Model-based RL with Optimistic Posterior Sampling: Structural Conditions and Sample Complexity
This work provides a unified theoretical framework for model-based RL algorithms, addressing sample efficiency challenges in reinforcement learning.
The authors tackled the problem of designing posterior sampling methods for model-based reinforcement learning by proposing a general framework that reduces regret to Hellinger distance in conditional probability estimation. They demonstrated that optimistic posterior sampling can control this distance using data likelihood, achieving state-of-the-art sample complexity guarantees across various model-based RL settings.
We propose a general framework to design posterior sampling methods for model-based RL. We show that the proposed algorithms can be analyzed by reducing regret to Hellinger distance in conditional probability estimation. We further show that optimistic posterior sampling can control this Hellinger distance, when we measure model error via data likelihood. This technique allows us to design and analyze unified posterior sampling algorithms with state-of-the-art sample complexity guarantees for many model-based RL settings. We illustrate our general result in many special cases, demonstrating the versatility of our framework.