You are here

Model-aware Machine Learning Methods for Sciences and Engineering Problems

Hwan Goh, University of Texas-Austin

Rebecca Morrison,University of Colorado-Boulder

Tan Bui-Thanh, University of Texas-Austin

Alex Gorodetsky, University of Michigan

Scientific machine learning is an emerging field at the intersection of scientific computing and machine learning. In scientific computing, large-scale models are derived from first principles or known physical phenomena, need little data to be calibrated, and can extrapolate, but development of such models is slow. On the other hand, in machine learning, data-driven models need minimal prior knowledge or assumptions, but require big data to calibrate many parameters. Through scientific machine learning, we retain the benefits of both disciplines with interpretable, data-driven models. These “model-aware machine learning methods” can prevent overfitting, respect physical constraints or conservation laws, capture model error, and possibly even extrapolate. This minisymposium invites researchers to present their latest work in this growing and exciting new field.