ArXiv Domain 2026-10-21

数据来源：ArXiv Domain

LLM Domain Papers

1. Modeling Human Behavior in the Prisoner’s Dilemma with Large Language Models

Abstract:The Prisoner’s Dilemma has long served as a foundational framework for studying human decision-making and cooperative outcomes across social scenarios, yet human behavior consistently deviates from predictions of fully rational economic theory. Previous works have studied human and LLM outcomes in the Prisoner’s Dilemma; however, these studies have lacked at least one key feature: a memory component for LLM agents, accessible open-source models, or comparison against empirical human data. This project investigates whether open-source LLMs, TII Falcon 7B Instruct, Google Gemma 7B, Meta Llama 3.1 8B Instruct, Microsoft Phi Mini MoE Instruct, and Qwen 2.5 7B Instruct, can model human cooperation patterns in repeated Prisoner’s Dilemma games, while examining how an LLM’s knowledge that its opponent will be shuffled affects its decisions. Using a standardized prompting framework, we simulated interactions in two trial types: a fixed-partner setting where both agents retain memory, and a shuffled-partner setting where a memory-enabled agent, informed that its opponent changes across rounds, interacts with a memoryless agent. For each trial, we measured cooperation rates over 100 rounds and 10-round increments and compared them to human data from Montero-Porras et al. (2022). Under both frameworks, all LLMs differed from the human cooperation rate. Open-source LLMs generally over-cooperated relative to humans, particularly in the shuffled framework, where human cooperation averaged 28.5% while several models exceeded 80%. Qwen-2.5-7B-Instruct showed deterministic tendencies, while Phi-mini-MoE-Instruct displayed greater variability and briefly approximated human rates. These findings suggest that LLMs have potential as behavioral simulation agents and provide a benchmark across two interaction environments, although open-source models are not yet reliable substitutes for human participants.

中文摘要

摘要：囚徒困境长期以来一直作为研究人类决策和社会情景中合作结果的基础框架，但人类行为始终偏离完全理性的经济理论预测。以往的研究已经探讨了人类和大语言模型（LLM）在囚徒困境中的表现；然而，这些研究至少缺少一个关键特征：LLM代理的记忆组件、可获取的开源模型，或与实证人类数据的对比。本项目研究开源LLM（包括TII Falcon 7B Instruct、Google Gemma 7B、Meta Llama 3.1 8B Instruct、Microsoft Phi Mini MoE Instruct和Qwen 2.5 7B Instruct）是否能够模拟重复囚徒困境游戏中的人类合作模式，同时考察LLM在知道对手会被随机更换的情况下其决策如何变化。使用标准化提示框架，我们模拟了两种试验类型的交互：固定伙伴设置（双方代理都具有记忆）和随机伙伴设置（拥有记忆的代理在知晓对手会在各轮中变化的情况下与无记忆代理互动）。在每次试验中，我们测量了100轮及10轮增量的合作率，并将其与Montero-Porras等人（2022）的实证人类数据进行了比较。在两种框架下，所有LLM的合作率均与人类不同。相较于人类，开源LLM的合作倾向普遍较高，尤其是在随机框架下，人类的平均合作率为28.5%，而若干模型的合作率超过80%。Qwen-2.5-7B-Instruct表现出确定性倾向，而Phi-mini-MoE-Instruct则显示出更大变化性，并短暂接近人类水平。这些发现表明，LLM具有作为行为模拟代理的潜力，并提供了两种交互环境下的基准，但开源模型尚不能可靠地替代人类参与者。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23860 (HTTP 429)

Authors: Raina Gao, Tristan Raffo

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.23860.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23860

Published: 2026-10-21T02:00:22.539Z

2. Composable Parametric Demographic Injection for Values Survey Simulation

Abstract:Large language models (LLMs) are increasingly used as simulated respondents in computational social science, yet accurately reproducing population-specific heterogeneous response distributions remains challenging. Existing methods such as demographic prompting are sensitive to prompt design and increase inference overhead, while parameter-based methods struggle to reuse demographic information efficiently, limiting generalization to unseen combinations. In this regard, we propose PARADE, a composable PARAmetric DEmographic injection framework that represents each demographic attribute value with a reusable and composable LoRA module. Specifically, PARADE separates task-common survey capability from demographic-specific value shifts via a shared Task LoRA and attribute-level Demographic LoRAs, with composition calibration to reduce parameter interference. At inference time, target population models can be constructed on demand by recombining the relevant modules, including demographic combinations not observed during training. Extensive experiments on the World Values Survey and SocioBench across three backbone models show that PARADE achieves strong population-level distribution alignment while maintaining competitive predictive performance, reducing KL divergence by up to 52.8% relative to the strongest baseline.

中文摘要

摘要：大型语言模型（LLMs）在计算社会科学中越来越多地被用作模拟受访者，但准确再现特定人群的异质性响应分布仍然具有挑战性。现有方法，如人口统计提示，对提示设计敏感并增加推理开销，而基于参数的方法难以高效地重用人口统计信息，从而限制了对未见组合的泛化能力。为此，我们提出了PARADE，一种可组合的参数化人口统计注入框架（composable PARAmetric DEmographic injection framework），它用可重用且可组合的LoRA模块表示每个人口统计属性值。具体而言，PARADE通过共享的任务LoRA和属性级人口统计LoRA，将任务通用的调查能力与人口统计特定的数值偏移分离，并通过组合校准减少参数干扰。在推理时，可以通过重新组合相关模块按需构建目标人群模型，包括训练期间未观察到的人口统计组合。在世界价值观调查（World Values Survey）和SocioBench上的三个主干模型的广泛实验表明，PARADE能够实现强大的人群级分布对齐，同时保持有竞争力的预测性能，相对于最强基线将KL散度最多降低52.8%。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23867 (HTTP 429)

Authors: Xinyuan Cao, Qingyi Pan, Bangde Du, Yujia Zhou, Hexi Wang, Weihang Su, Yiqun Liu, Yueyue Wu, Min Zhang, Qingyao Ai

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.23867.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23867

Published: 2026-10-21T02:00:22.539Z

3. Translating in style: An open-weight alternative to GPT-4 for culturally aligned machine translation

Abstract:Machine translation systems often preserve meaning but systematically flatten linguistic style, a failure that is most acute when translating into non-Western languages. The RASTA framework of Havaldar et al. (2025a) addresses this through retrieval-augmented stylistic alignment, but relies on GPT-4, a closed, proprietary model. We investigate whether an open-weight translation specialist, Hunyuan-MT-7B, can serve as an open-weight alternative. We evaluate three variants of the model, the base model, a LoRA supervised fine-tune, and a GRPO-steered policy. We do this across three style dimensions (formality, politeness, and intimacy) and twelve directed language pairs, measuring style alignment with the correlation-based metric $\mathcal{A}$. We find that fine-tuning does not reliably improve alignment: the three variants are statistically indistinguishable in aggregate. The open backbone does not match RASTA-GPT-4 overall, but the shortfall is concentrated in formality and largely closes for intimacy and Chinese directions, where the open model matches or surpasses GPT-4 on individual pairs. An accessible RASTA pipeline built on an open model is thus already viable for parts of the problem.

中文摘要

摘要：机器翻译系统通常保留意义，但系统性地扁平化语言风格，这种失败在翻译成非西方语言时尤为明显。Havaldar等人（2025a）的RASTA框架通过检索增强风格对齐解决了这个问题，但依赖于GPT-4，一个封闭的专有模型。我们研究了开放权重翻译专家Hunyuan-MT-7B是否可以作为开放权重的替代方案。我们评估了该模型的三种变体：基础模型、LoRA监督的微调和GRPO引导策略。我们在三个风格维度（正式性、礼貌性和亲密性）及十二对定向语言对中进行分析，并用基于相关性指标的$\mathcal{A}$测量风格对齐。我们发现微调并不能可靠地提升对齐：这三种变体在统计上总体上难以区分。开放骨干整体上与RASTA-GPT-4不匹配，但不足主要集中在形式上，且在亲密和中文方向方面大多封闭，开放模型在单个配对上匹配甚至超过GPT-4。基于开放模型构建的可访问RASTA流水线因此已适用于部分问题。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23868 (HTTP 429)

Authors: Thomas Rietman, Hadi Mohammadi, Robert A. Bagheri

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.23868.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23868

Published: 2026-10-21T02:00:22.539Z

4. The Window Is in the Reasoning: When LLM Agents Decide to Follow Prompt Injections

Abstract:LLM agents that read untrusted content can be hijacked by prompt injections. Runtime monitors that try to catch a hijack before the agent acts, and benchmarks that score each task with a single run, share an assumption: whether an agent follows an injection is settled when the injection is read. We test this assumption with branching rollouts that replay an episode exactly up to the step at which the agent reads an injection and then resample the rest of the episode many times, splitting the outcome variance into a part fixed at read time and a part decided afterwards. Across four AgentDojo suites, a Qwen3 agent without thinking is largely decided at read time (0.91 of the outcome variance lies between states); with thinking on, this share drops to 0.62 (95% CI of the drop [-0.38, -0.22]), in the same direction for three attack templates and two success labels. How much is fixed at read time is a property of the model that the same measurement reveals, and thinking lowers the attack success rate of Qwen3 but raises it for two other models. Resampling from reasoning prefixes shows that the decision forms gradually as the agent reasons. Zero-shot chain-of-thought monitors built on cheaper models miss much of what is already decided, whereas stronger monitors, and a lookahead monitor that judges simulated next actions, come close to what the prefix determines. Finally, single-run labels of thinking agents disagree with a rerun up to 35% of the time, so benchmarks need repeated runs per task.

中文摘要

摘要：阅读不受信任内容的LLM代理可能会被提示注入劫持。试图在代理采取行动前捕获劫持的运行时监控，以及使用单次运行评分每个任务的基准测试，共享一个假设：代理是否遵循注入在读取注入时就已决定。我们使用分支回放测试这一假设：回放一个剧集直到代理读取注入的步骤，然后对之后的剧集进行多次重新采样，将结果方差拆分为读取时固定的部分和之后决定的部分。在四个AgentDojo套件中，一个没有思考的Qwen3代理在读取时基本就已决定（结果方差中0.91位于状态之间）；开启思考时，这一比例降至0.62（下降的95%置信区间 [-0.38, -0.22]），对于三种攻击模板和两种成功标签方向相同。读取时固定的程度是模型的属性，相同的测量可以揭示这种特性，思考降低了Qwen3的攻击成功率，但提高了另外两个模型的成功率。从推理前缀重新采样显示，决策是在代理推理过程中逐渐形成的。基于更廉价模型的零样本链式思维监控错过了许多已决定的内容，而更强的监控，以及判断模拟后续动作的前瞻监控，接近前缀所决定的结果。最后，思考型代理的单次运行标签与重新运行存在高达35%的不一致，因此基准测试需要对每个任务进行多次运行。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24048 (HTTP 429)

Authors: Haoxuan Jia, Yang Liu, Bin Chong, Yingguang Yang, Yancheng Chen, Jiayu Liang, Qian Li, Hanning Lu, Kefu Xu, Hao Zheng, Chongyang Zhang, Yulin Huang, Jianshen Zhang, Yongzhi Qi, Yunuo Lin, Hao Peng

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.24048.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24048

Published: 2026-10-21T02:00:22.539Z

5. Discuss More, Comment Less: Boosting Automated Research via Multi-Perspective Collaborative Review

Abstract:As automated research is increasingly adopted for writing papers, automated review has become a critical link between paper generation and quality control. However, existing methods largely rely on single-model feedback or simple review aggregation, failing to explicitly model expert responsibilities or collective deliberation in actual peer reviews, which limits effective cross-examination and reliable decision-making. Therefore, we formulate paper review as a computational modeling problem of peer review mechanisms and propose RoundtableReview, a deliberative multi-agent framework for enhancing automated research. RoundtableReview structures each paper into eight semantic slots and coordinates six academic and industry reviewers with complementary perspectives, retrieval sources, and specialized responsibilities. Through independent assessment, cross-examination, roundtable deliberation, dynamically weighted voting, and chair synthesis, the framework generates evidence-based and traceable feedback. Experiments on two automated research backbones show that the method improves paper quality in both backbones and outperforms baselines in automatic and human evaluations with fewer comments.

