Lecture Bayesian Networks Winter Term 2015/2016

we finished to check your exams. You can look at the corrected exams on Thursday (18.02.2016) from 10am to 11am or from 1pm to 2pm in room G29-015. The grades will be created afterwards.

Summary

• General Information
• Schedule and Rooms
• Lecturers
• Conditions for Certificates (Scheine) and Exams
• Prerequisites
• Slides
• Assignment Sheets
• Additional Material
• Software
• References
• Links

News

General Information

This page contains information about the lecture "Bayesian Networks" (in German "Bayes-Netze") that is held in winter term 2015/2016 by Prof. Dr. Rudolf Kruse. This page is updated during the course.

Topics

• Modeling of uncertainty and vagueness in expert systems
• Representation of uncertain information in probabilistic networks (Bayesian networks / Markov networks)
• Evidence propagation in such networks
• Quantitative and structural induction of probabilistic networks from data
• Other uncertainty calculi (Dempster-Shafer)
• Applications

Schedule and Rooms

Lecturers

If you have questions regarding the lecture or exercise, please contact (via e-mail if possible) one of the persons named below.

• Prof. Dr. Rudolf Kruse, kruse@iws.cs.uni-magdeburg.de
• Pascal Held, pheld@ovgu.de
• Alexander Dockhorn alexander.dockhorn@ovgu.de

Conditions for Certificates (Scheine) and Exams

Certificate (Übungsschein): There are assignment sheets published every week. Assignments the solutions of which you want to present in the next exercise lecture have to be ticked beforehand on a votation sheet that is handed our prior to every exercise lecture. If ticked, you may be asked to present your solution in front of class. The solutions need not necessarily be completely correct, however, it should become obvious that you treated the assignment thoroughly. You are granted the certificate (Schein), if (and only if) you

• ticked at least two thirds of the assignments,
• presented at least two times a solution during the exercise, and
• pass the exam

Exam: If you intend to finish the course with an exam, your are required to meet the certificate conditions. There will be a written exam after the curse. You can use your own not graphical and not programmable calculator.

Prerequisites

You should have background knowledge on fundamentals of computer science such as algorithms, data structures etc. Also, insights into probability theory are highly recommended.

Slides

Note that the script may be subject to change (which will be stated in the news section above) during the course, i.e. page numbers may change.

• Slides 0 - Introduction
• Slides 1 - Uncertainty in Knowledge Based Systems
• Slides 2 - Basics of Applied Probability Theory
• Slides 3 - Separation Concepts
• Slides 4 - Decomposible Models
• Slides 5 - Probabilistic Graphical Models
• Slides 6 - Inference in Belief Trees
• Slides 7 - Inference in Bayes Networks
• Slides 8 - Belief Change in Probabilistic Graphical Models updated 11.01.2016
• Slides 9 - Manual Building of Bayes Networks
• Slides 10 - Building Bayes Networks: Parameter Learning updated 11.01.2016
• Slides 11 - Building Bayes Networks: Structure Learning updated 11.01.2016
• Slides 12 - Decision Graphs / Influence Diagrams
• Slides 13 - Dempster Shafer Theory and Belief Functions

Assignment Sheets

The assignment sheets will be published weekly at this location.

• Assignment Sheet 1 (Deadline 22.10.2015)
• Assignment Sheet 2 (Deadline 29.10.2015)
• Assignment Sheet 3 (Deadline 05.11.2015)
• Assignment Sheet 4 (Deadline 12.11.2015)
• Assignment Sheet 5 (Deadline 19.11.2015)
• Assignment Sheet 6 (Deadline 26.11.2015)
• Assignment Sheet 7 (Deadline 03.12.2015)
• Assignment Sheet 8 (Deadline 17.12.2015)
• Assignment Sheet 9 (Deadline 11.01.2016) changed!
• Assignment Sheet 10 (Deadline 28.01.2016)

Additional Material

Feel free to check out the following supplementary material that augment the lecture and exercise.

• Data Mining with Graphical Models dmwgm.pdf
• Kapitel Unsicheres und vages Wissen uvws.pdf
• Illustration of Simple Tree Propagation
• Illustration of Joint Tree Propagation
• Example Network used in the clique propagation lesson.
• Blood group determination of Danish Jersey cattle in the F-blood group system
• Introductory slides about belief functions (Source: Thierry Denoeux)

Software

Here you find links to programs with for learning and using Bayesian networks.

• Induction of Network Structures
• HUGIN Lite (Evidence Propagation)

References

• Computational Intelligence
  R. Kruse, C. Borgelt, F. Klawonn, C. Moewes, G. Ruß, M. Steinbrecher.
  Vieweg+Teubner, Wiesbaden, Germany 2011
• Graphical Models - Representations for Learning, Reasoning and Data Mining, 2nd Edition.
  C. Borgelt, M. Steinbrecher und R. Kruse.
  J. Wiley & Sons, Chichester, United Kingdom 2009
• Handbuch der künstlichen Intelligenz
  G. Görz, C.-R. Rollinger und J. Schneeberger (Hrsg.).
  Oldenbourg, München, 2000
  In particular: C. Borgelt, H. Timm und R. Kruse.
  Kapitel 9: Unsicheres und vages Wissen.
  uvws.pdf
• An Introduction to Bayesian Networks.
  F.V. Jensen.
  UCL Press, London, United Kingdom 1996
• Expert Systems and Probabilistic Network Models.
  E. Castillo, J.M. Gutierrez, and A.S. Hadi.
  Springer, New York, NY, USA 1997
• Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference.
  J. Pearl.
  Morgan Kaufmann, San Mateo, CA, USA 1988 (2nd edition 1992)

Links

• Otto-von-Guericke University of Magdeburg
• Faculty of Computer Science
• Department of Knowledge and Language Engineering
• Computational Intelligence Workgroup
• Other Lectures of the Workgroup