基于SVM的二次下降有效集算法  

Secondary Descent Active Set Algorithm Based on SVM

作　　者：丁晓剑[1] 赵银亮[1] 李远成[1] 

机构地区：[1]西安交通大学电子与信息工程学院,陕西西安710049

出　　处：《电子学报》2011年第8期1766-1770,共5页Acta Electronica Sinica

基　　金：国家863高技术研究发展计划(No.2008AA01Z136)

摘　　要：针对现有的有效集方法应用到支持向量机(support vector machine,SVM)优化问题时收敛速度较慢的问题,提出了一种基于二次下降法和推测赋值法的有效集算法.该算法在每次迭代过程中利用映射因子将迭代向量值限制在优化问题的不等式约束中,并通过调整步长使目标优化问题的函数值较传统的有效集算法进一步下降.由于函数值在每次迭代后保证了严格快速下降,所以提出的算法能够快速收敛到全局最优解.实验结果表明该方法的迭代次数和迭代时间有明显减少.To solve the slow convergence rate of the existing active set methods applied into optimization formulation of support vector machine,an active set algorithm based on the secondary descent method and the speculative assignment method is proposed.At each iteration of the algorithm,a projection operator is used to restrict the iterative vector onto the inequality constraints of optimization formulation,and then an adjustable step size is used to ensure the functional value of optimization formulation make further descent compared to the traditional active set method.As functional value ensure rapid and strictly descent at the end of each iteration,the global optimum solution can be obtained with rapid convergence rate.Experimental results show that iterations time and training time of the proposed method have been decreased obviously.

关 键 词：支持向量机 有效集 二次下降法 迭代 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]