Séance Séminaire

Séminaire de Probabilités et Statistique

Building a decision theoretic approach to privacy

While several results in the literature (e.g., Dimitrakakis et al., 2017; Zhang and Zhang, 2023) demonstrate that Bayesian inference approximated by MCMC output can achieve differential privacy with zero or limited impact on the ensuing posterior, we argue that the ensuing privacy is mostly related to a slowing-down of MCMC convergence rather than a generic gain in protecting data privacy. We then propose an alternative setting where privacy and inferential goals are brought together to achieve Bayesian efficiency. The case of insufficient statistics is processed as a specific illustration of the approach.

This is joint work with Joshua Bon and Stanislas du Ché (Université Paris Dauphine), supported by the 2023-2029 ERC Synergy Grant OCEAN.

Séminaire en salle 109, également retransmis sur zoom : https://umontpellier-fr.zoom.us/j/7156708132