大模型推理能力 - arXivDaily 专题

2510.08517 2025-10-10 cs.AI cs.CL cs.LG 67%

CaRT: Teaching LLM Agents to Know When They Know Enough

Grace Liu, Yuxiao Qu, Jeff Schneider, Aarti Singh, Aviral Kumar

机构 * Carnegie Mellon University（卡内基梅隆大学）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

2505.18356 2025-10-09 cs.CL cs.AI cs.LG 67%

The Unreasonable Effectiveness of Model Merging for Cross-Lingual Transfer in LLMs

Lucas Bandarkar, Nanyun Peng

机构 * University of California, Los Angeles（加州大学洛杉矶分校）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments MRL Workshop at EMNLP 2025

2502.15436 2025-10-07 cs.LG cs.AI cs.CL cs.DC 67%

Fed-SB: A Silver Bullet for Extreme Communication Efficiency and Performance in (Private) Federated LoRA Fine-Tuning

Raghav Singhal, Kaustubh Ponkshe, Rohit Vartak, Lav R. Varshney, Praneeth Vepakomma

机构 * Mohamed bin Zayed University of Artificial Intelligence（穆罕默德·本·扎耶德人工智能大学）； Duke University（杜克大学）； University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）； Massachusetts Institute of Technology（麻省理工学院）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Raghav Singhal and Kaustubh Ponkshe contributed equally to this work

2507.17746 2025-10-06 cs.LG cs.AI cs.CL 67%

Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains

Anisha Gunjal, Anthony Wang, Elaine Lau, Vaskar Nath, Yunzhong He, Bing Liu, Sean Hendryx

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments preprint

2510.01624 2025-10-03 cs.LG cs.AI cs.CL 67%

Quagmires in SFT-RL Post-Training: When High SFT Scores Mislead and What to Use Instead

Feiyang Kang, Michael Kuchnik, Karthik Padthe, Marin Vlastelica, Ruoxi Jia, Carole-Jean Wu, Newsha Ardalani

机构 * FAIR at Meta（Meta 的 FAIR）； Meta

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Preprint. Under Review

2411.19557 2025-10-03 cs.CL cs.AI cs.LG 67%

Initialization using Update Approximation is a Silver Bullet for Extremely Efficient Low-Rank Fine-Tuning

Kaustubh Ponkshe, Raghav Singhal, Eduard Gorbunov, Alexey Tumanov, Samuel Horvath, Praneeth Vepakomma

机构 * Mohamed bin Zayed University of Artificial Intelligence（穆罕默德·本·扎耶德人工智能大学）； Georgia Institute of Technology（佐治亚理工学院）； Massachusetts Institute of Technology（麻省理工学院）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Kaustubh Ponkshe and Raghav Singhal contributed equally to this work

2504.02670 2025-10-02 cs.AI cs.CL cs.IR cs.LG 67%

Affordable AI Assistants with Knowledge Graph of Thoughts

Maciej Besta, Lorenzo Paleari, Jia Hao Andrea Jiang, Robert Gerstenberger, You Wu, Jón Gunnar Hannesson, Patrick Iff, Ales Kubicek, Piotr Nyczyk, Diana Khimey, Nils Blach, Haiqiang Zhang, Tao Zhang, Peiran Ma, Grzegorz Kwaśniewski, Marcin Copik, Hubert Niewiadomski, Torsten Hoefler

机构 * ETH Zurich（苏黎世联邦理工学院）； Cledar（Cledar公司）； IDEAS Research Institute（IDEAS研究所）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

2509.25849 2025-10-01 cs.LG cs.AI cs.CL 67%

Knapsack RL: Unlocking Exploration of LLMs via Optimizing Budget Allocation

Ziniu Li, Congliang Chen, Tianyun Yang, Tian Ding, Ruoyu Sun, Ge Zhang, Wenhao Huang, Zhi-Quan Luo

机构 * The Chinese University of Hong Kong, Shenzhen（香港中文大学（深圳））； Shenzhen Research Institute of Big Data（深圳大数据研究院）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

