High Dimensional Decomposition Of Coherent/Structured Matrices Via Sequential Column/Row Sampling
Keywords
Low Rank Matrix; Matrix Decomposition; Randomized Method; Robust PCA; Subspace Recovery
Abstract
This paper focuses on the low rank plus sparse matrix decomposition problem in big data settings. Conventional algorithms solve high-dimensional optimization problems that scale with the data dimension, which limits their scalability. In addition, existing randomized approaches mostly rely on blind random sampling. In this paper, the drawbacks of random sampling from coherent/structured data matrices are analyzed showing that random sampling cannot provide efficient descriptive sketches of coherent data. In addition, a column/row subspace pursuit algorithm which recovers the low rank component via a small set of informative columns/rows of the data is proposed. The obtained column and row spaces are updated in each iteration to converge to the column and row spaces of the low rank matrix. The informative columns are located using the information embedded in the row space while the informative rows are identified using the information embedded in the column space.
Publication Date
6-16-2017
Publication Title
ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings
Number of Pages
6419-6423
Document Type
Article; Proceedings Paper
Personal Identifier
scopus
DOI Link
https://doi.org/10.1109/ICASSP.2017.7953392
Copyright Status
Unknown
Socpus ID
85023772831 (Scopus)
Source API URL
https://api.elsevier.com/content/abstract/scopus_id/85023772831
STARS Citation
Rahmani, Mostafa and Atia, George, "High Dimensional Decomposition Of Coherent/Structured Matrices Via Sequential Column/Row Sampling" (2017). Scopus Export 2015-2019. 7482.
https://stars.library.ucf.edu/scopus2015/7482