Research map / Federated learning and privacy / Differential privacy
Pure Differential Privacy for Functional Summaries with a Laplace-like Process
Haotian Lin, Matthew Reimherr
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 Gaussian Process noise instead of ICLP noise | ICLP mechanism · Computation time for 100 samples | computation time: generating 100 Gaussian Processes is about 30% to 50% faster | Improved |
| raise cutoff K from 100 to 500 (ablation) | ICLP sampling · Computation time, Exponential kernel | seconds for 100 samples: 0.567688 -> 29.98454 | Hurt |
| select regularization by PSS instead of PCV (ablation) | ICLP-QR and FRL · Mean function privacy (S-1 to S-3) | privacy-estimation error | No clear effect |
| select regularization by PSS instead of PCV (ablation) | ICLP and FRL · Mean function privacy (S-4) | privacy-estimation error | Hurt |
| use ICLP-QR instead of FRL | ICLP mechanism · Mean function privacy | estimation error | Improved |
| use ICLP-AR (soft thresholding) instead of ICLP-QR | ICLP mechanism · Mean function privacy | privacy-estimation error | Hurt |
| relax pure DP to (epsilon, delta)-DP Gaussian mechanism | ICLP-QR · Mean function privacy | privacy-estimation error | Improved |
| increase smoothing eta in ICLP-QR (ablation) | ICLP-QR · KDE protection | smoothness vs peak bias | Mixed |
| use ICLP-QR instead of FRL | ICLP-QR · Mortality KDE, developing regions (Table 3) | expected L2-distance | Improved |
| use ICLP-QR instead of FRL | ICLP-QR · Mortality KDE, developed regions (Table 3) | expected L2-distance | Hurt |
| increase privacy budget epsilon (ablation) | FRL · Electricity demand and DTI mean function | expected L2-distance: 0.1366 -> 0.1362 | No clear effect |
Closest papers by meaning
- Elliptical Perturbations for Differential Privacy (NeurIPS 2019 · similarity 0.88)
- Benefits and Pitfalls of the Exponential Mechanism with Applications to Hilbert Spaces and Functional PCA (ICML 2019 · similarity 0.87)
- Differentially Private Distributed Data Summarization under Covariate Shift (NeurIPS 2019 · similarity 0.87)
- Differential Privacy Over Riemannian Manifolds (NeurIPS 2021 · similarity 0.87)
- Minimax Risks and Optimal Procedures for Estimation under Functional Local Differential Privacy (NeurIPS 2023 · similarity 0.86)
- Exactly Minimax-Optimal Locally Differentially Private Sampling (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.
