Practical Defenses against Adversarial ML

Most attacks proposed in ML Research literature do not work IRL