Nesterov Accelerated Gradient and Scale Invariance for Adversarial Attacks
Jiadong Lin, Chuanbiao Song, Kun He, Liwei Wang, John E. Hopcroft
TL;DR
The paper addresses the limited transferability of adversarial examples in black-box settings by introducing two complementary methods: NI-FGSM, which incorporates a Nesterov-style look-ahead into iterative perturbations, and SIM, which exploits scale-invariance by optimizing over scale copies of the input. These approaches are combined into SI-NI-FGSM and can be extended with existing gradient-based attacks (DIM, TIM, TI-DIM) to form a robust, transferable adversarial attack. Extensive ImageNet experiments show higher transferability across normally trained, adversarially trained, and other advanced defenses, with SI-NI-TI-DIM achieving particularly high success rates. The results underscore the need for stronger defenses and suggest directions for future work, including exploring Adam-like momentum and a deeper investigation into the origins of scale-invariance in deep networks.
Abstract
Deep learning models are vulnerable to adversarial examples crafted by applying human-imperceptible perturbations on benign inputs. However, under the black-box setting, most existing adversaries often have a poor transferability to attack other defense models. In this work, from the perspective of regarding the adversarial example generation as an optimization process, we propose two new methods to improve the transferability of adversarial examples, namely Nesterov Iterative Fast Gradient Sign Method (NI-FGSM) and Scale-Invariant attack Method (SIM). NI-FGSM aims to adapt Nesterov accelerated gradient into the iterative attacks so as to effectively look ahead and improve the transferability of adversarial examples. While SIM is based on our discovery on the scale-invariant property of deep learning models, for which we leverage to optimize the adversarial perturbations over the scale copies of the input images so as to avoid "overfitting" on the white-box model being attacked and generate more transferable adversarial examples. NI-FGSM and SIM can be naturally integrated to build a robust gradient-based attack to generate more transferable adversarial examples against the defense models. Empirical results on ImageNet dataset demonstrate that our attack methods exhibit higher transferability and achieve higher attack success rates than state-of-the-art gradient-based attacks.