计算机应用与软件Issue(9)：181-183,212,4.DOI:10.3969/j.issn.1000-386x.2014.09.044
基于最小剩余方差的LLE改进算法
AN IMPROVED LLE ALGORITHM BASED ON MINIMUM RESIDUAL VARIANCE
摘要
Abstract
When the manifold is curled and the distance between two surfaces is relatively small,the locally linear embedding algorithm(LLE)may cause distortions in manifold structural in the process of reconstruction;moreover,there are not the consistent standards inselecting the value of nearest number K and the dimension number D of dimensionality reduction,which lead to the reduction indimensionality reduction effect.In order to deal with these problems,we present an LLE algorithm that is based on improved distance and onintelligently selecting the parameter values according to residual variance.The algorithm introduces new distance metric formula to replace theEuclidean distance in original algorithm,and evaluates the effect of embedding the high-dimensional data structure into a low-dimensionalspace by introducing residual variance according to the values of K,D.The method is verified on UCI dataset and Yale face database.Experimental results of MATLAB programming illustrate that this method achieves better performance than the traditional methods in selectingthe parameters values and recognition rate.关键词
局部线性嵌入算法(LLE)/近邻个数K/降维维数D/距离度量/剩余方差Key words
Locally linearembedding(LLE)/Nearestnumber K/Dimensionality reductionnumber D/Distancemetric/Residualvari-ance分类
信息技术与安全科学引用本文复制引用
吴学斌,肖迪..基于最小剩余方差的LLE改进算法[J].计算机应用与软件,2014,(9):181-183,212,4.基金项目
江苏省高校自然科学研究项目 （）
