Skip to Content

Differentially Private Graph Learning via Sensitivity-Bounded Personalized PageRank

Posted on One min read
Charts with DPPPR performance.
Approximation quality of differentially-private personalized pagerank.

paper · pdf

TL;DR

Novel differentially-private primitive for private graph learning.

In this paper, we introduce DPPPR, an algorithm that approximates the PPR vector with a provably bounded sensitivity to edge changes. We also introduce its joint version, which is a more practical yet natural definition of privacy in graphs.