ArXiv Domain 2026-10-09

数据来源：ArXiv Domain

LLM Domain Papers

1. Diffu-LoRA: A Novel Low-Rank Adaptation for Personalized Diffusion Models

Abstract:Personalizing text-to-image diffusion models from a few reference images requires preserving subject identity while following prompts that describe new contexts. Full-model fine-tuning is parameter-intensive, whereas low-rank adaptation (LoRA) reduces the number of trainable parameters but leaves open how adaptation capacity should be distributed across layers. We introduce Diffu-LoRA, a parameter-efficient method that learns this allocation through gated low-rank adaptation. Diffu-LoRA inserts trainable low-rank components into the linear layers of Transformer blocks and assigns a learnable gate to each component. Bilevel optimization updates the adaptation weights and gate parameters on separate data splits, while progressive pruning removes components with the lowest gate values to meet a prescribed rank budget. This procedure allocates adaptation capacity nonuniformly across layers while keeping the pretrained backbone frozen. Experiments with Stable Diffusion on subjects from DreamBooth and additional collected datasets show improved overall subject fidelity and prompt alignment relative to the evaluated fine-tuning baselines. Ablation studies examine the contributions of bilevel optimization, progressive pruning, and adapter placement. These results support learned rank allocation as a practical approach to parameter-efficient diffusion model personalization.

中文摘要

摘要：从少量参考图像个性化文本到图像的扩散模型需要在保持主体身份的同时遵循描述新场景的提示。全模型微调需要大量参数，而低秩适配（LoRA）减少了可训练参数的数量，但如何在各层分配适配容量仍未确定。我们提出了 Diffu-LoRA，一种参数高效的方法，通过门控低秩适配学习这种分配。Diffu-LoRA 在 Transformer 块的线性层中插入可训练的低秩组件，并为每个组件分配一个可学习的门。双层优化在不同的数据分割上更新适配权重和门参数，同时逐步剪枝移除门值最低的组件以满足规定的秩预算。该过程在各层中非均匀分配适配容量，同时保持预训练骨干网络冻结。使用 Stable Diffusion 对 DreamBooth 的主体和额外收集的数据集进行实验表明，与评估的微调基线相比，能够提高整体主体忠实度和提示一致性。消融研究考察了双层优化、逐步剪枝和适配器放置的贡献。这些结果支持将学习的秩分配作为参数高效扩散模型个性化的实用方法。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10550 (HTTP 429)

Authors: Tianjing Li, Wei Zhu

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.10550.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10550

Published: 2026-10-09T03:04:34.191Z

2. Wieszcz-XIX: A 3.1-Billion-Word Corpus of Pre-1918 Polish and Temporally Bounded Language Models Trained From Scratch

Abstract:Historical Polish is well documented as a language but annotated in machine-readable form only to about a million words for the period this paper covers; the rest sits behind optical character recognition of variable quality. We present Wieszcz-XIX, a corpus of 6.75 billion tokens (about 3.1 billion words) in 294,369 documents, most of them periodical issues, of Polish published from 1800 to 1918, assembled from Wolne Lektury and the Internet Archive by a pipeline that filters, deduplicates, audits for post-1918 leakage and splits at the document level. It is over three orders of magnitude larger than the annotated corpus of the same period, and we quantify its defects: recognition corruption against a false-positive floor, near-identical duplication, which is removed, and post-1918 leakage, which is excluded from the training corpus itself down to a known residue of 0.04 to 0.38% of its bytes, found in the transcribed source, so the published corpus is the trained one document for document. On a hand-corrected sample the character error rate is 0.68% where the text is legible, and 45% of the sampled passages cannot be corrected. On it we train a ladder of decoder-only models from 47M to 349M parameters from scratch, and measure their temporal boundedness. Against two modern Polish base models, one far larger, the 349M shows a crossover, as does the 107M against the comparator of its size: post-1918 vocabulary costs them about 3.1 bits per byte more than period vocabulary, a gap the comparators do not show, and period vocabulary costs them fewer bits than it costs the comparators. Shown period text, the models keep its spelling and the comparators only partly. Adding parameters gains about twice as much as a second pass over the data. We release the corpus, code and weights. Content warning: the models reproduce period prejudice, including antisemitic statements.

中文摘要

摘要：历史波兰语作为一种语言有完整的文献记录，但在机器可读形式下注释的仅约有论文涵盖期间的百万字，其余的则只能通过质量不一的光学字符识别获取。我们提出了 Wieszcz-XIX 语料库，其包含 67.5 亿个标记（约 31 亿个单词），来源于 294,369 篇文档，其中大多数为期刊出版物，时间跨度为 1800 年至 1918 年，这些文档由 Wolne Lektury 和互联网档案馆汇集而来，经过一条处理流程，包括过滤、去重、审核 1918 年后的泄漏，并在文档层面拆分。该语料库的规模比同期注释语料大三到四个数量级，我们对其缺陷进行了量化：相对于误报下限的识别错误、几乎相同的重复内容（已移除），以及 1918 年后的泄漏（在训练语料中被排除，仅在转录源中残留 0.04% 到 0.38% 的字节），因此公开发布的语料与训练语料在文档层面是一致的。在人工校正样本中，可读文本的字符错误率为 0.68%，但 45% 的抽样段落无法纠正。在此语料上，我们从零开始训练了参数量从 4700 万到 3.49 亿的仅解码器模型，并测量其时间边界性。与两个现代波兰语基础模型（其中一个远大于我们的模型）相比，3.49 亿的模型显示出交叉点，107M 参数的模型也与同规模比较对象类似：1918 年后的词汇使它们每字节成本约增加 3.1 比特，而比较模型未显示该差距，同时同期词汇使它们的成本低于比较模型。面对同期文本时，这些模型能保持原有拼写，而比较模型只能部分保持。增加参数带来的收益约为对数据进行第二次训练的两倍。我们发布了语料库、代码和模型权重。内容警告：模型会再现当时期的偏见，包括反犹言论。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10592 (HTTP 429)

Authors: Szymon Kocur

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.10592.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10592

Published: 2026-10-09T03:04:34.191Z

3. Large Language Model-Assisted Preparation of Transportation Management Plans: A Case Study with WisDOT WisTMP System

Abstract:Work zones are critical yet hazardous components of transportation infrastructure, requiring carefully designed Transportation Management Plans (TMPs) to ensure safety and mobility. However, TMP preparation remains labor-intensive and heavily dependent on practitioner expertise. This paper proposes a Large Language Model (LLM)-assisted framework to automate TMP content generation, leveraging the WisDOT WisTMP system as the application context. The framework fine-tunes multiple open-source LLMs across different model scales and deploys them locally to ensure data security. To support model training, we construct a domain-specific dataset from historical WisTMP documents by converting PDF files into structured question-answer pairs in JSON format. Experimental results show that fine-tuning significantly improves performance across standard text generation metrics. Further section-wise and strategy-level analyses reveal that, while LLMs achieve strong overall performance, they tend to over-generate strategies and struggle to produce project-specific justifications and accurate cost estimates. In addition, scaling from 7B/8B to 14B yields limited gains. These findings demonstrate the potential of LLMs to improve TMP preparation efficiency while highlighting remaining challenges in LLM-assisted TMP development. The source code and demo videos will be publicly available at this https URL.

中文摘要

摘要：施工区域是交通基础设施中关键但危险的组成部分，需要精心设计的交通管理计划（TMP）以确保安全和通行。然而，TMP 的准备工作仍然劳动密集，并且高度依赖从业人员的专业知识。本文提出了一个基于大型语言模型（LLM）的辅助框架，以自动生成 TMP 内容，并以 WisDOT WisTMP 系统作为应用背景。该框架在不同模型规模上微调多个开源 LLM，并在本地部署以确保数据安全。为了支持模型训练，我们从历史 WisTMP 文档构建了一个领域特定的数据集，通过将 PDF 文件转换为 JSON 格式的结构化问答对。实验结果表明，微调显著提高了标准文本生成指标的性能。进一步的章节级和策略级分析显示，虽然 LLMs 在整体上表现出色，但它们倾向于生成过多策略，并且在提供项目特定的理由和准确的成本估算方面存在困难。此外，从 7B/8B 模型扩展到 14B 模型带来的提升有限。这些发现展示了 LLM 在提高 TMP 准备效率方面的潜力，同时也突出了 LLM 辅助 TMP 开发中仍存在的挑战。源代码和演示视频将在此 https URL 公开提供。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10650 (HTTP 429)

Authors: Zihao Sheng, Pei Li, Zilin Huang, Yen-Jung Chen, Yuhao Luo, Zhengyang Wan, Steven T. Parker, David A. Noyce, Sikai Chen

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.10650.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10650

Published: 2026-10-09T03:04:34.191Z

4. Cognitive Thermometers: Machine Learning and Logical Complexity

Abstract:How does the human mind represent semantic categories? Why do natural languages favor certain meanings over others? Prior explanations have relied on logical definability and complexity, but these are highly sensitive to the choice of logical language, rendering some design choices unmotivated. In this article, we propose that machine learning provides a somewhat more agnostic approach to measuring semantic complexity. We review emerging evidence that logic and machine learning often yield converging results on relative complexity and its resulting effects in semantic typology. Where they diverge, learning appears to be a better explanation than logical complexity. We argue that treating machine learning models as ``cognitive thermometers’’ enables a unified approach to complexity that bridges symbolic logic and connectionist AI.

中文摘要

摘要：人类的思维如何表征语义类别？为什么自然语言偏好某些含义而非其他含义？以往的解释依赖于逻辑可定义性和复杂性，但这些高度依赖于逻辑语言的选择，从而使某些设计选择缺乏依据。在本文中，我们提出，机器学习提供了一种在一定程度上更为中立的测量语义复杂性的方法。我们回顾了新兴的证据，表明逻辑与机器学习在相对复杂性及其对语义类型学的影响上常常产生趋同的结果。在它们出现分歧的地方，学习似乎比逻辑复杂性提供了更好的解释。我们认为，将机器学习模型视为“认知温度计”能够提供一种统一的复杂性方法，桥接符号逻辑与联结主义人工智能。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10724 (HTTP 429)

Authors: Shane Steinert-Threlkeld, Jakub Szymanik

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.10724.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10724

Published: 2026-10-09T03:04:34.191Z

5. Lossy Compressive Text Autoencoders

Abstract:Our work explores learning a compressed latent representation of text, at the intersection of data compression and representation learning. We propose an autoencoder architecture that performs residual downscaling and upscaling of hidden representations along the time axis, with a residual low-dimension discrete bottleneck. We analyze our approach for different quantization methods, training objectives, and datasets. For different levels of compression, we evaluate the similarity between the original and reconstructed text both at the surface-level (BLEU) and at the semantic-level (LLM-based judge). Additionally, we evaluate our models on downstream question-answering and semantic text similarity benchmarks. Our approach results in compressed representations which are on par with lossless text compression algorithms at 2.24 bits per byte on web text data, while having good reconstruction and downstream task performance.

