大模型推理能力

The Capacity for Moral Self-Correction in Large Language Models

Deep Ganguli, Amanda Askell, Nicholas Schiefer, Thomas I. Liao, Kamilė Lukošiūtė, Anna Chen, Anna Goldie, Azalia Mirhoseini, Catherine Olsson, Danny Hernandez, Dawn Drain, Dustin Li, Eli Tran-Johnson, Ethan Perez, Jackson Kernion, Jamie Kerr, Jared Mueller, Joshua Landau, Kamal Ndousse, Karina Nguyen, Liane Lovitt, Michael Sellitto, Nelson Elhage, Noemi Mercado, Nova DasSarma, Oliver Rausch, Robert Lasenby, Robin Larson, Sam Ringer, Sandipan Kundu, Saurav Kadavath, Scott Johnston, Shauna Kravec, Sheer El Showk, Tamera Lanham, Timothy Telleen-Lawton, Tom Henighan, Tristan Hume, Yuntao Bai, Zac Hatfield-Dodds, Ben Mann, Dario Amodei, Nicholas Joseph, Sam McCandlish, Tom Brown, Christopher Olah, Jack Clark, Samuel R. Bowman, Jared Kaplan

专题命中 复杂问题求解 ：self-correction(title,abstract)；分类 cs.CL

Achieving and Understanding Out-of-Distribution Generalization in Systematic Reasoning in Small-Scale Transformers

Andrew J. Nam, Mustafa Abdool, Trevor Maxfield, James L. McClelland

专题命中 复杂问题求解 ：reasoning(title,abstract)；分类 cs.LG

Disentangling Reasoning Capabilities from Language Models with Compositional Reasoning Transformers

Wanjun Zhong, Tingting Ma, Jiahai Wang, Jian Yin, Tiejun Zhao, Chin-Yew Lin, Nan Duan

专题命中 复杂问题求解 ：reasoning(title,abstract)；分类 cs.CL

Comments 12 pages

Maieutic Prompting: Logically Consistent Reasoning with Recursive Explanations

Jaehun Jung, Lianhui Qin, Sean Welleck, Faeze Brahman, Chandra Bhagavatula, Ronan Le Bras, Yejin Choi

专题命中 复杂问题求解 ：reasoning(title,abstract)；分类 cs.CL

Comments EMNLP 2022

Relational Reasoning via Set Transformers: Provable Efficiency and Applications to MARL

Fengzhuo Zhang, Boyi Liu, Kaixin Wang, Vincent Y. F. Tan, Zhuoran Yang, Zhaoran Wang

专题命中 复杂问题求解 ：reasoning(title,abstract)；分类 cs.LG

Reflection of Thought: Inversely Eliciting Numerical Reasoning in Language Models via Solving Linear Systems

