ECNet
Efficient communication in multi-agent RL via learned communication gates
Presented at BayLearn 2021.
Communication between agents in MARL is expensive. ECNet learns communication gates that allow agents to selectively communicate only when necessary, optimizing the trade-off between task performance and communication cost.
Key result: 75% reduction in inter-agent communication cost without sacrificing task performance.
Stack: Python, PyTorch, multi-agent RL