Logical Representational and Computational Methods for Markov Decision Processes

Craig Boutilier

Department of Computer Science
University of Toronto
Toronto, Ontario, Canada M5S 3H5

(cebly@cs.toronto.edu)
http://www.cs.toronto.edu/~cebly

Prerequisites: This course presumes a familiarity (or comfort level) with elementary (discrete) probability theory and inference, and propositional and predicate logic. Some familiarity with Bayesian networks will be helpful, but will not be assumed.

Summary: Markov decision processes (MDPs) have become standard models for sequential decision problems involving uncertainty within the planning and probabilistic reasoning communities. The aim of the course is:

(a) to provide an introduction to Markov decision processes;

(b) to survey some of the recent advances that have been made in the concise and natural representation of MDPs using logical techniques; and

(c) to survey some of the computational methods for solving MDPs that exploit this logical structure.

Representations to be discussed include propositional methods (e.g., probabilistic STRIPS, dynamic Bayesian networks), first-order representations, representations using temporal logics, and computationally-motivated representations drawn from the verification community (e.g., BDDs). Computational techniques include regression, abstraction, and decomposition methods that rely on these specific logical representations.

Course Notes:
Lecture 1
Lecture 2
Lecture 3
Lecture 4
Lecture 5

Back to course listing.