MIA GNN Project Starter

If you meet the version mismatch error for Lasagne library, please use following command to upgrade Lasagne library.

pip install --upgrade https://github.com/Lasagne/Lasagne/archive/master.zip

MIA Attack Process

Step1: Training Target Model using Target Dataset
Step2: Training Shadow Model using Shadow Dataset
Step3: Training Attack Model using Posteriors retrieved from Shadow Model

Here, Target Dataset and Shadow Dataset are disjoint.

Training Target and Shadow Model by GCN model

TUs: DD, PROTEINS_full, ENZYMES

# 10: run 10 times ;100:start from 100 epochs; DD : dataset DD
sh run_TUs_target_shadow_training.sh 10 100 DD

SPs: CIFAR10, MNIST

# 10: run 10 times ;100:start from 100 epochs; DD : dataset DD
sh run_SPs_target_shadow_training.sh 10 100 DD

Membership Inference Attack

# For transfer based attack, run 15 times
sh run_transfer_attach.sh 15 

Acknowledge

This project references from benchmarking-gnn and DeeperGCN

If it has any issues, please let me know.