Adapted filter banks in machine learning: applications in biomedical signal processing

Abstract

The theory of signal-adapted filter banks has been developed in signal compression in recent years and only rarely be applied to other applications fields such as machine learning. In this paper, we propose lattice structure based signal-adapted filter banks and time-scale atoms, respectively, for the construction of morphological local discriminant bases and hybrid wavelet-support vector classifiers. The first mentioned method is a more powerful construction of the recently introduced local discriminant bases algorithm which employs, in addition to the conventional wavelet-packet tree adjustment, an adaptation of the analyzing time-scale atoms. The latter mentioned method utilizes adapted wavelet decompositions which are tailored for support vector classifiers with radial basis functions as kernels. For both methods, we present applications in biomedical signal processing.