タツミ　ケイジ   Keiji Tatsumi   巽　啓司    所属   追手門学院大学  理工学部 数理・データサイエンス学科    職種   教授
言語種別	英語
発行・発表の年月	2019/03
形態種別	外国学会誌（First author）
査読	査読あり
標題	Approximate Multiobjective Multiclass Support Vector Machine Restricting Classifier Candidates Based on k-means Clustering
執筆形態	共著・編著（代表編著を除く）
掲載誌名	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
巻・号・頁	11471 LNAI,pp.271-283
著者・共著者	Keiji Tatsumi,Takahumi Sugimoto,Yoshifumi Kusunoki
概要	© Springer Nature Switzerland AG 2019. In this paper, we propose a reduction method for the multiobjective multiclass support vector machine (MMSVM), one of all-together method of the SVM. The method can maintain the discrimination ability, and reduce the computational complexity of the original MMSVM. First, we derive an approximate convex multiobjective optimization problem for the MMSVM by linearizing some constraints, and we secondly restrict the normal vectors of classifier candidates by using centroids obtained from the k-means clustering for each class dataset. The derived problem can be solved by the reference point method based on the centers of gravity of class datasets, in which the geometric margins between all pairs are exactly maximized. Some numerical experiments for benchmark problems show that the proposed method can reduce the computational complexity without decreasing its generalization ability widely.
DOI	10.1007/978-3-030-14815-7_23
ISSN	0302-9743
PermalinkURL	https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85064213572&origin=inward