Offre en lien avec l’Action/le Réseau : BigData4Astro/– — –
Laboratoire/Entreprise : Institut d’Astrophysique de Paris
Durée : 3 à 6 mois
Contact : florent.leclercq@iap.fr
Date limite de publication : 2022-12-26
Contexte :
Surveys of the cosmic large-scale structure carry rich scientific opportunities. Advancing the research frontier requires solving unique and challenging statistical problems, to unlock the information content of massive and complex data vectors. The recently-proposed machine learning technique BOLFI (Bayesian optimisation for likelihood-free inference) makes inference of complex Bayesian hierarchical models under the constraint of a very limited simulation budget possible. Unfortunately, its use is currently hampered by several theoretical and practical challenges.
Sujet :
The goal of this interdisciplinary project is to upgrade the BOLFI algorithm for the extraction of information distributed in massive and heterogeneous data, in the context of expensive and imperfect data models. Motivating problems and applications will come from upcoming galaxy survey data such as Euclid. We will address several issues, including: (i) the parallel acquisition of simulations when only a limited number of noisy likelihood evaluations can be obtained, (ii) the robustification of the technique against model mis-specification, (iii) the definition of summary statistics that maximise the extraction of information, e.g. via information-maximising neural networks (IMNN). The proposed algorithm will be applied to the inference of cosmological parameters using a realistic simulator. Ultimately, the developed method will be an important tool for the extraction of physical information from Euclid data, which has the potential to influence the design of future data analysis pipelines.
Related links and literature / Version française : https://florent-leclercq.eu/supervision.php#internship-2023-info
Profil du candidat :
The student will get experience of statistical modelling, machine learning, data mining, cosmology, and astronomical observations. They should be comfortable with computing (preferably, experience with python and git). This work could naturally lead to a PhD project in data science and/or cosmology, for example in the large-scale structure and distant Universe group of the Institut d’Astrophysique de Paris (IAP).
Formation et compétences requises :
Interest in information science, machine learning, data science, and a taste for (astro)physics.
Adresse d’emploi :
Institut d’Astrophysique de Paris, 98 bis boulevard Arago, 75014 Paris, France