Keywords: Bayesian Optimization, Self-Supervised Learning, Deep Representation Learning, Materials Science, Materials Discovery, Bandit Algorithm
TL;DR: Discrete assumption of Bayesian Optimization succeeded in outperforming the materials discovery efficiency of human experts, whereas the conventional continuous one did not.
Abstract: A material exploration model based on high-dimensional discrete Bayesian optimization is introduced. Features were extracted from a large-scale database of ab-initio calculations by self-supervised representation learning. Material exploration was carried out based on 100 prior target values from 6,218 candidate materials. As a baseline, ten human experts of materials science were selected and evaluated their exploration efficiency. Under the same conditions, the proposed discrete algorithm was 1.93 times as efficient as human experts on average, while the conventional continuous algorithm could not outperform them.
Track: Original Research Track