A Study on Generative Adversarial Perturbations Attacks

In last few years Generative model known as Generative adversarial networks (GANs). GANs are architecture to train generative models. GANs uses two models: Generative model and Discriminator model, where Generative model create new images by adding some random noise in existing image and Discriminator model check whether the image is real or fake. In Deep Convolution Neural Network, Generative Adversarial Network is one of the most dynamic analysis, possibilities, and its outstanding image generation capability has received wide attention. In GANs there are two approaches: Generator model and Discriminator model. An Adversarial Networks are classified as Targeted attack and Untargeted attack. This research summarized existing work of the Adversarial Networks from the Generative model and Discriminator model work. Nowadays, Adversarial Networks are commonly used in the industry.