ArXiv Domain 2026-10-05

数据来源：ArXiv Domain

LLM Domain Papers

1. HakemBench: A Turkish Benchmark of Typed Decisions

Abstract:HakemBench is a Turkish benchmark of typed decisions, in which the model under test reads a text, a question and a fixed set of options and returns a probability for every option. Version 1.0 is released fully open under CC BY 4.0, with 2,346 items and 4,275 choice, yes/no and score questions in seven tracks (fact-check triage, education, guardrails, legal routing, moderation, spam and phishing, and customer support). One harness scores decision quality (macro F1), calibration (from the normalised Brier score) and selective automation (from the normalised area under the generalised risk-coverage curve), combines them by a geometric mean and reports intervals from 2,000 bootstrap draws; probes for option order, paraphrase, English translation and substituted names are reported alongside. Most gold labels come from blind passes of one AI model family compared with the votes of a panel of large language models from other model families; they are not human-verified. On a board of 16 rows the leader scores a composite of 0.888 and the lab’s own model is 7th at 0.660. Its numbers are not blind. Earlier runs’ test results shaped its training data, so its guardrail, moderation and customer support numbers are flagged; with every model scored on the other four tracks only, its composite is 0.678, 6th of 16.

中文摘要

摘要：HakemBench 是一个土耳其语的类型化决策基准，其中受测模型需要读取一段文本、一个问题和一组固定选项，并返回每个选项的概率。版本 1.0 在 CC BY 4.0 协议下完全开放发布，包含 2,346 个条目和 4,275 个选择题、是/否题和评分题，涵盖七个领域（事实核查分流、教育、护栏机制、法律路由、内容审核、垃圾邮件与网络钓鱼以及客户支持）。一个评估工具衡量决策质量（宏 F1）、校准（根据归一化布里尔分数）和选择性自动化（根据归一化广义风险-覆盖曲线下的面积），将其通过几何平均结合，并报告 2,000 次自助抽样的区间；同时报告了选项顺序、改写、英文翻译和替换名称的测试。大多数金标签来自同一家 AI 模型家族的盲测结果，与其他模型家族的大型语言模型小组投票进行对比；这些标签没有经过人工验证。在一个 16 行的排行榜上，领先者的综合得分为 0.888，该实验室自有模型排名第七，得分为 0.660，其得分结果不是盲测。早期运行的测试结果塑造了其训练数据，因此其护栏机制、内容审核和客户支持的得分被标记；每个模型仅在其他四个领域上得分时，其综合得分为 0.678，在 16 个模型中排名第六。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02293

Authors: Sait Furkan Teke

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.02293.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02293

Published: 2026-10-05T02:23:35.825Z

2. Finding the Move Is Not Winning the Game: XiangqiBench for Closed-Loop Evaluation of LLM Agents

Abstract:Static evaluations credit a language model for naming the right move, but an agent must carry a plan through to a verified outcome while an opponent responds. We introduce XiangqiBench, an executable benchmark that measures this difference in Chinese chess: starting from 119 tactical endgames with forced mates supported by engine or checks-only search, an LLM agent must deliver checkmate against an engine defender. An interactive REPL interface separates real moves, state queries, and forward simulation, and we record 8,568 multi-turn trajectories from 12 frontier LLMs under two observation protocols. Three signals that look like competence each overstate closed-loop success. (i) The Conversion Gap: models play the stored reference first move in 26.1\% of Sighted trials, yet only 13.9\% of these trials end in a win. (ii) The Consistency Gap: the leading model reaches 38.7\% pass@3 but only 5.9\% pass^3, winning all three trials on 7 of the 46 positions it ever wins. (iii) The Simulation Gap: 32.3\% of accepted simulation calls stop on an illegal move, and in 49.3\% of comparable cases the real defender replies differently from the line the agent simulated; self-authored rollouts check legality but cannot anticipate the opponent. Finding the move is not winning the game: agent evaluations should score closed-loop outcomes and report reliability alongside coverage.

中文摘要

摘要：静态评估会因为语言模型命中了正确的走法而给予其评分，但一个代理必须在对手响应的情况下，将计划执行到经过验证的结果。我们引入了象棋基准(XiangqiBench)，这是一个可执行的基准，用于衡量中国象棋中的这种差异：从119个带强制将死的战术残局开始，这些残局由引擎支持或仅通过将军搜索得到，LLM代理必须对抗引擎防守实现将死。一个交互式REPL界面将实际走法、状态查询和前向模拟分开，我们在两种观察协议下记录了12个前沿LLM的8,568条多回合轨迹。三类看似表现能力的信号都高估了闭环成功率。(i) 转换差距：模型在有视野的实验中先走存储的参考第一步的比例为26.1%，但这些实验中只有13.9%以胜利告终。(ii) 一致性差距：领先模型在pass@3中达到38.7%，而在pass^3中仅为5.9%，它在赢得过的46个残局中，只有7个残局在三次实验中全胜。(iii) 模拟差距：32.3%的接受的模拟调用因非法走法而中止，在可比案例中，49.3%的情况下，真实防守者的反应与代理模拟的走法不同；自我生成的回溯可以检查合法性，但无法预判对手。找到走法不等于赢得比赛：代理评估应对闭环结果进行评分，并在报告覆盖率的同时说明可靠性。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02425

Authors: Yekun Chai, Qiwei Peng, Haoyi Xiong

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.02425.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02425

Published: 2026-10-05T02:23:35.825Z

3. Counterexample Generation via Per-Theorem Symbolic Verifiers: When Imitation Hurts and Reinforcement Repairs

Abstract:Large language models often solve a theorem forward yet fail to disprove a closely related false one: a falsification gap that supervised fine-tuning does not close and can actively worsen. We frame counterexample generation as constrained witness emission against a deterministic per-theorem Python verifier, and release SymCE, a corpus of 4,707 false undergraduate-algebra and real-analysis conjectures, each paired with executable verifiers. The verifier also serves as the reward function, making SymCE a training environment. Training Qwen3-4B with SFT followed by GRPO under this oracle reveals an imitation trap: counterexample-only SFT collapses true-theorem recognition from 0.27 to 0.00, while RLVR with a sparse outcome-only reward repairs this and exceeds the base, to 0.66. The collapse replicates across four seeds and on Gemma-3-4B. Sparse and dense rewards yield statistically indistinguishable in-domain success yet diverge by 33 points on a held-out calibration probe, a dissociation we trace to the partial-credit term. Our 4B model outperforms every evaluated 7B open-weights math specialist, remains competitive with six frontier commercial APIs, and transfers under unchanged prompting to GSM8K, MATH-500 and MMLU-college-math. A human audit of 177 verifier decisions finds 97.7% accuracy. Code, data, verifier modules and annotations: this https URL.

中文摘要

摘要：大型语言模型通常可以正向解决定理，但无法反驳一个密切相关的错误定理：这种伪证缺口（falsification gap）是监督微调无法弥补的，甚至可能使情况恶化。我们将反例生成框架化为针对每个定理的确定性 Python 验证器的受约束证人输出，并发布了 SymCE，它包含 4707 条错误的本科代数和实分析猜想，每条都配有可执行验证器。验证器也充当奖励函数，使 SymCE 成为训练环境。在这一验证器下，对 Qwen3-4B 进行先 SFT 再 GRPO 的训练揭示了模仿陷阱：仅反例的 SFT 将真定理识别率从 0.27 崩溃到 0.00，而带有稀疏结果奖励的 RLVR 修复了这一问题，并超过了基础水平，达到 0.66。这种崩溃在四个随机种子以及 Gemma-3-4B 上得到重复验证。稀疏与密集奖励在域内成功率上统计上无显著差异，但在保留的校准探针上相差 33 分，我们将这一脱钩归因于部分积分项。我们的 4B 模型在评估过的所有 7B 开放权重数学专用模型中表现最佳，与六个前沿商业 API 保持竞争力，并可在不改变提示的情况下迁移到 GSM8K、MATH-500 和 MMLU-college-math。对 177 个验证器决策的人类审核显示准确率为 97.7%。代码、数据、验证器模块和注释请访问该 https URL。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02444

Authors: Omar Farouk Zouak, Houssam Eddine Boukhalfa, Soumaya Lakehal, Shiv Katiyar, Samia Nefti-Meziani

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.02444.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02444

Published: 2026-10-05T02:23:35.825Z

4. FinDialogLens: Event Extraction over Multi-Party Dialogue for Missed-Trade Identification in Financial Chatrooms

Abstract:Multi-party financial chatrooms are vital for sales-and-trading professionals, but their complexity makes manual recovery of missed trades infeasible: each Request for Quote (RFQ) is an event whose final price and trade outcome appear many messages after the RFQ-trigger message (the inquiry message), interleaved with concurrent RFQs from other participants. We cast this as event extraction (EE) over multi-party dialogue and present FinDialogLens, a hybrid LLM pipeline in which compact fine-tuned classifiers act as inference-time scaffolds: they detect RFQ-triggers and price/trade outcome metadata, an RFQ-Level Module segments per-event RFQ windows, and a Trade Engine fills argument roles. With GPT-4o, FinDialogLens reaches 92.1% and 94.3% accuracy on final price and trade outcome, respectively, outperforming full-chatroom CoT prompting methods; fine-tuned open-source LLMs with as few as 3B parameters achieve comparable performance with modest in-domain data. To make the LLM-based solution practical at scale, a difficulty-aware router balances cost and accuracy by allocating RFQs between a low-cost rule-based engine and the higher-performing LLM-powered Trade Engine, cutting LLM calls by 85% on final price while recovering half of the accuracy gap to FinDialogLens (GPT-4o), saving over $300/day at our 70,000-RFQ/day scale.

中文摘要

摘要：多方金融聊天群对销售和交易专业人员至关重要，但其复杂性使得手动恢复错过的交易几乎不可能：每个报价请求（RFQ）都是一个事件，其最终价格和交易结果往往出现在RFQ触发消息（询价消息）之后的许多消息中，并与来自其他参与者的并发RFQ交错出现。我们将此问题视为多方对话中的事件抽取（EE），并提出了FinDialogLens，这是一种混合型大型语言模型（LLM）管道，其中紧凑型微调分类器在推理时充当辅助：它们检测RFQ触发器和价格/交易结果元数据，RFQ级模块对每个事件的RFQ窗口进行分段，交易引擎则填充参数角色。使用GPT-4o时，FinDialogLens在最终价格和交易结果上的准确率分别达到92.1%和94.3%，超越了全聊天群链式推理（CoT）提示方法；使用仅3B参数的微调开源LLM在适量领域数据下也能达到相当的性能。为了使基于LLM的解决方案在大规模应用中可行，困难感知路由器通过在低成本基于规则的引擎和高性能的LLM驱动交易引擎之间分配RFQ，实现成本与准确率的平衡，在最终价格计算中减少85%的LLM调用，同时恢复了FinDialogLens（GPT-4o）一半的准确率差距，在我们每天处理70,000个RFQ的规模下每天节省超过300美元。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02455

Authors: Chin-Lun Fu, Hong Ni, Behrouz Madahian

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.02455.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02455

Published: 2026-10-05T02:23:35.825Z

5. CUEing User Simulators: Calibrated User Embeddings for Multi-Turn Benchmarking

Abstract:Recent benchmarks rely on user simulators to evaluate AI agents in multi-turn interaction. While existing simulation techniques demonstrate surface fidelity to human style and behavior, ecologically valid interactive benchmarking also requires alignment in when and how agents fail across simulated and real user populations. We find that existing simulators lack outcome calibration: agreement with observed success rates and failure patterns when real users interact with the same agent. We introduce Calibrated User Embeddings (CUE), a framework that both encodes observed sessions and samples continuous representations, then decodes them into persona commands to steer LLMs to act as user simulators without training. Through this, we evaluate user-conditioned replay of past sessions and aggregate metric agreement when sampling novel personas for the same tasks. On $\tau^2$-Bench, CUEd simulators commit fewer simulator-attributed errors and more faithfully reproduce real-user agent failure modes, aggregate success rates, and outcomes for specific task-user pairs than other persona-based simulation methods. These gains coexist with competitive user fidelity as measured using metrics established in prior work. After being fit to mostly customer support interactions, the same CUEd simulators generalize to document creation, math tutoring, and casual conversation, and remain effective across different simulator LLMs without CUE retraining.

