为CNN基础Grad-CAM的一个简单扩展，尽管实际上它需要非平凡的选择，这些选择会影响严谨性、可重复性和解释性。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Casey Wall, Longwei Wang, Rodrigue Rizk, KC Santosh

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2608.05258.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05258

Published: 2026-08-07T02:17:08.233Z

7. A Paragraph is Worth a Thousand Captions: Rethinking Text Supervision for Vision-Language Retrieval

Abstract:Contrastive vision-language models such as CLIP and BLIP are typically trained on short image captions, limiting their ability to retrieve images from detailed textual descriptions. While methods such as Long-CLIP extend the token limit through positional embedding interpolation, we ask a simpler question: does training text granularity alone determine long-text retrieval performance? We present a systematic study of supervision ranging from single captions to multi-sentence paragraphs for contrastive image-text retrieval. Using a synthetic pipeline based on Qwen2-VL and Llama 3.2 Vision, we generate diverse captions, hard negatives, and quality-scored paragraphs for 500K CC3M images. To isolate the effect of text granularity, we fine-tune only the BLIP text encoder while keeping the vision encoder frozen across 10 training configurations. Our paragraph-supervised models match Long-CLIP-L on ShareGPT4V and outperform it by more than 14 points on DOCCI for image-to-text retrieval, without architectural changes. We further show that paragraph supervision enables effective use of long token sequences, whereas caption-only training degrades beyond 60 tokens. Increasing caption diversity improves short-caption retrieval with diminishing returns, while paragraph supervision consistently benefits long-description benchmarks and hard negatives prove detrimental in text-only fine-tuning. Evaluations on Flickr30k, COCO, ShareGPT4V, and DOCCI provide a comprehensive analysis of the trade-offs between text granularity, retrieval direction, and description length.

中文摘要

摘要：对比视觉-语言模型如 CLIP 和 BLIP 通常在简短的图像标题上进行训练，这限制了它们从详细文本描述中检索图像的能力。虽然像 Long-CLIP 这样的方法通过位置嵌入插值扩展了 token 上限，但我们提出一个更简单的问题：仅通过训练文本粒度是否决定了长文本检索的性能？我们对从单个标题到多句段落的对比图文检索监督进行了系统研究。基于 Qwen2-VL 和 Llama 3.2 Vision，我们构建了一个合成管线，为 50 万张 CC3M 图像生成多样化标题、困难负样本和质量评分段落。为了隔离文本粒度的影响，我们在保持视觉编码器冻结的情况下，仅微调 BLIP 文本编码器，覆盖了 10 种训练配置。我们的段落监督模型在 ShareGPT4V 上与 Long-CLIP-L 相当，并在 DOCCI 的图像到文本检索中超出其 14 分以上，而无需修改架构。我们进一步表明，段落监督能够有效利用长 token 序列，而仅用标题训练在超过 60 个 token 后性能下降。增加标题多样性会改善短标题检索，但收益递减，而段落监督始终对长描述基准有益，同时困难负样本在纯文本微调中反而有害。对 Flickr30k、COCO、ShareGPT4V 和 DOCCI 的评估提供了文本粒度、检索方向和描述长度之间权衡的全面分析。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Mahyar Ghazanfari, Amin Tabrizian, Arsyi Aziz, Binshuai Wang, Peng Wei

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2608.05260.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05260

Published: 2026-08-07T02:17:08.233Z

8. Context Matters: Support Set Selection and Failure Detection for In-Context Medical Image Segmentation

Abstract:In-context learning (ICL) adapts medical image segmentation models to unseen structures and modalities without retraining by conditioning on a task-specific support set of image-mask exemplars. Because this support set is the model’s only task-specific signal, its composition directly influences segmentation performance. In this work, we investigate the support set as a controllable determinant of ICL reliability. First, we compare random sampling against similarity-based selection, where exemplars are retrieved based on their visual similarity to the query image. Second, we train a transformer-based classifier to predict, from the query and support images alone, whether a segmentation will fall below a specified Intersection-over-Union (IoU) threshold. Using MultiverSeg with DINOv3 embeddings across four benchmarks and three imaging modalities, we show that similarity-based selection consistently matches or outperforms random sampling, with the largest gains at the smallest support set sizes. Furthermore, our classifier predicts segmentation failure above chance on all four benchmarks. Ultimately, these results demonstrate that the reliability of in-context segmentation can be both improved via informed support selection and anticipated before use, providing practical mechanisms for safer clinical deployment.

