arXivDaily arXiv每日学术速递 周一至周五更新

期刊&会议

International Conference on Machine Learning · 会议 · Machine Learning

共收录 11797
2511.18507 2026-03-16 cs.CV cs.AI

Multimodal Continual Learning with MLLMs from Multi-scenario Perspectives

多场景视角下的多模态持续学习与MLLMs

Kai Jiang, Siqi Huang, Xiangyu Chen, Jiawei Shao, Hongyuan Zhang, Ping Luo, Xuelong Li

AI总结 本文提出UNIFIER框架,通过视觉表征扩展和视觉一致性约束解决多场景下的持续学习问题,提升跨场景视觉问答和F1分数。

Comments 22 pages, 17 figures. This is a preprint version of a paper submitted to ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2308.08705 2026-03-16 cs.LG cs.GT cs.MA

Partially Observable Multi-Agent Reinforcement Learning with Information Sharing

部分可观测多智能体强化学习与信息共享

Xiangyu Liu, Kaiqing Zhang

AI总结 本文研究了在部分可观测随机游戏框架下可证明的多智能体强化学习问题,提出利用智能体间的信息共享来解决计算复杂性问题,并设计了具有准多项式时间复杂度的算法,扩展至合作部分可观测马尔可夫决策过程。

Comments Final journal version of the ICML 2023 conference paper, accepted to SIAM Journal on Control and Optimization (SICON)

详情

展开后加载摘要…

URL PDF HTML 收藏
2406.01076 2026-03-13 cs.CV cs.AI cs.LG

Estimating Canopy Height at Scale

大规模估算树冠高度

Jan Pauls, Max Zimmer, Una M. Kelly, Martin Schwartz, Sassan Saatchi, Philippe Ciais, Sebastian Pokutta, Martin Brandt, Fabian Gieseke

AI总结 本文提出基于卫星数据的全球树冠高度估算框架,通过改进的损失函数和数据预处理技术提升预测精度,实现了比现有方法更优的MAE/RMSE结果,助力全球生态分析。

Comments ICML Camera-Ready, 17 pages, 14 figures, 7 tables

详情

展开后加载摘要…

URL PDF HTML 收藏
2501.19328 2026-03-13 cs.LG cs.AI cs.CV

Capturing Temporal Dynamics in Large-Scale Canopy Tree Height Estimation

在大规模树冠高度估计中捕捉时间动态

Jan Pauls, Max Zimmer, Berkant Turan, Sassan Saatchi, Philippe Ciais, Sebastian Pokutta, Fabian Gieseke

AI总结 本文提出了一种基于Sentinel-1和Sentinel-2数据生成欧洲大陆2019-2022年10米分辨率时间树冠高度地图的方法,利用GEDI激光雷达数据提升精度,实现大范围森林监测。

Comments ICML Camera-Ready, 9 pages main paper, 8 pages references and appendix, 9 figures, 8 tables

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.03782 2026-03-12 cond-mat.supr-con cond-mat.str-el cs.AI

Expert Evaluation of LLM World Models: A High-$T_c$ Superconductivity Case Study

专家评估LLM世界模型:一个高温超导体案例研究

Haoyu Guo, Maria Tikhanovskaya, Paul Raccuglia, Alexey Vlaskin, Chris Co, Daniel J. Liebling, Scott Ellsworth, Matthew Abraham, Elizabeth Dorfman, N. P. Armitage, Chunhan Feng, Antoine Georges, Olivier Gingras, Dominik Kiese, Steven A. Kivelson, Vadim Oganesyan, B. J. Ramshaw, Subir Sachdev, T. Senthil, J. M. Tranquada, Michael P. Brenner, Subhashini Venugopalan, Eun-Ah Kim

AI总结 本文通过专家评估LLM在高温超导领域的问题回答能力,发现基于RAG的系统在全面性和证据支持方面优于封闭模型。

Comments (v1) 9 pages, 4 figures, with 7-page supporting information. Accepted at the ICML 2025 workshop on Assessing World Models and the Explorations in AI Today workshop at ICML'25

Journal ref Proceedings of the National Academy of Sciences 123, e2533676123 (2026)

详情

展开后加载摘要…

URL PDF HTML 收藏
2601.22607 2026-03-11 cs.AI cs.CL

From Self-Evolving Synthetic Data to Verifiable-Reward RL: Post-Training Multi-turn Interactive Tool-Using Agents

从自演化合成数据到可验证奖励强化学习:训练后多轮交互工具使用智能体

Jiaxuan Gao, Jiaao Chen, Chuyi He, Shusheng Xu, Di Jin, Yi Wu

AI总结 本文提出EigenData框架,结合自演化数据代理与验证器强化学习,通过合成数据提升多轮交互工具使用智能体的训练效率和性能。

Comments Submitted to ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.09014 2026-03-11 cs.LG cs.CV