Fan Zhou, Haoyu Dong, Qian Liu, Zhoujun Cheng, Shi Han, Dongmei Zhang

专题命中 复杂问题求解 ：reasoning(title,abstract)；分类 cs.CL

ChemAlgebra: Algebraic Reasoning on Chemical Reactions

Andrea Valenti, Davide Bacciu, Antonio Vergari

专题命中 复杂问题求解 ：reasoning(title,abstract)；分类 cs.LG

MMKGR: Multi-hop Multi-modal Knowledge Graph Reasoning

Shangfei Zheng, Weiqing Wang, Jianfeng Qu, Hongzhi Yin, Wei Chen, Lei Zhao

专题命中 复杂问题求解 ：reasoning(title,abstract)；分类 cs.AI

A Benchmark for Compositional Visual Reasoning

Aimen Zerroug, Mohit Vaishnav, Julien Colin, Sebastian Musslick, Thomas Serre

专题命中 复杂问题求解 ：reasoning(title,abstract)；分类 cs.AI

Legal Judgment Prediction (LJP) Amid the Advent of Autonomous AI Legal Reasoning

Lance Eliot

专题命中 复杂问题求解 ：reasoning(title,abstract)；分类 cs.AI

Comments 39 pages, 13 figures

DMRM: A Dual-channel Multi-hop Reasoning Model for Visual Dialog

Feilong Chen, Fandong Meng, Jiaming Xu, Peng Li, Bo Xu, Jie Zhou

专题命中 复杂问题求解 ：reasoning(title,abstract)；分类 cs.CL

Comments Accepted at AAAI 2020

Towards Abstraction in ASP with an Application on Reasoning about Agent Policies

Zeynep G. Saribatur, Thomas Eiter

专题命中 复杂问题求解 ：reasoning(title,abstract)；分类 cs.AI

Comments Proceedings of the 11th Workshop on Answer Set Programming and Other Computing Paradigms 2018

Structured Factored Inference: A Framework for Automated Reasoning in Probabilistic Programming Languages

Avi Pfeffer, Brian Ruttenberg, William Kretschmer

专题命中 复杂问题求解 ：reasoning(title,abstract)；分类 cs.AI

Implementing Evidential Reasoning in Expert Systems

John Yen

专题命中 复杂问题求解 ：reasoning(title,abstract)；分类 cs.AI

Comments Appears in Proceedings of the Third Conference on Uncertainty in Artificial Intelligence (UAI1987)

ViSS-R1: Self-Supervised Reinforcement Video Reasoning

Bo Fang, Yuxin Song, Qiangqiang Wu, Haoyuan Sun, Wenhao Wu, Antoni B. Chan

机构 * City University of Hong Kong（香港城市大学） ； Baidu Inc.（百度公司） ； Tsinghua University（清华大学） ； The University of Sydney（悉尼大学） 纠错

专题命中 复杂问题求解 ：reasoning(title,abstract)

Comments Our paper was initially titled "Video-SSR1: Self-Supervised Reinforcement Video Reasoning." Upon noticing its close resemblance to the title of a recently released paper, we have decided to rename our work as "ViSS-R1."

Hierarchical Vision-Language Planning for Multi-Step Humanoid Manipulation

André Schakkal, Ben Zandonati, Zhutian Yang, Navid Azizan

机构 * Massachusetts Institute of Technology（麻省理工学院） ； École Polytechnique Fédérale de Lausanne（瑞士联邦理工学院） 纠错

专题命中 复杂问题求解 ：planning(title,abstract)

Comments Accepted at the RSS 2025 Workshop on Robot Planning in the Era of Foundation Models

Linguistically Driven Graph Capsule Network for Visual Question Reasoning

Qingxing Cao, Xiaodan Liang, Keze Wang, Liang Lin

专题命中 复杂问题求解 ：reasoning(title,abstract)

Comments Submitted to TPAMI 2020. We have achieved an end-to-end interpretable structural reasoning for general images without the requirement of layout annotations

The Illusion of Multi-Agent Advantage

多智能体优势的错觉

Prathyusha Jwalapuram, Hehai Lin, Chuyuan Li, Fangkai Jiao, Sudong Wang, Yifei Ming, Zixuan Ke, Chengwei Qin, Giuseppe Carenini, Shafiq Joty

机构 * Salesforce Research（Salesforce研究院） ； HKUST (Guangzhou)（香港科技大学（广州）） ； University of British Columbia（不列颠哥伦比亚大学） ； Nanyang Technological University（南洋理工大学） 纠错

专题命中 复杂问题求解 ：reasoning(abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.CL、cs.AI

