基于商空间粒度理论的大规模SVM分类算法

Large-scale SVM classification algorithm based on granularity of quotient space theory

利用商空间粒度理论对已有的SVM分类算法进行改进,给出了一种新的SVM分类算法——SVM-G。该算法将SVM分类问题划分成两个或多个子问题,从而降低了SVM分类复杂度。实验表明,改进的算法适用于处理大数据量的样本,能在保持分类精度的情况下有效地提高支持向量机的学习和分类速度。

This paper improved the existing SVM algorithm with the granularity of the quotient space theory, proposed a new SVM algorithm（SVM-G）. The improved algorithm divided SVM classification problem into two or more sub-issues, thereby re- ducing the computation complexity of SVM classification. Experimental results indicate that the improved algorithm is suitable for processing large number of observations and can effectively accelerate the speeds of SVM learning and classifying while keeping the classification precision.