INI logo
WS 2012/2013 (310003)

Machine Learning: Unsupervised Methods

Lecture and Tutorial
Prof. Dr. Laurenz Wiskott
RUB logo

Lecture (2 SWS, 2 credit points): Tuesdays 12:00-13:30 o'clock in the larger INI seminar room NB 3/57. First time 16.10.2012.
Tutorial (2 SWS, 4 credit points): Tuesdays 10:15-11:45 o'clock in the smaller INI seminar room NB 3/72. First time 16.10.2012.

Questions and answers session for Exam 1: Thursday 2013-02-14 15:00-16:00 in NB 3/29.
Exam 1: Monday 2013-02-18 10:00-12:30 in HZO 60.
Exam 2: Wednesday 2013-09-18 11:00-13:30 in NC 5/99.
For the exams, only pen and ruler are permitted plus a handwritten formulary (Formelsammlung) of one page DIN A4 (only one side, not two sides written). Paper will be provided.


Advertisement: StuSer.de Onlineplattform für Studieninteressierte und Studenten


Language: This course can be given in English upon request. Course material (lecture notes and exercise sheets) will be in English in any case.

Goal: (i) The students should get to know a number of unsupervised learning methods. (ii) They should be able to discuss which of the methods might be appropriate for a given data set. (iii) They should understand the mathematics of these methods.

Content: This course covers a variety of unsupervised methods from machine learning such as principal component analysis, independent component analysis, vector quantization, clustering, self-organizing maps, growing neural gas, Bayesian theory and graphical models. We will also briefly discuss reinforcement learning.

Requirements: The mathematical level of the course is mixed but generally high. The tutorial is almost entirely mathematical. Mathematics required include calculus (functions, derivatives, integrals, differential equations, ...), linear algebra (vectors, matrices, inner product, orthogonal vectors, basis systems, ...), and a bit of probability theory (probabilities, probability densities, Bayes' theorem, ...).

Certificate: Criteria for a certificate for the tutorial are an active participation, in particular presentation of selected exercises, and at least 50% in the final exam.


Schedule

Rules of the course.

# date topic
- UNSUPERVISED LEARNING
1 16.10.2012 Introductory remarks
Principal component analysis I
2 23.10.2012 Principal component analysis II
3 30.10.2012 Principal component analysis III
Fisher discriminant analysis
4 06.11.2012 Independent component analysis I
5 13.11.2012 Independent component analysis II
6 20.11.2012 Vector quantization
7 27.11.2012 Clustering
8 04.12.2012 Self Organizing Maps
Slow Feature Analysis
9 11.12.2012 Bayesian inference
10 18.12.2012 Inference in Bayesian networks
11 08.01.2013 Inference in Gibbsian networks
12 15.01.2013 Learning in Bayesian networks
- REINFORCEMENT LEARNING
13 22.01.2013 Reinforcement Learning
14 29.01.2013 Summary and Review of All Lectures

Laurenz Wiskott, http://www.ini.rub.de/PEOPLE/wiskott/