Soft-magin support vector machine maximizing geometric margins for multiclass classification

2015 Fiscal Year Final Research Report

• Project/Area Number: 24500275
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Sensitivity informatics/Soft computing
• Research Institution: Osaka University
• Principal Investigator: Tatsimi Keiji 大阪大学, 工学(系)研究科(研究院), 准教授 (30304017)
• Project Period (FY): 2012-04-01 – 2016-03-31
• Keywords: 多クラス識別問題 / 教師有り学習 / サポートベクトルマシン / 多目的最適化 / マージン最大化 / クラスタリング / 2次錐計画問題

Outline of Final Research Achievements

We had proposed the multiobjective multiclass support vector machine (MMSVM), which can maximize the geometric margins for multiclass classification problem. In this study, we developed reduction methods of time and space computational complexity of the MMSVM, and extended the MMSVM into soft-margin models which can learn training data including outliers, where we compare the performances of various combinations from some restriction methods of the constraints of the MMSVM, several solving method of the multiobjective optimization, and two kinds of penalty functions. Among them, we verified that some propose methods, a data reduction method based on support vectors for the soft-margin MMSVM, an improved MMSVM based on the one-against-all method, an MMSVM based on the k-means method are more effective than existing methods.

Free Research Field

機械学習，最適化