中文摘要

摘要：近期的基准测试依赖用户模拟器来评估人工智能代理在多轮交互中的表现。虽然现有的模拟技术在表面上展示了与人类风格和行为的相似性，但生态有效的交互基准测试还要求模拟与真实用户群体在代理失败的时间和方式上保持一致。我们发现，现有的模拟器缺乏结果校准：即在真实用户与相同代理交互时，与观察到的成功率和失败模式的一致性。我们提出了校准用户嵌入（CUE）框架，它能够编码观察到的会话并采样连续表示，然后将其解码为人格命令，以引导大型语言模型（LLM）作为用户模拟器工作而无需训练。通过这种方式，我们评估了基于用户条件的过去会话重放，并在为相同任务采样新的人格时汇总指标一致性。在$ au^2$-Bench上，CUEd模拟器比其他基于人格的模拟方法产生更少的模拟器归因错误，更忠实地再现真实用户代理的失败模式、整体成功率以及特定任务-用户对的结果。这些提升与以往工作中建立的指标衡量的用户忠实度指标同时共存。在主要适配于客户支持交互后，相同的CUEd模拟器能够推广到文档创建、数学辅导和日常对话，并且在不同的模拟器LLM上无需CUE重新训练也保持有效。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02460

Authors: Anjali Kantharuban, Jonas Mueller

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.02460.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02460

Published: 2026-10-05T02:23:35.825Z

6. APDMem: Agent-Controlled Progressive Disclosure for Query-Adaptive Long-Term Memory

Abstract:Personalized LLM assistants must recover sparse evidence from long conversation histories across queries of varying complexity. We introduce APDMem (Agent-controlled Progressive Disclosure Memory), a hierarchical long-term memory architecture that applies progressive disclosure to memory retrieval. Rather than relying on a flat memory store or fixed retrieval granularity, APDMem represents conversation history as four progressively detailed layers: thematic summaries, personalized key facts, turn-level evidence notes, and raw messages. At inference time, a controller applies progressive disclosure to the memory hierarchy: it first reads high-level summaries and drills into finer evidence only when needed. This creates an adaptive cost-fidelity trade-off: simple queries can terminate early, while complex temporal, multi-hop, or exact-evidence queries trigger deeper inspection. A note synthesizer converts retrieved evidence into a query-focused structure that consolidates facts, orders events, and flags contradictions before final answer generation. Experiments on LongMemEval show that APDMem achieves strong performance for long-context memory reasoning while accessing only 8% of the total conversations.

中文摘要

摘要：个性化大型语言模型助手必须从不同复杂度查询的长对话历史中恢复稀疏证据。我们提出了 APDMem（代理控制的渐进式披露记忆），这是一种分层长期记忆架构，将渐进式披露应用于记忆检索。APDMem 并不依赖于扁平的记忆存储或固定的检索粒度，而是将对话历史表示为四个逐步详细的层次：主题摘要、个性化关键事实、轮次级证据笔记和原始消息。在推理时，控制器将渐进式披露应用于记忆层次结构：它首先读取高级摘要，并仅在需要时深入更具体的证据。这创造了一种自适应的成本-精度权衡：简单查询可以提前结束，而复杂的时间、跨步或精确证据查询则触发更深入的检查。笔记合成器将检索到的证据转换为聚焦于查询的结构，整合事实、排列事件顺序，并在最终答案生成前标注矛盾。LongMemEval 的实验表明，APDMem 在长上下文记忆推理中表现出强大性能，同时仅访问总对话的 8%。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02472

Authors: Chin-Lun Fu, Anagha Kulkarni, Hong Ni, Behrouz Madahian

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.02472.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02472

Published: 2026-10-05T02:23:35.825Z

7. From Retrieval to Typed Decisions: Calibrated System One Models from Biomedical Sentence Encoders

Abstract:Typed decision models answer schema-constrained questions about a text in one forward pass and return probabilities meant to be thresholded. We ask whether biomedical sentence encoders trained for retrieval are good starting points for such models. We present SBERT2S1, which converts Sentence-Transformers encoders into bi-encoder, cross-head (C) and prior-fused residual (PFR) decision models, together with BIODECIDE, a biomedical typed-decision suite, and MEDLINE-S1, 243k training decisions derived from NLM indexing. Across six parent-retriever pairs, retrieval training improves zero-shot matching of content-bearing options. After fine-tuning, its effect depends on the head: across five pairs and three training-set sizes, retrieval training significantly helps PFR, which keeps the retrieval prior, in 10 of 15 comparisons, but helps C in one and hurts it in five. A matched grid of two heads and five training objectives shows that C outperforms PFR under every objective, and that the released RLCD recipe of open System One models trails cross-entropy by 2.5-3.0 points. The deficit stems mainly from its reward normalisation, which inflates the noisy score-function term 3.6-15-fold; an unbiased leave-one-out estimator recovers most of the gap. After temperature scaling, no objective is clearly better calibrated than cross-entropy. We release the code, the MEDLINE-S1 labels and a model.

中文摘要

摘要：类型化决策模型在一次前向计算中回答受模式约束的文本问题，并返回需要阈值处理的概率值。我们探讨了为检索训练的生物医学句子编码器是否是此类模型的良好起点。我们提出了 SBERT2S1，它将 Sentence-Transformers 编码器转换为双编码器、跨头（C）和先验融合残差（PFR）决策模型，同时提供 BIODECIDE，一个生物医学类型化决策套件，以及 MEDLINE-S1，包含由 NLM 索引生成的 243k 个训练决策。在六个父检索器对中，检索训练能改善内容承载选项的零样本匹配。微调后，其效果取决于头类型：在五对和三种训练集规模下，检索训练在 15 次比较中，有 10 次显著帮助保留检索先验的 PFR，帮助 C 仅一次且在五次中有负面影响。一个涵盖两种头和五个训练目标的匹配网格显示，在每个目标下，C 都优于 PFR，同时开放系统 One 模型发布的 RLCD 配方比交叉熵落后 2.5-3.0 个点。差距主要源于其奖励归一化，将噪声得分函数项放大了 3.6-15 倍；无偏的一次留出估计器恢复了大部分差距。经过温度缩放后，没有任何目标明显比交叉熵校准更好。我们发布了代码、MEDLINE-S1 标签和一个模型。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02486

Authors: Pritam Deka

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.02486.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02486

Published: 2026-10-05T02:23:35.825Z

8. A generative-informed neuro-symbolic framework for syntactic ambiguity resolution: Evidence from Arabic DPs

Abstract:Syntactic ambiguity poses a persistent challenge for Arabic NLP, particularly in morphologically rich nominal constructions where multiple structu6ral interpretations may be compatible with the same surface sequence. This study proposes a generatively informed neuro-symbolic framework for resolving structural ambiguity in Modern Standard Arabic (MSA) DPs. The framework integrates generative syntactic notions with AraBERT by representing ambiguity as a candidate-based decision task in which linguistically motivated alternatives are explicitly constructed and evaluated through candidate-conditioned input representations. Findings indicate that the model achieved 96.88% accuracy, 95.92% macro-F1, 96.83% weighted F1, and 93.94% binary F1 on the unseen evaluation set. Class-level analysis revealed asymmetric performance, with recall of 99.71% for High/VP Attachment (N1) and 89.26% for Low/NP/Embedded Attachment (N2), indicating greater difficulty in recovering the embedded interpretation. The study concludes that formal syntactic representations can be operationalized within Transformer-based NLP as an explicit interface between linguistic structure and contextual neural modeling, providing a controlled and interpretable approach to Arabic syntactic ambiguity resolution and beyond.

中文摘要

摘要：句法歧义对阿拉伯语自然语言处理（NLP）构成了持续的挑战，尤其是在形态丰富的名词结构中，其中多个结构解释可能与相同的表层序列兼容。本研究提出了一种受生成语法启发的神经符号框架，用于解决现代标准阿拉伯语（MSA）名词短语（DPs）中的结构歧义。该框架将生成句法概念与AraBERT相结合，通过将歧义表示为基于候选的决策任务来操作，其中语言学动机的备选方案被明确构建，并通过候选条件输入表示进行评估。研究结果表明，该模型在未见评估集上实现了96.88%的准确率，95.92%的宏平均F1，96.83%的加权F1，以及93.94%的二分类F1。类别级分析显示性能存在不对称性，高/动词短语附着（N1）的召回率为99.71%，低/名词短语/嵌入附着（N2）的召回率为89.26%，表明恢复嵌入解释更具挑战性。研究结论认为，形式句法表示可以在基于Transformer的NLP中被操作化，作为语言结构与上下文神经建模之间的显式接口，为阿拉伯语句法歧义解析及更广泛的应用提供了一种可控且可解释的方法。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02529

Authors: Mohammed Damom, Muneef Y. Alshawsh, Ashraf A. Naji, Mustafa Ali Alhamzi, Fawwaz An-Nashef, Jameel Ahmed Elayah, Mohammed Q. Shormani, Noman AL-Sayadi

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.02529.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02529

Published: 2026-10-05T02:23:35.825Z

9. Evaluating Multi-Dimensional Generalization of Large Language Models in Temporal Extraction Tasks

Abstract:Time and event expression extraction are fundamental temporal reasoning tasks, but the problem remains difficult due to annotation ambiguity, domain sensitivity, and unstable model behavior. Existing evaluations focus on in-domain performance, offering limited insight into reliability under distribution shifts. We evaluate multiple model configurations across families, architectures, and reasoning strategies over four dimensions of generalization, examining transfer from base performance, cross-dimensional correlations, and the effects of scale, architecture, and prompting. This provides a systematic study of how prompted LLMs generalize in time and event expression extraction tasks. We find that strong base-task performance generally predicts better generalization. However, this relationship weakens under substantial distribution shifts. Inductive prompting performs most consistently across domain shift, adversarial perturbations, compositionality, and length increase, while gains from scale, architecture, and deductive and abductive prompting strategies are uneven and dimension-specific. We conclude that LLM generalization in temporal extraction tasks cannot be predicted from any single dimension alone and cannot be reliably inferred from in-domain or single-dimension evaluations, highlighting the need for reasoning strategies that generalize across dimensions.

