OR-Bench: An Over-Refusal Benchmark for Large Language Models
Comments Accepted to ICML 2025, we thank everyone for their valuable suggestions and feedback!
期刊&会议
ICML
International Conference on Machine Learning · 会议 · Machine Learning
Rethinking Explainable Machine Learning as Applied Statistics
机构 * University of Tübingen, Tübingen AI Center, Germany（图宾根大学，图宾根人工智能中心，德国） ； University of Tübingen, Germany（图宾根大学，德国）
Comments ICML 2025 camera ready
EgoPrivacy: What Your First-Person Camera Says About You?
机构 * University of California, San Diego（加州大学圣地亚哥分校） ； Qualcomm AI Research, an initiative of Qualcomm Technologies, Inc（高通人工智能研究） ； University of Electronic Science and Technology of China（电子科技大学）
Comments ICML 2025
Long-Short Alignment for Effective Long-Context Modeling in LLMs
机构 * State Key Lab of General Artificial Intelligence, School of Intelligence Science and Technology, Peking University, China（人工智能国家重点实验室，智能科学与技术学院，北京大学） ； MIT CSAIL, USA（麻省理工学院计算机科学与人工智能实验室） ； Institute for Artificial Intelligence, Peking University, China（人工智能研究院，北京大学）
Comments ICML 2025
Configurable Preference Tuning with Rubric-Guided Synthetic Data
机构 * Komorebi AI Technologies, Madrid, Spain（Komorebi人工智能技术公司）
Comments Accepted to ICML 2025 Workshop on Models of Human Feedback for AI Alignment
Dynamic Mixture of Curriculum LoRA Experts for Continual Multimodal Instruction Tuning
机构 * Department of Computer Science and Technology, BNRist, Tsinghua University, Beijing, China（计算机科学与技术系，BNRist，清华大学，北京，中国） ； Alibaba Group, Hangzhou, China（阿里巴巴集团，杭州，中国）
Comments Accepted by ICML 2025
A Comparative Analysis of Influence Signals for Data Debugging
机构 * Department of Computer Science & ETIS Lab University of Crete, Greece & ENSEA, France（计算机科学系及ETIS实验室 希腊克里特大学及法国ENSEA） ； Department of Computer Science University of Crete, Greece（计算机科学系 希腊克里特大学） ； ETIS Lab ENSEA, France（ETIS实验室 法国ENSEA）
Comments Accepted and presented at the Data-centric Machine Learning Research (DMLR) Workshop at ICML 2024
RollingQ: Reviving the Cooperation Dynamics in Multimodal Transformer
机构 * Gaoling School of Artificial Intelligence Renmin University of China（中国人民大学人工智能学院） ； Beihang University（北京航空航天大学） ； Xiamen University（厦门大学） ； Beijing Key Laboratory of Research on Large Models（北京市大模型研究关键实验室） ； Engineering Research Center of Next-Generation Intelligent Search（下一代智能搜索工程研究中心） ； Tencent（腾讯）
Comments Accepted by ICML 2025
GaussMarker: Robust Dual-Domain Watermark for Diffusion Models
机构 * Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所） ； School of Artificial Intelligence, University of Chinese Academy of Sciences（中国科学院大学人工智能学院）
Comments Accepted at ICML 2025
Multimodal Modeling of CRISPR-Cas12 Activity Using Foundation Models and Chromatin Accessibility Data
Comments This manuscript has been accepted by ICML workshop 2025
Targeted control of fast prototyping through domain-specific interface
机构 * Department of Advanced Manufacturing and Robotics, College of Engineering, Peking University, Beijing, China（工程学院先进制造与机器人系，北京大学） ； Department of Computer Science and Engineering, The Hong Kong University of Science and Technology, Hong Kong SAR（计算机科学与工程系，香港科学与技术大学） ； School of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan, China（计算机科学与技术学院，华中科技大学） ； Division of Emerging Interdisciplinary Areas, The Hong Kong University of Science and Technology, Hong Kong SAR（新兴交叉学科领域，香港科学与技术大学）
Comments In International Conference on Machine Learning (ICML'25)
Diffuse Everything: Multimodal Diffusion Models on Arbitrary State Spaces
机构 * Machine Learning Center, Georgia Institute of Technology, Atlanta, GA ； School of Mathematics, Georgia Institute of Technology, Atlanta, GA ； Department of Mathematics \& Statistics, SUNY Albany, NY
Comments Accepted to ICML 2025. Code available at https://github.com/KevinRojas1499/Diffuse-Everything
Is Your LLM-Based Multi-Agent a Reliable Real-World Planner? Exploring Fraud Detection in Travel Planning
