Backdoor Attacks on Discrete Graph Diffusion Models

Name: Backdoor Attacks on Discrete Graph Diffusion Models
Uploaded: 2026-10-06T20:52:05.105373+00:00
Duration: 710 s
Description: This work introduces a novel backdoor attack against Discrete Graph Diffusion Models (DGDMs) for graph generation. The attack manipulates both training and inference using a subgraph trigger to create stealthy, persistent backdoored graphs. Crucially, the generated graphs maintain essential properties like permutation invariance and exchangeability while preserving the model's core utility.

This work introduces a novel backdoor attack against Discrete Graph Diffusion Models (DGDMs) for graph generation. The attack manipulates both training and inference using a subgraph trigger to create stealthy, persistent backdoored graphs. Crucially, the generated graphs maintain essential properties like permutation invariance and exchangeability while preserving the model's core utility.

Open the full episode