CSCE 582—Bayesian Networks and Decision Graphs [=STAT 582] (3) (Prereq: CSCE 350 and STAT 509) Normative approaches to uncertainty in artificial intelligence. Probabilistic and causal modeling with Bayesian networks and influence diagrams. Applications in decision analysis and support. Algorithms for probability update in graphical models.

Course Homepage: Spring 2006 (Past Pages: Fall 2003 )

Usually Offered: Once every two years, in the Computer Science Department

Purpose: To appreciate the foundations, power, and limitations of probabilistic and causal modeling with Bayesian networks, solve computer-based decision analysis problems using a Bayesian network and influence diagram tool, and understand and implement structure-based (non-iterative) algorithms for probability update in graphical models.

Current Textbook: Bayesian Networks and Decision Graphs, (2nd ed.) Finn V. Jensen and Thomas D. Nielsen, Springer, 2007.

 Topics Covered



Uncertainty in Artificial Intelligence: symbolic, non-probabilistic, and probabilistic approaches; review of relevant probability theory


1 week

Causal and Bayesian networks: reasoning under uncertainty, d-separation, factorization of joint probability in graphical models, the chain rule for Bayesian networks, findings and evidence, the variable elimination algorithm for computing posterior marginal probabilities; review of relevant graph theory


3 weeks

Building models: catching the structure, determining the conditional probabilities; modeling methods, including Kalman filters, hidden Markov models, noisy-Or, divorcing, noisy functional dependencies, interventions


3 weeks

Learning, adaptation, and tuning

parts of 6 and 7

2 weeks

Graphical languages for specification of decision problems: decision trees and influence diagrams


2 weeks

Belief updating in Bayesian networks: triangulated (chordal) graphs, junction trees, Lauritzen-Spiegelhalter, Shenoy-Shafer, and Hugin propagation in junction trees


2 weeks