Research question
Can provable robustness guarantees scale to ImageNet?
Key findings
- Gaussian noise smoothing yields a classifier with a certified L2 robustness radius - First certified defence on ImageNet scale
Certified Adversarial Robustness via Randomized Smoothing
Randomized smoothing gave the first scalable provable robustness guarantee, the foundation of certified defences.
Contribute
Spot an error, or know a record we should add? Members can propose additions and submit corrections.