This course begins by providing an overview of white-box and black-box adversarial attacks on machine learning systems. It will then guide you through using the Fast Gradient Signed Method (FGSM) white-box attack on a keras machine learning model. Next, we will cover black-box attacks. You will be guided on using a machine learning as a service system called Clarif.AI and then performing a black-box adversarial attack to trick this service into labeling a benign image as dangerous. Finally, to solidify learning, the student is given an assignment on tricking a MNIST keras classifier via a white-box adversarial attack.
