In this work, we show that adversarial examples can be generated using a generic approach that relies on the perturbation analysis of learning algorithms. Formulated as a convex program, the proposed ...