TY - JOUR AU - Wang, Yu-Xiang AU - Balle, Borja AU - Kasiviswanathan, Shiva PY - 2021/02/03 Y2 - 2024/03/28 TI - Subsampled Rényi Differential Privacy and Analytical Moments Accountant JF - Journal of Privacy and Confidentiality JA - JPC VL - 10 IS - 2 SE - TPDP 2018 DO - 10.29012/jpc.723 UR - https://journalprivacyconfidentiality.org/index.php/jpc/article/view/723 SP - AB - <p>We study the problem of subsampling in differential privacy (DP), a question that is the centerpiece behind many successful differentially private machine learning algorithms. Specifically, we provide a tight upper bound on the Renyi Differential Privacy (RDP) [Mironov, 2017] parameters for algorithms that: (1) subsample the dataset, and then (2) apply a randomized mechanism M to the subsample, in terms of the RDP parameters of M and the subsampling probability parameter.<br>Our results generalize the moments accounting technique, developed by [Abadi et al. 2016] for the Gaussian mechanism, to any subsampled RDP mechanism.</p> ER -