Research map / Federated learning and privacy / Differential privacy
Best Linear Unbiased Estimate from Privatized Contingency Tables
Jordan Awan, Adam Edwards, Paul Bartholomew, Andrew Sillers
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 SEA BLUE instead of full matrix projection | Census contingency table post-processing · 5x5 table | runtime (s): 60.45 -> 0.90 | Improved |
| use SEA BLUE instead of full matrix projection | Census contingency table post-processing · 5x5 table | memory (MiB): 947.54 -> 4.15 | Improved |
| use SEA BLUE instead of full matrix projection on PL94 block | PL94 demonstration product · PL94 Block | runtime (s): 70.55 -> 0.49 | Improved |
| apply SEA BLUE to a 6x6 table where projection runs out of memory | SEA BLUE · 6x6 table | memory (MiB): 30.98 | Improved |
| violate equal-variance assumption A3 within more tables (All Counts 1 Var) (ablation) | SEA BLUE · 4x4 table, unequal variances | MSE to true counts: 0.7932 vs projection 0.7456 | Hurt |
| violate equal variance in only one marginal table (ablation) | SEA BLUE · 4x4 table, unequal variances | MSE to true counts: 0.8132 vs 0.8131 | No clear effect |
| increase Monte Carlo replicates R from 19 to 99 to 199 for T-based interval (ablation) | Monte Carlo T confidence interval · normal data, 100,000 replicates | relative width vs exact z interval: 1.055 -> 1.010 -> 1.005 | Improved |
| increase Monte Carlo replicates R from 19 to 99 to 199 for distribution-free interval (ablation) | Monte Carlo DF confidence interval · normal data, 100,000 replicates | relative width vs exact z interval: 1.094 -> 1.017 -> 1.009 | Improved |
| clip confidence intervals to non-negative integers (ablation) | SEA BLUE exact confidence interval · PL94 Block (supplied tables) | CI width: 19.342 -> 9.319 | Improved |
| use distribution-free Monte Carlo interval instead of T-based interval | SEA BLUE confidence interval · PL94 Multi | empirical coverage: 0.9490 -> 0.9733 | Mixed |
| apply SEA BLUE estimates to DHC raw intervals | SEA BLUE exact confidence interval · DHC State (supplied tables) | raw CI width: 13.044 -> 12.887 | No clear effect |
Closest papers by meaning
- A Simple and Practical Method for Reducing the Disparate Impact of Differential Privacy (AAAI 2024 · similarity 0.86)
- Fairness Issues and Mitigations in (Differentially Private) Socio-Demographic Data Processes (AAAI 2025 · similarity 0.85)
- Analysis of Differentially Private Synthetic Data: A Measurement Error Approach (AAAI 2024 · similarity 0.85)
- Private Statistical Estimation via Truncation (NeurIPS 2025 · similarity 0.83)
- On Inherent Privacy of Posterior Sampling: A Unified R\'enyi-Divergence Framework (NeurIPS 2026 · similarity 0.83)
- Optimal Survey Design for Private Mean Estimation (ICML 2025 · similarity 0.83)
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.
