Research map / Label-efficient and robust learning / Domain adaptation and generalization
Evading Data Provenance in Deep Neural Networks
Hongyu Zhu, Sichu Liang, Wenwen Wang, Zhuomeng Zhang, Fangqi Li, Shi-Lin Wang
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 transfer set curation (vs random LSVRC samples) (ablation) | Escaping DOV · CIFAR-10 / Narcissus | ACC: 91.63 ± 0.49 -> 94.37 ± 0.18 | Improved |
| add transfer set curation (vs random LSVRC samples) (ablation) | Escaping DOV · CIFAR-10 / Narcissus | VSR: 7.87 ± 0.52 -> 4.59 ± 0.64 | Improved |
| use in-domain-like LSVRC curated set instead of GTSRB as transfer set (ablation) | Escaping DOV · CIFAR-10 / Narcissus | ACC: 66.80 ± 0.41 -> 94.37 ± 0.18 | Improved |
| add transfer set curation (vs GTSRB) (ablation) | Escaping DOV · CIFAR-10 / Narcissus | VSR: 8.96 ± 1.63 -> 4.59 ± 0.64 | Improved |
| add selective knowledge transfer (SKT) (ablation) | Escaping DOV · CIFAR-10 / Narcissus, varying temperature | VSR: VSR rises sharply without SKT at higher temperatures | Improved |
| raise distillation temperature (ablation) | Escaping DOV · CIFAR-10 / Narcissus | ACC: test accuracy remains nearly unchanged | No clear effect |
| replace SKT offline perturbation pool with direct UAT on student (ablation) | Escaping DOV · CIFAR-10 / Narcissus | VSR / ACC: VSR 2.07%, accuracy reduction 3.54% | Mixed |
| use curated transfer set instead of entropy-based DFND selection | Escaping DOV · CIFAR-10 / semantic backdoor | VSR: 70.2% -> 17.4% | Improved |
| add mixup in teacher training | Escaping DOV · CIFAR-10 / Narcissus | VSR: 4.59 -> 3.66 | Improved |
| add mixup in teacher training | Escaping DOV · CIFAR-10 / Isotope | p-value: 0.2845 -> 0.7013 | Improved |
| add mixup in teacher training | Escaping DOV · CIFAR-10 / Dataset Inference | p-value: 0.4759 -> 0.5196 | Improved |
| increase ADB hyperparameter mu from 0.1 to 10 | Escaping DOV · CIFAR-10 / ADB watermark | VSR: 94.42% -> 14.21% | Hurt |
Closest papers by meaning
- ZeroMark: Towards Dataset Ownership Verification without Disclosing Watermark (NeurIPS 2024 · similarity 0.89)
- EntropyMark: Towards More Harmless Backdoor Watermark via Entropy-based Constraint for Open-source Dataset Copyright Protection (CVPR 2025 · similarity 0.86)
- Data Taggants: Dataset Ownership Verification Via Harmless Targeted Data Poisoning (ICLR 2025 · similarity 0.86)
- Dataset Ownership Verification in Contrastive Pre-trained Models (ICLR 2025 · similarity 0.85)
- Tracking the Copyright of Large Vision-Language Models through Parameter Learning Adversarial Images (ICLR 2025 · similarity 0.84)
- Efficient Model Stealing Defense with Noise Transition Matrix (CVPR 2024 · 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.
