Hi, this is Mark. I document my learning journey here.
I hope you can find something useful from this blog.
The Story behind WGAN
This post explores the theory and some follow-up papers of one of the most influential machine learning papers: Generative Adversarial Networks (GANs). Contrary to other deep learning models, I find that generative models are supported by more rigorous mathematics that are easily digestible. As we understand the theory behind GANs (mostly through the paper by Arjovsky and Bottou, 2017), we will recognize its limitations and the reason behind its instability. This naturally leads us to Wasserstein Generative Adversarial Networks (WGANs), which apply useful concepts from Optimal Transport (OT). ...