BRExIt: On Opponent Modelling in Expert Iteration

TL;DR

BRExIt enhances expert iteration in multi-agent learning by integrating opponent modeling, leading to faster and more effective policy learning against fixed agents.

Contribution

It introduces BRExIt, a novel method that incorporates opponent models into expert iteration to improve policy learning in game settings.

Findings

BRExIt outperforms standard Expert Iteration in empirical tests.

Opponent modeling improves feature shaping and policy accuracy.

BRExIt converges faster to high-performing policies.

Abstract

Finding a best response policy is a central objective in game theory and multi-agent learning, with modern population-based training approaches employing reinforcement learning algorithms as best-response oracles to improve play against candidate opponents (typically previously learnt policies). We propose Best Response Expert Iteration (BRExIt), which accelerates learning in games by incorporating opponent models into the state-of-the-art learning algorithm Expert Iteration (ExIt). BRExIt aims to (1) improve feature shaping in the apprentice, with a policy head predicting opponent policies as an auxiliary task, and (2) bias opponent moves in planning towards the given or learnt opponent model, to generate apprentice targets that better approximate a best response. In an empirical ablation on BRExIt's algorithmic variants against a set of fixed test agents, we provide statistical evidence that BRExIt learns better performing policies than ExIt.