ColloquiaTalk by John Langford - Wednesday, October 30th, 2002

[Margery Ishmael]

---------------------------------------------------------------------------------------- 

TOYOTA TECHNICAL INSTITUTE
----------------------------------------------------------------------------------------

Date: Wednesday, October 30th, 2002

Time: 2:30 p.m.

Place: Ryerson Hall 251

Speaker: JOHN LANGFORD, Carnegie Mellon University

Title: A Tight, Simple Margin Bound

Abstract: A "margin bound" is a statistical technique for predicting
future error rates in machine learning classification problems. The
margin bound I present here is an improvement on the state of the art
in several ways:
1) It is several orders of magnitude tighter.
2) It yields functional improvements.
3) It is proved with a simpler argument.
Alterations (1) and (2) are sufficient to make the bound
quantitatively applicable to real-world learning problems unlike
previous margin bounds which were typically not interesting when
applied. Alteration (2) gives new insight into quantifying the
'hardness' of learning itself. Alteration (3) is important as it
further opens a new approach in learning theory which it significantly
more applicable.

This is joint work with Rich Caruana and John Shawe-Taylor.

*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