BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Inductive Logic Programming and Kernel methods - David Kirchheimer
 \, Bristol University
DTSTART:20070328T130000Z
DTEND:20070328T140000Z
UID:TALK6763@talks.cam.ac.uk
CONTACT:David MacKay
DESCRIPTION:With the introduction of kernel methods\, statistical approach
 es to machine learning have increasingly become the focus of research: Ker
 nel methods routinely outperform other approaches at classification in ter
 ms of accuracy\, and rest on a well-understood and long-researched statist
 ical basis. However\, each application to a problem requires the manual de
 finition of a meaningful kernel given expert knowledge to achieve good sep
 arability. While Inductive Logic Programming (ILP) has long been establish
 ed as a versatile tool for hypothesis generation given expert knowledge es
 pecially given structured data\, on its own it performs poorly on noisy / 
 numeric data. Proposals have been made and experimentally validated to com
 bine the two approaches- using ILP to extract the features on which SVMs c
 an then operate. This talk aims to highlight the advantages of such hybrid
  approaches\, introduce some relevant work\, and discuss open challenges /
  areas of subsequent research.
LOCATION:TCM Seminar Room\, Cavendish Laboratory\, Department of Physics
END:VEVENT
END:VCALENDAR