语言大模型 / LLM - arXivDaily 专题

2508.08701 2025-11-26 cs.CV cs.AI cs.LG 62%

SafeFix: Targeted Model Repair via Controlled Image Generation

SafeFix: 通过受控图像生成实现目标模型修复

Ouyang Xu, Baoming Zhang, Ruiyu Mao, Yunhui Guo

机构 * The University of Texas at Dallas（德克萨斯大学达拉斯分校）

专题命中预训练与数据：language model(abstract)；分类 cs.AI、cs.LG

AI总结 SafeFix通过生成语义忠实的图像来修复模型，提升模型对罕见案例的鲁棒性，减少系统性错误。

2511.17854 2025-11-25 cs.CL cs.AI cs.CY cs.HC cs.MA 62%

A superpersuasive autonomous policy debating system

一种超级有说服力的自主辩论系统

Allen Roush, Devin Gonier, John Hines, Judah Goldfeder, Philippe Martin Wyder, Sanjay Basu, Ravid Shwartz Ziv

专题命中预训练与数据：LLM(abstract)；分类 cs.CL、cs.AI

AI总结 DeepDebater是一种能够自主参与并赢得完整政策辩论的AI系统，通过多智能体协作生成高质量论证并战胜人类对手。

Comments Accepted to CLIP workshop at AAAI 2026

2511.16912 2025-11-24 cs.LG cs.AI 62%

PepEVOLVE: Position-Aware Dynamic Peptide Optimization via Group-Relative Advantage

PepEVOLVE: 通过组相对优势实现位置感知的动态肽优化

Trieu Nguyen, Hao-Wei Pang, Shasha Feng

机构 * Merck & Co., Inc.（默克公司）； University of South Florida（佛罗里达州立大学）

专题命中预训练与数据：pretraining(abstract)；分类 cs.AI、cs.LG

AI总结 PepEVOLVE通过位置感知和动态优化方法，实现多目标肽优化，优于PepINVENT。

2511.03601 2025-11-20 cs.CL cs.AI cs.HC cs.SD eess.AS 62%

Step-Audio-EditX Technical Report

Chao Yan, Boyong Wu, Peng Yang, Pengfei Tan, Guoqiang Hu, Li Xie, Yuxin Zhang, Xiangyu, Zhang, Fei Tian, Xuerui Yang, Xiangyu Zhang, Daxin Jiang, Shuchang Zhou, Gang Yu

机构 * StepFun

专题命中预训练与数据：LLM(abstract)；分类 cs.CL、cs.AI

2511.13685 2025-11-18 cs.LG cs.AI 62%

Protein Secondary Structure Prediction Using 3D Graphs and Relation-Aware Message Passing Transformers

Disha Varshney, Samarth Garg, Sarthak Tyagi, Deeksha Varshney, Nayan Deep, Asif Ekbal

机构 * Banaras Hindu University, Varanasi, India（班纳拉锡大学）； Indiana University, Bloomington（印第安纳大学）； Indian Institute of Technology Jodhpur, India（印度理工学院朱罗尔分校）； Indian Institute of Technology Patna, India（印度理工学院帕坦分校）

专题命中预训练与数据：language model(abstract)；分类 cs.AI、cs.LG

Comments 40 pages

2503.09443 2025-11-18 cs.CL cs.LG 62%

Scaling Laws for Conditional Emergence of Multilingual Image Captioning via Generalization from Translation

Julian Spravil, Sebastian Houben, Sven Behnke

专题命中预训练与数据：language model(abstract)；分类 cs.CL、cs.LG

2511.08544 2025-11-17 cs.LG cs.AI cs.CV stat.ML 62%

LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics

Randall Balestriero, Yann LeCun

机构 * Brown University（布朗大学）； New York University (NYU)（纽约大学）； Meta-FAIR

专题命中预训练与数据：pretraining(abstract)；分类 cs.AI、cs.LG

2511.09923 2025-11-17 cs.LG cs.AI cs.RO 62%

Harnessing Bounded-Support Evolution Strategies for Policy Refinement

Ethan Hirschowitz, Fabio Ramos

机构 * The University of Sydney（悉尼大学）； NVIDIA（NVIDIA公司）

专题命中预训练与数据：pretraining(abstract)；分类 cs.AI、cs.LG

Comments 10 pages, 6 figures, to be published in Australasian Conference on Robotics and Automation (ACRA 2025)

2511.07461 2025-11-12 cs.CL cs.AI 62%

It Takes Two: A Dual Stage Approach for Terminology-Aware Translation

Akshat Singh Jaswal

机构 * PES University（PES大学）

专题命中预训练与数据：LLM(abstract)；分类 cs.CL、cs.AI

Comments Accepted to WMT 2025. Code availavle at https://github.com/akshat-sj/duterm

Journal ref In Proceedings of the Tenth Conference on Machine Translation, pages 1344-1350, Suzhou, China. Association for Computational Linguistics (2025)

2510.16227 2025-11-10 cs.CL cs.AI 62%

What Can String Probability Tell Us About Grammaticality?

