Adversarial Examples Are Not Bugs, They Are Features

Final Project for AM 207, Fall 2021. Review & experimentation with paper "Adversarial Examples Are Not Bugs, They Are Features" (link) by Ilyas et al. (2019).

Group Members (alphabetical): Connor Capitolo, Kevin Hare, Mark Penrod, Sivananda Rajananda

Teaching Fellow: Max Cembalest

Instructor: Weiwei Pan

Summary

The final deliverables for this report are contained in three files. First, the summary of the paper, experiments, and evaluation can be found here. Due to issues with HTML rendering on GitHub, we recommend that this be run as a local Jupyter Notebook or by accessing the Deepnote notebook here. Because of the verbosity of the code to reproduce the experiments, the notebook used to generate the results cited in the report can be found here. Finally, we have maintained all code to run these reports under source control here.

In summary:

• Final Report
• Code Submission
• Helper Python Files