Seminar in Computational Learning

Dr. Amos Beimel

Fall 2000-01

The ability of a computer to learn by interacting with the environment is a fascinating subject. There are a few theoretical models of the learning process. The most important model is the Probably Approximately Correct (PAC) model: A learner gets examples labeled according to some target function. His purpose is to "learn" the target function, i.e., to output a hypothesis that approximates this function. In this seminar we will describe learning algorithms in this model, discuss learning in the presence of noise, and show how to boost weak learning algorithms to strong learning algorithms.

Requirements:

This is a 2-credit course, consisting of weekly 2-hour meetings. I will give the first two lecture. Each student will give one lecture based on papers that I'll give him. In addition the student will submit a summary of the paper they presented. The students are required to attend (and hopefully understand) all lectures. There maybe one or two homework assignments (if students won't be active in the lectures).

Pre-requirements:

The course in intended for third year undergrad students and graduate students. The course "design of algorithms" is a pre-requisite.

Lectures and papers

Book:

M. J. Kearns and U. V. Vazirani. An Introduction to Computational Learning Theory. MIT Press, 1994.

Information:

Lecture hours:	Monday 10:00-12:00.
Reception hours:	Sunday 16:00-18:00, Building 58 (Math and CS building), Room 205.
E-mail:	beimel at cs.bgu.ac.il
Phone:	647 7858
Course homepage:	www.cs.bgu.ac.il/~beimel/Courses/Learning2000/learning.html

Links:

Home page of previous seminar