We develop a Local Deep Kernel Learning (LDKL) technique for efficient non-linear SVM prediction while maintaining classification accuracy above an acceptable threshold. LDKL learns a tree-based primal feature embedding which is high dimensional and sparse.

We develop a Local Deep Kernel Learning (LDKL) technique for efficient non-linear SVM prediction while maintaining classification accuracy above an acceptable threshold. LDKL learns a tree-based primal feature embedding which is high dimensional and sparse.

2014年10月2日 · In this talk, we develop LDKL – an efficient non-linear SVM classifier with prediction costs that grow logarithmically with the number of training points. We generalize Localized Multiple Kernel Learning so as to learn a deep primal feature embedding which is high dimensional and sparse.

Machine & Deep learning: Extreme classification, neural language modeling, multi-label learning, resource-efficient machine learning, machine learning for the Internet of Things, supervised learning.

LDKL could significantly bring down prediction costs as compared to RBF-SVMs while maintaining classi-fication accuracy above an acceptable threshold. The experiments also revealed that LDKL could yield bet-ter classification accuracies as compared to state-of-the-art methods for speeding up SVM prediction. For

In this paper, we propose an efficient nonlinear classifica- tion algorithm based on the Locally Linear Classifiers with Supervised Anchor Point Learning (LLC-SAPL). We take a fully supervised approach for learning both anchor points and …

http://research.microsoft.com/~manik/code/LDKL/download.html • LDKL learns a local, deep composite kernel for efficient non-linear SVM prediction • LDKL can be exponentially faster than the state-of-the-art • Efficiency is important during both training and prediction

The LDKL[1] learns a non-linear kernel K as a product of a global and a local kernel K(x i;x j) = K L(x i;x j)K G(x i;x j) LDKL y(x) = sign(Wt(x)˚ G(x)) (7) w k = X i iy i˚ L k (x i)˚ G(x i);˚ L 2RM (8) W = [w 1;:::;w M] (9) W(x) = W˚ L(x) (10) Dhruv Singal, Pranav Maneriker A study of kernel SVM approximation methods

2016年5月12日 · LDKL learns a tree-based primal feature embedding which is high dimensional and sparse. Primal based classification decouples prediction costs from the number of support vectors and the size of the training set and LDKL’s tree-structured features efficiently encode non-linearities while speeding up prediction exponentially over the state-of ...

Random features for large-scale kernel machines. In Advances in neural information processing systems (pp. 1177-1184).