Your Discrete Graph Diffusion Model is Secretly Memorizing

In 30 Sekunden

Was passiert ist

Researchers found that discrete graph diffusion models memorize training data when reverse trajectories are conditioned on corrupted training graphs, despite appearing to generalize.

Warum es zählt

Matters for engineers building graph generation systems who need to understand whether their models truly learn generalizable patterns or exploit training data memorization.

Achtung

The memorization is conditional and hidden during normal inference, only surfacing when analyzing reverse diffusion trajectories, making it easy to miss in standard evaluation.