Research Of Random Feature Based Multiple Kernel Collaborative Fuzzy Clustering Method In P2p Distributed Network

In recent years,the rapid development of cloud computing,Internet of Things,big data and other technologies has promoted the application of the P2 P distributed network.This paper aims to propose a novel random feature based multiple kernel distributed collaborative fuzzy clustering method,which can analyze the data in peer-to-peer distributed networks.Finally,this paper proposes four random feature based centralized kernel fuzzy clustering methods and a distributed collaborative fuzzy clustering method.Firstly,this paper proposes a novel random feature based single kernel fuzzy clustering method,in which the random feature method is employed to map the original data into a low-rank randomized space and the fuzzy c-means method is implemented in this space.The experimental results prove the superiority and efficiency of the proposed method.What's more,two random feature methods,the random Fourier feature and the Quasi-Monte Carlo feature,are used in the test.The results show that the Quasi-Monte Carlo method can approximate the kernel function more precisely and lead to better clustering results.Considering the importance of the different dimension of the random feature space to the formation of the clusters is not equivalent,this paper proposes a random feature based single kernel fuzzy clustering method with attribute weights.In this method,the random feature method is employed to map the original data into the random feature space,and the maximum-entropy regularization method is used to optimize the distribution of the attribute weights.The experimental results show the proposed method can extract the important attributes to improve the clustering results.To relief the influence of the selection of the kernel function on the clustering results,this paper proposes a random feature based multiple kernel fuzzy clustering method,in which the random feature method is used to approximate multiple kernel functions and the maximum-entropy regularization method is utilized to assign the kernel weights.In the experiments,the proposed method achieves stable clustering results and is much more efficient than the traditional multiple kernel clustering methods.To improve the clustering results of the multiple kernel fuzzy clustering method,this paper proposes a random feature based multiple kernel fuzzy clustering method with all attribute weights.In this method,the random feature method is used to map the data into multiple random feature spaces,and then the multiple random feature spaces are merged as a combined feature space.The maximum-entropy regularization method is employed to optimize the weights of all the attributes.In the experiments,the proposed method achieves significant improvement on clustering results on most data sets.Finally,this paper proposes a novel random feature based multiple kernel collaborative fuzzy clustering method in P2 P distributed network.In this method,each peer of the network first maps its data into the combined random feature space and assigns the weights for attributes.Then,the peers implement clustering process and communication process repeatedly.In the clustering process,the random feature based multiple kernel fuzzy clustering method with all attribute weights method is performed at each peer independently with its optimization pursuits by focusing on the local data and the findings communicated by neighbor peers at this point of time.In the communication process,each peer exchanges its cluster centers and attribute weights with its neighbor peers.Finally,the global clustering results are achieved at all peers.The experiments show the superiority and efficiency of the proposed method.