Most Convolutional Networks Suffer from Small Adversarial Perturbations

Amit Daniely, Idan Mehalel

The existence of adversarial examples is relatively understood for random fully connected neural networks, but much less so for convolutional neural networks (CNNs). The recent wor…