Notes on Causally Correct Partial Models

I recently encountered this cool paper in a reading group presentation:Rezende et al (2020) Rezende Causally Correct Partial Models for Reinforcement LearningIt's frankly taken me a long time to understand what was going on, and it took me weeks to write this half-decent explanation of it. The first notes I wrote followed the logic of the paper more, this in this post I thought I'd just focus on the high level idea, after which hopefully the paper is more straightforward. I wanted to c...