ChangeOrange的博客

1. IntroductionIn recommendation systems, two types of users’ private information are leaked in traditional MF（Matrix Factorization） ：(i) users’ raw preference data(ii)users’ learned latent feature vectorsPrior studies have studied privacy-preserving M

目录Introduction关于ML privacy risk，作者的思考方向与答案：GMIA--Generalized MIAContributionsAssumptiontheoryIntroduction背景论文：Membership inference attacks against machine learning models1.Membership inference attack (MIA) can succeed on overfitted models with only blac