Self-driving Multimodal Studies at User Facilities

Abstract

Multimodal characterization is commonly required for understanding materials. User facilities possess the infrastructure to perform these measurements, albeit in serial over days to months. In this paper, we describe a unified multimodal measurement of a single sample library at distant instruments, driven by a concert of distributed agents that use analysis from each modality to inform the direction of the other in real time. Powered by the Bluesky project at the National Synchrotron 36th Conference on Neural Information Processing Systems (NeurIPS 2022).

Accurate, Uncertainty-Aware Classification of Molecular Chemical Motifs from Multimodal X-ray Absorption Spectroscopy

“…Gaussian Processes (GPs) are popular choices due to their statistical rigor (GP uncertainty is derived directly from a covariance kernel, and GPs themselves are properly generalizations of the multivariate normal distribution) and the ability to imbue them with prior physical knowledge . GPs have been widely adopted in autonomous experimentation ,,,− as they are fast (assuming small data sets and limited numbers of features), interpretable, and easy to use. Other popular methods are ensembling (whether it be an ensemble of decision trees or random forests), Monte Carlo dropout, and Bayesian neural networks .…”

“…Artificial intelligence and machine learning (AI/ML), and more broadly data-driven methods, have made a huge impact in our society over the past decade, with exciting applications in image processing, self-driving vehicles, and natural language processing using generative AI. The adaptation of AI/ML methods in scientific research has quickly spread to a wide range of domains, from fundamental research in physics, − chemistry, − biology, and materials science − to applications in drug discovery pipelines , and autonomous experimentation/self-driving laboratories. − …”