AI总结 通过系统评估，发现自动生成的多智能体系统在性能和成本效率上均不如单智能体基线（如思维链自一致性），揭示了现有评估框架的缺陷和架构膨胀问题。

Hybrid Training for Vision-Language-Action Models

视觉-语言-动作模型的混合训练

Pietro Mazzaglia, Cansu Sancaktar, Markus Peschl, Daniel Dijkman

机构 * Qualcomm AI Research（高通AI研究） 纠错

专题命中 复杂问题求解 ：CoT(abstract,abstract_cn)；chain-of-thought(abstract)；分类 cs.AI、cs.LG

AI总结 本文提出混合训练框架，旨在使视觉-语言-动作模型在推理时能够根据需要生成思考过程或直接预测动作，从而在保持性能提升的同时提高推理效率。

Comments Published as a conference paper at ICLR 2026

Fast-Slow Thinking RM: Efficient Integration of Scalar and Generative Reward Models

快速-缓慢思考RM：标量与生成奖励模型的有效整合

Jiayun Wu, Peixu Hou, Shan Qu, Peng Zhang, Ning Gu, Tun Lu

机构 * Fudan University（复旦大学） ； Meituan（美团） 纠错

专题命中 复杂问题求解 ：reasoning(abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.CL、cs.LG

AI总结 本文提出F/S-RM，结合快速和缓慢思考机制，提升奖励模型性能并降低计算成本。

When Silence Is Golden: Can LLMs Learn to Abstain in Temporal QA and Beyond?

沉默即是黄金：大语言模型能否在时间问答中学会退避？

Xinyu Zhou, Chang Jin, Carsten Eickhoff, Zhijiang Guo, Seyed Ali Bahrainian

机构 * HKUST (GZ)（香港科技大学） ； Tongji University（同济大学） ； University of Tübingen（图宾根大学） ； HKUST（香港科技大学） 纠错

专题命中 复杂问题求解 ：reasoning(abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.CL、cs.AI

AI总结 本文探讨了如何通过强化学习训练大语言模型在时间问答中学会退避，发现RL在推理和可靠性方面优于SFT。

Comments Accepted to ICLR2026

Iterative Refinement Improves Compositional Image Generation

迭代细化改进组合图像生成

Shantanu Jaiswal, Mihir Prabhudesai, Nikash Bhardwaj, Zheyang Qin, Amir Zadeh, Chuan Li, Katerina Fragkiadaki, Deepak Pathak

机构 * Carnegie Mellon University（卡内基梅隆大学） ； Lambda AI 纠错

专题命中 复杂问题求解 ：reasoning(abstract)；chain-of-thought(abstract)；self-correction(abstract)；分类 cs.AI、cs.LG

AI总结 本文提出了一种迭代细化策略，通过视觉-语言模型反馈提升组合图像生成质量，实验显示在多个基准上均优于并行采样方法。

Comments Project webpage: https://iterative-img-gen.github.io/

Rethinking Supervised Fine-Tuning: Emphasizing Key Answer Tokens for Improved LLM Accuracy

重新思考监督微调：强调关键答案标记以提高LLM准确性

Xiaofeng Shi, Qian Kou, Yuduo Li, Hua Zhou

机构 * Beijing Academy of Artificial Intelligence (BAAI)（北京人工智能研究院） ； Beijing Jiaotong University (BJTU)（北京交通大学） 纠错

专题命中 复杂问题求解 ：reasoning(abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.CL、cs.AI

AI总结 SFTKey通过两阶段训练方案，强调关键答案标记以提高LLM在复杂推理任务中的准确性。

Auto-Prompting with Retrieval Guidance for Frame Detection in Logistics

基于检索引导的自动提示法用于物流帧检测

Do Minh Duc, Quan Xuan Truong, Nguyen Tat Dat, Nguyen Van Vinh

机构 * Faculty of Information Technology, VNU University of Engineering and Technology（信息技术学院，越南工程大学） 纠错

专题命中 复杂问题求解 ：reasoning(abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.CL、cs.AI

AI总结 本文提出基于检索引导的自动提示法，通过优化提示提升物流文本帧检测的准确性和效率。

Towards Efficient LLM-aware Heterogeneous Graph Learning

面向高效LLM-aware异构图学习

Wenda Li, Tongya Zheng, Shunyu Liu, Yu Wang, Kaixuan Chen, Hanyang Yuan, Bingde Hu, Zujie Ren, Mingli Song, Gang Chen

