Applied Math Seminar-Physics informed deep learning methods: a solution to bridge data gap in computational mechanics.

Date and Time: Friday, November 5, 3:15-4:15 pm

Speaker: Goswami Somdatta, Brown University

Place: zoom

Zoom link: 

https://gwu-edu.zoom.us/j/92385444375?pwd=SktyVElKc2VPOUc1RzN1bVM2WnlOZz09

 

Title: Physics informed deep learning methods: a solution to bridge data gap in computational mechanics.

 

Abstract: Artificial neural networks (ANN) can be seen as a way of storing and processing information that is loosely modeled after the biological structure of the human brain. ANNs can be ``trained'' to solve problems by recognizing patterns in the given data, sometimes exceeding the abilities of humans or that of conventional computer programs. Despite its excellent performance in several fields, it had limited applications in the field of computational mechanics until 2017, because we worked in small data regimes and simple boundary conditions. However, the gap between the physical models and the observational data was bridged by scientific machine learning which works on a known model, partially known constitutive relationship or closures, and some (or zero) high-fidelity data. In this presentation, the primary focus would be on solving coupled PDEs in computational mechanics using physics-informed deep learning. The strength of methods would be presented through their computational efficiency, accuracy, and extrapolation capability.