[14a-A205-1~10] 23.1 Joint Session N "Informatics"

[14a-A205-7] Prediction of properties from carbon-K edge ELNES/XANES via neural network

〇Kakeru Kikumasa¹, Shin Kiyohara^2,3, Kiyou Shibata^1,3, Teruyasu Mizoguchi^1,3 (1.Grad. Sch. of Eng., the Univ. of Tokyo, 2.IIR, Tokyo Tech., 3.IIS, the Univ. of Tokyo)

Keywords:ELNES XANES, Machine learning, Prediction of properties

Although ELNES and XANES are powerful for investigating local electronic structure, they require lots of time and efforts because extracting properties needs high expertise and theoretical calculations. In this study, we tried fast prediction of molecular properties from carbon-K edge spectra using machine learning. We succeeded in predicting properties with high accuracy using feed forward neural network, whereas prediction of some properties such as internal energy is inaccurate. We improved accuracy by adding molecular composition to input data of neural network.