Henry Chan1,Aditya Koneru2,Subramanian Sankaranarayanan1,2,Valeria Molinero3,Jie Xu1
Argonne National Laboratory1,University of Illinois at Chicago2,The University of Utah3
Henry Chan1,Aditya Koneru2,Subramanian Sankaranarayanan1,2,Valeria Molinero3,Jie Xu1
Argonne National Laboratory1,University of Illinois at Chicago2,The University of Utah3
Advancement in high-performance computing and the availability of materials databases have empowered data-driven approaches for predicting the structure-property relationships of materials and their processing conditions. However, successful application of these methods typically rely on an abundant quantity of good-quality data, which is limited in real-world scenario and often comprised of multiple data types from different sources. Coupling physics-based simulations with data-driven approaches has the potential to address these issues and open a path towards building efficient models that can accurately predict multiple material properties across scales or digital twins that can be used in experiments. Here, we will discuss our work incorporating heterogeneous data in the development of accurate multi-scale models for molecular dynamics simulations and the use of physics-based simulations for handling the small data problem.