About us
Espace utilisateur
INSTN offers more than 40 diplomas from operator level to post-graduate degree level. 30% of our students are international students.
Professionnal development
Professionnal development
Find a training course
INSTN delivers off-the-self or tailor-made training courses to support the operational excellence of your talents.
Human capital solutions
At INSTN, we are committed to providing our partners with the best human capital solutions to develop and deliver safe & sustainable projects.
Home   /   Thesis   /   Radio Image Reconstruction for Multi-Messenger Astronomy

Radio Image Reconstruction for Multi-Messenger Astronomy


Cosmology in the 21st century aims to better our understanding of the Universe by seeking to answer open questions concerning the nature of dark matter and dark energy, and the precise expansion rate of the Universe. In order to tackle these questions, it is essential to take advantage of all the data made available in the current era of multi-messenger astronomy, capitalising on the latest advances in signal processing, machine learning and the handling of big data.

Current and upcoming optical surveys, such as KiDS-450 [1], the Dark Energy Survey Year 1 [2], the Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST) [3] and Euclid [4], are probing wider and deeper patches of the late-time Universe to improve constraints on cosmological parameters by measuring the shapes and distributions of galaxies. These parameters can be compared to the latest analyses of the cosmic microwave background radiation (i.e. the early Universe) by surveys like Planck [5]. Recent studies [6] highlight some discrepancy between early and late-time analyses indicating some systematic uncertainty or an incomplete model of cosmology.

Additionally, In recent years gravitational wave interferometers, such as LIGO and Virgo, have pushed astronomical observations beyond the electromagnetic spectrum. This has made it possible to detect the interaction of distant neutron stars and black holes.

Radio wavelengths provide a complementary and independent probe of the late-time Universe. Radio astronomy provides the advantage of probing higher redshifts, having a deterministic point spread function (PSF) and being less sensitive to PSF anisotropies [7]. Cross-correlations between radio and optical surveys can additionally alleviate systematics effects such as intrinsic alignments improving cosmological constraints [8-9]. Upcoming radio surveys, such as the Square Kilometre Array (SKA), are designed to reach an order of magnitude greater sensitivity and survey speed than existing instruments. SKA has the potential to add significant additional constraints on cosmological parameters given the vast sky area it will cover (~75%). This, however, comes at the cost of having to manage extremely large scales of data and complicated image reconstruction. SKA is expected to produce ~1 TB of data every second. With typical observations taking ~6h and a total lifespan of 15 years, SKA will produce data in the Exabyte (1018 bytes) scale [10], making it one of the biggest data management problems in modern science.

The CosmoStat team has been pioneering the use of signal processing and machine learning techniques for solving inverse problems in astronomical image reconstruction. Applying these methods to radio-interferometric data, however, brings a host of new challenges, in part due to the additional complexity of the inverse problems to solve, but also due to the extremely large-scale dimensions of the problem. Conventional deep learning approaches will not be able to scale to the typical size of an SKA field, and developing efficient model-parallelism approaches will be necessary.


Institut de recherche sur les lois fondamentales de l’univers
Direction d’Astrophysique
Laboratoire CosmoStat
Top envelopegraduation-hatlicensebookuserusersmap-markercalendar-fullbubblecrossmenuarrow-down