Off-Policy Actor-Critic with Shared Experience Replay

Simon Schmitt, Matteo Hessel, Karen Simonyan

Keywords: Reinforcement Learning, Off-Policy Learning, Experience Replay

TL;DR: We investigate and propose solutions for two challenges in reinforcement learning: (a) efficient actor-critic learning with experience replay (b) stability of very off-policy learning.

Abstract:

We investigate the combination of actor-critic reinforcement learning algorithms with uniform large-scale experience replay and propose solutions for two challenges: (a) efficient actor-critic learning with experience replay (b) stability of very off-policy learning. We employ those insights to accelerate hyper-parameter sweeps in which all participating agents run concurrently and share their experience via a common replay module.

To this end we analyze the bias-variance tradeoffs in V-trace, a form of importance sampling for actor-critic methods. Based on our analysis, we then argue for mixing experience sampled from replay with on-policy experience, and propose a new trust region scheme that scales effectively to data distributions where V-trace becomes unstable.

We provide extensive empirical validation of the proposed solution. We further show the benefits of this setup by demonstrating state-of-the-art data efficiency on Atari among agents trained up until 200M environment frames.