中文摘要

摘要：我们的工作探索了文本的压缩潜在表示的学习，位于数据压缩和表示学习的交叉点。我们提出了一种自编码器架构，该架构沿时间轴对隐藏表示进行残差降采样和升采样，并带有残差低维离散瓶颈。我们分析了不同量化方法、训练目标和数据集下的方法表现。对于不同压缩级别，我们评估了原始文本与重建文本在表层（BLEU）和语义层面（基于大语言模型的评判）的相似性。此外，我们还在下游问答和语义文本相似性基准上评估了我们的模型。我们的方法在网络文本数据上实现的压缩表示与无损文本压缩算法相当，达到每字节2.24比特，同时具有良好的重建性能和下游任务表现。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10738 (HTTP 429)

Authors: Vinko Sabolčec, Angelos Katharopoulos, David Grangier

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.10738.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10738

Published: 2026-10-09T03:04:34.191Z

6. Clarify, Then Focus: Statement Normalization for Conversation Analytics at Scale

Abstract:Enterprise conversation analytics asks many questions of millions of interactions. Each question can require reconstructing what people mean and identifying which information matters, repeating costly interpretive work across the same transcripts. We propose a simple principle: clarify the text, then focus the reader. Statement normalization transforms dialogue into short, speaker-attributed statements with source references and semantic tags. The statements make meaning more explicit; the tags support selecting evidence for a particular question. Downstream models can use the full representation or a relevant subset, depending on what helps them make the decision. In an offer-suppression task on customer-service calls, normalization improves a supervised classifier without selection, while weaker prompted readers benefit from both normalization and selection. A small model can learn the normalization contract, while lightweight encoders handle tagging and downstream decisions. Sharing this preparation across questions supports an inference pipeline built entirely from small models, making analytics over millions of conversations substantially less expensive.

中文摘要

摘要：企业对话分析对数百万次互动提出了许多问题。每个问题可能都需要重建人们的意图并识别哪些信息重要，从而在相同的文本记录上重复进行高成本的解释工作。我们提出一个简单的原则：澄清文本，然后聚焦读者。陈述规范化将对话转换为简短的、带有说话人归属、来源引用和语义标签的陈述。这些陈述使含义更明确；标签有助于为特定问题选择证据。下游模型可以使用完整表示或相关子集，取决于什么对它们做出决策有帮助。在客户服务电话的报价抑制任务中，规范化在没有选择的情况下改善了监督分类器的性能，而较弱的提示式阅读器则从规范化和选择中都获益。小型模型可以学习规范化规范，而轻量编码器处理标签和下游决策。在问题之间共享这种准备工作支持了一个完全由小型模型构建的推理管道，从而使对数百万次对话的分析成本大幅降低。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10758 (HTTP 429)

Authors: Mikhail L. Arbuzov, Karan Dave, Evgeniya Dontsova, Yaodong Hu, Vincent Lao, Navita Jain, Sisong Bei, Dmitry Dimov

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.10758.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10758

Published: 2026-10-09T03:04:34.191Z

7. Grammar Concept Annotation at Scale: Deployed Fine-Tuned Small Language Models Outperform Prompted Frontier Models

Abstract:Corrective feedback is among the best-evidenced drivers of second-language acquisition, yet corrections delivered during lessons rarely accumulate into an actionable view of grammar mastery. Prompted frontier models can provide such a view from learner—tutor lesson transcripts, but they are costly at scale. We close this gap by fine-tuning Qwen3.5 small language models (SLMs) on filtered and rebalanced teacher-generated supervision, then deploying an efficient 0.8B model in an end-to-end grammar mastery tracker for all English learners on our platform. Internalizing the annotation contract into adapter weights enables pairing the 0.8B model with a compact matched prompt rather than verbose instructions. On two human-curated benchmarks, both the deployed 0.8B model and a 4B reference comparator outperform prompted GPT-5.4 and GPT-5.6 Sol in precision and recall under nested matching criteria of increasing strictness: concept, evidence span, and correctness. The deployed 0.8B SLM reduces serving cost by approximately 16$\times$. A feature-level online experiment shows significant gains in learner engagement ($+15.8\%$) and key business metrics, including scheduled hours ($+2.1\%$) and GMV from new lessons ($+13.2\%$).

中文摘要

摘要：纠正性反馈是第二语言习得中证据最充分的驱动因素之一，然而在课程中提供的纠正通常很少会积累成可操作的语法掌握视图。受提示的前沿模型可以从学习者-导师课程记录中提供这种视图，但在大规模应用时成本很高。我们通过在过滤和重新平衡的教师生成监督数据上微调Qwen3.5小型语言模型（SLMs），然后在我们平台上为所有英语学习者部署高效的0.8B模型至端到端语法掌握追踪器，从而弥补了这一差距。将标注协议内化到适配器权重中，使0.8B模型能够配对紧凑的匹配提示，而不是冗长的指令。在两个人工策划的基准测试中，无论是部署的0.8B模型还是4B参考对照模型，在概念、证据跨度和正确性等递进严格的嵌套匹配标准下，其精确率和召回率均优于提示的GPT-5.4和GPT-5.6 Sol。部署的0.8B SLM将服务成本降低了约16倍。一个特征级别的在线实验显示学习者参与度显著提升（+15.8%），以及关键业务指标的改善，包括预约学时（+2.1%）和新课程产生的GMV（+13.2%）。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10827 (HTTP 429)

Authors: Marjan Celikik, Ana Peleteiro Ramallo, Javier Morales

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.10827.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10827

Published: 2026-10-09T03:04:34.191Z

8. Real Long-Term Memory for AI: A 50-Million-Token Window That Is Faster and Cheaper Than Recompute

Abstract:A large language model can only use the text that fits in its context window, and it recomputes its internal key-value (KV) state for a prompt every time the prompt is sent. We test a memory layer, the public package galahad-kv, that saves the KV state of each block of about 16,000 tokens to encrypted local NVMe disk and loads it back later, byte-exact, without recomputing it. We ran it on 50,000,000 tokens of real public text, served through vLLM on one NVIDIA H100, with Gemma 4 12B and Gemma 4 31B. Every block we probed was loaded back from the encrypted store with no recompute (100 of 100, at depths from 0 to 50M tokens) on both models. Loading a block was 2.8x to 4.3x faster than recomputing it and used 8.8x to 12.3x less GPU energy, and GPU memory stayed flat over the whole 50M-token stream. Asked about facts planted millions of tokens earlier, the 12B model gave the right answer 82 times out of 100 and the 31B model 98 times out of 100. Neither model made up an answer. The limits are as follows. This is reuse of stored state, not a wider attention window: one block is loaded at a time, and how well a question is answered depends on the model. Writing the memory is a one-time cost, and the store takes terabytes of local NVMe disk. We describe the test protocol, which is built to resist common ways of gaming long-context benchmarks, and give a single-GPU reproduction that uses public software and a free licence for the package.

中文摘要

摘要：大型语言模型只能使用适合其上下文窗口的文本，并且每次发送提示时，它都会重新计算自身的内部键值（KV）状态。我们测试了一个内存层——公共软件包 galahad-kv，它将每个大约 16,000 令牌的块的 KV 状态保存到加密的本地 NVMe 磁盘，并在稍后按字节精确地加载回，而无需重新计算。我们在 50,000,000 令牌的真实公共文本上运行它，通过 vLLM，在一块 NVIDIA H100 上使用 Gemma 4 12B 和 Gemma 4 31B。我们测试的每个块都从加密存储中加载回，没有重新计算（100 次中的 100 次，深度从 0 到 50M 令牌）在两个模型上都是如此。加载一个块比重新计算快 2.8 倍到 4.3 倍，并且使用的 GPU 能量少 8.8 倍到 12.3 倍，同时整个 50M 令牌流中 GPU 内存保持稳定。对于几百万令牌之前植入的事实问题，12B 模型在 100 次中给出了 82 次正确答案，31B 模型在 100 次中给出了 98 次正确答案。两个模型都没有编造答案。其限制如下：这是存储状态的重用，而不是更广的注意力窗口：一次只加载一个块，问题回答的好坏取决于模型。写入内存是一次性成本，存储需要数 TB 的本地 NVMe 磁盘。我们描述了测试协议，该协议旨在抵御常见的长上下文基准作弊方式，并提供了一个使用公共软件和免费许可包的单 GPU 复现示例。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10845 (HTTP 429)

Authors: Sietse Schelpe

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.10845.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10845

Published: 2026-10-09T03:04:34.191Z

9. Disentangling Linguistic and Paralinguistic Information with Routed Sparse Autoencoders

Abstract:Self-supervised speech encoders contain linguistic and paralinguistic information in a shared, entangled representation space. We combine a TopK sparse autoencoder with route-specific supervision and cross-factor adversaries. Across frozen SPEAR and WavLM encoders, independent probes show factor-specific retention and suppression: linguistic information remains stronger in the linguistic route, while paralinguistic factors, including speaker identity, emotion, and prosody, are retained in the paralinguistic route and substantially reduced in the linguistic route. The route organisation learned on LibriSpeech persists on MSP-Podcast without representation-side retraining. Feature-space route interventions further transfer the swapped factor while largely preserving the information carried by the unchanged route. These results show consistent route-selective separation across encoders, corpora, independent probes, and representation-level interventions.

中文摘要

摘要：自监督语音编码器在共享的、纠缠的表示空间中包含语言信息和副语言信息。我们将TopK稀疏自编码器与特定路径的监督和跨因素对抗结合。在冻结的SPEAR和WavLM编码器中，独立探测器显示了因素特定的保留与抑制：语言信息在语言路径中保持较强，而副语言因素，包括说话人身份、情绪和韵律，在副语言路径中被保留，而在语言路径中大幅减少。在LibriSpeech上学习的路径组织在MSP-Podcast上仍然有效，无需对表示侧进行重新训练。特征空间的路径干预进一步传递被交换的因素，同时在很大程度上保持未更改路径所携带的信息。这些结果表明，在编码器、语料库、独立探测器和表示层干预中都存在一致的路径选择性分离。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10865 (HTTP 429)

Authors: Beimnet Bekele Guta, Xiaoyu Yang, Guangzhi Sun, Philip C. Woodland

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.10865.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10865

Published: 2026-10-09T03:04:34.191Z

10. Sparse Attention Is Matrix Approximation, Not Choosing from a Bag of Values

Abstract:Large Language Models (LLMs) achieve strong performance across many domains, but their efficiency is limited by the quadratic cost of attention with respect to prompt length. Sparse attention reduces this cost by retaining only a small fraction of query-key interactions to approximate the full attention matrix. However, existing methods are trapped in a mathematically wrong view: they simply keep large scalar entries or high-mass regions of the attention matrix. This treats the attention matrix as a bag of values, ignoring that it is used as a structured matrix whose entries jointly determine the attention output through multiplication with value vectors. We argue that this is the core conceptual issue: sparse attention should be formulated as matrix approximation, not as blindly choosing the largest values from a bag of entries. Based on this view, we propose Matrix Approximation Sparse Attention (MASA). MASA replaces raw attention-mass ranking with a closed-form score that measures how much each sparse unit reduces matrix-product approximation error. As a theory-grounded plug-in correction, MASA can be added to existing sparse attention frameworks without changing their sparse kernels or budgets. Extensive experiments across multiple sparse attention methods, benchmarks, and LLM backbones show consistent accuracy gains, supporting both MASA and the matrix-approximation view of sparse attention.

