[22a-B201-1~11] 10.1 Emerging materials in spintronics and magnetics (including fabrication and characterization methodologies)

▼ [22a-B201-2] Materials exploration using ensemble machine learning with small dataset

〇Kenji Nawa^1,2, Katsuyuki Hagiwara¹, Yoshio Miura², Kohji Nakamura¹ (1.Mie Univ., 2.NIMS)

Keywords:machine learning, first-principles calculations, magnetism

Materials informatics combined with machine learning and density functional theory (DFT) calculations has a potential to accelerate materials design. Here, we propose a neural network (NN) approach for machine learning in which database is limited due to huge costs in the DFT calculations. We introduced ensemble machine learning algorithm into the NN and applied to typical ferromagnetic multilayers CoFe. Significant improvement of prediction accuracy, for example, for magnetic moment in CoFe was demonstrated by using the ensemble algorithm. Therefore, the present method can be utilized in materials informatics with small dataset.