NoLiMa: Long-Context Evaluation Beyond Literal Matching
机构 * LMU（莱茵-穆尔大学） ； Center for Information and Language Processing（信息与语言处理中心） ； Munich Center for Machine Learning (MCML)（慕尼黑机器学习中心） ； Adobe（Adobe公司）
Comments Accepted at ICML 2025
期刊&会议
ICML
International Conference on Machine Learning · 会议 · Machine Learning
Hybrid Quantum-Classical Multi-Agent Pathfinding
机构 * University of Bonn（波恩大学）
Comments 11 pages, accepted at ICML 2025
Emergence in non-neural models: grokking modular arithmetic via average gradient outer product
机构 * UC San Diego（圣迭戈大学） ； The Broad Institute of MIT and Harvard（哈佛大学与麻省理工学院 Broad 实验室） ； IIT Bombay（孟买印度理工学院）
Comments Accepted to ICML 2025 (Oral presentation & Spotlight poster)
Near-Optimal Consistency-Robustness Trade-Offs for Learning-Augmented Online Knapsack Problems
Comments 31 pages, 16 figures, Accepted at ICML 2025
MetaOptimize: A Framework for Optimizing Step Sizes and Other Meta-parameters
机构 * Openmind Research Institute, Canada（开放心智研究机构，加拿大） ； Leiden Institute of Advanced Computer Science, Leiden University, Leiden, Netherlands（莱顿先进计算机科学研究所，莱顿大学，莱顿，荷兰） ； Department of Computing Science, University of Alberta, Edmonton, Canada（计算科学系，阿尔伯塔大学，爱德蒙顿，加拿大）
Journal ref ICML 2025
Best-of-N through the Smoothing Lens: KL Divergence and Regret Analysis
机构 * The Alan Turing Institute, London, UK（艾伦·图灵研究所） ； Statistical Laboratory, University of Cambridge, Cambridge, UK（剑桥大学统计实验室） ； Massachusetts Institute of Technology, USA（麻省理工学院） ； IBM Research, USA（IBM研究院）
Comments Workshop on Efficient Systems for Foundation Models at iCML
Beyond Communication Overhead: A Multilevel Monte Carlo Approach for Mitigating Compression Bias in Distributed Learning
机构 * Viterby Faculty of Electrical and Computer Engineering（电气与计算机工程学院）
Comments Accepted to ICML 2025
Predicting mutational effects on protein binding from folding energy
机构 * Stanford University（斯坦福大学）
Comments Code: https://github.com/LDeng0205/StaB-ddG
Journal ref ICML 2025
Foreground-aware Virtual Staining for Accurate 3D Cell Morphological Profiling
机构 * Broad Institute of MIT and Harvard（MIT和哈佛大学Broad研究所） ； University of Michigan Medical School（密歇根大学医学学院） ； Institute of Molecular Biotechnology of the Austrian Academy of Sciences（奥地利科学院分子生物技术研究所） ； University of Michigan（密歇根大学） ； Shenzhen Research Institute of Big Data, Chinese University of Hong Kong–Shenzhen（深圳大学深圳大数据研究所） ； University of Michigan School of Nursing（密歇根大学护理学院）
Comments ICML 2025 Generative AI and Biology (GenBio) Workshop
Causal Foundation Models: Disentangling Physics from Instrument Properties
机构 * Massachusetts Institute of Technology（麻省理工学院） ； Harvard University（哈佛大学） ； New York University（纽约大学） ； Flatiron Institute（Flatiron研究所） ； The NSF AI Institute for Artificial Intelligence and Fundamental Interactions（NSF人工智能与基本相互作用研究所）
Comments 8 pages, 5 figures. Accepted to the ICML 2025 Foundation Models for Structured Data Workshop and accepted to the Machine Learning for Astrophysics Workshop 2025
AGACCI : Affiliated Grading Agents for Criteria-Centric Interface in Educational Coding Contexts
机构 * AA LAB, MODULABS（AA实验室，MODULABS） ； Aiffel, MODULABS（Aiffel，MODULABS） ； Department of Statistics, Inha university（统计系，Inha大学）
Comments Accepted at ICML 2025 Workshop on Multi-Agent Systems in the Era of Foundation Models: Opportunities, Challenges and Futures (MAS)
Speeding up Speculative Decoding via Sequential Approximate Verification
机构 * Department of Electrical and Computer Engineering, University of Arizona, Tucson, US（电气与计算机工程系，亚利桑那大学，图森，美国）
Comments ICML 2025, Workshop on Efficient Systems for Foundation Models
Learning to Plan & Reason for Evaluation with Thinking-LLM-as-a-Judge
机构 * FAIR at Meta（Meta 的 FAIR）
Comments ICML 2025
Balancing Efficiency and Expressiveness: Subgraph GNNs with Walk-Based Centrality
机构 * Imperial College London（帝国理工学院伦敦校区） ； Technion -- Israel Institute of Technology（技术学院——以色列理工学院） ； University of Oxford（牛津大学） ； NVIDIA Research（NVIDIA研究）
Comments ICML 2025
Regression for the Mean: Auto-Evaluation and Inference with Few Labels through Post-hoc Regression
机构 * Columbia University（哥伦比亚大学） ； Google DeepMind（谷歌DeepMind）
