Fuzzy skeletons and statistical learning theory

Pando Georgiev, Anca Ralescu.

We define a fuzzy subspace skeleton of data points and propose an algorithm for finding it. Such a skeleton has direct applications in statistical learning theory. We propose a new type of classifiers: fuzzy skeleton classifiers, which might be a better alternative to Support Vector Machines in some cases. Another application is presented to the unsupervised learning - Blind Signal Separation, based on mild sparsity assumptions. Our methods are illustrated by examples. Potential application include problems from bioinformatics as separation of protein spectra, gene expressions, etc., as well as any problems requiring signal separation in which Independent Component Analysis doesn’t work, or gives unsatisfactory results.

PDF full paper