Some Algorithms Of Support Vector Machine To Solve Large-scale Problems

Support Vector Machine (SVM) is a new generation of learning algorithm developed on the basis of the statistical learning theory. In 1992 Vapnik introduce it into the fields of machine learning. It got comprehensive attention after that. In the late 1990s SVM have been comprehensive and in-depth development. Now it becomes the standard technology in the fields of machine learning and data mining. It integrates the largest interval hyper-plane, Mercer nuclear, convex quadratic programming, sparse and relaxation of a number of variables and so on. so far. This paper introduces the basic support vector machine algorithm and some deformation. On the basis of the introduction we mainly study two methods to solve large-scale problems: one is support vector clustering technology, and the other is the support vector incremental and decrement learning algorithm.Support Vector clustering technology (SVC) has three main steps: first go through the dual problem of an optimization problem to find the minimum hyper-plane; secondly affirm the kind of the point through allocating it; finally adjust the parameters. The scale of the dual problem in first step is the scale of the input data. It is often the bottleneck of the whole operation. The solution of problems only relies on the support vector in the input space while the other input doesn't affect any results but increases in the clustering complexity. In this paper, based on a heuristic data reprocessing we improve the technology to avoid all the samples are involved in the training and narrow the sample space. The improvement greatly enhances the training speed.In this paper we improve the support vector decrement algorithm when we calculate the inverse of the matrix. We make use of the last result of the inverse after transform the matrix. We get the next inverse avoid compute it through a theorem by matrix inverse replace. We realized it in LSVM and find the runtime much less than it's before.