Generating Adversarial Examples with Graph Neural Networks

This repository contains the implementation of the paper Generating Adversarial Examples with Graph Neural Networks in PyTorch. If you use this work for your research, please cite the paper:

@article{jaeckle2021generating,
  title={Generating Adversarial Examples with Graph Neural Networks},
  author={Jaeckle, Florian and Kumar, M Pawan},
  journal={Conference on Uncertainty in Artificial Intelligence},
  year={2021}
}

Repository structure

• ./scripts/ is a set of python scripts. The README in the directory provides further details of how to launch the experiments described in the paper
• ./adv_exp/ contains implementations of various adversarial attack methods including our own AdvGNN method. It further contains files to train new GNNs and has a folder with trained GNNs used in the UAI paper.
• ./tools/' contains auxiliary functions to run attacks
• ./datasets/ contains the Verification and Adversarial Attack Datasets referenced in the paper
• ./models/ contains the trained Neural Networks that we are attacking in our experiments
• ./UAI_experiments/ is the directory for result files generated while running attacks on CIFAR-10
• ./lp_solver/ contains an implementation of

Running the code

Dependencies

The code was implemented assuming to be run under python3.8. We have a dependency on:

• Pytorch to represent the Neural networks and to use as a Tensor library.

Installation

We assume the user's Python environment is based on Anaconda.

git clone --recursive https://github.com/oval-group/AdvGNN.git

cd AdvGNN

#Create a conda environment
yes | conda create -n advgnn python=3.8
conda activate advgnn

# Install pytorch to this virtualenv
# (or check updated install instructions at http://pytorch.org)
pip install torch==1.5.1+cu92 torchvision==0.6.1+cu92 -f https://download.pytorch.org/whl/torch_stable.html

# Install the code of this repository
python setup.py install

Execution

Finally, all experiments can be replicated by the different python scripts as explained further in the directory scripts

OVAL Dataset

The OVAL adversarial dataset can be found in the ./datasets/ directory.