Nonlinear modelling of back-propagation neural networks

Summary form only given, as follows. A method for approximating an existing N-input backpropagation neural network (NN) using an N-dimensional (N-D) polynomial discriminant (PD) function is discussed. First, the hidden unit activation functions are approximated by polynomials. Then, after multiplying out the resulting composition of polynomials, the final network PD results. This technique is a practical method for developing the N-D PD. The approximation is applied to NN´s designed to perform classification and filtering tasks. The resulting polynomials differ substantially from PDs developed via other more traditional techniques