Approximating Shapley Explanations in Reinforcement Learning

TL;DR

This paper introduces FastSVERL, a scalable method for approximating Shapley value explanations in reinforcement learning, addressing computational challenges and enabling real-time, interpretable decision-making in complex environments.

Contribution

FastSVERL is a novel, scalable approach that efficiently approximates Shapley explanations specifically tailored for reinforcement learning settings.

Findings

FastSVERL significantly reduces computation time for Shapley explanations.

It effectively handles temporal dependencies and off-policy data.

Enables real-time interpretability in reinforcement learning applications.

Abstract

Reinforcement learning has achieved remarkable success in complex decision-making environments, yet its lack of transparency limits its deployment in practice, especially in safety-critical settings. Shapley values from cooperative game theory provide a principled framework for explaining reinforcement learning; however, the computational cost of Shapley explanations is an obstacle to their use. We introduce FastSVERL, a scalable method for explaining reinforcement learning by approximating Shapley values. FastSVERL is designed to handle the unique challenges of reinforcement learning, including temporal dependencies across multi-step trajectories, learning from off-policy data, and adapting to evolving agent behaviours in real time. FastSVERL introduces a practical, scalable approach for principled and rigorous interpretability in reinforcement learning.