Research map / Image generation / Text-to-image diffusion
Invisible Watermarks, Visible Gains: Steering Machine Unlearning with Bi-Level Watermarking Design
Yuhao Sun, Yihua Zhang, Gaowen Liu, Hongtao Xie, Sijia Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Water4MU watermarking | GA · CIFAR-10 (10% random data forgetting) | UA (%): 0.80 -> 1.92 | Improved |
| add Water4MU watermarking | Retrain · CIFAR-10 (10% random data forgetting) | UA (%): 6.78 -> 10.01 | Improved |
| add Water4MU watermarking | FT · CIFAR-10 (class-wise forgetting) | UA (%): 37.29 -> 53.25 | Improved |
| add Water4MU watermarking | Retrain · CIFAR-10 (class-wise forgetting) | UA (%): 100.00 -> 100.00 | No clear effect |
| add Water4MU watermarking | Retrain · CIFAR-10 (10% random data forgetting) | TA (%): 92.02 -> 90.63 | Hurt |
| add Water4MU watermarking | IU · CIFAR-10 (10% random data forgetting) | TA (%): 94.51 -> 92.87 | Hurt |
| add Water4MU watermarking | Sparse · CIFAR-10 (class-wise forgetting) | RA (%): 99.58 -> 95.27 | Hurt |
| add Water4MU watermarking | FT · ImageNet (class-wise forgetting, 10% of classes) | UA (%): 45.52 -> 53.71 | Improved |
| add Water4MU watermarking | GA · CIFAR-10 (worst-case challenging forgets) | UA (%): 0.01 -> 1.73 | Improved |
| add Water4MU watermarking | Retrain · ImageNet (worst-case class-wise forgetting) | UA (%): 48.49 -> 51.32 | Improved |
| add watermarking of the evaluation sets at test time (ablation) | Retrain + Water4MU · CIFAR-10 (10% random data forgetting) | UA (%): 10.01 -> 15.84 | Improved |
| add watermarking of the evaluation sets at test time (ablation) | Retrain + Water4MU · CIFAR-10 (10% random data forgetting) | TA (%): 90.63 -> 88.79 | Hurt |
Closest papers by meaning
- Lightweight-Mark: Rethinking Deep Learning-Based Watermarking (ICML 2025 · similarity 0.88)
- On the Learnability of Watermarks for Language Models (ICLR 2024 · similarity 0.87)
- Transferable Black-Box One-Shot Forging of Watermarks via Image Preference Models (NeurIPS 2025 · similarity 0.87)
- Learning to Follow In-Context Watermark Instructions via Self-Distillation (NeurIPS 2026 · similarity 0.87)
- The Illusion of Unlearning: The Unstable Nature of Machine Unlearning in Text-to-Image Diffusion Models (CVPR 2025 · similarity 0.86)
- WMCopier: Forging Invisible Watermarks on Arbitrary Images (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