中文摘要

摘要：时间和事件表达提取是基本的时间推理任务，但由于标注歧义、领域敏感性和模型行为不稳定，这一问题仍然困难。现有评估主要关注域内性能，对于分布变化下的可靠性提供的见解有限。我们在四个泛化维度上评估了跨模型家族、架构和推理策略的多种模型配置，考察了从基础性能的迁移、跨维度相关性以及规模、架构和提示方式的影响。这提供了对提示型大语言模型在时间和事件表达提取任务中泛化的一项系统研究。我们发现，强大的基础任务性能通常预示着更好的泛化能力。然而，在存在显著分布变化的情况下，这种关系会减弱。归纳提示在域变化、对抗扰动、组合性和长度增加方面表现最为一致，而规模、架构以及演绎和溯因提示策略的提升则不均衡且具有维度特异性。我们总结认为，LLM在时间提取任务中的泛化能力不能仅通过单一维度预测，也不能可靠地从域内或单维度评估中推断，这突显了开发能够跨维度泛化的推理策略的必要性。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02549

Authors: Fahmid Shahriar Iqbal, Ritam Dutt, Soumitra Das, Arnav Verma, Sagnik Ray Choudhury

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.02549.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02549

Published: 2026-10-05T02:23:35.825Z

10. Learning When to Commit from Partial Speech for End-to-End Simultaneous Speech Translation

Abstract:Simultaneous speech translation must emit useful target text before the source is complete while preserving every committed token. We adapt a full-utterance speech language model using prefix supervision derived from its own complete- and partial-waveform translations, requiring neither transcripts nor human translations. We compare single-turn forced-prefix and multi-turn append-only decoding, use a confidence threshold to control the inference-time quality—latency trade-off, and vary the density of training prefixes with a separate synthesis margin. On FLEURS and CoVoST2 in three language directions, prefix training improves quality—latency frontiers over the unadapted model, and confidence provides the broadest consistently competitive operating range. Multi-turn decoding is generally stronger at low latency; under multi-turn training, commit-calibration error falls by 63—68% overall and 68—80% at early prefixes, whereas single-turn training provides only modest overall calibration gains and no early-prefix improvement. A small synthesis margin sometimes extends the frontier to lower latency, particularly on shorter utterances, while a larger margin degrades translation quality and calibration. Prefix adaptation therefore improves simultaneous speech translation, especially under multi-turn append-only decoding, while synthesis density introduces a non-monotonic quality—latency trade-off.

中文摘要

摘要：即时语音翻译必须在源语言尚未完成时就输出有用的目标文本，同时保留每一个已提交的标记。我们使用从自身完整和部分波形翻译中获得的前缀监督来调整完整话语的语音语言模型，这不需要转录稿或人工翻译。我们比较了单轮强制前缀解码和多轮仅追加解码，使用置信度阈值来控制推理时的质量–延迟权衡，并通过单独的合成边距来改变训练前缀的密度。在FLEURS和CoVoST2的三种语言方向上，前缀训练改善了未调整模型的质量–延迟前沿，而置信度提供了最广泛且持续竞争的操作范围。多轮解码在低延迟下通常更强；在多轮训练下，提交校准误差整体下降63–68%，在早期前缀下降68–80%，而单轮训练仅提供适度的整体校准提升，并且对早期前缀没有改善。一些小的合成边距有时会将前沿延伸到更低延迟，尤其是在较短的语句上，而较大的边距会降低翻译质量和校准。因此，前缀适配提升了即时语音翻译性能，尤其是在多轮仅追加解码下，而合成密度引入了非单调的质量–延迟权衡。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02612

Authors: Hieu Hoang, Amittai Axelrod

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.02612.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02612

Published: 2026-10-05T02:23:35.825Z

Agent Domain Papers

1. MintFlow: Minimal Trajectory Intervention for Constrained Flow Matching

Abstract:Flow matching models excel at generative modeling, and many downstream applications require their samples to satisfy prescribed constraints, such as observed measurements and physical laws. However, existing constrained samplers often face a trade-off: \textit{enforcing constraints can substantially displace samples from the pretrained data distribution}. To address this trade-off, we introduce \textbf{MintFlow}, a training-free constrained sampling framework that formulates constraint enforcement as a minimal intervention on the pretrained flow trajectory. MintFlow seeks the minimal perturbation of an intermediate flow state such that its subsequent evolution under the pretrained flow field satisfies the target constraint. By minimally perturbing the flow state while keeping the pretrained flow field unchanged, MintFlow enforces the constraint while minimizing unnecessary deviation from the pretrained distribution. An adjoint formulation yields a closed-form expression for this perturbation, eliminating expensive iterative optimization. Furthermore, MintFlow adaptively selects the intervention time to balance the required perturbation magnitude with its amplification by the remaining flow. Across a range of tasks in generative vision and physical system modeling, MintFlow achieves competitive constraint satisfaction while preserving the pretrained generative distribution substantially better than state-of-the-art constrained methods.

中文摘要

摘要：流匹配模型在生成建模方面表现出色，许多下游应用要求其样本满足预设约束条件，例如观测测量值和物理定律。然而，现有的受约束采样器常常面临一个权衡：执行约束可能会显著偏离预训练数据分布。为了解决这一权衡，我们提出了MintFlow，一种无需训练的受约束采样框架，它将约束执行表述为对预训练流轨迹的最小干预。MintFlow旨在对中间流状态进行最小扰动，使其在预训练流场下的后续演化满足目标约束。通过最小扰动流状态同时保持预训练流场不变，MintFlow在最小化偏离预训练分布的同时实现约束。伴随方法的表述为该扰动提供了闭式表达式，避免了昂贵的迭代优化。此外，MintFlow自适应地选择干预时间，以平衡所需扰动幅度与剩余流的放大效应。在生成视觉任务和物理系统建模的多个任务中，MintFlow在实现竞争性的约束满足的同时，比最先进的受约束方法更好地保留了预训练生成分布。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02260

Authors: Yesom Park, Kelvin Kan, Qifan Chen, Thomas Flynn, Hayden Schaeffer. Xihaier Luo

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02260.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02260

Published: 2026-10-05T02:24:01.367Z

2. Fast Models, Slow Evidence: A Paired and Self-Audited Evaluation of System-1 Decision Models for LLM Agent Harnesses

Abstract:Agent harnesses make many small, typed decisions per task: which model to call, which tool to use, whether retrieved text is relevant, whether an input carries an injection. System-1 decision models answer such questions in a single forward pass with class probabilities, promising large cost and latency savings over LLM calls. We present a paired evaluation of an open-weight (Laya) and a hosted (Jev) System-1 model on 11 agent decision points built from 18 public sources: 7,283 base cases plus 6,640 robustness variants, with byte-identical inputs, paired tests, and cross-hardware and cross-day reproducibility checks. Jev is significantly more accurate on 9 of 11 decision points (+10.8 to +46.0 pp). Neither model beats chance on zero-shot model routing, and they tie on RAG relevance gating. Laya changes 30% of its answers when the option order is reversed and degrades sharply with many or similar candidates (31% at 50 nearest-neighbour tools, vs. 98% for Jev on items with a unique correct tool). We also audit our own pipeline. Three analysis errors and one design confound distorted headline deployment claims: an omitted pre-screen cost (reported 23.9% saving, actual 4.3%), gate accuracy reported as end-to-end quality (58% vs. 98%), in-sample thresholds (5% target, up to 17% held-out misses), and a “channel effect” on injection false positives that vanishes with channel-native content. Two other suspected confounds did not change the conclusions. All cases, raw outputs and analysis code are available at this https URL.

中文摘要

摘要：代理机束在每个任务中做出许多小的类型决策：调用哪个模型、使用哪个工具、检索文本是否相关、输入是否携带注入。System-1决策模型通过一次前向传递以类概率回答这些问题，承诺相较LLM调用节省大量成本和延迟。我们对开放权重（Laya）和托管（Jev）System-1模型进行了配对评估，基于11个代理决策点，基于18个公开来源构建：7,283个基案例和6,640个鲁棒性变体，具备字节相同输入、配对测试以及跨硬件和跨日重复性检查。Jev在11个决策点中的9个（+10.8至+46.0 pp）上显著更准确。这两个模型在零样本模型路由上均不及偶然性，且在RAG相关性门控上并列。当选项顺序反转且对多个或相似候选工具出现显著下降时，Laya会改变30%的答案（50个最近邻工具时为31%，Jev在唯一正确工具下为98%）。我们还审计了自己的流程。三个分析错误和一个设计混淆了扭曲的标题部署声明：遗漏的预筛成本（报告节省23.9%，实际4.3%），门准确率报告为端到端质量（58%对98%），样本内阈值（5%目标，最高17%未中目标），以及注入误报的“通道效应”，随着通道原生内容消失。另外两个疑似混淆未改变结论。所有案例、原始输出和分析代码均可在此 https 网址获取。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02267

Authors: Jiawei Li

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02267.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02267

Published: 2026-10-05T02:24:01.367Z

3. The AI Risk Observatory: What Can We Learn from AI Disclosures in Annual Reports About Societal Resilience?

Abstract:Societal resilience research relies on access to useful and actionable data, which motivates our main research question: Can annual reports, processed at scale with LLMs, provide a useful signal about how companies disclose their response to AI? We test this by applying a reproducible two-stage classification pipeline to 9,821 annual reports from 1,362 UK listed companies (2020-2025, with partial 2026 data). We first validate the method against 474 human-annotated passages, finding high recall and moderate label-level agreement. We then report three empirical patterns: (i) between 2020 and 2025, the share of reports mentioning AI risk rose from 2.8% to 41.2%, while AI adoption disclosure also rose, from 13.8% to 45.2%, and named vendor mentions cluster around a small set of major providers led by Microsoft; (ii) disclosure varies substantially by Critical National Infrastructure sector and market segment: AIM reports disclose AI risk at far lower rates than Main Market reports, and sectors such as Energy and Data Infrastructure lag behind the rest in AI risk disclosure; and (iii) harm disclosures are near-absent (seven reports across the entire corpus). We develop a substantiveness classification to assess the quality of the disclosure and find that most AI risk disclosure is not substantive: in 2025, 41.2% of all reports mention AI as a risk, but only 4.3% contain AI risk disclosure we classify as substantive.

中文摘要

摘要：社会韧性研究依赖于获取有用且可操作的数据，这促使我们提出了主要研究问题：年度报告在大规模使用大型语言模型（LLMs）处理后，能否提供有关公司披露其应对人工智能措施的有用信号？我们通过将可重复的两阶段分类流程应用于1,362家英国上市公司（2020-2025年，部分2026年数据）的9,821份年度报告来测试这一问题。我们首先将该方法与474段人工标注的文本进行验证，发现召回率高，标签级别一致性中等。然后，我们报告三个实证模式：(i) 在2020至2025年间，提及AI风险的报告比例从2.8%上升至41.2%，同时AI采用披露也从13.8%上升至45.2%，且所提及的供应商主要集中在以微软为主的一小部分主要供应商；(ii) 披露情况在关键国家基础设施部门和市场细分中差异显著：AIM市场报告披露AI风险的比例远低于主板市场报告，能源和数据基础设施等部门在AI风险披露方面落后于其他部门；(iii) 损害披露几乎不存在（整个语料库中仅有七份报告）。我们开发了一种实质性分类方法来评估披露质量，发现大多数AI风险披露并不具有实质性：在2025年，41.2%的报告提及AI作为风险，但仅有4.3%的报告包含我们认定为实质性的AI风险披露。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02281

Authors: Bart Jaworski

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02281.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02281

Published: 2026-10-05T02:24:01.367Z

4. Keep It CALM: Analyzing the Limits of Global Unsafety in Text-to-Image Generation

Abstract:Training-free safeguards for text-to-image generation often rely on a reusable safety signal, such as an unsafe direction or global toxic subspace, applied broadly across prompts. We provide a controlled geometric analysis of this global-unsafety assumption and reveal a consistent coverage-selectivity trade-off: compact unsafe subspaces fail to cover heterogeneous unsafe semantics, whereas broader aggregation increasingly distorts safety-adjacent benign prompts. Motivated by this finding, we propose CALM (Counterfactual Adaptive Local Modulation), a training-free safeguard that replaces uniform global removal with prompt-local counterfactual correction. Using matched unsafe-benign anchors, CALM routes each prompt to active unsafe categories, minimally edits only violating token representations toward the safe side, and suppresses positively aligned unsafe residual components. Across broad evaluation, CALM significantly improves unsafe content suppression while preserving benign utility, demonstrating that local counterfactual correction provides a more selective alternative to global unsafe signal removal.

