Median Generalized Learning Vector Quantization for Distance Data

This is a Java 7, fully MATLAB (R)-compatible implementation of _median generalized learning vector quantization for dissimilarity data_ (MGLVQ) as proposed by [Nebel, Hammer, Frohberg, and Villmann (2015)](https://doi.org/10.1016/j.neucom.2014.12.096). _Learning vector quantization_ (LVQ) is a classification algorithm which represents classes in terms of _prototypes_ and classifies data by assigning each data point to the class of the closest prototype ([Kohonen, 1995](https://doi.org/10.1007/978-3-642-97610-0_6)). Median versions of LVQ use data points as prototypes, that is: Each prototype corresponds exactly to a data point from the training data. This particular implementation of median LVQ represents the data in terms of pairwise distances or dissimilarities. The input to this algorithm is a m x m distance matrix D, a number of prototypes per class K and a m x 1 vector of training data labels Y, and the output is an array of prototypes W with K prototypes per class, given as data point indices. The training is performed according to an expectation maximization scheme suggested by ([Nebel, Hammer, Frohberg, and Villmann, 2015](https://doi.org/10.1016/j.neucom.2014.12.096)).