An Off-policy Policy Gradient Theorem Using Emphatic Weightings
- Ehsan Imani
- Eric Graves
- Martha White, University of Alberta
Policy gradient methods are widely used for control in reinforcement learning, particularly for the continuous action setting. There have been a host of theoretically sound algorithms proposed for the on-policy setting, due to the existence of the policy gradient theorem which provides a simplified form for the gradient. In off-policy learning, however, where the behaviour policy is not necessarily attempting to learn and follow the optimal policy for the given task, the existence of such a theorem has been elusive. In this work, we solve this open problem by providing the first off-policy policy gradient theorem. The key to the derivation is the use of emphatic weightings. We develop a new actor-critic algorithm—called Actor Critic with Emphatic weightings (ACE)—that approximates the simplified gradients provided by the theorem. We demonstrate in a simple counterexample that previous off-policy policy gradient methods—particularly OffPAC and DPG—converge to the wrong solution whereas ACE finds the optimal solution.
Citation
E. Imani, E. Graves, M. White. "An Off-policy Policy Gradient Theorem Using Emphatic Weightings". Neural Information Processing Systems (NIPS), (ed: Samy Bengio, Hanna M. Wallach, Hugo Larochelle, Kristen Grauman, Nicolo Cesa-Bianchi, Roman Garnett), pp 96-106, December 2018.Keywords: | |
Category: | In Conference |
Web Links: | NeurIPS |
BibTeX
@incollection{Imani+al:NIPS18, author = {Ehsan Imani and Eric Graves and Martha White}, title = {An Off-policy Policy Gradient Theorem Using Emphatic Weightings}, Editor = {Samy Bengio, Hanna M. Wallach, Hugo Larochelle, Kristen Grauman, Nicolo Cesa-Bianchi, Roman Garnett}, Pages = {96-106}, booktitle = {Neural Information Processing Systems (NIPS)}, year = 2018, }Last Updated: February 25, 2020
Submitted by Sabina P