Séminaire IGE

---

Physics-informed machine learning for the Geosciences: an overview

vendredi 15 décembre 2023 - 13h30

Jordi Bolibar - University Utrecht
--- 
Machine learning is becoming increasingly popular as a result of the ever growing datasets in the Geosciences. While machine learning models are very flexible and enable the exploitation of massive datasets, they have often stood as an alternative to the classically used mechanistic models based on differential equations representing long standing domain-specific knowledge. While each of the two approaches has some specific advantages, the combination of the flexibility and data assimilation capabilities of machine learning models with the interpretability and reliability of mechanistic models holds an immense potential. In this talk, I will provide an overview, illustrated with examples, of diverse strategies that meld machine learning with domain-specific knowledge, particularly in physics, for geoscientific modelling. These hybrid methodologies range from predominantly data-driven to deeply rooted in domain knowledge, offering a versatile toolkit tailored to address specific modelling needs across a spectrum of scientific problems. The dichotomy between 'machine learning' and 'physics' is dissolving, with scientific models seamlessly integrating both in hybrid approaches.

---

Equipe organisatrice : Organisation labo

Salle de conférence MCP, IGE MCP, 38400 Saint Martin d'Hères

Informations de visio :

Register here : https://univ-grenoble-alpes-fr.zoom.us/meeting/register/tJMvdu2opz8jHd2lTeg8i9W1A1EuJftY4CpJ#/registration