a materials database and ran density-functional calculations using supercomputers at the National Energy Research Scientific Computing Center (NERSC) at Berkeley Lab to identify their defects. The small data sample allowed the team to use a forest approach called gradient boosting to develop their machine-learning method to a high accuracy, Medasani said.
For this approach, researchers built additional machine-learning models successively and combined them with prior models to minimize the difference between the predictions obtained from the models and the results from density-functional calculations, he said. The team then repeated the process until they achieved a high level of accuracy in their predictions.
The end result is a tool that enables researchers to predict metallic defects faster and robustly, which will in turn accelerate materials design, according to researchers.