Deep Learning

22 Generative Adversarial Networks

Generating data is exciting. It lets us produce new paintings, songs, and sculptures that have a resemblance to their inputs. In Chapter 18 we saw how to use autoencoders to generate new data that was like the training data. In this chapter we explore a completely different approach to data generation. The type of system we look at is called a Generative Adversarial Network, or GAN. It’s based on a clever strategy where two different deep networks are pitted against one another, with the goal of getting one network to create new samples that are not from the training data, but are so much like the training ...