On the Convergence and Sample Efficiency of Variance-Reduced Policy Gradient MethodDownload PDF

21 May 2021, 20:42 (edited 19 Oct 2021)NeurIPS 2021 SpotlightReaders: Everyone
  • Keywords: Reinforcement Learning, Policy Gradient Method, Variance Reduction, Gradient Truncation
  • TL;DR: We derive a novel form of variance reduced policy gradient method for solving RL beyond cumulative reward.
  • Abstract: Policy gradient (PG) gives rise to a rich class of reinforcement learning (RL) methods. Recently, there has been an emerging trend to augment the existing PG methods such as REINFORCE by the \emph{variance reduction} techniques. However, all existing variance-reduced PG methods heavily rely on an uncheckable importance weight assumption made for every single iteration of the algorithms. In this paper, a simple gradient truncation mechanism is proposed to address this issue. Moreover, we design a Truncated Stochastic Incremental Variance-Reduced Policy Gradient (TSIVR-PG) method, which is able to maximize not only a cumulative sum of rewards but also a general utility function over a policy's long-term visiting distribution. We show an $\tilde{\mathcal{O}}(\epsilon^{-3})$ sample complexity for TSIVR-PG to find an $\epsilon$-stationary policy. By assuming the \emph{overparameterization} of policy and exploiting the \emph{hidden convexity} of the problem, we further show that TSIVR-PG converges to global $\epsilon$-optimal policy with $\tilde{\mathcal{O}}(\epsilon^{-2})$ samples.
  • Supplementary Material: pdf
  • Code Of Conduct: I certify that all co-authors of this work have read and commit to adhering to the NeurIPS Statement on Ethics, Fairness, Inclusivity, and Code of Conduct.
  • Code: zip
10 Replies

Loading