Improvement and Implementation of Manifold Learning Algorithms ISOMAP
|School||Dalian University of Technology|
|Course||Applied Computer Technology|
|Keywords||Dimension of simplicity Manifold Learning Isometric Mapping Sector punctuation|
Dimension Simplicity is an effective means to deal with these high-dimensional data , and can effectively avoid the \Its purpose : the premise does not change the essence of the original high -dimensional data structure to minimize or remove the redundant information to reduce the dimensionality of the original data , so as to achieve the purpose of simplicity dimension . Through the existing the linear simple method of analysis, able to draw the real geometry of linear high-dimensional data . Most of the data in the real world is nonlinear , we need to be able to effectively deal with the nonlinear high-dimensional data dimension simple method , traditional linear simple method , however , the linear nature of the decision , such methods can only be used to find global linear structure in high-dimensional data , can not find the nonlinear structure in high-dimensional data . In this context , the manifold learning methods came into being used to solve the existing problems in the analysis of nonlinear high -dimensional data can effectively find the internal geometry of nonlinear high-dimensional data . Isometric feature mapping algorithm is a typical global optimization algorithm of manifold learning the embedded results can reflect the distance between the manifold in high - dimensional data samples , able to get the ideal embedding results . An important issue in the algorithm is the computation time required . To address this issue , this paper presents a fuzzy C - means clustering method to select representative sector punctuation to improve isometric feature mapping algorithm . First using fuzzy clustering algorithm simple sample set of high-dimensional data obtained cluster centers of the various types of sample points as a landmark isometric mapping algorithm to construct distance matrix , last LMDS method for solving the final result of the embedding . , ISOMAP algorithm whether accurate results of the low- dimensional embedding high -dimensional data sets , mainly depends on the selection of the number of points in the neighborhood , and how to select the appropriate points in the neighborhood , is still an open question . Combination of fuzzy clustering theory and graph theory , the proposed tentative neighborhood value estimation algorithm TNVE, to determine the the ISOMAP algorithm parameters - neighborhood values ??.