Probably Approximately Global Robustness Certification

Peter Blohm, Patrick Indri, Thomas Gärtner +1

We propose and investigate probabilistic guarantees for the adversarial robustness of classification algorithms. While traditional formal verification approaches for robustness are…