Support Vector Machine Kernel Function Selection With Sparse Representation

Support vector machine is a new learning method based on the kernel, the kernel function plays the role in the play a decisive role in support vector machine. Geometry of different kernel functions contain metric characteristic different, choose different kernel function support vector machine leads to generalization ability difference. Due to the choice of kernel function has an important effect on the performance of support vector machine model construction, how to effectively carry out the selection of kernel function is currently an important research topic in the field of research on vector machine.At present, about the selection of kernel function of support vector machine mostly depends on experience, lack of theoretical guidance of concrete, which obviously has great limitations and uncertainties. Therefore, to construct a can with the sample prior information,but also consider the geometric kernel function is the metric features supervised SVM kernel function selection mechanism, effectively avoid the blindness. This paper made some research on SVM kernel function selection method, the main work is:1、the kernel function theory, to measure the characteristics of different kernel functions are analyzed and discussed. The geometrical features from the kernel contains start with,respectively, introduced the commonly used kernel function, the distance and the angle of the Riemann metric three feature, geometric meaning and measurement features of the described.2,、in the early stage of the research work presented in a selection method of kernel function of support vector machine based on the characteristics of the sample distribution.The method for a given sample data are hyper sphere description, establish the sample distribution of energy entropy function, and calculate the energy entropy of each sample,build the sample distribution of discriminant function and calculation of its results; then the similarity selection the type of kernel function and kernel function according to the determination result of geometric properties; finally, parameter optimization and support vector machine model determine. This is a kind of can according to the specific object information of the samples, and can both contain the metric properties of kernel function of support vector machine is the guidance of kernel function selection method, and has the characteristics of fast operation, very suitable for real-time online support vector machine model predictive control places.3 、 the sparse representation attribute theory to the sample data representation and modeling ability, this paper puts forward a design method of support vector machine for sparse representation under the. The first sample data given by the specific issues of constructing sparse dictionary different kernel function model; then said coefficient according to sparse attributes, the selection of kernel function of support vector machine model suitable for practical problems; finally based on correlation method and optimization of the corresponding parameters and ultimately determine the support vector machine model. The method is a kind of kernel function of support vector machine instruction selection and practical methods, effectively overcomes the defects of traditional support vector machine model selection method is artificially specified the type of kernel function and cause model cannot achieve optimal performance shortcomings.Proved by the simulation results, the method can effectively use the sample data of prior information of practical problems, both the metric properties of different kernel function of support vector machine modeling, has strong generalization ability, is a kind of practical method of kernel function selection guidance, can improve the generalization ability. The contents of this study, help to enrich the SVM kernel function selection, the model of support vector machine has a certain guiding significance.