Provably Efficient Multi-Agent Reinforcement Learning with Fully Decentralized Communication

TL;DR

This paper demonstrates that fully decentralized communication among multiple agents in reinforcement learning can significantly improve learning efficiency and reduce regret, with theoretical bounds and numerical validation.

Contribution

It provides provable regret and sample complexity bounds for decentralized multi-agent Q-learning with limited communication, highlighting the benefits of increased agents and information sharing.

Findings

Communication improves group performance and reduces regret.

More agents and information sharing accelerate convergence.

Theoretical bounds depend on network structure and communication range.

Abstract

A challenge in reinforcement learning (RL) is minimizing the cost of sampling associated with exploration. Distributed exploration reduces sampling complexity in multi-agent RL (MARL). We investigate the benefits to performance in MARL when exploration is fully decentralized. Specifically, we consider a class of online, episodic, tabular $Q$-learning problems under time-varying reward and transition dynamics, in which agents can communicate in a decentralized manner.We show that group performance, as measured by the bound on regret, can be significantly improved through communication when each agent uses a decentralized message-passing protocol, even when limited to sending information up to its $\gamma$-hop neighbors. We prove regret and sample complexity bounds that depend on the number of agents, communication network structure and $\gamma.$ We show that incorporating more agents and more information sharing into the group learning scheme speeds up convergence to the optimal policy. Numerical simulations illustrate our results and validate our theoretical claims.