The Coupling Within: Flow Matching via Distilled Normalizing Flows

内部耦合:通过蒸馏的规范化流进行流匹配

David Berthelot, Tianrong Chen, Jiatao Gu, Marco Cuturi, Laurent Dinh, Bhavik Chandna, Michal Klein, Josh Susskind, Shuangfei Zhai

AI总结 本文提出规范化流匹配(NFM),通过蒸馏预训练的规范化流模型的耦合来训练学生模型,从而在性能和效率上均优于传统方法。

Comments Submitted to ICML 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.08658 2026-03-10 cs.LG

Context-free Self-Conditioned GAN for Trajectory Forecasting

无上下文自条件生成对抗网络用于轨迹预测

Tiago Rodrigues de Almeida, Eduardo Gutierrez Maestro, Oscar Martinez Mozos

机构 * Knut and Alice Wallenberg Foundation(瓦伦贝格基金会)

AI总结 本文提出了一种无上下文自条件GAN方法,用于从轨迹中学习不同模式,从而提升轨迹预测的准确性。

Comments Accepted at the 2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA)

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.08506 2026-03-10 cs.LG cs.AI

Oracle-Guided Soft Shielding for Safe Move Prediction in Chess

由Oracle引导的软屏蔽用于国际象棋中的安全移动预测

Prajit T Rajendran, Fabio Arnez, Huascar Espinoza, Agnes Delaborde, Chokri Mraidha

AI总结 OGSS通过学习概率安全模型,在国际象棋中实现安全探索,减少战术失误并提升探索效率。

Comments Accepted for publication at the 24th International Conference on Machine Learning and Applications (ICMLA), 2025. Preprint version in Arxiv

详情

展开后加载摘要…

URL PDF HTML 收藏
2505.16321 2026-03-10 cs.CV

Efficient Motion Prompt Learning for Robust Visual Tracking

高效运动提示学习用于鲁棒视觉跟踪

Jie Zhao, Xin Chen, Yongsheng Yuan, Michael Felsberg, Dong Wang, Huchuan Lu

AI总结 本文提出一种高效运动提示学习方法,通过整合运动和视觉线索提升视觉跟踪的鲁棒性,实验表明其在多个基准测试中表现优异,且训练成本低。

Comments Accepted by ICML2025

Journal ref Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:77353-77370, 2025, https://proceedings.mlr.press/v267/zhao25e.html

详情

展开后加载摘要…

URL PDF HTML 收藏
2508.02879 2026-03-10 cs.LG cs.AI

CauKer: Classification Time Series Foundation Models Can Be Pretrained on Synthetic Data

CauKer:分类时间序列基础模型可以在合成数据上进行预训练

Shifeng Xie, Vasilii Feofanov, Ambroise Odonnat, Lei Zan, Marius Alonso, Jianfeng Zhang, Themis Palpanas, Lujia Pan, Keli Zhang, Ievgen Redko

机构 * Université Paris Cité(巴黎大学) Huawei Noah’s Ark Lab(华为诺亚实验室)

AI总结 CauKer通过生成因果一致的合成时间序列数据,实现对分类时间序列基础模型的样本高效预训练。

Comments This manuscript combines material from the ICML 2025 TSFM Workshop paper and the ICLR 2026 Main Track paper

Journal ref ICLR 2026 Oral

详情

展开后加载摘要…

URL PDF HTML 收藏
2501.10466 2026-03-10 cs.LG cs.AI cs.CR cs.CV

Efficient Semi-Supervised Adversarial Training via Latent Clustering-Based Data Reduction

通过基于潜在聚类的数据减少实现高效的半监督对抗训练

Somrita Ghosh, Yuelin Xu, Xiao Zhang

机构 * CISPA Helmholtz Center for Information Security(CISPA信息安全赫尔姆霍兹中心)

AI总结 本文提出基于潜在聚类的数据减少方法,有效降低半监督对抗训练的数据和计算需求,同时保持鲁棒性优势。

Comments Shorter version of this work accepted by NextGenAISafety Workshop at ICML 2024

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.04343 2026-03-05 cs.CV cs.LG

Enhancing Authorship Attribution with Synthetic Paintings

通过合成绘画增强作者归属

Clarissa Loures, Caio Hosken, Luan Oliveira, Gianlucca Zuin, Adriano Veloso

AI总结 本研究通过结合真实和合成数据提升绘画作者归属的准确率和泛化能力。

Comments Accepted for publication at the 24th IEEE International Conference on Machine Learning and Applications (ICMLA 2025)

