• Construct a dataset of materials with quantum mechanical simulations using high-performance computing resources. • Develop a semi-supervised ML model to predict a "synthesizability" score for materials. • Identify most interesting materials with high synthesizability for laboratory experiments.
• Positive and Unlabeled learning is a powerful method in materials science where datasets are often small, diverse, and not labeled. • ML can accelerate the slow, traditional approach to developing new materials. • PUMML can be applied to new materials systems and entirely new problems.
Don't forget to add the tag @ncfrey in your comments.