Reading Assignment 4: Generative Adversarial Networks

This time we will be using the book Probabilistic Machine Learning: Advanced Topics by Kevin P. Murphy. Chapter 26 is on GANs. Please read:

• Up to and including section 26.3
  • You can skip the latter parts of 26.2 (f-divergences and following sections)
• 26.6 on neural architectures

The book treats GANs with a lot of “hindsight”, embedding them in a broader context of so-called implicit likelihood models, as well as providing a unified framework for many different GAN variants. If you want a simpler, more heuristic approach, consider reading the original paper. As another alternative, you could consider this blog post.

Finally, DCGAN was the first paper to demonstrate successful use of CNNs in GANs. The paper includes many “design principles” to stabilize GAN training. It’s a bit outdated, but can still be useful.