|
Department of
Computational Perception
|
|
PROBABILISTIC MODELS
344.036, VO, 2 hrs. (3 ECTS), WS 2017/18
Univ.-Prof. Dr. Gerhard Widmer
This class will be taught in English.
NOTE: Time and place had to be changed due to the large number of
persons interested in participating:
NEW TIME: Monday, 13:45 - 15:15
NEW PLACE: Science Park 2, Room 048
START: Monday, Oct. 2, 2017
Motivation:
Here is a little set of
Introduction & Motivation Slides ...
Goals and Contents of this Class:
This course is a gentle introduction to one of the most important and
central classes of methods in present-day Artificial Intelligence.
It will introduce students to the basic concepts of
Probabilistic Graphical Models as representations of uncertain knowledge
in complex domains. All three aspects related to such models will be covered:
model semantics, inference, and learning.
In particular, the following topics will be covered (in more or less detail):
- Elementary Concepts:
Basics on Probability Distributions, Density Functions,
Probabilistic Reasoning and Inference.
- Bayesian Networks:
Representation, Semantics, Conditional Independence, Factorisation.
- Inference in Bayesian Networks:
Exact Inference, Variable Elimination Algorithms;
Approximate Inference via Stochastic Sampling,
Markov Chain Monte Carlo (MCMC) Methods.
- Learning Bayesian Networks:
Parameter Learning, Structure Learning, Learning Generative vs. Discriminative Models.
- Modelling and Reasoning about Temporal Phenomena:
Kalman Filters, Hidden Markov Models, Dynamic Bayes Nets, Particle Filters.
- (Briefly, if there is time:) Semi-directed and Undirected Models:
Conditional Bayes Nets, Conditional Random Fields, Markov Random Fields.
- Selected Applications of Probabilistic Graphical Models.
It is strongly recommended to take this VO together with the
"Practical Excercises in Probabilistic Models" (UE, 1h) in the same semester.
There, the students will perform practical experiments with some of the methods
taught in the VO.
Teaching materials:
Pdf versions of the Powerpoint slides used in the lecture will be made
available via KUSSS (weekly).
Recommended reading (will not be needed if the lectures are
attended on a regular basis):
Koller, Daphne and Friedman, Nir (2009).
Probabilistic Graphical Models: Principles and Techniques.
Cambridge, MA: MIT Press.
Russell, Stuart J. and Norvig, Peter (2003).
Artificial Intelligence: A Modern Approach.
Englewood Cliffs, NJ: Prentice Hall.
Questions, suggestions, complaints, etc. to:
Gerhard Widmer
Tel. 2468-4701
gerhard dot widmer at jku dot at
last edited by gw on Aug 16, 2017