Towards Data-Driven Predictive Multiscale Materials Modelling

Nicholas Zabaras
University of Warwick

In this presentation, we will highlight some of the many bottlenecks in predictive multiscale materials modelling and potential approaches to resolve them. They include (i) addressing the curse of stochastic dimensionality using approximate inference methods on graphs, (ii) addressing the phenomenology of constitutive equations in most deterministic multiscale materials models, (ii) quantifying epistemic uncertainty in surrogate materials models trained with limited simulation and/or experimental data, and (iv) introducing a Bayesian generative approach to stochastic coarse graining using latent variable models. Examples will be shown from diverse areas ranging from electronic structure calculations to predictive modelling of polycrystalline materials.

Back to Uncertainty Quantification for Multiscale Stochastic Systems and Applications