Unified Structure to Generate Sparsity-Inducing Regularizers for Enhanced Low-Rank Matrix Completion

Zhi-Yong Wang; Hing Cheung So

doi:10.1109/ICECET61485.2024.10698010

Unified Structure to Generate Sparsity-Inducing Regularizers for Enhanced Low-Rank Matrix Completion

^*Corresponding author for this work

Department of Electrical Engineering

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

Abstract

Applying half-quadratic optimization to loss functions can generate the associated regularizers, while these reg-ularizers are usually not able to give a sparse solution. To address it, we put forth an unified structure to produce sparsity-promoting regularizers with closed-form proximity operators. In addition, three commonly-used loss functions are adopted in our structure to generate the corresponding sparsity-inducing regu-larizers, which are then employed as nonconvex rank surrogates to achieve enhanced low-rank matrix completion. Furthermore, algorithms with convergence guarantees are developed, and numerical results demonstrate the effectiveness of our methods in terms of recovery performance and runtime. © 2024 IEEE.

Original language	English
Title of host publication	International Conference on Electrical, Computer and Energy Technologies (ICECET 2024)
Publisher	IEEE
Number of pages	6
ISBN (Electronic)	9798350395914
ISBN (Print)	9798350395921
DOIs	https://doi.org/10.1109/ICECET61485.2024.10698010
Publication status	Published - 2024
Event	4th IEEE International Conference on Electrical, Computer, and Energy Technologies (ICECET 2024) - Sydney, Australia Duration: 25 Jul 2024 → 27 Jul 2024 https://www.icecet.com/2024/

Publication series

Name	International Conference on Electrical, Computer, and Energy Technologies, ICECET

Conference

Conference	4th IEEE International Conference on Electrical, Computer, and Energy Technologies (ICECET 2024)
Country/Territory	Australia
City	Sydney
Period	25/07/24 → 27/07/24
Internet address	https://www.icecet.com/2024/

Funding

This work was supported in part by the Research Grants Council of the Hong Kong Special Administrative Region, China [Project No. CityU 11207922], and in part by the Research Grants of Shenzhen Research Institute, City University of Hong Kong, Shenzhen, China [Project No. R-IND25501].

Research Keywords

matrix completion
rank minimization
rank surrogate
Sparsity-inducing regularizer

Access Output

10.1109/ICECET61485.2024.10698010

Other Access Options

Check CityUHK Library

Permanent Link

Right click to copy link

Cite this

@inproceedings{99b07534b4ec49528e8a11bf162a6c48,

title = "Unified Structure to Generate Sparsity-Inducing Regularizers for Enhanced Low-Rank Matrix Completion",

abstract = "Applying half-quadratic optimization to loss functions can generate the associated regularizers, while these reg-ularizers are usually not able to give a sparse solution. To address it, we put forth an unified structure to produce sparsity-promoting regularizers with closed-form proximity operators. In addition, three commonly-used loss functions are adopted in our structure to generate the corresponding sparsity-inducing regu-larizers, which are then employed as nonconvex rank surrogates to achieve enhanced low-rank matrix completion. Furthermore, algorithms with convergence guarantees are developed, and numerical results demonstrate the effectiveness of our methods in terms of recovery performance and runtime. {\textcopyright} 2024 IEEE.",

keywords = "matrix completion, rank minimization, rank surrogate, Sparsity-inducing regularizer",

author = "Zhi-Yong Wang and So, {Hing Cheung}",

year = "2024",

doi = "10.1109/ICECET61485.2024.10698010",

language = "English",

isbn = "9798350395921",

series = "International Conference on Electrical, Computer, and Energy Technologies, ICECET",

publisher = "IEEE",

booktitle = "International Conference on Electrical, Computer and Energy Technologies (ICECET 2024)",

address = "United States",

note = "4th IEEE International Conference on Electrical, Computer, and Energy Technologies (ICECET 2024) ; Conference date: 25-07-2024 Through 27-07-2024",

url = "https://www.icecet.com/2024/",

}

Wang, Z-Y & So, HC 2024, Unified Structure to Generate Sparsity-Inducing Regularizers for Enhanced Low-Rank Matrix Completion. in International Conference on Electrical, Computer and Energy Technologies (ICECET 2024). International Conference on Electrical, Computer, and Energy Technologies, ICECET, IEEE, 4th IEEE International Conference on Electrical, Computer, and Energy Technologies (ICECET 2024), Sydney, Australia, 25/07/24. https://doi.org/10.1109/ICECET61485.2024.10698010

