/adversarial-detection

11-785 Deep Learning Project

Primary LanguageHTMLMIT LicenseMIT

Adversarial Attack Detection

11-785 Deep Learning Project (F20)

Adversarial Detector

This implements the baseline from On Detecting Adversarial Perturbations - Jan Hendrik Metzen, Tim Genewein, Volker Fischer, Bastian Bischoff.

The code for this is available in detector/. Refer to Readme file inside for more details.

Adversarial Data Generation

In the adv_generation folder. The original code for this is from https://github.com/sanilpande/adversarial-defense. Slight modifications were made to use the FGSM based generator on the other classifier models.