Material Detail

Distance Metric Learning for Kernel Machines

Distance Metric Learning for Kernel Machines

This video was recorded at NIPS Workshops, Whistler 2010. Recent work in metric learning has significantly improved the state-of-the-art in k-nearest neighbor classification. However, Support vector machines (with RBF kernels) are arguably the most popular class of classification algorithms that uses distance metrics to compare examples. In this talk I will introduce support vector metric learning (SVML), an algorithm that seamlessly combines both by learning a Mahalanobis metric at the same time as the RBF-SVM decision boundary. SVML is an effective tool for automatically pre-processing data sets for classification, as well as visualizing the structure of SVM decision boundaries. We demonstrate the capabilities (and shortcomings) of our algorithm on 10 benchmark data sets of varying sizes and difficulties.


  • User Rating
  • Comments
  • Learning Exercises
  • Bookmark Collections
  • Course ePortfolios
  • Accessibility Info

More about this material


Log in to participate in the discussions or sign up if you are not already a MERLOT member.