Research map / Federated learning and privacy / Privacy attacks and secure learning
Curation Leaks: Membership Inference Attacks against Data Curation for Machine Learning
Dariush Wahdany, Matthew Jagielski, Adam Dziedzic, Franziska Boenisch
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 deterministic voting attack instead of LiRA on scores | Image-based curation · CIFAR-10 scores | attack AUC: 0.6 -> 0.95 | Improved |
| combine LiRA and voting scores by weighted averaging (ablation) | voting attack on Image-based scores · Image-based scores | attack success | No clear effect |
| add differential privacy noise (eps=100) (ablation) | Image-based curation · CIFAR-10 | TPR at 1% FPR (%): 98.4 -> 5.4 | Improved |
| add differential privacy noise (eps=100) (ablation) | TRAK curation · CIFAR-10 | TPR at 1% FPR (%): 100 -> 33.2 | Improved |
| use stricter DP (eps=10) (ablation) | Image-based and TRAK curation · CIFAR-10 | TPR at 1% FPR (%): 1.1 and 1.7 | Improved |
| grow target dataset size (ablation) | TRAK end-to-end attack · end-to-end attack | attack success | Hurt |
| increase embedding dimensions from small to 128 (ablation) | LiRA vs Image-based curation · score attack | attack success | Mixed |
| double shadow models from 128 to 256 (ablation) | LiRA vs Image-based curation · CIFAR-10 vs PCAM | AUC: +10% (CIFAR-10), +1% (PCAM) | Mixed |
| add more shadow models (ablation) | LiRA vs TRAK curation · score attack | attack success | No clear effect |
Closest papers by meaning
- Students Parrot Their Teachers: Membership Inference on Model Distillation (NeurIPS 2023 · similarity 0.87)
- Privacy Backdoors: Enhancing Membership Inference through Poisoning Pre-trained Models (NeurIPS 2024 · similarity 0.86)
- Forget to Flourish: Leveraging Machine-Unlearning on Pretrained Language Models for Privacy Leakage (AAAI 2025 · similarity 0.85)
- Data Taggants: Dataset Ownership Verification Via Harmless Targeted Data Poisoning (ICLR 2025 · similarity 0.85)
- Assessing Per-Sample Membership Inference Vulnerability without Retraining (NeurIPS 2026 · similarity 0.85)
- Chameleon: Increasing Label-Only Membership Leakage with Adaptive Poisoning (ICLR 2024 · 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.
