Machine Learning Group

Automated Sorting of High Repetition Rate Coherent Diffraction Data from XFELs

This project is harnessing novel machine learning algorithms to analyze the substantial data influx generated by X-ray Free Electron Laser (XFEL) experiments. These algorithms, centered around deep neural networks and other advanced machine learning methods, will play a pivotal role in tasks, such as clustering, filtering, and reconstructing real-time coherent diffraction data. The overarching goal is to distinguish effectively between the genuine fluctuations exhibited by a material sample and the inherent beam fluctuations intrinsic to XFELs. Through the application of these algorithms, the processed data will yield a heightened clarity in observing ultrafast dynamics and nanoscale critical phenomena within quantum materials. This is particularly significant as such phenomena have previously remained obscured due to the presence of measurement uncertainties.

Consequently, the ramifications of this project extend toward ushering in transformative capabilities within XFEL and synchrotron facilities. By harnessing the power of machine learning, the project facilitates the uncovering of presently inaccessible ultrafast fluctuations and dynamics. These algorithms present a solution to the longstanding challenges posed by measurement limitations, enabling the observation of critical phenomena within materials that have eluded comprehensive understanding until now. The key distinction between intrinsic sample behavior and beamline instabilities, achieved through machine learning techniques, opens avenues for novel scientific discoveries. Ultimately, this endeavor may expedite the acquisition of nanoscale imaging data, providing invaluable insights into the intricate behaviors exhibited by quantum materials.

As the first step in applying machine learning methods to the coherent x-ray diffraction problem, we are developing a Complex-valued Convolutional Neural Network (C-CNN)-based phase retrieval method for inverting the three-dimensional (3D) diffraction patterns from nanocrystals.