Machine learning as a contributor to physics: Understanding Mg alloys

• Main Authors: Zongrui Pei, Junqi Yin
• Format: Article
• Language: English
• Published: Elsevier 2019-06-01
• Series: Materials & Design
• Online Access: http://www.sciencedirect.com/science/article/pii/S0264127519301960

Machine learning (ML) methods have played an increasingly important role in materials design. Take Mg alloys as an example, we show the ML methods not only supply mathematical solutions but more importantly also contribute to understand the physics in the problem. Hitherto, the role of ML methods is widely applied in high-throughput predictions, while their contribution to understand the physical mechanisms has been rarely explored. In this study, we firstly demonstrate that the Gaussian Process Classification algorithm reliably and efficiently predicts promising solutes for ductile Mg alloys, and then use these results to evaluate the correlation between two recently proposed mechanisms. Our results help clarify the controversy regarding the ductility mechanisms that can be used as the guide for materials design. Keywords: Mg alloys, Machine learning, Gaussian process classification