Keywords: Bayesian optimization, materials optimization, human expert, perovskites, accelerated design
TL;DR: Bayesian optimization can guide materials exploration but the varying quality of experimental samples is a potential issue - we show that human in the loop can help avoiding convergence of the search to false optima with low quality samples.
Abstract: Bayesian optimization (BO) is a popular sequential machine learning optimization strategy for black-box functions. BO has proven to be an effective approach for guiding sample-efficient exploration of materials domains and is increasingly being used in automated materials optimization set-ups. However, when exploring novel materials, sample quality may vary unexpectedly, which, in the worst case, can invalidate the optimization procedure if undetected. This limits the use of highly-automated optimization loops, especially in high-dimensional materials spaces that require more samples. Sample quality may be hard to define unequivocally for a machine but human scientists are usually good at quality assurance, at least on a cursory yet often sufficient level. In this work, we demonstrate that humans can be added into the BO loop as experts to comment on the sample quality, which results in more trustworthy BO results. We implement human-in-the-loop BO via a data fusion approach and simulate BO of experimental perovskite film stability (data from the literature). Our human-in-the-loop approach facilitates automated materials design and characterization by reducing the occurrence of invalid optimization results.
Paper Track: Papers
Submission Category: AI-Guided Design
Supplementary Material: pdf