Property-Oriented Material Design Based on a Data-Driven Machine Learning Technique

Abstract

Property-oriented material design is a persistent pursuit for material scientists. Recently, machine learning (ML) as a powerful new tool has attracted world-wide attentions in material design field. Based on statistics instead of solving physical equations, ML can predict material properties faster with less cost. Just because of its data-driven characteristics, the quantity and quality of material data become the key to the practical applications of this technique. In this perspective, problems caused by lack of data and diversity of data are discussed. Various approaches including high-throughput calculations, database construction, feedback loop algorithms and better descriptors have been exploited to address these problems. It is expected that this perspective will bring data itself to the forefront of ML-based material design.