KWIK Learning: Explicit Uncertainty and Decision Making


Invited Talk
Michael Littman (Rutgers University)

Abstract:

I will present KWIK (Knows What It Kkows) framework for supervised learning and relate it to the problem of efficient reinforcement learning.  I argue that KWIK learning algorithms lie at the heart of existing polynomial-time learning results and also point the way to approaches to learning in novel models.  I will try to seuqeeze in as many robot videos as I can to illustrate these formal results.