We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Methodology

Title: Re-weighting of Vector-weighted Mechanisms for Utility Maximization under Differential Privacy

Abstract: We present practical aspects of implementing a pseudo posterior synthesizer for microdata dissemination under a new re-weighting strategy for utility maximization that applies to any vector-weighting approach. We demonstrate on two different vector-weighted schemes. Both schemes target high-risk records by exponentiating each of their likelihood contributions with a record-level weight, $\alpha_i \in [0,1]$ for record $i \in (1,\ldots,n)$. The first vector-weighted synthesizing mechanism computes the maximum (Lipschitz) bound, $\Delta_{x_{i}}$, of each log-likelihood contribution over the space of parameter values, and sets the by-record weight $\alpha_{i} \propto 1 / \Delta_{x_{i}}$. The second vector-weighted synthesizer is based on constructing an identification disclosure risk probability, $IR_i$ of record $i$, and setting the by-record weight $\alpha_i \propto 1 / IR_i$. We compute the overall Lipschitz bound, $\Delta_{\boldsymbol{\alpha},\mathbf{x}}$, for the database $\mathbf{x}$, under each vector-weighted synthesizer such that both provide an $(\epsilon = 2 \Delta_{\boldsymbol{\alpha},\mathbf{x}})-$differential privacy (DP) formal guarantee. Our new vector re-weighting strategy maximizes the data utility given any privacy budget for the vector-weighted synthesizers by adjusting the by-record weights, $(\alpha_{i})_{i = 1}^{n}$, such that their individual Lipschitz bounds, $\Delta_{\boldsymbol{\alpha},x_{i}}$, approach the bound for the entire database, $\Delta_{\boldsymbol{\alpha},\mathbf{x}}$. We illustrate our methods using simulated count data with and without over-dispersion-induced skewness and compare the results to a scalar-weighted synthesizer under the Exponential Mechanism (EM). We further demonstrate our methods on a sample of the Survey of Doctorate Recipients.
Subjects: Methodology (stat.ME); Applications (stat.AP)
Cite as: arXiv:2006.01230 [stat.ME]
  (or arXiv:2006.01230v2 [stat.ME] for this version)

Submission history

From: Jingchen Hu [view email]
[v1] Mon, 1 Jun 2020 20:03:37 GMT (1480kb,D)
[v2] Fri, 3 Jul 2020 14:36:33 GMT (2171kb,D)
[v3] Tue, 23 Mar 2021 15:56:21 GMT (2513kb,D)
[v4] Fri, 5 Nov 2021 12:43:26 GMT (2521kb,D)
[v5] Thu, 28 Apr 2022 18:17:11 GMT (2162kb,D)

Link back to: arXiv, form interface, contact.