| Abstract Scope |
Since the 2000s, powerful AI techniques have accelerated progress in physics, materials science, chemistry, and biology, aiding material design and discovery for various applications. However, machine learning/deep learning (ML/DL) models lack in integration of causal, hypothesis-driven nature of physical sciences. There is a need to bring in explainable and interpretable methods to understand underlying cause-and-effect relationships to gain more understanding into the decision-making process. This presentation will focus on integrating causal ML models and hypothesis-driven active learning with materials representation to extract fundamental atomistic mechanisms in systems such as perovskite oxides and two-dimensional materials with focus on corresponding electronic and magnetic properties.
Acknowledgments:This research is sponsored by the Laboratory Directed Research and Development Program of Oak Ridge National Laboratory, managed by UT-Battelle, LLC, for the U.S. Department of Energy. |