【摘要】 在SVM训练过程中,二次规划问题的求解制约着SVM应用于大规模数据.SVM的决策函数由邻近分类超平面的部分训练样本——支持向量决定.基于减小训练样本数目、加快SVM训练过程的目的,提出一种基于二分网格的边界样本提取方法.数据仿真实验表明,该方法具有边界样本提取准确、效率高、速度快、能够自适应样本分布的优点,而且不会显著降低SVM分类器的性能.更多还原

【Abstract】 In training a SVM, the solution of a quadratic programming (QP) problem constrains the SVM in large scale problem. The decision fuction of SVM is determined by a subset of the training patterns, support vectors, which lies closer to the classification hyperplane. Motivated by reducing the training patterns, a grid-based algorithm is proposed to pre-select a subset of patterns near the decision boundary. Simulation experiments show that BS-grid efficiently extracts accurate boundary patterns without degrading the classification ability of SVM and has good adaptive capacity to pattern’s density and outline.更多还原