Efficient sparse approximation of support vector machines solving a kernel Lasso

Marcelo Aliquintuy, Emanuele Frandi, Ricardo Ñanculef, Johan A.K. Suykens

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

© Springer International Publishing AG 2017. Performing predictions using a non-linear support vector machine (SVM) can be too expensive in some large-scale scenarios. In the non-linear case, the complexity of storing and using the classifier is determined by the number of support vectors, which is often a significant fraction of the training data. This is a major limitation in applications where the model needs to be evaluated many times to accomplish a task, such as those arising in computer vision and web search ranking. We propose an efficient algorithm to compute sparse approximations of a non-linear SVM, i.e., to reduce the number of support vectors in the model. The algorithm is based on the solution of a Lasso problem in the feature space induced by the kernel. Importantly, this formulation does not require access to the entire training set, can be solved very efficiently and involves significantly less parameter tuning than alternative approaches. We present experiments on well-known datasets to demonstrate our claims and make our implementation publicly available.
Original languageEnglish
Title of host publicationEfficient sparse approximation of support vector machines solving a kernel Lasso
Pages208-216
Number of pages9
ISBN (Electronic)9783319522760
DOIs
Publication statusPublished - 1 Jan 2017
EventLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) -
Duration: 1 Jan 2019 → …

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10125 LNCS
ISSN (Print)0302-9743

Conference

ConferenceLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Period1/01/19 → …

Fingerprint Dive into the research topics of 'Efficient sparse approximation of support vector machines solving a kernel Lasso'. Together they form a unique fingerprint.

  • Cite this

    Aliquintuy, M., Frandi, E., Ñanculef, R., & Suykens, J. A. K. (2017). Efficient sparse approximation of support vector machines solving a kernel Lasso. In Efficient sparse approximation of support vector machines solving a kernel Lasso (pp. 208-216). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10125 LNCS). https://doi.org/10.1007/978-3-319-52277-7_26