Slides

Short Abstract: In this talk, we consider the setting of generative model, where we can directly get samples from any state-action pair. An improved analysis based on absorbing state is presented, which is novel and might be extended to the other settings like online RL.

Reference: http://proceedings.mlr.press/v125/agarwal20b/agarwal20b.pdf http://www.liziniu.org/docs/rl-generative-model.pdf