GNNBleed: Inference Attacks to Unveil Private Edges in Graphs with Realistic Access to GNN Models

Name: GNNBleed: Inference Attacks to Unveil Private Edges in Graphs with Realistic Access to GNN Models
Uploaded: 2026-08-27T00:29:31.900846+00:00
Duration: 585 s
Description: The episode discusses 'GNNBleed,' a paper detailing inference attacks that unveil private edges in graphs using realistic access to Graph Neural Network (GNN) models. Hosts explore how gradient information reveals secret connections, and improved methods like relative, node-specific thresholds are proposed to mitigate this vulnerability.

The episode discusses 'GNNBleed,' a paper detailing inference attacks that unveil private edges in graphs using realistic access to Graph Neural Network (GNN) models. Hosts explore how gradient information reveals secret connections, and improved methods like relative, node-specific thresholds are proposed to mitigate this vulnerability.

Open the full episode