Model-based Offline Reinforcement Learning with Count-based Conservatism

Byeongchan Kim, Min-Hwan Oh
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:16728-16746, 2023.

Abstract

In this paper, we present a model-based offline reinforcement learning method that integrates count-based conservatism, named $\texttt{Count-MORL}$. Our method utilizes the count estimates of state-action pairs to quantify model estimation error, marking the first algorithm of demonstrating the efficacy of count-based conservatism in model-based offline deep RL to the best of our knowledge. For our proposed method, we first show that the estimation error is inversely proportional to the frequency of state-action pairs. Secondly, we demonstrate that the learned policy under the count-based conservative model offers near-optimality performance guarantees. Through extensive numerical experiments, we validate that $\texttt{Count-MORL}$ with hash code implementation significantly outperforms existing offline RL algorithms on the D4RL benchmark datasets. The code is accessible at https://github.com/oh-lab/Count-MORL.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-kim23q, title = {Model-based Offline Reinforcement Learning with Count-based Conservatism}, author = {Kim, Byeongchan and Oh, Min-Hwan}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {16728--16746}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/kim23q/kim23q.pdf}, url = {https://proceedings.mlr.press/v202/kim23q.html}, abstract = {In this paper, we present a model-based offline reinforcement learning method that integrates count-based conservatism, named $\texttt{Count-MORL}$. Our method utilizes the count estimates of state-action pairs to quantify model estimation error, marking the first algorithm of demonstrating the efficacy of count-based conservatism in model-based offline deep RL to the best of our knowledge. For our proposed method, we first show that the estimation error is inversely proportional to the frequency of state-action pairs. Secondly, we demonstrate that the learned policy under the count-based conservative model offers near-optimality performance guarantees. Through extensive numerical experiments, we validate that $\texttt{Count-MORL}$ with hash code implementation significantly outperforms existing offline RL algorithms on the D4RL benchmark datasets. The code is accessible at https://github.com/oh-lab/Count-MORL.} }
Endnote
%0 Conference Paper %T Model-based Offline Reinforcement Learning with Count-based Conservatism %A Byeongchan Kim %A Min-Hwan Oh %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-kim23q %I PMLR %P 16728--16746 %U https://proceedings.mlr.press/v202/kim23q.html %V 202 %X In this paper, we present a model-based offline reinforcement learning method that integrates count-based conservatism, named $\texttt{Count-MORL}$. Our method utilizes the count estimates of state-action pairs to quantify model estimation error, marking the first algorithm of demonstrating the efficacy of count-based conservatism in model-based offline deep RL to the best of our knowledge. For our proposed method, we first show that the estimation error is inversely proportional to the frequency of state-action pairs. Secondly, we demonstrate that the learned policy under the count-based conservative model offers near-optimality performance guarantees. Through extensive numerical experiments, we validate that $\texttt{Count-MORL}$ with hash code implementation significantly outperforms existing offline RL algorithms on the D4RL benchmark datasets. The code is accessible at https://github.com/oh-lab/Count-MORL.
APA
Kim, B. & Oh, M.. (2023). Model-based Offline Reinforcement Learning with Count-based Conservatism. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:16728-16746 Available from https://proceedings.mlr.press/v202/kim23q.html.

Related Material