Extending Defensive Distillation

TL;DR

This paper revisits defensive distillation to improve its effectiveness against adversarial examples, highlighting the importance of advanced training techniques in enhancing model robustness.

Contribution

It identifies limitations of existing defensive distillation and proposes improvements to better defend against adversarial attacks.

Findings

Enhanced defense against recent adversarial attacks

Reinforces the significance of training techniques in robustness

Addresses limitations of previous distillation methods

Abstract

Machine learning is vulnerable to adversarial examples: inputs carefully modified to force misclassification. Designing defenses against such inputs remains largely an open problem. In this work, we revisit defensive distillation---which is one of the mechanisms proposed to mitigate adversarial examples---to address its limitations. We view our results not only as an effective way of addressing some of the recently discovered attacks but also as reinforcing the importance of improved training techniques.