Code Repository for Knowledge Enhanced Machine Learning Pipeline against Diverse Adversarial Attacks

• Complete repository with full data and pretrained models :

  https://www.dropbox.com/sh/4xxz10xxi0o5kjb/AAAhkYrCtHWD-huP3D_FCK1Ga?dl=0

• Evaluation against $\mathcal{L}_\infty$, unforeseen attacks and common corruptions : ./SettingA-linf,unforeseen,corruptions

  • Setup

    • Download ./SettingA-linf,unforeseen,corruptions/data/data from the dropbox directory, which contains all the clean data, pregenerated adversarial examples, preporcessed sensory information.

    • Download ./SettingA-linf,unforeseen,corruptions/pipeline/sensor from the dropbox directory, which contains all the sensors (submodels) for KEMLP (SettingA).

  • Reimplement tables : Please refer to ./SettingA-linf,unforeseen,corruptions/readme.md

• Evaluation against physical stop sign attack : ./SettingB-stop_sign_attack

  • Setup
    • Download ./SettingB-stop_sign_attack/data/data from the dropbox directory, which contains all the clean data, pregenerated adversarial examples, preporcessed sensory information.
    • Download ./SettingB-stop_sign_attack/pipeline/sensor from the dropbox directory, which contains all the sensors (submodels) for KEMLP (SettingB).
  • Reimplement tables : Please refer to ./SettingB-stop_sign_attack/readme.md

• Environment Dependency

  • python 3.6
  • pytorch 1.7.0
  • cv2