Universalization of Any Adversarial Attack using Very Few Test Examples

This repository is the preliminary codebase for the CODS-COMAD 2022 research track paper Universalization of any adversarial attack using very few test examples.

Overview

The paper gives a simple SVD based algorithm to obtain an universal attack from well known adversarial directions like Gradients, FGSM and DeepFool directions.

Dependencies

Codebases used in the paper as is or modified accordingly.

• [DeepFool] (https://github.com/LTS4/DeepFool)
• [CIFAR10] (https://github.com/kuangliu/pytorch-cifar)

Code documentation.

• Instructions to construct the SVD-Attack for CIFAR10 dataset
  • Load a trained model as obtained in [CIFAR10] (https://github.com/kuangliu/pytorch-cifar)
  • Collect the attack vectors (python/collect-attack-vectors.py - code snippet)
  • Obtain the top SVD vectors using the given script (python/svd-uap.py)
  • Apply the SVD-Attack with scale factor and obtain the fooling rate. (python/fooling-rate.py - code snippet)

Citation

If the code related to our work is useful for your work, kindly cite this work as given below:

@inproceedings{kamath2020universalization,
  title={Universalization of Any Adversarial Attack Using Very Few Test Examples}, 
  author={Sandesh Kamath and Amit Deshpande and K V Subrahmanyam and Vineeth N Balasubramanian},
  booktitle = {5th Joint International Conference on Data Science & Management of Data (9th ACM IKDD CODS and 27th COMAD)},
  year = {2022},
  pages = {72–80},
  howpublished={arXiv preprint arXiv:2005.08632},
  url={https://dl.acm.org/doi/abs/10.1145/3493700.3493718}
}