Understanding the future of Deep Reinforcement Learning from the perspective of Game Theory

Abstract

This paper is to discuss the development of Deep Reinforcement Learning and the future of it from the perspective of Game Theory. The relationship and potential interaction between these two areas are also introduced, especially the optimization method. This paper discusses about the situations both under non-cooperative and cooperative game. Recently, Artificial Intelligence (AI) and Machine Learning (ML) have grasp sufficient attention from various research areas. Deep Reinforcement Learning, as one of the most promising ML methods, enlighten more researchers to devote themselves in this area. However, even such accomplishment could not belie that, for most kinds of real-life problems, DRL is still unable to provide with an optimal strategy. Because unlike the well-Adjusted environment in laboratory, real life problems are not always able to be converted into mathematical problems. Under such circumstances, most of real-life problems have no nominal "optimal solution". Game Theory provides potential solutions to covert "real issues" into "mathematical problems", then it is easier for researchers to handle.