An ANN model perceptron algorithm using generalized matrix inversion

Abstract

The application of generalized matrix inversion to artificial neural network model perceptron algorithm is described. This method permits real number inputs to the perceptron besides binary inputs, and can provide a solution depending upon error. Further it can provide recursive improvement to the solution depending upon whether the new input vector is linearly dependent or independent of the previous, input set of vectors; also this method permits to check for consistency of the solution. Thus redundancy and inconsistency can be checked. Also we point out the close relationship of Chu-Hsich algorithm to Ivakhnenko Group method of data handling (GMDH) that builds up a multinomial combination of input components. We illustrate this method using the problem of reconstruction of a magic square.