Research map / Federated learning and privacy / Differential privacy
PEARL: Data Synthesis via Private Embeddings and Adversarial Reconstruction Learning
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 |
|---|---|---|---|
| Add the adversarial critic to PEARL (ablation) | PEARL · MNIST | FID: 3.79 ± 0.06 -> 3.52 ± 0.06 | Helped |
| Add the adversarial critic to PEARL (ablation) | PEARL · MNIST | KID (× 10^3): 77.8 ± 9.88 -> 70.5 ± 10.3 | Helped |
| Add the adversarial critic to PEARL (ablation) | PEARL · Fashion-MNIST | FID: 1.99 ± 0.04 -> 1.92 ± 0.04 | Helped |
| Add the adversarial critic to PEARL (ablation) | PEARL · Fashion-MNIST | KID (× 10^3): 24.0 ± 6.90 -> 26.9 ± 6.80 | Hurt |
| Use PEARL instead of DP-MERF | DP-MERF · MNIST | FID: 49.9 ± 0.22 -> 3.52 ± 0.06 | Helped |
| Use PEARL instead of DP-MERF | DP-MERF · Fashion-MNIST | FID: 37.0 ± 0.15 -> 1.92 ± 0.04 | Helped |
| Use PEARL instead of DP-MERF | DP-MERF · Adult (average across classifiers) | ROC: 0.641 ± 0.044 -> 0.721 ± 0.035 | Helped |
| Use PEARL instead of DP-MERF | DP-MERF · Adult (average across classifiers) | PRC: 0.536 ± 0.034 -> 0.618 ± 0.033 | Helped |
| Use PEARL instead of DP-MERF | DP-MERF · Credit | ROC: 0.66 -> 0.84 | Helped |
| Use PEARL instead of DP-MERF | DP-MERF · Credit | PRC: 0.26 -> 0.68 | Helped |
Closest papers by meaning
- Privacy without Noisy Gradients: Slicing Mechanism for Generative Model Training (NeurIPS 2024 · similarity 0.89)
- Differentially Private Sum-Product Networks (ICML 2024 · similarity 0.88)
- A Privacy-Preserving Framework for Generative Model-driven Synthetic Datasets (AAAI 2025 · similarity 0.87)
- Contrastive Private Data Synthesis via Weighted Multi-PLM Fusion (ICML 2025 · similarity 0.87)
- Privacy Amplification Persists under Unlimited Synthetic Data Release (NeurIPS 2026 · similarity 0.87)
- PCEvolve: Private Contrastive Evolution for Synthetic Dataset Generation via Few-Shot Private Data and Generative APIs (ICML 2025 · similarity 0.86)
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.