The Sample Complexity of Policy Learning with Mu-Resets

Gene Li

We study policy-based reinforcement learning under the $\mu$-resets interaction protocol of Kakade and Langford [KL02]. This interaction protocol enables the learner to sample trajec…