Research map / Adversarial robustness and security / Attacks on text and graphs
Towards Robust Model Watermark via Reducing Parametric Vulnerability
Guanhao Gan, Yiming Li, Dongxian Wu, Shu-Tao Xia
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 adversarial parametric perturbation (APP) alone, without c-BN (ablation) | vanilla watermarking · CIFAR-10 Content watermark | average WSR drop: 59.15 -> 62.91 | Hurt |
| add clean-sample BatchNorm (c-BN) alone, without APP (ablation) | vanilla watermarking · CIFAR-10 Content watermark | average WSR drop: 59.15 -> 82.24 | Hurt |
| combine APP with c-BN (ablation) | vanilla watermarking · CIFAR-10 Content watermark | average WSR drop: 59.15 -> 10.10 | Improved |
| APP + c-BN versus baselines on unrelated-domain watermark | model watermark · CIFAR-10 Unrelated watermark | average WSR drop: 50.90 -> 6.20 | Improved |
| APP + c-BN on ImageNet-subset content watermark | model watermark · ImageNet-subset Content watermark | average WSR drop: 78.20 -> 39.40 | Improved |
| APP + c-BN effect on benign accuracy | model watermark · CIFAR-10 | benign accuracy: 93.86 -> 93.42 | No clear effect |
| APP + c-BN on ImageNet-subset benign accuracy | model watermark · ImageNet-subset Content watermark | benign accuracy: 74.81 -> 72.29 | Hurt |
| apply APP-based method to different target classes (ablation) | model watermark · CIFAR-10 Content watermark | average WSR under removal attacks: 32.91% -> 94.87% (class 1) | Improved |
| APP versus NNL attack on noise watermark | APP + c-BN · CIFAR-10 Noise watermark | WSR after NNL: 1.00 -> 3.22 | No clear effect |
| vary APP perturbation magnitude within epsilon <= 4.0e-2 (ablation) | APP + c-BN · CIFAR-10 Content watermark | robustness against attacks | Improved |
Closest papers by meaning
- Safe and Robust Watermark Injection with a Single OoD Image (ICLR 2024 · similarity 0.90)
- EntropyMark: Towards More Harmless Backdoor Watermark via Entropy-based Constraint for Open-source Dataset Copyright Protection (CVPR 2025 · similarity 0.88)
- ZeroMark: Towards Dataset Ownership Verification without Disclosing Watermark (NeurIPS 2024 · similarity 0.88)
- Lightweight-Mark: Rethinking Deep Learning-Based Watermarking (ICML 2025 · similarity 0.87)
- IPRemover: A Generative Model Inversion Attack against Deep Neural Network Fingerprinting and Watermarking (AAAI 2024 · similarity 0.87)
- Defense against Model Extraction Attack by Bayesian Active Watermarking (ICML 2024 · similarity 0.87)
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.
