Colloques du Collège de France - Collège de France

Grand événement - À la recherche d'un Avenir Commun DurableL'IA et les mathématiques pour la météorologie et la climatologieAI and math for meteorology and climatologyCollège de FranceAnnée 2024-20255 mai 2025Grand événement - AI and math for meteorology and climatology - Marc Bocquet : Artificial intelligence for geophysical data assimilationMarc BocquetCEREA, ENPC, EdF R&D, Institut Polytechnique de ParisRésuméData assimilation is the set of key mathematical methods used to optimally combine observations and numerical model outputs. Data assimilation (DA) is critical to adjust the initial condition of meteorological forecasts, to estimate model parameters, and produce accurate re-analysis datasets. It has been at the heart of all operational weather forecasts for the past 50 years. Very recently, artificial intelligence (AI) and in particular deep learning, has begun being used as a tool to improve classical DA, to be combined with DA algorithmic schemes, or even to offer a substitute for DA.I will give an overview of the recent achievements and promising routes offered by AI into DA.For instance, ML can be leveraged in the regularisation of ensemble-based DA, in the solvers of variational DA methods, for generating or augmenting ensembles in DA, for building surrogates of the tangent linear and adjoint meteorological models to be used within DA, to learn a model error correction within a weak-constraint 4D-Var framework, or, ultimately, as a replacement for the DA analysis. I will also present an example where AI unveils new DA methods that were overlooked so far by the research community.Marc BocquetMarc Bocquet holds a Ph.D. from École Polytechnique and has an Habilitation delivered by Sorbonne University. He was a postdoctoral fellow at the University of Warwick and then at the University of Oxford. He is currently deputy director of CEREA, a laboratory of École nationale des ponts et chaussées and EDF R&D, and a professor at École nationale des ponts et chaussées, Institut Polytechnique de Paris. He works on data assimilation, inverse problems and statistical learning applied to the geosciences. He develops mathematical methods to better estimate the state of the atmosphere, of the ocean and the climate, as well as their constituents, using massive observations and complex models. He has published 115 papers and two books. He is associate editor of several peer-reviewed journals in the geosciences, and a Distinguished Research Fellow of the world's most renowned weather forecasting centre, the European Centre for Medium-Range Weather Forecasts (ECMWF).

Grand événement - À la recherche d'un Avenir Commun DurableL'IA et les mathématiques pour la météorologie et la climatologieAI and math for meteorology and climatologyCollège de FranceAnnée 2024-20255 mai 2025Grand événement - AI and math for meteorology and climatology - Remi Lam : Learning global weather forecasting from dataRemi LamMassachusetts Institute of Technology, Staff Research Scientist, Google DeepMindRésuméThis presentation will cover some of the recent advances in weather forecasting, learning directly from data using machine learning techniques.It will discuss some of the limitations and pitfalls of training ML models for scientific applications, and will highlight new research opportunities.Rémi LamRémi Lam is a Staff Research Scientist at Google DeepMind working on making weather forecasting faster and more accurate.His research leverages machine learning techniques such as adversarial neural networks, graph neural networks and diffusion models to design tools for precipitation nowcasting (DGMR) and global medium range weather prediction (GraphCast, GenCast).

Grand événement - À la recherche d'un Avenir Commun DurableL'IA et les mathématiques pour la météorologie et la climatologieAI and math for meteorology and climatologyCollège de FranceAnnée 2024-20255 mai 2025Grand événement - AI and math for meteorology and climatology - Claire Monteleoni : Confronting climate change with generative and self-supervised machine learningClaire MonteleoniResearch Director, INRIA Paris & Professor, University of Colorado BoulderRésuméRésuméThe stunning recent advances in AI content generation rely on cutting-edge, generative deep learning algorithms and architectures trained on massive amounts of text, image, and video data. With different training data, these algorithms and architectures can also be used to confront climate change. As opposed to text and video, the relevant training data includes weather and climate data from observations, reanalyses, and even physical simulations. As in many massive data applications, creating "labeled data" for supervised machine learning is often costly, time-consuming, or even impossible. Fortuitously, in very large-scale data domains, "self-supervised" machine learning methods are now actually outperforming supervised learning methods. In this lecture, I will survey our lab's work developing generative and self-supervised machine learning approaches for applications addressing climate change, including downscaling and temporal interpolation of spatiotemporal data and generating probabilistic weather predictions.Claire MonteleoniClaire Monteleoni is a Choose France Chair in AI and a Research Director at INRIA Paris, a Professor in the Department of Computer Science at the University of Colorado Boulder (on leave), and the founding Editor in Chief of Environmental Data Science, a Cambridge University Press journal launched in December 2020. Her research on machine learning for the study of climate change helped launch the interdisciplinary field of Climate Informatics. She co-founded the International Conference on Climate Informatics, which will hold its 14th annual event in 2025. She gave an invited tutorial: Climate Change: Challenges for Machine Learning, at NeurIPS 2014. She currently serves on the U.S. National Science Foundation's Advisory Committee for Environmental Research and Education, and as Tutorials co-Chair for the International Conference on Machine Learning (ICML) 2024 and 2025.

