Statistics and Samples in Distributional Reinforcement Learning

TL;DR

This paper introduces a unifying statistical framework for distributional reinforcement learning, enabling improved analysis and the development of new algorithms like EDRL and ER-DQN, with demonstrated benefits on various benchmarks.

Contribution

It provides a novel perspective by decomposing DRL algorithms into statistical estimators and distribution imputation methods, leading to new algorithms and enhanced understanding.

Findings

EDRL outperforms existing methods on multiple MDPs

ER-DQN achieves strong results on Atari-57 games

The framework improves analysis and design of DRL algorithms

Abstract

We present a unifying framework for designing and analysing distributional reinforcement learning (DRL) algorithms in terms of recursively estimating statistics of the return distribution. Our key insight is that DRL algorithms can be decomposed as the combination of some statistical estimator and a method for imputing a return distribution consistent with that set of statistics. With this new understanding, we are able to provide improved analyses of existing DRL algorithms as well as construct a new algorithm (EDRL) based upon estimation of the expectiles of the return distribution. We compare EDRL with existing methods on a variety of MDPs to illustrate concrete aspects of our analysis, and develop a deep RL variant of the algorithm, ER-DQN, which we evaluate on the Atari-57 suite of games.