Research map / Adversarial robustness and security / Black-box and transfer attacks
MOVE: Effective and Harmless Ownership Verification via Embedded External Features
Yiming Li, Linghui Zhu, Xiaojun Jia, Yang Bai, Yong Jiang, Shu‐Tao Xia, Xiaochun Cao, Kui Ren
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 |
|---|---|---|---|
| replace white-square patch with style-transfer external feature (ablation) | MOVE (white-box) · CIFAR-10 distillation | p-value: 0.17 -> 10^{-7} | Improved |
| replace white-square patch with style-transfer external feature (ablation) | MOVE (white-box) · ImageNet distillation | p-value: 0.13 -> 10^{-5} | Improved |
| add meta-classifier to BadNets verification (ablation) | BadNets watermarking · CIFAR-10 and ImageNet | p-value | Improved |
| increase transformation rate gamma (ablation) | MOVE (white-box) · CIFAR-10 | p-value | Improved |
| increase transformation rate gamma (ablation) | MOVE (black-box) · CIFAR-10 | p-value | Mixed |
| increase number of sampled images m (ablation) | MOVE · CIFAR-10 | p-value | Improved |
| use a different style image (Pattern a vs b) (ablation) | MOVE · CIFAR-10 | p-value: Fine-tuning 10^{-7} -> 10^{-9} | No clear effect |
| use multiple stealing stages | MOVE (white-box) · CIFAR-10 multi-stage | p-value: 10^{-7} -> 10^{-5} -> 10^{-4} | Hurt |
| use harder multi-stage attack (Logit-query then Zero-shot) | MOVE (white-box) · CIFAR-10 multi-stage | p-value: 10^{-4} -> 0.01 | Hurt |
| distill the watermarked model | BadNets watermark · CIFAR-10 | ASR: 100.00 -> 3.84 | Hurt |
| train on half-split different data with different structure | Dataset Inference · CIFAR-10 | p-value: 10^{-7} | Hurt |
| apply MOVE style transformation to training data | victim model · CIFAR-10 | accuracy: 91.79 -> 91.99 | No clear effect |
Closest papers by meaning
- Defending against Model Stealing via Verifying Embedded External Features (AAAI 2022 · similarity 0.92)
- Dataset Inference: Ownership Resolution in Machine Learning (ICLR 2021 · similarity 0.88)
- Efficient Model Stealing Defense with Noise Transition Matrix (CVPR 2024 · similarity 0.87)
- Stealix: Model Stealing via Prompt Evolution (ICML 2025 · similarity 0.86)
- ZeroMark: Towards Dataset Ownership Verification without Disclosing Watermark (NeurIPS 2024 · similarity 0.85)
- GraphIP–Bench: How Hard Is It to Steal a Graph Neural Network, and Can We Stop It? (NeurIPS 2026 · similarity 0.85)
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.
