Computer Engineering and Applications ›› 2011, Vol. 47 ›› Issue (3): 190-195.DOI: 10.3778/j.issn.1002-8331.2011.03.057

• 图形、图像、模式识别 • Previous Articles     Next Articles

Comparison of three methods for training smooth support vector classification

TU Wengen,XIONG Jinzhi,YUAN Huaqiang   

  1. Engineering and Technology Institute,Dongguan University of Technology,Dongguan,Guangdong 523808,China
  • Received:2009-04-29 Revised:2009-06-21 Online:2011-01-21 Published:2011-01-21
  • Contact: TU Wengen



  1. 东莞理工学院 工程技术研究院,广东 东莞 523808
  • 通讯作者: 涂文根

Abstract: Smooth Support Vector Classification(SSVC) originally is unconstrained mathematical programming with convex and smooth,and is a method for solving Support Vector Classification(SVC) quickly.The BFGS-Armijo and Newton-Armijo algorithms have been used to train SSVC,and the latter has faster speed.Newton-PCG algorithm is just enough method for unconstrained problem which has better speed than Newton in theory.On the numerical experimentation of using BFGS-Armijo,Newton-Armijo and Newton-PCG to train SSVC,this paper gives analysis and comparison among the three algorithms,and proves that Newton-PCG has the best result.

Key words: pattern recognition, Smooth Support Vector Machine(SSVM), classification, Newton-PCG

摘要: 光滑支持向量分类机(SSVC)是支持向量分类机(SVC)的快速求解模型,本质上是求解数学规划中具有光滑性和强凸性的无约束最优化问题。BFGS-Armijo和Newton-Armijo算法被用来训练SSVC,相比而言后者拥有更快的训练速度;牛顿-预优共轭梯度法(Newton-PCG)适用于求解无约束的最优化问题,理论上快于一般的Newton类算法。使用Newton-Armijo、BFGS-Armijo和Newton-PCG三种算法来训练光滑支持向量分类机,根据数值实验结果进行分析比较,证明了Newton-PCG算法有更优的效果。

关键词: 模式识别, 光滑支持向量机, 分类, Newton-PCG算法

CLC Number: