Awesome

Certifiable Robustness of GCN under Structure Perturbations

Implementation of the paper:
Certifiable Robustness of Graph Convolutional Networks under Structure Perturbations

by Daniel Zügner and Stephan Günnemann.
Published at KDD'20, August 2020 (virtual event)

Additional resources

[Paper | Slides (KDD 2020)]

Run the code

The fastest way to try our code is to use the Jupyter notebook demo.ipynb.
If you would like to reproduce our numbers shown in the paper, use reproduce.ipynb. Note that in order to reproduce the results exactly you may need to install the solver CPLEX, which is proprietary. The open-source linear program solver ECOS is slower and fails on some of the instances and therefore results in slightly lower numbers than reported in the paper.

Installation

python setup.py install

If you just want to add a symbolic link to your package directory run
python setup.py develop

Contact

Please contact zuegnerd@in.tum.de in case you have any questions.

References

Datasets

In the datasets folder we provide the following datasets originally published by

Cora

McCallum, Andrew Kachites, Nigam, Kamal, Rennie, Jason, and Seymore, Kristie.
Automating the construction of internet portals with machine learning.
Information Retrieval, 3(2):127–163, 2000.

and the graph was extracted by

Bojchevski, Aleksandar, and Stephan Günnemann. "Deep gaussian embedding of
attributed graphs: Unsupervised inductive learning via ranking." ICLR 2018.

Citeseer

Sen, Prithviraj, Namata, Galileo, Bilgic, Mustafa, Getoor, Lise, Galligher, Brian, and Eliassi-Rad, Tina.
Collective classification in network data.
AI magazine, 29(3):93, 2008.

PubMed

Sen, Prithviraj, Namata, Galileo, Bilgic, Mustafa, Getoor, Lise, Galligher, Brian, and Eliassi-Rad, Tina.
Collective classification in network data.
AI magazine, 29(3):93, 2008.

Graph Convolutional Networks

Our implementation of the GCN algorithm is based on the authors' implementation, available on GitHub here.

The paper was published as

Thomas N Kipf and Max Welling. 2017.
Semi-supervised classification with graph convolutional networks. ICLR (2017).

Cite

Please cite our paper if you use the model or this code in your own work:

@inproceedings{10.1145/3394486.3403217, 
author = {Z\"{u}gner, Daniel and G\"{u}nnemann, Stephan},
title = {Certifiable Robustness of Graph Convolutional Networks under Structure Perturbations},
year = {2020},
isbn = {9781450379984},
doi = {10.1145/3394486.3403217}, 
booktitle = {Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining},
pages = {1656–1665},
location = {Virtual Event, CA, USA},
series = {KDD '20} }
}