Nanoengineers at the University of California, San Diego (UCSD) have developed a new machine learning (ML) method that uses low-quality data to predict material properties with greater accuracy than existing models.
UCSD's Shyue Ping Ong said the method works to "combine the large low-fidelity data and the smaller high-fidelity data to improve the models' accuracy in high value predictions."
The UCSD team examined the materials' band gaps to determine their electrical conductivity and other properties as a proof of concept; their multi-fidelity graph networks led to a 22% to 45% reduction in the mean absolute errors of experimental band-gap predictions, versus a single-fidelity model.
The approach also accurately forecast high-fidelity molecular energies, and could generate a predictive model for disordered compounds.
Said Ong, "Now we are able to do materials discovery and prediction across the entire space of both ordered and disordered materials, rather than just ordered materials."
From UC San Diego News Center
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA