AI data analysis

The project will develop new algorithms for the optimal exploitation of astronomical observations, enabling the detection and characterization of exoplanets with unrivalled sensitivity.

Loïc DENIS, Jordan FRECON-DELOIRE, Rémi EMONET, Laboratoire Hubert Curien (UMR CNRS-UJM-IOGS), Université Jean Monnet Saint-Etienne

Olivier FLASSEUR, Ferréol SOULEZ, Maud LANGLOIS, Éric THIEBAUT, Centre de Recherche Astrophysique de Lyon (UMR CNRS-UCBL-ENS Lyon), CNRS

Anne-Marie LAGRANGE, Laboratoire d’Études Spatiales et d’Instrumentation en Astrophysique (UMR CNRS-Obs. Paris PSL-Sorbonne Univ.-Univ. Paris Cité), CNRS

Olivier MICHEL, Florent CHATELAIN, Laboratoire Grenoble Images Parole Signal Automatique (UMR CNRS-Grenoble INP-Univ. Grenoble-Alpes), CNRS

Nelly PUSTELNIK, Julian TACHELLA, Laboratoire de Physique (UMR CNRS-ENS Lyon), CNRS

Hervé LE COROLLER, Laboratoire d’Astrophysique de Marseille (UMR CNRS-Aix Marseille Univ-CNES), CNRS

François ORIEUX, Laboratoire des Signaux et Systèmes (UMR CNRS-CentraleSupélec-Univ Paris Saclay), CNRS

Laurent MUGNIER, Olivier HERSCOVICI-SCHILLER, Cyril PETIT, Département Optique et Techniques Associées, ONERA

David MOUILLET, Xavier BONFILS, Faustine CANTALLOUBE, Mickael BONNEFOY, Julien MILLI, Alexis CARLOTTI, Xavier DELFOSSE, Jean-Philippe BERGER, Institut de Planétologie et d’Astrophysique de Grenoble (UMR CNRS-Univ Grenoble Alpes), Université Grenoble Alpes

Jean-François GIOVANNELLI, Laboratoire Intégration du Matériel au Système (UMR CNRS-Univ Bordeaux, Bordeaux INP), Université de Bordeaux

Hervé CARFANTAN, Institut de Recherche en Astrophysique et Planétologie (UMR CNRS-Univ Toulouse III-CNES), Université Paul Sabatier Toulouse III

André FERRARI, Simon PRUNET, Laboratoire J-L Lagrange (UMR CNRS Obs Côte d’Azur, Univ Côte d’Azur), Observatoire de la Côte d’Azur

Jean PONCE, Julien MAIRAL, Équipes WILLOW (Inria Rocquencourt, ENS Paris, CNRS) et THOTH (Inria Grenoble), Inria

Artificial intelligence coupled with advanced signal processing techniques will jointly analyze multivariate measurements (spatial, spectral, temporal), including physics, instrumental models and knowledge from archives of past observations. Particular emphasis will be placed on the automatic estimation of instrumental parameters and algorithms, to improve the robustness of processing, and on the characterization of uncertainties, sensitivity limits and false alarm probabilities, which are essential information for the astrophysical exploitation of results.


Develop methods for detecting and characterizing exoplanets and their environments

Design new self-supervised algorithms for source detection and image reconstruction. These algorithms must counter the effects of instrumental blur, fluctuations (notably due to adaptive optics corrections) and measurement noise. Exploit self-supervised learning techniques (i.e., with no information other than observations and an instrumental model) and algorithms unrolled in deep-network AI architectures. Model and learn about statistical fluctuations in measurements and their correlation structures (spatial, spectral, temporal).


Advanced signal processing for high-resolution spectroscopy

Detect exoplanets by coupling high-contrast imaging observations with high-resolution spectroscopy. Develop new methods for detecting and characterizing exoplanets using spectro-velocimetry.


Combine multi-date observations and exploit observation archives

Detect exoplanets by combining sets of multi-date observations and estimation of orbital parameters. Deep learning based on the coupling between instrumental models and large archives of observations. Learning instrumental fluctuations from scientific observations.


Multi-instrument and multi-modality fusion

Fusion of high-contrast imaging and optical interferometry observations.


Université Jean Monnet Saint-Etienne, CNRS, ONERA, Université Grenoble Alpes, Université de Bordeaux, Université Paul Sabatier Toulouse III, Observatoire de la Côte d’Azur, INRIA

Saint-Etienne (Lab. Hubert Curien), Lyon (CRAL, Lab. de Physique), Paris (LESIA, Inria, L2S, ONERA), Grenoble (IPAG, GIPSA-lab, Inria), Marseille (LAM), Bordeaux (IMS), Toulouse (IRAP), Nice (Lagrange).


Scientific expectations

The project will provide the astronomical community with methods to better process data from their instruments dedicated to the search for exoplanets, their characterization and the analysis of their interactions with their environment (proto-planetary disk, other exoplanets). The source code of the algorithms developed in the project will be freely distributed. The expertise acquired in data processing will help guide the design of future generations of instruments.


Societal impacts

The problems of detection, estimation and characterization using artificial intelligence methods integrating instrumental models can find numerous outlets beyond astronomy, in particular in Earth observation (satellite imagery for environmental studies), biomedical imagery (localization of markers in fluorescence microscopy, image reconstruction in the medical field, multi-modality fusion).


Skills development

A multi-disciplinary community of 31 researchers and teacher-researchers specializing in astronomy, instrumentation, signal and image processing, and artificial intelligence, as well as 2 post-docs and 7 PhD students (including one co-funded by ONERA and one co-funded by the University of Bordeaux).