Tensor Equation Based Multi-attribute Feature Recognition With Its Efficient Implementations

With the initial formation of the domestic sensor industry cluster,the rapid development of information technology,and the digital transformation of the modern society based on the Internet,a system that integrates digital information,physical information,and social information has gradually formed a system called ”Cyber-Physical-Social Systems”,also known as CPSS.Driven by the continuous development of CPSS,the processing and analysis of high-order and high-dimensional signals has become a new hot research direction in the field of computer science and signal processing.The main research of this thesis can be summarized as follows:1.A multi-attribute feature recognition algorithm based on tensor equation is proposed,and the overall framework is divided into three parts,which are multi-attribute feature extraction,multi-attribute feature matching,and multi-attribute feature ranking.The two processes of multi-attribute feature matching and multi-attribute feature ranking are introduced in detail.Then this paper proposes a method,namely the high-order GMRES subspace algorithm,to solve the main high-order tensor linear equation problem in multi-attribute feature matching.The complexity of the algorithm is analyzed in detail,and application experiments are performed.2.A scheme based on distributed parallelism is proposed to solve the high-order GMRES subspace algorithm.The algorithm is optimized from two aspects of spatial parallelism and temporal parallelism.Among them,spatial parallelism includes optimization of tensor multimode multiplication and optimization of for loop body,and temporal parallel optimization is realized by analyzing synchronization points.Finally,a complete distributed parallel scheme is given.3.Optimization of distributed parallel algorithms.The distributed parallel algorithm has greater limitations on the two models A and B.The A model is optimized in parallel by the inner product of tensor,and the B model is optimized by the coarse-grained parallel optimization.The tensor multimode is given by combining the two schemes.Multiplication parallel optimization algorithm.