Advancing Discovery with Artificial Intelligence and Machine Learning at NSLS-II

Abstract

With the National Synchrotron Light Source II (NSLS-II) coming online in 2015 as the brightest source in the world, the imminent up-grades at the Advanced Photon Source, Advanced Light Source, and Linear Coherent Light Source, and advances in detector technology, the data generation rates at the U.S. Department of Energy (DOE) Basic Energy Sciences’ X-ray light sources are skyrocketing. At NSLS-II, over 1 petabyte of raw data was produced last year, and that rate is expected to increase as the facility matures [1]. Despite such huge data generation rates, approaches to both experimental control and data analysis have not kept pace. Consequently, data collected in seconds to minutes may take weeks to months of analysis to understand. Due to such limita-tions, knowledge extraction is often divorced from the measurement process. The lack of real-time feedback forces users into flying blind at the beamline, leading to missed opportunities, mistakes, and inefficient use of beamtime as a resource—as all beamlines are oversubscribed. This is a challenge facing nearly all users of light sources. One promising path forward to solve this challenge—both during data collection and post-experiment analysis—is the use of artificial intelligence (AI) and machine learning (ML) methods [1, 2]. In this contribution, we review recent developments employing AI/ML methods at the NSLS-II, tackling the