Distributionally Robust Model-based Reinforcement Learning with Large State Spaces

Shyam Sundhar Ramesh, Pier Giuseppe Sessa, Yifan Hu, Andreas Krause, Ilija Bogunovic
Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, PMLR 238:100-108, 2024.

Abstract

Three major challenges in reinforcement learning are the complex dynamical systems with large state spaces, the costly data acquisition processes, and the deviation of real-world dynamics from the training environment deployment. To overcome these issues, we study distributionally robust Markov decision processes with continuous state spaces under the widely used Kullback-Leibler, chi-square, and total variation uncertainty sets. We propose a model-based approach that utilizes Gaussian Processes and the maximum variance reduction algorithm to efficiently learn multi-output nominal transition dynamics, leveraging access to a generative model (i.e., simulator). We further demonstrate the statistical sample complexity of the proposed method for different uncertainty sets. These complexity bounds are independent of the number of states and extend beyond linear dynamics, ensuring the effectiveness of our approach in identifying near-optimal distributionally-robust policies. The proposed method can be further combined with other model-free distributionally robust reinforcement learning methods to obtain a near-optimal robust policy. Experimental results demonstrate the robustness of our algorithm to distributional shifts and its superior performance in terms of the number of samples needed.

Cite this Paper

BibTeX
@InProceedings{pmlr-v238-sundhar-ramesh24a, title = { Distributionally Robust Model-based Reinforcement Learning with Large State Spaces }, author = {Sundhar Ramesh, Shyam and Giuseppe Sessa, Pier and Hu, Yifan and Krause, Andreas and Bogunovic, Ilija}, booktitle = {Proceedings of The 27th International Conference on Artificial Intelligence and Statistics}, pages = {100--108}, year = {2024}, editor = {Dasgupta, Sanjoy and Mandt, Stephan and Li, Yingzhen}, volume = {238}, series = {Proceedings of Machine Learning Research}, month = {02--04 May}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v238/sundhar-ramesh24a/sundhar-ramesh24a.pdf}, url = {https://proceedings.mlr.press/v238/sundhar-ramesh24a.html}, abstract = { Three major challenges in reinforcement learning are the complex dynamical systems with large state spaces, the costly data acquisition processes, and the deviation of real-world dynamics from the training environment deployment. To overcome these issues, we study distributionally robust Markov decision processes with continuous state spaces under the widely used Kullback-Leibler, chi-square, and total variation uncertainty sets. We propose a model-based approach that utilizes Gaussian Processes and the maximum variance reduction algorithm to efficiently learn multi-output nominal transition dynamics, leveraging access to a generative model (i.e., simulator). We further demonstrate the statistical sample complexity of the proposed method for different uncertainty sets. These complexity bounds are independent of the number of states and extend beyond linear dynamics, ensuring the effectiveness of our approach in identifying near-optimal distributionally-robust policies. The proposed method can be further combined with other model-free distributionally robust reinforcement learning methods to obtain a near-optimal robust policy. Experimental results demonstrate the robustness of our algorithm to distributional shifts and its superior performance in terms of the number of samples needed. } }
Endnote
%0 Conference Paper %T Distributionally Robust Model-based Reinforcement Learning with Large State Spaces %A Shyam Sundhar Ramesh %A Pier Giuseppe Sessa %A Yifan Hu %A Andreas Krause %A Ilija Bogunovic %B Proceedings of The 27th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2024 %E Sanjoy Dasgupta %E Stephan Mandt %E Yingzhen Li %F pmlr-v238-sundhar-ramesh24a %I PMLR %P 100--108 %U https://proceedings.mlr.press/v238/sundhar-ramesh24a.html %V 238 %X Three major challenges in reinforcement learning are the complex dynamical systems with large state spaces, the costly data acquisition processes, and the deviation of real-world dynamics from the training environment deployment. To overcome these issues, we study distributionally robust Markov decision processes with continuous state spaces under the widely used Kullback-Leibler, chi-square, and total variation uncertainty sets. We propose a model-based approach that utilizes Gaussian Processes and the maximum variance reduction algorithm to efficiently learn multi-output nominal transition dynamics, leveraging access to a generative model (i.e., simulator). We further demonstrate the statistical sample complexity of the proposed method for different uncertainty sets. These complexity bounds are independent of the number of states and extend beyond linear dynamics, ensuring the effectiveness of our approach in identifying near-optimal distributionally-robust policies. The proposed method can be further combined with other model-free distributionally robust reinforcement learning methods to obtain a near-optimal robust policy. Experimental results demonstrate the robustness of our algorithm to distributional shifts and its superior performance in terms of the number of samples needed.
APA
Sundhar Ramesh, S., Giuseppe Sessa, P., Hu, Y., Krause, A. & Bogunovic, I.. (2024). Distributionally Robust Model-based Reinforcement Learning with Large State Spaces . Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 238:100-108 Available from https://proceedings.mlr.press/v238/sundhar-ramesh24a.html.

Related Material