AI Glossary Term
What does “Robustness” mean?
Definition
A model's ability to maintain performance under noise, shifts, or adversarial inputs.
How Robustness works
Example
A road-sign classifier works on clear photos but fails when a sticker is added or the image is blurry. A more robust version keeps working under such changes.
Why it matters
Real inputs are messy. A model needs to be dependable outside the clean conditions of testing.
Common questions
What threatens robustness?
Noise, new conditions, shifts in the data, and deliberately crafted adversarial inputs.
How is it improved?
Training on varied and augmented data, adversarial training, and testing under stress.
Robustness vs accuracy?
Accuracy measures performance on typical data. Robustness measures how well it holds up when conditions change.