How do multi-agent systems use unsupervised learning for coordination?

Multi-agent systems use unsupervised learning to discover coordination strategies without explicit rewards or labels.

In multi-agent environments, agents often lack centralized supervision. Unsupervised objectives such as mutual information maximization, agreement learning, or self-prediction allow agents to learn coordination patterns. Techniques like Independent Component Learning or emergent communication protocols enable agents to develop shared representations. This is crucial in swarm robotics, distributed AI systems, and decentralized decision-making.