Multi-fidelity Linear Regression for Scientific Machine Learning from Scarce Data

Explore cutting-edge research on multi-fidelity linear regression for scientific machine learning in data-scarce environments during this one-hour talk by Elizabeth Qian from Georgia Tech. Delve into innovative approaches for developing surrogate models for complex engineering systems when traditional high-fidelity simulations are costly. Learn about a novel multifidelity training method that leverages data of varying fidelities and costs to improve model accuracy and robustness. Discover how multifidelity control variate estimators are used to enhance linear regression models, and gain insights into theoretical analyses guaranteeing improved performance with limited training budgets. Examine numerical results demonstrating significant error reduction compared to standard training approaches when high-fidelity data is scarce. Gain valuable knowledge from Dr. Qian's expertise in model reduction, scientific machine learning, and multifidelity methods, applicable to engineering design and decision-making for complex systems.

DDPS | “Multi-fidelity linear regression for scientific machine learning from scarce data”