Learning to Communicate in Multi-Agent Reinforcement Learning : A Review
It addresses the problem of improving multi-agent coordination and efficiency in AI systems, but is incremental as it primarily reviews and builds upon existing methods.
This paper reviews algorithms for enabling multiple agents to learn communication strategies through reinforcement learning in partially observable environments, and proposes a novel entropy-based metric for evaluating these strategies.
We consider the issue of multiple agents learning to communicate through reinforcement learning within partially observable environments, with a focus on information asymmetry in the second part of our work. We provide a review of the recent algorithms developed to improve the agents' policy by allowing the sharing of information between agents and the learning of communication strategies, with a focus on Deep Recurrent Q-Network-based models. We also describe recent efforts to interpret the languages generated by these agents and study their properties in an attempt to generate human-language-like sentences. We discuss the metrics used to evaluate the generated communication strategies and propose a novel entropy-based evaluation metric. Finally, we address the issue of the cost of communication and introduce the idea of an experimental setup to expose this cost in cooperative-competitive game.