1: \begin{abstract}

2: The goal of this paper is to find a low-rank approximation for a given tensor.

3: Specifically, we give a computable strategy on calculating the rank of a given tensor,

4: based on approximating the solution to an

5: NP-hard problem. In this paper, we formulate a sparse optimization problem via an $l_1$-regularization to find a low-rank approximation of tensors.

6: To solve this sparse optimization problem, we propose a rescaling algorithm of the proximal alternating minimization

7: and study the theoretical convergence of this algorithm.

8: Furthermore, we discuss the probabilistic consistency of the sparsity result

9: and suggest a way to choose the regularization parameter for practical computation.

10: In the simulation experiments, the performance of our algorithm supports that our method provides

11: an efficient estimate on the number of rank-one tensor components in a given tensor.

12: Moreover, this algorithm is also applied to surveillance videos for low-rank approximation.

13: \end{abstract}

14: