Bayesian optimization is an experimental design technique to find the object with optimal property with minimum number of experiments. It requires an accurate prediction model of the property of interest. We applied this technique to find the material with highest melting temperature from a set of single and binary compounds. We first built a prediction model by combination of systematic density functional theory (DFT) calculations and regression techniques. The inclusion of physical properties computed by the DFT calculation substantially enhanced the prediction accuracy. As a result of Bayesian optimization, the number of required measurements was dramatically smaller in comparison to random designs. This result suggests that Bayesian optimization can play a central role in cost effective material design.
Back to Machine Learning for Many-Particle Systems