Talk: RL, fast and slow
I'm a researcher who's just getting started, so there might be inaccuracies in my talk. Please comment if you find any and I'll be happy to correct them.
===
Correction 1: The number of possible functions is 2^(256^10000)
This is a recording of talk I gave at Cognitive AI reading group at MILA.
Because we had diverse audience (and technical concepts in the paper are advanced), I'll be focusing more on intuitions than math.
Prerequisites for understanding better: Basic RL (value functions, policy, temporal difference learning for prediction and control)
Slides: https://docs.google.com/presentation/...