AtmoRep: A probabilistic foundation model for atmospheric dynamics
par
Amphi Recherche
Foundation models represent a new class of artificial intelligence models designed to encapsulate information from a diverse set of data sources. The most prominent examples are large language models, which have demonstrated remarkable capabilities for a wide range of applications.
In this talk I will introduce AtmoRep, a foundation model for atmospheric dynamics released by a multidisciplinary collaboration between Magdeburg University, CERN and the Jülich Supercomputing Center (https://arxiv.org/abs/2308.13280). Starting from a single pre-trained model as a backbone, AtmoRep shows skilful results in multiple zero-shot applications such as nowcasting, temporal interpolation and scenario generation, compared to the current numerical approaches.
In the last part of the talk I will discuss the foreseen future developments of AtmoRep, opening the floor for a broader discussion on the potential and the limitations of such models in fundamental science.
