Big Data and Machine Learning

Research

In our Galaxy, star formation is resolved on all spatial scales (from the milli to the kiloparsec) and a wealth of existing data allows to describe star formation. However, the way to analyse these data to extract information from it is out of reach for « classical » human research and statistical analysis. Recent progress in Machine Learning allow to envision the treatment of this galactic information to build a model of the observed star formation.
We have started to work on this project with a proof-of-concept study, BigSF, funded by A*Midex (2020-2022, PI : A. Zavagno). This work has been done in collaboration with F-X Dupé, S. Bensaid (A*Midex post-doc) M. Gray, S. Schisano (INAF-IAPS), G. Li Causi.
We used the sample of Galactic filaments studied by Schisano et al. (2020) and test different approches in supervised machine learning to classify the pixels as belonging (or not) to the filament class. The output (the segmented map) is a map showing the probability, for a pixel, to belong to the filament class.

This image shows the comparison between the machine learning approach and the filaments identified previsously (from the Hi-GAL survey) from the work of Schisano et al.(2020). The 2MASS K image is shown for comparison but is not used in the learning process. This figure is extracted from the paper Zavagno, Dupé, Bensaid et al. 2023 (A&A 669, A120).
The next step that we explore now is to work using a multi wavelength and multi view representation of these Galactic filaments. This project is ongoing with Loris Berthelot, a PhD student working at LIS and LAM under the supervision of T. Artières and A. Zavagno. The project is funded by the CNRS 80PRIME program that promotes interdisciplinary research.

Results obtained using supervised learning on Galactic filaments

Results presented at the conference SFtools-bigdata


References
Schisano et al. 2020, MNRAS, 492, 5420
Signoroni et al. 2019, Journal of Imaging, 5, 52
N'Diaye-Faye, P. 2019, Master 2 Internship Report (in French)
Zavagno, A. et al., A&A, 669, A120
INSU Press release, January 23, 2023, in french , in english

Back to Research page