Is Noise Conditioning Necessary for Denoising Generative Models?
机构 * MIT（麻省理工学院）
Comments Update ImageNet experiments (SiT with CFG). Update Appendix
Journal ref Proceedings of the 42nd International Conference on Machine Learning (ICML), 2025
期刊&会议
ICML
International Conference on Machine Learning · 会议 · Machine Learning
Speculative Decoding in Decentralized LLM Inference: Turning Communication Latency into Computation Throughput
机构 * Department of XXX, University of YYY, Location, Country（XXX系，YYY大学，地点，国家） ； School of ZZZ, Institute of WWW, Location, Country（ZZZ学院，WWW研究所，地点，国家）
Comments 6 pages, 2 figures, 2 tables. Uses ICML 2025 style
A Global Geometric Analysis of Maximal Coding Rate Reduction
机构 * Department of Electrical Engineering and Computer Science, University of Michigan, Ann Arbor（电气工程与计算机科学系，密歇根大学） ； Antai College of Economics and Management, Shanghai Jiao Tong University, Shanghai（经济管理学院，上海交通大学） ； Department of Electrical Engineering and Computer Science, University of California, Berkeley（电气工程与计算机科学系，加州大学伯克利分校） ； Department of Computer Science and Engineering, Ohio State University, Columbus（计算机科学与工程系，俄亥俄州立大学） ； Institute of Data Science, University of Hong Kong, Hong Kong（数据科学研究所，香港大学）
Comments This work has been accepted for publication in the Proceedings of the 41st International Conference on Machine Learning (ICML 2024)
Distributed Linear Bandits under Communication Constraints
Journal ref Proceedings of the 40th International Conference on Machine Learning, PMLR 202:29845-29875, 2023
Logistic Variational Bayes Revisited
机构 * Department of XXX, University of YYY, Location, Country（XXX系，YYY大学，地点，国家） ； School of ZZZ, Institute of WWW, Location, Country（ZZZ学院，WWW研究所，地点，国家） ； Department of Mathematics, Imperial College London, United Kingdom（数学系，伦敦帝国学院，英国）
Comments Accepted at the 41st International Conference on Machine Learning
Provably Scalable Black-Box Variational Inference with Structured Variational Families
机构 * University of Pennsylvania（宾夕法尼亚大学）
Comments Accepted to ICML'24; v3, v4: fixed typos
Fundamental limits of learning in sequence multi-index models and deep attention networks: High-dimensional asymptotics and sharp thresholds
机构 * Emanuele Troiani Statistical Physics Of Computation Laboratory, École Polytechnique Fédérale de Lausanne (EPFL)（埃因斯坦理工大学统计物理计算实验室） ； Hugo Cui Center of Mathematical Sciences and Applications, Harvard University（哈佛大学数学科学中心） ； Yatin Dandi Statistical Physics Of Computation Laboratory, École Polytechnique Fédérale de Lausanne (EPFL) Information Learning and Physics Laboratory, École Polytechnique Fédérale de Lausanne (EPFL)（埃因斯坦理工大学统计物理计算实验室） ； Florent Krzakala Information Learning and Physics Laboratory, École Polytechnique Fédérale de Lausanne (EPFL)（埃因斯坦理工大学信息学习与物理实验室） ； Lenka Zdeborová Statistical Physics Of Computation Laboratory, École Polytechnique Fédérale de Lausanne (EPFL)（埃因斯坦理工大学统计物理计算实验室）
Journal ref ICML 2025
Counting in Small Transformers: The Delicate Interplay between Attention and Feed-Forward Layers
机构 * Statistical Physics of Computation Laboratory, Ècole polytechnique fédérale de Lausanne (EPFL), Lausanne, Switzerland（计算物理学实验室，瑞士洛桑联邦理工学院） ； Department of Computing Sciences, Università Bocconi, Milan, Italy（计算科学系，意大利博科尼大学）
Comments 9 pages main, 33 including appendix, ICML 2025
Journal ref ICML 2025
DP-AdamW: Investigating Decoupled Weight Decay and Bias Correction in Private Deep Learning
机构 * Harvard University（哈佛大学）
Comments 19 pages, 5 appendices; presented at ICML 2025 DIG-BUGS Workshop
DRAGON: Guard LLM Unlearning in Context via Negative Detection and Reasoning
机构 * University of California, Santa Cruz（加州大学圣克ruz分校） ； Center for Advanced AI, Accenture（Accenture高级人工智能中心）
Comments Please refer to the NeurIPS 2025 submission: https://openreview.net/forum?id=FNuul0hlin The paper has been accepted to the ICML 2025 MUGen Workshop: https://openreview.net/forum?id=ET24oKP23c
Internal Causal Mechanisms Robustly Predict Language Model Out-of-Distribution Behaviors
机构 * stanford（斯坦福大学）
Comments ICML 2025
GPT, But Backwards: Exactly Inverting Language Model Outputs
Comments 7 pages, ICML 2025 Workshop on Reliable and Responsible Foundation Models
Understanding and Mitigating Memorization in Diffusion Models for Tabular Data
机构 * Department of Computer ； Data Sciences, Case Western Reserve University, Cleveland, USA ； Department of Computer Science \& Engineering, Texas A\&M University, College Station, USA ； Department of Biochemistry, Case Western Reserve University, Cleveland, USA ； Center for RNA Science ； Therapeutics, Case Western Reserve University, Cleveland, USA ； Department of Biomedical Engineering, Case Western Reserve University, Cleveland, USA
