A new orthogonal neural network
Ching‐Shiow Tseng, Shiow-Shung Yang · 2002
This paper presents a new neural network based on orthogonal functions. This single-layer neural network may avert the problems of traditional feedforward neural networks such as the determination of the numbers of layers and processing elements, and the initial values of weights. The processing elements of the neural network are composed of the expansion terms of Legendre polynomials. The required number of processing elements is determined according to the desired output accuracy. Because the weights are unique, the training of the weights will converge rapidly. Two experiments are given to demonstrate the performance of the proposed neural network. The results show that the neural network has excellent performance in convergence time and in finding a near-global solution.