Graph-Attention Communication for Decentralized Multi-Agent Reinforcement Learning
A learned communication framework for decentralized multi-agent reinforcement learning on dynamic networks: a graph attention network layer inside a recurrent message-passing scheme lets each agent aggregate neighbours' messages by learned attention weights, and a multi-round targeting mechanism decides whom to communicate with. Trained end-to-end with reinforcement learning alone in a sparse-reward packet-routing environment subject to node failures, it improves routing reward while cutting communication overhead versus a non-attention baseline.
Dynamic Graph Communication for Decentralised Multi-Agent Reinforcement Learning
This work presents a novel communication framework for decentralized multi-agent systems operating in dynamic network environments. Integrated into a multi-agent reinforcement learning system, the fr…