Private Synthetic Data Generation in Bounded Memory

We propose $P r i v H P$ , a lightweight synthetic data generator with $differential privacy$ guarantees. $P r i v H P$ uses a novel hierarchical decomposition that approximates the input’s cumulative distribution function (CDF) in bounded memory. It balances hierarchy depth, noise addition, and pruning of low-frequency subdomains while preserving frequent ones. Private sketches estimate subdomain frequencies efficiently without full data access. A key feature is the pruning parameter $k$ , which controls the trade-off between space and utility. We define the skew measure ${t a i l}_{k}$ , capturing all but the top $k$ subdomain frequencies. Given a dataset $X$ , $P r i v H P$ uses $M = O (k \log^{2} ∣ X ∣)$ space and, for input domain $Ω = [0, 1]$ , ensures $ε$ -differential privacy. It yields a generator with expected Wasserstein distance: $O (\frac{\log^{2} M}{ε n} + \frac{∣ ∣ {t a i l}_{k} (X) ∣ ∣_{1}}{M n})$ from the empirical distribution. This parameterized trade-off offers a level of flexibility unavailable in prior work. We also provide interpretable utility bounds that account for hierarchy depth, privacy noise, pruning, and frequency estimation errors.

Private Synthetic Data Generation in Bounded Memory

文献摘要

讨论

讲解博客

AI智能校验

Private Synthetic Data Generation in Bounded Memory

文献摘要

讨论

讲解博客

AI智能校验

Private Synthetic Data Generation in Bounded Memory

文献摘要

讨论

|讲解博客

|AI智能校验

Private Synthetic Data Generation in Bounded Memory

文献摘要

讨论

|讲解博客

|AI智能校验

讲解博客

AI智能校验

讲解博客

AI智能校验