[16p-K505-1~17] 23.1 Joint Session N "Informatics"

1:45 PM - 2:00 PM JST(4:45 AM - 5:00 AM UTC)

[16p-K505-4] Improving the efficiency of searching for highly magnetic alloy materials using TargetLearning VAE and Bayesian optimization

〇Naoki Yoshida¹, Yuma Iwasaki², Yasuhiko Igarashi¹ (1.Univ of Tsukuba, 2.NIMS)

Keywords:

VAE,Bayesian Optimization

In research into multi-component alloys, the combinatorial explosion problem has made conventional manual material development difficult. Therefore, autonomous material search that combines first-principles calculations and machine learning is important. In particular, as the input space increases, the combination of dimensionality reduction using Autoencoder or Variational Auto-Encoder (VAE) and Bayesian optimization has attracted attention. In this study, we focus on an efficient material search method that combines a dimensionality reduction method and Bayesian optimization. By utilizing the correlation between the physical properties of magnetic alloys, we embed non-target property information using Target Learning VAE (TL-VAE) and design a latent space that improves the efficiency of Bayesian optimization. We also use semi-supervised VAE to verify the case where there is a small amount of non-target property information.