Research map / Federated learning and privacy / Differential privacy
Iterative Methods for Private Synthetic Data: Unifying Framework and New Methods
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 |
|---|---|---|---|
| replace MWEM with PEP (adaptively reuse past query measurements) | MWEM · ACS (reduced) and ADULT (reduced), 3-way marginals | max error | Helped |
| use GEM (generator network with exponential mechanism) instead of RAP-softmax | RAP-softmax · ACS and ADULT, 3-way marginals, high dimension | max error | Helped |
| pretrain generator on public data CA-18 (GEM^Pub) vs GEM without public data (ablation) | GEM · ACS PA-18, 3-way marginals | max error | Helped |
| GEM^Pub vs PMW^Pub with insufficient-support public data CA-18 | PMW^Pub · ACS PA-18 with CA-18 public data, eps=1 | max error: 9.23x lower | Helped |
| GEM^Pub vs PMW^Pub with sufficient-support public data PA-10 | PMW^Pub · ACS PA-18 with PA-10 public data | max error | No clear effect |
| pretrain on public data with only 7 of 13 attributes (GEM^Pub-reduced) (ablation) | GEM · ADULT, 3-way marginals | max error | Helped |
| select whole marginal workload with exponential mechanism and measure all its queries (marginal trick) (ablation) | GEM · k-way marginals | max error | Helped |
Closest papers by meaning
- Private Evolution Converges (NeurIPS 2025 · similarity 0.90)
- Differentially Private Sum-Product Networks (ICML 2024 · similarity 0.89)
- Synthesizing Privacy-Preserving Text Data via Finetuning *without* Finetuning Billion-Scale LLMs (ICML 2025 · similarity 0.89)
- Privacy Amplification Persists under Unlimited Synthetic Data Release (NeurIPS 2026 · similarity 0.89)
- PCEvolve: Private Contrastive Evolution for Synthetic Dataset Generation via Few-Shot Private Data and Generative APIs (ICML 2025 · similarity 0.89)
- A Privacy-Preserving Framework for Generative Model-driven Synthetic Datasets (AAAI 2025 · similarity 0.89)
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.