A Research On Fast Algorithms Of Support Vector Machine

Support vector machine (SVM) is introduced by Vapnik based on StatisticalLearning Theory (SLT). It is a new machine learning algorithm. SVM solves the smallsample problem mainly and finds the best compromise between the complexity of themodel and the learning ability in order to obtaining the best generalization ability. AndSVM has been applied in many fields successfully, such as written digit recognition,image detection and text classification so on. However, in practical applications, SVMreveals some shortcomings, such as the slow training speed, the large calculation andthe parameters selection based on experience so on. These shortcomings inhibit thesuccess of the SVM method in large-scale practical application. Thus, abandon theuseless history of the sample points under the premise of what is no loss ofclassification accuracy to extract support vectors is an important issue to consider in theSVM learning algorithm.In order to improve the training efficiency of SVM, First, a new algorithm basedon the some rules has been proposed by extracting a majority of support vectors,denoted by BD-SVM(Based on Distance) algorithm and BS-SVM(Based on theSimilarity)algorithm. An experiment, which was done on large-scale datasets, wascomparing with the standard SVM. And we should recognize that it’s an effective roadto make the standard SVM better with the similar accuracy.