大模型推理能力 - arXivDaily 专题

2602.00574 2026-02-03 cs.AI 89%

Learning Modal-Mixed Chain-of-Thought Reasoning with Latent Embeddings

通过潜在嵌入学习模态混合的思考链推理

Yifei Shao, Kun Zhou, Ziming Xu, Mohammad Atif Quamar, Shibo Hao, Zhen Wang, Zhiting Hu, Biwei Huang

机构 * University of California, San Diego（加州大学圣地亚哥分校）

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.AI

AI总结本文提出模态混合CoT方法，通过潜在嵌入结合视觉与文本信息，提升多模态推理性能。

2601.02170 2026-01-06 cs.AI 89%

Streaming Hallucination Detection in Long Chain-of-Thought Reasoning

流式长推理链中的幻觉检测

Haolang Lu, Minghui Pan, Ripeng Li, Guoshun Nan, Jialin Zhuang, Zijie Zhao, Zhongxiang Sun, Kun Wang, Yang Liu

机构 * Beijing University of Posts and Telecommunications（北京邮电大学）； Nanyang Technological University（南洋理工大学）； Southwest Jiaotong University（西南交通大学）； Renmin University of China（中国人民大学）

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.AI

AI总结本研究提出通过跟踪推理状态演变来检测长推理链中的幻觉，提供实时可解释的检测证据。

2511.12001 2025-11-20 cs.CL cs.HC 89%

Critical or Compliant? The Double-Edged Sword of Reasoning in Chain-of-Thought Explanations

Eunkyu Park, Wesley Hanwen Deng, Vasudha Varadarajan, Mingxi Yan, Gunhee Kim, Maarten Sap, Motahhare Eslami

机构 * Seoul National University（首尔国立大学）； Language Technologies Institute, Carnegie Mellon University（语言技术研究所，卡内基梅隆大学）； Human-Computer Interaction Institute, Carnegie Mellon University（人机交互研究所，卡内基梅隆大学）

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.CL

Comments Under review; 16 pages, 15 figures

2510.19851 2025-10-24 cs.CR cs.AI 89%

Can Reasoning Models Obfuscate Reasoning? Stress-Testing Chain-of-Thought Monitorability

Artur Zolkowski, Wen Xing, David Lindner, Florian Tramèr, Erik Jenner

机构 * ML Alignment & Theory Scholars (MATS)（机器学习对齐与理论学者（MATS））； ETH Zurich（苏黎世联邦理工学院）； Google DeepMind（谷歌DeepMind）

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.AI

2510.10103 2025-10-14 cs.CL 89%

Stop When Enough: Adaptive Early-Stopping for Chain-of-Thought Reasoning

Renliang Sun, Wei Cheng, Dawei Li, Haifeng Chen, Wei Wang

机构 * UCLA ； NEC Labs America ； Arizona State University（亚利桑那州立大学）

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.CL

2509.21284 2025-09-26 cs.CL 89%

Bounds of Chain-of-Thought Robustness: Reasoning Steps, Embed Norms, and Beyond

Dingzirui Wang, Xuanliang Zhang, Keyan Xu, Qingfu Zhu, Wanxiang Che, Yang Deng

机构 * Harbin Institute of Technology（哈尔滨工业大学）； Singapore Management University（新加坡国立大学）

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.CL

2506.04205 2025-06-05 cs.LG 89%

EPiC: Towards Lossless Speedup for Reasoning Training through Edge-Preserving CoT Condensation

Jinghan Jia, Hadi Reisizadeh, Chongyu Fan, Nathalie Baracaldo, Mingyi Hong, Sijia Liu

机构 * Michigan State University（密歇根州立大学）； Amazon（亚马逊）； University of Minnesota（明尼苏达大学）； IBM Research（IBM研究院）

专题命中其他推理：reasoning(title,abstract)；CoT(title,abstract)；chain-of-thought(abstract)；分类 cs.LG

