Research map / Label-efficient and robust learning / Self-supervised and contrastive
Dataset Ownership Verification for Pre-trained Masked Models
Yuechen Xie, Jie Song, Yicheng Shan, Xiaoyan Zhang, Yuanyu Wan, Shengxuming Zhang, Jiarui Duan, Mingli Song
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 |
|---|---|---|---|
| compute the t-test on relative difficulty delta R instead of raw difficulty R (ablation) | DOV4MM · ImageNet-1K, SimMIM Swin-B | p-value: 0.10 -> 0.03 | Improved |
| increase decoder training set D_t from 10,000 to 20,000 images (ablation) | DOV4MM · ImageNet-1K, CAE ViT-B/16 | p-value: 0.02 -> 10^-3 | Improved |
| increase sampling iterations K from 10 to 30 (ablation) | DOV4MM · ImageNet-1K, CAE ViT-B/16 | p-value: 0.05 -> 10^-3 | Improved |
| increase samples per iteration N from 256 to 4,096 (ablation) | DOV4MM · ImageNet-1K, MAE ViT-B/16 | p-value: 0.02 -> 10^-18 | Improved |
| vary decoder width, depth and heads (ablation) | DOV4MM · ImageNet-1K, MAE/CAE/iBOT | p-value | No clear effect |
| vary masking ratio from 30% to 90% (ablation) | DOV4MM · ImageNet-1K, MAE/CAE/iBOT | p-value: CAE 10^-3 -> 0.04 at 90% | No clear effect |
| fine-tune suspicious ViT-L/16 iBOT on Places365 downstream task | DOV4MM · ImageNet-1K pretrained, fine-tuned | p-value: 10^-3 -> 0.01 | No clear effect |
| add early stopping (patience=15) to suspect training | DOV4MM · ImageNet-100, MAE ViT-L/16 | p-value: 10^-6 -> 10^-4 | No clear effect |
| add adaptive-attack loss L that narrows seen/unseen reconstruction gap | DOV4MM · ImageNet-50, CAE | p-value: 10^-10 -> 10^-8 | No clear effect |
Closest papers by meaning
- Dataset Ownership Verification in Contrastive Pre-trained Models (ICLR 2025 · similarity 0.91)
- ZeroMark: Towards Dataset Ownership Verification without Disclosing Watermark (NeurIPS 2024 · similarity 0.88)
- Data Taggants: Dataset Ownership Verification Via Harmless Targeted Data Poisoning (ICLR 2025 · similarity 0.88)
- EntropyMark: Towards More Harmless Backdoor Watermark via Entropy-based Constraint for Open-source Dataset Copyright Protection (CVPR 2025 · similarity 0.85)
- Dataset Distillation for Pre-Trained Self-Supervised Vision Models (NeurIPS 2025 · similarity 0.84)
- STAMP Your Content: Proving Dataset Membership via Watermarked Rephrasings (ICML 2025 · similarity 0.84)
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.
