[22p-M206-1~16] 23.1 Joint Session N "Informatics"

[22p-M206-1] [Young Scientist Presentation Award Speech] Sparse-Modeling based Bayesian Optimization Algorithm for Experimental Design and Prospect of Optimization for Chemical Additives

〇Ryuji Masui¹, Unseo Lee¹, Ryo Nakayama², Taro Hitosugi^2,3 (1.HACARUS, 2.Tokyo Tech., 3.Univ. Tokyo)

Keywords:sparse modeling, experimental design, materials informatics

For the efficiency of materials development, machine learning techniques such as Bayesian optimization are applied to optimize synthesis conditions. However, in order to obtain the best materials within a realistic number of experiments, we need to restrict the search space by eliminating unnecessary parameters among a large number of experimental parameters. In this study, we propose a sparse modeling-based experimental design algorithm to achieve optimization with fewer experiments. We evaluate this method by comparing the number of experiments needed to achieve the optimal chemical additives, compared with the conventional method.