Generalized potential function neural net classification

A new method of construction of neural nets is presented, based on a generalization of potential function classification. The construction is direct and much simpler computationally than backpropagation training. The method has demonstrated superior classification performance and more reliable indication of the confidence of classification for complex classes, compared to backpropagation training, Specht´s probabilistic neural network, nearest neighbor, and simple Gaussian parametric classifiers. An example of classification of vehicle vibration spectra is presented