中文摘要

摘要：面向文本到图像生成的免训练安全防护通常依赖可重复使用的安全信号，例如不安全方向或全局有害子空间，并广泛应用于各种提示。我们对该全局不安全假设提供了受控几何分析，并揭示了一个一致的覆盖-选择性权衡：紧凑的不安全子空间无法覆盖异质的不安全语义，而更广的聚合则会越来越扭曲安全邻近的良性提示。受此发现启发，我们提出了 CALM（反事实自适应局部调制），这是一种免训练安全防护方法，将统一的全局移除替换为提示局部的反事实修正。通过匹配的不安全-良性锚点，CALM 将每个提示路由至活跃的不安全类别，仅对违反规则的 token 表示进行最小化编辑以朝安全方向调整，并抑制正向对齐的不安全残余成分。在广泛评估中，CALM 显著改善了不安全内容的抑制，同时保留了良性效用，证明局部反事实修正提供了比全局不安全信号移除更具选择性的替代方案。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02300

Authors: NaHyeon Park, Minhyun Lee, Hyunjung Shim

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02300.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02300

Published: 2026-10-05T02:24:01.367Z

5. Choosing Before Acting: Comparative Value Estimation for Long-Horizon Tool-Use Agents

Abstract:Large language models (LLMs) rely on long-horizon tool invocation sequences for complex tasks, where each invocation can alter the task state and condition subsequent decisions. In long-horizon tool use, final-outcome rewards provide weak credit assignment over long interaction traces. Step-level rewards can offer more targeted feedback, but obtaining reliable step supervision often requires human or LLM judgment, or additional rollouts to estimate the downstream effect of an intermediate decision. In this paper, we argue that effective tool-use agents should estimate the long-horizon value of a possible next tool invocation before executing it. This objective requires comparative supervision over alternative invocations under the same context, while logged trajectories only contain the invocation that was actually taken. Therefore, we propose Comparative Inference for Tool-use Agents (CITA). CITA trains a Comparative Inference Model (CIM) from paired signals that combine observed tool behavior, scalable supervision from a Bayesian tool-graph simulator, and semantic judgments from LLM-based comparison. The resulting CIM learns to estimate how likely a possible next tool invocation is to support final task success under the current context. Across three tool-use benchmarks and multiple backbone LLMs, CITA consistently improves Tool F1 and task success. Additional analysis shows that CIM learns accurate step-level value estimates for comparative tool choices.

中文摘要

摘要：大型语言模型（LLM）在执行复杂任务时依赖于长周期的工具调用序列，其中每一次调用都可能改变任务状态并影响随后决策。在长周期的工具使用中，最终结果的奖励在长时间的交互轨迹中很难提供有效的信用分配。步骤级奖励可以提供更有针对性的反馈，但获取可靠的步骤监督通常需要人工或LLM判断，或者需要额外的回滚以估计中间决策的下游影响。在本文中，我们认为有效的工具使用代理应在执行可能的下一步工具调用之前，评估其长周期价值。这个目标要求在相同上下文中对不同调用进行比较监督，而日志轨迹中仅包含实际发生的调用。因此，我们提出了工具使用代理的比较推理方法（CITA）。CITA利用观测到的工具行为、来自贝叶斯工具图模拟器的可扩展监督以及基于LLM的语义比较等配对信号训练比较推理模型（CIM）。由此得到的CIM能够估计在当前上下文下，可能的下一步工具调用支持最终任务成功的可能性。在三个工具使用基准测试以及多个主干LLM中，CITA持续提升了工具F1分数和任务成功率。进一步分析显示，CIM能够为比较工具选择学习准确的步骤级价值估计。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02330

Authors: Yu Li, Zheng Zhang, Xin Liu, Shengtian Yang, Guangfeng Cai, Lei Feng

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02330.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02330

Published: 2026-10-05T02:24:01.367Z

6. World Editing: Intervening on Executable Worlds at Increasing Depth

Abstract:Interactive world models are increasingly capable of generating environments and acting within them, yet deliberately editing an existing executable world remains underexplored. We formulate world editing as intervening on an existing world while preserving properties that should remain unchanged, and introduce intervention depth as an axis describing how strongly an edit couples world entities, dynamics, and systems. We instantiate this capability through industry-grade game modding and introduce IGMWorld, together with IGMBench, a benchmark of 110 tasks and over 1.1K executable state and behavioral criteria across Minecraft and Terraria. The tasks span property, entity, dynamics, and system interventions and are evaluated through deterministic executability, behavioral, preservation, and visual checks. Frontier coding agents already exhibit substantial world-editing capability: the strongest configuration solves 78.2% of tasks under a strict task-level criterion, while criterion-level performance reaches 94.8%. Reliability generally decreases with intervention depth, and this pattern persists even among tasks with similar numbers of evaluation criteria. Most failed edits still build and load successfully, suggesting that the main difficulty is making the edited world behave as requested. Visual consistency remains a separate weakness, with all evaluated configurations below 50% joint visual pass rate. These results show that world editing is a distinct capability from world generation and interaction, and that executable games provide a practical testbed for studying it.

中文摘要

摘要：交互式世界模型越来越能够生成环境并在其中行动，但有意地编辑一个现有的可执行世界仍未被充分探索。我们将世界编辑定义为在干预现有世界的同时保持应保持不变的属性，并引入干预深度作为描述编辑如何强烈地耦合世界实体、动态和系统的维度。我们通过行业级游戏模组实现了这一能力，并推出了IGMWorld，以及IGMBench，这是一个包含110个任务和超过1.1K可执行状态和行为标准的基准，覆盖Minecraft和Terraria。这些任务涵盖了属性、实体、动态和系统干预，并通过确定性的可执行性、行为、保持性和视觉检查进行评估。前沿编码代理已经展现出显著的世界编辑能力：最强配置在严格的任务级标准下解决了78.2%的任务，而标准级性能达到了94.8%。可靠性通常随干预深度而下降，这种模式即使在评估标准数量相似的任务中也会持续存在。大多数失败的编辑仍然能够成功构建和加载，这表明主要困难在于使编辑后的世界按要求表现。视觉一致性仍然是一个独立的弱点，所有评估配置的联合视觉通过率均低于50%。这些结果表明，世界编辑是一种不同于世界生成和交互的能力，而可执行游戏为研究其提供了一个实际的测试平台。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02331

Authors: Max Ku, Nok-Kan Law, Yu-Chien Tang, Shih-Ying Yeh, Ping Nie, Andy Zheng, Tat Hei Lai, Fei-Yueh Chen, Nikko Yu, Wei-Chieh Sun, Suzy Huang, Chiao-Wei Hsu, Chih-Chuan Huang, Chak-Wing Mak, Ho Yin Sam Ng, Edisy Kin Wai Chan, Min-Hung Chen, Ho Kei Cheng

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02331.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02331

Published: 2026-10-05T02:24:01.367Z

7. A Multi Method Importance and Performance Efficiency Analysis of Topological Metrics for Natural Visibility Graph Based Cyber Attack Detection

Abstract:Natural Visibility Graph (NVG) based analysis characterizes network traffic through topological descriptors reflecting different structural properties. However, not all descriptors contribute equally to cyber-attack classification, and extracting a large metric set can increase computational cost. This study evaluates 21 NVG derived topological metrics and investigates whether a compact subset can preserve classification capability while improving computational efficiency. Four importance analysis methods SHAP, grouped Permutation Importance, Boruta, and Recursive Feature Elimination (RFE) are integrated through a Consensus Ranking strategy. Based on this ranking, Full21, Top15, Top10, Top7, Top5, and Top3 configurations are evaluated using the CICIDS2018 dataset, a CNN classifier, and stratified 5 fold cross validation. The three highest ranked metrics are avg_clustering_coeff_median, avg_clustering_coeff_std, and avg_clustering_coeff_mean. Top3 achieved the highest observed mean performance, with 97.148% accuracy, 97.055% weighted F1 score, and an MCC of 0.9675, compared with 95.999%, 95.521%, and 0.9549 for Full21, respectively. It also reduced total runtime from 14,961.39 s to 589.22 s (96.06%). These results indicate that importance guided metric reduction can provide a compact NVG representation with higher observed mean predictive performance and substantially lower computational cost under the evaluated setting.

中文摘要

摘要：基于自然可视化图（NVG）的分析通过反映不同结构特性的拓扑描述符来表征网络流量。然而，并非所有描述符在网络攻击分类中贡献相同，而提取大量指标会增加计算成本。本研究评估了21个来源于NVG的拓扑指标，并探讨了是否可以通过一个紧凑的子集在保持分类能力的同时提高计算效率。将四种重要性分析方法——SHAP、分组置换重要性（grouped Permutation Importance）、Boruta和递归特征消除（RFE）——通过共识排名策略进行整合。基于此排名，对Full21、Top15、Top10、Top7、Top5和Top3配置使用CICIDS2018数据集、CNN分类器和分层5折交叉验证进行了评估。排名前三的指标为平均聚类系数中位数（avg_clustering_coeff_median）、平均聚类系数标准差（avg_clustering_coeff_std）和平均聚类系数均值（avg_clustering_coeff_mean）。Top3配置实现了最高的观测平均性能，准确率为97.148%，加权F1得分为97.055%，MCC为0.9675，而Full21配置分别为95.999%、95.521%和0.9549。Top3配置的总运行时间也从14,961.39秒减少到589.22秒（减少96.06%）。这些结果表明，在所评估的情况下，基于重要性指导的指标缩减可以提供一个紧凑的NVG表示，同时获得更高的观测平均预测性能和显著降低的计算成本。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02342

Authors: Ali Melih Kanca, Ilker Turker

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02342.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02342

Published: 2026-10-05T02:24:01.367Z

8. DeReAct: Decomposed Reasoning and Acting for Reliable AI Agents

Abstract:ReAct-based agents typically rely on a single LLM policy to propose actions, interact with the environment, and decide when a task is complete. This coupling makes action authorization and completion control difficult to enforce independently, allowing errors to propagate and unsupported completion claims to terminate execution. We introduce DeReAct, a modular agent architecture that externalizes two gating policies: a Critic that validates proposed actions before execution, and a Context Manager that reconstructs an environment-supported \textsc{State} and certifies task completion. Across GAIA and SWE-bench Verified, DeReAct improves Pass@1 most for weaker Brain models, with gains of 6.5—7.0 points for Qwen3-Coder-480B and 4.2—5.2 points for Claude Sonnet~4.5; gains diminish as Brain capability increases. Trajectory and ablation analyses show that external gating is effective when targeted failures are sufficiently prevalent and the gating policy is itself sufficient. With Claude Opus~4.5, Pass@1 remains comparable to ReAct, while DeReAct produces more evidence-complete and constraint-satisfying trajectories, indicating that completion control can trade earlier termination for stronger grounding. Overall, DeReAct improves weaker agents while retaining grounding benefits as models strengthen.

