OVERVIEW ON KERNELS FOR LEAST- SQUARES SUPPORT-VECTOR-MACHINE- BASED CLUSTERING: EXPLAINING KERNEL SPECTRAL CLUSTERING

Abstract

This letter presents an overview on some remarkable basics on kernels as well as the formulation of a clustering approach based
on least-squares support vector machines. Specifically, the method known as kernel spectral clustering (KSC) is of interest. We
explore the links between KSC and a weighted version of kernel principal component analysis (WKPCA). Also, we study the
solution of the KSC problem by means of a primal-dual scheme. All mathematical developments are carried out following an
entirely matrix formulation. As a result, in addition to the elegant KSC formulation, important insights and hints about the use
and design of kernel-based approaches for clustering are provided.