About Sparsity in Functional Relevance Learning in Generalized Learning Vector Quantization
Marika Kästner, Thomas Villmann, Michael L. Biehl · 2011
We propose a functional approach to relevance learning and matrix adaptation for learning vector quantization of high-dimensional functional data. We show how parametrization of the functional relevance profile or functional matrix learning can be established for a reasonable number of adaptive parameters. In particular we empha-size model sparsity in terms of structural sparsity and feature selection.