Self-Play and Fictitious Play
Chenning Xu · Applied and Computational Engineering · 2023
As one of the most important algorithms, Fictitious Play lays the foundation of improving agents’ performance by anticipating adversaries’ strategies and making coping strategies. However, few applications of Fictitious Play have made due to its inefficiency when handling massive data. Recently techniques including Model-Based Reinforcement Learning and Q-learning pave way for enhancement of Fictitious Play. Different algorithms have been proposed in order to improve agents’ performance and efficiency based on Fictitious Play in the past few decades. This paper firstly summarizes Fictitious Play and other algorithms, followed by discussing some of the main variants based on Fictitious Play. Analysis of defects, including robustness and excessive calculation, are also presented in the paper.