Mois : novembre 2020

Orateur : Catuscia Palaminessi
Titre : A new mechanism for distributed Differential Privacy
Résumé : Differential Privacy (DP) is one of the most successful proposal to protect the privacy of the sensitive data while preserving their utility. In this talk, we will briefly introduce the DP framework, and then present a new proposal for a mechanism to implement DP in the distributed case, namely in a scenario in which the data collections are distributed across different organizations, which do not wish to disclose the original data but only their sanitized version, and still benefit from the advantages of combining the information coming from different sources. The mechanism we propose is particularly suitable for the application of a variant of the statistical Expectation-Maximization method, thanks to which the utility of the original data can be retrieved to an arbitrary degree of approximation, without affecting the privacy of the original data owners.