Research

As artificial intelligence (AI) models become more advanced, extracting physical understanding from those models becomes more challenging. Explainable AI concepts overcome these challenges by increasing their transparency (the potential of a model to be understood), interpretability (how well a human can understand the interpretations from a model), and explainability (how easily a human can understand the decisions of an AI system). These concepts, as well as the accuracy and a models ability to know how reliable it is also form NIST’s Four Principles of Explainable AI. The Purcell lab focuses on developing more advanced symbolic regression methods that are inherently transparent, while using and improving postprocessing techniques to make these transparent models explainable.

High-throughput workflows and computational funnels comprised the foundation of materials discovery over the past several decades. The success of these frameworks has resulted in repositories containing several million calculations for hundreds of thousands of materials, such as the materials project, Aflow, and NOMAD. However, to efficiently explore the practically infinite materials space, new active learning frameworks that leverage AI to intelligently select the next set of candidates are necessary. The Purcell group is developing these frameworks to facilitate the discovery of new materials.

Utilizing the frameworks and methods developed in the our lab, we will discover novel energy and sustainability materials. The initial focus of our efforts will be discovering new thermoelectric materials, to make use of the over 50% of energy given off as waste heat globally. Thermoelectric materials convert a temperature gradient into an electric potential, and are described by their figure of merit, ZT. However, to date, the widespread adoption of this technology is limited by the efficiency and processability of the available materials, and discovering better thermoelectric materials is of the upmost importance. We are also open to applying these techniques to other key technological areas, such as organic electronics.