中文摘要

摘要：随着自动化研究在论文撰写中的日益应用，自动化审稿已成为论文生成与质量控制之间的关键环节。然而，现有方法大多依赖单一模型反馈或简单的评论聚合，未能明确建模实际同行评审中的专家职责或集体审议，这限制了有效的交叉审查和可靠的决策。因此，我们将论文审稿问题表述为同行评审机制的计算建模问题，并提出了RoundtableReview，这是一种旨在增强自动化研究的多智能体审议框架。RoundtableReview将每篇论文结构化为八个语义槽，并协调六位具有互补视角、检索来源和专业职责的学术与行业审稿人。通过独立评估、交叉审查、圆桌审议、动态加权投票和主席综合，该框架生成基于证据且可追溯的反馈。在两个自动化研究基础模型上的实验表明，该方法在提升两种基础模型论文质量方面表现良好，并在自动和人工评估中优于基线方法，且评论数量更少。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24050 (HTTP 429)

Authors: Junkai Zhou, Jiajia Wu, Zhaoyi Zhang, Shiting Guan, Chuanguang Yang

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.24050.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24050

Published: 2026-10-21T02:00:22.539Z

6. OR-License: Evaluating LLM Agents as Real-World Operations Research Practitioners

Abstract:Can large language model (LLM) agents deliver expert-level operations research (OR) solutions directly from original client materials, without prior expert reformulation? Existing OR benchmarks typically begin with self-contained descriptions or predefined optimization tasks, underrepresenting the translation of original business materials into professionally validated solutions. We introduce OR-License, an end-to-end benchmark of 31 instances across 10 industrial scenarios from paid OR engagements totaling USD 3.68 million. It preserves original requirements, input data, deliverable formats, and information boundaries after de-identification. To study information acquisition, we introduce ORSim, a controlled client simulator that discloses business facts without supplying OR expertise. Hidden evaluation checks feasibility and compares objective quality against original expert solutions accepted by clients before benchmark construction. Deployment-Adjusted Quality (DAQ) averages expert-relative quality across instances, assigning zero to infeasible deliveries. Across 11 frontier LLMs using coding-agent harnesses, several exceed the expert reference in conditional objective quality, but none matches its DAQ; the strongest primary Claude Code configuration achieves 88.39% strict feasibility and a DAQ of 0.87 against the expert reference of 1. Client information acquisition extends beyond clarification into solution development and validation, with some later stages showing higher confirmed issue-resolution proportions. Trajectory analysis identifies formulation and implementation failures, while task-equalized research-chain completion rates range from 24% to 45% across models on evaluable tasks. These findings motivate sustained information-gap resolution, faithful problem formulation, and evidence-linked self-improvement for reliable long-horizon agents.

中文摘要

摘要：大型语言模型（LLM）代理能否直接从原始客户材料提供专家级运筹学（OR）解决方案，而无需事先由专家进行重述？现有的OR基准测试通常以自包含的描述或预定义的优化任务开始，这低估了将原始商业材料转化为专业验证解决方案的过程。我们引入了OR-License，这是一个端到端基准测试，共涵盖来自价值368万美元的付费OR项目的10个行业场景的31个实例。它保留了原始需求、输入数据、交付格式以及去标识化后的信息边界。为了研究信息获取，我们引入了ORSim，这是一个受控的客户模拟器，它披露业务事实但不提供OR专长。隐藏评估检查可行性，并将目标质量与基准建立前由客户接受的原始专家解决方案进行比较。部署调整质量（DAQ）在各实例中平均专家相对质量，对不可行的交付赋值为零。在使用编码代理工具的11个前沿LLM中，多个模型在条件目标质量上超过了专家参考，但没有一个模型的DAQ达到专家参考值；最强的主要Claude Code配置实现了88.39%的严格可行性和0.87的DAQ，而专家参考为1。客户信息获取从澄清扩展到解决方案开发和验证，其中某些后期阶段显示更高的确认问题解决比例。轨迹分析识别了公式化和实施失败，而在可评估任务上的任务均衡研究链完成率在模型间范围为24%至45%。这些发现促使持续的信息差距解决、忠实的问题公式化以及基于证据的自我改进，以实现可靠的长期代理。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24087 (HTTP 429)

Authors: Peisong Wang, Feixue Liu, Yize Meng, Bowen Liu, Ruoyu Chen, Gaoyang Dai, Xing He, Aiqian Li, Fengchi Yuan, Aochuan Chen, Chenyi Zi, Yuhan Li, Hongchuan Zeng, Yimo Yan, Nuo Chen, Jie Fu, Zuo-Jun Max Shen, Yong-Hong KUO, Jia Li

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.24087.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24087

Published: 2026-10-21T02:00:22.539Z

7. Do LMs Know What Variables Do? Interpreting Code Representations via Variable Roles

Abstract:Variable roles describe recurring behaviors such as accumulation, traversal, and retaining the best candidate encountered. We use Sajaniemi’s independently defined taxonomy to study which role distinctions can be recovered from large language models, how sparse autoencoders (SAEs) retain those distinctions, and whether associated directions support control of generation. We introduce static annotation rules for all eleven roles and a corpus of 126,926 natural-code files across five languages. Across five model-SAE systems, residual probes outperform individual SAE features and tested lexical baselines, with modest losses after neutral target renaming. Combining SAE features closes much of the single-feature gap, but probes on SAE reconstructions are less sensitive to role-changing edits reviewed by LLMs. High paired ranking accuracy coexists with frequent false positives at fixed thresholds. During generation, interventions often change role annotations at the cost of correctness; changes accompanied by passing tests show no clear aggregate advantage over random controls. The benchmark separates role-related recoverability, SAE fidelity, and correctness-preserving control through matched prediction, source edits, and generation interventions.

中文摘要

摘要：变量角色描述了诸如累积、遍历以及保留遇到的最佳候选项等重复性行为。我们使用Sajaniemi独立定义的分类法来研究哪些角色区分可以从大型语言模型中恢复，稀疏自编码器（SAEs）如何保留这些区分，以及相关方向是否支持生成控制。我们为所有十一种角色引入了静态标注规则，并提供了跨五种语言的12.6926万个自然代码文件语料库。在五个模型-SAE系统中，残差探测器优于单独的SAE特征和测试的词汇基线，在目标重命名中存在适度损失。结合SAE特征可以弥合大部分单特征差距，但对SAE重建的探测对LLM审查的更改角色的编辑不太敏感。在固定阈值下，高配对排序准确率与频繁的假阳性并存。在生成期间，干预往往会改变角色标注，但会以正确性为代价；伴随通过测试的更改在整体上没有明显优于随机控制。该基准通过匹配预测、源编辑和生成干预，将角色相关的可恢复性、SAE保真度和保持正确性的控制区分开来。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24118 (HTTP 429)

Authors: Célian Vasson, Daniil Ognev, Kentaro Inui, Benjamin Heinzerling

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.24118.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24118

Published: 2026-10-21T02:00:22.539Z

8. Prefix-Tuning for Arbitrary Output Sequences on Pretrained Transformers

Abstract:Prefix-tuning is a parameter-efficient fine-tuning method for large language models (LLMs) that steers model outputs by prepending task-specific prefix representations to the input sequence. Despite its practical success, the theoretical understanding of the expressiveness of prefixes remains incomplete. In particular, the following question remains open: Given a pretrained autoregressive Transformer, for any input-output sequence pair $(\mathtt{x},\mathtt{y})$, does there always exist a prefix $\mathcal{S}$ such that prepending $\mathcal{S}$ to $\mathtt{x}$ induces the model to generate the target output $\mathtt{y}$? In this work, we prove that under the assumption that $\mathtt{y}$ is in the range of the pretrained Transformer, such a prefix $\mathcal{S}$ exists and can be explicitly computed by solving multiple linear systems. We further show that this assumption is necessary, in the sense that such a prefix may not exist when the assumption fails, thereby providing a complete answer to the above question. Our analysis is constructive and the obtained theoretical results hold under fairly general assumptions on model architecture. Extensive experiments on GPT-2 and Qwen3-4B demonstrate that our algorithm produces prefixes with strong control over model outputs.

中文摘要

摘要：前缀调优是一种对大型语言模型（LLMs）参数高效的微调方法，通过在输入序列前添加任务特定的前缀表示来引导模型输出。尽管这种方法在实践中取得了成功，但关于前缀表达能力的理论理解仍不完整。特别是，以下问题仍未解决：给定一个预训练的自回归Transformer，对于任意输入-输出序列对$(\mathtt{x},\mathtt{y})$，是否总存在一个前缀$\mathcal{S}$，使得将$\mathcal{S}$添加到$\mathtt{x}$前能够诱导模型生成目标输出$\mathtt{y}$？在本文工作中，我们证明了在假设$\mathtt{y}$属于预训练Transformer的输出范围的前提下，这样的前缀$\mathcal{S}$是存在的，并且可以通过解多个线性方程组显式计算得到。我们进一步表明，这一假设是必要的，即当该假设不成立时，这样的前缀可能不存在，从而对上述问题提供了完整的答案。我们的分析是构造性的，所获得的理论结果在模型架构的相当一般的假设下成立。在GPT-2和Qwen3-4B上的大量实验表明，我们的算法可以生成对模型输出具有强控制力的前缀。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24123 (HTTP 429)

Authors: Peter Cho-Ho Lam, Ziyi Wang, Zirui Zhou

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.24123.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24123

Published: 2026-10-21T02:00:22.539Z

9. Preservation and loss of algorithmic structure in experimental diffusion of joint-action norms

Abstract:Everyday norms of social interaction - e.g. norms of sharing space, taking turns, or queuing - are essential to large societies, offering structured procedures for common interactions. However, it remains unclear how the underlying algorithms of these implicit joint plans evolve as they diffuse through populations. Here we show that this process of diffusion can select against complex algorithms in the earliest stages. We conducted a large-scale study of experimental diffusion of novel social norms among human participants, leveraging advances in program synthesis to characterize the procedures people developed in independent populations. Inherited procedures became structurally simplified and lost during interactions between participants who inherited different variants of joint-action norms; however, opportunities for interactive repair offset this pressure, preserving complex algorithms despite misalignment. These results demonstrate that interactive reasoning mediates selective pressures on structural complexity during cultural evolution of procedural norms.

