Adversarial Ranking Attack and Defense (ECCV2020)

Adversarial Ranking Attack and Defense, ECCV, 2020.

Adversarial Ranking Attack and Defense (ECCV2020)

Materials for ECCV-2020 Paper #2274.

NEWS & Updates

  1. Substantial progress (at least 60% and at most 540% robustness improvement) based on this conference paper is available as a preprint. NOTE: If you want to do some further research based on this ECCV 2020 paper, please have a look at this preprint paper to better understand the limitations of the ECCV 2020 work.

Demonstration

Contributions

Definition of Adversarial ranking attack: adversarial ranking attack aims raise or lower the ranks of some chosen candidates C={c₁,c₂, … ,cₘ} with respect to a specific query set Q={q₁,q₂, … ,qw}. This can be achieved by either Candidate Attack (CA) or Query Attack (QA).

  1. The adversarial ranking attack is defined and implemented, which can intentionally change the ranking results by perturbing the candidates or queries.

  2. An adversarial ranking defense method is proposed to improve the ranking model robustness, and mitigate all the proposed attacks simultaneously.

License and Bibtex

The paper (PDF file) is distributed under the CC BY-SA-NC 4.0 License.

The code is published under the Apache-2.0 License.

Bibtex for the ECCV version:

@InProceedings{advrank,
  title={Adversarial Ranking Attack and Defense},
  author={Zhou, Mo and Niu, Zhenxing and Wang, Le and Zhang, Qilin and Hua, Gang},
  booktitle={ECCV},
  year={2020},
  pages={781--799},
  isbn={978-3-030-58568-6}
}

Bibtex for the ArXiv preprint version:

@article{zhou2020advrank,
  title={Adversarial Ranking Attack and Defense},
  author={Zhou, Mo and Niu, Zhenxing and Wang, Le and Zhang, Qilin and Hua, Gang},
  journal={arXiv preprint arXiv:2002.11293},
  year={2020}
}

References

  1. A. Madry et.al. Towards Deep Learning Models Resistant to Adversarial Attacks