Differentially Private Bayesian Linear Regression

Naysan Saran, Joseph Antoun, Mohamed Amine Arfaoui

Abstract:

Linear regression is an important tool across many
fields that work with sensitive human-sourced data. Significant
prior work has focused on producing differentially private point
estimates, which provide a privacy guarantee to individuals
while still allowing modelers to draw insights from data by
estimating regression coefficients. The problem of Bayesian linear
regression, with the goal of computing posterior distributions that
correctly quantify uncertainty given privately released statistics,
was investigated in [1]. In the aforementioned paper, it was shown
that a naive approach that ignores the noise injected by the
privacy mechanism does a poor job in realistic data settings. This
has motivated the corresponding authors to develop noise-aware
methods that perform inference over the privacy mechanism and
produce correct posteriors across a wide range of scenarios. In
this paper, the methods developed in [1] are first reproduced
and then enhanced through various experiments. Finally, based
on the outcome of the present paper and the current state of the
art, some future research directions are proposed

Track: Ablation

NeurIPS Paper Id: /forum?id=ryeGE4rg8B