2505.03320 2025-06-04 cs.CL 89%

Recall with Reasoning: Chain-of-Thought Distillation for Mamba's Long-Context Memory and Extrapolation

Junyu Ma, Tianqing Fang, Zhisong Zhang, Hongming Zhang, Haitao Mi, Dong Yu

机构 * Tencent AI Lab（腾讯AI实验室）

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.CL

2506.01683 2025-06-03 cs.AI 89%

Reasoning-Based Approach with Chain-of-Thought for Alzheimer's Detection Using Speech and Large Language Models

Chanwoo Park, Anna Seo Gyeong Choi, Sunghye Cho, Chanwoo Kim

机构 * Korea University（韩国大学）； Cornell University（康奈尔大学）； University of Pennsylvania（宾夕法尼亚大学）

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.AI

Comments Accepted to INTERSPEECH 2025

2410.17885 2025-06-02 cs.AI cs.CV 89%

Theorem-Validated Reverse Chain-of-Thought Problem Generation for Geometric Reasoning

Linger Deng, Linghao Zhu, Yuliang Liu, Yu Wang, Qunyi Xie, Jingjing Wu, Gang Zhang, Yingying Zhu, Xiang Bai

机构 * Huazhong University of Science and Technology（华中科技大学）； Department of Computer Vision Technology, Baidu Inc（百度公司计算机视觉技术部门）

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.AI

2505.17650 2025-05-26 cs.AI 89%

Does Chain-of-Thought Reasoning Really Reduce Harmfulness from Jailbreaking?

Chengda Lu, Xiaoyu Fan, Yu Huang, Rongwu Xu, Jijie Li, Wei Xu

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.AI

2411.15382 2025-04-01 cs.CL 89%

On the Impact of Fine-Tuning on Chain-of-Thought Reasoning

Elita Lobo, Chirag Agarwal, Himabindu Lakkaraju

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.CL

Comments This paper is a work in progress with findings based on limited evidence. Please exercise discretion when interpreting the findings

2406.14197 2025-01-27 cs.CL cs.FL 89%

On the Representational Capacity of Neural Language Models with Chain-of-Thought Reasoning

Franz Nowak, Anej Svete, Alexandra Butoi, Ryan Cotterell

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.CL

Comments Published at ACL 2024

2407.06309 2024-07-10 cs.CY cs.AI 89%

Multimodal Chain-of-Thought Reasoning via ChatGPT to Protect Children from Age-Inappropriate Apps

Chuanbo Hu, Bin Liu, Minglei Yin, Yilu Zhou, Xin Li

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.AI

2402.05376 2024-02-09 cs.CL 89%

Zero-Shot Chain-of-Thought Reasoning Guided by Evolutionary Algorithms in Large Language Models

Feihu Jin, Yifan Liu, Ying Tan

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)；分类 cs.CL

Comments 17 pages, 5 figures, 16 tables

2604.25800 2026-08-11 cs.LG cs.CL 版本更新 89%

Barriers to Universal Reasoning With Transformers (And How to Overcome Them)

变压器通用推理的障碍（以及如何克服它们）

Oliver Kraus, Yash Sarrof, Yuekun Yao, Alexander Koller, Michael Hahn

机构 * Saarland University（萨尔兰大学）； Ohio State University（俄亥俄州立大学）

专题命中其他推理：CoT(summary_cn,abstract)；reasoning(title)；chain-of-thought(abstract)；分类 cs.CL、cs.LG

AI总结研究探讨了Transformer在长CoT轨迹上的泛化能力限制，并提出通过动态词汇增长和符号标记实现长度可扩展性。

Comments Accepted at COLM 2026

2511.07368 2026-06-15 cs.LG cs.AI 版本更新 89%

Distributional Biases in Post-Training: A Markovian Analysis of Reasoning Trajectories

后训练中的分布偏差：推理轨迹的马尔可夫分析

Dake Bu, Wei Huang, Andi Han, Atsushi Nitanda, Bo Xue, Qingfu Zhang, Hau-San Wong, Taiji Suzuki

机构 * City University of Hong Kong（香港城市大学）； Center for Advanced Intelligence Project, RIKEN（RIKEN高级智能研究中心）； The Institute of Statistical Mathematics（统计数学研究所）； University of Sydney（悉尼大学）； CFAR and IHPC, Agency for Science, Technology and Research (A*STAR)（A*STAR的CFAR和IHPC）； Nanyang Technological University（南洋理工大学）； The University of Tokyo（东京大学）

专题命中其他推理：CoT(summary_cn,abstract)；reasoning(title,abstract)；分类 cs.AI、cs.LG

AI总结通过马尔可夫链模型分析后训练策略（如RLVR和ORM/PRM）如何强化高概率路径而遗忘稀有但关键的推理步骤，并证明探索策略（如拒绝简单实例和KL正则化）有助于保留稀有CoT。

2607.21433 2026-07-24 cs.CL cs.AI cs.LG 新提交 89%

Token Budget Saturation and Mechanistic Early Detection of Reasoning Non-Convergence in Chain-of-Thought Models

思维链模型中令牌预算饱和与推理不收敛的机制早期检测

Renuka Oladri, Niveda Jawahar, Abdirisak Mohamed

机构 * University of Maryland（马里兰大学）； SAP Labs, LLC（思爱普实验室有限责任公司）

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结研究思维链模型的双峰收敛模式，通过在特定令牌位置的隐藏状态激活上训练线性探针，发现收敛命运可在生成结束前于中间表征中部分编码，为早期退出推理和自适应计算分配开辟道路。

2510.21122 2026-06-08 cs.CV 89%

NoisyGRPO: Incentivizing Multimodal CoT Reasoning via Noise Injection and Bayesian Estimation

NoisyGRPO：通过噪声注入和贝叶斯估计激励多模态Co T推理

Longtian Qiu, Shan Ning, Jiaxuan Sun, Xuming He

机构 * ShanghaiTech University（上海科技大学）； Shanghai Engineering Research Center of Intelligent Vision and Imaging（上海智能视觉与成像工程研究中心）； Lingang Laboratory（临港实验室）

专题命中其他推理：reasoning(title,abstract)；CoT(title,abstract)；chain-of-thought(abstract)

AI总结 NoisyGRPO通过引入可控噪声增强探索并利用贝叶斯框架建模优势估计，提升多模态大语言模型的泛化能力和鲁棒性，尤其在小规模模型上表现突出。

Comments Accepted by Neurips 2025, Project page is available at https://artanic30.github.io/project_pages/NoisyGRPO/

Journal ref Advances in Neural Information Processing Systems 38 (2026) 124239-124267

2603.17199 2026-03-19 cs.LG cs.AI cs.CL 89%

Catching rationalization in the act: detecting motivated reasoning before and after CoT via activation probing

在行动中捕捉理性化：通过激活探测检测推理前后的动机推理

Parsa Mirtaheri, Mikhail Belkin

机构 * UC San Diego（加州大学圣迭戈分校）

专题命中其他推理：reasoning(title,abstract)；CoT(title,abstract)；分类 cs.CL、cs.AI、cs.LG

AI总结研究通过激活探测检测LLM在Cot中的动机推理现象，证明内部表示比CoT监控更可靠，且预生成探测能早期发现动机行为。

2603.02872 2026-03-09 cs.CV 89%

Think-as-You-See: Streaming Chain-of-Thought Reasoning for Large Vision-Language Models

Think-as-You-See: 为大视觉-语言模型设计的流式推理链式思维

Jialiang Zhang, Junlong Tong, Junyan Lin, Hao Wu, Yirong Sun, Yunpu Ma, Xiaoyu Shen

