A New Neuron Model Based on Multilayer Perceptron and Radial Basis Transfer Function

In order to effectively optimize the solution of feedforward neural network, a new general transfer function is proposed that effectively unifies the inputs of multiplayer perceptron and radial basis function to provide flexible decision border. Based on this, a new learning algorithm based on gradient descent and error propagation is proposed. Several pattern classification examples simulations are made to verify the validity of the proposed algorithm by comparing the proposed transfer function and learning algorithm with BP algorithm adding momentum term, CSFN and RBF. The experimental results show that the proposed method has the merits of simple network structure, quick training speed and high classification accuracy