Machine learning, assimilation de données et systèmes dynamiques
Auteur / Autrice : | Quentin Malartic |
Direction : | Marc Bocquet, Fabio D'andrea |
Type : | Projet de thèse |
Discipline(s) : | Physique |
Date : | Inscription en doctorat le 01/10/2019 |
Etablissement(s) : | Marne-la-vallée, ENPC |
Ecole(s) doctorale(s) : | École doctorale Sciences, Ingénierie et Environnement |
Partenaire(s) de recherche : | Laboratoire : CEREA |
Mots clés
Mots clés libres
Résumé
Data assimilation aims at estimating and forecasting a geophysical system by combining in a mathematically optimal way a high-dimensional model and a large observation dataset of that system. This is especially useful for chaotic systems whose horizon of predictability is intrinsically limited. Data assimilation has been hugely successful in improving the skill of weather forecasting for the past 30 years. More recently, machine learning techniques, and especially deep learning, made impressive breakthroughs in image and speech recognition. These techniques are spreading to many fields in sciences. Data assimilation and machine learning share common goals and part of their mathematical foundations. The general objective of this thesis is to look at the potential of machine learning techniques for data assimilation and for modelling chaotic geophysical fluids such as the atmosphere.