Exploiting Structure in Offline Multi-Agent RL: The Benefits of Low Interaction Rank

TL;DR

This paper introduces the concept of interaction rank in offline multi-agent reinforcement learning, showing that low interaction rank functions improve robustness and enable efficient decentralized learning, supported by theoretical analysis and experiments.

Contribution

It proposes the interaction rank as a structural assumption, demonstrating its benefits for robustness and efficiency in offline MARL, and explores low interaction rank critic architectures.

Findings

Low interaction rank functions are more robust to distribution shifts.

Utilizing low interaction rank function classes enables decentralized, efficient offline MARL.

Experiments show low interaction rank critic architectures outperform standard approaches.

Abstract

We study the problem of learning an approximate equilibrium in the offline multi-agent reinforcement learning (MARL) setting. We introduce a structural assumption -- the interaction rank -- and establish that functions with low interaction rank are significantly more robust to distribution shift compared to general ones. Leveraging this observation, we demonstrate that utilizing function classes with low interaction rank, when combined with regularization and no-regret learning, admits decentralized, computationally and statistically efficient learning in offline MARL. Our theoretical results are complemented by experiments that showcase the potential of critic architectures with low interaction rank in offline MARL, contrasting with commonly used single-agent value decomposition architectures.