Research And Application Of Multi-View Clustering Based On Multi-Channel Tissue-Like P System

Clustering is a tool of machine learning and artificial intelligence,which divides a set of data points into corresponding clusters,so that the similarity of data points in the clusters is high,while the similarity of data points between clusters is low.This is an unsupervised learning technique.With the development of science and technology,more and more data are represented by multiple views,that is,the so-called multi-view data.The clustering operation of this kind of data is multiview clustering.The use of multi-view clustering can make full use of the consistency information and complementary information among multiple views.Compared with single view clustering,multi-view clustering is widely concerned because of its better clustering performance.The weight allocation method of each view,the construction way of similarity matrix and the way of outputting clustering results will have more or less influence on the performance of multi-view clustering.How to optimize these problems has become a research hotspot in this field.As a branch of natural computing,P system is a distributed parallel computing model abstracted from the structure and function of biological cells and the collaboration of cell populations such as organs and tissues.P system has computing power equivalent to Turing machine.Due to the uncertainty and computation parallelism of P system,we embed the multiview clustering algorithm into the frame of P system to run and improve the computational efficiency of the multi-view clustering algorithm.In this paper,we propose a variant of the tissuelike P system,and combine it with the improved multi-view subspace clustering algorithm and multi-view spectral clustering algorithm,as follows:1.A multi-channel tissue-like P system with weight and rule triggering mechanism(WRMCTP)is proposed.Multiple channels are set up between cells and each channel can only transmit specific objects.On the other hand,each channel has weights that can control the number of objects transferred.For some rules,the execution of the rule is controlled by setting the trigger condition.2.A reweighted multi-view subspace clustering algorithm based on global information is proposed.The relationship between one sample and all other samples is emphasized.Firstly,each self-representation matrix(regarded as the similarity matrix)is constructed by the selfrepresentation method,and each view is fused to obtain the unified similarity matrix and the updated similarity matrix of each view.The updated similarity matrix for each view obtained in the first step is then taken as input,and the view fusion operation is performed to obtain the final similarity matrix.At the same time,the Constrained Laplacian Rank(CLR)is applied to the final matrix so that the clustering results can be obtained directly without additional clustering steps.We construct a WRMCTP system,combine it with the reweighted multi-view subspace clustering algorithm to improve the computational efficiency of the algorithm,and explain its running process in detail.3.An improved multi-view spectral clustering algorithm based on local information is proposed.The relationship between central vertices and their neighbors is emphasized.Firstly,the k-nearest neighbor method is applied to initialize the similarity matrix of each view,and then it is fused into the unified similarity matrix.It should be noted that the unified similarity matrix and the similarity matrix of each view are mutually promoted and updated,so we use the iterative cycle update method to optimize the similarity matrix of each view.In the stage of obtaining the clustering results,the spectral clustering algorithm is combined with the symmetric non-negative matrix decomposition method,and the discrete clustering label matrix is output directly to obtain the clustering results.We design a WRMCTP system and combine it with improved multi-view spectral clustering to improve the efficiency of the algorithm.4.In terms of application,the reweighted multi-view subspace clustering algorithm and the improved multi-view spectral clustering algorithm are used to solve the problem of face recognition classification and text clustering respectively.ORL data set is used for face recognition classification experiment.The data set contains 400 face images of 40 different people,and multiple views are constructed through different feature extraction methods.Experimental results show that the reweighted multi-view subspace clustering algorithm is more accurate than other multi-view clustering algorithms in face recognition.Text clustering experiment is carried out using BBCSports dataset.The BBCSports dataset is composed of 544 news documents from the BBCSports website in 2004-2005,including five topics respectively,athletics,cricket,football,rugby and tennis.Experiments show that the improved multi-view spectral clustering algorithm has better performance of text clustering than other multi-view clustering algorithms.