Transfer learning for materials informatics using crystal graph convolutional neural network

For successful applications of machine learning in materials informatics, it is necessary to overcome the inaccuracy predictions ascribed insufficient amount data. In this study, we propose a transfer using crystal graph convolutional neural network (TL-CGCNN). Herein, TL-CGCNN pretrained with big data such as formation energies for structures, and then used predicting target properties relatively small We confirm that can improve various bulk moduli, dielectric constants, quasiparticle band gaps, which are computationally demanding, construct materials. Moreover, quantitatively observe prediction models via becomes more accurate an increase size training dataset models. Finally, superior other regression methods properties, suffer from Therefore, conclude promising along compiling easy accumulate relevant properties.