Generating Minimal Adversarial Perturbations with Integrated Adaptive Gradients

Abstract

Deep neural networks are easily fooled high confidence predictions for adversarial samples

Tasks

Reproductions