If you've ever wondered what "bits-per-pixel" actually means, why logistic distributions are awesome, how to improve training stability of normalizing flows and more... Check out this blog post on training likelihood models blog.evjang.com/2019/07/like…