2509.19249 2025-09-26 cs.CL cs.AI cs.LG 67%

Reinforcement Learning on Pre-Training Data

Siheng Li, Kejiao Li, Zenan Xu, Guanhua Huang, Evander Yang, Kun Li, Haoyuan Wu, Jiajia Wu, Zihao Zheng, Chenchen Zhang, Kun Shi, Kyrierl Deng, Qi Yi, Ruibin Xiong, Tingqiang Xu, Yuhao Jiang, Jianfeng Yan, Yuyuan Zeng, Guanghui Xu, Jinbao Xue, Zhijiang Xu, Zheng Fang, Shuai Li, Qibin Liu, Xiaoxue Li, Zhuoyu Li, Yangyu Tao, Fei Gao, Cheng Jiang, Bo Chao Wang, Kai Liu, Jianchen Zhu, Wai Lam, Wayyt Wang, Bo Zhou, Di Wang

机构 * LLM Department, Tencent（腾讯大模型部门）； HunYuan Infra Team（文心一言基础设施团队）； The Chinese University of Hong Kong（香港中文大学）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Work in progress

2509.02333 2025-09-09 cs.CL cs.AI cs.LG 67%

DCPO: Dynamic Clipping Policy Optimization

Shihui Yang, Chengfeng Dou, Peidong Guo, Kai Lu, Qiang Ju, Fei Deng, Rihui Xin

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

2504.15266 2025-09-01 cs.LG cs.AI cs.CL 67%

Roll the dice & look before you leap: Going beyond the creative limits of next-token prediction

Vaishnavh Nagarajan, Chen Henry Wu, Charles Ding, Aditi Raghunathan

机构 * Google Research, US（谷歌研究）； Carnegie Mellon University, Pittsburgh, US（卡内基梅隆大学）

专题命中数学推理：planning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments ICML 2025 (oral)

2508.15096 2025-08-22 cs.CL cs.AI cs.LG 67%

Nemotron-CC-Math: A 133 Billion-Token-Scale High Quality Math Pretraining Dataset

Rabeeh Karimi Mahabadi, Sanjeev Satheesh, Shrimai Prabhumoye, Mostofa Patwary, Mohammad Shoeybi, Bryan Catanzaro

机构 * NVIDIA（英伟达）； Boston University（波士顿大学）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

2405.17604 2025-08-21 cs.LG cs.AI cs.CL 67%

LoRA-XS: Low-Rank Adaptation with Extremely Small Number of Parameters

Klaudia Bałazy, Mohammadreza Banaei, Karl Aberer, Jacek Tabor

机构 * Jagiellonian University（雅盖隆大学）； EPFL（苏黎世联邦理工学院）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

2508.12790 2025-08-19 cs.AI cs.CL cs.LG 67%

Reinforcement Learning with Rubric Anchors

Zenan Huang, Yihong Zhuang, Guoshan Lu, Zeyu Qin, Haokai Xu, Tianyu Zhao, Ru Peng, Jiaqi Hu, Zhanming Shen, Xiaomeng Hu, Xijun Gu, Peiyi Tu, Jiaxin Liu, Wenyu Chen, Yuzhuo Fu, Zhiting Fan, Yanmei Gu, Yuanyuan Wang, Zhengkai Yang, Jianguo Li, Junbo Zhao

机构 * Inclusion AI ； Ant Group（蚂蚁集团）； Zhejiang University（浙江大学）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments technical report

2411.02083 2025-08-19 cs.CL cs.AI cs.CE cs.LG 67%

Regress, Don't Guess -- A Regression-like Loss on Number Tokens for Language Models

Jonas Zausinger, Lars Pennig, Anamarija Kozina, Sean Sdahl, Julian Sikora, Adrian Dendorfer, Timofey Kuznetsov, Mohamad Hagog, Nina Wiedemann, Kacper Chlodny, Vincent Limbach, Anna Ketteler, Thorben Prein, Vishwa Mohan Singh, Michael Morris Danziger, Jannis Born

机构 * Technical University of Munich（慕尼黑技术大学）； Munich Center for Machine Learning (MCML)（慕尼黑机器学习中心（MCML））； IBM Research Europe（IBM欧洲研究院）； ETH Zurich（苏黎世联邦理工学院）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments ICML 2025

2508.04138 2025-08-07 cs.LG cs.AI cs.CL 67%

COPO: Consistency-Aware Policy Optimization

Jinghang Han, Jiawei Chen, Hang Shao, Hao Ma, Mingcheng Li, Xintian Shen, Lihao Zheng, Wei Chen, Tao Wei, Lihua Zhang