中文摘要

摘要：大型语言模型（LLMs）在许多领域表现出强大的性能，但它们的效率受到注意力机制相对于提示长度的二次成本限制。稀疏注意力通过仅保留一小部分查询-键交互来近似完整的注意力矩阵，从而降低了这一成本。然而，现有方法陷入了一个数学上的错误观点：它们只是保留注意力矩阵中的大标量条目或高质量区域。这将注意力矩阵视为一个值的集合，而忽略了它被用作一个结构化矩阵，其条目通过与值向量的乘法共同决定注意力输出。我们认为这是核心概念性问题：稀疏注意力应被表述为矩阵近似，而不是盲目地从值集合中选择最大值。基于这一观点，我们提出了矩阵近似稀疏注意力（MASA）。MASA用一个闭式评分替代了原始的注意力质量排名，该评分衡量每个稀疏单元减少矩阵乘积近似误差的程度。作为一个理论支持的插件式修正，MASA可以被加入到现有的稀疏注意力框架中，而无需更改它们的稀疏内核或预算。在多种稀疏注意力方法、基准测试和大型语言模型骨干网络上的大量实验证明了持续的准确性提升，支持了MASA以及稀疏注意力的矩阵近似观点。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10871 (HTTP 429)

Authors: Fang Wan, Xufeng Liu, Fan Li, Yi Liu

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.10871.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10871

Published: 2026-10-09T03:04:34.191Z

Agent Domain Papers

1. An Explainable Header-Centric Framework for Large-Scale Semantic Table Interpretation and Data Quality Assessment

Abstract:Knowledge Graph (KG) quality depends not only on downstream graph validation, but also on the quality of tabular metadata used before integration. In metadata-only Semantic Table Interpretation (STI), where cell values are unavailable, noisy, or unsuitable, column headers become a critical source of semantic evidence for traceable KG preparation. We present an explainable, header-centric framework for metadata-only Column Type Annotation (CTA) and Data Quality Assessment (DQA). The framework maps headers to 39 interpretable FinalFormat types using curated lexical resources and preserves token-level traceability through SourceKeywords. Each assigned type activates validation rules based on a taxonomy of Data Quality Issues (DQIs), producing detections such as missing data, duplicates, domain violations, wrong data type, and temporal mismatch. These detections are aggregated into HeadersIQ, a lightweight, unweighted data source-level quality metric. The framework was evaluated across heterogeneous benchmarks, including UCI, Prague, Kaggle, VizNet/Sato, SOTAB, T2Dv2, and the SemTab 2024 Metadata-to-KG track, comprising around 120,000 header columns. The results show broad practical coverage across noisy real-world metadata, while a parallel KG-mapping pathway supports alignment to DBpedia and this http URL. On the SemTab 2024 Metadata-to-KG track, the official GT-strict evaluation was modest. However, a blinded diagnostic audit indicates that many mismatches reflect benchmark granularity, aliasing, and ontology-selection effects rather than wholly implausible header-centric predictions. We report this audit as diagnostic evidence on disagreement patterns, not as revised benchmark performance. Overall, the paper presents a reusable workflow for metadata-driven semantic annotation, data source-level quality monitoring, and KG-oriented benchmark diagnosis.

中文摘要

摘要：知识图谱（KG）的质量不仅取决于下游的图验证，还取决于整合前使用的表格元数据的质量。在仅元数据的语义表格解释（STI）中，当单元格值不可用、噪声较大或不适用时，列标题成为可追踪KG准备的关键语义证据来源。我们提出了一个可解释、以标题为中心的仅元数据列类型注释（CTA）和数据质量评估（DQA）框架。该框架使用精选词汇资源将标题映射到39个可解释的最终格式类型，并通过源关键词（SourceKeywords）保持令牌级可追踪性。每个分配的类型会根据数据质量问题（DQIs）分类激活验证规则，从而产生缺失数据、重复数据、领域违规、数据类型错误和时间不匹配等检测结果。这些检测结果被汇总为HeadersIQ，一种轻量级、未加权的数据源级质量指标。该框架在异构基准上进行了评估，包括UCI、Prague、Kaggle、VizNet/Sato、SOTAB、T2Dv2以及SemTab 2024 Metadata-to-KG赛道，共涵盖约12万列标题。结果显示在噪声现实世界元数据中具有广泛的实际覆盖能力，同时并行的KG映射通路支持与DBpedia和该http URL的对齐。在SemTab 2024 Metadata-to-KG赛道上，官方GT-strict评估结果适中。然而，盲法诊断审核表明，许多不匹配反映了基准粒度、别名和本体选择效应，而非完全不合理的以标题为中心的预测。我们将该审核报告作为对争议模式的诊断证据，而非修正后的基准性能。总体而言，本文提出了一个可复用的工作流程，用于元数据驱动的语义注释、数据源级质量监控以及KG导向的基准诊断。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10541 (HTTP 429)

Authors: Marcelo Valentim Silva, Hannes Herrmann, Valerie Maxville

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10541.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10541

Published: 2026-10-09T03:04:56.129Z

2. Synthesis Through Simulation: Generating Coherent Enterprise Data via Scalable Agent-System Interaction

Abstract:Tool-calling agents have become central to enterprise AI, yet training and evaluating them at scale remains severely constrained due to business and legal restrictions on enterprise systems, data, and database schemas. Tabular data synthesis offers a natural alternative, but its effectiveness is fundamentally limited by structural validity and schema availability, while procedure-based approaches yield the opposite weakness, typically lacking distributional fidelity without per-domain authoring. We introduce Synthesis Through Simulation (STS), a schema—free data synthesis paradigm in which an LLM agent generates data by executing operations against policy-enforcing APIs within simulated enterprise environments. Because data is generated through the same environment that defines what is valid, STS guarantees structural validity by construction while decoupling validity enforcement from distribution modeling, allowing each to be addressed independently. The Generalist Populator (GP), STS’s domain-agnostic agent, addresses the remaining challenges of distributional fidelity and synthesis scalability: GP achieves 0.88 average marginal fidelity and 100\% constraint satisfaction across all ten environments without access to DB schemas, while statistical synthesizers are inapplicable to seven due to necessary seed data requirements, and schema-privileged agents fail 82\% of trajectories on airline environment’s tightly coupled workflows due to brittle task composition. We open-source the full framework, all ten environments, and generated datasets at this https URL.

中文摘要

摘要：调用工具的代理已经成为企业 AI 的核心，但由于企业系统、数据和数据库模式的业务和法律限制，对它们进行大规模训练和评估仍然受到很大限制。表格数据合成提供了一种自然的替代方案，但其有效性从根本上受到结构有效性和模式可用性的限制，而基于流程的方法则表现出相反的弱点，通常在没有每个领域创作的情况下缺乏分布一致性。我们提出了通过模拟合成（Synthesis Through Simulation，STS），这是一种无模式的数据合成范式，其中 LLM 代理通过在模拟企业环境中对执行政策的 API 执行操作来生成数据。由于数据是通过同一环境生成的，这个环境定义了什么是有效的，因此 STS 从构建上保证了结构有效性，同时将有效性执行与分布建模解耦，使两者可以独立处理。通用填充器（Generalist Populator，GP），STS 的领域无关代理，解决了分布一致性和合成可扩展性方面的剩余挑战：GP 在十个环境中实现了0.88的平均边际一致性和100%约束满足率，且无需访问数据库模式，而统计合成器由于需要种子数据在七个环境中无法应用，具有模式特权的代理在航空环境的紧耦合工作流中，由于任务组合脆弱失败了 82% 的轨迹。我们在此 URL 开源了完整框架、所有十个环境以及生成的数据集。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10549 (HTTP 429)

Authors: Yipeng Li, Ashutosh Hathidara, Jane Lo, Harshavardhan Abichandani, Gunraj Singh, Atin Ghosh

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10549.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10549

Published: 2026-10-09T03:04:56.129Z

3. Agent-Controlled Forgetting for Tool-Using Agents: Reversible Context Curation in Practice

Abstract:Tool-using agents repeatedly carry observations whose useful content can be much smaller than their original payload. We study agent-controlled forgetting: the acting model selects previously observed tool results, replaces each with a short note at its original position, and retains the exact original in a recoverable archive. A Python harness exposes batch archival and explicit recovery without task-specific model training, while protecting user instructions and assistant messages from these operations. In an exploratory OpenTelemetry debugging case followed by an unrelated implementation task, the method ended with 231,951 provider-reported prompt tokens versus 912,492 under retained history, used 50% fewer cumulative input tokens, and had an estimated API cost of USD 1.28-1.44 versus approximately USD 4.38. Both arms passed the two-case primary behavioral oracle; neither fully satisfied the follow-up evaluation. The method made more requests and took 17% longer. A contrasting application-development pair produced no context or cost saving, and an earlier continuation exhibited lower manually assessed quality despite reduced context. These observations demonstrate substantial resource savings in noisy tool-use trajectories and identify workload dependence as a central consideration for reversible context management.

中文摘要

摘要：使用工具的智能体会反复携带观测信息，而其有用内容可能远小于原始数据量。我们研究了智能体控制的遗忘：执行模型选择先前观察到的工具结果，将每个结果替换为原位置的简短注释，并将精确原始内容保存在可恢复档案中。一个 Python 框架提供批量归档和显式恢复功能，无需针对特定任务进行模型训练，同时保护用户指令和助手消息不受这些操作影响。在一次探索性的 OpenTelemetry 调试案例及随后的一项无关实现任务中，该方法以 231,951 个提供者报告的提示令牌结束，而保留历史情况下为 912,492 个，累计输入令牌使用量减少了 50%，估计 API 成本为 1.28-1.44 美元，而保留历史大约为 4.38 美元。两种方法都通过了两个案例的主要行为预言测试；但在后续评估中均未完全满足要求。该方法发出的请求更多，耗时增加 17%。一对对比的应用开发任务未产生上下文或成本节约，而早期一次延续实验在减少上下文的情况下手动评估质量较低。这些观察结果表明在嘈杂的工具使用轨迹中可以显著节省资源，并指出工作负载依赖性是可逆上下文管理的核心考虑因素。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10590 (HTTP 429)

Authors: Jan-Peter Franke

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10590.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10590

Published: 2026-10-09T03:04:56.129Z

4. Verification and Self-Improvement in Agentic AI: Foundations and Limits

Abstract:Agentic AI systems can improve by searching longer, receiving additional support, or modifying how they propose and verify outputs. A performance score does not distinguish these mechanisms. We compare these changes through bounded verification with hidden terminal randomness. A stage specifies admissible transcripts, polynomial bounds, an alternating verification protocol, and a terminal checker. Its native reach uses default support; its closure frontier permits all support already admitted by the interface. Under a uniform pointwise probability gap and task-relative soundness, these are well-defined languages. We prove that independent majority amplification preserves both languages, whereas existential acceptance over random tapes can admit incorrect outputs. Exact verification is the zero-randomness case, with placement and completeness results. The randomized-verifier classes satisfy $\Sigmak^{\mathrm{P}}\subseteq\Sigma_k^{\mathrm{RV}}\subseteq\Sigma{k+1}^{\mathrm{P}}$; strict enlargement and depth separation require explicit complexity assumptions, while $\mathrm{BPP}=\mathrm{P}$ yields exact companions with the same frontiers. Representation analysis separates invariant acceptance from core-versus-support labels that can change under refactoring. For recursive self-improvement, uniformly bounded self-modification under a common sound interpreter and fixed verification protocol remains within the same verification class. A separate conditional-error budget controls false selection across adaptively chosen candidates. A quota-enforced XOR-synthesis family separates unbounded ratios of search success from changes in the accepted languages; exact and probabilistic audits check the resulting evidence requirements. The framework ties self-improvement claims to obligations on correctness, admissible evidence, verification resources, and selection error.