Unified Structure to Generate Sparsity-Inducing Regularizers for Enhanced Low-Rank Matrix Completion. / Wang, Zhi-Yong ; So, Hing Cheung.
International Conference on Electrical, Computer and Energy Technologies (ICECET 2024). IEEE, 2024. (International Conference on Electrical, Computer, and Energy Technologies, ICECET).

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

TY - GEN

T1 - Unified Structure to Generate Sparsity-Inducing Regularizers for Enhanced Low-Rank Matrix Completion

AU - Wang, Zhi-Yong

AU - So, Hing Cheung

PY - 2024

Y1 - 2024

N2 - Applying half-quadratic optimization to loss functions can generate the associated regularizers, while these reg-ularizers are usually not able to give a sparse solution. To address it, we put forth an unified structure to produce sparsity-promoting regularizers with closed-form proximity operators. In addition, three commonly-used loss functions are adopted in our structure to generate the corresponding sparsity-inducing regu-larizers, which are then employed as nonconvex rank surrogates to achieve enhanced low-rank matrix completion. Furthermore, algorithms with convergence guarantees are developed, and numerical results demonstrate the effectiveness of our methods in terms of recovery performance and runtime. © 2024 IEEE.

AB - Applying half-quadratic optimization to loss functions can generate the associated regularizers, while these reg-ularizers are usually not able to give a sparse solution. To address it, we put forth an unified structure to produce sparsity-promoting regularizers with closed-form proximity operators. In addition, three commonly-used loss functions are adopted in our structure to generate the corresponding sparsity-inducing regu-larizers, which are then employed as nonconvex rank surrogates to achieve enhanced low-rank matrix completion. Furthermore, algorithms with convergence guarantees are developed, and numerical results demonstrate the effectiveness of our methods in terms of recovery performance and runtime. © 2024 IEEE.

KW - matrix completion

KW - rank minimization

KW - rank surrogate

KW - Sparsity-inducing regularizer

UR - http://www.scopus.com/inward/record.url?scp=85207380674&partnerID=8YFLogxK

UR - https://www.scopus.com/record/pubmetrics.uri?eid=2-s2.0-85207380674&origin=recordpage

U2 - 10.1109/ICECET61485.2024.10698010

DO - 10.1109/ICECET61485.2024.10698010

M3 - RGC 32 - Refereed conference paper (with host publication)

SN - 9798350395921

T3 - International Conference on Electrical, Computer, and Energy Technologies, ICECET

BT - International Conference on Electrical, Computer and Energy Technologies (ICECET 2024)

PB - IEEE

T2 - 4th IEEE International Conference on Electrical, Computer, and Energy Technologies (ICECET 2024)

Y2 - 25 July 2024 through 27 July 2024

ER -

Unified Structure to Generate Sparsity-Inducing Regularizers for Enhanced Low-Rank Matrix Completion

Abstract

Publication series

Conference

Funding

Research Keywords

Access Output

Other Access Options

Permanent Link

Other Files and Links

Fingerprint

GRF: Advanced Factorization Approaches for Low-Rank Matrix Recovery

Cite this

Unified Structure to Generate Sparsity-Inducing Regularizers for Enhanced Low-Rank Matrix Completion

Abstract

Publication series

Conference

Funding

Research Keywords

Access Output

Other Access Options

Permanent Link

Other Files and Links

Fingerprint

Projects

GRF: Advanced Factorization Approaches for Low-Rank Matrix Recovery

Cite this