ABC Reinforcement Learning

TL;DR

This paper presents ABC-RL, a likelihood-free Bayesian reinforcement learning framework that leverages simulators and Approximate Bayesian Computation, enabling Bayesian methods without explicit probabilistic models.

Contribution

It introduces ABC-RL, a novel approach that extends Bayesian reinforcement learning to complex models using simulation and likelihood-free inference techniques.

Findings

Demonstrates ABC-RL's effectiveness compared to LSPI

Shows ABC is theoretically sound with non-sufficient statistics

Enables Bayesian RL without explicit probabilistic models

Abstract

This paper introduces a simple, general framework for likelihood-free Bayesian reinforcement learning, through Approximate Bayesian Computation (ABC). The main advantage is that we only require a prior distribution on a class of simulators (generative models). This is useful in domains where an analytical probabilistic model of the underlying process is too complex to formulate, but where detailed simulation models are available. ABC-RL allows the use of any Bayesian reinforcement learning technique, even in this case. In addition, it can be seen as an extension of rollout algorithms to the case where we do not know what the correct model to draw rollouts from is. We experimentally demonstrate the potential of this approach in a comparison with LSPI. Finally, we introduce a theorem showing that ABC is a sound methodology in principle, even when non-sufficient statistics are used.