A Parallel CP Decomposition Algorithm for Sparse Tensor

Abstract

With the advent of high-dimensional data, tensor and tensor decomposition methods have received widespread attention in the field of data analysis. However, the high-dimensional and sparse characteristics of tensor data lead to the high computation complexity of algorithm, which became an obstacle to the actual application of the tensor decomposition algorithm. In this paper, we start from the high sparsity of real tensor, and give a GPU parallel sparse tensor CP decomposition algorithm called ParSCP-ALS.