SGN-9406 Signal Processing Graduate Seminar IV, 3-8 cr

.2007  Course requirements updated.
.2007  Preliminary schedule is now available.


Probabilistic graphical models

Time and Place:
Periods II & III, Wednesdays 10:15-12:00, TB219

First meeting:

October 17, 2007

Harri Lähdesmäki
office: TF412

This seminar course provides an introduction to the theory and algorithms of probabilistic graphical models. These models include both directed and undirected graphical models, such as Bayesian networks and Markov random fields (emphasis on directed models). Topics of the course include background, representation, inference, parameter and structural learning, and extensions to dynamic graphical models (e.g. dynamic Bayesian networks, hidden Markov models, etc.). If time permits, we will also study some important applications, with an emphasis on biologically oriented ones.

Active participation in the lectures
and seminar presentations (at least 70%) and one oral presentation. (More than one presentation might be possible with extra credit points.) There will be no final exam. If you give one presentation, you will get 3 credit points. By giving two presentations will give you 5 credit points.

Practical implementation:
The first two lectures and one lecture in the middle of the course are given by the instructor. The rest of the seminar consists of student presentations.

Basic knowledge of probability, statistics and signal processing is assumed.

At the first meeting. If you are unable to attend the first meeting, please email the course instructor.


The course is accepted as a postgraduate course (but M.Sc. students can participate as well).

Probabilistic Networks and Expert Systems,
Robert G. Cowell, A. Philip Dawid, Steffen L. Lauritzen, and David J. Spiegelhalter,
Springer, 2003
Amazon, Google books

Course covers part of the book. The book can be borrowed from the course instructor. The material will be supplemented by recent articles.

Additional reading:
  1. D. Heckerman, A tutorial on learning with Bayesian networks. Technical report, Microsoft Research, 1995.
  2. D. Husmeier, Introduction to learning Bayesian networks from data (Chapter 2 of the book Probabilistic Modeling in Bioinformatics and Medical Informatics by Dirk Husmeier, Richard Dybowski and Stephen Roberts, Springer, 2006).
  3. N. Friedman and D. Koller, Being Bayesian about network structure: a Bayesian approach to structure discovery in Bayesian networks, Machine Learning, 50:95-126, 2003.
  4. M. Koivisto and K. Sood, Exact Bayesian structure discovery in Bayesian networks, Journal of Machine Learning Research, 5(May):549-573, 2004.
  5. D. Eaton and K. Murphy, Exact Bayesian structure learning from uncertain interventions, In AI & Statistics, 2007.
  6. K. Murphy, Active learning of causal Bayes net structure, Tehcnical report, March 2001.
  7. E. Segal, D. Pe'er, A. Regev, D. Koller, and N. Friedman. Learning Module Networks, Journal of Machine Learning Research (JMLR), 2005 April, 6(Apr): 557-88.
  8. Perhaps parts of a Ph.D. Thesis by K. Murphy, Dynamic Bayesian Networks: Representation, Inference and Learning, University of California, Berkeley, 2002.

17.10.2007  First meeting, Chapters 1-2, Harri Lähdesmäki
24.10.2007  Chapter 3, Harri Lähdesmäki
31.10.2007  Sections 4-4.2 (pages 43-52), Kirsti Laurila
07.11.2007  Sections 4.3-4.5 (pages 52-61) Xiaofeng Dai
14.11.2007  Sections 5-5.3 (pages 63-75) Jyrki Selinummi
21.11.2007  Sections 6-6.3 (pages 83-95) Jukka Intosalmi
28.11.2007  No presentation.
12.12.2007  Section 6.4 (pages 95-109) Antti Larjo, and Sections 6.5-6.7 (pages 109-123) Timo Erkkilä
19.12.2007  Sections 9-9.4 (pages 189-199) Alireza Razavi
09.01.2008  Sections 9.5-9.10 (pages 200-223) Xiaofeng Dai
16.01.2008  Chapter 10 (pages 225-241) Vinod Kumar
23.01.2008  Chapter 11 (pages 243- ) Stefan Uhlmann
**.**.2008  Articles ??