Design And Implementation Of Distributed Higher-order Singular Value Decomposition Algorithm

With the development of Internet of Things,Cyber-Physical-Social Systems(CPSS)is becoming a new research field and hotspot.It has been shown that tensor decomposition,as a new data representation and analysis method,has the capability to organize and process multisource,multimode,spatially distributed sensor data in CPSS.Higher-order Singular Value Decomposition(HOSVD)is one of the most widely used tensor decomposition algorithm for data analysis and processing.However,for large-scale tensor data,due to the limitations include computation ability,memory capability,the computing efficiency of the traditional HOSVD algorithm is low.Therefore,it is very meaningful to design a distributed HOSVD algorithm for the tensor data.In this paper,we propose a new distributed HOSVD algorithm based on divide and conquer strategy,one-sided jacobi svd algorithm and tree structure.It is used to simplify large computations by breaking original tensor down into many smaller sub-tensors,the results of which can then be recombined to generate the decomposition results of the original tensor.Through the analysis of tensor unfolding principles,the law of recombination is found,and the corresponding algorithm is designed.Based on these work,the paper further optimizes the distributed HOSVD algorithm,and proposes two distributed HOSVD algorithms with embedded tree structure,which greatly improves the efficiency of the algorithm.The experiments and simulations show that the proposed algorithms has the merits of high precision and calculation efficiency,and can deal with large scale data.