/linear-region-attack

A powerful white-box adversarial attack that exploits knowledge about the geometry of neural networks to find minimal adversarial perturbations without doing gradient descent

Primary LanguagePython

Watchers