WORKSHOP
Digital twins for inverse problems in Earth science
Jumeaux numériques pour les problèmes inverses en science de la Terre
22 – 26 July, 2024
Organizing Committee
Comité d’organisation
Amy Braverman (NASA Jet Propulsion Laboratory)
Owhadi Houman (California Institute of Technology)
Jouni Susiluoto (NASA Jet Propulsion Laboratory)
Olivier Zahm (INRIA Grenoble)
Digital twins have emerged as central to studying complex phenomena with feedbacks. Starting in the 2010’s there has been new interest in digital twin technology for Earth System science for which complex models and copius data resources exist. Both NASA and ESA have new initiatives in this area, and so the time is right to revisit the mathematical foundations with an eye towards Earth science applications. Such applications require methods for large-scale, computationally efficient modeling and inference; areas in which significant advances have been made in the last decade via machine learning.
To advance foundational understanding and practical application to real-world problems, we propose a five-day workshop to study how a) cutting-edge inverse problems research can aid the advancement of digital twin technologies, particularly in Earth science, and b) practical problems encountered in this application motivate future research.
The workshop will focus on the following scientific themes: large-scale Bayesian methods, including data assimilation, reduced order modeling, error estimation, and dimension reduction. We are particularly interested in understanding how the intersection of numerically efficient algorithms, complex models, and massive data, can be pushed forward. Many of the current issues have to do with computational constraints that limit the utility of digital twins.
Les jumeaux numériques se sont imposés comme des éléments centraux pour l’étude de phénomènes complexes avec des rétroactions. Depuis le début des années 2010, un nouvel intérêt à émergé pour la technologie des jumeaux numériques dans le domaine des géosciences, pour lequel des modèles complexes et d’abondantes ressources de données existent. La NASA et l’ESA ont toutes deux lancé de nouvelles initiatives dans ce domaine, et il est donc opportun de revisiter les fondements mathématiques en mettant l’accent sur les applications en sciences de la Terre. De telles applications nécessitent des méthodes de modélisation et d’inférence à grande échelle et computationnellement efficaces, des domaines dans lesquels des progrès significatifs ont été réalisés au cours de la dernière décennie grâce à au machine learning.
Pour faire progresser la compréhension fondamentale et l’application pratique à l’échelle opérationnelle, nous proposons un atelier de cinq jours pour étudier comment a) la recherche de pointe sur les problèmes inverses peut contribuer à l’avancement des technologies des jumeaux numériques, en particulier dans le domaine des sciences de la Terre, et b) les problèmes pratiques rencontrés dans cette application motivent la recherche future.
L’atelier portera sur les thèmes suivants : les m´méthodes bayésiennes à grande échelle, y compris l’assimilation de données, la construction de modèles réduits, l’estimation d’erreur, la réduction de dimension. Nous nous intéresserons particulièrement à comprendre comment des algorithmes numériquement efficaces peuvent être mis en œuvre en présence de modèles complexes et de données massives.
SPEAKERS
Ricardo Baptista (CALTECH) & Olivier Zahm ((INRIA Grenoble) Gradient-based dimension reduction with guarantees
Paul Battle (CALTECH) Optimization-Based Frequentist Confidence Intervals for Functionals in Constrained Inverse Problems: Resolving the Burrus Conjecture
Théo Bourdais (CALTECH) Computational Hypergraph Discovery: A Gaussian Process framework for connecting the dots
Amy Braverman (California Institute of Technology) Spatial Uncertainty Quantification for Remote Sensing Data
Edoardo Calvello (CALTECH) Continuum Attention for Neural Operators
Nisha Chandramoorthy (Georgia Tech) A dynamical systems approach to sampling and surrogate modeling
Qiao Chen (INRIA Université Grenoble Alpes) Coupled Input-Output Dimension Reduc on: Applica on to goal-oriented Bayesian experimental design and global sensi vity analysis
Tiangang Cui (University of Sydney) Tensor-Train Methods for Sequential State and Parameter Estimation in State-Space Models
Matthieu Darcy (CALTECH) Kernel methods and PINNS for solving rough nonlinear partial differential equations
Sylvain Douté (Université Grenoble Alpes) Massive analysis of multi-angular images by inverse regression of reflectance models for the physical characterization of planetary surfaces
Nina Fischer (University of Edinburgh) History matching as a calibration method for carbon cycle models
Roger Ghanem (University of Southern California) Lessons from high-speed reactive flows
Omar Ghattas (University of Texas) A Digital Twin for Real Time : Bayesian Inference and Prediction of Tsunamis
Heikki Haario (LUT University) Keynote 7
Tapio Helin (LUT University) Bayesian optimal experimental design in inverse problems
Owhadi Houman (CALTECH) Talk
Mike Kirby (University of Utah) Physics-Informed Machine Learning As Part of The Digital Twin Toolbox
Alex Konomi (University of Cincinnati) Bayesian spanning treed for high dimensional output emulation
Otto Lamminpää (NASA Jet Propulsion Laboratory) Using Gaussian Mixture Models to Solve Inverse Problems with Full Bayesian UQ
Grigorios Lavrentiadis (CALTECH) Development of a physics-guided non-ergodic ground motion model for the Groningen, Netherlands region
Mathieu Le Provost (MIT) Preserving linear invariants in ensemble filtering methods Digital twins for inverse problems in Earth science
Matthew Li (MIT) Talk
Youssef Marzouk (MIT) Sampling and generative modeling using dynamical representations of transport
Kelli McCoy (Nasa Jet Propulsion Laboratory) Inferring Spatial Patterns of Aerosols with PACE Mission Data
Tuhin Sahai (SRI International) Talk
Eric Savin (CentraleSupélec) Kinetic models for wave inversion in anisotropic elastic media
Benjamin D Smith (NASA Jet Propulsion Laboratory) Earth System Digital Twins: Technical Challenges and Opportunities
Jouni Susiluoto (California Institute of Technology) Keynote 4
Tong Xin (National University of Singapore) Ensemble Kalman inversion for high dimensional problems
Olivier Zahm (INRIA, Université Grenoble Alpes) Gradient-based dimension reduction with guarantees
Benjamin Zanger (INRIA Université Grenoble Alpes) Sequential density estimation using measure transport
Shandian Zhe (University of Utah) Multi-fidelity and Active Surrogate Learning for Scientific Machine Learning
Qianyu Zhu (MIT) On the design of scalable, high-precision spherical-radial Fourier features