Dedicate 1hr to learn about Variational Autoencoders

[jvican]

Variational Autoencoders have showed up in the footnotes of the Pyro SVI documentation. Given these connections and their presence in current ML research, I've decided I should become more familiar with them.

Paper https://paperswithcode.com/method/vae Blog summary with diagrams https://zhenye-na.github.io/blog/2018/05/18/intro-to-variational-autoencoders.html

This 1hr will include:

1. Peeking at the paper and trying to read the core of the approach as quickly as possible
2. Get familiar with pseudo code that explains how they work
3. Be able to answer: why are VAEs so broadly used? What problem do they solve? How do they solve it?
4. What are the alternatives to VAE nowadays?

At the end of the session, i'll decide whether I should have a follow-up session on it or not.