Robustness
A model's ability to perform reliably on noisy, unexpected, or adversarial input.
Robustness is a model’s capacity to maintain reliable, consistent behavior when inputs deviate from the clean training distribution — noisy data, out-of-distribution examples, and adversarial attacks such as prompt injection. A robust model degrades gracefully rather than producing wildly incorrect or unsafe outputs when conditions shift. On the AIF-C01 exam, robustness is often contrasted with accuracy: a model can score high on benchmarks yet collapse on slightly perturbed inputs, so high accuracy does not guarantee robustness. Amazon SageMaker Model Monitor detects data drift and quality issues in production, while SageMaker Clarify surfaces bias and feature-attribution drift after deployment.
PlayPrepHQ study notes are written and reviewed against primary exam sources. How we create & review content →