Reproduce “HYGMA: Hypergraph Coordination Networks with Dynamic Grouping for Multi-Agent Reinforcement Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chiqiang Liu, Dazi Li
HYGMA improves multi-agent reinforcement learning by employing dynamic spectral clustering for adaptive agent grouping and hypergraph attention networks for expressive, higher-order information exchange. It achieved superior sample efficiency and performance across challenging cooperative tasks like StarCraft II and Predator-Prey, showing faster convergence and higher success rates compared to existing methods.
Key insights
- Dynamic grouping, based on spectral clustering applied to agent state histories, offers a principled way to automatically identify and adapt coordination patterns in multi-agent systems.
- Hypergraph neural networks provide a more expressive and efficient architecture for modeling complex, higher-order relationships among multiple agents, outperforming traditional pairwise graphs for intricate team coordination.
- Integrating dynamic group formation with hypergraph-based information processing yields a robust, sample-efficient, and interpretable framework for cooperative MARL, capable of revealing emergent group behaviors and roles.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.