Awesome

CertifyGNN

Source code of "Certified Robustness of Graph Neural Networks against Adversarial Structural Perturbation", accepted by ACM SIGKDD'21

1. There are two different ways to train the GNN model: I) Train on the raw clean graph (train_edge); II) Train on the perturbed graph, where the graph is dynamically changed by randomly perturbing the connection status between the edge (train_edge_noise).

2. The detailed steps are as follows: I) Train the GNN model (either train_edge.py or train_edge_noise.py); II) Divide the binary graph structure space by running count.py or/and obtain the base probability (P_base) by running threshold.py ; III) Obtain the certified accuracy by running calculate_certify_K.py if not running threshold.py in II), or by running calculate_certify_K_efficient.py where I already stored the P_base.

3. Note that the certified accuracy could be slightly changed due to 1) randomness during the training and 2) random perturbations

4. CertifyGraphClassification.zip contains the source code of certifying robustness of graph classification (i.e, GIN).