Multi-view Clustering Based On Deep Learning

In the information age,the amount of data is exploding exponentially.It is very necessary to reasonably aggregate similar objects,namely clustering,to reduce the confusion of data and help researchers to distinguish their internal logic more easily.Clustering is one of the most basic unsupervised learning tasks in the fields of machine learning,pattern recognition,data mining,etc.Its purpose is to group data into a specific category without supervised information.Single view clustering is not competent to deal with complex scenes due to its single acquisition perspective and incomplete data information.Multi-view clustering,which can integrate heterogeneous and complementary information of multi-view data,has attracted wide attention in recent years.However,the existing multi-view clustering based on traditional machine learning cannot obtain the nonlinear structure of multi-view data,resulting in poor performance.Although the multi-view clustering based on deep learning can obtain the nonlinear structure of multi-view data,it fails to comprehensively consider the structural relations between views and the discriminant structure within clusters.In order to solve these problems,this paper,by means of artificial intelligence,studies and mines the inter-view and intra-view structural relations of multi-view data,and constructs a more effective multi-view clustering model.The specific content is as follows:(1)To solve the problem that the existing algorithms do not consider local structure and discriminant structure,this paper proposes a novel method,named Deep Multi-view Subspace Clustering with Unified and Discriminative Learning.The model combines the global and local structures with the self-expression layer,and the global and local structures interact with each other to make full use of the potential feature information of the original multi-view data,so that the distance between samples in the same cluster is closer,while the distance between samples in different clusters is longer.Meanwhile,considering the expression of the consistency among views,the clustering discriminant constraint was used to make the distance between different clusters become longer,and the clustering performance was improved.Therefore,the model can adaptively learn the weight relationship between samples,the weight of similar samples is large,the weight of different samples is small,so as to better learn the multi-view clustering shared connection matrix.The experimental results show that the proposed method is superior to several existing multi-view clustering methods in performance.(2)In view of the existing algorithm neglecting the problem of multi-level feature extraction in network structure,a deeply Structured Multi-view Subspace Clustering network is constructed.The proposed method uses multi-path convolutional neural network to explicitly learn each view in a hierarchical way and extract the subspace representation of features at different levels.In this way,high-level and low-level structural features are integrated through a common connection matrix to explore the structure of a shared subspace between multiple views.In addition,the connection matrix is decomposed into the consistent connection matrix and the view difference connection matrix,and the difference information between different views is stripped from the common connection matrix to seek a more consistent representation.In addition,the model imposes a low-rank constraint on the consistent connection matrix to reduce the influence of noise and further highlight all view consensus information.Experimental results on four common datasets show the effectiveness of the proposed method.