Parallel faceted imaging in radio interferometry via proximal splitting (Faceted HyperSARA): when precision meets scalability

Pierre-Antoine Thouvenin, Abdullah Abdulaziz, Ming Jiang, Arwa Dabbech, Audrey Repetti, Adrian Jackson, Jean-Philippe Thiran, Yves Wiaux

Research output: Contribution to journalArticle

1 Downloads (Pure)

Abstract

Upcoming radio interferometers are aiming to image the sky at new levels of resolution and sensitivity, with wide-band image cubes reaching close to the Petabyte scale for SKA. Modern proximal optimization algorithms have recently shown a potential to significantly outperform clean thanks to their ability to inject complex image models to regularize the inverse problem for image formation from visibility data. They are also scalable to large data volumes thanks to a splitting functionality enabling the decomposition of data into blocks, for parallel processing of all block-specific data-fidelity terms of the objective function. In the present work, the same splitting functionality is further exploited to decompose the target image cube into spatio-spectral facets, and enable parallel processing of facet-specific regularization terms in the objective. The resulting algorithm, dubbed “Faceted HyperSARA”, was implemented in MATLAB (code available on the Puri-Psi webpage). Extensive simulation results on synthetic image cubes confirm that faceting can provide a major increase in scalability at no cost in imaging quality. A proof-of-concept reconstruction of a 15 GB image cube of Cyg A from 7.4 GB of VLA data, utilizing 496 CPU cores on a High Performance Computing system for 68 hours, confirms both scalability and a quantum jump in imaging quality from CLEAN. Last but not least, we also combined Faceted HyperSARA with a joint image and data dimensionality reduction technique, under the assumption of a slow spectral slope of Cyg A in the frequency range of interest. Our results show that dimensionality reduction enables utilizing no more than 31 CPU cores for 142 hours to form the image while preserving the overall reconstruction quality, thus demonstrating a second scalability feature beyond faceting. Reconstructed image cubes are available online (see attached dataset).
Original languageEnglish
JournalMonthly Notices of the Royal Astronomical Society
Publication statusSubmitted - Jan 2020

Fingerprint

Interferometry
Scalability
Imaging techniques
Program processors
Processing
Inverse problems
Visibility
Interferometers
MATLAB
Data reduction
Image processing
Decomposition
Costs

Keywords

  • techniques: image processing
  • techniques: interferometric

Cite this

@article{b0f5072dfc9a41a4a17e008d5ef154c0,
title = "Parallel faceted imaging in radio interferometry via proximal splitting (Faceted HyperSARA): when precision meets scalability",
abstract = "Upcoming radio interferometers are aiming to image the sky at new levels of resolution and sensitivity, with wide-band image cubes reaching close to the Petabyte scale for SKA. Modern proximal optimization algorithms have recently shown a potential to significantly outperform clean thanks to their ability to inject complex image models to regularize the inverse problem for image formation from visibility data. They are also scalable to large data volumes thanks to a splitting functionality enabling the decomposition of data into blocks, for parallel processing of all block-specific data-fidelity terms of the objective function. In the present work, the same splitting functionality is further exploited to decompose the target image cube into spatio-spectral facets, and enable parallel processing of facet-specific regularization terms in the objective. The resulting algorithm, dubbed “Faceted HyperSARA”, was implemented in MATLAB (code available on the Puri-Psi webpage). Extensive simulation results on synthetic image cubes confirm that faceting can provide a major increase in scalability at no cost in imaging quality. A proof-of-concept reconstruction of a 15 GB image cube of Cyg A from 7.4 GB of VLA data, utilizing 496 CPU cores on a High Performance Computing system for 68 hours, confirms both scalability and a quantum jump in imaging quality from CLEAN. Last but not least, we also combined Faceted HyperSARA with a joint image and data dimensionality reduction technique, under the assumption of a slow spectral slope of Cyg A in the frequency range of interest. Our results show that dimensionality reduction enables utilizing no more than 31 CPU cores for 142 hours to form the image while preserving the overall reconstruction quality, thus demonstrating a second scalability feature beyond faceting. Reconstructed image cubes are available online (see attached dataset).",
keywords = "techniques: image processing, techniques: interferometric",
author = "Pierre-Antoine Thouvenin and Abdullah Abdulaziz and Ming Jiang and Arwa Dabbech and Audrey Repetti and Adrian Jackson and Jean-Philippe Thiran and Yves Wiaux",
year = "2020",
month = "1",
language = "English",
journal = "Monthly Notices of the Royal Astronomical Society",
issn = "0035-8711",
publisher = "Oxford University Press",

}

Parallel faceted imaging in radio interferometry via proximal splitting (Faceted HyperSARA): when precision meets scalability. / Thouvenin, Pierre-Antoine; Abdulaziz, Abdullah; Jiang, Ming; Dabbech, Arwa; Repetti, Audrey; Jackson, Adrian; Thiran, Jean-Philippe; Wiaux, Yves.

