Backpropagation algorithm for a generalized neural network structure

Abstract

The author presents a generalized neural network structure and the associated backpropagation learning algorithm. This structure is an extension of a neural net structure presented by P.J. Werbos (1988, 1990). Generalization is aimed at ease of implementation and flexibility in structure selection. It is shown how certain network structures can be accommodated using this general structural form. Networks that are investigated include feedforward, recurrent, and memory networks.<>