Generative Deep Learning

Chapter 9. The Future of Generative Modeling

I started writing this book in May 2018, shortly after the “World Models” paper discussed in Chapter 8 was published. I knew at the time that I wanted this paper to be the focus of the final core chapter of the book, as it is the first practical example of how generative models can facilitate a deeper form of learning that takes place inside the agent’s own world model of the environment. To this day, I still find this example completely astonishing. It is a glimpse into a future where agents learn not only through maximizing a single reward in an environment of our choice, but by generating their own internal representation of an environment and therefore having the capability to create their own reward functions to optimize. In this chapter, we will run with this idea and see where it takes us.

First, we must place ourselves at the very edge of the generative modeling landscape, among the most radical, innovative, and leading ideas in the field. Since the inception of this book, significant advancements in GAN and attention-based methodologies have taken us to the point where we can now generate images, text, and music that is practically indistinguishable from human-generated content. We shall start by framing these advancements alongside examples that we have already explored and walking through the most cutting-edge architectures available today.

Five Years of Progress

The history of generative modeling in its current form is ...