Multi-agent Reinforcement Learning and Its Application to Wireless Network Communication

Real-world problems often involve many players interacting with each other while pursuing their own goals. Although the topic of Machine Learning (ML) is making great progress in picturing good learning strategies, there is a lack of consideration of many player strategies. Thus, the topic of Multi-Agent Reinforcement Learning (MARL) attempts to address this need. Here, the learning strategies of Reinforcement Learning (RL), where one agent interacts with an environment to solve a task, are extended to multiple players trying to solve their own tasks. To do so, algorithms from RL are used and applied to multi-agent settings, coupled with game-theoretic aspects to predict the multi-player behaviour. After explaining the main concepts of MARL with their challenges and advantages, we apply MARL to a setting of wireless communication. Here, the multi-agent setting can take into account the different number of communicating devices that communicate in real-world communication scenarios.