Discriminative Visualization by Limited Rank Matrix Learning

2008

We propose an extension of the recently introduced Generalized Matrix Learning Vector Quantization (GMLVQ) algorithm. The original algorithm provides a discriminative distance measure of relevance factors, aided by adaptive square matrices, which can account for correlations between different features and their importance for the classification. We extend the scheme to matrices of limited rank corresponding to low-dimensional representations of the data. This allows to incorporate prior knowledge of the intrinsic dimension and to reduce the number of adaptive parameters efficiently. The case of two- or three-dimensional representations constitutes an efficient visualization method. The identification of a suitable projection is not treated as a preprocessing step but as an integral part of the supervised training

Link

@techreport{MLR0308Bunte2008a,
author = {K. Bunte and P. Schneider and B. Hammer and F.-M. Schleif and T. Villmann and M. Biehl},
title = {Discriminative Visualization by Limited Rank Matrix Learning},
number = {MLR-03-2008},
issn = {1865-3960},
url = {https://www.techfak.uni-bielefeld.de/~fschleif/mlr/mlr_03_2008.pdf},
journal = {Machine Learning Reports},
institution = {Leipzig University},
volume = {2},
pages = {37--51},
year = {2008},
abstract = {We propose an extension of the recently introduced Generalized Matrix Learning Vector Quantization (GMLVQ) algorithm.  The original algorithm provides a discriminative distance measure of relevance factors, aided by adaptive square matrices,  which can account for correlations between different features and their importance for the classification.  We extend the scheme to matrices of limited rank corresponding to low-dimensional representations of the data.  This allows to incorporate prior knowledge of the intrinsic dimension and to reduce the number of adaptive parameters efficiently.  The case of two- or three-dimensional representations constitutes an efficient visualization method.  The identification of a suitable projection is not treated as a preprocessing step but as an integral part of the supervised training},
}