Jennifer Hu, Ethan Gotlieb Wilcox, Siyuan Song, Kyle Mahowald, Roger P. Levy

机构 * Kempner Institute for the Study of Natural and Artificial Intelligence at Harvard University（哈佛大学自然与人工智能研究 institute）

专题命中预训练与数据：language model(abstract)；分类 cs.CL、cs.AI

2511.04228 2025-11-07 cs.CL cs.LG 62%

REMIND: Input Loss Landscapes Reveal Residual Memorization in Post-Unlearning LLMs

Liran Cohen, Yaniv Nemcovesky, Avi Mendelson

机构 * Technion - Israel Institute of Technology（技术离子-以色列理工学院）

专题命中预训练与数据：language model(abstract)；分类 cs.CL、cs.LG

Comments Pre-print version under review

2504.12463 2025-11-05 cs.LG cs.AI 62%

Dense Backpropagation Improves Training for Sparse Mixture-of-Experts

Ashwinee Panda, Vatsal Baherwani, Zain Sarwar, Benjamin Therien, Sambit Sahu, Tom Goldstein, Supriyo Chakraborty

机构 * University of Maryland（马里兰大学）； Capital One ； University of Chicago（芝加哥大学）； Mila – Quebec AI Institute（魁北克人工智能研究所）

专题命中预训练与数据：pretraining(abstract)；分类 cs.AI、cs.LG

Comments NeurIPS 2025

2505.06111 2025-11-04 cs.RO cs.AI cs.LG 62%

UniVLA: Learning to Act Anywhere with Task-centric Latent Actions

Qingwen Bu, Yanting Yang, Jisong Cai, Shenyuan Gao, Guanghui Ren, Maoqing Yao, Ping Luo, Hongyang Li

专题命中预训练与数据：pretraining(abstract)；分类 cs.AI、cs.LG

Comments Accepted to RSS 2025. Code is available at https://github.com/OpenDriveLab/UniVLA

2511.00191 2025-11-04 cs.CV cs.AI cs.LG 62%

A Retrospect to Multi-prompt Learning across Vision and Language

Ziliang Chen, Xin Huang, Quanlong Guan, Liang Lin, Weiqi Luo

机构 * Jinan University（济南大学）； Sun Yat-sen University（中山大学）； Pazhou Laboratory（琶洲实验室）

专题命中预训练与数据：pretraining(abstract)；分类 cs.AI、cs.LG

Comments ICCV

2510.27062 2025-11-03 cs.LG cs.AI 62%

Consistency Training Helps Stop Sycophancy and Jailbreaks

Alex Irpan, Alexander Matt Turner, Mark Kurzeja, David K. Elson, Rohin Shah

机构 * Google（谷歌）

专题命中预训练与数据：LLM(abstract)；分类 cs.AI、cs.LG

Comments 19 pages

2508.07976 2025-10-28 cs.CL cs.AI 62%

Beyond Ten Turns: Unlocking Long-Horizon Agentic Search with Large-Scale Asynchronous RL

Jiaxuan Gao, Wei Fu, Minyang Xie, Shusheng Xu, Chuyi He, Zhiyu Mei, Banghua Zhu, Yi Wu

机构 * IIIS, Tsinghua University（清华大学智能技术学部）； Ant Group（蚂蚁集团）； University of Washington（华盛顿大学）

专题命中预训练与数据：LLM(abstract)；分类 cs.CL、cs.AI

2507.10741 2025-10-28 cs.LG cs.AI 62%

Ground-Compose-Reinforce: Grounding Language in Agentic Behaviours using Limited Data

Andrew C. Li, Toryn Q. Klassen, Andrew Wang, Parand A. Alamdari, Sheila A. McIlraith

机构 * Department of Computer Science, University of Toronto（计算机科学系，多伦多大学）； Vector Institute for Artificial Intelligence（人工智能矢量研究所）； Schwartz Reisman Institute for Technology and Society（技术与社会施瓦茨-雷曼研究所）

