Join or Log in
Universal Adversarial Perturbations 
Added 2 years ago by Francis Tseng
Show info
Universal Adversarial Perturbations 
Info
Smartphone demo showing the vulnerability of deep networks to a universal adversarial perturbation. This universal perturbation has been specifically designed to cause maximal harm to the classifier, while being barely perceptible. This demo is based on the paper http://arxiv.org/abs/1610.08401. The video is narrated by Eric Bezzam.
1 Connection