Research map / Learning theory and algorithms / Matrix and tensor methods
An Optimal and Scalable Matrix Mechanism for Noisy Marginals under Convex Loss Functions
Yingtai Xiao, Guanlin He, Danfeng Zhang, Daniel Kifer
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 ResidualPlanner instead of HDMM for selection (RMSE objective) | HDMM · Synth-nd, d=15 | selection time: 201.485 ± 13.697 sec -> 0.030 ± 0.017 sec | Improved |
| use ResidualPlanner instead of HDMM for selection (RMSE objective) | HDMM · Synth-nd, d=10 | selection time: 0.639 ± 0.059 sec -> 0.009 ± 0.001 sec | Improved |
| use ResidualPlanner where HDMM runs out of memory | HDMM · Synth-nd, d=20 to 100 | selection time: Out of memory -> 9.913 ± 0.246 sec | Improved |
| optimize max variance instead of RMSE objective in ResidualPlanner | ResidualPlanner · Synth-nd, d=100 | selection time: 9.913 ± 0.246 sec -> 121.224 ± 3.008 sec | Hurt |
| optimize max variance instead of RMSE objective in ResidualPlanner | ResidualPlanner · Synth-nd, d=2 | selection time: 0.001 ± 0.0008 sec -> 0.007 ± 0.001 sec | No clear effect |
| use ResidualPlanner instead of HDMM + PGM for reconstruction | HDMM + PGM · Synth-nd, d=15 | reconstruction time: 91.101 ± 7.621 sec -> 0.413 ± 0.006 sec | Improved |
| use ResidualPlanner instead of HDMM for reconstruction | HDMM · Synth-nd, d=10 | reconstruction time: Out of memory in reconstruction -> 0.125 ± 0.032 sec | Improved |
| use ResidualPlanner instead of HDMM at very small d | HDMM · Synth-nd, d=2 | reconstruction time: 0.003 ± 0.0006 sec -> 0.005 ± 0.003 sec | Hurt |
| optimize max variance directly (ResidualPlanner) instead of RMSE-optimal HDMM | HDMM · Adult, 3-way marginals | max variance: 5675.238 -> 236.843 | Improved |
| optimize max variance directly (ResidualPlanner) instead of RMSE-optimal HDMM | HDMM · CPS, <=3-way marginals | max variance: 415.073 -> 13.216 | Improved |
| optimize max variance directly (ResidualPlanner) instead of RMSE-optimal HDMM | HDMM · Loans, <=3-way marginals | max variance: 4317.709 -> 180.817 | Improved |
| optimize max variance directly (ResidualPlanner) instead of RMSE-optimal HDMM | HDMM · Adult, 1-way marginals | max variance: 41.772 -> 12.047 | Improved |
Closest papers by meaning
- Efficient and Private Marginal Reconstruction with Local Non-Negativity (NeurIPS 2024 · similarity 0.86)
- On Inherent Privacy of Posterior Sampling: A Unified R\'enyi-Divergence Framework (NeurIPS 2026 · similarity 0.85)
- Information-Computation Tradeoffs for Noiseless Linear Regression with Oblivious Contamination (NeurIPS 2025 · similarity 0.85)
- Perturb-and-Project: Differentially Private Similarities and Marginals (ICML 2024 · similarity 0.85)
- Privacy Guarantees in Posterior Sampling under Contamination (NeurIPS 2026 · similarity 0.85)
- Differential Private Stochastic Optimization with Heavy-tailed Data: Towards Optimal Rates (AAAI 2025 · 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.
