Research map / Federated learning and privacy / Differential privacy
SMOTE and Mirrors: Exposing Privacy Leakage from Synthetic Minority Oversampling
Georgi Ganev, MohammadReza Nazari, Rees Davison, Amirhassan Fallah Dizche, XINMIN WU, Ralph Abbey, Jorge G Silva, Emiliano De Cristofaro
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 naive distinguishing/DCR-style linkability to assess SMOTE privacy | SMOTE augmented/synthetic data · 8 imbalanced datasets | attack success: naive 0.01 precision vs DistinSMOTE 1.00 | Hurt |
| apply SMOTE augmentation before training classifiers (vs real data only) (ablation) | MIA target vulnerability · 8 imbalanced datasets | MIA AUC: 0.58 -> 0.68 (17% rise) | Hurt |
| higher imbalance ratio r (ablation) | ReconSMOTE · 8 imbalanced datasets, synthetic data | recall: 0.43 (r=8.6) -> 1.00 (r>=26) | Improved |
| reduce SMOTE neighbors from k=5 to k=2 (ablation) | ReconSMOTE · 8 imbalanced datasets | recall: 0.85 -> 0.52 | Hurt |
| reduce SMOTE neighbors from k=5 to k=2 (ablation) | DistinSMOTE · 8 imbalanced datasets | precision/recall: 1.00 -> 1.00 | No clear effect |
| use SMOTE-NC with mixed-type data, ignoring categorical features (ablation) | ReconSMOTE · cardio, churn | recall: 1.00 -> 0.98 | No clear effect |
| add small noise (<=1e-7) to SMOTE interpolation (ablation) | DistinSMOTE/ReconSMOTE · yeast_me2 | precision: 1.00 -> 0.96 (Recon, 1e-7) | No clear effect |
| add large noise (1e-3) to SMOTE interpolation (ablation) | ReconSMOTE · yeast_me2 | precision/recall: 1.00 -> 0.00 | Hurt |
| match attack tolerance to injected noise level (ablation) | DistinSMOTE/ReconSMOTE · yeast_me2 | precision: 0.07 vs 0.04 at noise 1e-5 (Distin) | Improved |
Closest papers by meaning
- Differential Privacy Under Class Imbalance: Methods and Empirical Insights (ICML 2025 · similarity 0.88)
- Privacy Amplification Through Synthetic Data: Insights from Linear Regression (ICML 2025 · similarity 0.87)
- Privacy Amplification Persists under Unlimited Synthetic Data Release (NeurIPS 2026 · similarity 0.87)
- Underestimated Privacy Risks for Minority Populations in Large Language Model Unlearning (ICML 2025 · similarity 0.87)
- The Canary’s Echo: Auditing Privacy Risks of LLM-Generated Synthetic Text (ICML 2025 · similarity 0.87)
- Concentration and excess risk bounds for imbalanced classification with synthetic oversampling (NeurIPS 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.
