Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : CRIStAL UMR 9189 (Lille)
Durée : 3 ans
Contact : pierre.chainais@centralelille.fr
Date limite de publication : 2022-09-15

Contexte :
The project is part of the ANR Chaire IA SHERLOCK (Fast inference with controlled uncertainty: application to astro- physical observations) led by Pierre Chainais (co-funded by Agence Nationale de la Recherche (ANR), ISITE, Centrale Lille Institut and Région Haut-de-France). The succesful candidate will be jointly supervised by Pierre Chainais and Pierre-Antoine Thouvenin in the CRIStAL lab (UMR 9189), Lille, France.
This work will be conducted in the continuity of an ongoing collaboration initiated by Pierre-Antoine Thouvenin with Audrey Repetti – research associate at Heriot-Watt University – and Pierre Chainais. There will be opportunities for short or longer stays at Heriot-Watt University.
The successful candidate will have access to the medium scale computing center from the Universtiy of Lille, and the national flagship Jean Zay supercomputer.

Sujet :
1 Project overview
This project is aimed at accelerating MCMC algorithms for fast Bayesian inference in large scale problems. Applications in astronomy (e.g., hyperspectral imaging) or in remote sensing (e.g., multimodal multi-temporal source separation) could be considered. The project is part of the ANR Chaire IA SHERLOCK led by Pierre Chainais (co-funded by ISITE, Centrale Lille Institut and Région Haut-de-France).
Many signal and image processing applications, ranging from astronomy (Abdulaziz et al. 2019; Cai et al. 2018) to remote sensing (Borsoi et al. 2021; Ghamisi et al. 2019), involve large datasets. In absence of ground truth, fast parameter inference under controlled uncertainty is critical to guarantee the quality of the resulting predictions.
Asynchronous (parallel or distributed) optimization algorithms have recently regained interest due to their potential of acceleration, in comparison with their synchronous counterparts (Hannah et al. 2017). However, optimization algorithms only bring a point estimate, such as the maximum a posteriori (MAP) estimator. Markov-chain Monte Carlo (MCMC) methods bring a richer information by sampling the posterior distribution of the model. MCMC methods are known to induce larger computational costs compared to optimization algorithms. Nevertheless, recent works at the interface between deterministic and stochastic optimization have introduced efficient samplers to address larger datasets (Durmus et al. 2018; Vono et al. 2020). With the exception of (Simsekli et al. 2018; Terenin et al. 2020), asynchronous MCMC algorithms largely remain to be investigated.
This PhD project is aimed at studying the potential of asynchronous MCMC algorithms for fast Bayesian inference in high dimensional problems.
Keywords. Bayesian inference, MCMC algorithms, asynchronous algorithms.

Profil du candidat :
Master 2 or last year engineering school students with major in applied mathematics, computer science or electrical engineering. The project requires a strong background in data science and/or machine learning (statistics, optimization), signal & image processing. Very good Python coding skills are expected. A B2 English level is mandatory.
Knowledge in C++ programming, as well as experience or interest in parallel/distributed code development (MPI, OpenMP, CUDA, …) will be appreciated.

Formation et compétences requises :
A B2 English level is mandatory.
Knowledge in C++ programming, as well as experience or interest in parallel/distributed code development (MPI, OpenMP, CUDA, …) will be appreciated.

Adresse d’emploi :
CRIStAL, Cité Scientifique, 59651 Villeneuve d’Ascq Cedex

Document attaché : 202201242015_phd_project_2022.pdf