Funding, software & past work
Projects
The research project funding my PhD, open-source software, and earlier explorations.
Funding project
MIGRATE
Multidisciplinary and InteGRated Approach for geoThermal Exploration
Geothermal energy is a renewable and stable alternative to fossil fuels, but its development is held back by poor knowledge of the subsurface, which makes exploration geologically and economically risky. MIGRATE brings together seismology (ambient-noise surface-wave tomography, dispersion-curve analysis) and machine learning to build automated, reproducible and high-resolution tools for exploring the Earth's upper crust from dense nodal networks.
My role
- Simulation-based inference under model misspecification — amortised inversion of surface-wave dispersion curves into crustal velocity models, robust to mismatch between the simulator and real data.
- Segmentation for dispersion-curve picking — deep segmentation models that help seismologists pick dispersion curves automatically and reproducibly.
- Outreach — co-organised the workshop Computational Earth: Integrating geology, seismic tomography and machine learning (Elba Island, Oct 2025), where I gave a talk on the challenges of nodal ambient-noise tomography and automating curve picking as a segmentation problem.
Software
PyBADS
A Python implementation of Bayesian Adaptive Direct Search, for fast and robust black-box optimization of rough, noisy and mildly expensive objectives — typical of model fitting. Developed during my MSc thesis with Luigi Acerbi at the University of Helsinki, including a study of convergence to stationary points of non-smooth stochastic functions.
Research code
VGB-DM
Official implementation of Variational Grey-Box Dynamics Matching (AISTATS 2026).
Earlier work