Interpretable, unsupervised machine learning for voluminous scattering data

The state-of-the-art scattering experiments generate complex, voluminous data at a rate outpacing the speed of conventional data analysis. This data bottleneck provides the opportunity and necessity for developing fast and interpretable machine-learning approaches to reveal their scientific story. In this talk, I will discuss a novel unsupervised machine learning technique called X-ray Temperature Clustering (X-TEC) [1] that separates the scattering data into clusters of distinct physical origins. Using X-TEC to analyze the temperature evolution of X-ray data from the Advanced Photon Source and CHESS, we identified scattering signatures of charge density waves, goldstone mode fluctuations, and quasi-long-range order in disordered systems [2,3,4]. A simple graphical interface enables the users to perform the X-TEC analysis on the fly. X-TEC is a fast and versatile technique that can be readily adapted to a broad range of experiments across scientific domains. I will discuss the successful application of X-TEC to other scattering probes like ultrafast X-ray diffraction and 4D scanning tunneling electron microscopy (4D-STEM).