Research map / Federated learning and privacy / Differential privacy
Cleaning the Null Space: A Privacy Mechanism for Predictors
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 Algorithm 2 (adversary-aware generalized eigenvectors) instead of Algorithm 1 (ablation) | null-space cleaning · CAL500 Nd=1 Nc=173, no attack | e_privacy: 0.845 -> 5.3E+10 | Helped |
| use Algorithm 2 instead of Algorithm 1 against the proposed retraining attack (ablation) | null-space cleaning · CAL500 Nd=1 Nc=173, with attack | e_privacy: 0.230 -> 0.005 | Hurt |
| use Algorithm 2 instead of Algorithm 1 (ablation) | null-space cleaning · wq Nd=13 Nc=1, no attack | e_privacy: 2.265 -> 1.403 | Hurt |
| use Algorithm 2 instead of Algorithm 1 (ablation) | null-space cleaning · scene Nd=1 Nc=5, no attack | e_privacy: 0.796 -> 0.796 | No clear effect |
| replace Laplace noise with null-space cleaning at equal utility error | privacy mechanism · scene Nd=1 Nc=5 | complete privacy (%): 31.5% -> 51.0% | Helped |
| replace Laplace noise with null-space cleaning at equal utility error | privacy mechanism · oes97 Nd=8 Nc=8 | complete privacy (%): 24.6% -> 43.6% | Helped |
| adversary retrains predictor on cleaned data (proposed attack) | Algorithm 1 · scene Nd=1 Nc=5 | e_privacy: 0.796 -> 0.289 | Hurt |
Closest papers by meaning
- Certification for Differentially Private Prediction in Gradient-Based Training (ICML 2025 · similarity 0.86)
- Instance-Specific Asymmetric Sensitivity in Differential Privacy (NeurIPS 2024 · similarity 0.86)
- The importance of feature preprocessing for differentially private linear optimization (ICLR 2024 · similarity 0.84)
- Private Truly-Everlasting Robust-Prediction (ICML 2024 · similarity 0.84)
- MaSS: Multi-attribute Selective Suppression for Utility-preserving Data Transformation from an Information-theoretic Perspective (ICML 2024 · similarity 0.84)
- Auditing Private Prediction (ICML 2024 · similarity 0.84)
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.