ColloquiaTalk by Shai Ben-David: Monday March 10th
Meridel Trimble
mtrimble at tti-c.org
Wed Mar 5 16:22:42 CST 2003
-------------------------------------------------------------------------------
TOYOTA TECHNOLOGICAL INSTITUTE
-------------------------------------------------------------------------------
Date: Monday, March 10th, 2003
Time: 2:30 p.m.
Place: Ryerson Hall 251
Speaker: Shai Ben-David
CS Department, Technion, Haifa, Israel and School of ECE, Cornell University, NY
Title: Computational Learning Theory
tradeoffs between Computational Complexity and Statistical Soundness
Abstract:
In the past decade machine learning witnessed a fascinating interplay between
theory and practice. Several ideas that were originally developed by
theoreticians were translated into popular andsuccessful practical algorithms.
Furthermore, the experimental success of these algorithms exceeds the most
optimistic theoretical predictions, thus creating a new challenge for
theoreticians - to justify the
unexpected success of their own algorithmic ideas. In some cases, further
research results in even more pessimistic theoretical predictions.
In this talk I will survey these developments from the point of view of the
interplay between the two main cost measures of learning; Computational
complexity and informational complexity.
Here, informational complexity refers to questions like How many training
examples are required to provide a certain statistical soundness guarantee?
And computational complexity refers to the computation applied to the
training data to extract from it the learners conclusions.
I shall discuss some recent insights and research directions concerning three
of the most common learning paradigms Neural Networks, Boosting, and Support
Vector Machines.
*Refreshments will be served after the talk in Ryerson 255*
If you wish to meet with the speaker, please send e-mail to Meridel Trimble
mtrimble at uchicago.edu
More information about the Colloquium
mailing list