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.03587 2026-03-05 stat.ME cs.LG stat.ML

Controllable Generative Sandbox for Causal Inference

可控制的生成沙盒用于因果推断

Qi Zhang, Harsh Parikh, Ashley Naimi, Razieh Nabi, Christopher Kim, Timothy Lash

机构 * Emory University(埃默里大学) Yale University(耶鲁大学) Amgen(安进公司)

AI总结 CausalMix是一种结合高斯潜在先验与数据类型特定解码器的变分生成框架,通过显式因果控制实现对因果机制的可控生成,适用于混合类型表格数据的高保真模拟及因果推断研究。

Comments 34 pages, 15 figures. Submitted to ICML 2026. Code available at https://github.com/zhangqiecho/causalmix

详情

展开后加载摘要…

URL PDF HTML 收藏
2310.04925 2026-03-05 cs.LG

Crystal-GFN: sampling crystals with desirable properties and constraints

Crystal-GFN: 生成具有理想性质和约束条件的晶体

Mila AI4Science, :, Alex Hernandez-Garcia, Alexandre Duval, Alexandra Volokhova, Yoshua Bengio, Divya Sharma, Pierre Luc Carrier, Yasmine Benabed, Michał Koziarski, Victor Schmidt, Gian-Marco Rignanese, Pierre-Paul De Breuck, Paulette Clancy

机构 * Mila AI4Science Université de Montréal(蒙特利尔大学) CentraleSupélec, Université Paris-Saclay(中央圣艾修伯里学院,巴黎萨克莱大学) Johns Hopkins University(约翰霍普金斯大学) UCLouvain(布鲁塞尔自由大学)

AI总结 Crystal-GFN是一种生成具有理想性质和约束条件的晶体结构模型,通过多环境连续-离散GFlowNet高效发现多样且有效的晶体。

Comments This is the version of the manuscript submitted (though not accepted) to ICML 2024 in February 2024

详情

展开后加载摘要…

URL PDF HTML 收藏
2506.17623 2026-03-04 cs.MM cs.CV

Synthetic Perception: Can Generated Images Unlock Latent Visual Prior for Text-Centric Reasoning?

合成感知:生成图像能否解锁潜在的视觉先验以用于以文本为中心的推理?

Yuesheng Huang, Peng Zhang, Xiaoxin Wu, Riliang Liu, Jiaqi Liang

AI总结 本文探讨生成图像能否解锁潜在视觉先验以提升文本中心推理,通过多模态融合架构和提示工程策略实现性能提升。

Comments Accepted as a poster at the International Conference on Machine Learning (ICML 2025) NewInML Workshop

详情

展开后加载摘要…

URL PDF HTML 收藏
2502.16894 2026-03-04 cs.CL

Make LoRA Great Again: Boosting LoRA with Adaptive Singular Values and Mixture-of-Experts Optimization Alignment

让LoRA更出色:通过自适应奇异值和专家混合优化对齐提升LoRA

Chenghao Fan, Zhenyi Lu, Sichen Liu, Chengfeng Gu, Xiaoye Qu, Wei Wei, Yu Cheng

机构 * School of Computer Science \& Technology, Huazhong University of Science The Chinese University of Hong Kong Zhejiang University

AI总结 GOAT通过自适应奇异值和专家混合优化对齐提升LoRA性能,实现在多个任务上的最优表现。

Comments Accepted by ICML 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2410.14632 2026-03-04 cs.CL

Diverging Preferences: When do Annotators Disagree and do Models Know?

分歧偏好:标注者何时分歧以及模型是否知道?

Michael JQ Zhang, Zhilin Wang, Jena D. Hwang, Yi Dong, Olivier Delalleau, Yejin Choi, Eunsol Choi, Xiang Ren, Valentina Pyatkin

机构 * New York University(纽约大学) Allen Institute for Artificial Intelligence(人工智能研究院) NVIDIA(NVIDIA公司) University of Washington(华盛顿大学) University of Southern California(南加州大学)

AI总结 本文研究了标注者分歧的来源,发现任务不明确等因素导致大多数分歧,并提出方法以减轻其在LLM评估和训练中的影响。

Comments ICML 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2603.01306 2026-03-03 math.OC cs.LG

GPU-friendly and Linearly Convergent First-order Methods for Certifying Optimal $k$-sparse GLMs

