| Statistical Learning Theory is put forward by Vapnik and others, which stressly research statistical rules and learning characters based on small samples. It sets up a preferable theoretical foundation for machine learning problem. At the same time, Support Vector Machine is no other than succeed realization of STL, which looks after a sort of balance on the basis of finite sample information between model complexity and learning ability for obtaining the optimal generalization quality. Compared with some traditional learning methods based ERM principle such as neural network, SVM has greater generalization performance.SVM was firstly used to settle classification problem. Recently, it is applied remarkably in field of regression. Therefore study and perfection of SVM and its realized algorithm have important significance. And research about non-linear SVM applicable system model can enrich support vector machine theory and algorithm, further prompting its application in some different fields.Paper content is as follows:(1) Embedded analysis and discuss about SVM from machine-learning and statistical-learning angle, in theory, giving some abstract and conclusions to svm.(2) Aiming at classification problem, based on the analysis of common norm in structural risk to control model complexity, 1-norm and 8-norm linear programming SVM are presented, and adopt simulation data to carry out numerical experiments for different svm models.(3) Aiming at regression, 1-norm and 8-norm linear programming regression SVM are advanced, through solving counterpart programming to improve the speed of training time. At the same time, simulation experience is carried out about 8-norm linear programming regression svm model.(4) Build up svm system model of 1 -norm linear programming and discuss some questions in the process of practical application, such as choose and optimization of parameter, choose of training sample, data pretreatment and comparison with classic svm. |
PDF Full Text Request