Neural nets versus synergetic computers. A practical comparison

In an increasing number of industrial application pattern recognition plays an important role. Conventional algorithms for this process, as e. g. the minimum distance and the nearest neighbour classifier, have lately been supplemented by neural nets like perceptions and restricted coulomb energy nets. Based on the mathematical description of self-organizing processes Haken recently proposed a new approach for pattern recognition: synergetic computers. In this paper we compare these classifiers with respect to real world classification problems. Optical and acoustic patterns taken from quality assurance tasks build the basis for this comparison. Apart from the error rates, we evaluate the number of prototypes and the effort for classification and training as important criteria for the applicability of theses classifiers.