(WORK IN PROGRESS)

This is the github repository for the ICLR 2022 paper "Attacking deep networks with surrogate-based adversarial black-box methods is easy", by Nicholas A. Lord, Romain Mueller, and Luca Bertinetto.

Please see environment.yml

Consult the following sets of instructions to reproduce the results of the paper:

• Main results: untargeted black-box attacks (Fig.2 in the paper) (this repo).
• Targeted black-box attacks (Fig.5 in the paper) (this repo).
• On the importance of input-specific priors (Fig.4 in the paper) (SimBA-PCA repo TODO).

@inproceedings{lord2022attacking,
    title={Attacking deep networks with surrogate-based adversarial black-box methods is easy},
    author={Nicholas A. Lord and Romain Mueller and Luca Bertinetto},
    booktitle={International Conference on Learning Representations},
    year={2022},
    url={https://openreview.net/forum?id=Zf4ZdI4OQPV}
}

This repository is based on ODS original implementation (https://github.com/ermongroup/ODS). We thank the authors for making their code available.