Mengwu Guo: Options of Bayesian Methods for Data-Driven Model Reduction

In this talk, two Bayesian methods of data-driven model order reduction will be presented, i.e., a black-box approach based on Gaussian process surrogate modeling and a gray-box approach that recovers the reduced-order operators through Bayesian inference, both featuring a non-intrusive nature. The presented Bayesian methods enable the quantification of uncertainties introduced by data-driven modeling and provide new perspectives for non-intrusive model reduction.

Furthermore, we will briefly discuss the potential Bayesian tools that can be used for the uncertainty quantification of deep learning and deep-learning-based reduced order modeling.