Webinar "Differential privacy and statistical inference"

The Working Group on Risk - CREAR, with the support of the IDS dpt , Institut des Actuaires, LabEx MME-DII and the group BFA-SFdS, has the pleasure to invite you to the seminar by:

Prof. Cristina BUTUCEA, Professor of statistics and machine learning at ENSAE, Institut Polytechnique de Paris.

Thursday February 3, 2022 12:30 pm (CET)
ESSEC Paris La Défense (CNIT), Amphi 201, and via Zoom, please click here (password/code : WGRisk)

Topics
Differential privacy (DP) has prevailed as the most convenient formalism to randomize sensitive data via privacy mechanisms submitted to some constraints. The concept of DP provides a rigorous formalism to randomize data and quantify the amount of privacy, but also leaves information on the initial sample.
We will introduce this concept and give several examples of privacy mechanisms that randomize the sample under the DP constraints and see how that affects the statistical learning. We distinguish global (or  central) differential privacy when the privacy mechanism uses all the original data, vs. local differential privacy when each sample from the original data is privatized on the user’s local machine before its release. In the sequel, we consider only the setup of local differential privacy (LDP) where slower rates are typically attained as compared to the optimal procedures that use the original data.
The problem of non-parametric estimation of functionals of the probability density will be further detailed.
We show that for estimation of a quadratic functional, interactive procedures that use previously released private data are faster than the non interactive ones.

Attending this webinar will enable you to acquire 6 points PPC CERA.