Beyond the Permutation Symmetry of Transformers: The Role of Rotation for Model Fusion
Comments ICML 2025
期刊&会议
ICML
International Conference on Machine Learning · 会议 · Machine Learning
Gravity-Bench-v1: A Benchmark on Gravitational Physics Discovery for Agents
机构 * University of Toronto（多伦多大学） ； New York University Abu Dhabi（纽约大学阿布扎克分校）
Comments Accepted at ICML 2025
Broadband Ground Motion Synthesis by Diffusion Model with Minimal Condition
Comments Accepted to ICML 2025
TINED: GNNs-to-MLPs by Teacher Injection and Dirichlet Energy Distillation
Comments 17 pages, published as a conference paper at ICML 2025
GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers
Comments 20 pages, 7 figures, 11 tables, Accepted by ICML 2025
Diffusion Sampling Correction via Approximately 10 Parameters
Comments Accepted at ICML 2025
Integer Programming for Generalized Causal Bootstrap Designs
Comments Accepted at ICML 2025
Guarantees of a Preconditioned Subgradient Algorithm for Overparameterized Asymmetric Low-rank Matrix Recovery
Journal ref International Conference on Machine Learning, 2025
A Bayesian Model Selection Criterion for Selecting Pretraining Checkpoints
Comments Accepted as an ICML 2025 paper
Epsilon-VAE: Denoising as Visual Decoding
Comments Accepted to ICML 2025. v2: added comparisons to SD-VAE and more visual results; v3: minor change to title; v4: camera-ready version
Understanding and Mitigating Miscalibration in Prompt Tuning for Vision-Language Models
机构 * Shuoyuan Wang（王舒元） ； Yixuan Li（李奕轩） ； Hongxin Wei（魏宏鑫）
Comments Accepted by ICML 2025
On-Device Collaborative Language Modeling via a Mixture of Generalists and Specialists
Comments Camera-ready version
Journal ref ICML 2025
QMamba: On First Exploration of Vision Mamba for Image Quality Assessment
机构 * Fengbin Guan（冯滨冠） ； Xin Li（李新） ； Zihao Yu（于子浩） ； Yiting Lu（卢艺婷） ； Zhibo Chen（陈志波）
Comments Accepted by ICML 2025
GuardAgent: Safeguard LLM Agents by a Guard Agent via Knowledge-Enabled Reasoning
机构 * University of Georgia（佐治亚大学） ； University of Chicago（芝加哥大学） ； UIUC（伊利诺伊大学香槟分校） ； University of Texas at Austin（德克萨斯大学奥斯汀分校） ； University of California, Berkeley（加州大学伯克利分校） ； Emory University（埃默里大学） ； Virtue AI
Comments Accepted to ICML 2025
Stochastic Variance-Reduced Forward-Reflected-Backward Splitting Methods for Nonmonotone Generalized Equations
Comments 38 pages, 12 figures
Journal ref ICML 2025
Improving Continual Learning Performance and Efficiency with Auxiliary Classifiers
机构 * Warsaw University of Technology（华沙技术大学） ； Computer Vision Center, Barcelona（巴塞罗那计算机视觉中心） ； IDEAS Research Institute（IDEAS研究机构） ； Institute of Artificial Intelligence and Robotics（人工智能与机器人研究所） ； Xi’an Jiaotong University（西安交通大学） ； Nankai University（南开大学）
Comments ICML 2025 (main track poster)
OmniArch: Building Foundation Model For Scientific Computing
Comments ICML 2025
Theoretical guarantees on the best-of-n alignment policy
Comments ICML 2025
Language Agents with Reinforcement Learning for Strategic Play in the Werewolf Game
机构 * Zelai Xu（泽来许） ； Chao Yu（姚朝） ； Fei Fang（方飞） ； Yu Wang（王宇） ； Yi Wu（吴毅）
Comments Published in ICML 2024
Unsupervisedly Learned Representations: Should the Quest be Over?
Comments published at The 6th International Conference on Machine Learning, Optimization and Data Science - LOD 2020
Position: Uncertainty Quantification Needs Reassessment for Large-language Model Agents
Comments Accepted at ICML 2025
Towards the Efficient Inference by Incorporating Automated Computational Phenotypes under Covariate Shift
Journal ref 2025 ICML
Demystifying the Paradox of Importance Sampling with an Estimated History-Dependent Behavior Policy in Off-Policy Evaluation
机构 * Hongyi Zhou（周洪毅） ； Josiah P. Hanna（哈尼安） ； Jin Zhu（朱锦） ； Ying Yang（杨莹） ； Chengchun Shi（石成春）
Comments Accepted by ICML 2025
Position: All Current Generative Fidelity and Diversity Metrics are Flawed
Comments ICML 2025
Synonymous Variational Inference for Perceptual Image Compression
Comments 31 pages, 20 figures. This paper is accepted by Proceedings of the 42nd International Conference on Machine Learning (ICML 2025) Poster
Suitability Filter: A Statistical Framework for Classifier Evaluation in Real-World Deployment Settings
机构 * ETH Zurich（苏黎世联邦理工学院） ； University of Toronto（多伦多大学） ； Vector Institute（向量研究所）
Comments Accepted to ICML 2025
D-Fusion: Direct Preference Optimization for Aligning Diffusion Models with Visually Consistent Samples
机构 * Zijing Hu（Hu Zijing） ； Fengda Zhang（Zhang Fengda） ； Kun Kuang（Kuang Kun）
Comments Accepted to ICML 2025
Representative Language Generation
机构 * Stanford University（斯坦福大学） ； University of Michigan（密歇根大学）
Comments Accepted to ICML 2025
Faster Rates for Private Adversarial Bandits
Comments Accepted to ICML 2025
Hierarchical Reinforcement Learning with Uncertainty-Guided Diffusional Subgoals
Comments ICML 2025