中文摘要

摘要：日常社会互动的规范——例如共享空间、轮流或排队的规范——对于大规模社会至关重要，它们为常见互动提供了结构化的程序。然而，目前尚不清楚这些隐含的联合计划的底层算法在群体中传播时是如何演化的。在这里，我们展示了在早期阶段，这种传播过程可能会对复杂算法形成选择压力。我们进行了大规模的人类实验研究，探讨新社会规范在参与者之间的扩散，并利用程序合成的进展来描述人们在独立群体中开发的程序。继承的程序在不同参与者继承不同联合行动规范变体的互动过程中变得结构上简化并丢失；然而，交互性修复的机会抵消了这种压力，使复杂算法得以保留，尽管存在不一致。这些结果表明，交互推理在程序性规范的文化演化过程中调节了对结构复杂性的选择压力。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24150 (HTTP 429)

Authors: Dhara Yu, Bill D. Thompson

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.24150.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24150

Published: 2026-10-21T02:00:22.539Z

10. CoDA: A Human-LM Collaborative Codebook Development Assistant for Domain-Specific Text Annotation

Abstract:We present CoDA, an open-source human-LM collaborative codebook development assistant that helps domain experts develop annotation guidelines for annotating new variables in unstructured text with language models. Existing language model annotation systems primarily focus on downstream classification performance, while assuming the availability of high-quality annotation guidelines, or codebooks. In practice, developing these codebooks is time-consuming and challenging, particularly for concepts that require expert knowledge. CoDA addresses this gap by collaborating with domain experts to iteratively develop codebooks through targeted review of incorrect language model predictions and incorporating expert feedback into the evolving codebook. We demonstrate CoDA on two real-world case studies: extracting health indicators from homelessness case management notes and classifying sentiment of wildlife trafficking posts on social media. Across both domains, CoDA produces codebooks that achieve downstream classification performance comparable to expert codebooks while reducing codebook development time from several weeks to two hours and requiring 5X fewer manually annotated examples. Applying CoDA to generate annotations at scale, we identified 2,410 case notes that contained health discussions or referrals and 10,196 social media posts expressing negative sentiment that may warrant investigation for potential wildlife trafficking. By accelerating codebook development, CoDA helps practitioners use language models to extract domain-specific insights from unstructured text at scale.

中文摘要

摘要：我们提出了 CoDA，一款开源的人类-语言模型协作代码本开发助手，帮助领域专家为使用语言模型对非结构化文本中新变量进行标注制定标注指南。现有的语言模型标注系统主要关注下游分类性能，同时假设高质量的标注指南或代码本可用。在实践中，开发这些代码本既耗时又具有挑战性，尤其是对于需要专家知识的概念。CoDA 通过与领域专家协作，通过针对语言模型错误预测的有针对性审查，并将专家反馈整合到不断演进的代码本中，从而解决了这一问题。我们在两个真实世界案例中展示了 CoDA：从无家可归者管理记录中提取健康指标，以及在社交媒体上对野生动物贩运相关帖子进行情感分类。在这两个领域中，CoDA 生成的代码本在下游分类性能上可与专家代码本相媲美，同时将代码本开发时间从数周缩短到两小时，并且所需的人工标注样本减少了 5 倍。在大规模生成标注时，我们使用 CoDA 识别了 2,410 条包含健康讨论或转诊的病例记录，以及 10,196 条可能涉及潜在野生动物贩运需进行调查的负面情绪社交媒体帖子。通过加速代码本开发，CoDA 帮助从业者利用语言模型大规模从非结构化文本中提取特定领域的洞见。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24157 (HTTP 429)

Authors: Jaspreet Ranjit, Swaminathan Chellappa, Augusto Rivas Costante, Chandan Manjunath, Laura Onasch-Vera, Hailey Winetrobe Nadel, Eric Rice, Swabha Swayamdipta

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2610.24157.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24157

Published: 2026-10-21T02:00:22.539Z

Agent Domain Papers

1. GardenBench: A lightweight, daily evaluation of LLM capabilities

Abstract:We introduce GardenBench, a benchmark with 11 test suites that evaluate the capabilities of Large Language Models (LLMs) across a range of tasks. By systematically prompting LLMs with various lightweight queries, we document day-to-day changes in model performance, including run costs and times, that other benchmarks miss. This allows us to compare the accuracy and features of responses across model providers and versions. We account for daily variations in model accuracy, capturing both model progress and degradation, for instance, when a model is changed post-deployment or when a new model is released. We base our evaluation on general-use tasks and make our results publicly accessible via a live dashboard, diversifying existing approaches to LLM testing. Our main contribution is an innovative, lightweight, and inexpensive benchmark, designed to be consumer-facing and for everyday tasks.

中文摘要

摘要：我们介绍了GardenBench，这是一个包含11个测试套件的基准，用于评估大语言模型（LLMs）在各种任务中的能力。通过系统地使用各种轻量级查询提示LLMs，我们记录了模型性能的日常变化，包括运行成本和时间，这些是其他基准测试所忽略的。这使我们能够比较不同模型提供商和版本之间的响应准确性和功能。我们考虑了模型准确性的每日变化，捕捉模型的进展和退化，例如当模型在部署后发生变化或发布新模型时。我们的评估基于通用任务，并通过实时仪表板公开我们的结果，从而丰富了现有的LLM测试方法。我们主要的贡献是一个创新的、轻量且低成本的基准，旨在面向消费者并用于日常任务。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23866 (HTTP 429)

Authors: Benedict Cummins-Mburu, Mariana Garcia Mejia, Oscar Heath, Sabrina Kreyzerman, Ellie Murray, Arusan Surendiran, Rohan Alexander

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.23866.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23866

Published: 2026-10-21T02:00:46.351Z

2. Doing Less with More: A First-Principles Exploration of the Suitability of Agentic Computing over Alternative Architectural Choices

Abstract:There is growing interest in automating business activities with Agentic Artificial Intelligence (AI) due to latter’s seeming ease of use. Never has it been easier, or costlier, to do less with more. However, little is known about when agents are preferable to established alternatives such as local computation, Representational State Transfer (REST), the Simple Object Access Protocol (SOAP), and the Model Context Protocol (MCP), particularly when development speed, performance, and operational cost are considered. We investigate this question using a controlled mathematical task that compares seven methods on a benchmark of 1,000 arithmetic expressions where semantics of operator precedence has to be preserved. We ran this setup across a native Function Calling baseline (Python library), REST, SOAP and MCP microservice architectures, direct responses from large language models (LLMs) using a Groq cloud API and five locally deployed Ollama models, and agentic systems in which five local LLMs invoke MCP tools. The deterministic methods achieve 100% accuracy with negligible latency. Direct LLM evaluation achieves at most 92.9% accuracy at over 22,000x the latency of a native function call. With MCP tools, the best agent reaches 99.9% accuracy, but incurs nearly 1,000,000x the baseline latency. These results demonstrate that LLMs approaches are poorly suited for exact arithmetic evaluation and motivate an informed architectural decision framework (ADF) for choosing an architecture based on accuracy and throughput requirements. We recognize that an actual interaction with an agent may span both reasoning and linguistic tasks. Hence our recommendation is to explore multiple architectural choices (based on task type) rather than one-size-fits-all solution (LLMs). Code is available at - this https URL

中文摘要

摘要：由于具代理性的人工智能（AI）似乎易于使用，自动化业务活动的兴趣日益增加。以前从未有过如此容易，或如此昂贵地用更多做更少。然而，关于代理何时优于现有替代方案（如本地计算、表征状态传输（REST）、简单对象访问协议（SOAP）以及模型上下文协议（MCP））知之甚少，尤其是在考虑开发速度、性能和运营成本时。我们使用一个受控的数学任务来研究这一问题，该任务将七种方法在1,000个算术表达式的基准测试中进行比较，这些表达式中必须保留操作符优先级的语义。我们在本地函数调用基线（Python库）、REST、SOAP和MCP微服务架构、通过Groq云API直接调用大型语言模型（LLM）的响应以及五个本地部署的Ollama模型，以及具代理系统中五个本地LLM调用MCP工具的情况下运行了此设置。确定性方法以可忽略的延迟实现100%的准确率。直接LLM评估的准确率最多为92.9%，延迟比本地函数调用高出22,000倍以上。使用MCP工具时，最佳代理达到了99.9%的准确率，但延迟几乎是基线的1,000,000倍。这些结果表明，LLM方法不适合精确算术评估，并促使提出一个根据准确率和吞吐量要求选择架构的知情架构决策框架（ADF）。我们认识到，与代理的实际交互可能同时涉及推理和语言任务。因此，我们的建议是根据任务类型探索多种架构选择，而不是采用一刀切的解决方案（LLM）。代码可通过以下URL获取 - this https URL

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23921 (HTTP 429)

Authors: Ritvik Garimella, Biplav Srivastava, Amit Sheth

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.23921.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23921

Published: 2026-10-21T02:00:46.351Z

3. A Controlled LLM Workflow for Evidence-Grounded Diagnosis of Risk Score Differences

Abstract:We propose a controlled Large Language Model (LLM) workflow for explaining risk score differences using similar historical transactions. The workflow combines reference retrieval and contrastive SHapley Additive exPlanations (SHAP) evidence with reference-based LLM reasoning and deterministic validation. Experiments on a test dataset show that the workflow produces more consistent, evidence-grounded and traceable diagnoses than the same LLM without structured workflow guidance.

中文摘要

摘要：我们提出了一种受控的大型语言模型（LLM）工作流程，用于使用相似的历史交易解释风险分数差异。该工作流程结合了参考检索和对比Shapley加性解释（SHAP）证据，以及基于参考的LLM推理和确定性验证。在测试数据集上的实验表明，该工作流程比没有结构化工作流程指导的相同LLM产生更一致、有证据依据且可追溯的诊断。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23962 (HTTP 429)

Authors: Hanqing Dou, Zijia Cao

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.23962.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23962

Published: 2026-10-21T02:00:46.351Z

Abstract:Prominent AI safety discussions emphasize existential threats and catastrophic events, such as AI-enabled nuclear conflict or engineered biological outbreaks. This position paper argues that there is another type of AI risk that arises and accumulates from the widespread use of AI. Drawing on lessons from social media, we argue that safety assessments must also address accumulative harms that develop through repeated use, recurring decisions, and feedback across interconnected deployments. Repeated reassurance, selective exposure, recurring decision biases, and growing personal profiles may gradually narrow perspectives, weaken independent judgment, or enable exploitative influence without a single clearly harmful event. We advance the Trajectory Neglect Thesis: a focus on vivid catastrophic events and identifiable incidents risks overlooking less visible harms that accumulate through the widespread daily use of AI. Social media research motivates this position by showing how reciprocal media effects, social reinforcement, and individual susceptibility shape outcomes beyond individual encounters. We examine how related mechanisms may operate in AI systems, using conversational AI as a concrete setting where generation, semantic inference, and persistent memory may tighten feedback loops. Sycophancy, manipulation, and privacy loss provide a case study of how these risks may interact and accumulate. This comparison identifies mechanisms to investigate rather than assuming that repeated AI use is harmful. We therefore call for evaluation of cumulative patterns of use and deployment, individual and societal outcomes, careful causal inference, and proportionate safeguards that preserve useful assistance and informed user choice.

中文摘要

摘要：突出的人机安全讨论强调存在性威胁和灾难性事件，例如由人工智能引发的核冲突或人为设计的生物疫情。本文立场认为，还有另一类人工智能风险，它源于人工智能的广泛使用并逐渐积累。借鉴社交媒体的经验教训，我们认为安全评估还必须关注通过重复使用、反复决策以及跨互联部署的反馈所产生的累积性伤害。反复的安慰、选择性接触、反复的决策偏差以及不断增长的个人资料，可能逐渐缩小视角、削弱独立判断，或在没有明显有害事件的情况下，使剥削性影响得以实现。我们提出了“轨迹忽视论”：过于关注生动的灾难性事件和可识别的个别事件，可能导致忽视那些通过人工智能广泛日常使用而积累的、较不显眼的危害。社交媒体研究为这一立场提供了动机，显示了互惠媒体效应、社会强化以及个体易感性如何塑造超越单一交互的结果。我们探讨了相关机制在人工智能系统中可能的运作方式，以对话式人工智能为具体场景，分析生成、语义推理和持久记忆如何可能加固反馈回路。阿谀奉承、操控行为和隐私损失提供了一个案例研究，展示了这些风险如何相互作用并积累。这一对照强调了需要调查的机制，而不是假设人工智能的重复使用必然有害。因此，我们呼吁评估使用和部署的累积模式、个体及社会结果、谨慎的因果推断，以及保持有益辅助和用户知情选择的适度防护措施。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24006 (HTTP 429)

Authors: Dawei Li, Bohan Jiang, Huan Liu

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.24006.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24006

Published: 2026-10-21T02:00:46.351Z

5. Beyond Explicit Intent: Discovering and Shaping Artistic Intent through Human-AI Co-Creation

Abstract:Generative AI is becoming an important medium for artistic creation, yet existing interactions often assume that creators can specify intentions in advance and accurately communicate them through text prompts. Artistic intentions, however, may be vague, latent, or recognizable only after seeing concrete visual possibilities. We propose a human-AI co-creation paradigm beyond explicit intent expression. By expanding model exploration, we expose artists to diverse visual possibilities, allowing them to identify previously unarticulated preferences through observation, comparison, and selection. We then integrate artists’ curatorial judgments into reinforcement learning, assigning different learning strengths based on preferences over generation trajectories, so recognized directions are more strongly reinforced and the model gradually adapts to individual aesthetics. This creates a continuous “explore-identify-shape-re-explore” loop in which artistic intent is progressively discovered, externalized, and shaped. Through three user studies, we examine latent-preference discovery, personalized preference alignment, and the externalization of vague, difficult-to-express creative intentions.

