This PhD thesis proposal addresses the limitations of numerical simulations in extreme-scale engineering modeling, particularly regarding uncertainty quantification and high computational costs. Neural operators will be investigated to design effective surrogate models that accelerate evaluations of partial differential equations, particularly in the context of broadband regional seismic wave propagation.
Key objectives of the PhD thesis include scaling neural operators for real-world 3D problems, overcoming spatial domain size limitations, and leveraging multiscale approaches. The proposal emphasizes developing transfer learning strategies to handle dataset shifts and enhance model efficiency. Additionally, it aims to investigate data augmentation techniques and super-resolution generative methodsfor realistic broadband solutions. Finally, a framework based on neural operators will be developed for uncertainty quantification purpose, focusing on posterior probability distributions in high-dimensional spaces through diffusion models. This combination of generative AI and neural operators will be tested on real case scenarios, involving probabilistic seismic hazard estimation over large regions with poor historical observations.
More details about the PhD topic: https://adum.fr/as/ed/voirproposition.pl?site=adumR&matricule_prop=59433#version
This PhD topic is participating to the Université Paris-Saclay EU COFUND DeMythif.AI program: https://www.dataia.eu/en/node/1162. It is especially advertised to international students who have spent less than 12 months in France in the last 3 years. The candidates will be evaluated by a jury who will select 20 PhD to start in fall 2025. The successful candidates will be fully funded for 3 years (monthly gross salary 2 430 €, mobility indemnities 3 000€ and financial support to the host laboratories 10 000 €), have access to specific scientific and non-scientific training, and be fully part of the Université Paris-Saclay AI community.
