Being Optimistic to Be Conservative: Quickly Learning a CVaR Policy

TL;DR

This paper introduces a novel, sample-efficient reinforcement learning algorithm for quickly learning CVaR-optimal policies, which are crucial for risk-sensitive decision-making in high-stakes environments.

Contribution

The paper presents the first optimistic algorithm for CVaR policy learning based on a new distributional Bellman operator, with proven convergence and improved speed over baselines.

Findings

Faster convergence to CVaR-optimal policies in simulated environments.

The novel operator effectively shifts probability mass to the tail, enhancing risk-sensitive learning.

The method applies to both discrete and continuous state spaces.

Abstract

While maximizing expected return is the goal in most reinforcement learning approaches, risk-sensitive objectives such as conditional value at risk (CVaR) are more suitable for many high-stakes applications. However, relatively little is known about how to explore to quickly learn policies with good CVaR. In this paper, we present the first algorithm for sample-efficient learning of CVaR-optimal policies in Markov decision processes based on the optimism in the face of uncertainty principle. This method relies on a novel optimistic version of the distributional Bellman operator that moves probability mass from the lower to the upper tail of the return distribution. We prove asymptotic convergence and optimism of this operator for the tabular policy evaluation case. We further demonstrate that our algorithm finds CVaR-optimal policies substantially faster than existing baselines in several simulated environments with discrete and continuous state spaces.