Milica Todorovic, University of Turku, Turku, Finland
Data-driven materials science based on artificial intelligence (AI) algorithms has facilitated breakthroughs in materials optimization and design. Of particular interest are active learning algorithms, where datasets are collected by smart sampling on-the-fly in the search for optimal solutions. We encoded such a probabilistic algorithm into the Bayesian Optimization Structure Search (BOSS) Python tool for materials research. We utilized this versatile tool in computational studies of functional materials, like molecular surface adsorbates, thin films, solid-solid interfaces, molecular conformers, and even to optimise experimental outcomes. Agreement between optimal solutions and experimental measurements suggests that active learning is capable of good accuracy at computational costs up to 10 times smaller than other approaches. In design-of-experiment tasks, BO delivers predictive models to optimize materials, processes and devices, while conducting as few experiments as possible.