Speaker: Anne-Sophie Charest, Professeure adjointe Département de mathématiques et de statistique Université Laval
Title: Statistical Privacy, Differential Privacy and an Application to Model Selection
Date: Tuesday, March 7, 2017
Time: 4:00 – 5:30 p.m.
Room: HP 4351 (Macphail Room), Carleton University
While statistical agencies would like to share their data with researchers, they must also protect the confidentiality of the data provided by their respondents. I will first briefly review the various techniques used in practice to satisfy these two conflicting objectives. I will then introduce the rigorous criterion of differential privacy, which is designed to guarantee confidentiality even in a worst-case scenario, by protecting the information of any individual in the database against an adversary with complete knowledge of the rest of the dataset. I will finally present a new method for differentially-private model selection, which combines constrained maximum likelihood estimation and noisy optimization of an information criterion.