A Review on Generative Adversarial Networks

Abstract

GenerativeAdversarial Networks (GAN) is currently one of the hottest subjects in the field of Artificial Intelligence; it has a significant impact on the development of generative models. The excellence of GAN is that it is based on zero-sum game theory and has a generator as well as a discriminator that optimize each other and finally receive the optimal result. In recent years, many different types of GAN optimization models have emerged, which can be classified by the different structure of their generators and discriminators. Since most of the experiments of the models are conducted on the datasets of MNIST, SVHN, CIFAR10, etc., the performance of each model on those datasets is evaluated. Then some of the applications and the methods of optimizing the models of GAN are explained. Finally, we propose challenges that GAN faces and the prospect of GAN.