中文摘要

摘要：代理型AI系统可以通过延长搜索时间、获得额外支持或修改其提出和验证输出的方式来改进。性能评分无法区分这些机制。我们通过有界验证与隐藏终端随机性进行比较。阶段指定可接受的转录、多项式界限、交替验证协议和终端检查器。其原生覆盖范围使用默认支持;其闭合前沿允许接口已允许的所有支持。在均匀的逐点概率差距和任务相对合理性下，这些是定义良好的语言。我们证明独立多数放大保留了两种语言，而对随机磁带的存在接受则允许错误输出。精确验证是零随机性情况，具有位置和完备性结果。随机验证类满足$\Sigmak^{\mathrm{P}}\subseteq\Sigma_k^{\mathrm{RV}}\subseteq\Sigma{k+1}^{\mathrm{P}}$;严格放大和深度分离需要显式复杂性假设，而$\mathrm{BPP}=\mathrm{P}$则给出具有相同前沿的完全伴随。表示分析将不变接受度与核心与支持标签区分开来，这些标签在重构下可能发生变化。对于递归自我改进，在共同声音解释器和固定验证协议下的统一边界自我修改仍属于同一验证类。一个独立的条件错误预算控制自适应候选对象间的错误选择。配额强制的异或综合家族将检索成功率的无界比率与被接受语言的变化区分开;精确和概率审计检查所得证据要求。该框架将自我改进主张与正确性、可采证据、验证资源和选择错误的义务联系起来。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10611 (HTTP 429)

Authors: Chien-Ping Lu

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10611.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10611

Published: 2026-10-09T03:04:56.129Z

5. The Harness as the Only Mutable Surface: Compliance-Bounded Self-Evolution of LLM Agents in Credit Pipelines, with a Measured Admission Gate

Abstract:Self-improving LLM agents can adapt a credit pipeline to a changed rule, but an agent that rewrites itself destroys the artefact a supervisor reviews: a named change, a recorded test, an approval. We argue that self-evolution is reviewable only if it is confined to the runtime harness (instruction text, tool-call logic and primitive composition) while model weights stay fixed, so that every adaptation is a diff with a cause and a test attached. We give a dual-loop engine built on that bound, with one admission gate that writes a hash-chained record before deployment, and we measure the gate in simulation, with a simulated agent and a seeded-search proposer rather than language models. Across three families of supervisory re-interpretation at three severities, 10 seeds each, the gated loop admitted 144 of 7,449 candidate changes, none of which worsened error on held-out history, and restored the false-positive rate to the oracle level without raising missed flags in every low- and mid-severity cell. With the gate replaced by the check an unbounded system applies (fewer errors visible in recent traces), the same loops admitted 309 harmful changes and left missed flags above 10% in 49 of 90 runs: false positives fell because the screen was loosened. Evaluated on pre-shift labels, the gate rejected every candidate, so a re-interpretation must be encoded as a rule that relabels history. Parametric and scope shifts were repaired locally, a structural one only by primitive replacement; at the highest structural severity the gate’s fixed tolerance blocked the correct replacement in half the seeds. We map the mechanisms to the EU AI Act’s provisions for high-risk credit scoring and note that the April 2026 US model-risk guidance excludes agentic AI from its scope.

中文摘要

摘要：自我提升的大型语言模型（LLM）代理可以将信用流程适应于改变的规则，但一个会自我重写的代理会破坏监督者审核的人工制品：命名的变更、记录的测试、批准。我们认为，只有当自我进化被限制在运行时框架（指令文本、工具调用逻辑和原始组合）内，而模型权重保持不变时，自我进化才是可审查的，这样每一次适应都有附带原因和测试的差异。我们在该限制下提供了一个双循环引擎，其中一个准入门在部署前写入哈希链记录，并在模拟中测量该门控，使用模拟代理和带种子搜索的提出者，而不是语言模型。在三类不同严重程度的监督性重新解释中，每类10个种子，门控循环从7,449个候选变更中准入了144个，没有一个在保留历史上的错误率增加，并将误报率恢复到理想水平，同时在每个低、中严重程度单元中没有产生遗漏标记。当用无限制系统所执行的检查替代门控（近期痕迹中可见错误更少）时，相同循环准入了309个有害变更，并在90次运行中有49次遗漏标记超过10%：误报下降是因为筛查放宽。对预转移标签评估时，门控拒绝每一个候选，因此重新解释必须被编码为重新标记历史的规则。参数和范围的变更在局部得到修复，结构性变更仅能通过原语替换修复；在最高结构严重性下，门控的固定容忍度在一半的种子中阻止了正确的替换。我们将这些机制映射到欧盟《人工智能法》中针对高风险信用评分的规定，并指出2026年4月的美国模型风险指南将具有代理能力的人工智能排除在其范围之外。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10629 (HTTP 429)

Authors: Ravil Akhtyamov

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10629.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10629

Published: 2026-10-09T03:04:56.129Z

6. Speaking the Navigator’s Language: Trajectory-Grounded Instruction Translation for Frozen Aerial VLN Agents

Abstract:Aerial vision-and-language navigation (VLN) agents are typically trained on detail-rich, trajectory-aligned commands, whereas users issue short, intent-driven instructions; on a frozen OpenFly navigator, this \emph{instruction gap} drops success rate (SR) from $31.03\%$ to $11.33\%$. To scale translator training, we prompt a language model with human-written style examples to convert original commands into paired, intent-centered Weak commands, which yield $15.27\%$ SR. We introduce the \textbf{Trajectory-Grounded Instruction Translator (TGIT)}, a front-end that keeps the navigator frozen and translates Weak inputs into agent-executable commands by learning from its trajectory outcomes. The resulting Weak-trained translator raises Weak-input SR to $37.93\%$ and transfers zero-shot to real human instructions ($11.33\%{\rightarrow}32.51\%$); it also improves held-out OpenFly ($4.95\%{\rightarrow}20.79\%$) and yields recovery on CityNav and AirVLN.

中文摘要

摘要：空中视觉与语言导航（VLN）代理通常在细节丰富、轨迹对齐的指令上进行训练，而用户则发出简短的、以意图为驱动的指令；在冻结的OpenFly导航器上，这种\emph{指令差距}会将成功率（SR）从$31.03\%$降至$11.33\%$。为了扩展翻译器训练，我们使用带有人类书写风格示例的语言模型，将原始指令转换为配对的、以意图为中心的弱指令，从而获得$15.27\%$的成功率。我们引入了\textbf{轨迹引导指令翻译器（TGIT）}，这是一个前端模块，在保持导航器冻结的情况下，通过学习其轨迹结果，将弱输入翻译为代理可执行的指令。由弱指令训练的翻译器将弱输入的成功率提高到$37.93\%$，并可零样本迁移到真实人类指令（$11.33\%{
ightarrow}32.51\%$）；它还提高了未见的OpenFly成功率（$4.95\%{
ightarrow}20.79\%$），并在CityNav和AirVLN中实现恢复效果。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10635 (HTTP 429)

Authors: Xi Chen, Zhe Liu, Xiaogang Xu, Jiafei Xu, Chunyi Zhou, Yuan Su, Rui Zeng, Tianyu Du, Kelu Yao, Chao Li, Shouling Ji

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10635.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10635

Published: 2026-10-09T03:04:56.129Z

7. Plan-and-Patch: Diffusion Language Models for Agentic Planning

Abstract:Planning is increasingly important for long-horizon agents, where successful execution requires coordinating subgoals, tool use, and intermediate outcomes over many steps. Yet assumptions made during planning may be invalidated by the environment, tools may return unexpected results, or actions may fail. Effective agents must therefore not only generate plans, but also revise them. Such revisions often affect only part of a plan, leaving the preceding and subsequent structure intact. Rather than regenerate the entire plan and risk unnecessary changes, repair can regenerate the affected region conditioned on the preserved prefix and suffix. We introduce Plan-and-Patch, a plan-and-act framework in which a diffusion language model (dLLM) generates a structured, program-like plan through parallel unmasking and repairs it by filling in selected regions while keeping the surrounding steps fixed. We compare DreamReasoner-8B and Qwen3-8B as diffusion and autoregressive (AR) planners. On Natural Plan without task-specific training, diffusion (53.7%) achieves nearly twice the plan repair success rate of AR (27.0%). After task-specific training on agentic benchmarks, ALFWorld and TextCraft, the planners achieve similar observed success in plan generation, while diffusion reduces mean plan-generation latency by 39-46% relative to AR. Our results show that Plan-and-Patch provides a framework for faster plan generation and effective plan repair in long-horizon agents.

中文摘要

摘要：规划对于长时程智能体越来越重要，在这种情况下，成功执行需要在多个步骤中协调子目标、工具使用和中间结果。然而，在规划过程中所做的假设可能会被环境所否定，工具可能返回意想不到的结果，或者动作可能会失败。因此，有效的智能体不仅必须生成计划，还必须对其进行修订。这些修订通常只影响计划的部分内容，而保持前后结构不变。与其重新生成整个计划并冒着不必要更改的风险，不如在保留的前缀和后缀条件下重新生成受影响的区域，从而进行修复。我们提出了“计划与补丁（Plan-and-Patch）”，这是一种计划与执行框架，其中扩散语言模型（dLLM）通过并行去掩码生成结构化、类似程序的计划，并通过在保持周围步骤固定的情况下填充选定区域来进行修复。我们比较了 DreamReasoner-8B 和 Qwen3-8B 作为扩散和自回归（AR）规划器。在未进行任务特定训练的 Natural Plan 上，扩散（53.7%）的计划修复成功率几乎是 AR（27.0%）的两倍。在针对代理基准 ALFWorld 和 TextCraft 进行任务特定训练后，这些规划器在计划生成上的观察成功率相似，而扩散相比 AR 将平均计划生成延迟减少了 39-46%。我们的结果表明，Plan-and-Patch 为长时程智能体提供了更快的计划生成和有效的计划修复框架。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10786 (HTTP 429)

Authors: Syamantak Kumar, Jiang Guo, Hassan Hamad, Hideo Kobayashi, Yi Xiang, Yezhou Yang, Yanjun Qi, Daniele Bonadiman, Jiarong Jiang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10786.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10786

Published: 2026-10-09T03:04:56.129Z

8. Whose Ground Truth? Embracing Ambiguity in Human-Centered AI

Abstract:As AI systems increasingly interact with people and make decisions about them, understanding human interpretations becomes an important part of developing human-centered AI. Conventional machine learning and AI systems are largely developed under the assumption that a single definitive ground truth exists, with variability in human annotations often resolved through aggregation or treated as noise. However, for many human-centered tasks, human interpretation is inherently ambiguous, and multiple interpretations of the same input may be simultaneously reasonable and valid. Reducing such ambiguity to a single target risks overlooking meaningful information about the diversity of human perception, judgment, and experience. In this position paper, we call for a shift towards modeling the interpretation space of plausible human judgments, while distinguishing meaningful ambiguity from annotation noise. We argue that this perspective should guide how AI systems are represented, learned, evaluated, deployed, and governed, supporting more human-centered AI that better reflects the diversity of human interpretation.

