Scaling Gaussian RBF kernel width to improve SVM classification

Qun Chang; Qingcai Chen; Xiaolong Wang

Scaling Gaussian RBF kernel width to improve SVM classification

Qun Chang^*
, Qingcai Chen
, Xiaolong Wang

^*Corresponding author for this work

School of Computer Science and Technology, Harbin Institute of Technology

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

43 Scopus citations

Abstract

Support vector classification with Gaussian RBF kernel is sensitive to the kernel width, Small kernel width may cause over-fitting, and large one under-fitting. The so-called optimal kernel width is merely selected based on the tradeoff between under-fitting loss and over-fitting loss. So, there exists urgent need to further reduce the tradeoff loss. To circumvent this, we scale the kernel width in a distribution-dependent way. Experiments validate the feasibility of this method. Existing problems are also discussed.

Original language	English
Title of host publication	Proceedings of 2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05
Pages	19-22
Number of pages	4
State	Published - 2005
Externally published	Yes
Event	2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05 - Beijing, China Duration: 13 Oct 2005 → 15 Oct 2005

Publication series

Name	Proceedings of 2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05
Volume	1

Conference

Conference	2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05
Country/Territory	China
City	Beijing
Period	13/10/05 → 15/10/05

Cite this

@inproceedings{2b0b1f752e4a4f7694485c07faa3954b,

title = "Scaling Gaussian RBF kernel width to improve SVM classification",

abstract = "Support vector classification with Gaussian RBF kernel is sensitive to the kernel width, Small kernel width may cause over-fitting, and large one under-fitting. The so-called optimal kernel width is merely selected based on the tradeoff between under-fitting loss and over-fitting loss. So, there exists urgent need to further reduce the tradeoff loss. To circumvent this, we scale the kernel width in a distribution-dependent way. Experiments validate the feasibility of this method. Existing problems are also discussed.",

author = "Qun Chang and Qingcai Chen and Xiaolong Wang",

year = "2005",

language = "英语",

isbn = "0780394224",

series = "Proceedings of 2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05",

pages = "19--22",

booktitle = "Proceedings of 2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05",

note = "2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05 ; Conference date: 13-10-2005 Through 15-10-2005",

}

Chang, Q, Chen, Q & Wang, X 2005, Scaling Gaussian RBF kernel width to improve SVM classification. in Proceedings of 2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05., 1614559, Proceedings of 2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05, vol. 1, pp. 19-22, 2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05, Beijing, China, 13/10/05.

Scaling Gaussian RBF kernel width to improve SVM classification. / Chang, Qun; Chen, Qingcai; Wang, Xiaolong.
Proceedings of 2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05. 2005. p. 19-22 1614559 (Proceedings of 2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05; Vol. 1).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Scaling Gaussian RBF kernel width to improve SVM classification

AU - Chang, Qun

AU - Chen, Qingcai

AU - Wang, Xiaolong

PY - 2005

Y1 - 2005

N2 - Support vector classification with Gaussian RBF kernel is sensitive to the kernel width, Small kernel width may cause over-fitting, and large one under-fitting. The so-called optimal kernel width is merely selected based on the tradeoff between under-fitting loss and over-fitting loss. So, there exists urgent need to further reduce the tradeoff loss. To circumvent this, we scale the kernel width in a distribution-dependent way. Experiments validate the feasibility of this method. Existing problems are also discussed.

AB - Support vector classification with Gaussian RBF kernel is sensitive to the kernel width, Small kernel width may cause over-fitting, and large one under-fitting. The so-called optimal kernel width is merely selected based on the tradeoff between under-fitting loss and over-fitting loss. So, there exists urgent need to further reduce the tradeoff loss. To circumvent this, we scale the kernel width in a distribution-dependent way. Experiments validate the feasibility of this method. Existing problems are also discussed.

UR - https://www.scopus.com/pages/publications/33847157959

M3 - 会议稿件

AN - SCOPUS:33847157959

SN - 0780394224

SN - 9780780394223

T3 - Proceedings of 2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05

SP - 19

EP - 22

BT - Proceedings of 2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05

T2 - 2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05

Y2 - 13 October 2005 through 15 October 2005

ER -

Scaling Gaussian RBF kernel width to improve SVM classification

Abstract

Publication series

Conference

Other files and links

Fingerprint

Cite this