![[2507.07139] Image Can Bring Your Memory Back: A Novel Multi-Modal Guided Attack against Image Generation Model Unlearning](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2507.07139] Image Can Bring Your Memory Back: A Novel Multi-Modal Guided Attack against Image Generation Model Unlearning
Summary
The paper presents Recall, a novel adversarial framework that targets the robustness of image generation model unlearning, revealing vulnerabilities in current methods.
Why It Matters
As image generation models become more prevalent, ensuring their ethical use is critical. This research highlights weaknesses in existing unlearning techniques, emphasizing the need for improved safety measures in AI systems to prevent misuse and enhance reliability.
Key Takeaways
- Recall is a new adversarial framework that enhances the effectiveness of attacks on unlearned image generation models.
- The study demonstrates significant vulnerabilities in current unlearning methods, particularly against multi-modal adversarial inputs.
- Recall outperforms existing techniques in adversarial effectiveness and computational efficiency.
- The findings stress the importance of developing more robust unlearning solutions in AI.
- Publicly available code and data support further research and validation of the proposed methods.
Computer Science > Computer Vision and Pattern Recognition arXiv:2507.07139 (cs) [Submitted on 9 Jul 2025 (v1), last revised 16 Feb 2026 (this version, v2)] Title:Image Can Bring Your Memory Back: A Novel Multi-Modal Guided Attack against Image Generation Model Unlearning Authors:Renyang Liu, Guanlin Li, Tianwei Zhang, See-Kiong Ng View a PDF of the paper titled Image Can Bring Your Memory Back: A Novel Multi-Modal Guided Attack against Image Generation Model Unlearning, by Renyang Liu and 3 other authors View PDF HTML (experimental) Abstract:Recent advances in image generation models (IGMs), particularly diffusion-based architectures such as Stable Diffusion (SD), have markedly enhanced the quality and diversity of AI-generated visual content. However, their generative capability has also raised significant ethical, legal, and societal concerns, including the potential to produce harmful, misleading, or copyright-infringing content. To mitigate these concerns, machine unlearning (MU) emerges as a promising solution by selectively removing undesirable concepts from pretrained models. Nevertheless, the robustness and effectiveness of existing unlearning techniques remain largely unexplored, particularly in the presence of multi-modal adversarial inputs. To bridge this gap, we propose Recall, a novel adversarial framework explicitly designed to compromise the robustness of unlearned IGMs. Unlike existing approaches that predominantly rely on adversarial text prompts, Recall ...