CNRS Research Scientist, Institut des Géosciences de l’Environnement, Grenoble (France)
Scientific Machine Learning
I investigate hybrid methods, combining mechanistic models based on differential equations and machine learning with differentiable programming. This new modelling paradigm represents the latest frontier in order to reuse previously existing scientific knowledge while harnessing the ever-increasing wealth of data and observations.
Glaciology
I investigate the use of differentiable programming, numerical methods, and inverse problems to improve our understanding of key physical processes of glaciers, ranging from glacier ice rheology variability and basal sliding to surface mass balance processes.
Climate
The strong response of glaciers to anthropogenic climate change is driving major changes in many regions of the world in terms of water availability. Our scientific models aim to better estimate and anticipate future water resources stemming from glaciers in order to improve the resilience of local populations under climate change.
ODINN
ODINN is an open-source glacier modelling project that investigates innovative hybrid methods to discover new laws governing glacier physics. By leveraging differentiable programming techniques, we are developing hybrid models that integrate differential equations describing ice flow with machine learning models to learn and parameterize specific components of these equations. This approach facilitates the discovery of parameterizations for glacier processes, helping to bridge the gap between our current mechanistic understanding of glacier physics and emerging observational data.
For more details about the model, please check our GitHub page.
Throughout the years I have been teaching and creating educational material for different workshops and short training courses. Here you will find a short list of the most relevant ones:
Machine Learning in Glaciology workshop
Theory and Jupyter notebooks for machine learning for glacier modelling
Jordi Bolibar 