自适应迭代算法支持向量集的特性研究

针对在支持向量机研究中,传统的优化方法无法处理规模不断扩大的分类问题,为设计适应大样本分类的训练算法,提出了基于块的自适应迭代算法。在该算法的训练过程中,块增量学习和逆学习交替进行,能够自动得到一个小的支持向量集。将该算法与SVML ight在支持向量数量方面进行了比较,计算了UC I(Un i-versity of Californ ia-Irvine)中的6个数据集和著名的Checkboard问题。结果表明:该自适应迭代算法确定的支持向量数一般不到SVML ight所得到的支持向量数的一半,其中70%多的支持向量被SVML ight所确定的支持向量集所包含,在支持向量选择方面具有高效性。

Study on Features of Support Vector Set of Adaptive and Iterative Algorithm

In the research of support vector machines,with increasing the scale,some classification problems cannot be solved by the classical optimization methods.In order to design the training algorithms for large classification problems,an adaptive and iterative support vector machine algorithm based on chunk(CAISVM) is proposed.During the training process,the chunk incremental and decremental procedures are performed alternatively,and a small support vector set can be obtained adaptively.Six UCI(University of California-Irvine) data sets and Checkboard benchmark problem are tested,and comparisons between CAISVM and SVM~(Light) are given focusing on the number of support vectors.The simulating results show that,in general,the number of support vectors obtained by CAISVM is smaller than half of that obtained by SVM~(Light),in which more than 70 percent of the support vectors are included in the support vector set obtained by SVM~(Light),which shows that CAISVM is efficient in selecting the support vectors.