中文摘要

摘要：上下文学习（ICL）通过依赖特定任务的图像-掩码示例支持集，使医学图像分割模型能够适应未见过的结构和模态，而无需重新训练。由于该支持集是模型唯一的任务特定信号，其组成直接影响分割性能。在本工作中，我们将支持集作为可控的ICL可靠性决定因素进行研究。首先，我们比较了随机采样与基于相似性的选择方法，其中示例根据其与查询图像的视觉相似性进行检索。其次，我们训练了一个基于Transformer的分类器，仅从查询图像和支持图像预测分割是否会低于指定的交并比（IoU）阈值。通过在四个基准测试和三种成像模态中使用带有DINOv3嵌入的MultiverSeg，我们表明，基于相似性的选择始终能与随机采样匹配或优于随机采样，在支持集最小时获得的增益最大。此外，我们的分类器在所有四个基准上预测分割失败的准确率均高于随机概率。最终，这些结果表明，通过信息化的支持集选择可以提高上下文分割的可靠性，并且可在使用前进行预测，从而为更安全的临床应用提供了实用机制。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Youssef Gehad, Emmanuel Zerefa, Krish Kabra, Guha Balakrishnan

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2608.05333.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05333

Published: 2026-08-07T02:17:08.233Z

9. Positive-Unlabeled Preference Optimization For Chest X-ray Report Generation

Abstract:Vision-Language Models (VLMs) for radiology report generation are typically trained on retrospective clinical reports, which suffer from omission noise: clinically present findings are left unreported due to the omission of subtle findings. For example, prior studies show that cardiomegaly may be omitted from ICU chest X-ray reports when the imaging request is focused on monitoring support device placement. As a result, models trained with standard approaches inherit these omissions, learning to under-report findings themselves. We propose PU-DPO, a preference optimization framework to prevent omission noise from corrupting the preference signal. We reformulate the objective under a positive-unlabeled (PU) learning framework, treating absent mentions as unlabeled rather than truly negative. Our framework provides preference supervision using constructed contrastive pairs, generated using edits to model responses, producing variants that explicitly mention or omit a specific finding. Generated responses that mention the finding are naturally preferred in the context of visual evidence. Across semi-synthetic experiments and analyses on real-world chest radiograph benchmarks where adjudicated labels are available, PU-DPO yields consistent gains in detection rates and recovery of hidden positives across multiple pathologies, and is more robust to omission noise than prior approaches.

中文摘要

摘要：用于放射学报告生成的视觉-语言模型（VLMs）通常在回顾性临床报告上进行训练，而这些报告存在遗漏噪声问题：由于某些细微发现被忽略，临床上存在的发现也未被报告。例如，先前的研究表明，当影像检查请求集中于监测支持器械的位置时，ICU胸部X光报告中可能会遗漏心脏增大。因此，采用标准方法训练的模型也会继承这些遗漏，学习到低报发现。我们提出了PU-DPO，一种偏好优化框架，用于防止遗漏噪声损害偏好信号。我们在正-未标记（PU）学习框架下重新表述了目标，将未出现的提及视为未标记而非真正的负样本。我们的框架通过构建对比对提供偏好监督，这些对比对通过对模型响应的编辑生成，产生明确提及或省略特定发现的变体。在视觉证据的背景下，自然更偏好提及发现的生成响应。在半合成实验以及带有裁定标签的真实世界胸片基准分析中，PU-DPO在多种病理发现的检测率和隐藏阳性恢复方面均取得一致提升，并且比以往方法对遗漏噪声更具鲁棒性。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Yuta Kobayashi, Pradyun Ramesh, Muhammad Ahmed Chaudhry, Vincent Jeanselme, Judy Wawira Gichoya, Sanmi Koyejo, Kathleen Capaccione, Shalmali Joshi

