The Weizmann Institute of Science
Faculty of Mathematics and Computer Science
Vision and Robotics Seminar
Shahar Mendelson
Research School of Information Sciences and Engineering
The Australian National University
Canberra, Australia
will speak on
Some remarks on Learning Theory
from a geometric viewpoint
Abstract:
Learning Theory deals with the ability to approximate an unknown random variable (or target function)
by an element in a given function class, where the information one can use to accomplish this task
is a random sample of the unknown function. The main question is ``how close" (as a function of the
size of the sample) can one approximate the random variable in an appropriate sense. It turns out
that the answer to this question is determined by the geometric structure of the given function class.
We will present some sharp estimates on the so-called ``error-rate" (i.e. degree of approximation as
a function of the sample size) based on the geometry of the function class.
The lecture will take place in the
Lecture Hall, Room 1, Ziskind Building
on Thursday, December 30, 2004
at noon