摘要
支持向量机在大规模训练集上学习时,存在学习时间长、泛化能力下降的问题。路径跟踪算法具有O(n L)的时间复杂度,能够在多项式时间内求解大规模QP问题。分析了影响SVM分类超平面的主要因素,使用路径跟踪内点算法和核距离矩阵快速约简训练集,再用约简后的训练集重新训练SVM。实验结果表明,重新训练后的SVM模型得到了简化,模型的泛化能力也得到提高。
If the Support Vector Machine（SVM） is trained on large-scale datasets,the training time will be longer and the generalization capability will be descended.The time complexity of the path following interior point method is O（n L）,so it has been used to solve many large-scale Quadratic Programming（QP） problems.The main factors for constructing the separating hyper-plane of SVM are stated.The path following method and kernel distance matrix are used to reduce the training datasets,and the SVM is retrained with the reduced datasets.The experimental results show that the SVM model is simpler and the generalization capability is enhanced after using the reduced datasets to train the SVM.
出处
《计算机工程与应用》
CSCD
北大核心
2011年第3期160-162,169,共4页
Computer Engineering and Applications
基金
广西高校人才小高地建设创新团队计划基金(桂教人[2007]71号)
关键词
支持向量机
路径跟踪算法
核距离矩阵
泛化能力
Support Vector Machine（SVM） path following method kernel distance matrix generalization capacity
