Why the Maximum Second Derivative of Activations Matters for Adversarial Robustness | ScienceToStartup | ScienceToStartup