机构 * State Key Laboratory of Blockchain and Data Security, Zhejiang University（区块链与数据安全国家重点实验室，浙江大学） ； Zhejiang Lab（浙江实验室） ； College of Computer Science and Technology, Zhejiang University（浙江大学计算机科学与技术学院） ； Zhejiang Provincial Engineering Research Center for Real-Time SmartTech in Urban Security Governance, School of Computer and Computing Science, Hangzhou City University（浙江省实时智能城市安全治理工程技术研究中心，杭州城市大学计算机与计算科学学院） ； Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security, Hangzhou, Zhejiang, China（杭州高新技术区（滨江）区块链与数据安全研究院，杭州，浙江，中国） ； Nanyang Technological University（南洋理工大学） ； Bangsun Technology（邦sun科技） 纠错

专题命中 复杂问题求解 ：reasoning(abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.CL、cs.AI

AI总结 本文提出ELLA框架，通过LLM-aware关系分词器和层次关系图变压器，高效解决异构图中复杂关系语义建模和任务间语义间隙问题，实现性能与效率的提升。

The Idola Tribus of AI: Large Language Models tend to perceive order where none exists

Shin-nosuke Ishikawa, Masato Todo, Taiki Ogihara, Hirotsugu Ohba

机构 * Graduate School of Artificial Intelligence and Science, Rikkyo University（人工智能与科学研究生院，立命馆大学） ； Strategic Digital Business Unit, Mamezou Co., Ltd.（战略数字商务部，麦迈株式会社） 纠错

专题命中 复杂问题求解 ：reasoning(abstract)；chain-of-thought(abstract)；logical reasoning(abstract)；分类 cs.CL、cs.AI

Comments 14 pages, 3 figures, accepted to Findings of EMNLP 2025

POT: Inducing Overthinking in LLMs via Black-Box Iterative Optimization

Xinyu Li, Tianjin Huang, Ronghui Mu, Xiaowei Huang, Gaojie Jin

专题命中 复杂问题求解 ：reasoning(abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.AI、cs.LG

When Chain of Thought is Necessary, Language Models Struggle to Evade Monitors

Scott Emmons, Erik Jenner, David K. Elson, Rif A. Saurous, Senthooran Rajamanoharan, Heng Chen, Irhum Shafkat, Rohin Shah

机构 * Google（谷歌） 纠错

专题命中 复杂问题求解 ：reasoning(abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.CL、cs.AI

Towards Analyzing and Understanding the Limitations of VAPO: A Theoretical Perspective

Jintian Shao, Yiming Cheng, Hongyi Huang, Beiwen Zhang, Zhiyu Wu, You Shan, Mingkai Zheng

专题命中 复杂问题求解 ：reasoning(abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.CL、cs.LG

Comments We are withdrawing this submission as the underlying experiment is currently incomplete. We require additional time to gather more data and supplement the existing findings to ensure a comprehensive and robust presentation. We intend to resubmit once these additions are finalized

Minimalist Softmax Attention Provably Learns Constrained Boolean Functions

Jerry Yao-Chieh Hu, Xiwen Zhang, Maojiang Su, Zhao Song, Han Liu

机构 * Center for Foundation Models and Generative AI & Department of Computer Science, Northwestern University, USA（基础模型与生成人工智能中心及计算机科学系，西北大学，美国） ； School of Mathematical Sciences, Fudan University（复旦大学数学科学学院） ； University of California, Berkeley, USA（加州大学伯克利分校，美国） ； Center for Foundation Models and Generative AI & Department of Computer Science & Department of Statistics and Data Science, Northwestern University, USA（基础模型与生成人工智能中心及计算机科学系与统计与数据科学系，西北大学，美国） 纠错

专题命中 复杂问题求解 ：reasoning(abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.AI、cs.LG