A Survey on the Applications of Generative Adversarial Networks

Generative Adversarial Networks are computational structures consisting of two neural networks, i.e., a generator network and a discriminator network. The networks compete against each other to create new and synthetic data instances (thus the term "adversarial"). The Generator network takes random noise as input, transforms and reshapes it into a recognizable structure using a differential function. The generator's output appears to be an actual data point. When the generator network runs with various input noise levels, it creates multiple realistic output data samples. The purpose of these created data samples is to be representative of the real-world data distribution. In addition, GANs employ an approximation in which the generator network is guided by a second network called the Discriminator to produce samples from the probability distribution of input data. The Discriminator is a standard neural network classifier that separates the genuine examples from the generator's bogus samples. Thus, it helps in identifying whether the given image is real or fake. Some of the GAN applications discovered are 3D object production, image processing, pandemics, face detection, medicine, and traffic control. This paper provides a systematic study and analysis of the recent GAN model and its applications.