专题命中预训练与数据：pretraining(abstract)；分类 cs.AI、cs.LG

Comments NeurIPS 2025

2507.07543 2025-10-28 cs.CL cs.AI cs.IR 62%

The Cross-Lingual Cost: Retrieval Biases in RAG over Arabic-English Corpora

Chen Amiraz, Yaroslav Fyodorov, Elad Haramaty, Zohar Karnin, Liane Lewin-Eytan

机构 * Technology Innovation Institute（技术创新研究所）

专题命中预训练与数据：pretraining(abstract)；分类 cs.CL、cs.AI

Comments Accepted to ArabicNLP 2025

2510.22141 2025-10-28 cs.CV cs.CL cs.LG cs.RO eess.IV 62%

LOC: A General Language-Guided Framework for Open-Set 3D Occupancy Prediction

Yuhang Gao, Xiang Xiang, Sheng Zhong, Guoyou Wang

专题命中预训练与数据：language model(abstract)；分类 cs.CL、cs.LG

2510.20002 2025-10-27 cs.CL cs.AI 62%

Forging GEMs: Advancing Greek NLP through Quality-Based Corpus Curation

Alexandra Apostolopoulou, Konstantinos Kanaris, Athanasios Koursaris, Dimitris Tsakalidis, George Domalis, Ioannis E. Livieris

机构 * Novelcore ； Department of Statistics & Insurance University of Piraeus（统计与保险系帕里埃斯大学）

专题命中预训练与数据：language model(abstract)；分类 cs.CL、cs.AI

Comments The manuscript is submitted to Applied Sciences

2506.10226 2025-10-27 cs.CV cs.AI cs.LG 62%

ScoreMix: Synthetic Data Generation by Score Composition in Diffusion Models Improves Recognition

Parsa Rahimi, Sebastien Marcel

机构 * EPFL（苏黎世联邦理工学院）； Idiap（日内瓦智能感知研究院）

专题命中预训练与数据：foundation model(abstract)；分类 cs.AI、cs.LG

Comments Extended version of ICMLw25 Oral

2506.01728 2025-10-27 cs.LG cs.AI 62%

Principled Data Augmentation for Learning to Solve Quadratic Programming Problems

Chendi Qian, Christopher Morris

机构 * Department of Computer Science（计算机科学系）； RWTH Aachen University（亚琛工业大学）

专题命中预训练与数据：pretraining(abstract)；分类 cs.AI、cs.LG

Comments Accepted at NeurIPS 2025 as spotlight

2510.20256 2025-10-24 cs.CV cs.CL cs.LG cs.MM 62%

Calibrating Multimodal Consensus for Emotion Recognition

Guowei Zhong, Junjie Li, Huaiyu Zhu, Ruohong Huan, Yun Pan

机构 * College of Information Science and Electronic Engineering, Zhejiang University（信息科学与电子工程学院，浙江大学）； College of Computer Science and Technology, Zhejiang University of Technology（计算机科学与技术学院，浙江工业大学）； Zhejiang University Jinhua Research Institute（浙江大学金华研究院）

专题命中预训练与数据：pretraining(abstract)；分类 cs.CL、cs.LG

2506.16349 2025-10-24 cs.LG cs.AI cs.CR cs.CV 62%

Watermarking Autoregressive Image Generation

Nikola Jovanović, Ismail Labiad, Tomáš Souček, Martin Vechev, Pierre Fernandez

机构 * Meta FAIR ； ETH Zurich

专题命中预训练与数据：language model(abstract)；分类 cs.AI、cs.LG

Comments NeurIPS 2025

2505.23883 2025-10-24 cs.CV cs.CL cs.LG 62%

BioCLIP 2: Emergent Properties from Scaling Hierarchical Contrastive Learning

Jianyang Gu, Samuel Stevens, Elizabeth G Campolongo, Matthew J Thompson, Net Zhang, Jiaman Wu, Andrei Kopanev, Zheda Mai, Alexander E. White, James Balhoff, Wasila Dahdul, Daniel Rubenstein, Hilmar Lapp, Tanya Berger-Wolf, Wei-Lun Chao, Yu Su

机构 * The Ohio State University（俄亥俄州立大学）； Smithsonian Institution（史密森学会）； UNC Chapel Hill（北卡罗来纳大学教堂山分校）； University of California, Irvine（加州大学伊市分校）； Princeton University（普林斯顿大学）； Duke University（杜克大学）

专题命中预训练与数据：foundation model(abstract)；分类 cs.CL、cs.LG

Comments NeurIPS 2025 Spotlight; Project page: https://imageomics.github.io/bioclip-2/

