IEICE TRANSACTIONS on Information
Block Randomized Singular Value Decomposition on GPUs
Summary :
Fast computation of singular value decomposition (SVD) is of great interest in various machine learning tasks. Recently, SVD methods based on randomized linear algebra have shown significant speedup in this regime. For processing large-scale data, computing systems with accelerators like GPUs have become the mainstream approach. In those systems, access to the input data dominates the overall process time; therefore, it is needed to design an out-of-core algorithm to dispatch the computation into accelerators. This paper proposes an accurate two-pass randomized SVD, named block randomized SVD (BRSVD), designed for matrices with a slow-decay singular spectrum that is often observed in image data. BRSVD fully utilizes the power of modern computing system architectures and efficiently processes large-scale data in a parallel and out-of-core fashion. Our experiments show that BRSVD effectively moves the performance bottleneck from data transfer to computation, so that outperforms existing randomized SVD methods in terms of speed with retaining similar accuracy.
- Publication
- IEICE TRANSACTIONS on Information Vol.E103-D No.9 pp.1949-1959
- Publication Date
- 2020/09/01
- Publicized
- 2020/06/08
- Online ISSN
- 1745-1361
- DOI
- 10.1587/transinf.2019EDP7265
- Type of Manuscript
- PAPER
- Category
- Dependable Computing
Authors
Yuechao LU
Osaka University
Yasuyuki MATSUSHITA
Osaka University
Fumihiko INO
Osaka University
Keyword
Latest Issue
Copyrights notice of machine-translated contents
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Cite this
Copy
Yuechao LU, Yasuyuki MATSUSHITA, Fumihiko INO, "Block Randomized Singular Value Decomposition on GPUs" in IEICE TRANSACTIONS on Information,
vol. E103-D, no. 9, pp. 1949-1959, September 2020, doi: 10.1587/transinf.2019EDP7265.
Abstract: Fast computation of singular value decomposition (SVD) is of great interest in various machine learning tasks. Recently, SVD methods based on randomized linear algebra have shown significant speedup in this regime. For processing large-scale data, computing systems with accelerators like GPUs have become the mainstream approach. In those systems, access to the input data dominates the overall process time; therefore, it is needed to design an out-of-core algorithm to dispatch the computation into accelerators. This paper proposes an accurate two-pass randomized SVD, named block randomized SVD (BRSVD), designed for matrices with a slow-decay singular spectrum that is often observed in image data. BRSVD fully utilizes the power of modern computing system architectures and efficiently processes large-scale data in a parallel and out-of-core fashion. Our experiments show that BRSVD effectively moves the performance bottleneck from data transfer to computation, so that outperforms existing randomized SVD methods in terms of speed with retaining similar accuracy.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2019EDP7265/_p
Copy
@ARTICLE{e103-d_9_1949,
author={Yuechao LU, Yasuyuki MATSUSHITA, Fumihiko INO, },
journal={IEICE TRANSACTIONS on Information},
title={Block Randomized Singular Value Decomposition on GPUs},
year={2020},
volume={E103-D},
number={9},
pages={1949-1959},
abstract={Fast computation of singular value decomposition (SVD) is of great interest in various machine learning tasks. Recently, SVD methods based on randomized linear algebra have shown significant speedup in this regime. For processing large-scale data, computing systems with accelerators like GPUs have become the mainstream approach. In those systems, access to the input data dominates the overall process time; therefore, it is needed to design an out-of-core algorithm to dispatch the computation into accelerators. This paper proposes an accurate two-pass randomized SVD, named block randomized SVD (BRSVD), designed for matrices with a slow-decay singular spectrum that is often observed in image data. BRSVD fully utilizes the power of modern computing system architectures and efficiently processes large-scale data in a parallel and out-of-core fashion. Our experiments show that BRSVD effectively moves the performance bottleneck from data transfer to computation, so that outperforms existing randomized SVD methods in terms of speed with retaining similar accuracy.},
keywords={},
doi={10.1587/transinf.2019EDP7265},
ISSN={1745-1361},
month={September},}
Copy
TY - JOUR
TI - Block Randomized Singular Value Decomposition on GPUs
T2 - IEICE TRANSACTIONS on Information
SP - 1949
EP - 1959
AU - Yuechao LU
AU - Yasuyuki MATSUSHITA
AU - Fumihiko INO
PY - 2020
DO - 10.1587/transinf.2019EDP7265
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E103-D
IS - 9
JA - IEICE TRANSACTIONS on Information
Y1 - September 2020
AB - Fast computation of singular value decomposition (SVD) is of great interest in various machine learning tasks. Recently, SVD methods based on randomized linear algebra have shown significant speedup in this regime. For processing large-scale data, computing systems with accelerators like GPUs have become the mainstream approach. In those systems, access to the input data dominates the overall process time; therefore, it is needed to design an out-of-core algorithm to dispatch the computation into accelerators. This paper proposes an accurate two-pass randomized SVD, named block randomized SVD (BRSVD), designed for matrices with a slow-decay singular spectrum that is often observed in image data. BRSVD fully utilizes the power of modern computing system architectures and efficiently processes large-scale data in a parallel and out-of-core fashion. Our experiments show that BRSVD effectively moves the performance bottleneck from data transfer to computation, so that outperforms existing randomized SVD methods in terms of speed with retaining similar accuracy.
ER -
English
