QianSilence / Wasserstein-Backdoor

Unofficial implementation of NeurIPS 2021 paper Backdoor Attack with Imperceptible Input and Latent Modification.

Wasserstein-Backdoor

Unofficial implementation of NeurIPS 2021 paper Backdoor Attack with Imperceptible Input and Latent Modification.

Implementations of Backdoor Attacks

Setup the environment

Solve the enviroments referring to the original repo https://github.com/khoadoan106/backdoor_attacks. Python 3.9 + torch 1.10+ should work well.

Implementations

Backdoor attack with imperceptible input and latent modification (NeurIPS 2021).

• Paper
• Stage 1: Trigger Generation - LIRA learns to generate the trigger in Stage 1. Examples:
  • MNIST

    python lira_trigger_generation.py --dataset mnist --clsmodel mnist_cnn --path experiments/ --epochs 10  --train-epoch 1 --mode all2one --target_label 0 --epochs_per_external_eval 10 --cls_test_epochs 5 --verbose 2 --batch-size 128 --alpha 0.5 --eps 0.1 --avoid_cls_reinit 2>&1 >experiments/logs/mnist_trigger_generation.log &
    

  • CIFAR10

    python lira_trigger_generation.py --dataset cifar10 --clsmodel vgg11 --path experiments/ --epochs 50 --train-epoch 1 --mode all2one --target_label 0 --epochs_per_external_eval 10 --cls_test_epochs 5 --verbose 2 --batch-size 128 --alpha 0.5 --eps 0.1 --avoid_cls_reinit 2>&1 >experiments/logs/cifar10_trigger_generation.log &	
    

• Stage 2: Backdoor Injection. After the trigger is learned, LIRA poison and fine-tune the classifier in Stage 2. Examples:
  • MNIST

     python lira_backdoor_injection.py --dataset mnist --clsmodel mnist_cnn --path experiments/ --epochs 50 --train-epoch 1 --mode all2one --target_label 0 --epochs_per_external_eval 10 --cls_test_epochs 5 --verbose 2 --batch-size 128 --alpha 0.5 --eps 0.1 --avoid_cls_reinit --test_eps 0.01 --test_alpha 0.5 --test_epochs 50 --test_lr 0.01 --schedulerC_lambda 0.1 --schedulerC_milestones 10,20,30,40 2>&1 >experiments/logs/mnist_backdoor_injection.log &	
    

  • CIFAR10

    python lira_backdoor_injection.py --dataset cifar10 --clsmodel vgg11 --path experiments/ --epochs 50 --train-epoch 1 --mode all2one --target_label 0 --epochs_per_external_eval 10 --cls_test_epochs 5 --verbose 2 --batch-size 128 --alpha 0.5 --eps 0.1 --avoid_cls_reinit --test_eps 0.01 --test_alpha 0.5 --test_epochs 500 --test_lr 0.01 --schedulerC_lambda 0.1 --schedulerC_milestones 100,200,300,400 2>&1 >experiments/logs/cifar10_backdoor_injection.log &		
    

Please cite the paper, as below, when using this repository:

@inproceedings{doan2021lira,
  title={Lira: Learnable, imperceptible and robust backdoor attacks},
  author={Doan, Khoa and Lao, Yingjie and Zhao, Weijie and Li, Ping},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={11966--11976},
  year={2021}
}
@article{doan2021backdoor,
  title={Backdoor attack with imperceptible input and latent modification},
  author={Doan, Khoa and Lao, Yingjie and Li, Ping},
  journal={Advances in Neural Information Processing Systems},
  volume={34},
  pages={18944--18957},
  year={2021}
}