中文摘要

摘要：随着人工智能系统越来越多地与人互动并对人做出决策，理解人类的解读已成为开发以人为本的人工智能的重要组成部分。传统的机器学习和人工智能系统在很大程度上是基于存在单一明确的客观真理的假设开发的，人类标注的差异通常通过汇总解决或被视为噪音。然而，对于许多以人为中心的任务来说，人类解读本质上是模糊的，对同一输入的多种解读可能同时合理且有效。将这种模糊性简化为单一目标可能忽视有关人类感知、判断和经验多样性的有意义信息。在这篇立场论文中，我们呼吁转向建模合理人类判断的解读空间，同时区分有意义的模糊性与标注噪音。我们认为，这一视角应指导人工智能系统的表示、学习、评估、部署和治理，从而支持更具以人为本的人工智能，更好地反映人类解读的多样性。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10805 (HTTP 429)

Authors: Jingyao Wu, Mohammad Tariqul Islam, Per Rådberg Nagbøl, Julie Gerlings, Giovanni Leoni, Georgiana Fryza, Lars Pilegaard Thomsen, Jakob Mainz

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10805.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10805

Published: 2026-10-09T03:04:56.129Z

9. On the Clock: Towards Punctual and Productive Time-Budgeted AI Agents

Abstract:We study whether small LLM agents can operate effectively under explicit wall-clock time budgets by both respecting the allocated runtime and using available time productively. We evaluate Qwen3.6-27B on five competitions from MLE-Bench Lite and Qwen3-4B on Zork I (Jericho), two agentic benchmarks where additional computational time can meaningfully improve performance. In the simplest setting, where the budget is stated only in the prompt, agents fail to translate the stated budget into controlled use of time. These failures arise from gaps in time awareness, since the harness provides no timing feedback, but also because they cannot reliably anticipate the duration of actions, and do not have a learned mapping from available time to an appropriate strategy. We investigate two complementary classes of interventions: harness-based mechanisms that expose timing information and enforce deadlines, and reinforcement learning with budget-aware rewards. Injecting timing information through the harness substantially improves budget adherence for Qwen3.6-27B without measurable loss in performance, while enforcement hooks tighten adherence further. RL with GRPO achieves near-perfect budget adherence on Zork I and generalizes to held-out budgets not seen during training, but does not improve task performance over the untrained harness on MLE-Bench. Once agents are made to respect the budget, they still fail to use additional time to improve task performance. RL-trained policies learn when to stop but often fill extra time with repeated actions, and GRPO training on multiple budgets tends to collapse toward the strategy learned for the shortest budget. Our results reveal a gap between time adherence and productive time allocation, which remains a central challenge for budget-conditioned agents.

中文摘要

摘要：我们研究小型大型语言模型（LLM）代理是否能够在明确的实际时间预算下有效运行，同时既遵守分配的运行时间，又能有效利用可用时间。我们在来自 MLE-Bench Lite 的五项竞赛中评估了 Qwen3.6-27B，并在 Zork I（Jericho）上评估了 Qwen3-4B，这两个代理基准中额外的计算时间可以显著提高性能。在最简单的设置中，当预算仅在提示中说明时，代理无法将所述预算转化为对时间的控制性使用。这些失败源于时间感知的缺失，因为测试环境没有提供时间反馈，同时也因为代理无法可靠地预估动作的持续时间，并且没有从可用时间到合适策略的学习映射。我们研究了两类互补的干预方法：基于测试环境的机制，用于暴露时间信息并执行截止时间，以及带有预算感知奖励的强化学习。通过测试环境注入时间信息显著提高了 Qwen3.6-27B 的预算遵守情况，同时性能没有可测的下降，而执行钩子进一步增强了预算遵守的严格性。使用 GRPO 的强化学习在 Zork I 上实现了几乎完美的预算遵守，并能推广到训练期间未见的保留预算，但在 MLE-Bench 上并未提升任务性能，相比未经训练的测试环境。一旦代理被迫遵守预算，它们仍然未能利用额外时间来提升任务性能。经过强化学习训练的策略学会了何时停止，但往往用重复动作来填充额外时间，并且在多个预算上进行 GRPO 训练时，策略往往会收敛到针对最短预算学到的策略。我们的结果揭示了时间遵守与有效时间分配之间的差距，这仍然是预算条件代理的核心挑战。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10833 (HTTP 429)

Authors: Aaron Wang, Neelabh Madan, Vlad Sobal, Matthew Trager, Michael Kleinman, Elman Mansimov, Wei Xia, Stefano Soatto

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10833.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10833

Published: 2026-10-09T03:04:56.129Z

10. Self-Supervised Keyframe Discovery for Horizon-Invariant Behavior Cloning

Abstract:Behavior cloning (BC) in non-Markovian environments is a challenging problem because policies have to reason over contextual information over long horizons. Existing policy architectures rely on recurrent or attention-based mechanisms to capture long-term dependencies. However, recurrent models suffer from hidden-state collapse and gradient instability under backpropagation through time, while attention-based models are fundamentally limited by context length. To address these issues, we propose Keyframe Mnemonics, a novel self-supervised method that $\textit{discovers}$ a set of information-critical observations ($\textit{mnemonics}$) by learning an objective from randomly sampled past observations and using it as a reward for keyframe selection. We then train a BC policy that conditions on the discovered keyframes to model the action distribution. Under certain task-structure assumptions, our formulation provides context retention guarantees over an infinite horizon, while maintaining a small set of decision-relevant keyframes in the policy’s working memory. We evaluate our method on synthetic memory domains, where mnemonic-conditioned BC policies achieve $100$% success rates (SR) and generalize to horizons orders of magnitude beyond training without performance degradation. Additionally, we evaluate on memory-intensive robot manipulation benchmark, achieving a $13.9$% average absolute SR improvement over the strongest baseline across $23$ tasks and retaining $80$% SR at $20\times$ longer horizons on a real robot. Code and videos are available at this https URL.

中文摘要

摘要：非马尔可夫环境中的行为克隆（BC）是一个具有挑战性的问题，因为策略必须在长期视野内对上下文信息进行推理。现有策略架构依赖于循环或基于注意力的机制来捕捉长期依赖关系。然而，循环模型在反向传播过程中存在隐藏状态崩溃和梯度不稳定性，而基于注意力的模型则在上下文长度的基础上受限。为解决这些问题，我们提出了关键帧助记法，这是一种新颖的自监督方法，通过从随机抽样的过去观察中学习目标并作为关键帧选择的奖励，对一组关键信息进行 $\textit{discovers}$。然后我们训练一个基于发现的关键帧来建模动作分布的 BC 策略。在某些任务结构假设下，我们的表述在无限视野内保证上下文保留，同时在策略工作内存中保留少量决策相关关键帧。我们在合成记忆域中评估了该方法，其中助记条件BC策略实现了100%的成功率（SR），并在训练后推广到超出训练的视野上多个数量级且性能不下降。此外，我们基于内存密集型机器人操作基准进行评估，在23美元任务中，平均绝对SR提升达到13.9%美元，且在真实机器人中，在20美元/倍值更长的时间线下保持80%的SR。代码和视频可在此 https URL 获取。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10857 (HTTP 429)

Authors: Prabin Kumar Rath, Omkar Patil, Nakul Gopalan

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10857.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10857

Published: 2026-10-09T03:04:56.129Z

Evaluation Domain Papers

1. An Explainable Header-Centric Framework for Large-Scale Semantic Table Interpretation and Data Quality Assessment

中文摘要

摘要：知识图谱（KG）的质量不仅取决于下游的图验证，还取决于整合前使用的表格元数据的质量。在仅元数据的语义表格解释（STI）中，当单元格值不可用、噪声较大或不适用时，列标题成为可追溯知识图谱准备的重要语义证据来源。我们提出了一种可解释的、以标题为中心的框架，用于仅基于元数据的列类型标注（CTA）和数据质量评估（DQA）。该框架使用策划的词汇资源将标题映射到39种可解释的最终格式类型，并通过源关键字（SourceKeywords）保留了令牌级的可追溯性。每个分配的类型都会基于数据质量问题（DQIs）分类激活验证规则，从而产生检测结果，如缺失数据、重复数据、域违规、数据类型错误和时间不匹配等。这些检测结果被汇总为HeadersIQ，一种轻量级、无权重的数据源级质量指标。该框架在异质基准测试中进行了评估，包括UCI、Prague、Kaggle、VizNet/Sato、SOTAB、T2Dv2，以及SemTab 2024 Metadata-to-KG赛道，共涉及约120,000个列标题。结果显示，该方法在噪声较大的真实世界元数据中具有广泛的实用覆盖度，同时平行的知识图谱映射路径支持与DBpedia和该http URL的对齐。在SemTab 2024 Metadata-to-KG赛道上，官方GT严格评估结果一般。然而，盲审诊断显示，许多不匹配反映了基准的粒度、别名和本体选择效应，而非完全不合理的以标题为中心的预测。我们将该审计报告作为关于不一致模式的诊断证据，而非修正的基准性能。总体而言，本文提出了一个可重用的工作流程，用于基于元数据的语义标注、数据源级质量监控以及面向知识图谱的基准诊断。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10541 (HTTP 429)

Authors: Marcelo Valentim Silva, Hannes Herrmann, Valerie Maxville

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10541.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10541

Published: 2026-10-09T03:05:19.700Z

2. Synthesis Through Simulation: Generating Coherent Enterprise Data via Scalable Agent-System Interaction

中文摘要

摘要：调用工具的代理已经成为企业 AI 的核心，但由于企业系统、数据和数据库模式的业务和法律限制，对它们进行大规模训练和评估仍然受到很大限制。表格数据合成提供了一种自然的替代方案，但其有效性从根本上受到结构有效性和模式可用性的限制，而基于流程的方法则表现出相反的弱点，通常在没有每个领域创作的情况下缺乏分布一致性。我们提出了通过模拟合成（STS），这是一种无模式的数据合成范式，其中 LLM 代理通过在模拟企业环境中针对执行政策的 API 执行操作来生成数据。由于数据是通过定义有效性的同一环境生成的，STS 凭借其构建确保了结构有效性，同时将有效性执行与分布建模解耦，使它们能够独立处理。STS 的领域无关代理 通用填充器（GP）解决了分布一致性和合成可扩展性的其余挑战：GP 在所有十个环境中实现了平均边际一致性0.88和100% 的约束满足度，无需访问数据库模式，而统计合成器由于必要的种子数据要求在七个环境中无法使用，模式优先代理在航空环境的紧密耦合工作流中 82% 的轨迹失败，原因是任务组合脆弱。我们已在此 https URL 开源了完整框架、全部十个环境及生成的数据集。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10549 (HTTP 429)

Authors: Yipeng Li, Ashutosh Hathidara, Jane Lo, Harshavardhan Abichandani, Gunraj Singh, Atin Ghosh

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10549.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10549

Published: 2026-10-09T03:05:19.700Z

3. Agent-Controlled Forgetting for Tool-Using Agents: Reversible Context Curation in Practice

中文摘要

摘要：使用工具的智能体会反复携带观测信息，而其有用内容可能远小于原始数据量。我们研究了智能体控制的遗忘：执行模型选择先前观察到的工具结果，将每个结果替换为原位置的简短注释，并将精确原始内容保存在可恢复档案中。一个 Python 框架提供批量归档和显式恢复，而无需任务特定的模型训练，同时保护用户指令和助手消息不受这些操作的影响。在一个探索性的 OpenTelemetry 调试案例之后，紧接着进行一个无关的实现任务，该方法最终生成 231,951 个提供者报告的提示 tokens，而保留历史记录时为 912,492 个，累计输入 tokens 使用量减少了 50%，估计 API 成本为 1.28-1.44 美元，而之前约为 4.38 美元。两个实验均通过了两个案例的主要行为预言测试；但没有一个完全满足后续评估。该方法发出的请求更多，耗时增加了 17%。一个对比的应用开发实例未产生上下文或成本节约，而早期的延续案例尽管减少了上下文量，但人工评估质量较低。这些观察表明，在噪声较大的工具使用路径中可实现显著的资源节约，并指出工作负载依赖性是可逆上下文管理的重要考量因素。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10590 (HTTP 429)

Authors: Jan-Peter Franke

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10590.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10590

Published: 2026-10-09T03:05:19.700Z

4. Verification and Self-Improvement in Agentic AI: Foundations and Limits

中文摘要

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10611 (HTTP 429)

Authors: Chien-Ping Lu

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10611.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10611

Published: 2026-10-09T03:05:19.700Z

5. The Harness as the Only Mutable Surface: Compliance-Bounded Self-Evolution of LLM Agents in Credit Pipelines, with a Measured Admission Gate

中文摘要

摘要：自我提升的大型语言模型（LLM）代理可以将信用流程适应于改变的规则，但一个自我重写的代理会破坏主管审核的工件：命名的变更、记录的测试、批准。我们认为，自我进化只有在限制在运行时框架（指令文本、工具调用逻辑和原始组合）内且模型权重保持固定时才可审核，这样每次适应都有原因和测试附带的差异。我们提供了一个基于该约束的双环引擎，在部署前通过一个准入门写入哈希链记录，并在模拟中测量该门，包括模拟代理和种子搜索提议者，而非语言模型。在三种监督性重新解释的三种严重程度下，每种情况10个种子，门控循环从7,449个候选变更中接纳了144个，其中没有一个在留出历史上的错误恶化，并且将假阳性率恢复到理想水平，同时在每个低、中严重程度单元中没有遗漏标志。在将门替换为无限制系统应用的检查（最近的跟踪中错误更少）情况下，同样的循环接纳了309个有害变更，并在90次运行中有49次遗漏标志超过10%：假阳性下降是因为筛选标准放宽。在迁移前标签上评估时，门拒绝了每个候选，因此重新解释必须被编码为重新标注历史的规则。参数和范围的变化可以局部修复，结构性变化只能通过原始组件替换修复；在最高结构严重程度下，门的固定容差阻止了一半种子中正确的替换。我们将这些机制映射到欧盟人工智能法案（EU AI Act）中对高风险信用评分的规定，并指出2026年4月的美国模型风险指南将代理式人工智能排除在其范围之外。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10629 (HTTP 429)

Authors: Ravil Akhtyamov

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10629.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10629

Published: 2026-10-09T03:05:19.700Z

6. Speaking the Navigator’s Language: Trajectory-Grounded Instruction Translation for Frozen Aerial VLN Agents

中文摘要

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10635 (HTTP 429)

Authors: Xi Chen, Zhe Liu, Xiaogang Xu, Jiafei Xu, Chunyi Zhou, Yuan Su, Rui Zeng, Tianyu Du, Kelu Yao, Chao Li, Shouling Ji

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10635.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10635

Published: 2026-10-09T03:05:19.700Z

7. Plan-and-Patch: Diffusion Language Models for Agentic Planning

中文摘要

摘要：规划对于长时程智能体越来越重要，在这种情况下，成功执行需要在多个步骤中协调子目标、工具使用和中间结果。然而，在规划过程中所做的假设可能会被环境所否定，工具可能返回意想不到的结果，或者动作可能会失败。因此，有效的智能体不仅必须生成计划，还需对其进行修订。这类修订通常只影响计划的部分区域，而保持前后结构不变。与其重新生成整个计划并冒着不必要更改的风险，不如在保留的前缀和后缀条件下重新生成受影响的区域，从而进行修复。我们提出了“计划与补丁（Plan-and-Patch）”，这是一种计划与执行框架，其中扩散语言模型（dLLM）通过并行去掩码生成结构化、类似程序的计划，并通过在固定周围步骤的同时填充选定区域进行修复。我们将DreamReasoner-8B和Qwen3-8B分别作为扩散型和自回归（AR）规划器进行比较。在未进行任务特定训练的Natural Plan上，扩散型（53.7%）几乎是自回归（27.0%）计划修复成功率的两倍。经过在智能体基准ALFWorld和TextCraft上的任务特定训练后，这些规划器在计划生成的实际成功率方面表现相似，而扩散型相对于自回归在平均计划生成延迟上减少了39-46%。我们的结果表明，Plan-and-Patch为长时程智能体提供了一个更快生成计划和有效修复计划的框架。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10786 (HTTP 429)

Authors: Syamantak Kumar, Jiang Guo, Hassan Hamad, Hideo Kobayashi, Yi Xiang, Yezhou Yang, Yanjun Qi, Daniele Bonadiman, Jiarong Jiang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10786.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10786

Published: 2026-10-09T03:05:19.700Z

8. Whose Ground Truth? Embracing Ambiguity in Human-Centered AI

中文摘要

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10805 (HTTP 429)

Authors: Jingyao Wu, Mohammad Tariqul Islam, Per Rådberg Nagbøl, Julie Gerlings, Giovanni Leoni, Georgiana Fryza, Lars Pilegaard Thomsen, Jakob Mainz

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10805.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10805

Published: 2026-10-09T03:05:19.700Z

9. On the Clock: Towards Punctual and Productive Time-Budgeted AI Agents

中文摘要

摘要：我们研究小型大型语言模型（LLM）代理是否能够在明确的实际时间预算下有效运行，同时既遵守分配的运行时间，又能有效利用可用时间。我们在来自 MLE-Bench Lite 的五项竞赛中评估了 Qwen3.6-27B，并在 Zork I（Jericho）上评估了 Qwen3-4B，这两个代理基准中额外的计算时间可以显著提高性能。在最简单的设置中，当预算仅在提示中说明时，代理无法将所述预算转化为对时间的控制性使用。这些失败源于时间感知的缺失，因为测试环境没有提供时间反馈，同时也因为代理无法可靠地预估动作的持续时间，并且没有从可用时间到合适策略的学习映射。我们研究了两类互补的干预方法：基于测试环境的机制，用于暴露时间信息并执行截止时间，以及带有预算感知奖励的强化学习。通过测试环境注入时间信息显著提高了 Qwen3.6-27B 的预算遵守情况，同时性能没有可测的下降，而执行钩子进一步增强了预算遵守的严格性。使用 GRPO 的强化学习在 Zork I 上实现了几乎完美的预算遵守，并能推广到训练期间未见的保留预算，但在 MLE-Bench 上并未提升任务性能，相比未经训练的测试环境。一旦代理被迫遵守预算，它们仍然无法利用额外时间来提升任务表现。经过强化学习训练的策略学会了何时停止，但常常用重复动作填充额外时间，而在多个预算上进行 GRPO 训练时，策略往往会趋向于最短预算所学的策略。我们的结果揭示了时间遵守与生产性时间分配之间的差距，这仍然是预算条件代理的一项核心挑战。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10833 (HTTP 429)

Authors: Aaron Wang, Neelabh Madan, Vlad Sobal, Matthew Trager, Michael Kleinman, Elman Mansimov, Wei Xia, Stefano Soatto

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10833.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10833

Published: 2026-10-09T03:05:19.700Z

10. Self-Supervised Keyframe Discovery for Horizon-Invariant Behavior Cloning

中文摘要

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10857 (HTTP 429)

Authors: Prabin Kumar Rath, Omkar Patil, Nakul Gopalan

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.10857.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10857

Published: 2026-10-09T03:05:19.700Z

VLM Domain Papers

1. SLVR: Structured Latent Visual Reasoning via Human-like Reasoning Flows

Abstract:Multimodal large language models (MLLMs) often answer visual reasoning questions by relying on linguistic priors rather than task-relevant visual evidence. Textual chain-of-thought reasoning can partially mitigate this issue by encouraging models to decompose visual questions into intermediate evidence-seeking steps, but generating these steps autoregressively increases inference cost. Latent reasoning avoids explicit rationale generation, but existing approaches provide limited control over what intermediate states encode, making it difficult to impose separate supervision for planning, grounding, and evidence selection. We propose Structured Latent Visual Reasoning (SLVR), a training framework that bridges explicit chain-of-thought and latent reasoning by organizing multimodal reasoning into typed latent stages for planning, grounding, evidence selection, and reasoning integration. SLVR first trains the model to rely on the image by masking answer-revealing text and contrasting the correct answer with visually plausible distractors. It then organizes reasoning into latent stages for planning, grounding, evidence selection, and integration, supervising each stage with the corresponding signal: plans, boxes, visual evidence, and final rationales. This gives latent reasoning an explicit functional structure while avoiding generated textual chains at inference time. Built on Qwen2.5-VL-7B, SLVR improves consistently across multimodal reasoning benchmarks, with absolute gains of +9.4 on MMVP and +14.2 on BLINK Relation, as well as improvements on V*, MathVista, and ChartQA. These results suggest that structured latent supervision can improve fine-grained visual reasoning without the decoding overhead of textual CoT. Project page is available \href{this https URL}{here}.

中文摘要

摘要：多模态大型语言模型（MLLM）通常依赖语言先验而非任务相关的视觉证据来回答视觉推理问题。文本思维链推理可以通过鼓励模型将视觉问题分解为中间的证据寻求步骤来部分缓解这一问题，但生成这些步骤会自回归增加推理成本。潜在推理避免了显式理据生成，但现有方法对中间状态编码内容的控制有限，难以对规划、基础和证据选择施加单独监督。我们提出了结构化潜在视觉推理（SLVR）的培训框架，通过将多模态推理组织为类型化的潜在阶段，用于规划、基础、证据选择和推理整合，桥接显式思维链和潜在推理。SLVR首先通过掩盖揭示答案的文本，并将正确答案与视觉上合理的干扰因素对比，来训练模型依赖图像。然后将推理组织为潜在阶段，用于规划、基础化、证据选择和整合，并用相应信号监督每个阶段：计划、方框、视觉证据和最终推理。这为潜在推理提供了显式的功能结构，同时避免了推理时生成的文本链。基于Qwen2.5-VL-7B，SLVR在多模态推理基准测试中持续提升，MMVP绝对提升为+9.4，BLINK关系提升+14.2，V*、MathVista和ChartQA也有改进。这些结果表明，结构化潜在监督可以在不承担文本CoT解码开销的情况下改善细粒度的视觉推理。项目页面可访问 \href{this https URL}{here}。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10563 (HTTP 429)

Authors: Albert Gao, Bing Xue, Andrea Zanette

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.10563.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10563

Published: 2026-10-09T03:05:44.122Z

2. A Camera-Native Stereo VR180 Dataset

Abstract:Immersive VR180 video is increasingly produced with professional stereo fisheye cameras, yet public VR180 research resources are mostly collected from online platforms such as YouTube: already stitched, projected and compressed by unknown pipelines, and without lens calibration. We present a firsthand-captured stereo VR180 dataset recorded with two Blackmagic URSA Cine Immersive cameras. It contains 1,211 samples — 636 stereo video clips (2,220.8 s, mostly 90 fps) and 575 stereo stills — each released as camera-native Blackmagic RAW, separate-eye native fisheye HEVC (8160x7200 per eye) and half-equirectangular HEVC (7200x7200 per eye), together with the factory lens calibration, portable fisheye/half-equirectangular conversion tools and AI-generated scene and visual-challenge annotations. Re-encoding the released fisheye and half-equirectangular renders with x265 over 24 clips, both eyes, four rate points and nine viewing directions, native-fisheye coding needed more bitrate than half-equirectangular coding at equal viewport quality for all 24 clips (median +38%), in every part of the field of view. Data: this https URL ; code: this https URL

