多核学习中基于复合梯度映射的学习算法研究  

Research on learning algorithm based on composite gradient mapping in multiple kernel learning

作　　者：龙文光[1] 刘益和[2] 

机构地区：[1]内江师范学院现代教育技术中心,四川内江641112 [2]内江师范学院计算机科学学院,四川内江641112

出　　处：《计算机应用研究》2015年第4期1019-1023,共5页Application Research of Computers

摘　　要：现有的多核学习算法大多假设训练样本分类完全正确,将其应用到受扰分类样本上时,由于分类存在差错,因此往往只能实现次优性能。为了解决这一问题,首先将受扰分类多核学习问题建模为随机规划问题,并得到一种极小极大表达式;然后提出基于复合梯度映射的一阶学习算法对问题进行求解。理论分析表明,该算法的收敛速度为O(1/T),大大快于传统算法的收敛速度O(1槡/T)。最后,基于五个UCI数据集的实验结果也验证了本文观点和优化算法的有效性。The existing multiple kernel learning algorithms assume that the training sample classification entirely correct,when they used to the classification of disturbed samples,the suboptimal performance could only be achieved due to the incorrect class assignments. In order to solve this problem,firstly,it modeled the multiple kernel learning problems from noisy labels into a stochastic programming problem,and presenting a min-max formulation,and then proposed a first order learning algorithm based on composite gradient mapping to solve this problem. The theoretical analysis shows that,the convergence rate of O（ 1 / T） for the proposed algorithm,significantly faster than the classical O（ 1槡/ T） rate. Finally,the experimental results on five UCI data sets confirm the effectiveness and the efficiency of the proposed framework and the optimization algorithm.

关 键 词：多核学习 训练样本 随机规划 复合梯度映射 收敛速度 UCI数据集 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]