局部切空间对齐算法的核主成分分析解释.pdfVIP

CN431258/ T P 2010 32 6 ISSN 1007130X COMPUTER ENGINEERING SCIENCE Vol 32, No 6, 2010 : 1007130X(2010) A Kernel PCA View of the Local T angent Space Alignment Algorithm , , ZHAN Yubin, YIN Jianping, LIU Xinwang ( , 4 10073) ( School o Computer Science, National University o De ense Technology, Changsha 410073, China) : 基于核方法的降维技术和流形学习是两类有效而广泛应用的非线性降维技术, 它们有着 自不同的出发点和 理论基础, 在以往的研究中很少有研究关注两者的联系LT SA 算法利用数据的局部结构构造一种特殊的核矩阵, 然后利 用该核矩阵进行核主成分分析本文针对局部切空间对齐这种流形学习算法, 重点研究了LT SA 算法与核PCA 的内在联 系研究表明, LTSA 在本质上是一种基于核方法的主成分分析技术 Abstract: Recent ly, nonlinear dimensionality reduction has attracted ext ensive interests of researchers in the machine learning community. The techniques for nonlinear dimensionality reduction can be divided into t wo categories: kernelbased met hods and manifold learning . These tw o methods have different mot ivations and derivat ions. This paper interprets the wellknow n manifold learning algorithm LTSA as a kernel method. We show t hat LTSA can be described as the kernel PCA, and LTSA utilizes the local neighborhood inf ormat ion to construct a special kernel matrix , and the global embedding obtained by LTSA w ith modified constraints is equivalent to principal coordinates by the kernel PCA w ith this special ker nel. : 降维; 流形学习; 核方法; 核主成分分析;局部切空间对齐 Key words: dimensionality reduct ion; manifold learning; kernel method; kernel principal component analysis; local tangent space alignment doi : 10. 3969/ j. issn. 1007130X. 2010.06.044 :T P18 :A ( PCA) [4] 1 ( MDS) [ 5] ( LDA) [6] ,