Active Inference just got real, thanks to Monte Carlo trees and information theory.
Like Markov blankets, but that actually do what you think they do.
Connections between information theory, supervised learning, and privacy
Using graph neural networks to understand and improve multi-agent RL.