中文摘要

摘要：生成式人工智能正成为艺术创作的重要媒介，但现有的交互方式通常假设创作者可以提前明确表达意图，并通过文本提示准确传达。然而，艺术意图可能是模糊的、潜在的，或者只有在看到具体视觉可能性之后才能识别。我们提出了一种超越明确意图表达的人机共创范式。通过扩展模型探索，我们让艺术家接触到多样的视觉可能性，使他们能够通过观察、比较和选择识别先前未表达的偏好。然后，我们将艺术家的策展判断纳入强化学习，根据对生成轨迹的偏好分配不同的学习强度，使被识别的方向得到更强的强化，模型逐渐适应个体审美。这创造了一个连续的“探索—识别—塑造—再探索”循环，在这个循环中，艺术意图得以逐步发现、外化和塑形。通过三项用户研究，我们考察了潜在偏好的发现、个性化偏好的对齐，以及难以表达的模糊创意意图的外化。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24010 (HTTP 429)

Authors: Renye Yan, Jialun Zhong, Jikang Cheng, Yi Gao, Yanzeng Li, Ling Liang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.24010.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24010

Published: 2026-10-21T02:00:46.351Z

6. Unified Multiparadigm Fusion Intelligence for Rare Disease Diagnosis

Abstract:Rare diseases (RDs) affect an estimated 300 million people globally, leaving patients to navigate a protracted diagnostic odyssey constrained by clinical scarcity and fragmented medical knowledge. Artificial intelligence (AI) has progressively advanced RD diagnostics, from symbolic methods and deep learning to large language model (LLM)-based systems. However, each paradigm carries structural limitations: symbolic methods are bounded by knowledge completeness, neural models sacrifice interpretability, and LLM-based systems are vulnerable to evaluation biases like data leakage. In this study, we introduce FusionRare, a unified multiparadigm fusion framework integrating knowledge-driven symbolic reasoning, deep-learning pattern recognition, case-based patient retrieval, and LLM-based heuristic inference via statistical meta-aggregation and knowledge-enhanced agentic reflection. Evaluated on 2,528 strictly isolated clinical cases, FusionRare achieves state-of-the-art RD prioritization, attaining a macro-averaged Recall @1 of 51% for phenotype-only diagnosis and 79% for phenotype-genotype joint analysis. These results represent decisive advantages over the advanced agentic system DeepRare, frontier LLMs including GPT 5.5, and specialized tools like PhenoBrain and LIRICAL under stringent anti-leakage protocols. A randomized, crossover clinician-AI validation demonstrated that FusionRare elevates the true diagnostic rank in 45% of decisions, improves physician confidence in 62%, and reduces time to diagnosis by 21%. These findings indicate that reconciling these orthogonal AI paradigms significantly enhances diagnostic capabilities, bridging the gap between computational benchmarks and actionable clinical reliability. More broadly, this work provides a replicable framework for deploying trustworthy diagnostic AI across complex, high-stakes, and underserved medical domains.

中文摘要

摘要：罕见疾病（RDs）全球估计影响约3亿人，使患者不得不在临床资源稀缺和医学知识分散的限制下，经历漫长的诊断之旅。人工智能（AI）在罕见疾病诊断方面逐步取得进展，从符号方法和深度学习到基于大型语言模型（LLM）的系统。然而，每种范式都有结构性局限：符号方法受限于知识完整性，神经网络模型牺牲了解释性，而基于LLM的系统易受到数据泄露等评估偏差的影响。在本研究中，我们提出了FusionRare，一种统一的多范式融合框架，整合了基于知识的符号推理、深度学习模式识别、基于病例的患者检索以及通过统计元聚合和知识增强的主动反思实现的LLM启发式推理。在对2,528个严格隔离的临床案例进行评估时，FusionRare实现了最先进的罕见疾病优先级排序，在仅表型诊断中Macro-averaged Recall @1达到51%，在表型-基因型联合分析中达到79%。这些结果在严格反泄露机制下，相较于先进的主动智能系统DeepRare、包括GPT 5.5在内的前沿LLM，以及专业工具如PhenoBrain和LIRICAL均显示出明显优势。随机交叉的临床医生-AI验证表明，FusionRare在45%的决策中提高了真实诊断排名，在62%的情况下增强了医生信心，并将诊断时间缩短了21%。这些发现表明，整合这些正交的AI范式显著提升了诊断能力，弥合了计算基准与可操作临床可靠性之间的差距。更广泛地说，这项工作为在复杂、高风险和服务不足的医疗领域部署可信赖的诊断AI提供了可复制的框架。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24024 (HTTP 429)

Authors: Jian Yang, Peiling Liu, Songda Li, Yu Cao, Yuqing Feng, Lan Yu, Jiarong Pan, Jing Ye, Xinwen Huang, Chaochun Zou, Xudong Lu, Huilong Duan, Haomin Li

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.24024.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24024

Published: 2026-10-21T02:00:46.351Z

7. LogicVal:A Logic-Formalized Benchmark for Constraint Satisfaction in Text-to-Image Generation

Abstract:Text-to-Image (T2I) generative models have made significant progress in producing high-fidelity images from natural-language prompts. Alongside this progress, numerous benchmarks have been developed to evaluate capabilities such as attribute binding, numeracy, spatial reasoning, and commonsense reasoning. In this work, we propose a new perspective on T2I evaluation by casting it as a constraint-satisfaction problem. We introduce a logical framework that formally specifies the benchmark domain, including objects, attributes, and relations, alongside the constraints that generated images must satisfy. We evaluated four state-of-the-art T2I models across 9 classes of constraints ranging from a single-object-attribute to multi-object relationships constraints involving logical operators ($\forall, \exists, count, \rightarrow$). Our results reveal a substantial gap between single-object and complex conditional multi-object relationship constraint classes, with the best-performing model achieving 98% accuracy on the former and just 53% on the latter. We assess compositional generalisation, finding that the best-performing model`s accuracy drops from 84% on single-constraint prompts to 20% when satisfying multiple constraints simultaneously. Identifying unsatisfiable constraints is also challenging, with the best model correctly recognizing only 23% of such prompts. Building on this framework, we introduce \textbf{LogicVal}, an evaluation methodology that combines neural visual detection with logical reasoning to assess constraint satisfaction in generated images. On CLEVR, LogicVal achieves ROC-AUC scores ranging from 0.88 to 0.98, indicating strong alignment with human judgments. These findings underscore the value of logic-based evaluation for characterizing T2I model capabilities and limitations.

中文摘要

摘要：文本到图像（T2I）生成模型在从自然语言提示生成高保真图像方面取得了显著进展。与此同时，开发了众多基准测试，用于评估属性绑定、数值能力、空间推理和常识推理等能力。本研究提出T2I评估的新视角，将其定位为约束满足问题。我们引入了一个逻辑框架，正式指定基准领域，包括对象、属性和关系，以及生成图像必须满足的约束条件。我们评估了四个最先进的T2I模型，涵盖9类约束，涵盖从单对象属性到多对象关系约束，涉及逻辑运算符（$\forall， \exists， count， \rightarrow$）。我们的结果显示，单对象与复杂条件多对象关系约束类之间存在显著差距，表现最佳的模型在前者中准确率为98%，后者仅为53%。我们评估了组合泛化，发现表现最佳模型在单一约束提示下准确率从84%降至同时满足多个约束时的20%。识别不可满足约束同样具有挑战性，最佳模型仅正确识别了23%的此类提示。基于该框架，我们引入了 \textbf{LogicVal}，这是一种结合神经视觉检测与逻辑推理的评估方法，用于评估生成图像中的约束满足度。在CLEVR上，LogicVal的ROC-AUC分数范围为0.88至0.98，显示出与人类判断高度契合。这些发现强调了基于逻辑的评估在描述T2I模型能力和局限性方面的价值。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24051 (HTTP 429)

Authors: Savitha Sam Abraham, Marjan Alirezaie, Hossein Rahnama

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.24051.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24051

Published: 2026-10-21T02:00:46.351Z

8. World of SWECraft: Agent Collaboration at Scale via Simulated Software Worlds

Abstract:As teams of AI agents take on larger projects and long-standing open problems, agent collaboration offers another dimension for scaling inference-time compute. To understand what makes such collaboration effective and scalable, we develop Software World, a framework for simulating multi-agent coworking in realistic software ecosystems. It comes with a pipeline to create environments of real, interdependent Python packages of different topics, sizes, and dependency structures, which we call software worlds. In a simulation, agents maintain packages in the world with a high-level goal of improving software efficiency, and they are free to discover tasks to work on and collaborate with others. We evaluate agents’ work in a world collectively by measuring speedup in hidden dependent packages. The framework allows us to study different elements of collaboration by varying communication, ownership, and resource allocation. We run simulations in three different worlds, totaling more than 10,000 hours of agent work across eight models. Overall, performance improvement scales with inference-time compute, with the best model improving downstream efficiency by up to 31%. We show that most agents can find meaningful tasks via collaboration, but some frontier models can prematurely stop looking for improvements and make little progress. We release the simulation framework and traces to support future studies of agent collaboration at this https URL.

中文摘要

摘要：随着AI代理团队承担更大型的项目和长期存在的开放问题，代理协作为扩展推理时计算提供了另一维度。为了理解使这种协作有效且可扩展的因素，我们开发了Software World，这是一个用于在真实软件生态系统中模拟多代理协同工作的框架。它配备了一条创建环境的管道，这些环境由不同主题、规模和依赖结构的真实相互依赖的Python包组成，我们称之为软件世界。在模拟中，代理以提高软件效率为高层目标，维护世界中的包，并可以自由发现要处理的任务并与他人协作。我们通过测量隐藏依赖包中的加速来集体评估代理在世界中的工作。该框架允许我们通过改变通信、所有权和资源分配来研究协作的不同元素。我们在三个不同的世界中运行模拟，总计超过10,000小时的代理工作，涵盖八种模型。总体而言，性能提升随着推理时计算的增加而提升，最佳模型将下游效率提高了最多31%。我们展示了大多数代理可以通过协作找到有意义的任务，但一些前沿模型可能会过早停止寻找改进，从而几乎没有进展。我们发布了模拟框架和轨迹，以支持未来对代理协作的研究，链接为该https URL。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24061 (HTTP 429)

Authors: Shannon Zejiang Shen, Yapei Chang, Qiyao Liang, Benjamin Newman, Jenny Liang, Rulin Shao, Ziqiao Ma, Junjie Oscar Yin, Yung-Sung Chuang, Weijia Shi, Yejin Choi, Luke Zettlemoyer, Pang Wei Koh, Mohit Iyyer, Tongshuang Wu, David Sontag

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.24061.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24061

Published: 2026-10-21T02:00:46.351Z

9. Persistent Memory and Iterative Improvement in an LLM Agent for Chinese-Character Drawing

Abstract:Does persistent memory accumulated from previous attempts help an LLM agent solve new problems, and does memory design affect that benefit? We study programmatic Chinese-character drawing across five agent conditions attempting the same 668 items. Accumulated memory produces modest aggregate gains despite substantial growth in the memory stores: free-form notes improve first-attempt success by 4.6 percentage points, while a design combining free-form memory with reusable code improves it by 2.8 points. The two designs do not differ significantly overall, but the code bank shows a suggestive advantage on later, harder items. A separate post-hoc comparison finds a much larger improvement when the code-bank agent receives privileged, target-specific structural supervision. The agent can therefore exploit highly informative structural guidance, although the privileged per-item information does not establish what could have been learned from experience alone. These results suggest that memory design may affect where accumulated experience becomes useful, even when its average benefit is modest.