中文摘要

摘要：基于ReAct的智能体通常依赖单一的大语言模型（LLM）策略来提出动作、与环境交互以及决定任务何时完成。这种耦合使得动作授权和完成控制难以独立执行，从而导致错误传播和不支持的完成声明终止执行。我们提出了DeReAct，一种模块化的智能体架构，将两个门控策略外部化：一个Critic用于在执行前验证提出的动作，另一个Context Manager用于重建环境支持的状态（ extsc{State}）并认证任务完成。在GAIA和SWE-bench Verified中，DeReAct在弱型Brain模型上最显著提升Pass@1，对于Qwen3-Coder-480B提升6.5—7.0点，对于Claude Sonnet~4.5提升4.2—5.2点；随着Brain能力的提升，增益减小。轨迹和消融分析表明，当目标性失败足够普遍且门控策略自身充分时，外部门控是有效的。对于Claude Opus~4.5，Pass@1与ReAct相当，而DeReAct生成的轨迹在证据完整性和约束满足性方面更好，表明完成控制可以用较早的终止换取更强的依据。总体而言，DeReAct在增强弱型智能体的同时，在模型能力增强时仍保留了依据优势。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02351

Authors: Ajay Vohra, Tao Chen, Neeti Narayan, Caron Zhang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02351.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02351

Published: 2026-10-05T02:24:01.367Z

9. Traversing the Satisfaction-Diversity Frontier in Text-to-Image Diffusion

Abstract:Text-to-image generation enables users to explore several images generated from the same prompt. For these generated images to be useful, each one must reflect the user’s preferences, measured by a learned reward, and differ visually from the others to maintain diversity. Existing methods are limited: they either address reward and diversity separately or combine them in one aggregate score, enabling high diversity to offset low rewards. In this paper, we address these limitations by formulating generation as satisficing: every image (candidate) must satisfy a reward floor and the batch of images must satisfy a diversity cutoff. The reward floor controls the balance between worst-candidate reward and batch diversity; we show that varying this floor defines a Pareto frontier. To traverse this frontier, we introduce SatisDive, a training-free inference-time method. SatisDive uses a batch-relative reward cutoff to distinguish lower- from higher-reward candidates, emphasizing reward improvement for candidates below the cutoff and diversity among candidates above it. On Pick-a-Pic, at matched DreamSim, SatisDive improves worst-candidate reward over FK steering by up to 0.43 with FLUX.1-dev as the base model and HPSv3 as the reward, and by up to 0.70 with SANA-1.6B as the base model and ImageReward as the reward. More broadly, across their overlapping DreamSim ranges, SatisDive’s satisfaction-diversity curve Pareto-dominates FK steering’s curve in each setting.

中文摘要

摘要：文本生成图像使用户能够探索由相同提示生成的多张图像。为了使这些生成的图像有用，每张图像必须反映用户的偏好，这通过学习的奖励来衡量，并且在视觉上要与其他图像不同，以保持多样性。现有方法存在局限：它们要么分别处理奖励和多样性，要么将它们组合成一个总体分数，从而允许高多样性来弥补低奖励。在本文中，我们通过将生成问题表述为满意化问题来解决这些局限：每张图像（候选）必须满足奖励下限，并且整批图像必须满足多样性阈值。奖励下限控制最差候选奖励与批量多样性之间的平衡；我们表明，改变该下限定义了帕累托前沿。为了遍历这一前沿，我们提出了SatisDive，这是一种无需训练的推理时方法。SatisDive 使用相对于批次的奖励下限来区分低奖励和高奖励候选，强调对低于下限的候选进行奖励提升以及对高于下限候选的多样性。在 Pick-a-Pic 上，在匹配的 DreamSim 条件下，SatisDive 在基模型为 FLUX.1-dev 且奖励为 HPSv3 时，比 FK 引导在最差候选奖励上提升多达 0.43；在基模型为 SANA-1.6B 且奖励为 ImageReward 时，提升多达 0.70。更广泛地说，在它们重叠的 DreamSim 范围内，SatisDive 的满意-多样性曲线在每种设置中帕累托支配 FK 引导的曲线。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02372

Authors: Kevin Zhai, Siva Rajesh Kasa, Soumya Roy, Sumit Negi, Mubarak Shah

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02372.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02372

Published: 2026-10-05T02:24:01.367Z

10. THPL: A Vision-to-Language Decision Support Framework for Rainbow Trout Feeding Management in RAS

Abstract:In Recirculating Aquaculture Systems (RAS), precision feeding is critical for minimizing costs and improving fish welfare. However, existing methods lack cognitive alignment between fish behaviors and management knowledge, impeding translation into executable, interpretable feeding decisions. To address this, we propose THPL, a generative feeding decision framework tailored for rainbow trout (Oncorhynchus mykiss) in RAS. First, Fishsort extracts trajectories to establish an Activity Coefficient (AC) quantifying feeding intensity. Second, a Hierarchical Behavior Encoder (HBE) models individual temporal progression and collective dynamics using Temporal and Set Transformers, transforming trajectory tensors into dual-evidence representations of explicit physical and implicit soft tokens. Finally, these tokens are integrated with environmental parameters, metadata, and expert rules to fine-tune an LLM via LoRA, followed by counterfactual multimodal Direct Preference Optimization (mDPO) to reinforce causal reasoning. Results show that AC exhibits a statistically significant monotonic positive correlation with expert-annotated feeding intensity (Spearman $\rho = 0.925$, $p < 0.001$). Ablations indicate that decision accuracy improves from 33.33% (text-only baseline) to 93.33% with dual-evidence tokens, confirming that continuous spatiotemporal tokens provide necessary physical grounding for LLMs. Compared with standard LoRA, counterfactual mDPO elevates decision accuracy from 93.33% to 96.67%, advances METEOR from 58.10% to 85.30%, reduces Self-BLEU-2 from 58.79% to 52.88%, and increases Distinct-3 from 6.68% to 7.81%, suppressing templating and actuation biases while reinforcing causal consistency and operational safety. Overall, by integrating continuous kinematics with LLM reasoning, this study provides a novel decision support paradigm for precision aquaculture.

中文摘要

摘要：在循环水养殖系统（RAS）中，精准饲喂对于降低成本和改善鱼类福利至关重要。然而，现有方法缺乏鱼类行为与管理知识之间的认知对齐，阻碍了可执行且可解释的饲喂决策的转化。为此，我们提出了THPL，一种针对循环水系统中虹鳟（Oncorhynchus mykiss）的生成式饲喂决策框架。首先，Fishsort提取轨迹以建立量化饲喂强度的活动系数（AC）。其次，层级行为编码器（HBE）使用时间和集合变换器建模个体时间进程和集体动态，将轨迹张量转化为明确的物理信息和隐含软标记的双重证据表示。最后，这些标记与环境参数、元数据及专家规则结合，通过LoRA微调大语言模型（LLM），随后使用反事实多模态直接偏好优化（mDPO）强化因果推理。结果显示，AC与专家标注的饲喂强度具有显著单调正相关（Spearman $
ho = 0.925$, $p < 0.001$）。消融实验表明，决策准确率从33.33%（仅文本基线）提高到93.33%，确认连续时空标记为LLM提供了必要的物理基础。与标准LoRA相比，反事实mDPO将决策准确率从93.33%提升至96.67%，将METEOR从58.10%提升至85.30%，将Self-BLEU-2从58.79%降低至52.88%，将Distinct-3从6.68%提高至7.81%，抑制了模板化和执行偏差，同时增强了因果一致性和操作安全性。总体而言，通过将连续运动学与LLM推理相结合，本研究为精准水产养殖提供了一种新型的决策支持范式。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02378

Authors: Meng Liang, Guanbo Feng, Haozhuang Chi, Shilong Zhao, Zhixin Xiong, Yuhang He, Wenfeng Han, Tianhao Zhao, Zhihong Ma, Ying Liu

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02378.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02378

Published: 2026-10-05T02:24:01.367Z

Evaluation Domain Papers

1. MintFlow: Minimal Trajectory Intervention for Constrained Flow Matching

中文摘要

摘要：流匹配模型在生成建模方面表现出色，许多下游应用要求其样本满足规定的约束条件，例如观测测量和物理定律。然而，现有的约束采样器往往面临一个权衡： extit{强制约束可能会显著偏离预训练数据分布}。为了解决这一权衡，我们提出了 extbf{MintFlow}，一个无需训练的约束采样框架，它将约束强制执行表述为对预训练流轨迹的最小干预。MintFlow 寻求对中间流状态的最小扰动，使其在预训练流场下的后续演化满足目标约束。通过在保持预训练流场不变的情况下最小化对流状态的扰动，MintFlow 在强制约束的同时尽量减少对预训练分布的不必要偏离。伴随（adjoint）公式提供了该扰动的闭式表达式，从而消除了昂贵的迭代优化。此外，MintFlow 自适应选择干预时间，以平衡所需扰动幅度与其在剩余流中的放大效应。在生成视觉和物理系统建模的一系列任务中，MintFlow 实现了具有竞争力的约束满足，同时在保持预训练生成分布方面显著优于最先进的约束方法。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02260

Authors: Yesom Park, Kelvin Kan, Qifan Chen, Thomas Flynn, Hayden Schaeffer. Xihaier Luo

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02260.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02260

Published: 2026-10-05T02:24:31.851Z

2. Fast Models, Slow Evidence: A Paired and Self-Audited Evaluation of System-1 Decision Models for LLM Agent Harnesses

中文摘要

摘要：代理程序在每个任务中会做许多小型、类型化的决策：调用哪个模型、使用哪个工具、检索的文本是否相关、输入是否包含注入。System-1 决策模型通过一次前向传播并输出类别概率来回答此类问题，相比调用大语言模型（LLM）有望节省大量成本和延迟。我们对一个开源权重模型（Laya）和一个托管模型（Jev）的 System-1 模型在 11 个代理决策点进行了配对评估，这些决策点由 18 个公共资源构建：7,283 个基础案例加 6,640 个鲁棒性变体，使用字节一致的输入、配对测试，并进行了跨硬件和跨日期的可重复性检查。在 11 个决策点中，Jev 在 9 个点上的准确率显著更高（高出 10.8 至 46.0 个百分点）。在零样本模型路由上，两者都未超越随机水平，在 RAG 相关性门控上则持平。当选项顺序反转时，Laya 会更改 30% 的答案，并在面对大量或相似候选项时迅速下降（在 50 个最近邻工具中为 31%，而 Jev 在具有唯一正确工具的条目上为 98%）。我们还审查了自己的流程。三处分析错误和一处设计混淆扭曲了头条部署声明：省略的预筛选成本（报告节省 23.9%，实际为 4.3%）、将门控准确率报告为端到端质量（58% 对比 98%）、样本内阈值（目标 5%，保留样本漏检高达 17%）、以及注入假阳性的“通道效应”，该效应在通道原生内容中消失。其他两个疑似混淆因素未改变结论。所有案例、原始输出和分析代码可通过此 https URL 获取。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02267

Authors: Jiawei Li

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02267.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02267

Published: 2026-10-05T02:24:31.851Z

3. The AI Risk Observatory: What Can We Learn from AI Disclosures in Annual Reports About Societal Resilience?

中文摘要

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02281

Authors: Bart Jaworski

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02281.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02281

Published: 2026-10-05T02:24:31.851Z

4. Keep It CALM: Analyzing the Limits of Global Unsafety in Text-to-Image Generation

中文摘要

摘要：面向文本到图像生成的免训练安全措施通常依赖于可重复使用的安全信号，例如不安全方向或全局有害子空间，并广泛应用于各种提示。我们对该全局不安全假设提供了受控几何分析，并揭示了一个一致的覆盖-选择性权衡：紧凑的不安全子空间无法覆盖异质的不安全语义，而更广的聚合则会越来越扭曲安全邻近的良性提示。受此发现启发，我们提出了 CALM（反事实自适应局部调制），这是一种免训练安全措施，用提示局部的反事实修正替代统一的全局移除。通过匹配的不安全-良性锚点，CALM 将每个提示路由到活跃的不安全类别，仅对违反的标记表示进行最小编辑以朝向安全方向，并抑制正向对齐的不安全残余成分。在广泛评估中，CALM 显著提高了不安全内容抑制能力，同时保持了良性实用性，证明局部反事实修正提供了比全局不安全信号移除更具选择性的替代方案。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02300

Authors: NaHyeon Park, Minhyun Lee, Hyunjung Shim

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02300.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02300

