Deep Learning for Reduced Order Modeling in Parametrized PDEs

Explore the intersection of reduced order modeling (ROM) techniques and deep learning algorithms in this comprehensive talk. Discover how deep neural networks can enhance ROM efficiency for real-time simulation of large-scale nonlinear time-dependent problems. Learn about deep learning-based ROMs, POD-enhanced DL-ROMs, and hyper-reduced order models enhanced by deep neural networks. Examine strategies for learning nonlinear ROM operators and improving low-fidelity ROMs through multi-fidelity neural network regression. Gain insights into applications in structural mechanics and fluid dynamics, highlighting opportunities and challenges in using deep learning for parametrized PDEs. Presented by Andrea Manzoni, Associate Professor of Numerical Analysis at Politecnico di Milano, this talk offers valuable knowledge for those interested in computational methods, applied sciences, and engineering.

Introduction
Welcome
What we need
Outline
Example
Limitations
Similarities
Results
Comments
Extensions
Operator approximation
Output approximation
Model approximation
Conclusion
Questions