L2-Loss nonparallel bounded support vector machine for robust classification and its DCD-type solver

Liming Liu

doi:10.1016/j.asoc.2022.109125

L2-Loss nonparallel bounded support vector machine for robust classification and its DCD-type solver

2022-09-01

2

PRE

AI

DOI：10.1016/j.asoc.2022.109125

Original

Original request for help

Save

Abstract

En 中文

In this paper, we study a new classification methodology to enhance the robustness and convergence of the nonparallel support vector machine (NPSVM), namely L2-Loss nonparallel bounded SVM (L2NPBSVM). We first define a L2-Loss and an adjustable hinge loss for NPSVM. Then, using the two loss functions, we propose the L2-NPBSVM algorithm. Both the L2-loss and the adjustable hinge loss are insensitive to the feature noise around the decision boundary. In addition, the L2-Loss can make full use of the margin distribution information in the training samples. The margin distribution is more important than margin maximization for generalization performance. Furthermore, an additional regularization term is added into the objective functions of L2-NPBSVM. It can ensure the global solution and stability of optimization problems. Thus, the classification performance of L2-NPBSVM can be further improved. In addition, in order to shorten the training time, the dual coordinate descent (DCD) algorithm is described and analyzed to optimize L2-NPBSVM. Numerical experiments on different datasets demonstrate that our L2-NPBSVM has the advantages of strong robustness, strong generalization ability and fast convergence. (c) 2022 Published by Elsevier B.V.

Keywords:

Pattern recognition

Nonparallel support vector machine

Noise insensitive

Optimal margin information

Dual coordinate descent

L2-Loss nonparallel bounded support vector machine for robust classification and its DCD-type solver

Abstract

Journal

Organization