ABSTRACT
In an APS, machine learning (ML) controls automated laboratory equipment, allowing for ML-driven experiment design, execution, and analysis in a closed loop. The ML-driven closed-loop experiment cycle of APS promises to allow researchers to perform the minimum number of experiments necessary to explore the search space and identify improved technology-relevant materials. The pipeline begins with data collection from the experimentfollowed by preprocessing the data to increase its utility for the experiment. Based on closed-loop results of the APS system, the pipeline is re-engineered to improve performance. Designing the machine learning pipeline requires the selection of multiple algorithms. A common first step is to identify easy-to-use, off-the-shelf machine learning tools that can be assembled into a preliminary ML pipeline.
