Research map / Federated learning and privacy / Differential privacy
Private Synthetic Data for Multitask Learning and Marginal Queries
Giuseppe Vietri, Cedric Archambeau, Sergul Aydore, William Brown, Michael Kearns, Aaron Roth, Ankit Siva, Shuai Tang, Steven Wu
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 |
|---|---|---|---|
| Anneal sigmoid inverse temperature instead of fixed sigma (ablation) | RAP++ · Mixed-marginal queries on ACS data | max and mean query error | Improved |
| Use native handling of numeric features with threshold queries instead of binning (RAP++ vs PGM) | RAP++ · Mixed marginal queries, ACS 25 datasets | average query error | Improved |
| RAP++ vs PGM on categorical marginal queries | RAP++ · Categorical marginal queries, ACS | average query error | Improved |
| RAP++ vs PGM on single-task ML evaluation | RAP++ · ACS single-task, logistic regression | macro F1 | Improved |
| RAP++ vs PGM on multitask income task | RAP++ · ACS multitask, Income | macro F1 | Improved |
| RAP++ vs PGM on coverage tasks at small epsilon | RAP++ · ACS multitask, Coverage | macro F1 | Mixed |
| RAP++ vs PGM when numerical features are uninformative | RAP++ · ACS multitask, Employment | macro F1 | Hurt |
| Add numerical features to categorical features for Income prediction | Logistic regression on real data · ACS NY multitask, Income | macro F1: 0.77 -> 0.93 | Improved |
| Add numerical features for Coverage prediction | Logistic regression on real data · ACS NY multitask, Coverage | macro F1: 0.77 -> 0.82 | Improved |
| Add numerical features for Employment, Travel time, Mobility | Logistic regression on real data · ACS NY multitask | macro F1: 0.78 -> 0.79 | No clear effect |
| Vary RAP++ linear-threshold epochs T_LT over 10, 30, 50, 70 | RAP++ · ACS single-task, F1 vs runtime | macro F1 | No clear effect |
| Vary PGM bin size over 10, 20, 50, 100 | PGM · ACS single-task, F1 vs runtime | macro F1 | Mixed |
Closest papers by meaning
- Private Evolution Converges (NeurIPS 2025 · similarity 0.85)
- Maximizing the Potential of Synthetic Data: Insights from Random Matrix Theory (ICLR 2025 · similarity 0.85)
- Differentially Private Sum-Product Networks (ICML 2024 · similarity 0.84)
- Bounding the Excess Risk for Linear Models Trained on Marginal-Preserving, Differentially-Private, Synthetic Data (ICML 2024 · similarity 0.84)
- Does Training with Synthetic Data Truly Protect Privacy? (ICLR 2025 · similarity 0.84)
- Efficiently Computing Similarities to Private Datasets (ICLR 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.
