Differentially Private Marginals Release with Mutual Consistency and Error Independent of Sample Size

We report on a result of Barak et al. on a privacy-preserving technology for release of mutually consistent multi-way marginals [1]. The result ensures differential privacy, a mathematically rigorous notion for privacy-preserving statistical data analysis capturing the intuition that essentially no harm can befall a respondent who accurately reports her data beyond that which would befall her should she refuse to respond, or respond completely inaccurately [7, 5]. In addition to differential privacy, the techniques described herein ensure consistency among released tables and, in many cases, excellent accuracy.