Comments Presented as a conference paper at ICML 2025
Are LLMs Prescient? A Continuous Evaluation using Daily News as the Oracle
机构 * New York University（纽约大学）
Comments ICML 2025
Liability and Insurance for Catastrophic Losses: the Nuclear Power Precedent and Lessons for AI
机构 * Independent Researcher, Cambridge Boston Alignment Initiative（独立研究者，剑桥波士顿对齐计划）
Comments Accepted to Generative AI and Law Workshop at the International Conference on Machine Learning (ICML 2024)
CoDy: Counterfactual Explainers for Dynamic Graphs
机构 * Centrum Wiskunde en Informatica, Amsterdam, Netherlands（荷兰阿姆斯特丹数学与信息学研究中心） ； University of Amsterdam, Amsterdam, Netherlands（荷兰阿姆斯特丹大学） ； Karlsruhe Institute of Technology, Karlsruhe, Germany（德国卡尔斯鲁厄技术大学）
Comments Proceedings in ICML 2025
Cascade: Token-Sharded Private LLM Inference
机构 * Ritual ； Stanford University（斯坦福大学） ； Columbia University（哥伦比亚大学）
Comments To be published in ICML 2025 Main Proceedings as "Hidden No More: Attacking and Defending Private Third-Party LLM Inference", together with arXiv:2505.18332
QuEst: Enhancing Estimates of Quantile-Based Distributional Measures Using Model Predictions
机构 * UNC Chapel Hill（UNC夏洛特山分校） ； Columbia University（哥伦比亚大学） ； Google Deepmind（谷歌DeepMind）
Comments Published as a conference paper at ICML 2025
Graph-Assisted Stitching for Offline Hierarchical Reinforcement Learning
机构 * Department of Computer Science and Engineering, Sungkyunkwan University, Suwon, Republic of Korea（计算机科学与工程系，成均馆大学，苏旺，韩国） ； Department of Artificial Intelligence, Sungkyunkwan University, Suwon, Republic of Korea（人工智能系，成均馆大学，苏旺，韩国）
Comments ICML 2025
Stop treating `AGI' as the north-star goal of AI research
Comments Position Paper accepted to ICML 2025. OpenReview: https://openreview.net/forum?id=1RlrtH6ydW
Machine Learning from Explanations
机构 * School of Computing, National University of Singapore, Singapore, Singapore（新加坡国立大学计算机学院）
Comments ICML 2025 AIW
MODA: MOdular Duplex Attention for Multimodal Perception, Cognition, and Emotion Understanding
机构 * VCIP \& TMCC \& DISSec, College of Computer Science, Nankai University ； Pengcheng Laboratory
Comments ICML 2025 (Spotlight, Top 2.6%)
any4: Learned 4-bit Numeric Representation for LLMs
机构 * FAIR at Meta（Meta 的 FAIR）
Comments ICML 2025
Accelerated Online Reinforcement Learning using Auxiliary Start State Distributions
Comments ICML ARLET Workshop 2024
Hierarchical Reinforcement Learning with Targeted Causal Interventions
机构 * School of Computer and Communication Sciences, EPFL, Lausanne, Switzerland（计算机与通信科学系，瑞士洛桑联邦理工学院） ； College of Management of Technology, EPFL, Lausanne, Switzerland（技术管理学院，瑞士洛桑联邦理工学院） ； Leiden Institute of Advanced Computer Science (LIACS), Leiden University, Leiden, The Netherlands（莱顿高级计算机科学研究所（LIACS），莱顿大学）
Comments 44 pages, 13 figures. Proceedings of ICML 2025. Proposes a new framework for hierarchical reinforcement learning (HRL) using targeted causal interventions, with theoretical analysis and experiments in synthetic and Minecraft environments
When Data-Free Knowledge Distillation Meets Non-Transferable Teacher: Escaping Out-of-Distribution Trap is All You Need
机构 * Sydney AI Centre, The University of Sydney（悉尼人工智能中心，悉尼大学） ； School of Computer Science, Wuhan University（武汉大学计算机学院） ； Department of Computer Science, Hong Kong Baptist University（香港 Baptist 大学计算机系）
Comments Accepted by ICML 2025
Evaluating Adversarial Protections for Diffusion Personalization: A Comprehensive Study
机构 * School of Cyberspace, Hangzhou Dianzi University, Hangzhou, China（信息空间学院，杭州电子大学，杭州，中国） ； Microsoft, Redmond, USA（微软，美国西雅图）
Comments Accepted to the 2nd Workshop on Reliable and Responsible Foundation Models (R2-FM 2025) at ICML. 8 pages, 3 figures
Return of the Latent Space COWBOYS: Re-thinking the use of VAEs for Bayesian Optimisation of Structured Spaces
机构 * School of Mathematical Sciences, Lancaster University, UK（数学科学学院，兰卡斯特大学） ； Department of Applied Maths and Theoretical Physics, University of Cambridge, UK（应用数学与理论物理系，剑桥大学）
Comments Accepted for publication at the International Conference on Machine Learning (ICML) 2025
