You are here

Data-driven Science with Uncertainty Quantification, Machine Learning, and Optimization

Jim Stewart, Sandia National Laboratories

Krishna Garikipati, University of Michigan

Roger Ghanem, University of Southern California

Miguel Bessa, TU Delft

Christophe Desceliers, Universite Gustave Eiffel

Alberto Figueroa, University of Michigan

Marc Mignolet, Arizona State University

Florent Pled, Universite Gustave Eiffel

Christian Soize, Universite Gustave Eiffel

Data-driven approaches are opening new avenues in computational mechanics and materials science. Uncertainty quantification and optimization under uncertainties in computational mechanics and engineering sciences can take advantage of all the recent fundamental advances in data sciences and stochastic modeling such as, non-exclusively, representation of stochastic vectorial spaces, polynomial chaos representations, Gaussian processes for regression, random matrix theory, non-Gaussian stochastic field theory, as well as computational statistics tools such as maximum likelihood and Bayesian approaches, probabilistic Neural Networks, probabilistic learning, etc. This minisymposium focuses on (1) recently developed methods for data-driven approaches, and (2) data-driven applications to fluids, structures and materials involving (but not limited to) machine learning, uncertainty quantification and/or optimization. Contributions addressing specific challenges relevant to this topic such as reduced order modeling and high-performance computing are also encouraged. Ideally, this minisymposium will reflect the generality of data-driven science and its broad applicability to the communities of computational mechanics, materials science, and more generally, engineering sciences.