Connections Between Inference and Control
Berkeley’s deep rl course covers soft optimality and various connections with different rl algorithms in this lecture.
The basis for this lecture comes from Probabilistic Graphical Models (PGM), which cover various Bayesian techniques like Bayesian Neural Nets and Variational Autoencoders. This makes it a little bit out of reach for me, since I’ve not yet covered these topics in depth, although most of this lecture still made sense. It’s rather concise (covering what soft optimality is and various intuitions behind some of the basic derivations) so I’ll just leave my notes as usual: