eprintid: 10198782 rev_number: 10 eprint_status: archive userid: 699 dir: disk0/10/19/87/82 datestamp: 2024-10-23 14:17:50 lastmod: 2024-10-23 14:17:50 status_changed: 2024-10-23 14:17:50 type: proceedings_section metadata_visibility: show sword_depositor: 699 creators_name: Ramesh, Shyam Sundhar creators_name: Sessa, Pier Giuseppe creators_name: Hu, Yifan creators_name: Krause, Andreas creators_name: Bogunovic, Ilija title: Distributionally Robust Model-based Reinforcement Learning with Large State Spaces ispublished: pub divisions: UCL divisions: B04 divisions: F46 note: © The Author 2024. Original content in this paper is licensed under the terms of the Creative Commons Attribution 4.0 International (CC BY 4.0) Licence (https://creativecommons.org/licenses/by/4.0/). 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. date: 2024 date_type: published publisher: PMLR official_url: https://proceedings.mlr.press/v238/sundhar-ramesh24a.html oa_status: green full_text_type: pub language: eng primo: open primo_central: open_green verified: verified_manual elements_id: 2282805 lyricists_name: Bogunovic, Ilija lyricists_name: Ramesh, Shyam Sundhar lyricists_id: IBOGU49 lyricists_id: SRAME02 actors_name: Ramesh, Shyam Sundhar actors_id: SRAME02 actors_role: owner full_text_status: public pres_type: paper publication: INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE AND STATISTICS, VOL 238 volume: 238 pagerange: 1-42 pages: 42 event_title: 27th International Conference on Artificial Intelligence and Statistics (AISTATS) event_location: Valencia, Spain event_dates: 2nd-4th May 2024 book_title: Proceedings of The 27th International Conference on Artificial Intelligence and Statistics editors_name: Dasgupta, S editors_name: Mandt, S editors_name: Li, Y citation: Ramesh, Shyam Sundhar; Sessa, Pier Giuseppe; Hu, Yifan; Krause, Andreas; Bogunovic, Ilija; (2024) Distributionally Robust Model-based Reinforcement Learning with Large State Spaces. In: Dasgupta, S and Mandt, S and Li, Y, (eds.) Proceedings of The 27th International Conference on Artificial Intelligence and Statistics. (pp. pp. 1-42). PMLR Green open access document_url: https://discovery.ucl.ac.uk/id/eprint/10198782/1/sundhar-ramesh24a.pdf