AI RESEARCH
Exponential-Family Membership Inference: From LiRA and RMIA to BaVarIA
arXiv CS.LG
•
ArXi:2603.11799v1 Announce Type: new Membership inference attacks (MIAs) are becoming standard tools for auditing the privacy of machine learning models. The leading attacks -- LiRA (Carlini, 2022) and RMIA (Zarifzadeh, 2024) -- appear to use distinct scoring strategies, while the recently proposed BASE (Lassila, 2025) was shown to be equivalent to RMIA, making it difficult for practitioners to choose among them.