Skip to content

Neural networks, but malefic! 😈

License

Notifications You must be signed in to change notification settings

briland/maleficnet

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

MaleficNet: Hiding Malware into Deep Neural Networks using Spread-Spectrum Channel Coding

Conference License

This is the demo of MaleficNet, a new spread-spectrum-based technique to hide malware into Deep Neural Networks (DNN). To test the demo you need to put the payload (e.g., downloaded from TheZoo), inside the payload/ folder and follow the instructions below.

DISCLAIMER: the samples from TheZoo are live and dangerous malware. Do NOT download them and run this demo unless you are absolutely sure of what you are doing! They are to be used only for educational purposes. We highlight that since this demo is for educational purposes, we included only the injection and extraction algorithms on a single DNN architecture/task (i.e., DenseNet and CIFAR10).

Requirements

In order to try the demo, you'll need to satisfy the following requirements.

Dependencies

Install PyTorch and torchvision following these instructions. Then install the remaining dependencies:

pip install pytorch-lightning bitstring pyldpc

Dataset

Torchvision should take care by itself about CIFAR10.

Usage

To test using DenseNet and payload.bin payload, run this command. It will use a pre-trained DenseNet model from torchvision.models and will fine-tune it using CIFAR10.

python maleficnet.py --epoch 10 --model densenet --payload payload.bin --gamma 0.0009 --dataset cifar10 --num_classes 10 --dim 32

Citation

@inproceedings{hitaj2022maleficnet,
  title={MaleficNet: Hiding Malware into Deep Neural Networks using Spread-Spectrum Channel Coding},
  author={Hitaj, Dorjan and Pagnotta, Giulio and Hitaj, Briland and Mancini, Luigi V. and Perez-Cruz, Fernando},
  booktitle={Proceedings of the 27th European Symposium on Research in Computer Security (ESORICS 2022)},
  year={2022}
}

License

MaleficNet was made with ♥ and it is released under the MIT license.

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%