CMStatistics 2019: Start Registration
View Submission - CMStatistics
B0352
Title: Privacy in data dissemination, differential privacy, and analysis of perturbed data Authors:  Yosef Rinott - The Hebrew University (Israel) [presenting]
Abstract: Privacy issues that arise when an agency disseminates data will be briefly reviewed, along with some of the methods used by statisticians to assess the disclosure risk, and to decrease it. In general, such methods depend on scenarios regarding prior knowledge of potential intruders and the nature of the disseminated data. Differential Privacy is an approach that avoids much of the need to consider such scenarios, and guarantees a well-defined notion of privacy by adding noise with a known distribution to all released data. Some basic results on differential privacy and applications to the release of contingency tables will be discussed. In many cases, the released data appears like real data, and many researchers tend to analyze it without taking the noise distribution into account. Ongoing work on data analysis that takes the added noise into account, and the loss incurred by ignoring it will be discussed.