Houdini: Fooling Deep Structured Prediction Models

Moustapha Cisse, Yossi Adi, Natalia Neverova +1

Generating adversarial examples is a critical step for evaluating and improving the robustness of learning machines. So far, most existing methods only work for classification and…