demoallan/fine-pruning-defense

Fine-Pruning: Defending Against Backdooring Attacks on Deep Neural Networks (RAID 2018)

Fine-Pruning Defense

This is the source code for the paper:

Fine-Pruning: Defending Against Backdooring Attacks on Deep Neural Networks (RAID 2018)

Kang Liu, Brendan Dolan-Gavitt and Siddharth Garg.

If you find this code useful, please cite the paper:

@InProceedings{liu2018fine-pruning,
    author="Liu, Kang
    and Dolan-Gavitt, Brendan
    and Garg, Siddharth",
    title="Fine-Pruning: Defending Against Backdooring Attacks on Deep Neural Networks",
    booktitle="Research in Attacks, Intrusions, and Defenses",
    year="2018",
    pages="273--294",
}

Training data/models for backdoor attacks on face/speech recognition can be found in the following link https://drive.google.com/drive/folders/1GBhKk2UdeU5cB7guE4oI469JuQ573XO6?usp=sharing.

Backdoor attacks on traffic sign classifiers can be found in https://github.com/Kooscii/BadNets.

Please run "conv_output_prune.py" in "face" or "speech" folder to prune the network, and run "test.py" to test the network accuracy.

Thanks to the helpful resource from https://github.com/jinze1994/DeepID1 and https://github.com/pannous/caffe-speech-recognition.