2506.21374 2025-10-23 cs.LG cs.AI 62%

Pay Attention to Small Weights

Chao Zhou, Tom Jacobs, Advait Gadhikar, Rebekka Burkholz

机构 * CISPA Helmholtz Center for Information Security（CISPA 情报安全研究中心）

专题命中预训练与数据：pretraining(abstract)；分类 cs.AI、cs.LG

2510.17720 2025-10-21 cs.CL cs.AI 62%

PANER: A Paraphrase-Augmented Framework for Low-Resource Named Entity Recognition

Nanda Kumar Rengarajan, Jun Yan, Chun Wang

机构 * Concordia University（康科德大学）

专题命中预训练与数据：instruction tuning(abstract)；分类 cs.CL、cs.AI

2510.14947 2025-10-21 cs.RO cs.AI cs.LG cs.SY eess.SY 62%

Architecture Is All You Need: Diversity-Enabled Sweet Spots for Robust Humanoid Locomotion

Blake Werner, Lizhi Yang, Aaron D. Ames

机构 * Caltech MCE（Caltech机械工程系）

专题命中预训练与数据：pretraining(abstract)；分类 cs.AI、cs.LG

Comments 8 pages

2510.14968 2025-10-17 cs.RO cs.AI cs.CV cs.LG cs.SY eess.SY 62%

RDD: Retrieval-Based Demonstration Decomposer for Planner Alignment in Long-Horizon Tasks

Mingxuan Yan, Yuping Wang, Zechun Liu, Jiachen Li

机构 * University of California, Riverside（加州大学河滨分校）； University of Michigan（密歇根大学）； Meta AI

专题命中预训练与数据：language model(abstract)；分类 cs.AI、cs.LG

Comments 39th Conference on Neural Information Processing Systems (NeurIPS 2025); Project Website: rdd-neurips.github.io

2510.14139 2025-10-17 cs.LG cs.AI 62%

Inferred global dense residue transition graphs from primary structure sequences enable protein interaction prediction via directed graph convolutional neural networks

Islam Akef Ebeid, Haoteng Tang, Pengfei Gu

机构 * Division of Computer Science Texas Woman’s University（计算机科学系德克萨斯女子大学）； Department of Computer Science The University of Texas Rio Grande Valley（计算机科学系德克萨斯大学里奥格兰德谷分校）

专题命中预训练与数据：language model(abstract)；分类 cs.AI、cs.LG

Comments under review in Frontiers in Bioinformatics

1. 预训练与数据 12486 篇

SafeFix: Targeted Model Repair via Controlled Image Generation

A superpersuasive autonomous policy debating system

PepEVOLVE: Position-Aware Dynamic Peptide Optimization via Group-Relative Advantage

Step-Audio-EditX Technical Report

Protein Secondary Structure Prediction Using 3D Graphs and Relation-Aware Message Passing Transformers

Scaling Laws for Conditional Emergence of Multilingual Image Captioning via Generalization from Translation

LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics

Harnessing Bounded-Support Evolution Strategies for Policy Refinement

It Takes Two: A Dual Stage Approach for Terminology-Aware Translation

What Can String Probability Tell Us About Grammaticality?

REMIND: Input Loss Landscapes Reveal Residual Memorization in Post-Unlearning LLMs

Dense Backpropagation Improves Training for Sparse Mixture-of-Experts

UniVLA: Learning to Act Anywhere with Task-centric Latent Actions

A Retrospect to Multi-prompt Learning across Vision and Language

Consistency Training Helps Stop Sycophancy and Jailbreaks

Beyond Ten Turns: Unlocking Long-Horizon Agentic Search with Large-Scale Asynchronous RL

Ground-Compose-Reinforce: Grounding Language in Agentic Behaviours using Limited Data

The Cross-Lingual Cost: Retrieval Biases in RAG over Arabic-English Corpora

LOC: A General Language-Guided Framework for Open-Set 3D Occupancy Prediction

Forging GEMs: Advancing Greek NLP through Quality-Based Corpus Curation

ScoreMix: Synthetic Data Generation by Score Composition in Diffusion Models Improves Recognition

Principled Data Augmentation for Learning to Solve Quadratic Programming Problems

Calibrating Multimodal Consensus for Emotion Recognition

Watermarking Autoregressive Image Generation

BioCLIP 2: Emergent Properties from Scaling Hierarchical Contrastive Learning

Pay Attention to Small Weights

PANER: A Paraphrase-Augmented Framework for Low-Resource Named Entity Recognition

Architecture Is All You Need: Diversity-Enabled Sweet Spots for Robust Humanoid Locomotion

RDD: Retrieval-Based Demonstration Decomposer for Planner Alignment in Long-Horizon Tasks

Inferred global dense residue transition graphs from primary structure sequences enable protein interaction prediction via directed graph convolutional neural networks