Tractable MCMC for Private Learning with Pure and Gaussian Differential Privacy

arXiv:2310.14661v2 Announce Type: replace
Abstract: Posterior sampling, i.e., exponential mechanism to sample from the posterior distribution, provides $\varepsilon$-pure differential privacy (DP) guarantees and does not suffer from potentially unbounded privacy breach introduced by $(\varepsilon,\delta)$-approximate DP. In practice, however, one needs to apply approximate sampling methods such as Markov chain Monte Carlo (MCMC), thus re-introducing the unappealing $\delta$-approximation error into the privacy guarantees. To bridge this gap, we propose the Approximate SAample Perturbation (abbr. ASAP) algorithm which perturbs an MCMC sample …

abstract apply arxiv breach cs.lg delta differential differential privacy distribution however markov mcmc posterior practice privacy sample sampling stat.ml tractable type