中文摘要

摘要：从之前尝试中积累的持久记忆是否有助于大型语言模型代理解决新问题，以及记忆设计是否会影响这种益处？我们研究了在五种代理条件下对相同的668个项目进行程序化汉字绘制的情况。尽管记忆存储显著增长，积累的记忆仍带来适度的整体收益：自由形式笔记将首次尝试成功率提高了4.6个百分点，而将自由形式记忆与可重用代码结合的设计将其提高了2.8个百分点。这两种设计总体上差异不大，但代码库在后期更难的项目上显示出一定优势。另一项事后比较发现，当代码库代理获得特权的、特定目标的结构性监督时，改进幅度更大。因此，代理可以利用高度信息化的结构性指导，尽管这些针对每个项目的特权信息并未说明仅通过经验可以学到的内容。这些结果表明，记忆设计可能影响积累经验的有用性位置，即使其平均效益有限。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24090 (HTTP 429)

Authors: Peilin Wu, Yi-Zhuang You

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.24090.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24090

Published: 2026-10-21T02:00:46.351Z

10. Cache2LoRA: Understanding and Improving Context Internalization

Abstract:Context-to-LoRA methods encode documents as low-rank parameter updates, allowing language models to answer without retaining the source text in their decoding context. How reliably do these updates preserve facts and support their use under different instructions? Controlled diagnosis of existing context-to-LoRA methods separates fact discrimination, target retention, and response adaptation. Competing sources influence answers, while target-only construction and memory-state optimization improve recovery at fixed update size. Recovering facts also does not ensure that answers follow response instructions. We introduce Cache2LoRA, which uses each query to aggregate evidence from a reusable frozen document cache before generating LoRA updates. Trainable encoding operates only on the query and memory-token suffix, and multi-style supervision teaches both evidence preservation and answer expression. Across 17 tasks, Cache2LoRA improves on existing context-to-LoRA methods and achieves macro F1 of 0.663, approaching full-context in-context learning at 0.671. For 32K documents and 256-token outputs at batch size 16, its decoding throughput is 5.9$\times$ that of full-context ICL. At matched memory size, generated LoRA preserves answer accuracy more reliably than the learned-KV variant under long, diverse intervening text and combines more effectively with retained document KV, yielding higher hybrid F1 and EM.

中文摘要

摘要：Context-to-LoRA 方法将文档编码为低秩参数更新，使语言模型在解码时无需保留源文本也能回答问题。这些更新在多大程度上可靠地保留事实并支持在不同指令下使用？对现有 context-to-LoRA 方法的受控诊断将事实辨别、目标保留和响应适应区分开来。竞争来源会影响答案，而仅目标构建和记忆状态优化能在固定更新规模下提升恢复效果。恢复事实也不保证答案会遵循响应指令。我们提出了 Cache2LoRA，它利用每个查询从可重用的冻结文档缓存中聚合证据，然后生成 LoRA 更新。可训练编码仅作用于查询和记忆标记后缀，多风格监督既教授证据保留，也教授答案表达。在 17 项任务中，Cache2LoRA 相较于现有 context-to-LoRA 方法有所提升，宏 F1 值达到 0.663，接近全上下文 In-Context Learning 的 0.671。在 32K 文档和批量大小为 16、输出 256 个标记的情况下，其解码吞吐量是全上下文 ICL 的 5.9 倍。在匹配的内存大小下，生成的 LoRA 在长且多样的插入文本下比 learned-KV 变体更可靠地保持答案准确性，并且与保留的文档 KV 结合更有效，产生更高的混合 F1 和 EM。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24107 (HTTP 429)

Authors: Yi Fang, Gaurav Srivastava, Zhenyu Bi, Sirui Ding, Xuan Wang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.24107.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24107

Published: 2026-10-21T02:00:46.351Z

Evaluation Domain Papers

1. GardenBench: A lightweight, daily evaluation of LLM capabilities

中文摘要

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23866 (HTTP 429)

Authors: Benedict Cummins-Mburu, Mariana Garcia Mejia, Oscar Heath, Sabrina Kreyzerman, Ellie Murray, Arusan Surendiran, Rohan Alexander

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.23866.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23866

Published: 2026-10-21T02:01:06.947Z

2. Doing Less with More: A First-Principles Exploration of the Suitability of Agentic Computing over Alternative Architectural Choices

中文摘要

摘要：由于具代理性的人工智能（AI）似乎易于使用，自动化业务活动的兴趣日益增加。以前从未有过如此容易，或代价如此高昂的情况下用更多做更少。然而，对于何时应优先选择代理方法而非已建立的替代方案（如本地计算、表征状态传输（REST）、简单对象访问协议（SOAP）和模型上下文协议（MCP））知之甚少，特别是在考虑开发速度、性能和运营成本时。我们使用一个受控数学任务来研究这个问题，该任务比较了七种方法在包含1,000个算术表达式的基准测试上的表现，并且必须保持运算符优先级的语义。我们在本地函数调用基线（Python库）、REST、SOAP和MCP微服务架构、大型语言模型（LLM）通过Groq云API的直接响应以及五个本地部署的Ollama模型，以及代理系统（五个本地LLM调用MCP工具）中运行了这一设置。确定性方法以几乎可忽略的延迟实现100%的准确率。直接的LLM评估最多只能达到92.9%的准确率，且延迟是本地函数调用的22,000倍以上。使用MCP工具时，表现最佳的代理准确率达到99.9%，但延迟接近基线的1,000,000倍。这些结果表明，LLM方法不适合精确算术评估，并促使制定一个基于准确性和吞吐量要求的有信息架构决策框架（ADF）。我们认识到，与代理的实际交互可能同时涉及推理和语言任务。因此，我们的建议是根据任务类型探索多种架构选择，而不是采取一刀切的解决方案（LLM）。代码可通过此https URL获得。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23921 (HTTP 429)

Authors: Ritvik Garimella, Biplav Srivastava, Amit Sheth

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.23921.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23921

Published: 2026-10-21T02:01:06.947Z

3. A Controlled LLM Workflow for Evidence-Grounded Diagnosis of Risk Score Differences

中文摘要

摘要：我们提出了一种受控的大型语言模型（LLM）工作流程，用于使用相似的历史交易解释风险分数差异。该工作流程结合了参考检索和对比Shapley加性解释（SHAP）证据，以及基于参考的LLM推理和确定性验证。在测试数据集上的实验表明，该工作流程比没有结构化工作流程指导的相同LLM产生更一致、有证据支撑且可追溯的诊断结果。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23962 (HTTP 429)

Authors: Hanqing Dou, Zijia Cao

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.23962.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23962

Published: 2026-10-21T02:01:06.947Z

中文摘要

摘要：突出的人机安全讨论强调存在性威胁和灾难性事件，例如由人工智能引发的核冲突或人为设计的生物疫情。本文立场认为，还有另一类人工智能风险，它源于人工智能的广泛使用并逐渐积累。借鉴社交媒体的经验教训，我们认为安全评估还必须关注通过重复使用、反复决策以及跨互联部署的反馈所产生的累积性伤害。反复的安慰、选择性接触、反复的决策偏差以及不断增长的个人资料，可能逐渐缩小视角、削弱独立判断，或在没有明显有害事件的情况下，使剥削性影响得以实现。我们提出了“轨迹忽视论”：过于关注生动的灾难性事件和可识别的个别事件，可能导致忽视那些通过人工智能广泛日常使用而积累的、较不显眼的危害。社交媒体研究为这一立场提供了动机，显示了互惠媒体效应、社会强化以及个体易感性如何塑造超越单一交互的结果。我们探讨了相关机制在人工智能系统中可能的运作方式，以对话式人工智能为具体场景，分析生成、语义推理和持久记忆如何可能加固反馈回路。阿谀奉承、操控行为和隐私损失提供了一个案例研究，展示了这些风险如何相互作用及积累。这一对照强调了需要调查的机制，而不是假设人工智能的重复使用必然有害。因此，我们呼吁评估使用和部署的累积模式、个体及社会结果、谨慎的因果推断，以及保持有益辅助和用户知情选择的适度防护措施。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24006 (HTTP 429)

Authors: Dawei Li, Bohan Jiang, Huan Liu

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.24006.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24006

Published: 2026-10-21T02:01:06.947Z

5. Beyond Explicit Intent: Discovering and Shaping Artistic Intent through Human-AI Co-Creation

中文摘要

摘要：生成式人工智能正成为艺术创作的重要媒介，但现有的交互方式通常假设创作者可以提前明确表达意图，并通过文本提示准确传达。然而，艺术意图可能是模糊的、潜在的，或者只有在看到具体视觉可能性之后才能识别。我们提出了一种超越明确意图表达的人机共创范式。通过扩展模型探索，我们让艺术家接触到多样的视觉可能性，使他们能够通过观察、比较和选择识别先前未表达的偏好。然后，我们将艺术家的策展判断纳入强化学习，根据对生成轨迹的偏好分配不同的学习强度，使被识别的方向得到更强的强化，模型逐渐适应个体审美。这创造了一个连续的“探索—识别—塑造—再探索”循环，在这个过程中艺术意图被逐步发现、外化并成形。通过三项用户研究，我们考察了潜在偏好的发现、个性化偏好的对齐，以及模糊、难以表达的创作意图的外化。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24010 (HTTP 429)

Authors: Renye Yan, Jialun Zhong, Jikang Cheng, Yi Gao, Yanzeng Li, Ling Liang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.24010.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24010

Published: 2026-10-21T02:01:06.947Z

6. Unified Multiparadigm Fusion Intelligence for Rare Disease Diagnosis

中文摘要

摘要：罕见疾病（RDs）全球估计影响约3亿人，使患者不得不在临床资源稀缺和医学知识分散的限制下，经历漫长的诊断之旅。人工智能（AI）在罕见疾病诊断方面逐步取得进展，从符号方法和深度学习到基于大型语言模型（LLM）的系统。然而，每种范式都有结构性局限：符号方法受限于知识完整性，神经网络模型牺牲了解释性，而基于LLM的系统易受到数据泄露等评估偏差的影响。在本研究中，我们提出了FusionRare，一种统一的多范式融合框架，整合了基于知识的符号推理、深度学习模式识别、基于病例的患者检索以及通过统计元聚合和知识增强的主动反思实现的LLM启发式推理。在对2,528个严格隔离的临床案例进行评估时，FusionRare在罕见疾病优先级排序方面达到了最先进水平，在仅表型诊断中宏平均Recall @1达到51%，在表型-基因型联合分析中达到79%。这些结果在严格反泄露机制下，相较于先进的主动智能系统DeepRare、包括GPT 5.5在内的前沿LLM，以及专业工具如PhenoBrain和LIRICAL均显示出明显优势。随机交叉的临床医生-AI验证表明，FusionRare在45%的决策中提高了真实诊断排名，在62%的情况下增强了医生信心，并将诊断时间缩短了21%。这些发现表明，整合这些正交的AI范式显著提升了诊断能力，弥合了计算基准与可操作临床可靠性之间的差距。更广泛地说，这项工作为在复杂、高风险和服务不足的医疗领域部署可信赖的诊断AI提供了可复制的框架。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24024 (HTTP 429)

