Research map / Federated learning and privacy / Privacy attacks and secure learning
The Privacy Onion Effect: Memorization is Relative
Nicholas Carlini, Matthew Jagielski, Chiyuan Zhang, Nicolas Papernot, Andreas Terzis, Florian Tramer
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 |
|---|---|---|---|
| remove the 5,000 most vulnerable examples and retrain | CIFAR-10 model, LiRA · CIFAR-10 LiRA, FPR 0.01% | true-positive rate on remaining examples: 1.5% -> 0.6% (expected 0.1%) | Mixed |
| remove 5,000 hardest-to-attack or random examples instead of the most vulnerable (ablation) | CIFAR-10 model, LiRA · CIFAR-10 LiRA | ROC curve / attack success | No clear effect |
| add 5,000 random-label CIFAR-100 outliers after removing CIFAR-10 outliers (ablation) | CIFAR-10 model, LiRA · CIFAR-10 LiRA | attack success on remaining 45,000 | No clear effect |
| deduplicate CIFAR-10 (remove 5,275 duplicates) (ablation) | CIFAR-10 LiRA · CIFAR-10 LiRA | onion effect presence / attack success | Mixed |
| remove 25 highest-PrivInf points per target | second-layer target points · CIFAR-10 LiRA targets | membership inference advantage: +0.22 (56% of the 0.38 increase from removing all 5,000) | Improved |
| remove 25 highest-PrivInf points per target | random target points · CIFAR-10 LiRA targets | membership inference advantage: 0.34 -> 0.42 | Mixed |
| remove 25 highest-PrivInf points per target | initially safe target points · CIFAR-10 LiRA targets | membership inference advantage: 0.02 -> 0.14 | Improved |
Closest papers by meaning
- Reconstruction Attacks on Machine Unlearning: Simple Models are Vulnerable (NeurIPS 2024 · similarity 0.86)
- Forget to Flourish: Leveraging Machine-Unlearning on Pretrained Language Models for Privacy Leakage (AAAI 2025 · similarity 0.86)
- Causal Evaluation of Membership Inference Attacks (NeurIPS 2026 · similarity 0.85)
- Assessing Per-Sample Membership Inference Vulnerability without Retraining (NeurIPS 2026 · similarity 0.85)
- Leave-one-out Distinguishability in Machine Learning (ICLR 2024 · similarity 0.85)
- Privacy Backdoors: Enhancing Membership Inference through Poisoning Pre-trained Models (NeurIPS 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.
