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

Qitan Shi, Cheng Jin, Jiawei Zhang, Yuantao Gu
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:2818-2826, 2026.

Abstract

Diffusion models excel at generating high-quality, diverse images but also suffer from undesirable training data memorization, raising critical privacy and safety concerns. Data unlearning has emerged to mitigate this issue by removing the influence of specific data through fine-tuning rather than 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 unbiased fine-tuning loss. This loss is further approximated by retaining only the dominant terms, thereby reducing computational cost. 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.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-shi26a, title = { ReTrack: Data Unlearning in Diffusion Models Through Redirecting the Denoising Trajectory }, author = {Shi, Qitan and Jin, Cheng and Zhang, Jiawei and Gu, Yuantao}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {2818--2826}, year = {2026}, editor = {Khan, Emtiyaz and Li, Yingzhen and Solin, Arno and Ramdas, Aaditya}, volume = {300}, series = {Proceedings of Machine Learning Research}, month = {02--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v300/main/assets/shi26a/shi26a.pdf}, url = {https://proceedings.mlr.press/v300/shi26a.html}, abstract = { Diffusion models excel at generating high-quality, diverse images but also suffer from undesirable training data memorization, raising critical privacy and safety concerns. Data unlearning has emerged to mitigate this issue by removing the influence of specific data through fine-tuning rather than 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 unbiased fine-tuning loss. This loss is further approximated by retaining only the dominant terms, thereby reducing computational cost. 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. } }
Endnote
%0 Conference Paper %T ReTrack: Data Unlearning in Diffusion Models Through Redirecting the Denoising Trajectory %A Qitan Shi %A Cheng Jin %A Jiawei Zhang %A Yuantao Gu %B Proceedings of The 29th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2026 %E Emtiyaz Khan %E Yingzhen Li %E Arno Solin %E Aaditya Ramdas %F pmlr-v300-shi26a %I PMLR %P 2818--2826 %U https://proceedings.mlr.press/v300/shi26a.html %V 300 %X Diffusion models excel at generating high-quality, diverse images but also suffer from undesirable training data memorization, raising critical privacy and safety concerns. Data unlearning has emerged to mitigate this issue by removing the influence of specific data through fine-tuning rather than 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 unbiased fine-tuning loss. This loss is further approximated by retaining only the dominant terms, thereby reducing computational cost. 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.
APA
Shi, Q., Jin, C., Zhang, J. & Gu, Y.. (2026). ReTrack: Data Unlearning in Diffusion Models Through Redirecting the Denoising Trajectory . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:2818-2826 Available from https://proceedings.mlr.press/v300/shi26a.html.

Related Material