适用于GPU的且线性收敛的一阶方法用于证明最优k-稀疏GLMs的最优性

Jiachang Liu, Andrea Lodi, Soroosh Shafiee

AI总结 本文提出了一种适用于GPU且线性收敛的一阶方法,用于高效证明稀疏GLMs的最优性,通过复合优化和对偶间隙理论提升计算效率。

Comments Extended version of the ICML 2025 conference paper

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.24195 2026-03-02 cs.AI cs.CL cs.CV cs.LG

Uncertainty Quantification for Multimodal Large Language Models with Incoherence-adjusted Semantic Volume

多模态大语言模型的不确定性量化:基于不一致性调整的语义体积

Gregory Kang Ruey Lau, Hieu Dao, Nicole Kan Hui Lin, Bryan Kian Hsiang Low

机构 * Department of Computer Science, National University of Singapore(新加坡国立大学计算机科学系)

AI总结 UMPIRE是一种无需训练的多模态大语言模型不确定性量化框架,通过内部特征有效捕捉语义多样性和响应不一致性,提升错误检测和不确定性校准性能。

Comments Earlier versions presented at ICLR 2025 QUESTION workshop and ICML 2025 R2-FM workshop

详情

展开后加载摘要…

URL PDF HTML 收藏
2402.08552 2026-03-02 cs.LG cs.CV

Confronting Reward Overoptimization for Diffusion Models: A Perspective of Inductive and Primacy Biases

对抗扩散模型中的奖励过度优化:从归纳偏置和优先偏置的角度出发

Ziyi Zhang, Sen Zhang, Yibing Zhan, Yong Luo, Yonggang Wen, Dacheng Tao

机构 * Institute of Artificial Intelligence, School of Computer Science, Wuhan University, China Hubei Luojia Laboratory, Wuhan, China The University of Sydney, Australia JD Explore Academy, Beijing, China Nanyang Technological University, Singapore

AI总结 本文提出TDPO-R算法,通过利用扩散模型的时间归纳偏置和抑制活跃神经元的优先偏置,有效缓解奖励过度优化问题。

Comments Accepted to ICML 2024

Journal ref International Conference on Machine Learning, pp. 60396-60413, 2024

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.22774 2026-02-27 eess.SY cs.SY

Transformer Actor-Critic for Efficient Freshness-Aware Resource Allocation

基于Transformer的Actor-Critic用于高效的 freshness-aware 资源分配

Maryam Ansarifard, Mohit K. Sharma, Kishor C. Joshi, George Exarchakos

AI总结 本文提出基于Transformer的Actor-Critic框架,用于高效实现 freshness-aware 资源分配,通过注意力机制提升策略性能和可扩展性。

Comments \c{opyright} 2026 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses. Accepted for publication in the 2026 IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN)

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.20031 2026-02-27 cs.AI cs.LG

Latent Introspection: Models Can Detect Prior Concept Injections

潜在自我反思:模型可以检测先前概念注入

Theia Pearson-Vogel, Martin Vanek, Raymond Douglas, Jan Kulveit

机构 * ACS Research, CTS, Charles University(ACS研究机构、CTs、查尔斯大学)

AI总结 Qwen 32B模型能检测并识别先前注入的概念,通过提供准确的AI自我反思机制信息可显著提升检测效果,同时提高注入概念间的互信息。

Comments 28 pages, 17 figures. Submitted to ICML 2026. Workshop version submitted to ICLR 2026 Workshop on Latent and Implicit Thinking

详情

展开后加载摘要…

URL PDF HTML 收藏
2504.20094 2026-02-26 cs.IR cs.CL cs.HC

Toward Safe and Human-Aligned Game Conversational Recommendation via Multi-Agent Decomposition

迈向安全且人本对齐的游戏对话推荐的多智能体分解

Zheng Hui, Xiaokai Wei, Yexi Jiang, Kevin Gao, Chen Wang, Frank Ong, Se-eun Yoon, Rachit Pareek, Michelle Gong

机构 * Roblox Corporation(Roblox公司) University of Cambridge(剑桥大学)

AI总结 MATCHA通过多智能体框架提升游戏对话推荐的安全性与人本对齐,实现更精细的个性化和更强的对抗防御能力。

Comments ICML 2025 MAS, EACL 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.21116 2026-02-25 eess.SP cs.AI

Attention-Based SINR Estimation in User-Centric Non-Terrestrial Networks

基于注意力机制的用户中心非地面网络中SINR估计

Bruno De Filippo, Alessandro Guidotti, Alessandro Vanelli-Coralli

机构 * Department of Electrical, Electronic, and Information Engineering (DEI), Univ. of Bologna(电子、电气与信息工程系,博洛尼亚大学) National Inter-University Consortium for Telecommunications (CNIT), Bologna, Italy(电信跨大学联合体(CNIT),博洛尼亚,意大利)

