Paper Info
Title
COFFIN: A Computational Framework for Linear SVMs
Abstract
In a variety of applications, kernel machines such as Support Vector Machines (SVMs) have been used with great success often delivering state- of-the-art results. Using the kernel trick, they work on several domains and even enable het- erogeneous data fusion by concatenating feature spaces or multiple kernel learning. Unfortu- nately, they are not suited for truly large-scale ap- plications since they suffer from the curse of sup- porting vectors, i.e., the speed of applying SVMs decays linearly with the number of support vec- tors. In this paper we develop COFFIN — a new training strategy for linear SVMs that effectively allows the use of on demand computed kernel feature spaces and virtual examples in the pri- mal. With linear training and prediction effort this framework leverages SVM applications to truly large-scale problems: As an example, we train SVMs for human splice site recognition in- volving 50 million examples and sophisticated string kernels. Additionally, we learn an SVM based gender detector on 5 million examples on low-tech hardware and achieve beyond the state- of-the-art accuracies on both tasks. Source code, data sets and scripts are freely available from http://sonnenburgs.de/soeren/coffin.
Year
Venue
Keywords
2010
ICML
data fusion,source code,feature space,kernel machine,support vector machine,string kernel
Field
DocType
Citations 
Graph kernel,Pattern recognition,Radial basis function kernel,Least squares support vector machine,Computer science,Support vector machine,Multiple kernel learning,Tree kernel,Polynomial kernel,Artificial intelligence,String kernel,Machine learning
Conference
39
PageRank 
References 
Authors
1.52
17
2
Name
Order
Citations
PageRank
Soeren Sonnenburg11104.54
Vojtěch Franc258455.78