This repo contains tutorial content, example code, and working loss function calculators for Invariant Representations without Adversarial Training.
A transcription of the tutorial content in non-interactive HTML form is hosted on my website in at test blogpost.
You may also be interested in the Echo Noise Distribution, a noise model for encodings z that provides closed form rate terms I(z,x), with nicely computable derivatives. This is exactly the term we would like to penalize in the Inv. Rep. paper, but instead had to bound.