Defensive distillation smooths model outputs to reduce sensitivity to adversarial perturbations.
- True
- False
Answer: True
Defensive distillation trains a student model to mimic a teacher's softened outputs, reducing gradient magnitudes and adversarial vulnerability.