Ablating Concepts in Text-to-Image Diffusion Models

By Nupur Kumari, Bingliang Zhang, Sheng-Yu Wang, Eli Shechtman, Richard Zhang, Jun-Yan Zhu. IEEE International Conference on Computer Vision, 2023

361 citations recorded by the source corpus.

Large-scale text-to-image diffusion models can generate high-fidelity images with powerful compositional ability. However, these models are typically trained on an enormous amount of Internet data, often containing copyrighted material, licensed images, and personal photos. Furthermore, they have been found to replicate the style of various living artists or memorize exact training samples. How can we remove such copyrighted concepts or images without retraining the model from scratch? To achieve this goal, we propose an efficient method of ablating concepts in the pretrained model, i.e., preventing the generation of a target concept. Our algorithm learns to match the image distribution for a target style, instance, or text prompt we wish to ablate to the distribution corresponding to an anchor concept. This prevents the model from generating target concepts given its text condition. Extensive experiments show that our method can successfully prevent the generation of the ablated concept while preserving closely related concepts in the model.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Signal Processing

View Paper: https://www.semanticscholar.org/paper/62b368417eca18fe68d209e5a077392748450a29

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

Rethinking machine unlearning for large language models (2024, neutral): “…of MU has expanded to encompass diverse domains, such as image classification Ginart et al. (2019); Golatkar et al. (2020) In the literature, ‘exact’ unlearning, which involves retraining the model from scratch after removing specific training data points, is often considered the gold standard.”
Forget-Me-Not: Learning to Forget in Text-to-Image Diffusion Models (2023, neutral): “Kumari et al. [26] and Heng and Soh [19] choose to replace target concept with a user-deﬁned concept, usually a hypernym.”
Unified Concept Editing in Diffusion Models (2023, methodology): “Another approach adjusts a model after training is complete, deleting an undesired concept by modifying model weights [13, 15, 20, 21, 46].”
MACE: Mass Concept Erasure in Diffusion Models (2024, methodology): “Additionally, it is noteworthy that while meth-ods like FMN [71] and AC [30] are proﬁcient in removing speciﬁc features of a subject, they fall short in completely eradicating the subject’s generation.”