Links: Course Outline, Lecture Notes, Homework, Free Software

STA 294: Special Topics

Belief Networks

Instructor: Mark Peot

MWF, 9:10AM- 10:00 in Old Chemistry, Room 025.

Synopsis: Belief networks[ are an increasingly popular tool for representing uncertainty in artificial intelligence, statistics, and engineering. Belief networks are finding application in many products that people use everyday, for example, Microsoft has developed belief network-driven applications for several products, including the Answer Wizard of Office 95, the Office Assistant (the bouncy paperclip guy) of Office 97, and over 30 Technical Support Troubleshooters. Belief networks have found application in a number of domains, including:

  • consumer help desks,
  • nuclear reactor diagnosis,
  • tissue pathology,
  • pattern recognition,
  • credit assessment, and
  • computer network diagnosis.

The belief network representation and inference algorithms subsume many special purpose algorithms; including, for example, those used for prediction, estimation, and smoothing in linear statistical models (including Kalman filters and mixture models).

STA 294 is designed to introduce the student to the theory and application of belief networks and other graphical models for joint probability distributions. Topics discussed include:

    • Directed and undirected graphical models;
    • Independence properties in graphical models;
    • Exact and approximate inference algorithms (including the polytree, cutset conditioning, factoring, join tree, likelihood weighting, bounded variance, and AA algorithms);
    • Reasoning with multinomial, continuous, normal and mixture distributions;
    • Complexity results for exact and approximate reasoning (including ed-bounds);
    • Probabilistic expert systems, and
    • Learning algorithms (including parameter learning and structure learning techniques).

Grading: Homework (50%) and a course project (50%).

Prerequisites: Probability, including Bayes Law, multinomial and normal distributions and basic statistics.


(Required) Castillo, Gutierrez, and Hadi, Expert Systems and Probabilistic Network Models, Springer-Verlag, 1997.

(Required) STA 294 Course Reader

(Recommended) Laurtizen, Graphical Models, Oxford University Press, 1996.


Course Outline

Lecture Notes


Free Software

[AKA Bayesian networks AKA Bayes nets AKA causal probabilistic networks AKA directed graphical models.

Copyright 1999, Mark Alan Peot