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

2507.03958 2025-08-07 cs.IR cs.AI cs.CL cs.LG 67%

A Comparative Study of Specialized LLMs as Dense Retrievers

Hengran Zhang, Keping Bi, Jiafeng Guo

机构 * Key Laboratory of Network Data Science and Technology（网络数据科学与技术重点实验室）； Institute of Computing Technology（计算技术研究所）； Chinese Academy of Sciences（中国科学院）； State Key Laboratory of AI Safety（人工智能安全国家重点实验室）； University of Chinese Academy of Science（中国科学院大学）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Accepted by CCIR25 and published by Springer LNCS or LNAI

2504.07448 2025-08-05 cs.LG cs.AI cs.CL 67%

LoRI: Reducing Cross-Task Interference in Multi-Task Low-Rank Adaptation

Juzheng Zhang, Jiacheng You, Ashwinee Panda, Tom Goldstein

机构 * University of Maryland（马里兰大学）； Tsinghua University（清华大学）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments COLM 2025

2501.05727 2025-08-05 cs.CL cs.AI cs.LG 67%

Self-Evolving Critique Abilities in Large Language Models

Zhengyang Tang, Ziniu Li, Zhenyang Xiao, Tian Ding, Ruoyu Sun, Benyou Wang, Dayiheng Liu, Fei Huang, Tianyu Liu, Bowen Yu, Junyang Lin

机构 * The Chinese University of Hong Kong, Shenzhen（香港中文大学（深圳））； Shenzhen Research Institute of Big Data（深圳大数据研究院）； Shenzhen International Center for Industrial and Applied Mathematics（深圳国际工业与应用数学中心）； Qwen Team, Alibaba Inc.（阿里云通义团队）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Accepted by COLM 2025

2507.20150 2025-07-29 cs.AI cs.CL cs.LG 67%

The Policy Cliff: A Theoretical Analysis of Reward-Policy Maps in Large Language Models

Xingcheng Xu

机构 * Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

2402.07625 2025-07-23 cs.CL cs.AI cs.LG 67%

Autonomous Data Selection with Zero-shot Generative Classifiers for Mathematical Texts

Yifan Zhang, Yifan Luo, Yang Yuan, Andrew C Yao

机构 * Tsinghua University（清华大学）； Shanghai Qi Zhi Institute（上海启智研究院）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Published in ACL 2025 Findings

2504.09763 2025-07-22 cs.CL cs.AI cs.LG 67%

Executable Functional Abstractions: Inferring Generative Programs for Advanced Math Problems

Zaid Khan, Elias Stengel-Eskin, Archiki Prasad, Jaemin Cho, Mohit Bansal

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Project Page: https://zaidkhan.me/EFAGen/

2507.08034 2025-07-14 cs.CL cs.AI cs.LG 67%

Integrating External Tools with Large Language Models to Improve Accuracy

Nripesh Niketan, Hadj Batatia

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments 9 pages, 3 figures, 2 tables. Extended version of paper published in Proceedings of International Conference on Information Technology and Applications, Springer Nature Singapore, 2025, pp. 409-421. This version includes additional experimental results comparing against GPT-4o, LLaMA-Large, Mistral-Large, and Phi-Large, expanded evaluation methodology, and enhanced analysis

Journal ref Proceedings of International Conference on Information Technology and Applications, Springer Nature Singapore, 2025, pp. 409-421

2507.02841 2025-07-04 cs.AI cs.CL cs.LG 67%

StepHint: Multi-level Stepwise Hints Enhance Reinforcement Learning to Reason

Kaiyi Zhang, Ang Lv, Jinpeng Li, Yongbo Wang, Feng Wang, Haoyuan Hu, Rui Yan

机构 * GSAI, Renmin University of China（清华大学）； Peking University（北京大学）； Ant Group（蚂蚁集团）； SCS, Wuhan University（武汉大学）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

2411.05943 2025-06-23 cs.AI cs.CL cs.LG cs.LO 67%

Quantifying artificial intelligence through algorithmic generalization

Takuya Ito, Murray Campbell, Lior Horesh, Tim Klinger, Parikshit Ram

机构 * Mathematics & Theoretical Computer Science Department, T.J. Watson Research Center, IBM Research（数学与理论计算机科学系，T.J.沃森研究中心，IBM研究）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

