2D5311 Computational learning theory, Fall 1994

This course in learning theory is based on Ron Rivests MIT course 18.428. The material will be illustrated with some recent applications.

Lecturer

Stefan Arnborg

Course start Oct 27

Tuesdays at 3pm (15:15)
Meeting room, Nada plan 5
Lindstedtsvägen 3.

Tentative Topics

Magic of learning
Concept learning (vocabulary, presentation, performance criteria)
Version spaces and learning from positive data
Learning with perceptrons and neural nets
Statistical learning concepts (Bayes, ML, MAP, MDLP)
Distribution-free learning (PAC)
Applications in pattern/signal recognition

Course Material

An Introduction to Computational Learning Theory, Mike Kearns and Umesh Vazirani, MIT Press 1994. isbn 0262-111934. USD 46.00. MIT Press Order Information - Books
Course notes (in preparation)

Prerequisites

Discrete Math (particularly propositional logic and finite set theory)
Probability and Statistics
Algorithms and Data Structures.
Complexity theory (the concepts P and NP)

Examination

Homework and a take-home exam. Homework must be turned in when due. Examination is only given when the course is running (every second year).

Newsgroup

There is also a newsgroup where you will find various announcements.

Related resources

Entrance to Nadas course web
Rivests course
Similar course by Mitchell/Blum at CMU
Some pointers to applied research centers

CBCL, the MIT-based Center for Biological and Computational learning, Web pages.
The Perceptual Science Laboratory at the University of California-Santa Cruz
The SANS group of NADA

What becomes of learning?

Stefan Arnborg <stefan@nada.kth.se>