
The author proposes the DiffCPS algorithm to address the limited expressivity problem in offline reinforcement learning using diffusion-based policies and a primal-dual method.


coremsg

DiffCPS-Diffusion-based-Constrained-Policy-Search-for-Offline-Reinforcement-Learning


DiffCPS: Diffusion-based Constrained Policy Search for Offline Reinforcement Learning


title_rewrite


DiffCPS löst das Problem der begrenzten Ausdrucksfähigkeit in der gewichteten Regression durch das Diffusionsmodell.
