 |
|||
|
|||
|
|||
Home > TERATEC FORUM >Workshops
 |
TERATEC 2025 Forum |
|
Continuum HPC / Edge / Embedded
Chaired by Thierry Collette, Director Information Science and Technology Group, Thales and Eric Monchalin, Vice-President, Head of Machine Intelligence, Eviden
Assessing seismic hazard in France using a neural meta-model
By Michael Bertin, Research Engineer, CEA DAM
In regions with low to moderate seismicity, such as mainland France, it is difficult to quantify precisely the seismic hazard, and in particular the local effects of seismic amplification. While high-fidelity, high-performance numerical simulations can now faithfully reproduce earthquakes, computation times remain prohibitive for studying the associated uncertainties.
We propose a meta-model that replaces the numerical simulator with a predictive method based on artificial intelligence. The MIFNO (Multi-Input Fourier Neural Operator) meta-model is an extension of Fourier neural operators. It is trained on a base of 30,000 high-fidelity simulations, and provides an accurate and fast alternative for studying seismic events such as the Le Teil earthquake (Ardèche, 2019) and quantifying uncertainties in the seismic hazard.
 |
BiographY: After a PhD in applied mathematics at the CEA, devoted to modeling the propagation of infrasound acoustic waves in the atmosphere, my post-doctoral research focused on numerical simulation, uncertainty analysis and machine learning in various fields of application: seismic hazard, acoustic propagation, composite materials and patent analysis. Since 2020, I have been a researcher at CEA, where my work focuses on numerical simulation and machine learning for the study of seismic propagation and the assessment of seismic hazard in France. |
 |
|
Register now and get your badge here >>>
For any other information regarding the workshops, please contact :
Jean-Pascal JEGU
Tel : +33 (0)9 70 65 02 10
jean-pascal.jegu@teratec.fr
Campus TERATEC
2, rue de la Piquetterie
91680 BRUYERES-LE-CHATEL
France