Adversarial training enhances the robustness of deep networks at the cost of reduced natural accuracy. Moreover, networks fortified struggle to simultaneously defend against both sparse and dense ...