A Brief History of Generative Adversarial Networks

Abstract

Generative Adversarial Networks (GANs) are a class of Deep Neural Networks that have the ability to generate realistic synthetic data including but not limited to images, text, or even videos. The ability to learn and generate new data with the same statistics as the training set has made it the de-facto standard for generative networks, making older architectures (like autoencoders) obsolete. This paper investigates GANs from the ground-up, analysing various architectures , their advantages, caveats, and use cases in pursuit of understanding its widespread success.