Published: 2026-10-05T02:24:31.851Z

5. Choosing Before Acting: Comparative Value Estimation for Long-Horizon Tool-Use Agents

中文摘要

摘要：大型语言模型（LLM）在执行复杂任务时依赖于长周期的工具调用序列，其中每一次调用都可能改变任务状态并影响随后决策。在长周期的工具使用中，最终结果的奖励在长时间的交互轨迹中很难提供有效的信用分配。步骤级奖励可以提供更有针对性的反馈，但获取可靠的步骤监督通常需要人工或LLM判断，或者需要额外的回滚以估计中间决策的下游影响。在本文中，我们认为有效的工具使用代理应在执行可能的下一步工具调用之前，评估其长周期价值。这个目标要求在相同上下文中对不同调用进行比较监督，而日志轨迹中仅包含实际发生的调用。因此，我们提出了工具使用代理的比较推理方法（CITA）。CITA利用观测到的工具行为、来自贝叶斯工具图模拟器的可扩展监督以及基于LLM的语义比较等配对信号训练比较推理模型（CIM）。由此得到的CIM能够估计在当前上下文下，可能的下一步工具调用支持最终任务成功的可能性。在三个工具使用基准和多种主干LLM上，CITA持续提升了工具F1分数和任务成功率。进一步分析显示，CIM能够为比较工具选择学习准确的步骤级价值估计。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02330

Authors: Yu Li, Zheng Zhang, Xin Liu, Shengtian Yang, Guangfeng Cai, Lei Feng

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02330.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02330

Published: 2026-10-05T02:24:31.851Z

6. World Editing: Intervening on Executable Worlds at Increasing Depth

中文摘要

摘要：交互式世界模型越来越能够生成环境并在其中行动，但有意地编辑一个现有的可执行世界仍未被充分探索。我们将世界编辑定义为在干预现有世界的同时保持应保持不变的属性，并引入干预深度作为描述编辑如何强烈地耦合世界实体、动态和系统的维度。我们通过行业级游戏模组实现了这一能力，并推出了IGMWorld，以及IGMBench，这是一个包含110个任务和超过1.1K可执行状态和行为标准的基准，覆盖Minecraft和Terraria。这些任务涵盖属性、实体、动态和系统的干预，并通过确定性的可执行性、行为、保持性和视觉检查进行评估。前沿编码代理已经展现出显著的世界编辑能力：最强配置在严格的任务级标准下解决了78.2%的任务，而在标准级别上的表现达到94.8%。可靠性通常随着干预深度而下降，即使在评估标准数量相似的任务中，这一模式仍然存在。大多数未成功的编辑仍然能够构建和加载成功，这表明主要困难在于使编辑后的世界按要求运行。视觉一致性仍然是一个独立的弱点，所有评估配置的联合视觉通过率均低于50%。这些结果表明，世界编辑是一种不同于世界生成和交互的能力，而可执行游戏为研究这一能力提供了实际的测试平台。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02331

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02331.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02331

Published: 2026-10-05T02:24:31.851Z

7. A Multi Method Importance and Performance Efficiency Analysis of Topological Metrics for Natural Visibility Graph Based Cyber Attack Detection

中文摘要

摘要：基于自然可视化图（NVG）的分析通过反映不同结构特性的拓扑描述符来表征网络流量。然而，并非所有描述符在网络攻击分类中贡献相同，而提取大量指标会增加计算成本。本研究评估了21个来源于NVG的拓扑指标，并探讨了是否可以通过一个紧凑的子集在保持分类能力的同时提高计算效率。四种重要性分析方法——SHAP、分组置换重要性（grouped Permutation Importance）、Boruta 和递归特征消除（RFE）——通过共识排名策略进行整合。基于该排名，分别对Full21、Top15、Top10、Top7、Top5和Top3配置进行了评估，使用CICIDS2018数据集、CNN分类器和分层5折交叉验证。排名前三的指标是avg_clustering_coeff_median、avg_clustering_coeff_std和avg_clustering_coeff_mean。Top3在观察到的平均性能上达到最高，准确率为97.148%，加权F1分数为97.055%，MCC为0.9675，而Full21分别为95.999%、95.521%和0.9549。Top3还将总运行时间从14,961.39秒减少到589.22秒（减少96.06%）。这些结果表明，在评估设置下，基于重要性的指标缩减可以提供一个更紧凑的NVG表示，同时具有更高的观察平均预测性能和显著较低的计算成本。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02342

Authors: Ali Melih Kanca, Ilker Turker

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02342.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02342

Published: 2026-10-05T02:24:31.851Z

8. DeReAct: Decomposed Reasoning and Acting for Reliable AI Agents

中文摘要

摘要：基于ReAct的智能体通常依赖单一的大语言模型（LLM）策略来提出动作、与环境交互以及决定任务何时完成。这种耦合使得动作授权和完成控制难以独立执行，从而导致错误传播和不支持的完成声明终止执行。我们提出了DeReAct，一种模块化的智能体架构，它外部化了两个门控策略：Critic（评论者）在执行前验证提议的动作，以及Context Manager（上下文管理器）重建环境支持的状态并认证任务完成情况。在GAIA和SWE-bench Verified中，DeReAct在弱型Brain模型上最显著提升了Pass@1，Qwen3-Coder-480B提升了6.5—7.0点，Claude Sonnet 4.5提升了4.2—5.2点；随着Brain能力提升，这些增益会减少。轨迹和消融分析表明，当目标失败足够普遍且门控策略自身足够有效时，外部门控是有效的。使用Claude Opus 4.5时，Pass@1仍与ReAct相当，而DeReAct生成的轨迹更为证据完整且满足约束，表明完成控制可以以更早的终止为代价换取更强的基础支持。总体而言，DeReAct改善了较弱智能体的表现，同时在模型增强时保留了基础支持的优势。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02351

Authors: Ajay Vohra, Tao Chen, Neeti Narayan, Caron Zhang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02351.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02351

Published: 2026-10-05T02:24:31.851Z

9. Traversing the Satisfaction-Diversity Frontier in Text-to-Image Diffusion

中文摘要

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02372

Authors: Kevin Zhai, Siva Rajesh Kasa, Soumya Roy, Sumit Negi, Mubarak Shah

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02372.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02372

Published: 2026-10-05T02:24:31.851Z

10. THPL: A Vision-to-Language Decision Support Framework for Rainbow Trout Feeding Management in RAS

中文摘要

摘要：在循环水养殖系统（RAS）中，精准饲喂对于降低成本和改善鱼类福利至关重要。然而，现有方法缺乏鱼类行为与管理知识之间的认知对齐，阻碍了可执行且可解释的饲喂决策的转化。为此，我们提出了THPL，一种针对循环水系统中虹鳟（Oncorhynchus mykiss）的生成式饲喂决策框架。首先，Fishsort提取轨迹以建立量化饲喂强度的活动系数（AC）。其次，层级行为编码器（HBE）使用时间和集合变换器建模个体时间进程和集体动态，将轨迹张量转化为明确的物理信息和隐含软标记的双重证据表示。最后，这些标记与环境参数、元数据及专家规则结合，通过LoRA微调大语言模型（LLM），随后使用反事实多模态直接偏好优化（mDPO）强化因果推理。结果显示，AC与专家标注的饲喂强度具有显著单调正相关（Spearman $
ho = 0.925$, $p < 0.001$）。消融实验表明，决策准确率从33.33%（仅文本基线）提高到93.33%，确认连续时空标记为LLM提供了必要的物理基础。与标准LoRA相比，反事实mDPO将决策准确率从93.33%提升至96.67%，将METEOR从58.10%提升至85.30%，将Self-BLEU-2从58.79%降低至52.88%，将Distinct-3从6.68%提高至7.81%，抑制了模板化和执行偏差，同时增强了因果一致性和操作安全性。总体而言，通过融合连续运动学与LLM推理，本研究为精准水产养殖提供了一种新颖的决策支持范式。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02378

Authors: Meng Liang, Guanbo Feng, Haozhuang Chi, Shilong Zhao, Zhixin Xiong, Yuhang He, Wenfeng Han, Tianhao Zhao, Zhihong Ma, Ying Liu

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.02378.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02378

Published: 2026-10-05T02:24:31.851Z

VLM Domain Papers

1. EditHero: A Benchmark for Long-Horizon Part-Level 3D Editing and Vibe Modeling

Abstract:3D editing methods are usually tested on a single edit, yet an asset is built through a long sequence of revisions, each of which must implement the requested change while leaving everything else unchanged. We introduce EditHero, to our knowledge the first benchmark for long-horizon, part-level 3D editing, with natural-language instructions and target images for both geometry and texture. A deterministic assembly engine produces the exact target after every edit, and every sequence is reviewed by hand. We use EditHero to compare 2 opposite approaches to 3D editing. Non-agentic methods operate top down, regenerating the object from a learned 3D representation and inferring what to keep. In contrast, LLM/VLM agents operate bottom up, editing through code that inspects the mesh and rewrites only the parts required by instructions. The non-agentic methods often miss the requested change and disturb regions that should stay fixed. Most LLMs follow instructions more closely, and all of them preserve the unedited parts better, but each of their edits takes minutes. We will release the engine and the edit sequences to support research on reliable iterative 3D editing.

中文摘要

摘要：3D 编辑方法通常只在单次编辑上进行测试，但一个资产是通过一长串修订构建的，每一次修订都必须实现所请求的更改，同时保持其他部分不变。我们介绍 EditHero，据我们所知，这是第一个用于长周期、部分级 3D 编辑的基准，提供几何和纹理的自然语言指令以及目标图像。一个确定性的装配引擎在每次编辑后生成确切的目标，每个序列都经过人工审查。我们使用 EditHero 对两种截然不同的 3D 编辑方法进行比较。非智能方法从上而下操作，从学习到的 3D 表示中重新生成对象并推断需要保留的部分。相比之下，LLM/VLM 代理从下而上操作，通过检查网格并仅重写指令要求修改的部分来进行编辑。非智能方法经常错过请求的更改，并扰乱应保持固定的区域。大多数 LLM 更严格地遵循指令，并且所有 LLM 都更好地保留未编辑的部分，但它们每次编辑都需要几分钟。我们将发布引擎和编辑序列，以支持可靠的迭代 3D 编辑研究。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02298

Authors: Ruihan Yu, Yu-Ju Tsai, Muyao Niu, Runyi Li, Lian Fu, Hanqing Liu, Zheng-Hui Huang, Yonghao Yu, Sho Kuno, Ming-Hsuan Yang, Kaipeng Zhang, Zhixiang Wang

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.02298.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02298

Published: 2026-10-05T02:24:56.485Z

2. DeskForge: Dense Supervision from Desktop Environments for Computer-Use Agents

Abstract:Computer-use agents need to reliably ground action targets in complex desktop scenes, where multiple applications, overlapping windows, and visually similar controls compete for attention. Existing training data rarely pair such scenes with dense annotations or vary them in a controlled way. We introduce DeskForge, a controllable desktop environment that composes and explores real applications to generate large-scale supervision for computer-use agents. It varies application states, content, window layout, appearance, and resolution, and fuses screenshots, accessibility trees, and window geometry into dense element annotations while recording the outcome of each executed action. Using this environment, we construct DeskForge-1M, a corpus of 1.2M annotated desktop observations containing 159.7M element instances. We fine-tune four vision-language models on 200K grounding examples drawn from DeskForge-1M. All four improve across held-out desktop conditions and on all five external GUI grounding benchmarks; for Qwen3.5-4B, accuracy increases by 11.51 percentage points on ScreenSpot-Pro and 10.11 points on OSWorld-G. The gains also translate to long-horizon task completion: under a fixed planner, the fine-tuned action models solve more WebArena-Infinity and OpenApps tasks, with Qwen3.5-4B increasing from 31 to 50 of 119 tasks and from 3 to 15 of 100 tasks, respectively. These results show that controllable composition of real desktop environments provides a scalable source of supervision for improving both GUI grounding and long-horizon computer use. The framework code, the dataset, and the fine-tuned model are available from the project page: this https URL

