Core flow inversion with machine learning

**Description of the project

"We want to test the possibilities of modern machine learning techniques applied to the reconstruction of the flows of (i) the ZoRo experiment from acoustic measurements, and (ii) the earth’s core from magnetic field measurements. This exploratory project involves researchers from 2 teams (geodynamo and cycle) on the transversal theme "Imaging & Dynamics of the Internal Earth". Sophie Giffard Roisin will bring her expertise in machine learning, Franck Thollard will participate operationally, the geodynamo team will produce and provide the synthetic and real data sets. An M2 trainee will test different approaches and ideas. For part (i), we want to study the feasibility of finding the flow in the ZoRo experiment by training a machine from synthetics (acoustic spectra produced from simulated flows). For part (ii), we want to free ourselves from the limitations of classical inversion techniques by having the machine learn from synthetics resulting from numerical simulations, which will thus serve as a priori instead of ad-hoc regulation choices."