A MEMORY-BASED SIGMA-PI-SIGMA NEURAL NETWORK (SPSNN)

Abstract

This paper presents a class of self-organized basis function networks called sigma-pi-sigma neural networks (SPSNNs). A neural network structure that uses small memory blocks plus additional operators as basic building blocks has been investigated. The output of the new structure is the sum of the outputs from several submodules. Each submodule consists of memory-based product-of-sum form neural networks. The memory contents in these submodules are adjusted during the learning process. The new structure can learn to implement static mapping that multilayer neural networks and radial basis function networks usually do. The new neural network structure demonstrates excellent learning convergence characteristics and requires small memory space.