Adversarial Robustness

The ability of a model to maintain stable predictions when facing maliciously constructed input perturbations.