Attention is not all you need:  Comparing the performance of transformers and CNNs in classifying space groups from powder diffraction data

ORAL

Abstract

There has been a great deal of work in accelerating the determination of space group from powder diffraction data. In this work, I discuss our detailed study on the performance of CNNs and Transformers. In particular, I discuss the role of extinction groups, architecture, instrument resolution, and other effects. Our performance is state-of-the-art on synthetic data. I will also discuss our performance on actual diffraction data.

*Support for Elizabeth Baggett and Edward Friedman was provided by the Center for High Resolution Neutron Scattering, a partnership between the National Institute of Standards and Technology and the National Science Foundation under Agreement No. DMR-2010792.  This work used computational resources at [resource TACC through allocation PHY250007 from the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) program, which is supported by U.S. National Science Foundation grants #2138259, #2138286, #2138307, #2137603, and #2138296.

–

Presenters

  • William Ratcliff

    • National Institute of Standards and Technology (NIST)
    • University of Maryland; National Institute of Standards and Technology (NIST)

Authors

  • William Ratcliff

    • National Institute of Standards and Technology (NIST)
    • University of Maryland; National Institute of Standards and Technology (NIST)

  • Elizabeth J Baggett

    • Boston College

  • Edward Friedmann

    • Carnegie Mellon University

  • Vanellsa Acha

    • University of California, Berkeley

  • Derrick Chan-Sew

    • University of California, Berkeley

  • Abhishek Shetty

    • University of California, Berkeley