Research map / Adversarial robustness and security / Attacks on text and graphs
Dataset Inference: Ownership Resolution in Machine Learning
Pratyush Maini, Mohammad Yaghini, Nicolas Papernot
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 Blind Walk instead of MinGD for embedding generation | Dataset inference · CIFAR10, data distillation | effect size (Δμ): 0.586 -> 0.778 | Improved |
| Use Blind Walk instead of MinGD for embedding generation | Dataset inference · CIFAR10, different architecture | effect size (Δμ): 0.645 -> 1.400 | Improved |
| Use Blind Walk instead of MinGD for embedding generation | Dataset inference · CIFAR10, zero-shot learning | effect size (Δμ): 0.371 -> 0.406 | Improved |
| Use Blind Walk instead of MinGD for embedding generation | Dataset inference · CIFAR10, fine-tuning | effect size (Δμ): 0.832 -> 1.839 | Improved |
| Use Blind Walk instead of MinGD for embedding generation | Dataset inference · CIFAR10, label-query extraction | effect size (Δμ): 0.475 -> 1.006 | Improved |
| Use Blind Walk instead of MinGD for embedding generation | Dataset inference · CIFAR10, logit-query extraction | effect size (Δμ): 0.563 -> 1.048 | Improved |
| Use Blind Walk instead of MinGD for embedding generation | Dataset inference · CIFAR100, data distillation | effect size (Δμ): 0.362 -> 1.098 | Improved |
| Use Blind Walk instead of MinGD for embedding generation | Dataset inference · CIFAR100, different architecture | effect size (Δμ): 1.016 -> 1.471 | Improved |
| Use Blind Walk instead of MinGD for embedding generation | Dataset inference · CIFAR100, zero-shot learning | effect size (Δμ): 0.466 -> 0.405 | Mixed |
| Use Blind Walk instead of MinGD for embedding generation | Dataset inference · CIFAR100, fine-tuning | effect size (Δμ): 1.047 -> 1.423 | Improved |
| Use Blind Walk instead of MinGD for embedding generation | Dataset inference · CIFAR100, label-query extraction | effect size (Δμ): 0.270 -> 0.107 | Mixed |
| Use Blind Walk instead of MinGD for embedding generation | Dataset inference · CIFAR100, logit-query extraction | effect size (Δμ): 0.385 -> 0.184 | Mixed |
Closest papers by meaning
- Reconstruction Attacks on Machine Unlearning: Simple Models are Vulnerable (NeurIPS 2024 · similarity 0.87)
- Stealix: Model Stealing via Prompt Evolution (ICML 2025 · similarity 0.87)
- Data Taggants: Dataset Ownership Verification Via Harmless Targeted Data Poisoning (ICLR 2025 · similarity 0.87)
- Dataset Ownership Verification in Contrastive Pre-trained Models (ICLR 2025 · similarity 0.87)
- Efficient Model Stealing Defense with Noise Transition Matrix (CVPR 2024 · similarity 0.86)
- Exploring Query Efficient Data Generation Towards Data-Free Model Stealing in Hard Label Setting (AAAI 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.
