**
Probabilistic Artificial Intelligence
Michael Kearns,
AT&T Labs
Annual Workshop on Computational Complexity
Bellairs Research Institute of McGill University
Holetown, St. James, Barbados
February 22 - 26, 1999
**

**
COURSE DESCRIPTION:
**
In the last decade or so, many of the central problems of "classical"
artificial intelligence - such as knowledge representation, inference,
learning and planning -
have been reformulated in frameworks with statistical
or probabilistic underpinnings. The benefits of this trend include
the adoption of a common set of mathematical tools for the various
AI subdisciplines, increased attention on central algorithmic issues,
and an emphasis on approximation algorithms for some notoriously hard
AI problems.

In this lecture series, I will survey the probabilistic frameworks and the central computational problems posed in several well-developed areas of AI. I will describe some of the algorithms proposed for these problems, overview what is formally known about them (and also what is suspected but not proven), and try to give a flavor of the mathematical techniques involved. The lectures will be self-contained, with an emphasis on the interesting open problems.

Below is a (very) approximate outline of what I hope to cover; undoubtedly, it is overly ambitious. I am also happy to let the interests of the participants influence the directions we pursue. The outline includes many related papers for your perusal; as we approach the dates of the meeting, I will be posting more material, and will let you know as I do so.

I have only one text recommondation for the course, but I do recommend it strongly. It covers many of the basics, and it is a very accessible introduction to an influential topic in AI:

See you in Barbados!

** COURSE OUTLINE **

**
Basics of Markov Decision Processes and Reinforcement Learning
**

**
Related Papers:
**

**
The Theory of Uniform Convergence
**

**
Related Papers:
**

**
Improved Theoretical Results for Learning in MDPs
**

**
Related Papers:
**

**
Getting (More) Realistic: Handling Large State Spaces
**

**
Related Papers:
**

**
Probabilistic Reasoning and Bayesian Networks
**

**
Related Papers:
**

**
Approximate Inference in Bayes Nets
**

**
Related Papers:
**

**
Combining MDPs and Bayes Nets
**

**
Related Papers:
**

**
More Realism: Partial Observability and Macro-Actions
**

**
Related Papers:
**

Michael Kearns, AT&T Labs - Research Tel: (973) 360-8322 180 Park Avenue, Room A235 Florham Park, NJ 07932 mkearns@research.att.com

Last Edited: February 26, 2001