Presentation 2004/1/19
Maximal Margin Classifier based on Geometric Methods (Neurocomputing)
Manabu MUKAIYAMA, Haruhisa TAKAHASHI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) The support vector machine provides good learning performance by maximizing the margin and use of the Kernel Method. To solve a quadratic programing for maximizing the margin requires the computational efforts of polynomial order O(l^3) for sample size l. Although several implementaion methods of SVM such as SMO are proposed, the computational efforts are not yet realistic for large sample size. In this report, we propose a geometric method which constracts a maximal margin hyperplane finding support vectors iteratively starting from the vector pair minimizing the distance between classes. The computational complexity of this method is O(l^2) or O(m^3), m being the number of support vectors. Computer experiments shows that our method is advantageous when the number of support vectors is relatively small.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Pattern Recognition / Maximal Margin / Low Computational Effort / SVM / SMO
Paper # NC2003-114
Date of Issue

Conference Information
Committee NC
Conference Date 2004/1/19(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Maximal Margin Classifier based on Geometric Methods (Neurocomputing)
Sub Title (in English)
Keyword(1) Pattern Recognition
Keyword(2) Maximal Margin
Keyword(3) Low Computational Effort
Keyword(4) SVM
Keyword(5) SMO
1st Author's Name Manabu MUKAIYAMA
1st Author's Affiliation Graduate School of Electro-Communications, The University of Electro-Communications()
2nd Author's Name Haruhisa TAKAHASHI
2nd Author's Affiliation The University of Electro-Communications
Date 2004/1/19
Paper # NC2003-114
Volume (vol) vol.103
Number (no) 601
Page pp.pp.-
#Pages 6
Date of Issue