Multiple self-organizing maps: A hybrid learning scheme

Abstract

A scheme for hybrid learning (combining supervised and unsupervised techniques) based on multiple self-organizing maps (MSOM) is presented and its performance is compared with other methods in several pattern classification benchmarks using both synthetic and real data. The advantage of this approach is that the learning method is simplified with respect to a single SOM as the problem is divided into several networks which are trained in the standard unsupervised way. Classification is based on the SOM approximation of the probability densities and Bayesian decision. The resulting system classifies with higher accuracy than the single SOM and is comparable to other supervised methods on a wide range of problems, while maintaining the original properties of the SOM-like clustering and dimensionality reduction