机构 * Institute of Digital Twin, Eastern Institute of Technology（数字孪生研究院，东技术学院）； Ocean University of China（中国海洋大学）； Shanghai Jiao Tong University（上海交通大学）； The Hong Kong Polytechnic University（香港理工大学）； Munich Center for Machine Learning, LMU Munich（慕尼黑机器学习中心，慕尼黑大学）； Ningbo Key Laboratory of Spatial Intelligence and Digital Derivative（宁波空间智能与数字衍生关键实验室）

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)

AI总结 TaYS为大视觉-语言模型设计了流式推理链式思维框架，通过并行化生成、流约束训练和解耦的KV缓存，提升视频理解的效率和响应性。

2510.08985 2025-10-13 cs.IR 89%

Rethinking Reasoning in Document Ranking: Why Chain-of-Thought Falls Short

Xuan Lu, Haohang Huang, Rui Meng, Yaohui Jin, Wenjun Zeng, Xiaoyu Shen

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)

2509.03321 2025-10-10 cs.CV 89%

Empowering Lightweight MLLMs with Reasoning via Long CoT SFT

Linyu Ou, YuYang Yin

机构 * Beijing Institute of Technology（北京理工大学）； Basic Algorithm Center, PCG, Tencent（腾讯基本算法中心）

专题命中其他推理：reasoning(title,abstract)；CoT(title,abstract)；chain-of-thought(abstract)

2410.03810 2025-05-30 cs.LG cs.AI cs.CL 89%

Exploring the Limitations of Mamba in COPY and CoT Reasoning

Ruifeng Ren, Zhicong Li, Yong Liu

机构 * Gaoling School of Artificial Intelligence（光明学校人工智能学院）； Renmin University of China（中国人民大学）

专题命中其他推理：reasoning(title,abstract)；CoT(title,abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Mamba, Chain of Thought

2502.15132 2025-05-23 cs.CL cs.AI cs.LG 89%

CoT-ICL Lab: A Synthetic Framework for Studying Chain-of-Thought Learning from In-Context Demonstrations

Vignesh Kothapalli, Hamed Firooz, Maziar Sanjabi

机构 * New York University（纽约大学）

专题命中其他推理：chain-of-thought(title,abstract)；CoT(title,abstract)；分类 cs.CL、cs.AI、cs.LG

Comments ACL Main 2025

2502.20826 2025-03-03 cs.CV cs.IR 89%

CoTMR: Chain-of-Thought Multi-Scale Reasoning for Training-Free Zero-Shot Composed Image Retrieval

Zelong Sun, Dong Jing, Zhiwu Lu

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；CoT(abstract)

2402.16048 2024-12-13 cs.CL cs.AI cs.LG 89%

How Likely Do LLMs with CoT Mimic Human Reasoning?

Guangsheng Bao, Hongbo Zhang, Cunxiang Wang, Linyi Yang, Yue Zhang

专题命中其他推理：reasoning(title,abstract)；CoT(title)；chain-of-thought(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments COLING 2025 Camera Version (8 pages, 3 figures, 18 tables)

2602.12287 2026-02-16 cs.CL cs.AI eess.AS 88%

Retrieval-Augmented Self-Taught Reasoning Model with Adaptive Chain-of-Thought for ASR Named Entity Correction

带有自适应思维链的检索增强自教推理模型用于ASR命名实体修正

Junjie An, Jingguang Tian, Tianyi Wang, Yu Gao, Xiaofeng Mou, Yi Xu

机构 * AI Research Institute, Midea Group (Shanghai) Co.,Ltd., China（美的集团（上海）研究院）； Zhejiang University, China（浙江大学）

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；分类 cs.CL、cs.AI

AI总结本文提出一种检索增强自教推理模型，用于改进ASR中命名实体修正，通过自适应思维链和音节级编辑距离提升纠错效果。

2508.01309 2026-02-09 cs.CL cs.AI 88%

D-SCoRE: Document-Centric Segmentation and CoT Reasoning with Structured Export for QA-CoT Data Generation

