Winning the MIDST Challenge: New Membership Inference Attacks on Diffusion Models for Tabular Data Synthesis

We find that membership attacks designed for image diffusion models do not transfer cleanly to tabular synthesis. A small MLP trained on loss features across noise levels and timesteps captures stronger membership signals and placed first in every challenge track.

Recommended citation: Wu, X., Pang, Y., Liu, T., and Wu, Z. S. (2025). "Winning the MIDST Challenge: New Membership Inference Attacks on Diffusion Models for Tabular Data Synthesis." TPDP 2025.
Download Paper