Olivier Absil
Salle des séminaires Évry Schatzman, Institut d'Astrophysique de Paris
|
| « Using deep learning for exoplanet high-contrast imaging » |
|
Olivier Absil |
| High-contrast imaging (HCI) is an important tool for the study of planetary systems, providing access to the outer part of those systems, as well as a direct view on planet formation processes in young systems. Yet, our current HCI view of planetary systems is hampered by the fact that state-of-the-art HCI instruments are struggling to push detection limits below Jupiter-mass objects and towards au-scale regions, where most planets are supposed to be formed. While the development of high-performance HCI instruments on ever bigger telescopes is an active field of research, one should not underestimate the untapped potential of current HCI instruments. In this context, the rapid development of deep learning techniques provides an opportunity to revisit HCI methods in various ways. In this seminar, I will describe how deep learning can be deployed to improve the four major pillars of HCI: adaptive optics, coronagraphy, on-sky operations, and image processing. My presentation will particularly focus on our early attempts to deploy supervised learning techniques to process HCI data, and the various hurdles we faced along the way. I will also give a detailed description of how supervised, unsupervised, and reinforcement learning techniques can be used to mitigate the harmful effects of quasi-static aberrations inside HCI instruments, which represent a well-know limitation to the HCI performance. |
|
vendredi 13 juin 2025 - 11:00 Salle des séminaires Évry Schatzman, Institut d'Astrophysique de Paris |
| Page web du séminaire / Seminar's webpage |