Authors: Jian Yang, Peiling Liu, Songda Li, Yu Cao, Yuqing Feng, Lan Yu, Jiarong Pan, Jing Ye, Xinwen Huang, Chaochun Zou, Xudong Lu, Huilong Duan, Haomin Li

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.24024.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24024

Published: 2026-10-21T02:01:06.947Z

7. LogicVal:A Logic-Formalized Benchmark for Constraint Satisfaction in Text-to-Image Generation

中文摘要

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24051 (HTTP 429)

Authors: Savitha Sam Abraham, Marjan Alirezaie, Hossein Rahnama

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.24051.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24051

Published: 2026-10-21T02:01:06.947Z

8. World of SWECraft: Agent Collaboration at Scale via Simulated Software Worlds

中文摘要

摘要：随着AI代理团队承担更大型的项目和长期存在的开放问题，代理协作为扩展推理时计算提供了另一维度。为了理解使这种协作有效且可扩展的因素，我们开发了Software World，这是一个用于在真实软件生态系统中模拟多代理协同工作的框架。它配备了一条创建环境的管道，这些环境由不同主题、规模和依赖结构的真实相互依赖的Python包组成，我们称之为软件世界。在模拟中，代理以提高软件效率为高层目标，维护世界中的包，并可以自由发现要处理的任务并与他人协作。我们通过测量隐藏依赖包中的加速来集体评估代理在世界中的工作。该框架允许我们通过改变通信、所有权和资源分配来研究协作的不同元素。我们在三个不同的世界中运行模拟，总计超过10,000小时的代理工作，涵盖八种模型。总体而言，性能提升随着推理时计算量的增加而扩大，最佳模型将下游效率提高了最多31%。我们展示了大多数代理能够通过协作找到有意义的任务，但一些前沿模型可能会过早停止寻找改进，从而进展甚微。我们发布了模拟框架和轨迹，以支持未来对代理协作的研究，链接为该https URL。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24061 (HTTP 429)

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.24061.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24061

Published: 2026-10-21T02:01:06.947Z

9. Persistent Memory and Iterative Improvement in an LLM Agent for Chinese-Character Drawing

中文摘要

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24090 (HTTP 429)

Authors: Peilin Wu, Yi-Zhuang You

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.24090.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24090

Published: 2026-10-21T02:01:06.947Z

10. Cache2LoRA: Understanding and Improving Context Internalization

中文摘要

摘要：Context-to-LoRA 方法将文档编码为低秩参数更新，使语言模型在解码时无需保留源文本也能回答问题。这些更新在多大程度上可靠地保留事实并支持在不同指令下使用？对现有 context-to-LoRA 方法的受控诊断将事实辨别、目标保留和响应适应区分开来。竞争来源会影响答案，而仅目标构建和记忆状态优化能在固定更新规模下提升恢复效果。恢复事实也不保证答案会遵循响应指令。我们提出了 Cache2LoRA，它利用每个查询从可重用的冻结文档缓存中聚合证据，然后生成 LoRA 更新。可训练编码仅作用于查询和记忆标记后缀，多风格监督既教授证据保留，也教授答案表达。在 17 项任务中，Cache2LoRA 相较于现有 context-to-LoRA 方法有所提升，宏 F1 达到 0.663，接近完整上下文的上下文学习的 0.671。在批量大小为 16、文档数为 32K、输出为 256 标记的情况下，其解码吞吐量是完整上下文 ICL 的 5.9 倍。在匹配的内存大小下，生成的 LoRA 在长且多样的干扰文本下比学习的 KV 变体更可靠地保持答案准确性，并且与保留的文档 KV 更有效结合，产生更高的混合 F1 和 EM。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.24107 (HTTP 429)

Authors: Yi Fang, Gaurav Srivastava, Zhenyu Bi, Sirui Ding, Xuan Wang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2610.24107.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.24107

Published: 2026-10-21T02:01:06.947Z

VLM Domain Papers

1. I$^2$-Med: Interpretable Medical Inference through Visual-Guided Dynamic Logits Calibration

Abstract:Medical vision language models have demonstrated strong potential for medical image understanding and reasoning. However, their interpretability and reliability remain limited because medical vision language tasks require precise visual grounding and clinically coherent explanations. This is particularly challenging due to complex imaging patterns and scarce expert annotations. These challenges reduce the effectiveness of conventional reasoning methods and lead to visual hallucination, defined as the generation of text that is not grounded in the given visual input, which further limits their applicability in clinical practice. To mitigate these challenges, we propose an interpretable medical inference (I$^2$-Med) framework that enhances visual faithfulness during inference without additional training to improve reasoning and explanation faithfulness in medical vision language models. During text generation, I$^2$-Med uses a visual guidance module to check whether each candidate word is supported by the medical image and consistent with the current reasoning. Based on this guidance, the model’s output scores are adjusted to favor visually grounded words, without changing the model parameters. Extensive experiments on the OmniMedVQA benchmark, covering eight medical imaging modalities and five clinical question types, demonstrate that our proposed I$^2$-Med framework generates more clinically meaningful and visually faithful reasoning chains and reduces visual hallucination under standard medical VQA evaluation metrics. The source code is publicly available at \url{this https URL}.

中文摘要

摘要：医学视觉语言模型已展现出在医学图像理解和推理方面的强大潜力。然而，由于医学视觉语言任务需要精确的视觉基础和临床连贯的解释，其可解释性和可靠性仍然有限。由于复杂的成像模式和稀缺的专家注释，这一点尤为具有挑战性。这些挑战降低了传统推理方法的有效性，并导致视觉幻觉，即生成的文本不基于给定的视觉输入，进一步限制了其在临床实践中的适用性。为缓解这些挑战，我们提出了一种可解释医学推理（I$^2$-Med）框架，在无需额外培训的情况下提升推理中的视觉忠实度，以提升医学视觉语言模型中的推理和解释忠实度。在文本生成过程中，I$^2$-Med 使用视觉指导模块检查每个候选词是否得到医学图像的支持且是否与当前推理一致。基于该指导，模型的输出分数被调整为有利于视觉基础的词汇，且不改变模型参数。在OmniMedVQA基准测试上进行了涵盖八种医学影像模态和五种临床问题类型的广泛实验，表明我们提出的I$^2$-Med框架能生成更具临床意义和视觉忠实的推理链，并在标准医学VQA评估指标下减少视觉幻觉。源代码公开于\url{this https URL}。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23858 (HTTP 429)

Authors: Md Sajid, Shreeyut Maheshwari, Akshat Mishra, Ritik Mishra, Mohammad Tanveer

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.23858.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23858

Published: 2026-10-21T02:01:28.599Z

2. Latent Multimodal Memory Reasoning

Abstract:Long-term personalized agents are required to reason over distributed history across modalities and evolving dialogue states. While existing flat cross-modal methods compress visual evidence prematurely and struggle with temporally stale observations, they neglect the necessity to unify multimodal and cross-session evidence. Crucially, raw multimodal memories are inherently entangled and heterogeneous, making fine-grained visual grounding and cross-modal reasoning computationally prohibitive. To this end, we present latent multimodal memory reasoning, i.e., LMMR, a novel neuro-symbolic framework that shifts long-term multimodal memory organization into a continuous latent space. Specifically, (i) instead of converting images to text and relying on static retrieval, LMMR maps both textual hidden states and visual patch features into a shared Sparse Autoencoder concept space, establishing sparse activations as common semantic addresses to dynamically construct a turn-level multimodal memory graph. (ii) A graph encoder then aggregates cross-modal and temporal dependencies across the task-specific latent subgraph, injecting the graph readout as a continuous residual rollout into the generation backbone while retaining native visual tokens to preserve fine-grained details. Extensive experiments on nine Mem-Gallery tasks and twelve MemEye settings demonstrate that LMMR significantly outperforms state-of-the-art baselines in capturing complex multimodal dependencies while enabling faithful and grounded responses.

中文摘要

摘要：长期个性化智能体需要在跨模态和不断发展的对话状态中推理分布式历史。现有的平面跨模态方法会过早压缩视觉证据，并且在处理时间上过时的观察时表现不佳，同时它们忽略了统一多模态和跨会话证据的必要性。关键是，原始多模态记忆本质上是纠缠和异质的，这使得细粒度视觉定位和跨模态推理在计算上变得不可行。为此，我们提出了潜在多模态记忆推理（Latent Multimodal Memory Reasoning，LMMR），这是一种新型神经符号框架，将长期多模态记忆的组织转移到连续潜在空间中。具体来说，(i) LMMR并非将图像转换为文本并依赖静态检索，而是将文本隐状态和视觉片段特征映射到共享的稀疏自编码器概念空间中，通过稀疏激活建立通用语义地址，以动态构建回合级多模态记忆图。 (ii) 图编码器随后对任务特定潜在子图中的跨模态和时间依赖关系进行聚合，在生成骨干中注入图读出作为连续残差展开，同时保留原生视觉标记以保留细粒度细节。在九个Mem-Gallery任务和十二个MemEye设置上的大量实验表明，LMMR在捕捉复杂多模态依赖关系方面显著优于最先进的基线方法，同时能够生成忠实且有依据的响应。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23890 (HTTP 429)

Authors: Cai Ke, Xinghao Chen, Yuanchen Bei, Yuni Lai, Keyu Chen, Ruizhi Qiao, Siyu An, Junnan Dong, Ruifeng Xu, Xing Sun

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.23890.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23890

Published: 2026-10-21T02:01:28.599Z

3. EMMA: Event-Driven Multimodal Memory System with Multi-Agent Retrieval

Abstract:Long video understanding requires reasoning over hours of content that far exceed the context window of multimodal large language models (MLLMs). Existing approaches compress video into fixed-length clips, yet events are fragmented and unrelated content is merged, yielding noisy retrieval. To overcome this challenge, we propose EMMA, an Event-driven Multimodal Memory system with Multi-Agent retrieval, which organizes memory around semantic events in three stages: (1) Event segmentation: a dual-stream detector adaptively locates event boundaries by fusing visual semantic drift with spoken transitions, rather than fixed windows. (2) Memory construction: built on the detected events, a three-level hierarchy of raw video, clip-level memory, and event-level memory stores content at progressively higher semantic abstraction. (3) Memory retrieval: a multi-agent framework collaboratively queries this hierarchy through query decomposition, tiered search, evidence validation, and answer generation, narrowing evidence from coarse events to precise clips. Extensive experiments on three challenging benchmarks demonstrate substantial improvements, showing the effectiveness of our event-aligned memory system. Code, prompt templates, and reproduction scripts are available at this https URL.