Grand événement - À la recherche d'un Avenir Commun DurableL'IA et les mathématiques pour la météorologie et la climatologieAI and math for meteorology and climatologyCollège de FranceAnnée 2024-20255 mai 2025Grand événement - AI and math for meteorology and climatology - Thomas Dubos : Hamiltonian insights and the challenge of unresolved processes in geophysical modelsThomas DubosProfesseur, École PolytechniqueRésuméMathematical and numerical models of the atmosphere and ocean rely on various assumptions, approximations, and simplifications. Over the past decade, significant progress has been made in elucidating their structure and interconnections, particularly for the resolved and reversible fluid components. This advancement has largely been driven by Hamiltonian approaches, encompassing both Hamilton's principle of least action and the associated symplectic structure. Moreover, these insights have influenced the development of numerical methods in production-ready models.This progress shifts attention toward the unresolved and irreversible processes, where mathematical and theoretical foundations remain scarce—and may continue to be so. I will challenge the notion that partial differential equations are all that is needed, and highlight areas where theoretical progress seems possible. Hopefully, this perspective can shed light on the respective roles of physics-based and data-driven components in comprehensive models.Thomas DubosThomas Dubos studied mathematics and physics at the École Normale Supérieure (Paris) and obtained his Ph.D. in 2002 at the Laboratoire de Météorologie Dynamique (LMD), focusing on transport through two-dimensional turbulence. As Assistant Professor and then Professor at LMD/École Polytechnique, his research focused on geophysical turbulence and hydrodynamics. More recently, he has used Hamiltonian methods to uncover the structure and connections of existing geophysical models, to derive new models, and to develop numerical methods with desirable physical properties. This fundamental work has contributed to the development of the LMDZ, the atmospheric general circulation model developed at LMD, which is part of the Earth System Model at the Institut Pierre-Simon Laplace (IPSL).

Grand événement - À la recherche d'un Avenir Commun DurableL'IA et les mathématiques pour la météorologie et la climatologieAI and math for meteorology and climatologyCollège de FranceAnnée 2024-20255 mai 2025AI and math for meteorology and climatology - Michael Brenner : The neural GCM, and other remarksMichael BrennerHarvard University, Research Scientist, Google ResearchRésuméI will discuss the Neural GCM, which we built by building a dynamical core in JAX and then training the parameterization on ERA5 on 5-day forecasts. The quality of the forecasts up to 1 year portends a potential revolution in improving model parameterizations of physical systems described by nonlinear partial differential equations, of which weather and climate models are only one example. I will discuss some other problems in this spirit we are working on and try to draw conclusions.Michael BrennerMichael Brenner is the Michael F. Cronin Professor of Applied Mathematics and Professor of Physics at Harvard University, and a Research Scientist at Google Research. He is broadly interested in finding new ways of applying mathematics to the sciences.

Grand événement - À la recherche d'un Avenir Commun DurableL'IA et les mathématiques pour la météorologie et la climatologieAI and math for meteorology and climatologyCollège de FranceAnnée 2024-20255 mai 2025AI AND MATH FOR METEOROLOY AND CLIMATOLOGYStéphane MallatProfesseur du Collège de FranceL'événement est en anglais.L'initiative Avenir Commun Durable bénéficie du soutien de la Fondation du Collège de France, de ses grands mécènes la Fondation Covéa et TotalEnergies et de ses mécènes FORVIA et Saint-Gobain.Recent advances in artificial intelligence (AI) have produced unexpected and impressive results for weather forecasting, despite the complexity of these multi-scale phenomena. AI is also playing an increasingly important role in climatology. These results raise profound questions about modelling. On the one hand, we know the physics equations that have so far been used in large-scale numerical models. On the other hand, many physical parameters are unknown, for example at interfaces, which motivates a learning approach based on past data. We can also learn the evolution equations indirectly, eliminating the need for physical modelling. The approaches developed in AI in recent years oscillate between these two strategies.This conference will present the state of the art at the interface between mathematics, physics and statistical learning using deep neural networks. It will highlight the advantages and disadvantages of the different modelling strategies, as well as the a priori incorporated in the form of equations or algorithmic architectures, in relation with physics. An objective is to encourage a dialogue between mathematicians, AI researchers, meteorologists and climatologists.

Denis DubouleCollège de FranceÉvolution du développement et des génomesAnnée 2024-2025Enhancers Sequences and Gene RegulationColloque - Alex Stark : Decoding Transcriptional RegulationAlex StarkIMP, Vienna BioCenter, Vienna, Austria

Denis DubouleCollège de FranceÉvolution du développement et des génomesAnnée 2024-2025Enhancers Sequences and Gene RegulationColloque - Justin Crooker : Exploring the Evolutionary Limits of Transcriptional EnhancersJustin CrookerEMBL, Heidelberg, Germany

Denis DubouleCollège de FranceÉvolution du développement et des génomesAnnée 2024-2025Enhancers Sequences and Gene RegulationColloque - Ana Pombo : Variations in 3D Genome Structure Between Cell Types and in Stimulus ResponsesAna PomboMax-Delbrück-Centrum for Molecular Medicine, Berlin, Germany

Denis DubouleCollège de FranceÉvolution du développement et des génomesAnnée 2024-2025Enhancers Sequences and Gene RegulationColloque - Alexandre Mayran : Cooperative Assignment of Enhancer Activity Underlies Pattern Formation and Cell Fate SpecificationAlexandre MayranÉcole polytechnique fédérale de Lausanne (EPFL) Suisse