Certified Patch Robustness via Smoothed Vision Transformers

Authors:
  • HSHadi Salman
    Hadi Salman
  • SJSaachi Jain
    Saachi Jain
  • EWEric Wong
    Eric Wong
  • AMAleksander Madry
    Aleksander Madry

Abstract
Certified patch defenses can guarantee robustness of an image classifier to arbitrary changes within a bounded contiguous region. But, currently, this robustness comes at a cost of degraded standard accuracies and slower inference times. We demonstrate how using vision transformers enables significantly better certified patch robustness that is also more computationally efficient and does not incur a substantial drop in standard accuracy. These improvements stem from the inherent ability of the vision transformer to gracefully handle largely masked images. 1 1 Our code is available at https://github.com/MadryLab/smoothed-vit..
Funding Information
  • NSF (CCF-1553428,CNS-1815221)
  • Defense Advanced Research Projects Agency (DARPA) (HR001120C0015)