Augmenting Model Robustness with Transformation-Invariant Attacks

TL;DR

This paper investigates how training neural networks against transformation-invariant adversarial attacks can enhance their robustness, showing modest improvements across multiple datasets and highlighting the importance of invariance in defense strategies.

Contribution

It introduces the concept of transformation-invariant adversarial training and demonstrates its effectiveness in improving model robustness against such attacks.

Findings

Transformation-invariant attacks improve robustness by up to 3.7%.

Transformations of attacks alone do not affect robustness.

Model robustness gains are consistent across MNIST, CIFAR-10, and ImageNet.

Abstract

The vulnerability of neural networks under adversarial attacks has raised serious concerns and motivated extensive research. It has been shown that both neural networks and adversarial attacks against them can be sensitive to input transformations such as linear translation and rotation, and that human vision, which is robust against adversarial attacks, is invariant to natural input transformations. Based on these, this paper tests the hypothesis that model robustness can be further improved when it is adversarially trained against transformed attacks and transformation-invariant attacks. Experiments on MNIST, CIFAR-10, and restricted ImageNet show that while transformations of attacks alone do not affect robustness, transformation-invariant attacks can improve model robustness by 2.5\% on MNIST, 3.7\% on CIFAR-10, and 1.1\% on restricted ImageNet. We discuss the intuition behind this phenomenon.