PAMD: Structured Adaptive Distances for Bisimulation Representations in Visual Reinforcement Learning
PAMD:视觉强化学习中用于双模拟表示的结构化自适应距离
AI总结 针对视觉强化学习中潜在距离选择影响性能的问题,提出PAMD(成对自适应马氏距离),它为测量潜在状态相似性参数化正定成对条件度量,是现有双模拟方法的简单插件,在视觉MuJoCo任务中显著提升了相关算法性能。
Comments 9 pages, 6 figures, plus appendix. Accepted to the 43rd International Conference on Machine Learning (ICML 2026)