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2608.05341.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05341

Published: 2026-08-07T02:17:08.233Z

10. LoDA: A Level of Detection Aware Method and a Multimodal Sensing Benchmark for Object Level Change Detection

Abstract:High-definition 3D LiDAR maps are important for autonomous driving and smart-city services, which require reliable detection of object-level changes in multi-temporal urban LiDAR to keep digital maps aligned with the physical world. Existing approaches from raster height differencing to depth image and point-cloud networks often remain tile-based and threshold-driven, yielding per-point scores without explicit detection limits or consistent object-level labels. We propose an object-level 3D change-detection pipeline that integrates detection-limit-aware registration, geometry-driven object proxies with rule-based semantic and instance segmentation, and displacement cues in height, volume, and surface-normal direction to assign five change labels with confidence. By decoupling registration, geometry, and semantics, the pipeline propagates pose uncertainty into spatially varying detection limits, stabilizes cross-epoch correspondences, and suppresses false changes caused by residual misalignment and density variation. We also present LoDA, a level-of-detection (LoD) aware benchmark for the Subiaco district with fused multi-temporal vehicle-LiDAR maps constructed with LiDAR, GNSS, and IMU support, semantic instances, and object-level annotations. On this benchmark, our method achieves 95.0% accuracy, 90.8% macro F1, and 83.0% macro IoU, exceeding the best baseline by 8.7 IoU points and 4.4 F1 points. On the public Urb3DCD-V2 benchmark evaluated under the official point-wise protocol, it reaches 96.81% mean accuracy and 89.52% mean change IoU, improving over the strongest reported baselines by 1.36 points in mAcc and 3.18 points in mIoUch.

中文摘要

摘要：高精度三维激光雷达（LiDAR）地图对于自动驾驶和智慧城市服务非常重要，这些服务需要在多时段城市LiDAR中可靠地检测对象级变化，以确保数字地图与物理世界保持一致。现有方法从栅格高度差异到深度图像和点云网络，通常仍基于瓦片并依赖阈值，产生每点评分而没有明确的检测限或一致的对象级标签。我们提出了一种对象级三维变化检测流程，融合了检测限感知的配准、基于几何的对象代理及规则驱动的语义和实例分割，以及在高度、体积和表面法向方向上的位移线索，以赋予五种变化标签及其置信度。通过解耦配准、几何和语义，该流程将姿态不确定性传播到空间上变化的检测限，稳定跨时期对应关系，并抑制由残余错位和密度变化引起的虚假变化。我们还提出了LoDA，一个面向检测水平（LoD）的基准，用于Subiaco区，该基准融合了多时段车辆LiDAR地图，支持LiDAR、GNSS和IMU，包含语义实例和对象级注释。在该基准上，我们的方法实现了95.0%的准确率、90.8%的宏F1和83.0%的宏IoU，分别比最佳基线高出8.7 IoU点和4.4 F1点。在官方点点协议下评估的公共Urb3DCD-V2基准上，方法达到了96.81%的平均准确率和89.52%的平均变化IoU，在mAcc和mIoUch上分别比最强的已报道基线提高了1.36点和3.18点。

LLM Analysis

LLM Analysis Failed: Error: 抓取失败（已重试2次）: Navigation timeout of 10000 ms exceeded

Authors: Haitian Wang, Xinyu Wang, Sheldon Fung, Xian Zhang, Zichen Geng

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2608.05356.pdf

CoolPaper URL: https://papers.cool/arxiv/2608.05356

Published: 2026-08-07T02:17:08.233Z

ArXiv Domain 2026-08-07

LLM Domain Papers

1. Simulator-Grounded Large Language Models for Industrial Causal Reasoning: Tool-Use, Structured Injection, and Plant-Portable Retrieval for Wastewater Treatment Decision Support