中文摘要

摘要：沉浸式 VR180 视频越来越多地使用专业立体鱼眼相机制作，但公开的 VR180 研究资源大多来自 YouTube 等网络平台：已经被未知流程拼接、投影和压缩，并且没有镜头校准。我们提供了一个由两台 Blackmagic URSA Cine 沉浸式相机首次拍摄的立体 VR180 数据集。它包含 1,211 个样本——636 个立体视频片段（2,220.8 秒，大部分为 90 fps）和 575 张立体静态图——每个样本都以相机原生 Blackmagic RAW、单眼原生鱼眼 HEVC（每眼 8160x7200）和半等矩形 HEVC（每眼 7200x7200）发布，同时附带工厂镜头校准、便携式鱼眼/半等矩形转换工具以及 AI 生成的场景和视觉挑战标注。将已发布的鱼眼和半等矩形渲染使用 x265 重新编码，涵盖 24 个片段、双眼、四个码率点和九个观看方向，对所有 24 个片段来说，在相同视口质量下，原生鱼眼编码所需码率比半等矩形编码更高（中位数 +38%），在视场的每个部分均如此。数据：https URL ; 代码：https URL

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10607 (HTTP 429)

Authors: Linxuan Lu

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.10607.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10607

Published: 2026-10-09T03:05:44.122Z

3. Seeing Through the Glare: A Multi-Source Benchmark and Ocular-Adaptive Pixel MeanFlow for Eyeglass Reflection Removal

Abstract:Eyeglass reflection removal is important across smartphone imaging, video conferencing, and other face-centric visual applications. The task is challenging because reflections range from mild photometric contamination to severe ocular occlusion, requiring selective correction and plausible reconstruction without altering identity or natural appearance. Existing datasets cover limited reflection conditions, constraining generalization to complex real-world scenes and systematic evaluation. We introduce \textbf{OcuBench}, a multi-source benchmark comprising 10,280 controllable synthetic pairs, 732 real-input pseudo-pairs, and 458 independent real-world test images, supporting both paired evaluation and assessment beyond generated supervision. We further propose \textbf{OcuFlow}, an ocular-adaptive pixel MeanFlow (pMF) framework for efficient, detail-preserving restoration. It combines geometry-adaptive representation with one-step pMF to focus reconstruction on reflection-obscured ocular regions, together with native-resolution frequency-preserving synthesis to retain reliable observed details. Experiments across diverse reflection conditions demonstrate that OcuFlow achieves consistent advantages in reflection removal quality, ocular fidelity, and efficiency. In a blind user study, it receives $67.32\%$ of selections, $6.2\times$ the next-best share. Both the code and dataset will be released.

中文摘要

摘要：眼镜反光去除在智能手机成像、视频会议以及其他以面部为中心的视觉应用中非常重要。这一任务具有挑战性，因为反光现象可能从轻度光度污染到严重眼部遮挡不等，需要在不改变身份或自然外观的情况下进行选择性校正和合理重建。现有数据集涵盖的反光条件有限，限制了其在复杂现实场景中的泛化能力和系统评估。我们引入了\textbf{OcuBench}，一个多源基准数据集，包括 10,280 对可控的合成图像对、732 对真实输入伪对以及 458 张独立的真实世界测试图像，支持配对评估和超越生成监督的评估。我们进一步提出了\textbf{OcuFlow}，一种眼部自适应像素平均流（pMF）框架，用于高效、细节保留的修复。它将几何自适应表示与一步 pMF 相结合，重点重建被反光遮挡的眼部区域，并通过原生分辨率频率保留合成保留可靠的观察细节。在多种反光条件下的实验表明，OcuFlow 在反光去除质量、眼部保真度和效率方面均表现出一致优势。在一项盲用户研究中，它获得了 $67.32\%$ 的选择份额，是次佳方案的 $6.2\times$。代码和数据集都将公开发布。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10703 (HTTP 429)

Authors: Tao Liu, Youwei Pang, Kailai Zhou, Jiaming Zuo, Hanqi Liu, Wei Ji, Peng-Tao Jiang, Xiaofeng Liu, Weisi Lin, Xiaoqi Zhao

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.10703.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10703

Published: 2026-10-09T03:05:44.122Z

4. LinSlot: Exploiting Linear Representation hypothesis for unsupervised attribute discovery from slot based object representation

Abstract:This paper studies the problem of learning disentangled representations of objects and their attributes from raw, unstructured image data. Slot-based methods have shown considerable success in unsupervised learning of object representations from images. Block-slot attention-based methods extend this framework to attribute representations by assuming a uniform factorization of object representations into attributes, which may be suboptimal and consequently limit the quality of the learned representations. We therefore investigate a framework for jointly discovering object and attribute representations. Our key contribution is leveraging the Linear Representation Hypothesis (LRH), which postulates that composable concepts can be represented as linearly additive subspaces in slot representations. Based on this insight, we propose a probabilistic model connecting images, slots (objects), and blocks (attributes). We present an architecture that leverages block attention to connect attribute representations to slots and incorporates LRH in both object and attribute representation spaces. This architecture effectively optimizes the Evidence Lower Bound (ELBO) of the proposed graphical model. Our experiments demonstrate (i) effective discovery of disentangled object and attribute representations, (ii) empirical evidence for LRH in slot space, and (iii) the ability to perform image editing owing to the disentangled and interpretable nature of the learned representations. Our experiments on multiple datasets demonstrate improvements in DCI scores over state-of-the-art methods.

中文摘要

摘要：本文研究了从原始、非结构化图像数据中学习对象及其属性的解耦表示的问题。基于插槽的方法在从图像中进行对象表示的无监督学习方面已显示出显著成功。基于块-插槽注意力的方法通过假设对象表示可以均匀分解为属性来将这一框架扩展到属性表示，这一假设可能不是最优的，从而限制了所学习表示的质量。因此，我们研究了一种联合发现对象和属性表示的框架。我们的关键贡献是利用线性表示假说（Linear Representation Hypothesis, LRH），该假说提出，可组合的概念可以在插槽表示中作为线性可加子空间进行表示。基于这一见解，我们提出了一个连接图像、插槽（对象）和块（属性）的概率模型。我们设计了一种架构，该架构利用块注意力将属性表示连接到插槽，并在对象和属性表示空间中都引入LRH。该架构有效地优化了所提出图模型的证据下界（ELBO）。我们的实验表明：(i) 能够有效发现解耦的对象和属性表示，(ii) 在插槽空间中为LRH提供了实证证据，(iii) 由于所学习表示的解耦和可解释性，可以进行图像编辑。我们在多个数据集上的实验显示，DCI得分较最先进方法有所提升。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10722 (HTTP 429)

Authors: Sanket Gandhi, Utkarsh Giri, Varun Subramanium, Rohan Paul, Parag Singla

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.10722.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10722

Published: 2026-10-09T03:05:44.122Z

5. MESSENGER: Memory-Enhanced Sequential Scene Flow Estimation via Autoregressive Next-Frame Forecasting

Abstract:Scene flow can capture low-level 3D motion displacements in dynamic scenarios. Early pairwise estimators relying on instantaneous two-frame motion lack long-term temporal correlation and also struggle with poor extrapolation ability in future prediction. Although some recent methods attempt to explore multi-frame scene flow estimation in a sequence-to-sequence manner, they typically suffer from heavy computational overhead with increasing input frames and long-horizon prediction degradation due to ineffective motion propagation. To address these problems, we propose a novel memory-enhanced sequential scene flow pipeline, called MESSENGER. To sufficiently mine long-term temporal dependencies naturally within consecutive sequences, a memory buffer is designed by explicitly storing multiple history flow estimates and latent states. For each input frame, the temporally stored flows and states are correlated and retrieved to predict the current initialized flow in a next-frame forecasting manner. Furthermore, we develop an uncertainty-aware reweighting module to filter unreliable retrievals and mitigate accumulated errors. Extensive experiments on nuScenes and Argoverse 2 demonstrate state-of-the-art performance of our MESSENGER, reducing EPE3D by 71.6% on nuScenes and 67.7% on Argoverse 2 in long-horizon future extrapolation. This superiority can be attributed to our designed autoregressive forecasting paradigm, which naturally forces the network to progressively learn the next-frame distribution based on history observations. Code will be released at this https URL.

中文摘要

摘要：场景流可以捕捉动态场景中的低级别三维运动位移。依赖于瞬时两帧运动的早期成对估计器缺乏长期时间相关性，并且在未来预测中表现出较差的外推能力。虽然一些近期方法尝试以序列到序列的方式探索多帧场景流估计，但随着输入帧数的增加，它们通常会面临计算开销大的问题，并且由于运动传播效率低，长时间预测表现下降。为了解决这些问题，我们提出了一种新颖的增强记忆的顺序场景流处理流程，称为MESSENGER。为了充分挖掘连续序列中的长期时间依赖性，我们设计了一个记忆缓冲区，通过显式存储多个历史流估计和潜在状态。对于每一帧输入，临时存储的流和状态会被关联检索，以下一帧预测的方式预测当前的初始化流。此外，我们开发了不确定性感知重加权模块，用于过滤不可靠的检索结果并减轻累计误差。在nuScenes和Argoverse 2上的大量实验表明，我们的MESSENGER表现出最先进的性能，在长时段未来外推中，在nuScenes上将EPE3D降低了71.6%，在Argoverse 2上降低了67.7%。这种优势可归因于我们设计的自回归预测范式，它自然地迫使网络基于历史观测逐步学习下一帧的分布。代码将通过此https网址发布。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10759 (HTTP 429)

Authors: Jiuming Liu, Jianing Li, Mengmeng Liu, Hongyang He, Hesheng Wang, Per Ola Kristensson

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.10759.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10759

Published: 2026-10-09T03:05:44.122Z

6. DOGS: Design-Space Sampling for Prompt-Driven Logo Generation

Abstract:Prompt optimization for text-to-image (T2I) generation has been pursued almost entirely as text rewriting, in which a short user brief is expanded into a longer, model-preferred token sequence. We argue that such a language-space formulation is ill-suited to structured visual design tasks such as logo creation, where a one-line brief leaves most design decisions unspecified. These decisions depend on relational priors that a linear sequence cannot encode, and they leave an uncontrolled channel through which protected marks may be reproduced. We therefore recast logo prompting as sampling within a structured design space, and instantiate this idea as DOGS (Design-space prompting with an Originality-aware GFlowNet Sampler). From a large corpus of real-world logos, we mine a typed, graph-structured design grammar whose edges record empirical co-occurrence. A GFlowNet sampler then generates design graphs with probability proportional to a terminal reward that combines recognizability, aesthetics, and corpus-relative originality. Every slot draws only from a closed design-level vocabulary, and any infringement-inducing or harmful token is removed during parsing. The originality reward further penalizes proximity to existing logos, thereby incorporating infringement avoidance into the method by construction. On two open-source renderers and against nine baselines, DOGS produces logos that are more recognizable and aesthetic, substantially more diverse, and far less prone to trademark infringement.

