Structural learning of neurofuzzy GMDH with Minkowski norm

Abstract

There have been many studies of mathematical models of neural networks. However, there always arises a problem of determining their optimal structures due to the lack of prior information. We propose a procedure for the structure clarification of neurofuzzy GMDH model whose building blocks are represented by the radial basis functions network. The proposed method prunes the unnecessary links and units from the larger network to identify, as well as to clarify the network structure by minimizing the Minkowski norm of the derivatives of the building blocks.