This is an implementation of paper Certified Edge Unlearning for Graph Neural Networks. We provide the code of graph edge unlearning. In addition, we present the code for reproducing the experiments.

• tqdm/numpy/pandas/sklearn/matplotlib/seaborn
• Pytorch >= 1.09
• pyg (torch_geometric)
• DeepRobust
• StellarGraph

python experiment.py -rq bound

python experiment.py -rq unlearn

python experiment.py -rq efficiency

python experiment.py -rq efficacy

python experiment.py --rq epsilon

python experiment.py --rq cgu_compare

• -g, the ID of a GPU you want to use. Default: -1 (using CPU)
• -edges, a list, indicates the numbers of edges you want to unlearn. Default: [100, 200, 400, 800, 1000].
• -targets, a list, indicates what target models you want to evaluate. Default:['gcn', 'sage', 'gin'].
• -datasets, a list, indicate what datasets you want to use. Default:['cora', 'citeseer', 'cs'].