Laboratory of Computer and Information Science

Tik-61.184 Special Course in Information Technology IV (4 cr) (L)

Lecturers: Prof. Mark Girolami and Doc. Aapo Hyvärinen

Assistant: MSc Patrik Hoyer
Semester: Autumn 2000
Credit points: 4 cr
Place: Seminar room A328 in the Computer Science Building (note: this may change!)
Time: Thursdays 14-16, starting 14th September
Language: English

Unsupervised Learning

Unsupervised learning means neural network learning in which there is no teacher or desired response. The neural network must find the best representation or the best features of the input data. This often means finding underlying factors or causes.

In this course, we consider in detail some methods of unsupervised learning, both the theory and the applications. Emphasis is laid on independent component analysis and feature extraction from images. These are also subject to intensive research in the Neural Networks Research Centre, which is the research division of the Lab of Computer and Information science. Therefore, the course also serves as an introduction to this topic of research in which our research centre is one of the leading in the world.

The course material is based on selected chapters from the following books:

and some related journal articles.

Prerequisites for the course:

Required: Tik-61.261 Principles of neural computing.
Helpful but not required: Tik-61.263 Advanced course in neural computing.

Requirements for passing the course:

To pass "with distinction", at least 95% of the exercises should be solved, and the presentation as well as the project work should be very good.

Signing up for the course:

Simply show up on the starting session (Sept. 14th). If you cannot make it there for some reason, email the course assistant:

Practical arrangements:

The course schedule (with topics) can be found here. The schedule also has links to the slides of the presentations which have been held.

The exercises are available here: (Note: no exercises for session 6)
Session: [#1] [#2] [#3] [#4] [#5] [#7] [#8] [#9]
(Note: In the handouts of #9, the Gaussian density was not properly normalized! This has been fixed in this online version.)
Note that they DO NOT have to be returned on a weekly basis, rather you may returned them all after the course (DL January 31st, 2001).

Project work:
Data: [#1] [#2] [#3] [#4]
Deadline: January 31st, 2001

Note that at each session we will hand out copies of the papers for the next session, so that you can read the papers in advance. (The papers for the first session will not be given in advance though.)

Each presentation should be roughly 35-40 minutes in length (except in those sessions containing three papers, where 25-30 minutes is the rule). Please try to emphasize the central ideas in your chapter, and not focus too much on technical details. You are required to email the electronic file containing your slides to

If you have any questions regarding the practical arrangements, please email the assistant (

Previous seminars: [99]
Other CIS courses
September 20th, 2000
Patrik Hoyer