Font Size: a A A

Research On Locally Linear Embedding Dimensionality Reduction Algorithms Based On Density

Posted on:2018-08-30Degree:MasterType:Thesis
Country:ChinaCandidate:K HanFull Text:PDF
GTID:2348330542990974Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
With the rapid development of science and technology,the arrival of the era of big data and cloud computing,the presentation of the data sets is tending towards massive data,high dimension data,nonlinear data.So data dimensionality reduction,the ancient but new data processing plays an irreplaceable role.Locally Linear Embedding(LLE)selects K-Nearest Neighbor of each sample,it need to make prior estimates for the data sets.Selecting the value of K largely needs to be confirmed artificially.Therefore,selecting the value of K is a little sensitive.The noises of the data sets have deeply effect on performance of dimensional reduction,and The image data set of dimension reduction effect of instability.This paper points to the shortcomings and deficiencies of Locally Linear Embedding(LLE)and study Local embedding dimension reduction algorithm based on density,a reasonable improvement algorithm is proposed.The specific contents are as follows:In this paper,we propose that the Euclidean distance is replaced by the Euclidean distance metric,which is applied to the field of image processing,and the improved LLE algorithm is stable.Combined with the noise reduction processing of data sets,pretreating the data sets with LDA make homogeneous data more cohesive and make inhomogeneous data more discrete.Meanwhile processing the data sets segmented to avoid the effect of noise on dimensional reduction method;aiming at the method of selecting parameter artificially about LLE,locally linear embedding dimensional Reduction Algorithms Based on Density(DALLE)based on density has been proposed.It's an adaptive parameter adjustment method which selects neighborhood with the density of points around a sample.The simulation results show that the complexity of DALLE algorithm is similar to the original LLE algorithm,but it is not sensitive to noise points and does not need to choose the parameter values artificially.Through the analysis and experiment,compared with the existing dimension reduction algorithm,DALLE algorithm has achieved great results in nonlinear data dimensional reduction.
Keywords/Search Tags:manifold learning, data dimensional reduction, nonlinear data, local embedding algorithm, adaptive parameter selection
PDF Full Text Request
Related items