Chapter 6. Using Generative Models

Generative models generate new data. In a way, they are the exact opposite of the models that we've dealt with in prior chapters. While an image classifier takes in a high-dimensional input, the image, and outputs a low-dimensional output such as the content of the image, a generative model goes about things in exactly the opposite way around. It might, for example, draw images from the description of what's in them.

Generative models are still in the experimental phase of their development, and are currently used mostly in image applications. However, they are an important model as shown by the fact that there have already been several applications that have used generative models that have caused an uproar within the industry.

In 2017, so-called DeepFakes began to appear on the internet. Generative Adversarial Networks (GANs), which we will cover later in this chapter, were used to generate pornographic videos featuring famous celebrities. The year before...