BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Learning Structural Kernels for Natural Language Processing - Dani
 el Beck\, University of Sheffield
DTSTART:20150911T110000Z
DTEND:20150911T113000Z
UID:TALK60616@talks.cam.ac.uk
CONTACT:Tamara Polajnar
DESCRIPTION:Structural kernels are a flexible learning\nparadigm that has 
 been widely used in Natural\nLanguage Processing. However\, the problem\no
 f model selection in kernel-based methods\nis usually overlooked. Previous
  approaches\nmostly rely on setting default values for kernel hyperparamet
 ers or using grid search\,\nwhich is slow and coarse-grained. In contrast\
 , Bayesian methods allow efficient model\nselection by maximizing the evid
 ence on the\ntraining data through gradient-based methods.\nIn this paper 
 we show how to perform this\nin the context of structural kernels by using
 \nGaussian Processes. Experimental results on\ntree kernels show that this
  procedure results\nin better prediction performance compared to\nhyperpar
 ameter optimization via grid search.\nThe framework proposed in this paper
  can be\nadapted to other structures besides trees\, e.g.\,\nstrings and g
 raphs\, thereby extending the utility of kernel-based methods.\n
LOCATION:FW26\, Computer Laboratory
END:VEVENT
END:VCALENDAR