中文摘要

摘要：长视频理解需要在数小时的内容上进行推理，这远远超出了多模态大语言模型（MLLM）的上下文窗口。现有方法通常将视频压缩为固定长度的片段，但事件被切碎且无关内容被合并，从而导致检索噪声。为了解决这一挑战，我们提出了EMMA，一种基于事件驱动的多模态记忆系统与多智能体检索机制，该系统通过三个阶段围绕语义事件组织记忆：(1) 事件分割：双流检测器通过融合视觉语义漂移与口语转变自适应地定位事件边界，而不是使用固定窗口。(2) 记忆构建：基于检测到的事件，建立由原始视频、片段级记忆和事件级记忆组成的三层层次结构，以逐步更高的语义抽象存储内容。(3) 记忆检索：多智能体框架通过查询分解、分层搜索、证据验证和答案生成协作查询该层次结构，从粗略事件逐步定位到精确片段。基于三个具有挑战性的基准进行的大量实验显示了显著的提升，证明了我们事件对齐记忆系统的有效性。代码、提示模板和复现脚本可在此https链接获取。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23892 (HTTP 429)

Authors: Ruochen Zhao, Jiazheng Kang, Gezi Li, Dongying Lin, Ting Bai

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.23892.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23892

Published: 2026-10-21T02:01:28.599Z

4. Path-Conditioned Attention Mobility: From Depth to Execution Hubs in SegFormer

Abstract:Transformer compression often couples two distinct quantities: retained attention capacity and the number of sequential depths at which that capacity is executed. We study whether useful attention functions in a hierarchical vision transformer are locally mobile across depth. Using SegFormer-B5 for human segmentation, we find that some task-relevant attention heads can be relocated to nearby depths, but only when their computational context is respected. Head-only relocation is insufficient, while joint adaptation of the relocated head and surrounding representation boundary improves functional preservation; replacing donor-specific normalization and spatial-reduction preprocessing causes large degradation. We use this path-conditioned mobility to reorganize retained attention from depth into width. Starting from a fixed parent with 120 retained Stage-3 heads across 40 attention execution depths, we consolidate the same 120-head budget into 20 execution hubs. Under a layout-matched three-seed recovery control, K20 reaches 0.960686 +/- 0.000059 mIoU versus 0.960400 +/- 0.000074 for the corresponding State120 control. The paired point difference is +0.000285, while a paired image-bootstrap confidence interval includes zero; we therefore interpret the result as no measurable quality penalty rather than statistical superiority. For deployment, FOLDED_VEC preserves donor-specific paths while vectorizing their execution. Relative to the same 120-head State120 parent, it reduces end-to-end NVIDIA L4 latency by 5.28%, 3.15%, and 3.58% at batch sizes 1, 4, and 8. These results suggest that attention redundancy includes not only removable capacity but also execution-depth redundancy, provided that the computational path defining a relocated function is preserved.

中文摘要

摘要：Transformer 压缩通常会耦合两个不同的量：保留的注意力容量以及该容量被执行的序列深度数量。我们研究了在分层视觉 Transformer 中，有用的注意力功能是否能够在深度上局部移动。使用 SegFormer-B5 进行人体分割时，我们发现一些与任务相关的注意力头可以重新定位到相邻的深度，但前提是尊重它们的计算上下文。仅重新定位注意力头是不够的，而同时调整重新定位的注意力头及其周围表示边界则有助于功能保持；替换捐赠端特定的归一化和空间降维预处理会导致性能大幅下降。我们利用这种基于路径的移动性，将保留的注意力从深度重新组织到宽度。以固定父模型为起点，其中在 40 个注意力执行深度上保留了 Stage-3 的 120 个注意力头，我们将相同的 120 头预算整合到 20 个执行中心。在布局匹配的三次随机种子恢复控制下，K20 达到 0.960686 +/- 0.000059 的 mIoU，而对应的 State120 控制为 0.960400 +/- 0.000074。配对点差值为 +0.000285，同时配对图像自助法置信区间包括零；因此我们将结果解释为没有可测量的质量损失，而非统计上的优越性。对于部署而言，FOLDED_VEC 在向量化执行的同时保留了捐赠端的特定路径。相较于相同的 120 头 State120 父模型，它在批量大小为 1、4 和 8 时，端到端 NVIDIA L4 延迟分别减少了 5.28%、3.15% 和 3.58%。这些结果表明，注意力冗余不仅包括可移除的容量，还包括执行深度冗余，前提是保留定义重新定位功能的计算路径。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23893 (HTTP 429)

Authors: Aleksandr Druzhinin

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.23893.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23893

Published: 2026-10-21T02:01:28.599Z

5. GeoForward: Geometry-Guided Cross-View Alignment for Feedforward 3D Gaussian Splatting

Abstract:Feedforward 3D reconstruction has largely followed an {appearance-driven paradigm}: multi-view images are encoded into visual representations, cross-view correspondence is implicitly inferred from learned features, and 3D primitives are subsequently predicted from these representations. However, calibrated cameras and sparse metric measurements already provide explicit constraints on where the same physical point should appear across views. We propose a novel paradigm, \textbf{Geometry-Conditioned Feedforward Reconstruction}, in which metric geometry establishes cross-view correspondence before the 3D representation is formed, while learned visual features refine rather than discover this correspondence. Based on this paradigm, we introduce \textbf{GeoForward}. Sparse LiDAR points are unprojected into 3D and analytically projected across calibrated cameras to provide metric correspondence anchors. A bounded continuous local search then refines these anchors using visual features, and the resulting geometry-conditioned representation is fused with global multi-view context from VGGT. The shared representation drives separate metric-depth and Gaussian-attribute prediction, establishing a direct path from explicit geometry to both Gaussian centers and spatial attributes. A sparse 3D refinement module subsequently improves local attribute compatibility while preserving the geometrically determined centers. For temporal reconstruction, endpoint Gaussians are composed using vehicle masks and motion priors. On nuScenes, GeoForward achieves the lowest cross-camera depth-consistency error and Chamfer distance among the compared methods while maintaining competitive reconstruction quality. These results demonstrate that explicitly conditioning correspondence formation on metric geometry provides an effective alternative to purely appearance-driven feedforward 3D reconstruction.

中文摘要

摘要：前馈式3D重建在很大程度上遵循了{基于外观的范式}：多视角图像被编码为视觉表示，跨视角对应关系从学习到的特征中隐式推断，并随后从这些表示中预测3D基元。然而，校准的相机和稀疏的度量测量已经提供了显式约束，说明同一物理点在不同视角中应出现的位置。我们提出了一种新范式，\textbf{几何条件前馈重建}（Geometry-Conditioned Feedforward Reconstruction），其中度量几何在形成3D表示之前建立跨视角对应关系，而学习到的视觉特征用于细化而非发现这种对应关系。基于这一范式，我们引入了\textbf{GeoForward}。稀疏的LiDAR点被反投影到3D空间，并在校准相机中解析投影，以提供度量对应锚点。随后，使用视觉特征通过有界连续局部搜索来优化这些锚点，得到的几何条件表示与来自VGGT的全局多视角上下文融合。共享表示驱动单独的度量深度和高斯属性预测，从而建立了从显式几何到高斯中心和空间属性的直接路径。随后，一个稀疏3D细化模块提高了局部属性的一致性，同时保持几何确定的中心。对于时间序列重建，端点高斯通过车辆掩码和运动先验进行组合。在nuScenes上，GeoForward在比较方法中实现了最低的跨相机深度一致性误差和Chamfer距离，同时保持了具有竞争力的重建质量。这些结果表明，明确将对应关系形成条件化于度量几何，为纯外观驱动的前馈3D重建提供了一种有效的替代方法。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23895 (HTTP 429)

Authors: Yiming Shu, Peng Sun, Lang Zhang, Mingyue Feng, Yaonong Wang, Pei Liu, Chen Sun, Jun Ma

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.23895.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23895

Published: 2026-10-21T02:01:28.599Z

6. RhinoNet: Unsupervised Learning for Phenoscaping and Generation of High Dimensional Point Clouds

Abstract:We present RhinoNet, a simplicial complex-based point cloud neural network, that creates a geometry-preserving embedding of high-dimensional point clouds. RhinoNet takes collections of datasets, also called {\em data cohorts}, and processes each dataset (point cloud) within the cohort as a simplicial complex. These simplicial complexes have datapoints as vertices and edges learned by Vietoris-Rips filtration. Using simplicial wavelets and MLP layers, RhinoNet computes a vector representation of each dataset in a latent space produced by an autoencoder. The latent space is optimized such that an optimal transport distance between pairs of point clouds is preserved between the latent representations. We apply RhinoNet to single-cell datasets collected on patients—-a common source of such high-dimensional data. RhinoNet creates a patient level representation, known in biological contexts as a phenoscape. We establish that RhinoNet generates organized phenoscapes, whereas other methods such as static maps do not. Next we show that we can generate entire point clouds by sampling and decoding this phenoscape. We show two modes of generation via 1) latent flow matching and 2) geodesic interpolation in latent space, followed by a decoding back into the point cloud space. RhinoNet achieves state-of-the-art performance on single-cell and spatial data cohorts relative to tested baselines. Overall, RhinoNet lifts the concepts of representation learning and generative modeling to the point cloud or population level and defines a neural model for phenoscaping.

中文摘要

摘要：我们提出了RhinoNet，一种基于单形复形的点云神经网络，它能够对高维点云进行几何保留的嵌入。RhinoNet接收数据集合，也称为数据队列，并将队列中的每个数据集（点云）作为单形复形进行处理。这些单形复形以数据点为顶点，通过Vietoris-Rips滤波学习边。利用单形小波和多层感知器（MLP）层，RhinoNet计算每个数据集在由自编码器生成的潜在空间中的向量表示。潜在空间被优化，使得点云对之间的最优传输距离在潜在表示中得以保留。我们将RhinoNet应用于收集自患者的单细胞数据集——这类高维数据的常见来源。RhinoNet生成患者级别的表示，在生物学背景下称为表型景（phenoscape）。我们证明了RhinoNet可以生成有组织的表型景，而其他方法如静态映射则不能。接着我们展示了可以通过对该表型景进行采样和解码生成整个点云。我们展示了两种生成模式：1）潜在流匹配，2）潜在空间中的测地线插值，随后解码回点云空间。与测试的基线方法相比，RhinoNet在单细胞和空间数据队列中实现了最先进的性能。总体而言，RhinoNet将表示学习和生成建模的概念提升到了点云或群体水平，并定义了用于表型景生成的神经模型。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23898 (HTTP 429)

Authors: Siddharth Viswanath, Hiren Madhu, Eddie Cavallin, Benjamin Jones, Alistair Wilkinson, Christopher Tape, Michael Perlmutter, Rex Ying, Ian Adelstein, Smita Krishnaswamy

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.23898.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23898

Published: 2026-10-21T02:01:28.599Z

7. EAGLE: Entropy-Aware Geometric Learning Enhancer for Partially Polygon-Supervised Object Localization

Abstract:Object detection is a critical application of computer vision, yet models trained uniformly on bounding boxes lack the geometric precision to be fully effective in inference settings, especially at high IoU thresholds. One common approach is the introduction of polygon-supervised localization, where a model is trained explicitly for localization but with access to polygon masks during training. Among such methods, partial polygon supervision is viable due to the higher annotation costs of developing fully supervised training datasets. However, this introduces a further challenge: in object classes of relatively low shape complexity, the addition of training masks and subsequent segmentation loss may obscure box localization, often worsening model performance despite access to deeper geometric information. To address the practical limitations of both full and partial polygon supervision, we introduce a mathematical framework for quantifying per-class geometric complexity by combining information theory and statistical shape analysis. With this framework, we develop Entropy-Aware Geometric Learning Enhancer (EAGLE), a lightweight, post-training model head that learns geometric information for the localization model, improving accuracy via edge refinement and localization rescoring using entropy-informed updates. Empirical analysis across MS COCO and Pascal SBD demonstrates the entropy framework meaningfully predicts shape reconstruction error and accuracy differences among mask-derived and standard box predictions, while the implemented EAGLE framework achieves consistent improvement on already SOTA detection frameworks (+0.9 mAP50-95 points on MS COCO) at minimal observed inference overhead.

