一种改进的分枝定界半监督支持向量机学习算法  

An Improved Learning Algorithm for Branch and Bound for Semi-Supervised Support Vector Machines

作　　者：赵莹 张健沛[1] 杨静[1] 王冠军[2] 

机构地区：[1]哈尔滨工程大学计算机科学与技术学院,黑龙江哈尔滨150001 [2]中国矿业大学计算机学院,江苏徐州221116

出　　处：《电子学报》2010年第2期449-454,共6页Acta Electronica Sinica

基　　金：国家自然科学基金(No.60873037;No.60673131)

摘　　要：分枝定界半监督支持向量机,由于其实现的是全局最优化,因而可以作为其他半监督学习算法的一个基准.针对分枝定界半监督支持向量机中存在的缺陷,提出一种改进的分枝定界半监督支持向量机学习算法.该算法重新对下界的估计进行定义,从而降低了各结点计算下界的时间复杂度;同时利用支持向量机的几何特点确定分枝结点,以提高算法的运算速度.实验分析表明本文提出的算法具有精度高、鲁棒性强等优点.Branch and bound semi-supervised support vector machines as an exact globally optimization is useful for benchmarking practical semi-supervised support vector machines implementations. An improved learning algorithm for branch and bound for semi-supervised support vector machines is presented,concerning the defects of the branch and bound for semi-supervised support vector machines. The estimations of the node lower bound are redefined, which can reduce time complexity of computing the lower bound on every node. Branching nodes are determined by using the geometric characteristic of the support vector machines, which can improve the operation speed simultaneously. Experimental results show that modified algorithm has high precision and strong robustness.

关 键 词：半监督学习 支持向量机 分枝定界 统计学习理论 

分 类 号：TP302.7[自动化与计算机技术—计算机系统结构]