Rachid EL MONTASSIR

Postodoctoral Researcher

CERFACS

Working on climate downscaling using deep learning models (foundation models).

Ph.D. Student

CERFACS

Thesis title: Hybrid Physics-AI architecture for cloud cover nowcasting.

This study introduces a hybrid approach combining Physics and AI for cloud cover nowcasting, aiming to address limitations of traditional deep learning methods. This method, called HyPhAI, enforces physical constraints in a differentiable way within a classical neural network model,showing superior performance compared to conventional methods and achieving better detail preservation with less data. This work led to a paper publication on Geoscientific Model Development (GMD) journal and a poster presentation at the ECMWF Machine Learning Workshop 2022. Defense expected in Oct. 2024.

Research Intern

CERFACS

Designing a hybrid Physics-AI approach for cloud cover nowcasting:
- Defining the Physical model in the form of partial differential equations (PDE).
- Writing the PDE in a differentiable way using PyTorch.
- Designing the hybrid architecture by associating trainable models with the physical model.
- Training, Testing and Adaptation.
Using: Python, PyTorch.

Research Intern

INP-ENSEEIHT

Adapting the practicals of mathematics subjects (PDE, Optimization, Optimal Control and Data Analysis) to remote work.