Our group focuses on developing novel informatic approaches of machine learning (ML) and applied statistics to accelerate the search and discovery of new materials and to better understand material structures and their relationship to material properties. The use of ML in materials science and engineering (MSE) is often hampered by a lack of accuracy and low interpretability of predictive models. In particular, the poor interpretability of most ML models prevents deeper mechanistic understanding, such as finding important constituents for desired material properties. To make accurate predictions, ML methods further rely on an adequately large volume of standardized and unbiased datasets for training and testing which also necessitates the integration of research data management (RDM) tools. Our goal is to develop ML models that provide both accurate prediction and good interpretability to increase our understanding the link between structure and properties of a material. This is of great importance to make critical advances in data-driven materials research: the development of accurate and interpretable AI-based methods will enable MSE researchers to better understand the behavior and mechanisms of materials, design new and safe materials, and to explain relevant material properties that were previously poorly understood.
In particular, we are active in the following areas: