Research On Multi-view Dimension Reduction Algorithm Based On Sparsity Preserving In Multiple Scenes

In recent years,with the continuous development of computer technology,there have been more and more multi-view high-dimensional data described from different angles.With the emergence of high-dimensional multi-view data,multi-view dimensionality reduction,clustering and classification and other multi-view learning has gradually become a research hotspot.In view of the multi-view data has the characteristics of view correlation and high dimensionality,more and more multi-view dimensionality reduction algorithms have been proposed.Multi-view data often has a high dimensions and semi-paired situation.Based on the research of previous study,this paper proposes a series of multi-view dimensionality reduction algorithms based on sparsity preserving in different scenes.(1)For multi-view high-dimensional data,a multi-view sparsity preserving projection based on collaborative training is proposed.This algorithm uses collaborative training to introduce sparsity reconstruction into multi-view data dimensionality reduction.The algorithm first constructs a sparsity reconstruction coefficient matrix on each view through sparse representation,and then uses collaborative training to achieve maximum consistency of the sparse coefficient matrix of the dimensionality reduction subspaces of different views during the iterative dimensionality reduction process.Finally,the superiority of the algorithm is verified through experiments.(2)The traditional multi-view dimensionality reduction algorithm requires that the data samples must be completely paired.Aiming at the problem that traditional algorithms cannot solve the problem of multi-view semi-paired data dimensionality reduction,a multi-view dimensionality reduction algorithm based on sparsity preserving in the semi-paired scenes is proposed.The algorithm recalculates the reconstruction coefficient fusion matrix in the public view by using the sparsity reconstruction relationship between the public samples and the paired sample.The algorithm can make better use of the data information of unpaired samples and has a better interpretability for the sparse coefficient matrix.It is verified by experiments that the algorithm can still preserve the sparsity relationship of data well in the semi-unpaired scenes.(3)In view of the problem of multi-view data dimensionality reduction in different pairing scenes,a multi-view sparsity preserving dimensionality reduction algorithm suitable for multiple pairing scenes is proposed.This algorithm utilizes the data information of the paired samples by constructing the structural consistency constraints between the views of the paired samples.At the same time,the algorithm further utilizes the data information of paired and unpaired samples by constructing a sparsity coefficient matrix of existing samples on each view.On this basis,the algorithm seeks to embed the low-dimensional subspace jointly,so that the sparsity relationship on the original high-dimensional space can be maintained in the low-dimensional subspace.In the experiments of different paired scenes,the experimental results show that the algorithm can not only make full use of the data information of the paired and unpaired samples,but also integrate the view correlation of the paired samples into the model.