Selectively Sharing Experiences Improves Multi-Agent Reinforcement Learning
This addresses the challenge of decentralized training in multi-agent systems, offering a robust performance uplift, though it is incremental as it builds on existing multi-agent RL methods.
The paper tackles the problem of inefficient learning in multi-agent reinforcement learning by introducing a selective experience sharing method, which outperforms baseline and state-of-the-art algorithms by sharing only a limited number of relevant transitions.
We present a novel multi-agent RL approach, Selective Multi-Agent Prioritized Experience Relay, in which agents share with other agents a limited number of transitions they observe during training. The intuition behind this is that even a small number of relevant experiences from other agents could help each agent learn. Unlike many other multi-agent RL algorithms, this approach allows for largely decentralized training, requiring only a limited communication channel between agents. We show that our approach outperforms baseline no-sharing decentralized training and state-of-the art multi-agent RL algorithms. Further, sharing only a small number of highly relevant experiences outperforms sharing all experiences between agents, and the performance uplift from selective experience sharing is robust across a range of hyperparameters and DQN variants. A reference implementation of our algorithm is available at https://github.com/mgerstgrasser/super.