More efficient sparse multi-kernel based least square support vector machine

Abstract

Multiple Kernel Learning (MKL) has been one of the most important methods for learning kernels. Multi-kernel based east square support vector machine (LSSVM-MK) can be handled by semi-definite programming (SDP) and quadratically constrained quadratic program (QCQP). Unfortunately SDP and QCQP can only handle the problem with small scale sample and kernels size. In this paper we introduce a more effective algorithm to solve the LSSVM-MK with larger kernel size and sample size. The experimental results show that the proposed algorithm is more effective than SDP and QCQP in terms of the number of kernel matrixes and samples. © 2012 Springer-Verlag Berlin Heidelberg.