计算机科学 ›› 2014, Vol. 41 ›› Issue (5): 239-242.doi: 10.11896/j.issn.1002-137X.2014.05.050

• 人工智能 • 上一篇    下一篇

一类改进的埃尔米特核函数

田萌,王文剑   

  1. 山西大学计算机与信息技术学院 太原030006;山西大学计算机与信息技术学院 太原030006
  • 出版日期:2018-11-14 发布日期:2018-11-14
  • 基金资助:
    本文受国家自然科学基金(60975035,61273291),山西省回国留学人员科研基金(2012-008)资助

A Set of Improved Hermite Kernel Function

TIAN Meng and WANG Wen-jian   

  • Online:2018-11-14 Published:2018-11-14

摘要: 核函数及其参数的选择是决定支持向量机(support vector machine,SVM)分类性能的关键。基于埃尔米特多项式,利用三角核函数构造并证明了一类改进的埃尔米特核函数——三角埃尔米特核函数。该类核函数含两个核参数,其中一个核参数可由样本点到样本均值的距离简单确定,而另一个核参数仅在自然数集中选取,从而简化了该类核函数的参数优化。在双螺线数据集、棋盘格数据集及7个UCI数据集上的实验表明,该类核函数比常见的多项式核函数、高斯核函数及文献[6]提出的埃尔米特核函数有着更好的泛化性能和鲁棒性。

关键词: 支持向量机,核选择,埃尔米特多项式,三角核函数

Abstract: The selection of kernel function and its parameters plays a significant role in support vector machine (SVM) classification algorithms.Based on Hermite polynomial and the triangular kernel function,a new set of kernel functions—triangular Hermite kernel was proposed.The triangular Hermite kernel has two parameters.One parameter is determined by the distance between sample points and sample mean,and the other parameter is chosen only from nonnegative integer.So the parameters of the triangular Hermite kernel can be optimized easily.The experimental results on bi-spiral data,checkerboard data and 7UCI data sets indicate that the new kernel achieves the competitive classification performance,compared with polynomial kernel,Gaussian kernel,and the previous Hermite kernel proposed in reference [6].

Key words: Support vector machine,Kernel selection,Hermite polynomial,Triangular kernel function

[1] Vapnik V.Statistical Learning Theory [M].New York:Wiley,1998
[2] 邓乃扬,田英杰.支持向量机——理论、算法与拓展[M].北京:科学出版社,2009:81-114
[3] Cristianini N,Shawe-Taylor J.An Introduction to Support Vector Machine[M].Cambridge:Cambridge University Press,2000
[4] Tsuda K,Akaho S,Kawanabe M,et al.Asymptotic properties of the Fisher kernel[J].Neural Computation,2004,6(1):115-137
[5] Sedat O,Chen C H,Cirpan H A.A set of new Chebyshev kernel functions for support vector machine pattern classification[J].Pattern Recognition,2011,44(4):1435-1447
[6] 张瑞,高红,张立伟.一类新的支持向量机核函数——埃尔米特核函数[J].山西大学学报:自然科学版,2012,5(1):38-42
[7] 张瑞,王文剑,张亚丹,等.基于支持向量机分类问题的勒让德核函数[J].计算机科学,2012,34(7):222-224 (下转第274页)(上接第242页)
[8] Schlkopf B,Smola A.Learning with Kernels[M].Cambridge:MIT Press,2002
[9] Cristianini N,Shawe-Taylor J,Elisseeff A,et al.On kernel-target alignment[C]∥Proceedings of Advances in Neural Information Processing Systems.2001:367-373
[10] Wang Wen-jian J,Xu Zong-ben,Lu Wei-zhen,et al.Determination of the spread parameter in the Gaussian kernel for classification and regression[J].Neurocomputing,2003,5(10):643-663
[11] Paclík P,Novoviˇ ová J,Pudil P,et al.Road sign classification using Laplace kernel classifier[J].Pattern Recognition Letters,2000,1(13/14):1165-1173
[12] Fleuret F,Sahbi H.Scale-invariance of support vector machines based on the triangular kernel[C]∥Proceedings of 3rd International Workshop on Statistical and Computational Theories of Vision.Nice,2003

No related articles found!
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!