Learning Complex Policies in Groups via Peer-to-Peer Action Recommendations

Peer learning enables a group of reinforcement learning agents to simultaneously learn complex policies by exchanging action recommendations, outperforming single-agent learning and state-of-the-art action advising baselines.