Game of GANs: Game-Theoretical Models for Generative Adversarial Networks

Generative Adversarial Networks (GANs) have recently attracted considerable\nattention in the AI community due to its ability to generate high-quality data\nof significant statistical resemblance to real data. Fundamentally, GAN is a\ngame between two neural networks trained in an adversarial manner to reach a\nzero-sum Nash equilibrium profile. Despite the improvement accomplished in GANs\nin the last few years, several issues remain to be solved. This paper reviews\nthe literature on the game theoretic aspects of GANs and addresses how game\ntheory models can address specific challenges of generative model and improve\nthe GAN's performance. We first present some preliminaries, including the basic\nGAN model and some game theory background. We then present taxonomy to classify\nstate-of-the-art solutions into three main categories: modified game models,\nmodified architectures, and modified learning methods. The classification is\nbased on modifications made to the basic GAN model by proposed game-theoretic\napproaches in the literature. We then explore the objectives of each category\nand discuss recent works in each category. Finally, we discuss the remaining\nchallenges in this field and present future research directions.\n