Multi-Agent Cooperation

2026-02-20

Building upon Google's Transformer architecture, researchers proposed a method for training sequence-model agents against diverse opponents. This approach enables agents to adapt within each game without relying on hardcoded assumptions about how other agents learn.