%A O. L. mangasarian and David R. Musicant
%T Successive Overrelaxation for Generalized Support Vector Machines
%D November 1998
%R 98-18
%I COMPUTER SCIENCES DEPARTMENT, UNIVERSITY OF WISCONSIN
%C MADISON, WI
%X
Successive overrelaxation (SOR) for symmetric linear complementarity
problems and quadratic programs \cite{man:77,man:91,lt:93c} is used to
train a support
vector machine (SVM) \cite{vap:95,cm:98} for discriminating between
the elements of two massive datasets, each with millions
of points. Because SOR handles one point at a time,
similar to Platt's sequential minimal optimization (SMO) algorithm
\cite{jp:98} which handles two constraints at a time, it can process
very large datasets that need not reside in memory. The algorithm
converges linearly to a solution. Encouraging numerical results on
very large datasets
that cannot be processed by conventional linear or quadratic
programming methods are presented.