Cheap and Robust Adaptive Reduced Order Models for Nonlinear Inversion and Design

Explore nonlinear inverse problems and PDE-constrained optimization in this 54-minute lecture from the Data-Driven Physical Simulations webinar series. Delve into the challenges of high computational costs associated with solving large linear and nonlinear systems in Newton-type methods. Discover how parametric model reduction and randomization techniques can significantly reduce these costs. Learn about innovative approaches to compute and update parametric reduced order models efficiently during optimization processes. Examine methods for monitoring the accuracy of reduced order models and their impact on objective function improvement. Gain insights into the interdisciplinary applications of these techniques, including structural design, tomography, and image reconstruction. Presented by Eric de Sturler, Professor of Mathematics at Virginia Tech, this talk offers valuable knowledge for researchers and practitioners in computational modeling, data analytics, and optimization.

DDPS | Cheap and robust adaptive reduced order models for nonlinear inversion and design