Machine Learning for Experimental Design of Ultrafast Electron Diffraction

Ultrafast Electron Diffraction (UED) experiments can generate several gigabytes of data that must be manually processed and analyzed to extract insights into materials behavior at ultrafast timescales. The lack of real-time data analysis precents in situ tuning of experimental parameters toward desirable outcomes or away from sample damage. Here, we demonstrate that machine learning methods based on Convolutional Neural Networks (CNN) trained on synthetic UED data can perform real-time analysis of diffraction data to resolve dynamical processes in the material and identify signs of material damage. Convolutional Variational Autoencoder (VAE) models showed the ability to track structural phase transformation in a model material system through the time trajectory of UED images in the low-dimensional latent space. By mapping experimental conditions to distinct regions of the latent space, such models enable real-time steering of experimental parameters towards conditions that realize phase transformations or other desirable outcomes. These examples show the ability of machine learning (ML) to design self-correcting diffraction experiments to optimize the use of large-scale user facilities. These methods can readily be extended to other experimental characterization methods, including microscopy and spectroscopy.