Abstract:
In offshore windfarm development, geophysical and geotechnical surveys are essential for assessing subsurface conditions. However, these datasets are typically sparse and irregularly distributed, posing challenges for accurate 3D predictions. Recent advances in generative modeling offer promising solutions by learning complex spatial distributions directly from data.
We present a framework based on flow matching to learn spatial random fields from partial measurements and demonstrate its effectiveness on seismic data. This approach enables realistic interpolation while providing quantifiable uncertainty estimates. Our results demonstrate the method’s effectiveness in capturing spatial variability and improving predictive accuracy.
Short bio:
I am a 1st year PhD student at Laboratoire Navier – École Nationale des Ponts et Chaussées and TotalEnergies, working at the intersection of generative modeling and geosciences. My research focuses on stochastic interpolation techniques for subsurface data, with applications in soil characterization for offshore wind farms. Prior to my PhD, I worked as a research engineer at TotalEnergies, where I developed machine learning methods for seismic and well log data analysis and defined my PhD project. I hold a Master in Applied Mathematics from Université Grenoble Alpes and an engineering degree from Grenoble INP – Ensimag, with a focus in machine learning, statistical modeling, and optimization. My academic journey also includes a research internship at the National Institute of Informatics in Tokyo, where I explored multimodal context analysis for live video comment generation.
