Variational Autoencoders

April 01, 2020

Abstract

There are traditionally two ways of seeing a Variational Autoencoder: as an autoencoder that uses variational inference loss, or as a variational inference algorithm that uses an autoencoder as the underlying architecture. I like the later, which comes from a more probabilistic school and gives (in my opinion) a better understanding of what is going on.

View Full Presentation