Distilling Policy Distillation

TL;DR

This paper provides a comprehensive analysis of policy distillation in Deep Reinforcement Learning, comparing various methods and introducing an expected entropy regularized approach that enhances learning speed and guarantees convergence.

Contribution

It offers a rigorous theoretical and empirical comparison of policy distillation variants and proposes a new method that improves learning efficiency and convergence guarantees.

Findings

Expected entropy regularized distillation speeds up learning.

Three distillation techniques are recommended based on task specifics.

The new method guarantees convergence while improving performance.

Abstract

The transfer of knowledge from one policy to another is an important tool in Deep Reinforcement Learning. This process, referred to as distillation, has been used to great success, for example, by enhancing the optimisation of agents, leading to stronger performance faster, on harder domains [26, 32, 5, 8]. Despite the widespread use and conceptual simplicity of distillation, many different formulations are used in practice, and the subtle variations between them can often drastically change the performance and the resulting objective that is being optimised. In this work, we rigorously explore the entire landscape of policy distillation, comparing the motivations and strengths of each variant through theoretical and empirical analysis. Our results point to three distillation techniques, that are preferred depending on specifics of the task. Specifically a newly proposed expected entropy regularised distillation allows for quicker learning in a wide range of situations, while still guaranteeing convergence.