A rapid learning orthogonal neural network for pattern recognition

Describes a neural network architecture similar to the one suggested by Kolmogorov´s (1957) existence theorem and a data processing method based on polynomial expansion. The resulting system, called an orthogonal neural network, can approximate any L₂ mapping function between the input and output vectors without using hidden layers or the backpropagation rule. A rapid learning algorithm that greatly reduces the training time is also introduced. Several systems built with this network are discussed. The rapid learning algorithm makes it quite possible to design a VLSI chip that could implement real-time testing (recall) and training