Reading Assignment 7: Generative Adversarial Networks (GANs) continued

Probabilistic Machine Learning: Advanced Topics – Remainder of Chapter 25

Former Reading Assignment (optional)

• Wasserstein GAN
• Improved WGAN

In both papers, concern yourself with the high-level ideas. You don’t need to understand all the mathematical details (but give it a shot!) and you can of course skip the appendices. There is also a less technical treatment of these concepts which may be easier to follow for you.

• Progressive Growing of GANs. These NVIDIA papers are usually very well written, providing justifications for their ideas, ablation studies for what happens when you leave them out, as well as noting all the “little tricks” that make the big idea actually work.
• If you still have time, check out StyleGAN (follow-up to progressive growing) as well.
• Finally, this is very readable and highly recommended for context: Wasserstein GANs work because they fail.

Optional Further Reading

• There is even a StyleGAN2!
• Open GAN Problems: To get a taste of unanswered questions.
• Adversarial Audio Synthesis and GANSynth if you want to see some non-image applications.

Feel free to share additional papers or other resources you come across!