Comments Accepted by ICML 2025 Workshop MAS
Enhancing Cooperative Multi-Agent Reinforcement Learning with State Modelling and Adversarial Exploration
机构 * School of Electrical ； Archimedes, Athena Research Center, Greece ； Department of Mathematics, ETH Zurich ； Dept. of Mechanical Engineering \& Materials Science, Duke University ； Dept. of Digital Systems, University of Piraeus, Greece
Comments Accepted at ICML 2025
The Sharpness Disparity Principle in Transformers for Accelerating Language Model Pre-Training
机构 * School of Mathematical Sciences, Peking University（北京大学数学科学学院） ； Center for Machine Learning Research, Peking University（北京大学机器学习研究中心） ； Shanghai Jiao Tong University（上海交通大学） ； AI for Science Institute, Beijing, China（北京人工智能科学研究院）
Comments 21 pages, accepted by ICML 2025
Understanding the Emergence of Multimodal Representation Alignment
机构 * Massachusetts Institute of Technology, USA（麻省理工学院） ； NTT Research, USA（NTT研究）
Comments To appear as a poster in ICML 2025. 21 pages, 22 figures, 3 tables
Near-Optimal Sample Complexity for MDPs via Anchoring
Journal ref International Conference on Machine Learning 2025
PANDAS: Improving Many-shot Jailbreaking via Positive Affirmation, Negative Demonstration, and Adaptive Sampling
机构 * University of Toronto, Vector Institute（多伦多大学，向量研究所） ； University of Oxford（牛津大学） ； Polytechnique Montréal, Mila - Quebec AI Institute, University of Toronto（蒙特利尔理工学院，魁北克人工智能研究所，多伦多大学）
Comments Accepted at ICML 2025 (Spotlight). Code: https://github.com/averyma/pandas
Tensor-Var: Efficient Four-Dimensional Variational Data Assimilation
机构 * University College London（伦敦大学学院）
Journal ref Proceedings of the Forty-second International Conference on Machine Learning (ICML), 2025
T1: Advancing Language Model Reasoning through Reinforcement Learning and Inference Scaling
机构 * Tsinghua University（清华大学）
Comments Accepted to ICML 2025
Entropy Controllable Direct Preference Optimization
机构 * The University of Tokyo（东京大学）
Comments ICML 2025 Workshop on Models of Human Feedback for AI Alignment
Taming Rectified Flow for Inversion and Editing
机构 * Tsinghua University（清华大学） ； ARC Lab, Tencent PCG（腾讯ARC实验室）
Comments ICML 2025; GitHub: https://github.com/wangjiangshan0725/RF-Solver-Edit
Persistent Topological Features in Large Language Models
机构 * Area Science Park, Trieste, Italy（特雷斯特科学公园） ； University of Trieste, Trieste, Italy（特雷斯特大学） ； University of Amsterdam, Amsterdam, the Netherlands（阿姆斯特丹大学）
Comments 10+17 pages, 17 figures, 3 tables. Accepted as poster at ICML 2025
AssistanceZero: Scalably Solving Assistance Games
机构 * ucb（加州大学伯克利分校）
Comments Presented at ICML 2025
Latent Action Learning Requires Supervision in the Presence of Distractors
机构 * Innopolis University（因诺波利斯大学） ； Research Center for Trusted Artificial Intelligence, ISP RAS（可信人工智能研究所以及ISP俄罗斯科学院）
Comments ICML 2025, Poster, Project Page: https://laom.dunnolab.ai/, Source code: https://github.com/dunnolab/laom
Hessian Geometry of Latent Space in Generative Models
机构 * Artificial Neural Computing Corp.（人工神经计算公司） ； School of Digital Technologies, Tallinn University（数字技术学院，塔林大学）
Comments ICML 2025
Non-stationary Online Learning for Curved Losses: Improved Dynamic Regret via Mixability
机构 * National Key Laboratory for Novel Software Technology, Nanjing University, China（新型软件技术国家重点实验室，南京大学，中国） ； The University of Tokyo, Japan（东京大学，日本）
Comments ICML 2025
FicGCN: Unveiling the Homomorphic Encryption Efficiency from Irregular Graph Convolutional Networks
Comments Accepted by ICML 2025
Interior-Point Vanishing Problem in Semidefinite Relaxations for Neural Network Verification
机构 * Institute of Science Tokyo（东京科学研究所） ； Keio University（庆应大学）
Comments 17 pages, 2 figures. Version revised after ICML 2025 reviews
Fundamental Limits of Learning High-dimensional Simplices in Noisy Regimes
Comments Extension of our ICML 2023 paper, 44 pages
