Building Generative Adversarial Networks

Summary

This chapter describes what Generative Adversarial Networks (GANs) are and demonstrates how they're capable of generating new data, especially images, from preexisting ones. Contrary to other deep learning neural networks that classify images or sequences, the specialty of GANs is their capability to generate new data by deriving inspiration from training data. Creating a basic GAN example has become quite accessible using Keras with TensorFlow, and learning by doing is a good way to explain details of the technology that would otherwise remain theoretical. The list of GANs and their applications grows every day, and keeping up is difficult. The chief application of GANs is to create images. Producing images of higher quality and possibly controlling the output generated has opened the way to more applications. The chapter concludes with the overview of GANs by building a network using Keras and TensorFlow.