AD-CERT uses logit-level adversarial distillation from a robust teacher combined with IBP to achieve state-of-the-art certified performance on robustness benchmarks, improving over feature-space distillation by up to 5.40 percentage points.
Generating less certain adversarial examples improves robust generalization.Transactions on Machine Learning Research, 2025
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Improving Certified Robustness via Adversarial Distillation
AD-CERT uses logit-level adversarial distillation from a robust teacher combined with IBP to achieve state-of-the-art certified performance on robustness benchmarks, improving over feature-space distillation by up to 5.40 percentage points.