Research map / Adversarial robustness and security / Certified and empirical robustness
CertDW: Towards Certified Dataset Ownership Verification via Conformal Calibration
Ting Qiao, Yiming Li, Jianbin Li, Y. Q. Wang, Leyi Qi, Junfeng Guo, Ruili Feng, Dacheng Tao
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 conformal calibration (CertDW vs CertDW-V) (ablation) | CertDW-V · GTSRB BadNets, sigma 2.5 | VSR: 52 -> 72 | Improved |
| add conformal calibration (CertDW vs CertDW-V) (ablation) | CertDW-V · GTSRB Blended (patch), sigma 2.5 | VSR: 46 -> 78 | Improved |
| add conformal calibration (CertDW vs CertDW-V) (ablation) | CertDW-V · GTSRB Blended (noise), sigma 2.5 | WCA: 54 -> 80 | Improved |
| add conformal calibration (CertDW vs CertDW-V) (ablation) | CertDW-V · CIFAR-10 BadNets, sigma 1.2 | WCA: 24 -> 36 | Improved |
| add conformal calibration (CertDW vs CertDW-V) (ablation) | CertDW-V · CIFAR-10 Blended (noise), sigma 1.2 | WCA: 78 -> 78 | No clear effect |
| use certified verification instead of vanilla verification (ablation) | Vanilla · GTSRB BadNets, sigma 2.5 | WCA: 0 -> 48 | Improved |
| use certified verification instead of vanilla verification (ablation) | Vanilla · GTSRB Blended (patch), sigma 1.5 | VSR: 80 -> 82 | No clear effect |
| use certified verification instead of vanilla verification (ablation) | Vanilla · GTSRB Blended (noise), sigma 2.5 | VSR: 0 -> 96 | Improved |
| add Gaussian noise to verification samples | vanilla dataset watermark · CIFAR-10 | WSR: reduce WSR by 60% | Hurt |
| add adversarial perturbation to verification samples | Blended watermark · verification sample space | WSR: drop by over 80% | Hurt |
| increase trigger size R from (0.3,0.8) to (0.8,1.3) | CertDW · GTSRB, sigma 1.5 | certified region area: 0.193 -> 0.133 | Hurt |
| increase noise level sigma from 1.5 to 3.5 | CertDW · GTSRB, R in (0.8,1.3) | certified region area: 0.133 -> 0.218 | Improved |
Closest papers by meaning
- ZeroMark: Towards Dataset Ownership Verification without Disclosing Watermark (NeurIPS 2024 · similarity 0.93)
- Domain Watermark: Effective and Harmless Dataset Copyright Protection is Closed at Hand (NeurIPS 2023 · similarity 0.92)
- EntropyMark: Towards More Harmless Backdoor Watermark via Entropy-based Constraint for Open-source Dataset Copyright Protection (CVPR 2025 · similarity 0.90)
- Untargeted Backdoor Watermark: Towards Harmless and Stealthy Dataset Copyright Protection (NeurIPS 2022 · similarity 0.89)
- Data Taggants: Dataset Ownership Verification Via Harmless Targeted Data Poisoning (ICLR 2025 · similarity 0.88)
- Dataset Ownership Verification in Contrastive Pre-trained Models (ICLR 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.
