/adversarial-attacks-project

Project repo for UWaterloo graduate course - ECE653. It involves comparing robustness, implementing new GAN based attack and extending CleverHans library with DeepFool attack.

Primary LanguageJupyter Notebook

Implementation of Adversarial Attacking Tools

The project is a part of UW ECE653, Software Testing, Quality Assurance and Maintenance. It involves:

  • Attacking state-of-the-art imagenet models with FDSM, PGD, and DeepFool methods
  • Using modern attacking tools to generate attacks on these models
  • Extending a tool with new attacking method

Attacks Performed

  • Fast Gradient Sign Method
  • Projected Gradient Descent
  • DeepFool

Tool Used

New Attack Methods Implemented

  • DeepFool Attack for CleverHans. Can be found here