Minimalist Concept Erasure in Generative Models

Yang Zhang, Er Jin, Yanfei Dong, Yixuan Wu, Philip Torr, Ashkan Khakzar, Johannes Stegmaier, Kenji Kawaguchi
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:75192-75215, 2025.

Abstract

Recent advances in generative models have demonstrated remarkable capabilities in producing high-quality images, but their reliance on large-scale unlabeled data has raised significant safety and copyright concerns. Efforts to address these issues by erasing unwanted concepts have shown promise. However, many existing erasure methods involve excessive modifications that compromise the overall utility of the model. In this work, we address these issues by formulating a novel minimalist concept erasure objective based only on the distributional distance of final generation outputs. Building on our formulation, we derive a tractable loss for differentiable optimization that leverages backpropagation through all generation steps in an end-to-end manner. We also conduct extensive analysis to show theoretical connections with other models and methods. To improve the robustness of the erasure, we incorporate neuron masking as an alternative to model fine-tuning. Empirical evaluations on state-of-the-art flow-matching models demonstrate that our method robustly erases concepts without degrading overall model performance, paving the way for safer and more responsible generative models.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-zhang25ai, title = {Minimalist Concept Erasure in Generative Models}, author = {Zhang, Yang and Jin, Er and Dong, Yanfei and Wu, Yixuan and Torr, Philip and Khakzar, Ashkan and Stegmaier, Johannes and Kawaguchi, Kenji}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {75192--75215}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/zhang25ai/zhang25ai.pdf}, url = {https://proceedings.mlr.press/v267/zhang25ai.html}, abstract = {Recent advances in generative models have demonstrated remarkable capabilities in producing high-quality images, but their reliance on large-scale unlabeled data has raised significant safety and copyright concerns. Efforts to address these issues by erasing unwanted concepts have shown promise. However, many existing erasure methods involve excessive modifications that compromise the overall utility of the model. In this work, we address these issues by formulating a novel minimalist concept erasure objective based only on the distributional distance of final generation outputs. Building on our formulation, we derive a tractable loss for differentiable optimization that leverages backpropagation through all generation steps in an end-to-end manner. We also conduct extensive analysis to show theoretical connections with other models and methods. To improve the robustness of the erasure, we incorporate neuron masking as an alternative to model fine-tuning. Empirical evaluations on state-of-the-art flow-matching models demonstrate that our method robustly erases concepts without degrading overall model performance, paving the way for safer and more responsible generative models.} }
Endnote
%0 Conference Paper %T Minimalist Concept Erasure in Generative Models %A Yang Zhang %A Er Jin %A Yanfei Dong %A Yixuan Wu %A Philip Torr %A Ashkan Khakzar %A Johannes Stegmaier %A Kenji Kawaguchi %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-zhang25ai %I PMLR %P 75192--75215 %U https://proceedings.mlr.press/v267/zhang25ai.html %V 267 %X Recent advances in generative models have demonstrated remarkable capabilities in producing high-quality images, but their reliance on large-scale unlabeled data has raised significant safety and copyright concerns. Efforts to address these issues by erasing unwanted concepts have shown promise. However, many existing erasure methods involve excessive modifications that compromise the overall utility of the model. In this work, we address these issues by formulating a novel minimalist concept erasure objective based only on the distributional distance of final generation outputs. Building on our formulation, we derive a tractable loss for differentiable optimization that leverages backpropagation through all generation steps in an end-to-end manner. We also conduct extensive analysis to show theoretical connections with other models and methods. To improve the robustness of the erasure, we incorporate neuron masking as an alternative to model fine-tuning. Empirical evaluations on state-of-the-art flow-matching models demonstrate that our method robustly erases concepts without degrading overall model performance, paving the way for safer and more responsible generative models.
APA
Zhang, Y., Jin, E., Dong, Y., Wu, Y., Torr, P., Khakzar, A., Stegmaier, J. & Kawaguchi, K.. (2025). Minimalist Concept Erasure in Generative Models. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:75192-75215 Available from https://proceedings.mlr.press/v267/zhang25ai.html.

Related Material