Research On Modeling And Optimization Of Self-expressive Subspace Clustering

People’s life produces a huge amount of data all the time,the data is text,or images,or video.And with the large-scale application of computer technology in different industries,the scale of data is getting larger and larger,and the dimension of data is getting higher and higher.With the advent of the era of big data,this trend is increasingly significant.Although modern machine learning techniques have achieved great success in analyzing big data,this approach requires a lot of annotated data,and the acquisition of annotated data often requires a lot of manpower and resources.Extracting patterns and features from unlabeled massive data has become an important research problem.Subspace clustering has attracted much attention because of its remarkable advantages in processing high-dimensional data.In particular,self-expressive subspace clustering proposed in recent years has become a research hotspot in the field of subspace clustering.In view of the problems that data collected in practical applications often contain different forms of noise,and it is difficult to clearly capture global and local information of samples,this paper studies how to improve clustering accuracy and robustness based on the self-expression model.The specific work of this paper is as follows:(1)One-step subspace clustering method based on adaptive graph regularization and subspace structure representation is proposed.In this method,subspace structure norm is used to improve the unity between the two steps of data representation and spectral clustering.The Frobenius norm is used to encourage grouping effects on the global structure,while the adaptive graph regularization is used to consider the local structure,that is,the coefficient matrix is learned by assigning an adaptive optimal neighborhood to each sample point according to local connectivity.Finally,experiments are carried out on three types of data sets,which proves that the proposed algorithm has certain advantages.(2)One-step subspace clustering method based on adaptive graph regularization and correntropy induced metric is proposed.Specifically,the method integrates the computational coefficient matrix,adaptive graph regularization and applied spectral clustering into a unified optimization framework.Nuclear norm is used to maintain the global structure of the data,while adaptive graph regularization is used to maintain the local connectivity of the data.At the same time,considering the complex noise that may exist in the actual data,the correlation measure which is insensitive to outliers is used to calculate the reconstruction error,which improves the robustness of the algorithm.An effective interactive algorithm is proposed to optimize the model.The method of artificially adding noise is used in the experiment design,and experiments are carried out on ten data sets of four types to verify the effectiveness of the algorithm.