README.md

---
license: cc-by-4.0
language:
- en
tags:
- art
- text-to-image
- stable-diffusion-diffusers
- unlearned-diffusion-model
- safe-diffusion-model
- unlearned-text-encoder
- defensive-unlearning
---


# Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models
Paper can be checked in [Arxiv Preprint](https://arxiv.org/abs/2405.15234). <br>
Code can be checked in [GitHub](https://github.com/OPTML-Group/AdvUnlearn). <br>

Our proposed robust unlearning framework, AdvUnlearn, enhances diffusion models' safety by robustly erasing unwanted concepts through adversarial training, achieving an optimal balance between concept erasure and image generation quality. 



<div align="center">
    <img src="https://cdn-uploads.huggingface.co/production/uploads/63e44e62789dcaae43c865d9/vad9l9ME0KD0OJKYLcFun.png" />
</div>


## Baselines

| DM Unlearning Methods | Nudity  | Van Gogh  | Objects |
|:-------|:----:|:-------:| :-------:|
| [ESD](https://github.com/rohitgandikota/erasing) (Erased Stable Diffusion)  | ✅  | ✅   | ✅ 
| [FMN](https://github.com/SHI-Labs/Forget-Me-Not) (Forget-Me-Not)  | ✅ | ✅   | ✅ 
| [AC](https://github.com/nupurkmr9/concept-ablation) (Ablating Concepts)  | ❌ | ✅   | ❌ 
| [UCE](https://github.com/rohitgandikota/unified-concept-editing) (Unified Concept Editing)  | ✅  |  ✅  |  ❌
| [SalUn](https://github.com/OPTML-Group/Unlearn-Saliency) (Saliency Unlearning)  | ✅  | ❌ |  ✅ 
| [SH](https://github.com/JingWu321/Scissorhands_ex) (ScissorHands)  | ✅  | ❌  | ✅ 
| [ED](https://github.com/JingWu321/EraseDiff) (EraseDiff)  | ✅  | ❌  | ✅ 
| [SPM](https://github.com/Con6924/SPM) (concept-SemiPermeable Membrane)   | ✅  | ✅   | ✅ 
| **AdvUnlearn (Ours)**  | ✅  | ✅   |  ✅ 


<br>


## Cite Our Work
The preprint can be cited as follows:
```
@misc{zhang2024defensive,
      title={Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models}, 
      author={Yimeng Zhang and Xin Chen and Jinghan Jia and Yihua Zhang and Chongyu Fan and Jiancheng Liu and Mingyi Hong and Ke Ding and Sijia Liu},
      year={2024},
      eprint={2405.15234},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}
```

---
license: cc-by-4.0
---