Tensor ring decomposition-based model with interpretable gradient factors regularization for tensor completion

Tensor ring (TR) decomposition, which factorizes a tensor into a sequence of cyclically interconnected third-order TR factors, is a powerful tool to capture the global low-rankness of high-dimensional data. However, the understanding of the physical interpretation of TR factors is not clear. In this...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Knowledge-based systems Ročník 259; s. 110094
Hlavní autoři: Wu, Peng-Ling, Zhao, Xi-Le, Ding, Meng, Zheng, Yu-Bang, Cui, Lu-Bin, Huang, Ting-Zhu
Médium: Journal Article
Jazyk:angličtina
Vydáno: Elsevier B.V 10.01.2023
Témata:
Interpretable gradient factors
Proximal alternating minimization
Tensor completion
Tensor ring decomposition
Tensor completion
Proximal alternating minimization
Tensor ring decomposition
Interpretable gradient factors
ISSN:0950-7051, 1872-7409
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Tensor ring (TR) decomposition, which factorizes a tensor into a sequence of cyclically interconnected third-order TR factors, is a powerful tool to capture the global low-rankness of high-dimensional data. However, the understanding of the physical interpretation of TR factors is not clear. In this paper, we first empirically discover the physical interpretation of TR factors in the gradient domain (termed as gradient factors) and then give the theoretical justification. Based on the interpretable gradient factors, we suggest a TR decomposition-based model with interpretable gradient factors regularization (TR-GFR) for tensor completion. To be specific, we consider the low-rankness and transformed sparsity priors of gradient factors to boost the performance and robustness of TR decomposition-based model. In addition, we develop an effective proximal alternating minimization algorithm to solve the proposed model. Numerical experiments validate that the proposed TR-GFR is superior to the compared state-of-the-art methods in terms of PSNR and SSIM values and more robust with TR rank.
ISSN:0950-7051
1872-7409
DOI:10.1016/j.knosys.2022.110094