Missions

More than 5,000 exoplanets have been discovered to date. The inventory is still ongoing, but our field is now heavily investing in the characterization of the physical and chemical properties of these objects through the use of sensitive imaging cameras and spectrographs with various spectral resolutions, operating from 0.5 to 28 µm.

Our group has developed the Bayesian modeling tool FORMOSA (Petrus et al. 2023). It allows the inference of low-resolution (R = λ/Δλ = 30) and high-resolution (R = 100,000) spectra of exoplanets using precomputed grids of models. The code can estimate global properties (effective temperature, pressure–temperature profiles) as well as atmospheric compositions, vertical cloud distributions, and particle sizes.

The nature of the data we interpret with FORMOSA is becoming increasingly complex as new instruments come online. Data formats can be heterogeneous, datasets unbalanced, and they may cover overlapping wavelength ranges with different signal-to-noise ratios and spectral resolutions. Moreover, the atmospheric models we use remain imperfect and can exhibit systematic deviations that bias the characterization process.

Data fusion is a branch of data science (see Lahat et al. 2015) that aims to combine datasets corresponding to the same phenomenon for decision-making or to obtain tighter constraints on a model (e.g., in weather forecasting). In this context, jointly considering data acquired by different instruments on the same objects has the potential to yield more robust and reliable estimates. Examples of data fusion in astrophysics have already been demonstrated in the context of disentangling galaxy spectra from different modalities (see the ODHIN method in Bacon et al., 2023), based on simplified assumptions. Overcoming the limitations inherent to these methods and adapting them to exoplanet imaging modalities represents a new challenge.

The postdoctoral researcher will investigate how data fusion techniques can be incorporated into FORMOSA to address the challenges described above. They will propose new methodologies based on statistical inference and machine learning applied to data fusion and adapt them to the field of exoplanet characterization. They will develop and maintain the FORMOSA code in coordination with the team of students working on its development in France, and will apply the new methodology to exoplanet characterization as a proof of concept.

References

Petrus et al. 2023, A&A, 670, 9.

Lahat et al. 2015, Proceedings of the IEEE, 103 (9), 1449–1477.

Bacon et al. 2023, A&A, 670, A4.

Activities

-Propose and adapt innovative approaches based on data fusion to address the challenges mentioned above.
-Publish the results in top-tier (A-ranked) journals.
-Communicate the results to both the data science and astronomy communities.
-Coordinate, maintain, and document developments within the FORMOSA code.

Skills

-PhD in Applied Mathematics, Computer Vision, or Data Science.
-Knowledge of statistical inference methods and machine learning.
-Experience in spectroscopy and imaging is an asset.
-Strong programming skills in Python; familiarity with Julia and Matlab is appreciated.
-Excellent written communication skills in English.

Work Context

The successful candidate will work within the research groups of Mickaël Bonnefoy, Mauro Dalla Mura, and Florent Chatelain at the Institut de Planétologie et d'Astrophysique de Grenoble (IPAG) and GIPSA-Lab (Grenoble Images Parole Signal Automatique), both located on the main campus in Grenoble. These laboratories provide a rich collaborative environment at the intersection of astrophysics and signal processing.
The work will be carried out as part of the ANR MIRAGES project, hosted by the LAM (Marseille), LESIA (Paris), IPAG (Grenoble), and Lagrange (Nice) laboratories, and coordinated by A. Vigan (LAM). MIRAGES focuses on the characterization of exoplanets and aims to exploit data from the newly commissioned HiRISE instrument.

Constraints and risks

None

Additional Information

Flexible starting date