Hi there 👋

I am Victor, an applied ML researcher working in AI for science, currently at the Information and Network Dynamics group at EPFL 🇨🇭.

I develop methods for high-dimensional, spatio-temporal forecasting and inverse problems that remain robust under low signal-to-noise ratios and distribution shift, with application in biology and environmental sciences. I am particularly interested in hybrid modelling, integrating domain priors with deep learning and GPU-accelerated differentiable computing. I ship open-source libraries and models, such as

ML Libraries

• AMJAX, a JAX library exposing GPU-accelerated and differentiable linear solvers for solving large-scale sparse linear systems
• HighDimPDE.jl, a deep learning-based solver for highly dimensional, non-local, nonlinear PDEs. The repo now belongs to the SciML ecosystem.
• HybridDynamicModels.jl, a Julia library for building and training hybrid state-space models which combine mechanistic and data-driven components; it supporting probabilistic programming.

Domain-centric libraries

• 🦌JAXScape, a GPU-accelerated and differentiable library for landscape genetics and ecological connectivity modelling.

ML models

• 🪻MuScaRi, a multi-scale model for spatial biodiversity estimation. Check out this HuggingFace Space to estimate the diversity of plant species in your garden 🏡.

Check out my website to know more.