Deep Generative Models Under GAN: Variants, Applications, and Privacy Issues

Deep learning has lately acquired a lot of attention in machine learning because of its capacity to train features and classifiers at the same time, resulting in a significant boost in accuracy. To attain a high level of accuracy, the models require huge amounts of data and processing capacity, both of which are now available due to the advancements in big data, Internet of Things, and cloud computing. Even though, some applications like medical diagnosis, image recognition, and biometric authentication faces the problem of data scarcity which affects the predictive analytics of deep learning. In order to tackle the issue, deep generative models like Generative Adversarial Networks (GAN) come into existence that are capable of artificially generating synthetic data for specific problems. In this article, various models of GAN and their applications were explored and a comparison of the models were also given. As the data increases, another issue faced by the applications is of data privacy. With rising privacy concerns, more priority has to be given for privacy issues while developing intelligent applications. GAN and its variants are nowadays used as an attacker as well as a defender against various privacy risks which were also presented in this review. As a future work, GANs potential to solve the issues of data privacy and security has to be deeply explored.