Generative Adversarial Networks: A Survey on Training, Variants, and Applications

In recent years, Generative Adversarial Network (GAN) and its variants have gained great popularity in both academia and industry. In this chapter, we survey different state-of-the-art GAN-based methods and their applications. These techniques vary in architecture and objective functions. The chapter firstly introduces generative models followed by the GAN’s usual training problems, such as vanishing gradients, mode collapse, and convergence. Then, the proposed solutions and strategies for improving GAN’s training and convergence are provided, including related tasks such as obtaining higher image quality when GANs are used in image processing applications. The chapter reviews state-of-the-art GANs and focuses on the main advancements that involve adjusting the loss function, modifying the training process, and adding auxiliary neural network(s). A summary of different applications of GANs is also provided.