中文摘要

摘要：计算机使用代理需要在复杂的桌面场景中可靠地定位操作目标，这些场景中多个应用程序、重叠窗口以及视觉上相似的控件会争夺注意力。现有的训练数据很少将此类场景与密集注释配对，或以受控方式对其进行变化。我们引入了DeskForge，这是一个可控的桌面环境，它组合并探索真实应用程序，为计算机使用代理生成大规模监督数据。它可以变化应用程序状态、内容、窗口布局、外观和分辨率，并将屏幕截图、辅助性树和窗口几何融合为密集元素注释，同时记录每个执行操作的结果。利用该环境，我们构建了DeskForge-1M，这是一个包含1.2M注释桌面观察的语料库，包含159.7M元素实例。我们在从DeskForge-1M抽取的200K定位示例上微调了四个视觉-语言模型。四个模型在分离的桌面条件和所有五个外部GUI定位基准上均有所提升；对于Qwen3.5-4B，在ScreenSpot-Pro上准确率提高了11.51个百分点，在OSWorld-G上提高了10.11个百分点。这些提升同样反映在长时任务完成能力上：在固定规划器下，微调后的操作模型完成更多WebArena-Infinity和OpenApps任务，Qwen3.5-4B分别从119个任务中的31个提升到50个，从100个任务中的3个提升到15个。这些结果表明，真实桌面环境的可控组合为提升GUI定位和长时计算机使用能力提供了一种可扩展的监督来源。框架代码、数据集和微调模型可在项目页面获取：此https URL

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02320

Authors: A. Said Gurbuz, Ahmed Nassar, Sunghwan Hong, Marc Pollefeys, Peter W. J. Staar

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.02320.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02320

Published: 2026-10-05T02:24:56.485Z

3. SCION: Scene Composition with Instanced Neural Primitives

Abstract:Real-world scenes are compositional: bricks, blades of grass, pebbles, and tree leaves recur across human-built and natural environments. Existing neural scene representations model these elements independently. Most 3D Gaussian Splatting and follow-up abstraction and compression methods treat each element as unique, fitting millions of independent Gaussians per scene. Prior methods like Splat and Replace fit template objects, but they require mostly manual selection of repeated elements. As a result, these representations store redundant parameters and provide weak manipulation handles for downstream tasks. We introduce SCION, a hier- archical compositional scene representation that replaces independent Gaussians with a compact vocabulary of reusable primitives and lightweight world-space instances that place transformed copies throughout the scene. We fit this represen- tation to multi-view captures via a joint optimization over discrete and continuous scene parameters, combining two-level densification over splats and instances with an adversarial loss that preserves detail across shared primitives. The recovered structure yields a compact, controllable representation while maintaining high quality even at 1.2 MB. SCION achieves rate-distortion favorable to existing Gaussian compression methods, and it enables instance-level scene editing and animation without retraining. Our results show that neural scene representations need not memorize scenes as independent primitives; they can discover reusable parts. Project webpage: this https URL

中文摘要

摘要：现实世界的场景是组合性的：砖块、草叶、鹅卵石和树叶在人类建造和自然环境中反复出现。现有的神经场景表示独立建模这些元素。大多数 3D 高斯点云（Gaussian Splatting）及其后续的抽象和压缩方法将每个元素视为独一无二的，为每个场景拟合数百万个独立的高斯分布。先前的方法如 Splat 和 Replace 拟合模板对象，但它们主要依赖手动选择重复元素。因此，这些表示存储了冗余参数，并为下游任务提供了较弱的操作手段。我们提出了 SCION，一种分层组合场景表示，它用可重用原语的紧凑词汇和轻量级的世界空间实例替代独立高斯分布，这些实例可以将变换后的副本分布整个场景中。我们通过对离散和连续场景参数的联合优化将这种表示拟合到多视角采集中，结合对点和实例的两级密集化以及保持共享原语细节的对抗损失。恢复的结构提供了紧凑、可控的表示，同时即使在 1.2 MB 的情况下仍保持高质量。SCION 在速率-失真方面优于现有的高斯压缩方法，并且无需重新训练即可实现实例级场景编辑和动画。我们的结果表明，神经场景表示不必将场景记忆为独立原语；它们可以发现可重复使用的部分。项目网页：此网址 https

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02322

Authors: William Koch, Amogh Joshi, Cyrus Vachha, Cheng Zheng, Felix Heide

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.02322.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02322

Published: 2026-10-05T02:24:56.485Z

4. SCOPE-4D: Endoscopic 4D Geometry Foundation Models

Abstract:Geometric understanding supports endoscopic navigation and robotic assistance, but learning reliable endoscopic geometry faces two challenges: scarce geometric annotations and ambiguity between camera motion and tissue deformation. We present SCOPE-4D, an endoscopic 4D geometry foundation model that jointly predicts camera parameters, dense geometry, and 3D tissue trajectories from monocular RGB video in a single forward pass. Our curation and annotation pipeline constructs SCOPE-5K, a collection of approximately 5,000 clips spanning real and synthetic gastrointestinal endoscopy and laparoscopy. The collection provides rich geometric supervision and includes newly collected phantom and real-colonoscopy evaluation sets. Geometric supervised fine-tuning on SCOPE-5K learns endoscopic priors that improve camera and depth estimation. Common—Residual Motion (CRM) further constrains local deformation relative to common tissue movement. Together with geometric supervision, CRM and trajectory supervision further improve camera and depth estimation over geometric fine-tuning alone while enabling dense 3D tissue tracking. Evaluations on public and newly collected benchmarks demonstrate strong in-domain and out-of-domain geometry, superior 3D tracking, and more stable long-sequence colon reconstruction. A blinded user study further supports the perceived reconstruction quality on real clinical video. Together, these results demonstrate the value of large-scale endoscopic supervision and motion constraints for joint geometry estimation and tissue tracking.

中文摘要

摘要：几何理解支持内窥镜导航和机器人辅助，但学习可靠的内窥镜几何面临两个挑战：几何标注稀缺以及相机运动与组织变形之间的歧义。我们提出了 SCOPE-4D，一种内窥镜 4D 几何基础模型，可以在单次前向推理中联合预测相机参数、密集几何和 3D 组织轨迹，输入为单目 RGB 视频。我们的策划和标注流程构建了 SCOPE-5K，这是一个包含约 5,000 个片段的集合，涵盖真实和合成的胃肠内镜以及腹腔镜视频。该集合提供了丰富的几何监督，并包括新收集的模拟器和真实结肠镜评价集。在 SCOPE-5K 上进行几何监督微调可以学习内窥镜先验，从而提升相机和深度估计。Common—Residual Motion（CRM）进一步约束了相对于常规组织运动的局部变形。结合几何监督，CRM 与轨迹监督能够在仅几何微调的基础上进一步提升相机和深度估计，同时实现密集的 3D 组织跟踪。在公共和新收集的基准数据集上的评估表明，该方法在域内和域外几何表现强劲，3D 跟踪优越，并且长序列结肠重建更稳定。盲法用户研究进一步支持了在真实临床视频上的重建质量感知结果。综合来看，这些结果证明了大规模内窥镜监督和运动约束在联合几何估计与组织跟踪中的价值。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02343

Authors: Chaoyi Zhou, Zhongpai Gao, Anwesa Choudhuri, Meng Zheng, Benjamin Planche, Run Wang, Terrence Chen, Siyu Huang, Ziyan Wu

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.02343.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02343

Published: 2026-10-05T02:24:56.485Z

5. Confidence-Controlled XAI Auditing for Pedestrian Detection under Domain Shift

Abstract:Explainability is increasingly required for perception models in intelligent vehicles, yet whether explanations remain faithful under driving domain shift is still poorly understood. This work audits post-hoc explanations of a fixed YOLOv8s pedestrian detector across PIE and JAAD using ROI-based D-Deletion, frozen confidence terciles, rank-based tests, bootstrap intervals, and Holm correction. The audit shows that deletion-based faithfulness is strongly coupled to detection strength at explanation time, with Spearman correlations between 0.70 and 0.82 for D-RISE, making naive confidence-stratified comparisons unreliable. After controlling for detection strength within fixed f0 bins, D-RISE faithfulness remains domain-dependent in the central f0 range, with PIE showing higher D-Deletion than JAAD and Holm-adjusted significance. A non-perturbative EigenCAM baseline is less faithful than D-RISE but also exhibits score coupling, suggesting that the effect is not specific to D-RISE and is related to the deletion-based evaluation setup. These results motivate confidence-controlled XAI audits for safety-critical perception under domain shift.

中文摘要

摘要：智能车辆的感知模型越来越需要可解释性，但在驾驶领域转移下解释是否仍然可信仍知之甚少。本研究通过基于ROI的D-Deletion、冻结置信度三分位、基于排序的检验、引导区间和Holm校正，对固定YOLOv8s行人检测器在PIE和JAAD数据集上的事后解释进行了审计。审计结果表明，基于删除的可信度在解释时与检测强度密切相关，D-RISE的斯皮尔曼相关系数在0.70到0.82之间，使得简单的置信度分层比较不可靠。在固定f0区间内控制检测强度后，D-RISE的可信度在中心f0范围内仍然依赖于领域，其中PIE表现出比JAAD更高的D-Deletion，并且Holm校正后具有显著性。非扰动的EigenCAM基线可信度低于D-RISE，但也表现出分数耦合，表明该效应并非D-RISE特有，而是与基于删除的评估设置相关。这些结果促使在领域转移下对安全关键感知进行置信度控制的XAI审计。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02364

Authors: Ruben Dario Florez-Zela

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.02364.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02364

Published: 2026-10-05T02:24:56.485Z

6. EviDent-CBCT: Evidence-Bottlenecked Report Generation from Dental CBCT under Non-Exhaustive Report Supervision

Abstract:Dento-maxillofacial cone-beam CT (CBCT) reports may contain dozens of tooth-specific, anatomical, and spatial findings from a single 3D scan. Learning to generate such reports from limited clinical data is challenging because routine reports may not exhaustively document image findings, and a non-mention may reflect either absence or non-reporting. We present EviDent-CBCT, an evidence-bottlenecked framework designed for this incomplete supervision. An anatomy-aware network maps each CBCT scan to a discrete record of tooth-level, global, and tooth-IAC evidence. A dental-logic consistency projection reconciles incompatible evidence before a deterministic renderer and an image-blind local language model generate the report using only this record. For tooth-level evidence, reliability-aware training uses eligible non-mentions as reduced-weight negatives, while unreported global and tooth-IAC labels remain unknown. A metal-sensitive input channel preserves intensity cues from dental materials. Across three validation runs, EviDent-CBCT achieves $0.666\pm0.006$ merged evidence set-F1 and $0.402\pm0.003$ RadFact-Lite-Dental logical-F1, versus $0.371\pm0.018$ for the strongest controlled direct baseline. In the ODIN 2026 challenge, it ranked second in automated evaluation and third in blinded clinical Arena comparison on the hidden test set. These results support the discrete evidence record as an effective and auditable interface for CBCT report generation.

