Research map / Federated learning and privacy / Differential privacy
Functional Renyi Differential Privacy for Generative Modeling
Dihong Jiang, Sun Sun, Yaoliang Yu
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 f-RDP kernel method instead of SoTA DPDM at eps=0.2 | DP-kernel (conditional) · MNIST eps=0.2 | FID: 61.9 -> 26.5 | Improved |
| use f-RDP kernel method instead of SoTA DPDM at eps=0.2 | DP-kernel (conditional) · FMNIST eps=0.2 | FID: 78.4 -> 46.2 | Improved |
| use f-RDP kernel method instead of DP-MERF at eps=0.2 | DP-kernel (conditional) · CelebA eps=0.2 | FID: 264.8 -> 85.9 | Improved |
| use f-RDP kernel method instead of DP-MERF at eps=0.2 | DP-kernel (conditional) · CelebA eps=0.2 | accuracy: 52.7 -> 86.5 | Improved |
| use f-RDP kernel method instead of DPDM at eps=0.2 | DP-kernel (conditional) · MNIST eps=0.2 | accuracy: 71.9 -> 91.3 | Improved |
| compare with DPDM at eps=1 | DP-kernel (conditional) · MNIST eps=1 | FID: 23.4 -> 29.5 | No clear effect |
| parallel per-class sub-models instead of conditional generator (ablation) | DP-kernel · MNIST eps=1 | FID: 29.5 -> 21.8 | Improved |
| parallel per-class sub-models instead of conditional generator (ablation) | DP-kernel · MNIST eps=0.2 | FID: 26.5 -> 37.3 | Hurt |
| parallel per-class sub-models instead of conditional generator (ablation) | DP-kernel · CelebA eps=0.2 | FID: 85.9 -> 95.8 | Hurt |
| parallel per-class sub-models instead of conditional generator (ablation) | DP-kernel · MNIST eps=0.2 | accuracy: 91.3 -> 93.7 | Improved |
| parallel per-class sub-models instead of conditional generator (ablation) | DP-kernel · CelebA eps=10 | FID: 77.2 -> 78.5 | No clear effect |
| lower privacy budget eps from 1 to 0.2 | DP-kernel (conditional) · MNIST | accuracy: 93.4 -> 91.3 | Hurt |
Closest papers by meaning
- Privacy without Noisy Gradients: Slicing Mechanism for Generative Model Training (NeurIPS 2024 · similarity 0.90)
- The Gaussian Mixing Mechanism: Renyi Differential Privacy via Gaussian Sketches (NeurIPS 2025 · similarity 0.88)
- Exactly Minimax-Optimal Locally Differentially Private Sampling (NeurIPS 2024 · similarity 0.88)
- 𝑓-Differential Privacy Filters: Validity and Approximate Solutions (NeurIPS 2026 · similarity 0.88)
- Privacy Amplification Persists under Unlimited Synthetic Data Release (NeurIPS 2026 · similarity 0.88)
- Differentially Private Stochastic Gradient Descent with Fixed-Size Minibatches: Tighter RDP Guarantees with or without Replacement (NeurIPS 2024 · similarity 0.88)
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.
