Chapter 3 - Generative adversarial networks and their variants

Abstract

Nowadays, generative adversarial networks (GANs) are one of the most exciting approaches in computer science. It is an algorithmic designs that utilize two neural networks, i.e., generator and discriminator, setting one in opposition to the next that's why adversarial to produce new content, synthetic instances of information that can pass for genuine data. This ability of GANs looks a little bit of magic, at least at first sight. The primary objective of GANs is to produce information from the beginning, consider mostly pictures but from other areas which also include music. Most commonly, they are utilized in voice generation, image generation, video generation, and many more. By and large, GANs are model design for preparing a generative model and it is generally necessary to utilize deep-learning models in this design. Some variants of GAN are Vari GAN, Temporal GAN (TGAN), Laplacian GAN (LAPGAN), VGAN, SRGAN, and FCGAN. This chapter provides an introduction to GANs, deep-learning methods with an overview of some variants and applications that have benefited from them.