LinkTeller: Recovering Private Edges from Graph Neural Networks via Influence Analysis

Explore a 16-minute IEEE conference talk on recovering private edges from Graph Neural Networks through influence analysis. Delve into the risks of edge privacy in vertically partitioned graph data and understand the overview of the LinkTeller attack. Examine the Differentially Private Graph Convolutional Network (DP GCN) framework and its practical implementation. Evaluate the effectiveness of LinkTeller against DP GCN and analyze the trade-off between model utility and privacy. Gain insights from researchers at the University of Illinois at Urbana-Champaign and ETH Zurich as they present their findings on this critical topic in graph data security.

Intro
Motivation
Vertically Partitioned Graph Data
Risks of Edge Privacy
Related Work & Preliminaries
Overview of the Attack
DP GCN Framework
Practical DP GCN
Evaluation of Link Teller
LinkTeller against DP GCN
Trade-off between Model Utility and Privacy
Summary