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:20140917T100000Z
DTEND:20140917T110000Z
UID:TALK53740@talks.cam.ac.uk
CONTACT:Dr Jes Frellsen
DESCRIPTION:*Abstract*\n\nThe time taken by an algorithm to make predictio
 ns is of critical importance as machine learning transitions to becoming a
  service available on the cloud. Algorithms that are efficient at predicti
 on can service more calls and utilize fewer cloud resources and thereby ge
 nerate more revenue. They can also be used in real time applications where
  predictions need to be made in micro/milliseconds.\n\nNon-linear SVMs  ha
 ve defined the state-of-the-art on multiple benchmark tasks. Unfortunately
 \, they  are slow at prediction with costs that are linear in the number o
 f training points. This reduces the attractiveness of non-linear SVMs trai
 ned on large amounts of data in cloud scenarios.\n\nIn this talk\, we deve
 lop LDKL -- an efficient non-linear SVM classifier with prediction costs t
 hat grow logarithmically with the number of training points. We generalize
  Localized Multiple Kernel Learning so as to learn a deep primal feature e
 mbedding which is high dimensional and sparse. Primal based classification
  decouples prediction costs from the number of support vectors and our tre
 e-structured features efficiently encode non-linearities while speeding up
  prediction exponentially over the state-of-the-art. We develop routines f
 or optimizing over the space of tree-structured features and efficiently s
 cale to problems with millions of training points. Experiments on benchmar
 k data sets reveal that LDKL can reduce prediction costs by more than thre
 e orders of magnitude over RBF-SVMs in some cases. Furthermore\, LDKL lead
 s to better classification accuracies as compared to leading methods for s
 peeding up non-linear SVM prediction.\n\nLDKL is available on AzureML\, Mi
 crosoft's cloud machine learning platform\, and I will briefly discuss how
  it can be used to develop a highly performant virus and malware classifie
 r which needs to predict potential threats every time files are opened\, s
 aved or executed on hundreds of millions of machines on a daily basis.\n\n
 \n*Spaker's brief biography*\n\nManik Varma is a researcher at Microsoft R
 esearch India where he helps manage the Machine Learning and Optimization 
 area. Manik received a bachelor's degree in Physics from St. Stephen's Col
 lege\, University of Delhi in 1997 and another one in Computation from the
  University of Oxford in 2000 on a Rhodes Scholarship. He then stayed on a
 t Oxford on a University Scholarship and obtained a DPhil in Engineering i
 n 2004. Before joining Microsoft Research\, he was a Post-Doctoral Fellow 
 at the Mathematical Sciences Research Institute Berkeley. He has been an A
 djunct Professor at the Indian Institute of Technology (IIT) Delhi in the 
 Computer Science and Engineering Department since 2009 and jointly in the 
 School of Information Technology since 2011. His  research interests lie i
 n the areas of machine learning\, computational advertising and computer v
 ision. He has served as an Area Chair for machine learning and computer vi
 sion conferences such as NIPS\, ICCV and CVPR. He has been awarded the Mic
 rosoft Gold Star award and has won the PASCAL VOC Object Detection Challen
 ge.\n
LOCATION:Engineering Department\, CBL Room BE-438
END:VEVENT
END:VCALENDAR