Constructing general partial differential equations using polynomial and neural networks (Q1669297)

The aim of the paper is to introduce and develop the concept of a differential polynomial neural network, D-PNN. One starts with some background considerations on partial derivative terms substitution and GMDH polynomial neural networks and one describes the principles of D-PNN. Further, the D-PNN multi-layer architecture is developed. The algorithm has been implemented and numerical examples for instance for function approximations are presented. Comparisons of D-PNN and other classical ANN implementation results are shown and discussed.