Berlin 2024 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MM: Fachverband Metall- und Materialphysik

MM 11: Data Driven Material Science: Big Data and Workflows II

MM 11.5: Vortrag

Montag, 18. März 2024, 17:00–17:15, C 243

A generic Bayesian Optimization framework for the inverse design of materials — •Zhiyuan Li, Yixuan Zhang, and Hongbin Zhang — Institute of Materials Science, TU Darmstadt, 64287 Darmstadt Germany

The traditional approach to develop materials relies on the time- and resource-costly trial-and-error experiments, as well as phenomenological theory with limited predictivity. Despite recent advances in high-throughput density functional theory calculations and statistical machine learning techniques, it is still a big challenge to efficiently explore a vast chemical space with a small number of initial samples to identify materials with optimized properties.

In this study, we propose and implement a comprehensive inverse design framework based on Bayesian optimization, integrating feature engineering, surrogate models, and acquisition functions, aiming to expedite the process of materials discovery. Focusing on the intrinsic physical properties such as formation energy, hardness, band gaps, and magnetization, it is demonstrated how such a framework can be applied to recommend optimal compositions in a vast chemical space exhibiting desired properties.

Keywords: Bayesian Optimization; active learning; inverse design