An Adversarial Robustness Perspective on the Topology of Neural Networks