VAE latent space and its influence on Bayesian optimization performance in magnetic alloy material search

Abstract

In materials informatics, Bayesian optimization is often used to search for alloy materials with high functional properties. In this study, we efficiently performed a Bayesian optimization search by reducing the dimensionality of the material search space, which is becoming increasingly high-dimensional, using deep generative learning and nonlinear dimension reduction techniques. In this presentation, we will visualize the search space for alloy materials and discuss how dimensionality reduction techniques affect Bayesian optimization. We also show the influence of Bayesian optimization by using objective property values and related information in VAE learning.