Research On Recursive And Prune Learning Algorithm Of Functional Neural Network And Its Ensemble

Functional network is a new extension of the Artificial Neural Network (ANN). It has been successfully applied to many fields and made a lot of achievements. However, there are still many deficiencies in theory and application since functional network is a new branch. It is necessary to further propose new theory and learning algorithm in order to perfect the theory system and broaden the application field.Aiming at the inadequate of the present study of functional network, such as in the learning algorithm of functional network, it undoubtedly reduced the memory usage due to the selection of base function is not suitable. Research on the problem of how to design an optimal or approximate optimal functional network structure to improve the generalization ability, this paper obtains the following results:1) According to the shortage of selection of base function in the existing learning algorithm of functional network, then we propose a functional network learning algorithm with recursively base functions by completely consideration of base function selection in its learning process. The algorithm use a recursive method for solving matrix’s pseudo-inverse to achieve adaptive adjustment of base functions in functional neural network. The experimental results show that the learning algorithm can achieve high accuracy of calculation.2) Taking the complexity of functional neural networks into consideration, we propose a pruning algorithm based on the complexity of functional neural networks combined with the pruning algorithm to solve the functional neural network structure optimization problems. The function approximation shows that the algorithm which combine pruning algorithm with information entropy can effectively remove the base function of functional neural networks, and determine its internal architecture. At the same time it can improve the generalization ability of the functional neural networks.3) Consider that several functional neural networks to assemble into a functional network due to functional neurons is assembly parts of functional network. Research on functional neurons ensemble theory, method, and apply the ensemble network in practical problems to solve some engineering and scientific computing problem.The results of this paper have the important theory significance and the application value in both theory and application scope.