This is the repository of the project of AML2021: Graph injection attack & defense.
The file gia_demo.ipynb provides an notebook example of the entire process of applying a graph injection attack:
- Data preparation: Use a Refined Cora dataset as an example. (Download the data and put them in
/data.) - Model preparation: GCN model as the surrogate model.
- Graph injection attack: Apply FGSM attack on the surrogate model and transfer the results to other models (e.g. GIN, TAGCN).
See more details in the file. The trained weights of GIN and TAGCN are saved in /saved_models. The example of generated attack features and adjacency matrix are saved in /results.
- scipy==1.5.2
- numpy==1.19.1
- torch==1.8.0
If you have any question, please raise an issue or concat qinkai@tsinghua.edu.cn or zoux18@tsinghua.edu.cn.