[Colloquium] McAllester talk today - 12:15 at TTI

[Meridel Trimble]

TOYOTA TECHNOLOGICAL INSTITUTE TALK 

Speaker: David McAllester 
TTI-C 

Speaker’s homepage: http://tti-c.org/mcallester.shtml 

Time: 12:15pm 
Date: Thursday, February 12th, 2004 
Place: TTI-C (1427 E. 60th St. 2nd Floor– University Press Building) 
LUNCH PROVIDED 

Title: Generalization bounds for phrase level precision and recall in 
statistical parsing 

Abstract: 
Learning theory has largely addressed classification --- predicting whether a 
given object, such as an image of hand written character, is in a certain 
class, such as the letter "w".  In many applications, however, the problem is 
to recover structured information such as a word sequence in speech 
recognition, a part of speech sequence in text tagging, or a parse tree in 
statistical parsing.  In these applications the traditional "0-1" notion of 
error rate seems inappropriate.  More appropriate is to measure the percentage 
of recovered structure that is correct.  For example, the word recognition 
rate in speech recognition or the phrase level precision and recall in 
statistical parsing.  This talk will give new theoretical bounds on the 
generalization error as measured by "Hamming loss" --- the Hamming distance 
from the recovered structure to the target structure.  These are large margin 
bounds proved with PAC-Bayesian methods.  The proof methods are perhaps as 
interesting as the results. 
  
If you have questions, or would like to meet the speaker, please contact 
Meridel  (4-9873/mtrimble at tti-c.org) 
  
For information on future TTI-C talks or events, please go to the TTI-C Events 
page: http://www.tti-c.org/events.shtml