Deep Learning with Limited Data for Battery Materials Design

The discovery and design of new materials is critical for advancing the state-of-the-art in batteries, which in turn are required for advancing a range of carbon-emission reducing technologies such as renewable energy and electric vehicles. Experimental discovery of new materials is typically slow and costly, quantum mechanics (QM) calculations have brought computational materials design within reach. However, QM calculations are often limited to relatively small sets of materials, as their computational costs are too great for large-scale screening, this is the case for calculating properties required for new battery materials. New methods in machine learning (ML) have emerged as a powerful complementary tool to QM calculations - learning rules from data calculated from QM and applying cheap, efficient models to explore large chemical spaces. However, these ML models have hitherto been restricted to instances where relatively large datasets of QM properties (tens of thousands or more instances) are available for training the ML, thus limiting their utility. In this project we will combine the expertise of our two groups (ML for materials design and computational modelling of battery materials) to tackle this important issue by using the approach of transfer learning (TL). In TL a prior model trained on a large dataset but on an apparently different problem, is used as a foundation to learn on a new, smaller dataset of direct relevance to the battery problem. TL has been transformative in many other fields and with this project we aim to bring this potential to materials design in general and battery materials in particular.