The Research On Neural Network With Quadratic Denominator Cubic Rational Spline Function Weight And Its Application

In order to solve the problem in traditional neural network, such as partial minimum frequently and low convergence speed, a new model and algorithm, named weight function neural network, is proposed in the monograph,"The new neural networks theory and method". The weight function neural network has a very sample structure and its training speed is very fast. It is proved in both theory and experimentation that the weight function neural network has a good approximation and generalization ability.Based on the research of weight function neural network, this paper combines the weight function neural network with the rational cubic spline with quadratic denominator, and constructs a new kind of artificial neural network - 3/2 type rational spline weight function neural network which weight function is rational cubic spline function with quadratic denominator. The contrast experiments with traditional neural network (BP and RBF) testified that the 3/2 type rational spline weight function neural network has better network performance.Combine with the application of neural network, in this paper, the 3/2 type rational spline weight function neural network's application in face recognition is given. The contrast experiments on ORL face database with traditional neural network testified at the 3/2 type rational spline weight function neural network has a much shorter training time and a higher recognition rate.