The course does not introduce lots of practical learning methods to be used as black-box methods. Theories of learning are not taught in detail either. Instead, the possibility of learning from data without any assumptions is examined. This is shown to be impossible from a certain point of view. Some difficulties of "weak" assumptions are also demonstrated. Next, Bayesian methods are developed as a well-justified approach to learning from data. Rest of the course deals with Bayesian methods with the emphasis on ideas and not on complicated details. One lecture will briefly examine Statistical Learning Theory, which appears to suggest that learning from data without assumptions is possible. This result is examined using Bayesian approach, which reveals that the theory actually does not support learning without assumptions.
As prerequisite information you will need the basic mathematics and probability courses, and know the basics of calculating with matrices. It is very useful to have taken some course which deals with modeling data nonlinearly (for example neural networks or pattern recognition).
Watch the webpage and the newsgroup opinnot.tik.informaatiotekniikka
You are at: CIS → T-61.5040 / Course Information
Page maintained by t615040 (at) cis.hut.fi, last updated Thursday, 11-Jan-2007 13:09:07 EET