Adaptive learning for complex-valued data

Apr, 2012

In this paper we propose a variant of the Generalized Matrix Learning Vector Quantization (GMLVQ) for dissimilarity learning on complex-valued data. Complex features can be encountered in various data domains, e.g. Fourier transformed mass spectrometry or image analysis data. Current approaches deal with complex inputs by ignoring the imaginary parts or concatenating real and imaginary parts in one real valued vector. In this contribution we propose a prototype based classication method, which allows to deal with complex-valued data directly. The algorithm is tested on a benchmark data set and for leaf recognition using Zernike moments. We observe that the complex version converges much faster than the original GMLVQ evaluated on the real parts only. The complex version has fewer free parameters than using a concatenated vector and is thus computationally more efficient than original GMLVQ

Link

@inproceedings{BunteESANN2012,
author = {Kerstin Bunte and Frank-Michael Schleif  and Michael Biehl},
title = {Adaptive learning for complex-valued data},
booktitle = {Proc. of the  20th "European Symposium on Artificial Neural Networks (ESANN)},
address = {Bruges, Belgium},
year = {2012},
pages = {387--392},
month = {"Apr."},
editor = {M. Verleysen},
publisher = {D-facto Publications},
isbn = {978-2-87419-049-0},
url = {http://www.cs.rug.nl/biehl/Preprints/2012-esann-complex.pdf},
note2 = {accepted for publication},
abstract = {In this paper we propose a variant of the Generalized Matrix Learning Vector Quantization (GMLVQ) for dissimilarity learning on complex-valued data.  Complex features can be encountered in various data domains, e.g. Fourier transformed mass spectrometry or image analysis data.  Current approaches deal with complex inputs by ignoring the imaginary parts or concatenating real and imaginary parts in one real valued vector.  In this contribution we propose a prototype based classication method, which allows to deal with complex-valued data directly. The algorithm is tested on a benchmark data set and for leaf recognition using Zernike moments.  We observe that the complex version converges much faster than the original GMLVQ evaluated on the real parts only.  The complex version has fewer free parameters than using a concatenated vector and is thus computationally more efficient than original GMLVQ},
}