[18p-B01-1~9] New material science and its practical application created by informatics

[18p-B01-8] Machine learning of experimental thermoelectric properties on literatures

〇Yukari Katsura^1,2,3, Masaya Kumagai^3,4, Riku Sato^1,2, Yuki Ando^3,2, Sakiko Gunji², Yoji Imai^3,2, Mitsunori Kaneshige⁵, Takushi Kodani^1,2, Kaoru Kimura¹, Koji Tsuda^1,3,2 (1.Univ. of Tokyo, 2.NIMS, 3.RIKEN, 4.Sakura Internet, 5.X-Ability)

Keywords:materials informatics, thermoelectric materials, machine learning

We developed an original web system named Starrydata2 (https://www.starrydata2.org/), to collect large-scale experimental data from plot images on published papers efficiently. By using this system, we collected experimental thermoelectric properties from over 18,000 samples in 2.5 years. By machine-learning of this dataset with a neural network, we succeeded to predict temperature dependences of thermoelectric properties only from chemical compositions.