Learning Multi-Agent Communication Protocol: Study on Information Entropy Efficiency in MARL
Researchers have introduced a new metric called the Information Entropy Efficiency Index (IEI) to evaluate the efficiency of communication protocols in multi-agent reinforcement learning (MARL). This metric quantifies the ratio between message entropy and task performance, aiming to encourage agents to learn more compact and efficient message representations. Experiments show that incorporating IEI into training can lead to equivalent or superior task performance with improved communication efficiency, challenging the notion that higher performance necessitates increased communication overhead. AI
IMPACT Introduces a novel metric for evaluating communication efficiency in MARL, potentially enabling more scalable multi-agent systems.