Radial basis function networks (RBFNN), with the simplicity of its single-hidden layer structure and the high speed of its training, is applied very extensively as a kind of neural networks. The key point in design of radial basis function networks is to specify the number and the locations of the centers. If the number of centers (or the hidden layer units) is chosen too much, over-fitting results and the generalization are getting worse. On the contrary, if the centers chosen is too little, the network is not enough to study the training samples that the performance of networks, for example, generalization will become bad. In recent years, a various kinds of algorithms to design radial basis function neural networks have been developed, but their common disadvantage is poor generalization, and most of them are used in the field of function approximation. Especially, in the application of classification, the advantage of radial basis function neural networks lies in achieving high accuracy by taking place of nonlinear algorithm with linear algorithm. So, radial basis function neural network is a kind of neural networks with the performance of high convergence and accuracy. But, when the radial basis function neural network is used to solve the classification of high-dimension data, the generalization of neural network determined by past center-chosen algorithms is very poor. Support vector machine is a new learning method based on VC theory, good generalization is required by minimizing the upper abound of expected risk. It has wider significance to mine the similarity and difference between new learning model and traditional one in order to improve the generalization of the latter.Two parts are involved in present paper:In chapter 2,firstly, it is to compare the discrepancy in solving linear classification problem between the linear-SVM methods and single layer perception. Secondly, by dwelling on the nonlinear SVM method, it is concluded that the decision function of nonlinear SVM method is similar to the ones of traditional RBF neural networks. Thirdly , the definition of center is presented and directly explained by two especial example . Finally an algorithm of RBF neural network is put forward .In chapS, the experiment showed that the method could fast construct an RBFN classifier and the performance of the classifier was better than the best result previously reported. |