Distributional Model Equivalence for Risk-Sensitive Reinforcement Learning

TL;DR

This paper explores how traditional model equivalence in reinforcement learning fails for risk-sensitive planning and introduces distributional approaches to define new, flexible model equivalences suitable for various risk measures.

Contribution

It introduces two new notions of model equivalence based on distributional reinforcement learning, enabling risk-sensitive planning for any risk measure and providing practical algorithms.

Findings

Distributional model equivalence can be used for risk-sensitive planning.

The proposed methods improve risk-sensitive reinforcement learning performance.

Framework is validated through tabular and large-scale experiments.

Abstract

We consider the problem of learning models for risk-sensitive reinforcement learning. We theoretically demonstrate that proper value equivalence, a method of learning models which can be used to plan optimally in the risk-neutral setting, is not sufficient to plan optimally in the risk-sensitive setting. We leverage distributional reinforcement learning to introduce two new notions of model equivalence, one which is general and can be used to plan for any risk measure, but is intractable; and a practical variation which allows one to choose which risk measures they may plan optimally for. We demonstrate how our framework can be used to augment any model-free risk-sensitive algorithm, and provide both tabular and large-scale experiments to demonstrate its ability.