In this paper we recall two kernel methods for discriminant analysis.
The first one is the kernel counterpart of the ubiquitous Linear Discriminant
Analysis (Kernel-LDA), while the second one is a method we named Kernel
Springy Discriminant Analysis (Kernel-SDA). It seeks to separate classes
just as Kernel-LDA does, but by means of defining attractive and repulsive
forces. First we give technical details about these methods and then
we employ them on phoneme classification tasks. We demonstrate that
the application of kernel functions significantly improves the recognition
accuracy.
