2017/2018 Edition

Probabilistic Graphical Models

Lecturer: Prof. Marek J. Dru┐d┐el (University of Pittsburgh, USA).

About the lecturer | Course Summary | Materials

About the lecturer: Marek Dru┐d┐el is an associate professor in the School of Information Sciences and in the Intelligent Systems Program and the director of the Decision Systems Laboratory at the University of Pittsburgh. He is also a visiting professor in the Department of Computer Science, Bialystok University of Technology. He got his M.S. degrees in Computer Science (1985) and Electrical Engineering (1987) from the Delft University of Technology in The Netherlands (both with distinction) and his Ph.D. in Engineering and Public Policy (1992) from Carnegie Mellon University, Pittsburgh, PA, USA. Prof. Druzdzel is a recipient of the Faculty Early Career Development Grant (known as CAREER grant) from the National Science Foundation (1996-2000), Outstanding Mentor Award (1997), and University of Pittsburgh's Chancellor's Distinguished Teaching Award (2007). He is spending his sabbatical year (2009-2010) at the Bialystok University of Technology on a Fulbright grant. His research interests concentrate on probabilistic and decision theoretic methods in decision support systems and human aspect of decision support. His laboratory is widely known for the graphical modeling software GeNIe and SMILE.

Course summary: Directed probabilistic graphs, such as Bayesian networks, are well known for their ability to tackle successfully hard practical problems. Their success can be attributed to a unique combination of the intuitive framework of directed graphs and to the sound foundations of probability theory and decision theory on which they are built. In this course, I will present the foundations of discrete probabilistic graphical models, show how they can be built based on expert knowledge and applied to problems such as classification or diagnosis. I will show how they extend to continuous variables and temporal domains. I will also show how their extension, influence diagrams, allow for modeling decision problems. Finally, I will focus on learning directed graphs and causal discovery from data. The course will be fairly self-contained and intuitive for computer science majors, although knowledge of elementary probability theory is a must.

Materials: Slides, exercises, assignments

Deadline for solutions: 10 May, 2010