AI总结 本文提出基于注意力机制的低复杂度SINR估计方法,通过直接提取用户间干扰特征,实现无需MMSE计算的高精度SINR估计,适用于用户中心非地面网络的调度优化。

Comments Paper accepted for presentation at IEEE International Conference on Machine Learning in Communications and Networking (ICMLCN) 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2309.13411 2026-02-25 cs.LG cs.AI cs.CV

Towards Attributions of Input Variables in a Coalition

朝向联合体中输入变量的归因

Xinhao Zheng, Huiqi Deng, Quanshi Zhang

机构 * Shanghai Jiao Tong University(上海交通大学)

AI总结 本文提出了一种新的归因度量,用于评估联合体变量的归因一致性,通过分析AND-OR交互影响,解决归因冲突问题,并在多个领域验证了方法的有效性。

Comments Accepted to the 2025 International Conference on Machine Learning (ICML 2025)

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.18868 2026-02-24 math.OC cs.LG

Limits of Convergence-Rate Control for Open-Weight Safety

开放权重安全性的收敛速率控制极限

Domenic Rosati, Xijie Zeng, Hong Huang, Sebastian Dionicio, Subhabrata Majumdar, Frank Rudzicz, Hassan Sajjad

机构 * Dalhousie University(达尔豪斯大学) Vector Institute(向量研究所) Indian Institute of Management Bangalore(班加罗尔印度管理学院)

AI总结 本文提出SpecDef算法,通过谱重参数化在非对抗性设置中减缓优化收敛速度,并揭示了对抗性环境下收敛速率控制方法的理论极限。

Comments Submitted to ICML 2026. 13 figures, 30 tables

详情

展开后加载摘要…

URL PDF HTML 收藏
2503.11842 2026-02-24 cs.LG stat.ML

Test-Time Training Provably Improves Transformers as In-context Learners

测试时训练可证明地提高变换器作为上下文学习者

Halil Alperen Gozeten, M. Emrullah Ildiz, Xuechen Zhang, Mahdi Soltanolkotabi, Marco Mondelli, Samet Oymak

AI总结 测试时训练通过调整模型权重提升变换器在上下文学习中的性能,减少样本需求并提高推理效率。

Comments Accepted at ICML 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2402.10758 2026-02-24 stat.ML cs.LG stat.CO

Stochastic Localization via Iterative Posterior Sampling

通过迭代后验采样实现随机定位

Louis Grenioux, Maxence Noble, Marylou Gabrié, Alain Oliviero Durmus

机构 * CMAP, CNRS, École polytechnique, Institut Polytechnique de Paris(CMAP、法国国家科学研究中心、巴黎高等学院、巴黎理工学院)

AI总结 本文提出SLIPS方法,通过迭代后验采样实现随机定位,用于从无规范目标密度中采样,适用于多模分布的基准测试。

Comments Accepted at ICML 2024, improved assumption A0 (and consequences), fixed corollary 11

详情

展开后加载摘要…

URL PDF HTML 收藏
2602.18426 2026-02-23 astro-ph.GA cs.CV

Spatio-Spectroscopic Representation Learning using Unsupervised Convolutional Long-Short Term Memory Networks

基于无监督卷积长短期记忆网络的时空表示学习

Kameswara Bharadwaj Mantha, Lucy Fortson, Ramanakumar Sankar, Claudia Scarlata, Chris Lintott, Sandor Kruk, Mike Walmsley, Hugh Dickinson, Karen Masters, Brooke Simmons, Rebecca Smethurst

机构 * Department of Physics and Astronomy, University of Minnesota Twin Cities(物理与天文学系,明尼苏达大学双城分校) Physics Department, Lancaster University(拉斯特纳大学物理系) European Space Agency (ESA), European Space Astronomy Centre (ESAC)(欧洲航天局(ESA)、欧洲空间天文学中心(ESAC)) Departments of Physics and Astronomy, Haverford College(物理与天文学系,哈弗德学院) School of Physical Sciences, The Open University(物理科学学院,开放大学)

AI总结 本文提出了一种基于无监督卷积长短期记忆网络的深度学习框架,用于在空间和光谱维度上学习星系的通用特征表示,并在活跃星系核样本上进行了演示。

Comments This manuscript was previously submitted to ICML for peer review. Reviewers noted that while the underlying VAE-based architecture builds on established methods, its application to spatially-resolved IFS data is promising for unsupervised representation learning in astronomy. This version is released for community visibility. Reviewer decisions: Weak accept and Weak reject (Final: Reject)

详情

展开后加载摘要…

URL PDF HTML 收藏