Off-Policy Actor-Critic with Shared Experience Replay
Simon Schmitt, Matteo Hessel, Karen Simonyan
Abstract
We investigate the combination of actor-critic reinforcement learning algorithms with a uniform large-scale experience replay and propose solutions for two ensuing challenges: (a) efficient actor-critic learning with experience replay (b) the stability of off-policy learning where agents learn from other agents behaviour. 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 solutions on DMLab-30 and further show the benefits of this setup in two training regimes for Atari: (1) a single agent is trained up until 200M environment frames per game (2) a population of agents is trained up until 200M environment frames each and may share experience. We demonstrate state-of-the-art data efficiency among model-free agents in both regimes.
BibTeX
@InProceedings{pmlr-v119-schmitt20a,
title = {Off-Policy Actor-Critic with Shared Experience Replay},
author = {Schmitt, Simon and Hessel, Matteo and Simonyan, Karen},
booktitle = {Proceedings of the 37th International Conference on Machine Learning},
pages = {8545--8554},
year = {2020},
editor = {III, Hal Daumé and Singh, Aarti},
volume = {119},
series = {Proceedings of Machine Learning Research},
month = {13--18 Jul},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v119/schmitt20a/schmitt20a.pdf},
url = {https://proceedings.mlr.press/v119/schmitt20a.html},
abstract = {We investigate the combination of actor-critic reinforcement learning algorithms with a uniform large-scale experience replay and propose solutions for two ensuing challenges: (a) efficient actor-critic learning with experience replay (b) the stability of off-policy learning where agents learn from other agents behaviour. 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 solutions on DMLab-30 and further show the benefits of this setup in two training regimes for Atari: (1) a single agent is trained up until 200M environment frames per game (2) a population of agents is trained up until 200M environment frames each and may share experience. We demonstrate state-of-the-art data efficiency among model-free agents in both regimes.}
}