Generative Adversarial Networks (GANs)

By 2014, a generative adversarial network (GAN) was proposed by Goodfellow et al. as an intelligent deep‐learning approach that could take the advantage of discriminative learners to build a well behaved generative learner. This chapter dives into the details of the standard GAN model as the baseline member of the family of generative deep networks. By covering the principles of GANs, it looks at such early GANs and shows how to obtain satisfactory training. The chapter focuses on two well‐known generative models, namely deep convolutional GAN and conditional GAN (CGAN). CGAN for simplicity, is a type of GAN that involves the conditional generation of data instances by a generator model. A conditional setup is used in CGANs, which means that both the generator and discriminator are contingent on auxiliary input from other modalities.