MindTrojan

A Collection of Backdoor/Trojan Learning Resources and Examples with MindSpore

Quick Start

• Installation

  pip install -r requirements.txt
  

• Quick demo:

  bash quick_demo.sh
  

• Supported Attacks: BadNets, Blended

• Supported Defense: Fine-tune, FT-SAM, SAU, NPD

• Note: By Default, the models are imported from MindCV (https://github.com/mindspore-lab/mindcv). In case some methods need to modify models, a local copy of models from MindCV is also included in this repo. Change the import part in the code to switch between the local models' folder and models in MindCV.

Citation

The default settings are in line with BackdoorBench (https://github.com/SCLBD/BackdoorBench) and we refer users to BackdoorBench for more details about the settings.

If interested, you can read our recent works about backdoor learning.

@inproceedings{backdoorbench,
  title={BackdoorBench: A Comprehensive Benchmark of Backdoor Learning},
  author={Wu, Baoyuan and Chen, Hongrui and Zhang, Mingda and Zhu, Zihao and Wei, Shaokui and Yuan, Danni and Shen, Chao},
  booktitle={Thirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmarks Track},
  year={2022}
}

@inproceedings{zhu2023enhancing,
  title={Enhancing fine-tuning based backdoor defense with sharpness-aware minimization},
  author={Zhu, Mingli and Wei, Shaokui and Shen, Li and Fan, Yanbo and Wu, Baoyuan},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={4466--4477},
  year={2023}
}

@article{wei2024shared,
  title={Shared adversarial unlearning: Backdoor mitigation by unlearning shared adversarial examples},
  author={Wei, Shaokui and Zhang, Mingda and Zha, Hongyuan and Wu, Baoyuan},
  journal={Advances in Neural Information Processing Systems},
  volume={36},
  year={2024}
}

@article{zhu2024neural,
  title={Neural polarizer: A lightweight and effective backdoor defense via purifying poisoned features},
  author={Zhu, Mingli and Wei, Shaokui and Zha, Hongyuan and Wu, Baoyuan},
  journal={Advances in Neural Information Processing Systems},
  volume={36},
  year={2024}
}