一种基于p-Laplacian的谱聚类维数约简算法  

Dimensionality reduction algorithm based on p-Laplacian in spectral clustering

作　　者：石远超[1] 马宏[1] 李海涛[1] 

机构地区：[1]国家数字交换系统工程技术研究中心,郑州450002

出　　处：《计算机应用研究》2012年第6期2126-2128,2131,共4页Application Research of Computers

基　　金：国家"863"计划资助项目(2011AA010603)

摘　　要：近年来,谱聚类因其深厚的理论基础而在机器学习和数据挖掘领域中引起了广泛的关注。针对谱聚类算法中采用Laplacian矩阵时无法获得较好的图切判据,通过引入p-Laplacian算子,提出了一种基于p-Laplacian的谱聚类维数约简算法。仿真实验表明,提出的方法可以获得一种优化图切的近似解,使得在降维后能够更为精确地求得原始数据在低维空间中的嵌入投影。In recent years,spectral analysis approaches have received much attention in machine learning and data mining areas,due to their rich theoretical foundations.This paper addressed the problem of Laplacian in spectral clustering,which couldn＇t get ideally graph cut criterion,by introducing the p-Laplacian operator,this paper proposed a new dimensionality reduction algorithm based on p-Laplacian.The experimental results denote that,the approach can get a approximation of the optimal graph cut,and can accurately get embedding mapped of original data in low dimensionality space.

关 键 词：谱聚类 图切判据 维数约简 P-LAPLACIAN 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]