Off-Policy Deep Reinforcement Learning without Exploration

Scott Fujimoto, David Meger, Doina Precup
Proceedings of the 36th International Conference on Machine Learning, PMLR 97:2052-2062, 2019.

Abstract

Many practical applications of reinforcement learning constrain agents to learn from a fixed batch of data which has already been gathered, without offering further possibility for data collection. In this paper, we demonstrate that due to errors introduced by extrapolation, standard off-policy deep reinforcement learning algorithms, such as DQN and DDPG, are incapable of learning with data uncorrelated to the distribution under the current policy, making them ineffective for this fixed batch setting. We introduce a novel class of off-policy algorithms, batch-constrained reinforcement learning, which restricts the action space in order to force the agent towards behaving close to on-policy with respect to a subset of the given data. We present the first continuous control deep reinforcement learning algorithm which can learn effectively from arbitrary, fixed batch data, and empirically demonstrate the quality of its behavior in several tasks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v97-fujimoto19a, title = {Off-Policy Deep Reinforcement Learning without Exploration}, author = {Fujimoto, Scott and Meger, David and Precup, Doina}, booktitle = {Proceedings of the 36th International Conference on Machine Learning}, pages = {2052--2062}, year = {2019}, editor = {Chaudhuri, Kamalika and Salakhutdinov, Ruslan}, volume = {97}, series = {Proceedings of Machine Learning Research}, month = {09--15 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v97/fujimoto19a/fujimoto19a.pdf}, url = {https://proceedings.mlr.press/v97/fujimoto19a.html}, abstract = {Many practical applications of reinforcement learning constrain agents to learn from a fixed batch of data which has already been gathered, without offering further possibility for data collection. In this paper, we demonstrate that due to errors introduced by extrapolation, standard off-policy deep reinforcement learning algorithms, such as DQN and DDPG, are incapable of learning with data uncorrelated to the distribution under the current policy, making them ineffective for this fixed batch setting. We introduce a novel class of off-policy algorithms, batch-constrained reinforcement learning, which restricts the action space in order to force the agent towards behaving close to on-policy with respect to a subset of the given data. We present the first continuous control deep reinforcement learning algorithm which can learn effectively from arbitrary, fixed batch data, and empirically demonstrate the quality of its behavior in several tasks.} }
Endnote
%0 Conference Paper %T Off-Policy Deep Reinforcement Learning without Exploration %A Scott Fujimoto %A David Meger %A Doina Precup %B Proceedings of the 36th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2019 %E Kamalika Chaudhuri %E Ruslan Salakhutdinov %F pmlr-v97-fujimoto19a %I PMLR %P 2052--2062 %U https://proceedings.mlr.press/v97/fujimoto19a.html %V 97 %X Many practical applications of reinforcement learning constrain agents to learn from a fixed batch of data which has already been gathered, without offering further possibility for data collection. In this paper, we demonstrate that due to errors introduced by extrapolation, standard off-policy deep reinforcement learning algorithms, such as DQN and DDPG, are incapable of learning with data uncorrelated to the distribution under the current policy, making them ineffective for this fixed batch setting. We introduce a novel class of off-policy algorithms, batch-constrained reinforcement learning, which restricts the action space in order to force the agent towards behaving close to on-policy with respect to a subset of the given data. We present the first continuous control deep reinforcement learning algorithm which can learn effectively from arbitrary, fixed batch data, and empirically demonstrate the quality of its behavior in several tasks.
APA
Fujimoto, S., Meger, D. & Precup, D.. (2019). Off-Policy Deep Reinforcement Learning without Exploration. Proceedings of the 36th International Conference on Machine Learning, in Proceedings of Machine Learning Research 97:2052-2062 Available from https://proceedings.mlr.press/v97/fujimoto19a.html.

Related Material