@inproceedings {pub2523,
	title = {Rejection Strategies for Learning Vector Quantization},
	author = {Lydia Fischer AND Barbara Hammer AND Heiko Wersing},
	year = {2014},
	month = {April},
	abstract = {Reject options in multiclass classification schemes enhance a given classifier
by a confidence value which quantifies the security of the classification decision
taken for a given data point.
Based on this value, man-machine interaction or an improvement of the classification
task with human expert knowledge becomes possible in regions with class overlap or outliers.
While there exist efficient reject options for popular classifiers such as
$k$- nearest neighbor classifiers, Bayes approaches, or support
vector machines, only few reject options have been proposd for prototype based classifiers.
In this contribution,
we propose two different reject options for prototype based classification
approaches and we demonstrate their suitability for recent learning vector quantization
schemes. The adequate behavior of the reject options is experimentally verified in
an artifical example where ground truth in the sense of an optimum
Bayesian decision is available, as well as in several benchmark data sets.},
	publisher = {d-Side},
	booktitle = {22th European Symposium on Artificial Neural Networks ESANN 2014},
	editor = {Michel Verleysen},
	city = {Bruges, Belgium},
	pages = {41-46}
}