所属单位：

理学院

发表刊物：

Journal of Computational Information Systems

关键字：

中文关键字：张量补全，英文关键字：Tensor Completion;

摘要：

Tensor nuclear norm minimization (TNNM) is a commonly-used model for solving low-rank tensor completion (LRTC) problems. Generally, algorithms to TNNM have very heavy computation burden due to the involvement of multiple singular value decompositions (SVDs) at each loop. To address efficiently LRTC, this paper proposes a Tucker decompositions technique which adopts the thin QR decompositions instead of SVDs. First, we establish a minimization model for LRTC based on Tucker decompositions and analyze its first-order optimality conditions. Then, we develop an iterative algorithm to the proposed optimization problem and prove its convergence. Finally, experimental results demonstrate that our method is competitive to other existing methods in computation complexity, completion accuracy and compression performance.

备注：

EI

合写作者：

雍龙泉

第一作者：

郑秀云,杨威,史加荣

论文类型：

期刊论文

卷号：

卷：11

期号：

期：10

页面范围：

页：3759-3768

是否译文：

否

发表时间：

2015-05-01