Research map / Probabilistic and scientific modeling / Molecules and proteins
DataGuard: A Non-intrusive Dataset Auditing Framework via Differential Information Forensics
Jiadong Lou, Wenxin Rong, Li Chen, Xing Gao, Rui Zhang, Xu Yuan
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 |
|---|---|---|---|
| use DataGuard non-intrusive auditing instead of injecting watermarks into the dataset | suspect models (ResNet-18, VGG16, DeiT) · CIFAR-10 | DBA (%): 0.00 vs 1.01 (BadNets), 9.59 (Label-Consistent) on RN-CF10 | Improved |
| audit with positive suspect model trained on target dataset | DataGuard · CIFAR-10/100, VGG/ResNet/DeiT | p-value: p-values < 10−308 for positive models; 1.0 for negative models | Improved |
| vary proportion of target dataset used by suspect model from 5% to 100% (ablation) | DataGuard · ResNet-CF100 (partial usage) | Hotelling T2 / p-value | Improved |
| dilute target dataset to 5% of suspect model's total training data (ablation) | DataGuard · ResNet-CF100 (mixed training) | Hotelling T2 / p-value | No clear effect |
| repeat training 100 times with different seeds (ablation) | DataGuard · CIFAR-10/100 stability | FDR / CDR: FDR 0%, CDR above 92% (100% on CIFAR-10/100) | No clear effect |
| stronger covariate shift, label noise or filtering in auxiliary datasets (ablation) | DataGuard · auxiliary mismatch (DeiT/Tiny-ImageNet) | FDR / CDR | Hurt |
| train suspect model with DP-SGD (epsilon 8) (ablation) | DataGuard · DP-SGD training | detection accuracy | No clear effect |
| stricter DP-SGD privacy budget (ablation) | DataGuard · DP-SGD training | CDR/FDR | Hurt |
Closest papers by meaning
- Robust ML Auditing using Prior Knowledge (ICML 2025 · similarity 0.85)
- How Well Can Differential Privacy Be Audited in One Run? (NeurIPS 2025 · similarity 0.84)
- SMI: Statistical Membership Inference for Reliable Unlearned Model Auditing (NeurIPS 2026 · similarity 0.83)
- Data Taggants: Dataset Ownership Verification Via Harmless Targeted Data Poisoning (ICLR 2025 · similarity 0.83)
- Rethinking Backdoor Attacks on Dataset Distillation: A Kernel Method Perspective (ICLR 2024 · similarity 0.83)
- Machine Unlearning Fails to Remove Data Poisoning Attacks (ICLR 2025 · similarity 0.83)
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.