2. Mean-Field Dynamics of Chain-of-Thought Reasoning in Large Language Models

3. Universal Pathologies, Conditional Consequences: A Triple-Robustness Analysis of RAG for Multi-Hop Traceability

4. RIG-RoPE: Relation- and Instance-Gated Rotary Positional Encoding with Duration-Aware Temporal Coordinates

5. Beyond Sentiment: Comparing Traditional NLP and LLM-Based Multi-Dimensional Analysis for Political News Evaluation

6. Scaffold-Mediated Post-Training: Co-Evolving Model Parameters and Procedural Scaffold Graphs

7. Large Language Models Threaten Double-blind Review

8. Safe Evolution with Circuit Anchors

9. SemiAdapt-Instruct: Extensible Instruction Tuning via Latent Domain-Specialised Adapters

10. PoolBench: A Benchmark for Pooling Strategies in Concept Representation Evaluation for Decoder-Only LLMs

Agent Domain Papers

1. Agentic Nesting: A New Methodology for Existing Enterprise Application Integration and Services

2. The Ignition Index: Measuring Global Workspace Dynamics in Language Models

3. Woodpecker Distillation: Weak Models Diagnose Reasoning Bugs in Strong Models

4. From Continuous Predictors to Clinical Thresholds: Early Evidence on Performance Trade-offs of Guideline-Based Categorisation for Ischaemic Stroke Outcome Prediction

5. SkillTrace: Multi-Trace Provenance Auditing for LLM-Agent Skill Reuse

6. Abstract Event Causal Rules: Induction and Application

7. Otter: A Time-Aware, History-Conditioned Human Chess AI

8. SearchAuditor: Auditing and Attributing Failures in Long-Horizon Search Agents

9. PD-GS: Phoneme-Driven 3DGS for Audio-Driven Talking Heads

10. When Privileged Guidance Misaligns: State-Matched Routing and Contextualized Self-Distillation for Multi-Turn Agents

Evaluation Domain Papers

1. Agentic Nesting: A New Methodology for Existing Enterprise Application Integration and Services

2. The Ignition Index: Measuring Global Workspace Dynamics in Language Models

3. Woodpecker Distillation: Weak Models Diagnose Reasoning Bugs in Strong Models

4. From Continuous Predictors to Clinical Thresholds: Early Evidence on Performance Trade-offs of Guideline-Based Categorisation for Ischaemic Stroke Outcome Prediction

5. SkillTrace: Multi-Trace Provenance Auditing for LLM-Agent Skill Reuse

6. Abstract Event Causal Rules: Induction and Application

7. Otter: A Time-Aware, History-Conditioned Human Chess AI

8. SearchAuditor: Auditing and Attributing Failures in Long-Horizon Search Agents

9. PD-GS: Phoneme-Driven 3DGS for Audio-Driven Talking Heads

10. When Privileged Guidance Misaligns: State-Matched Routing and Contextualized Self-Distillation for Multi-Turn Agents

VLM Domain Papers

1. MapTCL: Temporal Consistency Learning via Bidirectional Alignment for Vectorized HD Map Construction

2. Innocent Panels, Hateful Stories: Evaluating and Detecting Hateful Intent in Multi-Turn Visual Story Generation

3. StyleComposer: Training-Free Multi-Reference Style Composition

4. NeuroAdaptTrainer: A Fiji/ImageJ Plugin for YOLO-Based Neuron Segmentation, InteractiveCorrection and Transfer Learning

5. In-Context Forcing: Uncovering Context Effects in Autoregressive Video Diffusion

6. Grad-CAM for Vision Transformers: A Systematic Taxonomy and Audit of Methodological Ambiguity in Explainable AI

7. A Paragraph is Worth a Thousand Captions: Rethinking Text Supervision for Vision-Language Retrieval

8. Context Matters: Support Set Selection and Failure Detection for In-Context Medical Image Segmentation

9. Positive-Unlabeled Preference Optimization For Chest X-ray Report Generation

10. LoDA: A Level of Detection Aware Method and a Multimodal Sensing Benchmark for Object Level Change Detection