BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:(FPT Preview) A Scalable FPGA Architecture for Non-linear SVM Trai
 ning - Markos Papadonikolakis (Imperial College)
DTSTART:20081128T110000Z
DTEND:20081128T120000Z
UID:TALK15413@talks.cam.ac.uk
CONTACT:Dr George A Constantinides
DESCRIPTION:Support Vector Machines (SVMs) is a popular supervised learnin
 g method\, providing state-of-the-art accuracy in various classification t
 asks. However\, SVM training is a time-consuming task for large-scale prob
 lems. This work proposes a scalable FPGA architecture which targets a geom
 etric approach to SVM training based on Gilbert’s algorithm using kernel
  functions. The architecture is partitioned into floating-point and fixed-
 point domains in order to efficiently exploit the FPGA’s available resou
 rces for the acceleration of the non-linear SVM training. Implementation r
 esults present a speed-up factor up to three orders of magnitude of the mo
 st computational expensive part of the algorithm compared to the algorithm
 ’s software implementation.
LOCATION:Mahanakorn Laboratory\, EEE
END:VEVENT
END:VCALENDAR