系统工程与电子技术

• 软件、算法与仿真 • 上一篇    下一篇

一种基于QBC的SVM主动学习算法

徐海龙, 别晓峰, 冯卉, 吴天爱   

  1. 空军工程大学防空反导学院, 陕西 西安 710051
  • 出版日期:2015-11-25 发布日期:2010-01-03

Active learning algorithm for SVM based on QBC

XU Hai-long, BIE Xiao-feng, FENG Hui, WU Tian-ai   

  1. College of Air and Missile Defense, Air Force Engineering University, Xi’an 710051,China
  • Online:2015-11-25 Published:2010-01-03

摘要:

针对支持向量机(souport vector machine,SVM)训练学习过程中样本分布不均衡、难以获得大量带有类标注样本的问题,提出一种基于委员会投票选择(query by committee,QBC)的SVM主动学习算法QBC-ASVM,将改进的QBC主动学习方法与加权SVM方法有机地结合应用于SVM训练学习中,通过改进的QBC主动学习,主动选择那些对当前SVM分类器最有价值的样本进行标注,在SVM主动学习中应用改进的加权SVM,减少了样本分布不均衡对SVM主动学习性能的影响,实验结果表明在保证不影响分类精度的情况下,所提出的算法需要标记的样本数量大大少于随机采样法需要标记的样本数量,降低了学习的样本标记代价,提高了SVM泛化性能而且训练速度同样有所提高。

Abstract:

To the problem that large-scale labeled samples is not easy to acquire and the class-unbalanced dataset in the course of souport vector machine (SVM) training, an active learning algorithm based on query by committee (QBC) for SVM(QBC-ASVM) is proposed,which efficiently combines the improved QBC active learning and the weighted SVM.In this method,QBC active learning is used to select the samples which are the most valuable to the current SVM classifier,and the weighted SVM is used to reduce the impact of the unba-lanced-data set on SVMs active learning. The experimental results show that the proposed approach can considerably reduce the labeled samples and costs compared with the passive SVM, and at the same time, it can ensure that the accurate classification performance is kept as the passive SVM, and the proposed method improves generalization performance and also expedites the SVM training.