Project Grant 2154952

This National Science Foundation Project Grant award of $252,832 supports the development of a new machine learning architecture for chemistry and materials science applications by Dr. Matthew M. Montemore of Tulane University. The award is provided through the NSF's Mathematical and Physical Sciences program (CFDA 47.049), which aims to strengthen the nation's scientific enterprise through support of basic research in these fields.

Specifically, Dr. Montemore and his research group will create a machine learning framework that can generate reusable models across different applications involving guest-host chemical bonding, such as battery design. This represents a significant departure from existing screening approaches that require new bespoke models for each use case. The architecture leverages fundamental chemical principles like the discrete nature of elements to partially separate different guest and host elements, simplifying individual submodel training. This improved reusability and transfer learning capabilities are expected to significantly increase the efficiency of materials screening processes. The models may also be more interpretable and effective in predicting multiple quantities. Project outcomes will include publicly available code and trained models to enable broader use.