Tag
This paper introduces SLIM, a minimal architecture that decouples communication from policy representation in multi-agent reinforcement learning, achieving state-of-the-art performance under bandwidth constraints with minimal degradation.