Model-based Policy Optimization under Approximate Bayesian Inference

Chaoqi Wang, Yuxin Chen, Kevin Murphy
Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, PMLR 238:3250-3258, 2024.

Abstract

Model-based reinforcement learning algorithms (MBRL) present an exceptional potential to enhance sample efficiency within the realm of online reinforcement learning (RL). Nevertheless, a substantial proportion of prevalent MBRL algorithms fail to adequately address the dichotomy of exploration and exploitation. Posterior sampling reinforcement learning (PSRL) emerges as an innovative strategy adept at balancing exploration and exploitation, albeit its theoretical assurances are contingent upon exact inference. In this paper, we show that adopting the same methodology as in exact PSRL can be suboptimal under approximate inference. Motivated by the analysis, we propose an improved factorization for the posterior distribution of polices by removing the conditional independence between the policy and data given the model. By adopting such a posterior factorization, we further propose a general algorithmic framework for PSRL under approximate inference and a practical instantiation of it. Empirically, our algorithm can surpass baseline methods by a significant margin on both dense rewards and sparse rewards tasks from the Deepmind control suite, OpenAI Gym and Metaworld benchmarks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v238-wang24f, title = { Model-based Policy Optimization under Approximate {B}ayesian Inference }, author = {Wang, Chaoqi and Chen, Yuxin and Murphy, Kevin}, booktitle = {Proceedings of The 27th International Conference on Artificial Intelligence and Statistics}, pages = {3250--3258}, 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/wang24f/wang24f.pdf}, url = {https://proceedings.mlr.press/v238/wang24f.html}, abstract = { Model-based reinforcement learning algorithms (MBRL) present an exceptional potential to enhance sample efficiency within the realm of online reinforcement learning (RL). Nevertheless, a substantial proportion of prevalent MBRL algorithms fail to adequately address the dichotomy of exploration and exploitation. Posterior sampling reinforcement learning (PSRL) emerges as an innovative strategy adept at balancing exploration and exploitation, albeit its theoretical assurances are contingent upon exact inference. In this paper, we show that adopting the same methodology as in exact PSRL can be suboptimal under approximate inference. Motivated by the analysis, we propose an improved factorization for the posterior distribution of polices by removing the conditional independence between the policy and data given the model. By adopting such a posterior factorization, we further propose a general algorithmic framework for PSRL under approximate inference and a practical instantiation of it. Empirically, our algorithm can surpass baseline methods by a significant margin on both dense rewards and sparse rewards tasks from the Deepmind control suite, OpenAI Gym and Metaworld benchmarks. } }
Endnote
%0 Conference Paper %T Model-based Policy Optimization under Approximate Bayesian Inference %A Chaoqi Wang %A Yuxin Chen %A Kevin Murphy %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-wang24f %I PMLR %P 3250--3258 %U https://proceedings.mlr.press/v238/wang24f.html %V 238 %X Model-based reinforcement learning algorithms (MBRL) present an exceptional potential to enhance sample efficiency within the realm of online reinforcement learning (RL). Nevertheless, a substantial proportion of prevalent MBRL algorithms fail to adequately address the dichotomy of exploration and exploitation. Posterior sampling reinforcement learning (PSRL) emerges as an innovative strategy adept at balancing exploration and exploitation, albeit its theoretical assurances are contingent upon exact inference. In this paper, we show that adopting the same methodology as in exact PSRL can be suboptimal under approximate inference. Motivated by the analysis, we propose an improved factorization for the posterior distribution of polices by removing the conditional independence between the policy and data given the model. By adopting such a posterior factorization, we further propose a general algorithmic framework for PSRL under approximate inference and a practical instantiation of it. Empirically, our algorithm can surpass baseline methods by a significant margin on both dense rewards and sparse rewards tasks from the Deepmind control suite, OpenAI Gym and Metaworld benchmarks.
APA
Wang, C., Chen, Y. & Murphy, K.. (2024). Model-based Policy Optimization under Approximate Bayesian Inference . Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 238:3250-3258 Available from https://proceedings.mlr.press/v238/wang24f.html.

Related Material