Florian Tramèr

Latest

• Position: Membership Inference Attacks Cannot Prove that a Model Was Trained On Your Data
• Position: Considerations for Differentially Private Learning with Large-Scale Public Pretraining