Research map / Federated learning and privacy / Differential privacy
PrivSynth: Alternating and Control-Based Optimization for Privacy and Utility in Synthetic Data
Xinyuan Zhao, Hanlin Gu, Guibao Song, Gongxi Zhu, Yifei Zou, Lixin Fan, Yuxing Han
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 |
|---|---|---|---|
| apply two rounds of alternating optimization with PMP-based data selection (ablation) | PrivSynth · DomainNet | MIA ASR (%): 48.1% -> 1.9% | Improved |
| apply two rounds of alternating optimization with PMP-based data selection (ablation) | PrivSynth · DomainNet | 1-LPIPS: 0.67 -> 0.49 | Improved |
| apply two rounds of alternating optimization (ablation) | PrivSynth · DomainNet | downstream accuracy: remains above 67% | No clear effect |
| directly filter out privacy-sensitive data instead of utility-aware selection (ablation) | PrivSynth · downstream classification | utility | Hurt |
| increase ratio of selected synthetic data (ablation) | PrivSynth · downstream classification | accuracy and privacy leakage | Mixed |
| increase privacy weight phi1 (ablation) | PrivSynth · membership leakage | MIA ASR | Improved |
| increase privacy weight phi2 (ablation) | PrivSynth · pixel-level leakage | LPIPS | Improved |
| use PrivSynth selection instead of DP, Mask, Dedup on synthetic data | synthetic data generation · DomainNet and ImageNet | FID and KID | Improved |
| select data with classification loss as proxy for unknown downstream task | PrivSynth · ImageNet object detection | detection error (1-AP50) | Improved |
Closest papers by meaning
- Privacy-Preserving Data Release Leveraging Optimal Transport and Particle Gradient Descent (ICML 2024 · similarity 0.90)
- Privacy Amplification Persists under Unlimited Synthetic Data Release (NeurIPS 2026 · similarity 0.90)
- Privacy Amplification Through Synthetic Data: Insights from Linear Regression (ICML 2025 · similarity 0.89)
- CaPS: Collaborative and Private Synthetic Data Generation from Distributed Sources (ICML 2024 · similarity 0.89)
- A Privacy-Preserving Framework for Generative Model-driven Synthetic Datasets (AAAI 2025 · similarity 0.89)
- Private Evolution Converges (NeurIPS 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.
