Idealised Bayesian Neural Networks Cannot Have Adversarial Examples: Theoretical and Empirical Study

Open in new window