Research map / Federated learning and privacy / Privacy attacks and secure learning
Auditing Privacy Defenses in Federated Learning via Generative Gradient Leakage
Check 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 |
|---|---|---|---|
| Use GGL rather than the strongest baseline, IG | Gradient leakage reconstruction · CelebA, additive noise | PSNR: 3.1151 -> 11.0766 | Helped |
| Use GGL rather than the strongest baseline, IG | Gradient leakage reconstruction · CelebA, gradient clipping | PSNR: 12.6517 -> 11.1902 | Hurt |
| Use GGL rather than the strongest baseline, IG | Gradient leakage reconstruction · CelebA, gradient sparsification | PSNR: 5.2376 -> 11.1466 | Helped |
| Use GGL rather than the strongest baseline, IG | Gradient leakage reconstruction · CelebA, Soteria | PSNR: 4.8553 -> 10.1434 | Helped |
| Use GGL rather than the strongest baseline, IG | Gradient leakage reconstruction · ImageNet, additive noise | PSNR: 5.1120 -> 5.7089 | Helped |
| Use GGL rather than the strongest baseline, IG | Gradient leakage reconstruction · ImageNet, gradient clipping | PSNR: 8.4386 -> 6.5163 | Hurt |
| Use GGL rather than the strongest baseline, IG | Gradient leakage reconstruction · ImageNet, gradient sparsification | PSNR: 5.2376 -> 6.6920 | Helped |
| Use GGL rather than the strongest baseline, IG | Gradient leakage reconstruction · ImageNet, Soteria | PSNR: 4.3373 -> 6.0477 | Helped |
| Use gradient-free BO or CMA-ES instead of Adam (ablation) | GGL · ImageNet | PSNR: 2.4433 -> 5.7783; 2.4433 -> 5.7420 | Helped |
| Use Adam instead of BO or CMA-ES (ablation) | GGL · CelebA | PSNR: 10.9455 -> 13.6965; 11.4989 -> 13.6965 | Helped |
| Use squared L2 gradient loss with KL regularization rather than cosine loss or norm regularization (ablation) | GGL · CelebA | PSNR: 15.5405±2.7090 -> 15.6814±2.6387; 15.3471±3.1093 -> 15.6814±2.6387; 14.3249±4.1627 -> 15.6814±2.6387 | Helped |
Closest papers by meaning
- Concealing Sensitive Samples against Gradient Leakage in Federated Learning (AAAI 2024 · similarity 0.90)
- Leak and Learn: An Attacker's Cookbook to Train Using Leaked Data from Federated Learning (CVPR 2024 · similarity 0.89)
- PPIDSG: A Privacy-Preserving Image Distribution Sharing Scheme with GAN in Federated Learning (AAAI 2024 · similarity 0.88)
- One-shot Empirical Privacy Estimation for Federated Learning (ICLR 2024 · similarity 0.88)
- A New Federated Learning Framework Against Gradient Inversion Attacks (AAAI 2025 · similarity 0.88)
- Privacy without Noisy Gradients: Slicing Mechanism for Generative Model Training (NeurIPS 2024 · similarity 0.87)
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.