c0f9419caa85d7062c7e6d621a335726-Supplemental-Conference.pdf

Deep neural networks (DNNs) have been successfully applied in many safety-critical tasks, such asautonomous driving,facerecognition andverification,etc. Forreal-worldapplications, theDNN model aswell as the training dataset, are often hidden from users. Here, we analyze the computational cost of our method. The adversarial example generation process is conducted based on the offline surrogate models. Toachievethis goal, the authors multiply 16 skip connection by the random scalarr sampled from a uniform distribution.