中文摘要

摘要：牙颌面锥形束CT（CBCT）报告可能包含数十项牙齿特异性、解剖学和空间发现，来自单次3D扫描。学习从有限的临床数据生成此类报告具有挑战性，因为常规报告可能无法全面记录图像发现，且未提及可能反映缺失或未报告。我们介绍EviDent-CBCT，一个为这种不完整监督设计的证据瓶颈框架。一个解剖感知网络将每次CBCT扫描映射到牙齿层面、全局和牙齿IAC证据的离散记录。牙科逻辑一致性投影在确定性渲染器和无影像本地语言模型仅用该记录生成报告前，对不兼容证据进行调和。对于牙齿层面证据，可靠性感知训练将合格的未提及标记作为减权负面，而未报告的全局和牙齿IAC标签则未知。金属敏感输入通道保存了牙科材料的强度线索。在三次验证运行中，Evident-CBCT在合并证据集F1中取得了0.666\pm0.006$的成绩，RadFact-Lite-Dental logical-F1的0.402\pm0.003$，而最强的对照直接基线则为0.371\pm0.018美元。在ODIN 2026挑战中，它在隐藏测试集的自动化评估中排名第二，在盲法临床Arena比较中排名第三。这些结果支持了离散证据记录作为CBCT报告生成的有效且可审计的界面。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02375

Authors: Ruiyang Hao, Zhi Qin Tan, Yulan He, Owen Addison, Yunpeng Li

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.02375.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02375

Published: 2026-10-05T02:24:56.485Z

7. FactorSplat: Appearance-Controllable Gaussian Proxies for Medical Volume Rendering

Abstract:Transfer functions (TFs) control color and visibility in medical volume rendering, but image-trained Gaussian proxies typically bake one transfer function into their appearance. We present FactorSplat, a per-scene N-dimensional Gaussian splatting (N-DGS) proxy that accepts region-specific intensity-to-RGBA curves at inference. A local lookup applies the authored color and opacity change, while a shared functional encoder and low-rank per-Gaussian factors learn the residual appearance response. Geometry and directional appearance remain shared across presets, with visibility control and TF-aware pruning preserving the ability to hide and reveal structures. On seven CT and MR scans, FactorSplat improves mean PSNR and changed-region error over region-aware VEG across validation, interpolation, unseen composition, and out-of-distribution (OOD) edits. Across these four splits, seven-scan mean PSNR gains over VEG range from 1.10 to 1.52 dB. One checkpoint per scan supports unseen edits without retraining. At $1600^2$, the cached fast renderer averages 524 FPS with 1.17 ms TF switches. Project page: this https URL.

中文摘要

摘要：传递函数（TFs）在医学体积渲染中控制颜色和可见性，但经过图像训练的高斯代理通常将单个传递函数固定在其外观中。我们提出了 FactorSplat，一种针对每个场景的 N 维高斯散射（N-DGS）代理，它在推理时接受区域特定的强度到 RGBA 曲线。本地查找应用作者设定的颜色和不透明度变化，同时共享的函数编码器和每个高斯的低秩因子学习残差外观响应。几何和方向外观在预设间保持共享，并且可见性控制和 TF 感知的修剪保留了隐藏和显示结构的能力。在七个 CT 和 MR 扫描上，FactorSplat 在验证、插值、未见组合和分布外（OOD）编辑中，相较于区域感知的 VEG，提升了平均 PSNR 和修改区域误差。在这四个分割中，七份扫描的平均 PSNR 相较 VEG 的增益范围为 1.10 到 1.52 dB。每个扫描的一个检查点支持未见过的编辑而无需重新训练。在 $1600^2$ 分辨率下，缓存的快速渲染器平均达到 524 FPS，TF 切换时间为 1.17 毫秒。项目页面：此 https URL。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02382

Authors: Zhongpai Gao, Benjamin Planche, Meng Zheng, Anwesa Choudhuri, Terrence Chen, Ziyan Wu

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.02382.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02382

Published: 2026-10-05T02:24:56.485Z

8. Octrees as an Explicit 3D Language

Abstract:Existing 3D large language models (LLMs) compromise on two fronts: they compress shapes into latent codebook indices or coordinate text, which removes spatial structure from what the model observes, and they acquire the 3D modality by fine-tuning the backbone, which overwrites its general language ability. We present OctLLM, which addresses both limitations. Geometry enters as an explicit 3D sequence of octree occupancy tokens. However, full octree sequences grow rapidly with depth; OctLLM therefore randomly empties penultimate-level nodes and omits descendants while preserving shape, yielding a shorter coordinate- and depth-anchored Sparse Octree (S-Octree) for position-aware mask-modeling generation and 3D understanding. On the other front, existing methods introduce a new modality with full fine-tuning or LoRA, but full fine-tuning is costly, LoRA limits 3D capacity, and both modify the language pathway. OctLLM instead adds 3D capacity in parameters separate from the pretrained ones: mesh tokens are routed through independent trainable branches in a subset of blocks while text and image tokens retain the frozen vision-language pathway, and the two streams interact through shared self-attention. It trains far fewer parameters than full fine-tuning, yet sets a new state of the art among unified multimodal LLMs, lowering image-to-3D FID by $17.4\%$ and raising render-grounded captioning by $28.7$ points over ShapeLLM-Omni, while matching the backbone on general language benchmarks.

中文摘要

摘要：现有的 3D 大型语言模型（LLMs）在两个方面存在妥协：它们将形状压缩为潜在码本索引或坐标文本，这会移除模型观察到的空间结构；并且它们通过微调骨干网络来获取 3D 模态，这会覆盖其通用语言能力。我们提出了 OctLLM，解决了这两个限制。几何信息以八叉树占据标记的显式 3D 序列形式输入。然而，完整的八叉树序列会随着深度迅速增长；因此，OctLLM 随机清空倒数第二层节点并省略子节点，同时保留形状，从而生成更短的、具有坐标和深度锚定的稀疏八叉树（S-Octree），用于位置感知的掩码建模生成和 3D 理解。在另一方面，现有方法通过完全微调或 LoRA 引入新模态，但完全微调成本高昂，LoRA 限制了 3D 能力，并且两者都会修改语言路径。OctLLM 则在与预训练参数分离的参数中增加 3D 能力：网格标记通过部分块中的独立可训练分支路由，而文本和图像标记保留冻结的视觉-语言路径，两条流通过共享自注意力进行交互。它训练的参数远少于完全微调，但在统一多模态 LLM 中创造了新的最先进水平，使图像到 3D 的 FID 降低了 17.4%，并将基于渲染的图像描述提升了 28.7 分，超过 ShapeLLM-Omni，同时在通用语言基准上与骨干网络匹配。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02388

Authors: Ran Dan, Si-Tong Wei, Pengfei Xiong, Wei Zhang, Yadong Mu, Peng-Shuai Wang

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.02388.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02388

Published: 2026-10-05T02:24:56.485Z

9. An AI-Based Multi-Stage Approach for Androgenetic Alopecia Assessment from Low-Magnification Scalp Images

Abstract:Androgenetic alopecia (AGA) is characterized by patterned follicular miniaturization, increased single-hair follicular units, and altered hair-shaft diameter. We present an automated quantitative scalp-analysis and clinical decision-support framework combining FU localization, ordinal visible-shaft counting, calibrated shaft-width estimation, regional aggregation, and an interpretable rule layer. The clinical cohort comprised 243 patients (127 AGA, 116 non-AGA), while the computer-vision experiments used 160 expert-annotated patients, 2,400 trichoscopic images, and approximately 158,000 FU annotations. Under patientdisjoint evaluation, YOLOv8m achieved test mAP@0.5=0.920 and recall=0.860; EfficientNet-B5 with a support-map channel achieved 87.0% expert-box count accuracy (macro F1=0.85). A separate 500-image set was processed end-to-end with detector-generated boxes, yielding MAE of 6.56 for follicle detection and 16.59 for follicle classification relative to human-expert annotations. The system is intended to assist, rather than replace, dermatologist interpretation.

中文摘要

摘要：雄激素性脱发（AGA）的特征为毛囊呈模式化的微小化、单发毛囊单位增加以及发干直径改变。我们提出了一种自动化的定量头皮分析与临床决策支持框架，结合了毛囊单位定位、按序可见毛干计数、校准毛干宽度估算、区域汇总以及可解释规则层。临床队列包括243名患者（127名AGA患者，116名非AGA患者），而计算机视觉实验使用了160名专家标注的患者、2400张毛囊镜图像以及约158,000个毛囊单位标注。在患者不重叠评估下，YOLOv8m在测试中实现了mAP@0.5=0.920，召回率=0.860；带有支持地图通道的EfficientNet-B5实现了87.0%的专家框计数准确率（宏F1=0.85）。一个独立的500图像集采用检测器生成的框进行端到端处理，其毛囊检测的平均绝对误差（MAE）为6.56，毛囊分类的MAE为16.59，相较于人工专家标注。该系统旨在辅助，而非替代，皮肤科医生的解读。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02421

Authors: Mahmoud Raslan, Nada Omar, Omar Khaled, Tarek Waleed, Mohamed Hazem, Rania Mounir, Solwan Elsamanoudy, Ahmed Mourad, Noura Adel, Muhammad Rushdi

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.02421.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02421

Published: 2026-10-05T02:24:56.485Z

10. A Simulation-Grounded Agentic VLM Framework for Wildfire Monitoring and Reporting

Abstract:Effective wildfire monitoring requires relating visual evidence to physical fire dynamics, yet real videos with synchronized physical annotations are scarce and high-fidelity 3D simulation is costly. We present a simulation-grounded vision-language model (VLM) framework that automatically converts 2D wildfire simulations into labeled video episodes. A fixed Blender mapping produces low-detail 3D proxies aligned with simulator terrain, fuel layout, fire activity, and wind cues; controllable video generation supplies richer appearance. The proxies are intermediate representations rather than finely rendered final scenes. Generated videos and simulator labels form reusable multimodal memory for a training-free multi-agent VLM system that retrieves reference episodes, reconciles visual and memory-based predictions, and produces structured wildfire reports. On held-out generated episodes, video memory achieves 51.5% exact four-tag accuracy, compared with 22.6% for direct VLM querying and 16-17% for text-only memory; the complete system achieves 77.3% accuracy on six simulator-derived report fields. Component ablations, cross-generator tests, and three real-UAV evaluations assess retrieval, reporting, generator changes, and observable monitoring tasks. The framework connects automatic simulation-to-proxy conversion with memory-based VLM reasoning under scarce real-world physical annotations.

中文摘要

摘要：有效的野火监测需要将视觉证据与火灾物理动力学联系起来，但带有同步物理标注的真实视频稀缺，而高保真度的3D模拟成本高昂。我们提出了一种基于模拟的视觉-语言模型（VLM）框架，该框架能够自动将二维野火模拟转换为带标签的视频片段。固定的Blender映射生成与模拟器地形、燃料布局、火势活动和风力线索对齐的低细节3D代理；可控视频生成提供更丰富的外观。这些代理是中间表示，而不是精细渲染的最终场景。生成的视频和模拟器标签形成可重用的多模态记忆，用于无需训练的多智能体VLM系统，该系统能够检索参考片段，协调视觉与记忆基础的预测，并生成结构化的野火报告。在未见的生成片段上，视频记忆实现了51.5%的四标签精确准确率，而直接VLM查询为22.6%，纯文本记忆为16-17%；完整系统在六个模拟器衍生报告字段上实现了77.3%的准确率。组件消融实验、跨生成器测试和三次真实无人机评估用于评估检索、报告生成、生成器变化及可观测监测任务。该框架将自动的模拟到代理转换与基于记忆的VLM推理在现实物理标注稀缺的情况下联系起来。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.02451

Authors: Duowen Chen, Yuchen Sun, Zhiqi Li, Yuxuan Liao, Sinan Wang, Bart van Bloemen Waanders, Bo Zhu

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.02451.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.02451

Published: 2026-10-05T02:24:56.485Z