LiRA

[johanos1]

Feature

Desired Behavior / Functionality

We want to implement LiRA from "Carlini, Nicholas, et al. "Membership inference attacks from first principles." 2022 IEEE Symposium on Security and Privacy (SP). IEEE, 2022."

What Needs to Be Done

• Create a class LiRA that implements the attack
• Add it to attack factory

How Can It Be Tested

• Unit test on toy data
• Comparison to RMIA
• Compare to paper results