Distributional Model Equivalence for Risk-Sensitive Reinforcement Learning

Tyler Kastner, Murat A Erdogdu, Amir-massoud Farahmand

Keywords: Reinforcement learning, Risk-Sensitive Reinforcement Learning, Model-Based Reinforcement Learning, Distributional Reinforcement Learning

TL;DR: We combine distributional reinforcement learning with decision-aware model learning to design models which can plan for risk-sensitive objectives.

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 models can be used to augment any model-free risk-sensitive algorithm, and provide both tabular and large-scale experiments to demonstrate our method’s ability.

Submission Number: 12460