中文摘要

摘要：目标检测是计算机视觉中的一个关键应用，但在统一使用边界框进行训练的模型在推理场景中缺乏几何精度，尤其是在高 IoU 阈值下，难以达到最佳效果。一种常见的方法是引入多边形监督定位，即在训练过程中明确进行定位训练，同时使用多边形掩码。在这些方法中，由于完全监督训练数据集的标注成本较高，部分多边形监督是可行的。然而，这也带来了进一步的挑战：对于形状复杂度相对较低的目标类别，添加训练掩码及后续的分割损失可能会干扰边界框定位，即便模型获得了更深入的几何信息，也往往会导致性能下降。为了应对全多边形与部分多边形监督的实际局限性，我们提出了一个数学框架，通过结合信息论和统计形状分析，量化每个类别的几何复杂度。基于这一框架，我们开发了熵感知几何学习增强器（Entropy-Aware Geometric Learning Enhancer，EAGLE），它是一个轻量级的后训练模型头，能够为定位模型学习几何信息，并通过边缘精炼和基于熵的信息更新进行定位重评分，从而提升精度。在 MS COCO 和 Pascal SBD 上的实证分析表明，熵框架能够有意义地预测形状重构误差以及掩码生成预测与标准边界框预测间的准确性差异，而实现的 EAGLE 框架在已有的最先进检测框架上实现了持续改进（MS COCO 上 mAP50-95 提升 0.9 点），且在推理开销方面几乎没有增加。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23900 (HTTP 429)

Authors: Naveen Stewart

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.23900.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23900

Published: 2026-10-21T02:01:28.599Z

8. Rethinking weakly supervised localization on chest radiographs: position nulls, protocol dependence, and PACE

Abstract:Weakly supervised localization on chest radiographs is reported as a single overlap accuracy, and those numbers are not comparable across papers. We measure what one contains. Against position-null controls that keep a readout’s own box and move only its position, class activation mapping and its gradient-weighted variant beat the null on two to four of seven annotated findings on ChestX-ray14 and two of eight on CheXlocalize. The null is beatable: three board-certified radiologists clear it on four to seven. Three rarely stated choices each move a result by more than most reported method effects: the results regime, the extraction threshold, and the mask-to-box rule, whose cost is predicted by how fragmented the thresholded mask is. Predicted box area alone reverses a paired comparison. These measurements produced PACE, for \emph{population anomaly and class evidence}, which multiplies the finding identity from the label-trained head by the spatial evidence from a frozen radiograph backbone that never sees a pathology label. It is the case study that made the protocol dependence visible, and we report its costs beside its gain: in an exploratory matched-area analysis it improves strict-overlap accuracy on both corpora and beats a label-derived second map, but it is the least faithful of the three readouts to the classifier’s causal map, its advantage depends on the box rule, and, this benchmark releasing no localization validation set, every design choice behind it was made on the boxes it is scored on. Its one clean transfer beats one baseline and loses to the other.

中文摘要

摘要：胸片上的弱监督定位通常以单一的重叠精度进行报告，而这些数值在不同论文之间不可比。我们测量了其包含的内容。针对只保持读出自身框并仅移动其位置的位置空白控制，类别激活映射及其梯度加权变体在ChestX-ray14的七个标注发现中击败了空白控制的两个到四个，在CheXlocalize的八个中击败了两个。空白控制是可以被超越的：三名具有执照的放射科医师在四到七个中超越它。三个很少提及的选择各自使结果的变化大于大多数已报道方法的效果：结果方案、提取阈值和掩码到框的规则，其成本可通过阈值化掩码的碎片化程度预测。仅预测的框面积就可以反转配对比较。这些测量产生了PACE，即“人群异常与类别证据”，它将由标签训练的头部给出的发现身份乘以从从未见过病理标签的冻结胸片骨干得到的空间证据。正是这个案例研究使协议依赖性显现出来，我们报告了其代价及收益：在探索性的匹配区域分析中，它在两个语料库上的严格重叠精度都有提高，并且击败了一个由标签导出的第二个映射，但它在三种读出中对分类器因果映射的忠实度最低，其优势依赖于框规则，并且，由于该基准没有发布定位验证集，其背后的每一个设计选择都是基于其评分的框作出的。它的一次干净迁移击败了一个基线而输给了另一个。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23902 (HTTP 429)

Authors: Shixin Qiu

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.23902.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23902

Published: 2026-10-21T02:01:28.599Z

9. AI-Experiment: an augmented reality art project co-created with artificial intelligence

Abstract:We describe the creation of a physical artwork that was performed in synergy between an artist, a computer scientist and generative AI tools. A latent diffusion model was finetuned on photographs of the artwork of the specific artist and the generative model then generated synthetic artworks in a similar style. From the generated artworks a subset was selected to be painted again with acrylic on canvas by the original artist. In addition to the physical artifact an augmented reality component was added which brings the artwork to life with a set of generative video models. Together with this paper we release part of the artwork (training data), the code that was deployed to generate the artwork as well as the augmented reality experience for arts, research and education under this https URL

中文摘要

摘要：我们描述了一个物理艺术作品的创作过程，该过程由艺术家、计算机科学家和生成式人工智能工具协同完成。将一个潜在扩散模型在该艺术家的艺术作品照片上进行微调，然后该生成模型生成了风格相似的合成艺术作品。从生成的艺术作品中，选择了一个子集，由原艺术家再次用丙烯在画布上绘制。除了物理作品之外，还增加了增强现实组件，通过一组生成式视频模型让艺术作品栩栩如生。随本文一起，我们发布了部分艺术作品（训练数据）、用于生成艺术作品的代码，以及用于艺术、研究和教育的增强现实体验，网址如下：https URL

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23903 (HTTP 429)

Authors: Bernhard Egger, Hans Furer

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.23903.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23903

Published: 2026-10-21T02:01:28.599Z

10. LVMH: A Meta-Optimization Harness for Long-Video Agent Memory

Abstract:Long-video agents must map hours or days of multimodal evidence into a bounded context while respecting token and latency budgets, making memory design a context-management challenge. Existing memory systems largely rely on hand-designed policies, leaving open how to jointly optimize memory design for performance and efficiency. We propose Long Video Meta-memory Harness (LVMH), an automated multi-agent framework that formulates memory-policy development as multi-objective code search around a fixed answer model. To guard against self-certification and development-set overfitting, LVMH separates implementation from independent review and enforces protected observation boundaries. Persistent meta-memory records past proposals, traces, costs, and failures to guide subsequent policy revisions. Through this harness, we discover Self-Evolving Evidence Memory (SEEM), a “text-first, vision-second” policy that answers from compact structured event memory and evidence specialists, consulting localized frames through a selective visual gate. The policy evolved on EgoLifeQA generalizes to unseen benchmarks, outperforming Meta-Harness on both Ego-R1 Bench and Video-MME (L) with GPT-5 mini. Across the three benchmarks, SEEM with GPT-5 achieves 65.7% average accuracy, surpassing the previous state of the art by 7.7 percentage points with approximately 32x lower average evaluation token use and 91x shorter average response time.

中文摘要

摘要：长视频智能体必须将数小时或数天的多模态证据映射到一个有限的上下文中，同时遵守令牌和延迟预算，这使得记忆设计成为一个上下文管理的挑战。现有的记忆系统在很大程度上依赖于手工设计的策略，仍然没有回答如何联合优化记忆设计以提高性能和效率的问题。我们提出了长视频元记忆框架（Long Video Meta-memory Harness, LVMH），这是一个自动化的多智能体框架，将记忆策略开发形式化为围绕固定答案模型的多目标代码搜索。为了防止自我认证和开发集过拟合，LVMH将实现与独立评审分离，并强制执行受保护的观察边界。持久化的元记忆记录过去的提案、轨迹、成本和失败，以指导随后的策略修订。通过该框架，我们发现了自我进化证据记忆（Self-Evolving Evidence Memory, SEEM），这是一种“文本优先、视觉次之”的策略，从紧凑的结构化事件记忆和证据专家中生成答案，并通过选择性视觉闸门咨询局部帧。该策略在EgoLifeQA上进化，且可推广到未见的基准测试，在Ego-R1 Bench和Video-MME（L）上均优于Meta-Harness，使用GPT-5 mini。在三个基准测试中，SEEM使用GPT-5实现了65.7%的平均准确率，比之前的最优水平高出7.7个百分点，同时平均评估令牌使用量降低约32倍，平均响应时间缩短91倍。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2610.23904 (HTTP 429)

Authors: Liwei Che, Benlin Liu, Dai Li, Xiangyan Li, Zijian Wu, Changjiang Li, Minghao Guo, Chengzhi Mao, Vladimir Pavlovic, Ranjay Krishna

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2610.23904.pdf

CoolPaper URL: https://papers.cool/arxiv/2610.23904

Published: 2026-10-21T02:01:28.599Z

ArXiv Domain 2026-10-21

LLM Domain Papers

1. Modeling Human Behavior in the Prisoner’s Dilemma with Large Language Models

2. Composable Parametric Demographic Injection for Values Survey Simulation

3. Translating in style: An open-weight alternative to GPT-4 for culturally aligned machine translation

4. The Window Is in the Reasoning: When LLM Agents Decide to Follow Prompt Injections

5. Discuss More, Comment Less: Boosting Automated Research via Multi-Perspective Collaborative Review

6. OR-License: Evaluating LLM Agents as Real-World Operations Research Practitioners

7. Do LMs Know What Variables Do? Interpreting Code Representations via Variable Roles

8. Prefix-Tuning for Arbitrary Output Sequences on Pretrained Transformers

9. Preservation and loss of algorithmic structure in experimental diffusion of joint-action norms

10. CoDA: A Human-LM Collaborative Codebook Development Assistant for Domain-Specific Text Annotation

Agent Domain Papers

1. GardenBench: A lightweight, daily evaluation of LLM capabilities

2. Doing Less with More: A First-Principles Exploration of the Suitability of Agentic Computing over Alternative Architectural Choices

3. A Controlled LLM Workflow for Evidence-Grounded Diagnosis of Risk Score Differences

4. What Social Media Teaches Us About Accumulative AI Risks

5. Beyond Explicit Intent: Discovering and Shaping Artistic Intent through Human-AI Co-Creation

6. Unified Multiparadigm Fusion Intelligence for Rare Disease Diagnosis

7. LogicVal:A Logic-Formalized Benchmark for Constraint Satisfaction in Text-to-Image Generation

8. World of SWECraft: Agent Collaboration at Scale via Simulated Software Worlds

9. Persistent Memory and Iterative Improvement in an LLM Agent for Chinese-Character Drawing

10. Cache2LoRA: Understanding and Improving Context Internalization

Evaluation Domain Papers

1. GardenBench: A lightweight, daily evaluation of LLM capabilities

2. Doing Less with More: A First-Principles Exploration of the Suitability of Agentic Computing over Alternative Architectural Choices

3. A Controlled LLM Workflow for Evidence-Grounded Diagnosis of Risk Score Differences

4. What Social Media Teaches Us About Accumulative AI Risks

5. Beyond Explicit Intent: Discovering and Shaping Artistic Intent through Human-AI Co-Creation

6. Unified Multiparadigm Fusion Intelligence for Rare Disease Diagnosis

7. LogicVal:A Logic-Formalized Benchmark for Constraint Satisfaction in Text-to-Image Generation

8. World of SWECraft: Agent Collaboration at Scale via Simulated Software Worlds

9. Persistent Memory and Iterative Improvement in an LLM Agent for Chinese-Character Drawing

10. Cache2LoRA: Understanding and Improving Context Internalization

VLM Domain Papers

1. I$^2$-Med: Interpretable Medical Inference through Visual-Guided Dynamic Logits Calibration

2. Latent Multimodal Memory Reasoning

3. EMMA: Event-Driven Multimodal Memory System with Multi-Agent Retrieval

4. Path-Conditioned Attention Mobility: From Depth to Execution Hubs in SegFormer

5. GeoForward: Geometry-Guided Cross-View Alignment for Feedforward 3D Gaussian Splatting

6. RhinoNet: Unsupervised Learning for Phenoscaping and Generation of High Dimensional Point Clouds

7. EAGLE: Entropy-Aware Geometric Learning Enhancer for Partially Polygon-Supervised Object Localization

8. Rethinking weakly supervised localization on chest radiographs: position nulls, protocol dependence, and PACE

9. AI-Experiment: an augmented reality art project co-created with artificial intelligence

10. LVMH: A Meta-Optimization Harness for Long-Video Agent Memory