AI & ML interests

Inverse problem, Normalizing Flow

Recent Activity

schattinย  updated a collection 15 days ago
Surfdisp96 Datasets
schattinย  updated a dataset about 1 month ago
MIGRATE/Surfdisp96-Roccastrada-10k-2025
schattinย  updated a dataset about 1 month ago
MIGRATE/Surfdisp96-Roccastrada-10k-2025
View all activity

๐ŸŒ‹ MIGRATE Project โ€” Multidisciplinary and InteGRated Approach for geoThermal Exploration

MIGRATE (Multidisciplinary and InteGRated Approach for geoThermal Exploration) is a scientific project that bridges seismology and machine learning to develop a new generation of automated, reproducible and high-resolution exploration tools for the Earth's upper crust.

๐ŸŽฏ Motivation

Reducing the acceleration of climate change is one of the great challenges of our time. As part of the global transition toward sustainable energy, geothermal energy offers a renewable and stable alternative to fossil fuels.

However, its development is hindered by a lack of reliable subsurface knowledge, which creates high geological and economic risks. MIGRATE aims to address this gap by creating innovative methods that reduce uncertainty in passive seismic exploration, using dense nodal networks and state-of-the-art data-driven models.

๐Ÿ”ฌ Scientific Approach

MIGRATE integrates three complementary disciplines:

  • Seismology โ€” ambient noise surface wave tomography, dispersion curve analysis
  • Machine Learning โ€” generative modeling, contrastive learning, neural surrogate inversion, and digital twins

These domains are tightly coupled to:

  • Automate the inversion of surface wave dispersion curves
  • Learn expressive representations of crustal velocity models

๐Ÿง  AI for Earth Models

We develop machine learning methods to capture the physical structure of the subsurface, with:

  • ๐ŸŒช Normalizing Flows for probabilistic inversion and generative modeling
  • ๐ŸŽฏ Contrastive encoders to structure seismic representations
  • ๐Ÿงฉ Latent representations to compress complex velocity models
  • ๐Ÿ” Self-supervised learning for unsupervised geophysical understanding

These tools are released as open-source datasets and pretrained models on Hugging Face.

๐Ÿงพ Citation

If you use this project, please cite:

@misc{migrate2025,
  title={MIGRATE: A Multidisciplinary and Integrated Approach for Geothermal Exploration},
  author={SSTE and DMML-GE},
  year={2025},
  howpublished={\url{https://huggingface.co/MIGRATE}}
}