2506.09340 2025-06-12 cs.CL cs.AI cs.LG 67%

RePO: Replay-Enhanced Policy Optimization

Siheng Li, Zhanhui Zhou, Wai Lam, Chao Yang, Chaochao Lu

机构 * The Chinese University of Hong Kong（香港中文大学）； Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Project Page: https://github.com/SihengLi99/RePO

2506.06607 2025-06-10 cs.CL cs.AI cs.LG 67%

Training-Free Tokenizer Transplantation via Orthogonal Matching Pursuit

Charles Goddard, Fernando Fernandes Neto

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

2410.14817 2025-06-04 cs.CL cs.AI cs.LG 67%

A Complexity-Based Theory of Compositionality

Eric Elmoznino, Thomas Jiralerspong, Yoshua Bengio, Guillaume Lajoie

机构 * Mila – Quebec AI Institute（魁北克人工智能研究所）； Université de Montréal（蒙特利尔大学）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

2503.09501 2025-05-28 cs.AI cs.CL cs.LG cs.MA 67%

ReMA: Learning to Meta-think for LLMs with Multi-Agent Reinforcement Learning

Ziyu Wan, Yunxiang Li, Xiaoyu Wen, Yan Song, Hanjing Wang, Linyi Yang, Mark Schmidt, Jun Wang, Weinan Zhang, Shuyue Hu, Ying Wen

机构 * Shanghai Jiao Tong University（上海交通大学）； Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）； University of British Columbia（不列颠哥伦比亚大学）； University College London（伦敦大学学院）

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

2410.01335 2025-05-27 cs.CL cs.AI cs.LG 67%

Layer Swapping for Zero-Shot Cross-Lingual Transfer in Large Language Models

Lucas Bandarkar, Benjamin Muller, Pritish Yuvraj, Rui Hou, Nayan Singhal, Hongjiang Lv, Bing Liu

专题命中数学推理：reasoning(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments ICLR 2025, Spotlight Paper, In The Thirteenth International Conference on Learning Representations, 2025

Journal ref The Thirteenth International Conference on Learning Representations (2025)

1. 数学推理 3252 篇

CaRT: Teaching LLM Agents to Know When They Know Enough

The Unreasonable Effectiveness of Model Merging for Cross-Lingual Transfer in LLMs

Fed-SB: A Silver Bullet for Extreme Communication Efficiency and Performance in (Private) Federated LoRA Fine-Tuning

Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains

Quagmires in SFT-RL Post-Training: When High SFT Scores Mislead and What to Use Instead

Initialization using Update Approximation is a Silver Bullet for Extremely Efficient Low-Rank Fine-Tuning

Affordable AI Assistants with Knowledge Graph of Thoughts

Knapsack RL: Unlocking Exploration of LLMs via Optimizing Budget Allocation

Reinforcement Learning on Pre-Training Data

DCPO: Dynamic Clipping Policy Optimization

Roll the dice & look before you leap: Going beyond the creative limits of next-token prediction

Nemotron-CC-Math: A 133 Billion-Token-Scale High Quality Math Pretraining Dataset

LoRA-XS: Low-Rank Adaptation with Extremely Small Number of Parameters

Reinforcement Learning with Rubric Anchors

Regress, Don't Guess -- A Regression-like Loss on Number Tokens for Language Models

COPO: Consistency-Aware Policy Optimization

A Comparative Study of Specialized LLMs as Dense Retrievers

LoRI: Reducing Cross-Task Interference in Multi-Task Low-Rank Adaptation

Self-Evolving Critique Abilities in Large Language Models

The Policy Cliff: A Theoretical Analysis of Reward-Policy Maps in Large Language Models

Autonomous Data Selection with Zero-shot Generative Classifiers for Mathematical Texts

Executable Functional Abstractions: Inferring Generative Programs for Advanced Math Problems

Integrating External Tools with Large Language Models to Improve Accuracy

StepHint: Multi-level Stepwise Hints Enhance Reinforcement Learning to Reason

Quantifying artificial intelligence through algorithmic generalization

RePO: Replay-Enhanced Policy Optimization

Training-Free Tokenizer Transplantation via Orthogonal Matching Pursuit

A Complexity-Based Theory of Compositionality

ReMA: Learning to Meta-think for LLMs with Multi-Agent Reinforcement Learning

Layer Swapping for Zero-Shot Cross-Lingual Transfer in Large Language Models