A unified framework for connectionist models

Abstract

A novel connectionist neural network model based an a propagation rule which contains Multilayer Perceptron (MLP) and Radial Basis Function (RBF) parts is introduced in this paper. The network using this propagation rule is known as a Conic Section Function Network (CSFN). Two different strategies have been used for training the network. The contact lens fitting problem has been considered to demonstrate the performance of the training algorithms. The performances of a standard MLP trained by back propagation, a fast back propagation with adapted learning rates, a standard RBFN using Matlab Neural Network software toolbox, and the proposed algorithm are compared for this particular problem.