Visual Privacy Auditing with Diffusion Models

Data reconstruction attacks on machine learning models pose a substantial threat to pri- vacy, potentially leaking sensitive information. Although defending against such attacks using differential privacy (DP) provides theoretical guarantees, determining appropriate DP parameters remains challenging. Current formal guarantees on the success of data recon- struction suffer from overly stringent assumptions regarding adversary knowledge about the target data, particularly in the image domain, raising questions about their real-world applicability. In this work, we empirically investigate this discrepancy by introducing a re- construction attack based on diffusion models (DMs) that only assumes adversary access to real-world image priors and specifically targets the DP defense. We find that (1) real-world data priors significantly influence reconstruction success, (2) current reconstruction bounds do not model the risk posed by data priors well, and (3) DMs can serve as heuristic auditing tools for visualizing privacy leakage.

Download paper here

Recommended citation: Schwethelm, K., Kaiser, J., Knolle, M., Lockfisch, S., Rueckert, D., & Ziller, A. (2024). Visual Privacy Auditing with Diffusion Models. arXiv preprint arXiv:2403.07588.