Comments Published in ICML 2025 (PMLR Volume 267): https://proceedings.mlr.press/v267/fang25f.html
Journal ref Proceedings of the 42nd International Conference on Machine Learning (ICML 2025), PMLR 267, Vancouver, Canada
Probably Approximately Global Robustness Certification
Comments ICML 2025
LIVS: A Pluralistic Alignment Dataset for Inclusive Public Spaces
机构 * Mila–Quebec AI Institute（魁北克AI研究所）
Comments ICML 2025
The Right to AI
机构 * Mila – Quebec AI Institute（魁北克AI研究所）
Comments ICML 2025
Conceptual Belief-Informed Reinforcement Learning
机构 * University of California, Berkeley（加州大学伯克利分校） ； Columbia University（哥伦比亚大学） ； Johns Hopkins University（约翰霍普金斯大学）
Comments Accepted by ICML 2025 Workshop on Models of Human Feedback for AI Alignment
Journal ref ICML 2025 Workshop on Models of Human Feedback for AI Alignment
Optimization without Retraction on the Random Generalized Stiefel Manifold
机构 * ICTEAM Institute, UCLouvain（ICTEAM学院，UCLouvain大学） ； Department of Statistics, University of Oxford（统计系，牛津大学） ； Apple Machine Learning Group（苹果机器学习小组） ； Academy of Mathematics and Systems Science, Chinese Academy of Sciences（数学与系统科学学院，中国科学院）
Comments This v4 is a corrected version of the ICML 2024 paper (PMLR 235:49226-49248); see the errata at the end for the list of changes
Journal ref Proceedings of the 41st International Conference on Machine Learning (ICML 2024), PMLR 235 (2024), 49226-49248
Sample Complexity of Distributionally Robust Off-Dynamics Reinforcement Learning with Online Interaction
机构 * Duke University（杜克大学）
Comments 53 pages, 6 figures, 3 tables. Published in Proceedings of the 42nd International Conference on Machine Learning (ICML 2025)
Journal ref Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:22595-22646, 2025
MorphTok: Morphologically Grounded Tokenization for Indian Languages
Comments Accepted at Tokenization Workshop (TokShop), ICML 2025
TAMAS: Benchmarking Adversarial Risks in Multi-Agent LLM Systems
机构 * International Institute of Information Technology, Hyderabad（国际信息科技研究所，海得拉巴） ； Microsoft Research, India（微软研究院，印度）
Comments Accepted at ICML 2025 MAS Workshop. This version includes additional experiments and analysis
Prediction-Powered Adaptive Shrinkage Estimation
机构 * Data Science Institute, The University of Chicago（芝加哥大学数据科学研究所） ； Department of Statistics, The University of Chicago（芝加哥大学统计学系）
Comments Accepted as poster in ICML 2025
Asymptotically Unbiased Synthetic Control Methods by Moment Matching
Comments This study was presented at the Workshop on Counterfactuals in Minds and Machines at the International Conference on Machine Learning in July 2023 and at the International Conference on Econometrics and Statistics in August 2023
Complexity as Advantage: A Regret-Based Perspective on Emergent Structure
机构 * IBM Research（IBM研究院）
Comments 15 pages. Under preparation for submission to ICML 2026. Feedback welcome
Fraud-Proof Revenue Division on Subscription Platforms
机构 * University of Oxford, UK（牛津大学） ； Harvard University, USA（哈佛大学）
Comments Appears in the 42nd International Conference on Machine Learning (ICML), 2025
Learning Filter-Aware Distance Metrics for Nearest Neighbor Search with Multiple Filters
机构 * Microsoft Research India（微软印度研究院） ； Microsoft Corporation（微软公司）
Comments 1st Workshop on Vector Databases at International Conference on Machine Learning, 2025
Navigating High Dimensional Concept Space with Metalearning
机构 * Department of Computer Science（计算机科学系） ； Princeton University（普林斯顿大学）
Comments 7 pages, 3 figures. Presented at the ICML 2025 HiLD Workshop
Decision-aware training of spatiotemporal forecasting models to select a top K subset of sites for intervention
机构 * Department of Computer Science, Tufts University, Medford, Massachusetts, United States（计算机科学系，塔夫茨大学，马萨诸塞州梅德福，美国） ； Department of Public Health and Community Medicine, Tufts University School of Medicine, Boston, Massachusetts, United States（公共卫生与社区医学系，塔夫茨大学医学院，马萨诸塞州波士顿，美国）
Comments 9 pages, 3 figures
Journal ref Forty-second International Conference on Machine Learning, 2025
Dissecting FLOPs along input dimensions for GreenAI cost estimations
机构 * University of Bologna Department of Informatics: Science ； Engineering (DISI) University of Bologna Department of Mathematics
Comments Article accepted at the 7th International Conference on Machine Learning, Optimization, and Data Science. October 4-8, 2021, Grasmere, Lake District, UK
Learning Monotonic Probabilities with a Generative Cost Model
Journal ref Proceedings of the 42nd International Conference on Machine Learning (ICML), PMLR 267:58789-58804 (2025)
