基于双重近邻图的有监督流形学习算法  

A Supervised Manifold Learning Algorithm Based on 2-fold Neighbor Graph

作　　者：符茂胜[1] 傅思勇[1] 金星[1] 吴其平[1] 

机构地区：[1]皖西学院信息工程学院,安徽六安237012

出　　处：《皖西学院学报》2015年第2期47-50,共4页Journal of West Anhui University

基　　金：安徽省科技厅自然科学基金面上项目(1308085MF97);国家级创新创业训练计划项目(201210376022)

摘　　要：提出了一种有监督的流形学习算法,算法首先构建双重近邻图,即类内近邻图和类间近邻图,从而获得相应的类内邻接矩阵和类间邻接矩阵,并在LPP框架下构建最优的低维嵌入。人工合成数据和实际数据上的实验都表明了所提算法优于一些线性和非线性的嵌入算法。A supervised manifold learning algorithm is proposed in the paper. Firstly, 2 fold neighbor graph is conducted, namely the within-class graph and between class graph, and then the within-class adjacency matrix and between-class adjacency matrix are obtained. The optimal low dimensional embedding based on the framework of LPP is derived. Experiments were conducted on synthetic and real data and the results demonstrate that the proposed algorithm significantly outperforms many linear and non- linear embedding techniques.

关 键 词：非线性维数约简 保局投影 流形学习 双重临近图 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]