Exploiting Structure in Offline Multi-Agent RL: The Benefits of Low Interaction Rank
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 cooperative and competitive 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.
Cite
Text
Zhan et al. "Exploiting Structure in Offline Multi-Agent RL: The Benefits of Low Interaction Rank." International Conference on Learning Representations, 2025.Markdown
[Zhan et al. "Exploiting Structure in Offline Multi-Agent RL: The Benefits of Low Interaction Rank." International Conference on Learning Representations, 2025.](https://mlanthology.org/iclr/2025/zhan2025iclr-exploiting/)BibTeX
@inproceedings{zhan2025iclr-exploiting,
title = {{Exploiting Structure in Offline Multi-Agent RL: The Benefits of Low Interaction Rank}},
author = {Zhan, Wenhao and Fujimoto, Scott and Zhu, Zheqing and Lee, Jason D. and Jiang, Daniel and Efroni, Yonathan},
booktitle = {International Conference on Learning Representations},
year = {2025},
url = {https://mlanthology.org/iclr/2025/zhan2025iclr-exploiting/}
}