Electrical Engineering
      and Computer Sciences

Electrical Engineering and Computer Sciences

COLLEGE OF ENGINEERING

UC Berkeley

Support Vector Machine Approximation using Kernel PCA

Narayanan Sundaram

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2009-94
June 18, 2009

http://www.eecs.berkeley.edu/Pubs/TechRpts/2009/EECS-2009-94.pdf

Support Vector Machine is a very important technique used for classification and regression. Although very accurate, the speed of SVM classification decreases with increase in the number of support vectors. This paper describes one method for reducing the number of support vectors through the application of Kernel PCA. This method is different from other proposed methods as we show that the exact choice of the reduced support vectors is not important as long as the vectors span a fixed subspace. This method reduces the number of support vectors by upto 90% without any significant degradation in performance. We also propose a heuristic to determine the reducibility of an SVM.


BibTeX citation:

@techreport{Sundaram:EECS-2009-94,
    Author = {Sundaram, Narayanan},
    Title = {Support Vector Machine Approximation using Kernel PCA},
    Institution = {EECS Department, University of California, Berkeley},
    Year = {2009},
    Month = {Jun},
    URL = {http://www.eecs.berkeley.edu/Pubs/TechRpts/2009/EECS-2009-94.html},
    Number = {UCB/EECS-2009-94},
    Abstract = {Support Vector Machine is a very important technique
used for classification and regression. Although
very accurate, the speed of SVM classification decreases
with increase in the number of support vectors.
This paper describes one method for reducing the
number of support vectors through the application of
Kernel PCA. This method is different from other proposed
methods as we show that the exact choice of the
reduced support vectors is not important as long as the
vectors span a fixed subspace. This method reduces
the number of support vectors by upto 90% without
any significant degradation in performance. We also
propose a heuristic to determine the reducibility of an
SVM.}
}

EndNote citation:

%0 Report
%A Sundaram, Narayanan
%T Support Vector Machine Approximation using Kernel PCA
%I EECS Department, University of California, Berkeley
%D 2009
%8 June 18
%@ UCB/EECS-2009-94
%U http://www.eecs.berkeley.edu/Pubs/TechRpts/2009/EECS-2009-94.html
%F Sundaram:EECS-2009-94