A Geometric Approach to Train SVM on Very Large Data Sets

摘要：

Reduced set method is an important approach tospeed up support vector machine(SVM)training onlarge data sets.Existing works mainly focused onselecting patterns near the decision boundary forSVM training by applying clustering,nearestneighbor algorithm and so on.However,on verylarge data sets,these algorithms require hugocomputational overhead,and thus the totalrunning time is still enormous.In this paper,an intuitive geometric method is developed to selectconvex hull samples in the feature space for SVMtraining,which has a time complexity that is linearwith training set size n.Experiments on real datasets show that the proposed method not onlypreserves the generalization performance of theresult SVM classifiers,but outperforms existingscale-up methods in terms of training time andnumber of support vectors.

作者: Zhi-Qiang Zeng Hua-Rong Xu Yan-Qi Xie Ji Gao

作者单位: Department of Computer Science and Technology,Xiamen University of Technology,Xiamen 361024,China Department of Physics and Electromechanical,Xiamen University,Xiamen 361005,China Department of Computer Science and Engineering,Zhejiang University,Hangzhou 310027,China

会议类型: 国际会议

会议名称: 2008 3rd International Conference on Intelligent System and Knowledge Engineering(第三届智能系统与知识工程国际会议)(ISKE 2008)

会议地点: 厦门

会议语种:英文

页码: 991-996

在线出版日期: 2008-11-17（万方平台首次上网日期，不代表论文的发表时间）