Optimizing Stochastic Policy Geometry for Few-Step Flow Reinforcement Learning
Geometry-Optimized Few-Step Reinforcement Learning is introduced, built on a selection principle: at a fixed conditioning state and timestep pair, choose the admissible policy that minimizes the conditional KL divergence induced by any given nonzero velocity-output update.
Zhuguanyu Wu, Cong-Jia Chen, Rui-Hao Gong et al.
· 0 citations