摘要
提出一种基于再生核随机投影的集成分类方法,该方法将原始数据投影到特征空间上,利用表示定理和核技巧转化为使用随机投影对Gram矩阵降维,然后利用集成学习方法构造集成分类器.利用随机的线性映射进行降维,再利用核技巧发展了核随机投影方法.还证明了核随机投影的集成学习方法泛化误差的极限性质,得到了在一定条件下的关于泛化误差的收敛速度性质.模拟研究和实证分析的结果表明该方法相较于一些常用方法具有更好的表现.
An ensemble classification method based on reproducing kernel random projection was presented.This method projects the original data onto the feature space,uses the representation theorem and kernel trick to convert the use of random projection to the reduction of the dimensionality of the Gram matrix,and then uses an ensemble learning method to construct an ensemble classifier.Random linear mapping was used for dimensionality reduction and a kernel random projection method was developed by using the kernel trick.The limit property of generalization error was proved and the convergence rate of generalization error was obtained in some certain conditions for the kernel random projection ensemble learning method.Finally,simulation research and empirical analysis are carried out.The results show that this method has better performance than some common methods.
作者
崔文泉
张枫
徐建军
CUI Wenquan;ZHANG Feng;XU Jianjun(Department of Statistics and Finance,School of Management,University of Science and of Technology of China,Hefei 230026,China)
基金
国家自然科学基金(71873128)
安徽省自然科学基金(1308085MA02)资助
关键词
随机投影
核技巧
集成学习
random projection
kernel trick
ensemble learning
