Adaptive least square kernel algorithms and applications
Anthony Kuh · 2003
This paper discusses adaptive online kernel algorithms and an application of these algorithms to signal processing problems. The support vector machine (SVM) is a kernel method technique that has gained widespread acceptance in solving pattern classification and regression problems. SVM solutions generally involve solving a quadratic programming problem making it more difficult for applying these methods to adaptive signal processing problems. In previous work a variant of the SVM has been developed called the least squares SVM (LS-SVM). A solution to the algorithm can be found by solving a set of linear equations which makes an online adaptive implementation of the algorithm feasible. After discussing some of the differences between the SVM and the LS-SVM we present an adaptive LS-SVM solution and discuss signal processing applications of these algorithms.