Is Diffusion Model Safe? Severe Data Leakage via Gradient-Guided Diffusion Model

arXiv:2406.09484v1 Announce Type: cross
Abstract: Gradient leakage has been identified as a potential source of privacy breaches in modern image processing systems, where the adversary can completely reconstruct the training images from leaked gradients. However, existing methods are restricted to reconstructing low-resolution images where data leakage risks of image processing systems are not sufficiently explored. In this paper, by exploiting diffusion models, we propose an innovative gradient-guided fine-tuning method and introduce a new reconstruction attack that is capable of stealing …

adversary arxiv breaches can cs.cr cs.cv data data leakage diffusion model image images leaked low privacy resolution restricted risks safe systems training