Supervised growing neural gas

Abstract

We present a new approach to supervised vector quantization inspired on growing neural gas network. An advantage of the new method is that it reduces the need for prior knowledge about the problem under study because it is able to determine at runtime the size of the codebook. Another advantage is that the training is less dependent on the initial state of the codebook vectors in contrast to methods like Learning Vector Quantization. Finally, it is shown that for some real datasets the classification performance is superior to other methods of supervised vector quantization. © 2012 Springer-Verlag.