A practical comparison of synergetic computer, restricted coulomb energy networks and multilayer perceptron

In this paper the performance of the recently proposed Synergetic Computer (SC) is discussed for the first time with respect to classification problems. A striking feature of SCs is their mathematical similarity to real physical effects and therefore the possibility of constructing synergetic hardware (nearest neighbour) as well as with feed forward networks (multilayered perceptron, restricted coulomb energy). Real world applications from the field of automated visual inspection are the base of our study. A practical evaluation with respect to error rate and computational effort for classification and learning shows the principal differences of the approaches.