About kernel latent variable approaches and SVM

It is demonstrated, that kernel latent variables approaches have a comparable predictive power with the set of kernel approaches based on regularization (e.g. Support Vector Machines).

Abstract

Abstract The aim of this paper is to demonstrate, that kernel latent variables approaches have a comparable predictive power with the set of kernel approaches based on regularization (e.g. Support Vector Machines). Kernel latent variable approaches are an alternative to kernel ridge regression, in the same way as PCR or PLS are the alternative approaches to Ridge Regression. Performance of these approaches is demonstrated for simulated data sets and microarray data set. Copyright © 2006 John Wiley & Sons, Ltd.