"Real Attackers Don't Compute Gradients": Bridging the Gap Between Adversarial ML Research and Practice

Open in new window