Supervised dimension reduction mappings

Apr, 2011

We propose a general principle to extend dimension reduction tools to explicit dimension reduction mappings and we show that this can serve as an interface to incorporate prior knowledge in the form of class labels. We explicitly demonstrate this technique by combining locally linear mappings which result from matrix learning vector quantization schemes with the t-distributed stochastic neighbor embedding cost function. The technique is tested on several benchmark data sets

Link

@INPROCEEDINGS{Bunte_ESANN11_Direduct,
author = {K. Bunte and M. Biehl and B. Hammer},
title = {Supervised dimension reduction mappings},
booktitle = {Proc. of the  19th "European Symposium on Artificial Neural Networks (ESANN)},
address = {Bruges, Belgium},
year = {2011},
pages = {281--286},
month = {"Apr."},
editor = {M. Verleysen},
publisher = {D-facto Publications},
url = {http://hdl.handle.net/11370/8a73386a-e6d3-4f19-a191-cd44a81d343e},
abstract = {We propose a general principle to extend dimension reduction tools to explicit dimension reduction mappings and we show that this can  serve as an interface to incorporate prior knowledge in the form of class labels.  We explicitly demonstrate this technique by combining locally linear mappings which result from matrix learning vector quantization schemes with  the t-distributed stochastic neighbor embedding cost function.  The technique is tested on several benchmark data sets},
}