Efficient Bayes-Adaptive Reinforcement Learning using Sample-Based Search

TL;DR

This paper introduces a sample-based Monte-Carlo tree search method for efficient approximate Bayes-optimal reinforcement learning, significantly improving performance on benchmark problems and enabling exploration in infinite state spaces.

Contribution

It presents a tractable, sample-based approach that avoids costly Bayes rule computations, outperforming previous Bayesian RL algorithms in various domains.

Findings

Outperformed prior Bayesian RL algorithms on benchmark problems

Enabled exploration in infinite state space domains

Reduced computational complexity by lazy sampling of models

Abstract

Bayesian model-based reinforcement learning is a formally elegant approach to learning optimal behaviour under model uncertainty, trading off exploration and exploitation in an ideal way. Unfortunately, finding the resulting Bayes-optimal policies is notoriously taxing, since the search space becomes enormous. In this paper we introduce a tractable, sample-based method for approximate Bayes-optimal planning which exploits Monte-Carlo tree search. Our approach outperformed prior Bayesian model-based RL algorithms by a significant margin on several well-known benchmark problems -- because it avoids expensive applications of Bayes rule within the search tree by lazily sampling models from the current beliefs. We illustrate the advantages of our approach by showing it working in an infinite state space domain which is qualitatively out of reach of almost all previous work in Bayesian exploration.