D-SCoRE：基于文档中心分割与推理的结构化导出QA-CoT数据生成

Weibo Zhou, Lingbo Li, Shangsong Liang

专题命中其他推理：CoT(title,abstract)；reasoning(title)；chain-of-thought(abstract)；分类 cs.CL、cs.AI

AI总结 D-SCoRE通过文档中心分割与推理生成高质量QA-CoT数据，提升领域适应微调性能。

2506.22777 2025-07-15 cs.CL cs.AI 88%

Teaching Models to Verbalize Reward Hacking in Chain-of-Thought Reasoning

Miles Turpin, Andy Arditi, Marvin Li, Joe Benton, Julian Michael

机构 * Scale AI ； Harvard College（哈佛学院）； Independent（独立）

专题命中其他推理：reasoning(title,abstract)；chain-of-thought(title,abstract)；分类 cs.CL、cs.AI

Comments Published at ICML 2025 Workshop on Reliable and Responsible Foundation Models

1. 其他推理 5757 篇

Learning Modal-Mixed Chain-of-Thought Reasoning with Latent Embeddings

Streaming Hallucination Detection in Long Chain-of-Thought Reasoning

Critical or Compliant? The Double-Edged Sword of Reasoning in Chain-of-Thought Explanations

Can Reasoning Models Obfuscate Reasoning? Stress-Testing Chain-of-Thought Monitorability

Stop When Enough: Adaptive Early-Stopping for Chain-of-Thought Reasoning

Bounds of Chain-of-Thought Robustness: Reasoning Steps, Embed Norms, and Beyond

EPiC: Towards Lossless Speedup for Reasoning Training through Edge-Preserving CoT Condensation

Recall with Reasoning: Chain-of-Thought Distillation for Mamba's Long-Context Memory and Extrapolation

Reasoning-Based Approach with Chain-of-Thought for Alzheimer's Detection Using Speech and Large Language Models

Theorem-Validated Reverse Chain-of-Thought Problem Generation for Geometric Reasoning

Does Chain-of-Thought Reasoning Really Reduce Harmfulness from Jailbreaking?

On the Impact of Fine-Tuning on Chain-of-Thought Reasoning

On the Representational Capacity of Neural Language Models with Chain-of-Thought Reasoning

Multimodal Chain-of-Thought Reasoning via ChatGPT to Protect Children from Age-Inappropriate Apps

Zero-Shot Chain-of-Thought Reasoning Guided by Evolutionary Algorithms in Large Language Models

Barriers to Universal Reasoning With Transformers (And How to Overcome Them)

Distributional Biases in Post-Training: A Markovian Analysis of Reasoning Trajectories

Token Budget Saturation and Mechanistic Early Detection of Reasoning Non-Convergence in Chain-of-Thought Models

NoisyGRPO: Incentivizing Multimodal CoT Reasoning via Noise Injection and Bayesian Estimation

Catching rationalization in the act: detecting motivated reasoning before and after CoT via activation probing

Think-as-You-See: Streaming Chain-of-Thought Reasoning for Large Vision-Language Models

Rethinking Reasoning in Document Ranking: Why Chain-of-Thought Falls Short

Empowering Lightweight MLLMs with Reasoning via Long CoT SFT

Exploring the Limitations of Mamba in COPY and CoT Reasoning

CoT-ICL Lab: A Synthetic Framework for Studying Chain-of-Thought Learning from In-Context Demonstrations

CoTMR: Chain-of-Thought Multi-Scale Reasoning for Training-Free Zero-Shot Composed Image Retrieval

How Likely Do LLMs with CoT Mimic Human Reasoning?

Retrieval-Augmented Self-Taught Reasoning Model with Adaptive Chain-of-Thought for ASR Named Entity Correction

D-SCoRE: Document-Centric Segmentation and CoT Reasoning with Structured Export for QA-CoT Data Generation

Teaching Models to Verbalize Reward Hacking in Chain-of-Thought Reasoning