MEMS Graduate Student Seminar: A Primer on Reduced-order Modeling
Friday, March 6, 2020 - 12:00pm to 1:00pm
Hudson Hall 224
Reduced-order models (ROMs) are a way of simulating a potentially very complicated system with very little computational cost. They are often derived from expensive experimental or computational data and serve as valuable utilities in pattern recognition and parametric perturbation studies. Mathematically they can be derived through variable separation methods and established linear algebra techniques including machine learning. This talk will outline what exactly reduced-order modeling is in the context of a turbulent fluid simulation; the cost to simulate the system will be shown to drop from an order of weeks to an order of seconds with potentially minimal losses in accuracy. The differences between reduced-order modeling and physics simplification will be discussed. It will be emphasized how ROMs can be employed in myriad research avenues ranging from protein folding to stock prediction to fluid dynamics. The known limitations of reduced-order models will also be explored and several active research avenues will be introduced.