FGSM

New attack method predicts gradients, boosting adversarial generation speed

Researchers have developed a new method for generating adversarial examples in machine learning models by predicting gradients from forward-pass hidden states. This technique bypasses the computationally expensive backw…