基于支持向量机与极端保守在线算法相结合的多类分类器

A MULTI-CLASS CLASSIFIER BASED ON SUPPORT VECTOR MACHINES AND ULTRACONSERVATIVE ONLINE ALGORITHM

本文给出了一种将SVM和极端保守在线算法相结合的通用多类分类算法,算法利用最大置信度原则将离线训练的多个SVM组合成一个多类分类器.为了提高在线学习过程的实时性,同时保证分类器的推广能力,我们将K.Crammer等人提出的极端保守在线算法思想引入到分类器修正过程当中,修正过程中采用对应SVM的支持向量和错分样本作为训练集.实验表明,算法具有良好的实时性能,且具有良好的推广能力。

A new general multiclass classifying algorithm that combines Support Vector Machines (SVM) and ultra-conservative online algorithm is presented in this paper. Using maximum confidence principle, the algorithm combines several SVMs to make a multiclass classifier. In order to improve the algorithm's real time ability in learning process as well as the generalization ability of the classifier, the idea of ultraconservative online algorithm presented by K. Crammer, et al. is introduced into the modifying process of the classifier. During the course of modifying the classifier, only the Support Vectors (SVs) correspond to SVMs and the samples'that is classified incorrectly are used as the training sample set. The experiment result shows that the algorithm has good performance in real time ability as well as in the generalization ability.