Théo SANTOS - Thesis Award in Geophysics CNFGG 2026

How can we obtain an image of the Earth’s interior from seismic signals? How can we correct the blurring of astrophysical images caused by the Earth’s atmosphere?
These questions were at the heart of my thesis. They fall under the same approach: inverse problems, which involve estimating a model (of the Earth or an astrophysical object) based on observations (seismic signals or degraded images). I developed a method based on a generative neural network, an AI model capable of generating images similar to the training images. Trained to produce realistic structures, it is then used to describe the models in solving the inverse problem. I demonstrated the method’s potential and its advantages (implicit constraint toward physically plausible solutions, fast execution, low number of parameters...) on synthetic data in two contexts: 
- high-resolution imaging of the Earth’s mantle using large-scale seismic tomography images
- modeling the instrumental response of a telescope for denoising images of stellar systems.