中文摘要

摘要：文本到图像（T2I）生成的提示优化几乎完全以文本重写的方式进行，其中用户提供的简短说明被扩展为更长的、模型偏好的代币序列。我们认为，这种语言空间的形式不适合于结构化的视觉设计任务，例如标志设计，其中一行简短说明无法确定大多数设计决策。这些决策依赖于线性序列无法编码的关系先验，并且它们会留下一个不受控制的通道，从而可能再现受保护的标志。因此，我们将标志提示重新定义为在结构化设计空间中的采样，并将这一思想实现为DOGS（具有原创性意识的GFlowNet采样器的设计空间提示）。我们从一个大型真实世界标志语料库中挖掘出类型化、图结构的设计文法，其边记录了经验共现情况。然后，GFlowNet采样器以终端奖励的概率生成设计图，该奖励结合了可识别性、美学以及相对于语料库的原创性。每个插槽仅从封闭的设计级词汇中选择，并且在解析过程中会移除任何可能引发侵权或有害的代币。原创性奖励进一步惩罚与现有标志的相似性，从而在方法构建中就纳入了侵权规避。在两个开源渲染器上，并与九个基线方法对比，DOGS生成的标志更具可识别性和美感，显著更加多样化，并且远不易侵犯商标权。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10760 (HTTP 429)

Authors: Ganyu Zou, Chen Dai, Nathan Self, Kevin Piper, Ramachandra Rao Seethiraju, Karthik Shyamsunder, Chang-Tien Lu, Naren Ramakrishnan

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.10760.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10760

Published: 2026-10-09T03:05:44.122Z

7. VICO: Visual Environments Co-Evolving for Vision-Language Model Reasoning

Abstract:Reinforcement learning with verifiable rewards (RLVR) has become a standard recipe for post-training vision-language models (VLMs), but it typically assumes a static training environment. As the actor improves, fixed tasks drift out of its learning frontier: many become trivial, others remain unsolvable; and the learning signal collapses. We argue that VLM post-training should evolve the visual environment alongside the actor, not just the actor itself. We propose VICO, a co-evolutionary framework in which an actor and an Environment-as-Rewriter (EnvRewriter) are trained jointly: the EnvRewriter edits verifiable image-side structures, such as scene graphs, chart tables, or protected region masks, and re-renders them to produce label-valid training samples whose difficulty is calibrated to the actor’s current ability through a pass-rate-based reward. This loop continuously realigns task difficulty with actor capability without any additional human annotation. Across nine multimodal benchmarks spanning mathematical reasoning and visually grounded understanding, VICO-8B improves over its base model by up to +5.0% on out-of-domain tasks, surpasses the strongest self-evolution and text-editing co-evolution baselines by +4.3% and +8.4% respectively, and stays comparable to chart-specialized RLVR methods using 16-160 times fewer labeled samples. By shifting from human-labeled supervision to image-editing co-evolution, VICO offers a scalable path beyond static-corpus RLVR for visual reasoning.

中文摘要

摘要：具有可验证奖励的强化学习（RLVR）已经成为后训练视觉-语言模型（VLMs）的标准方法，但它通常假设训练环境是静态的。随着执行者的提升，固定任务会偏离其学习边界：许多任务变得过于简单，其他任务仍然无法解决；学习信号因此崩溃。我们认为，VLM的后训练应与执行者同时演化视觉环境，而不仅仅是执行者本身。我们提出了VICO，一个协同演化框架，其中执行者与环境重写器（EnvRewriter）联合训练：EnvRewriter编辑可验证的图像端结构，例如场景图、图表表格或受保护区域掩码，并重新渲染它们以生成标签有效的训练样本，其难度通过基于通过率的奖励与执行者当前能力相匹配。这个循环不断将任务难度与执行者能力重新对齐，无需任何额外的人类标注。在覆盖数学推理和视觉理解的九个多模态基准测试中，VICO-8B在域外任务上相对于其基础模型提高了最多+5.0%，分别超越了最强的自我演化和文本编辑协同演化基线+4.3%和+8.4%，并使用16至160倍更少的标注样本保持与图表专用RLVR方法的可比性。通过从人工标注监督转向图像编辑协同演化，VICO为视觉推理提供了一条超越静态语料RLVR的可扩展路径。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10782 (HTTP 429)

Authors: Meng Lu, Ligeng Zhu, Olivia Xiao, Yuchen Zhuang, Zihan Wang, Kuncheng Wu, Bangya Liu, Yu Wang, Charles Fleming, Wenqi Shi, Xuan Wang

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.10782.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10782

Published: 2026-10-09T03:05:44.122Z

8. Velocity Scaling in Flow Matching

Abstract:Scaling a learned flow-matching velocity field $v_\theta$ by a gain $\gamma(t)$ was recently shown to greatly improve generation quality. Prior work argued that velocity fields trained with mean-squared error (MSE) systematically underestimate velocity magnitude and that scaling corrects this error. We show that MSE training does not create a velocity-magnitude deficit. We find instead that velocity scaling reduces population time lag: sampled states at model time $t$ resemble training states from an earlier time. Velocity scaling and moving model time back are two ways to address this population time lag. Across architectures and model sizes, measuring population time lag and using it to select a gain greatly improves generation quality, reducing FID from 28.0 to 12.2 (estimated by linear interpolation between FID measurements at neighboring gains) on ImageNet-256 at NFE 25 without guidance.

中文摘要

摘要：将学习到的流匹配速度场 $v_\theta$ 按增益 $\gamma(t)$ 进行缩放，最近被证明可以显著提高生成质量。先前的研究认为，用均方误差（MSE）训练的速度场系统性低估速度幅值，而缩放可以纠正这一误差。我们表明，MSE 训练并不会导致速度幅值不足。相反，我们发现速度缩放能够减少群体时间滞后：在模型时间 $t$ 采样的状态类似于训练中较早时间的状态。速度缩放和将模型时间向后移动是解决这种群体时间滞后的两种方法。在不同架构和模型规模中，测量群体时间滞后并使用其选择增益可以大幅提高生成质量，在 ImageNet-256 上 NFE 为 25 且不使用指导的情况下，将 FID 从 28.0 降低到 12.2（根据相邻增益的 FID 测量线性插值估计）。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10823 (HTTP 429)

Authors: Youssef Saied, François Fleuret

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.10823.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10823

Published: 2026-10-09T03:05:44.122Z

9. Enabling Preference-driven Unlearning in Few-step Distilled Text-to-Image Diffusion Models

Abstract:Text-to-image diffusion models are increasingly distilled into few-step variants and being deployed to enable fast inference. However, their ability to generate harmful or undesired content poses significant safety risks. Data-driven unlearning methods suppress targeted generations by fine-tuning model weights using specialized unlearning objectives. Crucially, these objectives implicitly rely on multi-step denoising dynamics, an assumption that breaks down for few-step distilled (FSD) models, resulting in ineffective forgetting. Furthermore, performing unlearning on the non-distilled base model and subsequently re-distilling it to obtain an unlearned FSD model incurs substantial computational and time overhead, making it impractical in many settings. Hence, we address this limitation with a preference-driven unlearning framework that revisits Direct Preference Optimization (DPO) for diffusion models. We show that standard DPO and its unlearning derivatives, formulated around noise-prediction error, transfer poorly to FSD models due to their altered generation dynamics. To overcome this, we introduce a modified preference optimization formulation explicitly aligned with the few-step generation properties, enabling direct concept removal in FSD models while preserving few-step efficiency and maintaining strong retention of desirable (non-targeted) capabilities. We evaluate our framework primarily on identity and NSFW (nudity) removal tasks and also extend our method to object-level unlearning. Extensive experiments demonstrate consistent and effective forgetting, and strong retention performance, establishing our method as a practical and principled solution for unlearning in FSD models.

中文摘要

摘要：文本到图像扩散模型正越来越多地被蒸馏成少步变体，并被部署以实现快速推理。然而，它们生成有害或不期望内容的能力带来了重大安全风险。数据驱动的消除学习方法通过使用专门的消除学习目标对模型权重进行微调，从而抑制特定生成内容。关键是，这些目标在隐含上依赖于多步去噪动态，而这种假设在少步蒸馏（FSD）模型上不成立，导致消除学习效果不佳。此外，在非蒸馏的基础模型上执行消除学习，然后再进行蒸馏以获得消除后的FSD模型，会产生巨大的计算和时间开销，在许多情况下不可行。因此，我们提出了一种基于偏好驱动的消除学习框架，重新审视了扩散模型的直接偏好优化（DPO）。我们展示了，标准DPO及其基于去噪误差的消除学习衍生方法，由于生成动态的改变，在FSD模型上转移效果较差。为了解决这一问题，我们引入了一种修改后的偏好优化公式，明确与少步生成特性对齐，使得在FSD模型中可以直接移除特定概念，同时保持少步效率并保持对期望（非目标）能力的强保留。我们主要在身份和NSFW（裸体）移除任务上评估了我们的框架，并将方法扩展到对象级消除学习。大量实验表明，我们的方法在遗忘效果和保留性能上具有一致性和有效性，确立了其作为FSD模型中消除学习的实用且有原则的解决方案。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10859 (HTTP 429)

Authors: Gaurav Patel, Jun Fang, Greg Ver Steeg, Qiang Qiu, Sravan Sripada

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.10859.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10859

Published: 2026-10-09T03:05:44.122Z

10. SPLIT-RL: Staged Perception-Language Reasoning Training with Claim-Level Advantages

Abstract:Vision-Language (VL) reasoning requires a model to both extract relevant and accurate information from an image (visual reasoning, VR), and to infer the answer from it (language reasoning, LR). Reinforcement learning with verifiable rewards typically trains both through a single chain-of-thought with a final-answer reward. This gives every CoT token the same sequence-level advantage, failing to distinguish capability specific errors. We propose SPLIT-RL, a staged post-training approach that trains VR and LR in disjoint phases. Because a group’s rollouts differ along one capability at a time, the group-relative advantage isolates it, and each phase is optimized using phase-specific reward. We further introduce Claim-Level Advantage (CLA-GRPO), which decomposes VR-phase rollouts into atomic visual claims and provides a fine-grained advantage at claim level based on visual-type group formation. Although trained in two phases, trained policy is evaluated like GRPO model, with a single CoT call at inference time. Under this protocol, SPLIT-RL improves average accuracy over GRPO by 1.4-6.1 points across Qwen3-VL models from 2B to 30B-A3B and InternVL3.5-8B. Evaluating each capability using an oracle based diagnostic shows that answer-only GRPO leaves perception unchanged, whereas SPLIT-RL improves both VR and LR.

中文摘要

摘要：视觉-语言（VL）推理要求模型既能从图像中提取相关且准确的信息（视觉推理，VR），又能从中推断答案（语言推理，LR）。利用可验证奖励的强化学习通常通过一个单一的思维链进行训练，并以最终答案作为奖励。这会使每个思维链（CoT） token 拥有相同的序列级优势，无法区分特定能力的错误。我们提出了 SPLIT-RL，这是一种分阶段的后训练方法，在不相交的阶段分别训练 VR 和 LR。由于每个组的 rollout 每次只在一个能力上有所不同，组相对优势能够将其隔离，并且每个阶段使用阶段特定的奖励进行优化。我们进一步引入了声明级优势（Claim-Level Advantage, CLA-GRPO），它将 VR 阶段的 rollout 分解为原子视觉声明，并根据视觉类型的组形成提供声明级的细粒度优势。尽管训练分为两个阶段，但训练后的策略在推理时像 GRPO 模型一样进行评估，仅需一次 CoT 调用。在该协议下，SPLIT-RL 在从 2B 到 30B-A3B 的 Qwen3-VL 模型以及 InternVL3.5-8B 上，相对于 GRPO 的平均准确率提升了 1.4-6.1 个百分点。使用基于 oracle 的诊断对每种能力进行评估表明，仅使用答案的 GRPO 不会改变感知，而 SPLIT-RL 则在 VR 和 LR 两方面均有提升。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.10889 (HTTP 429)

Authors: Raja Kumar, Rajat Koner, Ritwick Chaudhry, Zhuowei Li, Nishant Sankaran, Yifan Xing

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.10889.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.10889

Published: 2026-10-09T03:05:44.122Z