This repository is the official implementation of [Evaluating Membership Inference Through Adversarial Robustness].
To install requirements:
pip install -r requirements.txt
To train victim model:
Get_Test_Model.py
📋 you could change dataset and victim model by commenting directly on the code.
To get directional distance with different T and \lambda:
robust_sphere.py
To evaluate the victim model though I_{dd}:
eval_sphere.py
To get PGD-AT trained model and evaluate its privacy though difference traditional metric based membership inference methods:
AT_train_test.py