On the Effectiveness of Defensive Distillation

Key Points

• This research examines the effectiveness of defensive distillation in improving model defenses against adversarial attacks.
• Conducted experiments to evaluate defensive distillation's impact on adversarial samples.
• Tested samples crafted using both the fast gradient sign method and the Jacobian-based iterative attack.
• Defensive distillation significantly reduced the efficacy of adversarial samples from both attack methods, showing strong protective performance.