ReTrack: Data Unlearning in Diffusion Models through Redirecting the Denoising Trajectory

View PDF HTML (experimental)

Abstract:Diffusion models excel at generating high-quality, diverse images but suffer from training data memorization, raising critical privacy and safety concerns. Data unlearning has emerged to mitigate this issue by removing the influence of specific data without retraining from scratch. We propose ReTrack, a fast and effective data unlearning method for diffusion models. ReTrack employs importance sampling to construct a more efficient fine-tuning loss, which we approximate by retaining only dominant terms. This yields an interpretable objective that redirects denoising trajectories toward the $k$-nearest neighbors, enabling efficient unlearning while preserving generative quality. Experiments on MNIST T-Shirt, CelebA-HQ, CIFAR-10, and Stable Diffusion show that ReTrack achieves state-of-the-art performance, striking the best trade-off between unlearning strength and generation quality preservation.

Comments: 22 pages, 12 figures, accepted by AISTATS 2026

Subjects:

Machine Learning (cs.LG)

Cite as: arXiv:2509.13007 [cs.LG]

(or arXiv:2509.13007v2 [cs.LG] for this version)

https://doi.org/10.48550/arXiv.2509.13007

arXiv-issued DOI via DataCite

Submission history

From: Qitan Shi [view email] [v1] Tue, 16 Sep 2025 12:20:15 UTC (2,131 KB) [v2] Sun, 29 Mar 2026 15:08:09 UTC (2,700 KB)