BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Local Deep Kernel Learning for Efficient Non-linear SVM Prediction
   - Manik Varma (Microsoft Research India)
DTSTART:20140915T100000Z
DTEND:20140915T110000Z
UID:TALK54337@talks.cam.ac.uk
CONTACT:Microsoft Research Cambridge Talks Admins
DESCRIPTION:The time taken by an algorithm to make predictions is of criti
 cal importance as machine learning transitions to becoming a service avail
 able on the cloud. Algorithms that are efficient at prediction can service
  more calls and utilize fewer cloud resources and thereby generate more re
 venue. They can also be used in real time applications where predictions n
 eed to be made in micro/milliseconds.\n\nNon-linear SVMs have defined the 
 state-of-the-art on multiple benchmark tasks. Unfortunately\, they are slo
 w at prediction with costs that are linear in the number of training point
 s. This reduces the attractiveness of non-linear SVMs trained on large amo
 unts of data in cloud scenarios.\n\nIn this talk\, we develop LDKL -- an e
 fficient non-linear SVM classifier with prediction costs that grow logarit
 hmically with the number of training points. We generalize Localized Multi
 ple Kernel Learning so as to learn a deep primal feature embedding which i
 s high dimensional and sparse. Primal based classification decouples predi
 ction costs from the number of support vectors and our tree-structured fea
 tures efficiently encode non-linearities while speeding up prediction expo
 nentially over the state-of-the-art. We develop routines for optimizing ov
 er the space of tree-structured features and efficiently scale to problems
  with millions of training points. Experiments on benchmark data sets reve
 al that LDKL can reduce prediction costs by more than three orders of magn
 itude over RBF-SVMs in some cases. Furthermore\, LDKL leads to better clas
 sification accuracies as compared to leading methods for speeding up non-l
 inear SVM prediction. \n
LOCATION:Auditorium\, Microsoft Research Ltd\, 21 Station Road\, Cambridge
 \, CB1 2FB
END:VEVENT
END:VCALENDAR