Boosting the Transferability of Adversarial Attacks with Reverse Adversarial Perturbation

Authors: Zeyu Qin, Yanbo Fan, Yi Liu, Li Shen, Yong Zhang, Jue Wang, Baoyuan Wu

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experimental comparisons demonstrate that RAP can significantly boost adversarial transferability. Furthermore, RAP can be naturally combined with many existing black-box attack techniques, to further boost the transferability. When attacking a real-world image recognition system, i.e., Google Cloud Vision API, we obtain 22% performance improvement of targeted attacks over the compared method.
Researcher Affiliation Collaboration 1School of Data Science, Shenzhen Research Institute of Big Data, The Chinese University of Hong Kong, Shenzhen 2Tencent AI Lab 3JD Explore Academy
Pseudocode Yes Algorithm 1 Reverse Adversarial Perturbation (RAP) Algorithm
Open Source Code Yes Our codes are available at: https://github.com/SCLBD/Transfer_attack_RAP.
Open Datasets Yes We conduct the evaluation on the Image Net-compatible dataset 1 comprised of 1,000 images. Publicly available from https://github.com/cleverhans-lab/cleverhans/tree/master/ cleverhans_v3.1.0/examples/nips17_adversarial_competition/dataset
Dataset Splits No The paper does not explicitly provide validation dataset splits in the main text.
Hardware Specification No The paper does not explicitly specify hardware used for experiments such as GPU models or CPU types in the main text.
Software Dependencies No The paper does not explicitly provide specific software dependencies with version numbers in the main text.
Experiment Setup Yes Implementation Details. For untargeted attack, we adopt the Cross Entropy (CE) loss. For targeted attack, apart from CE, we also experiment with the logit loss... The adversarial perturbation ϵ is restricted by ℓ = 16/255. The step size α is set as 2/255 and number of iteration K is set as 400 for all attacks... For RAP, we set KLS as 100 and αn as 2/255. We set ϵn as 12/255 for I and TI in untargeted attack and 16/255 for other attacks in all other settings.