In: Monthly Notices of the Royal Astronomical Society, 01.2020.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Parallel faceted imaging in radio interferometry via proximal splitting (Faceted HyperSARA): when precision meets scalability

AU - Thouvenin, Pierre-Antoine

AU - Abdulaziz, Abdullah

AU - Jiang, Ming

AU - Dabbech, Arwa

AU - Repetti, Audrey

AU - Jackson, Adrian

AU - Thiran, Jean-Philippe

AU - Wiaux, Yves

PY - 2020/1

Y1 - 2020/1

N2 - Upcoming radio interferometers are aiming to image the sky at new levels of resolution and sensitivity, with wide-band image cubes reaching close to the Petabyte scale for SKA. Modern proximal optimization algorithms have recently shown a potential to significantly outperform clean thanks to their ability to inject complex image models to regularize the inverse problem for image formation from visibility data. They are also scalable to large data volumes thanks to a splitting functionality enabling the decomposition of data into blocks, for parallel processing of all block-specific data-fidelity terms of the objective function. In the present work, the same splitting functionality is further exploited to decompose the target image cube into spatio-spectral facets, and enable parallel processing of facet-specific regularization terms in the objective. The resulting algorithm, dubbed “Faceted HyperSARA”, was implemented in MATLAB (code available on the Puri-Psi webpage). Extensive simulation results on synthetic image cubes confirm that faceting can provide a major increase in scalability at no cost in imaging quality. A proof-of-concept reconstruction of a 15 GB image cube of Cyg A from 7.4 GB of VLA data, utilizing 496 CPU cores on a High Performance Computing system for 68 hours, confirms both scalability and a quantum jump in imaging quality from CLEAN. Last but not least, we also combined Faceted HyperSARA with a joint image and data dimensionality reduction technique, under the assumption of a slow spectral slope of Cyg A in the frequency range of interest. Our results show that dimensionality reduction enables utilizing no more than 31 CPU cores for 142 hours to form the image while preserving the overall reconstruction quality, thus demonstrating a second scalability feature beyond faceting. Reconstructed image cubes are available online (see attached dataset).

AB - Upcoming radio interferometers are aiming to image the sky at new levels of resolution and sensitivity, with wide-band image cubes reaching close to the Petabyte scale for SKA. Modern proximal optimization algorithms have recently shown a potential to significantly outperform clean thanks to their ability to inject complex image models to regularize the inverse problem for image formation from visibility data. They are also scalable to large data volumes thanks to a splitting functionality enabling the decomposition of data into blocks, for parallel processing of all block-specific data-fidelity terms of the objective function. In the present work, the same splitting functionality is further exploited to decompose the target image cube into spatio-spectral facets, and enable parallel processing of facet-specific regularization terms in the objective. The resulting algorithm, dubbed “Faceted HyperSARA”, was implemented in MATLAB (code available on the Puri-Psi webpage). Extensive simulation results on synthetic image cubes confirm that faceting can provide a major increase in scalability at no cost in imaging quality. A proof-of-concept reconstruction of a 15 GB image cube of Cyg A from 7.4 GB of VLA data, utilizing 496 CPU cores on a High Performance Computing system for 68 hours, confirms both scalability and a quantum jump in imaging quality from CLEAN. Last but not least, we also combined Faceted HyperSARA with a joint image and data dimensionality reduction technique, under the assumption of a slow spectral slope of Cyg A in the frequency range of interest. Our results show that dimensionality reduction enables utilizing no more than 31 CPU cores for 142 hours to form the image while preserving the overall reconstruction quality, thus demonstrating a second scalability feature beyond faceting. Reconstructed image cubes are available online (see attached dataset).

KW - techniques: image processing

KW - techniques: interferometric

M3 - Article

JO - Monthly Notices of the Royal Astronomical Society

JF - Monthly Notices of the Royal Astronomical Society

SN - 0035-8711

ER -