Research map / Federated learning and privacy / Privacy attacks and secure learning
Evaluating Gradient Inversion Attacks and Defenses in Federated 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 |
|---|---|---|---|
| gradient pruning p=0.999 as defense | ResNet-18 FL training · CIFAR-10 attack bs 32 | avg LPIPS: 0.45 -> 0.63 | Helped |
| gradient pruning p=0.999 | ResNet-18 · CIFAR-10 | test accuracy: 93.37 -> 83.58 | Hurt |
| gradient pruning p=0.7 as defense | ResNet-18 · CIFAR-10 attack bs 32 | avg LPIPS: 0.45 -> 0.48 | No clear effect |
| MixUp k=4 as defense | ResNet-18 · CIFAR-10 attack bs 16 | avg LPIPS: 0.45 -> 0.34 | Hurt |
| Intra-InstaHide k=4 as defense | ResNet-18 · CIFAR-10 attack bs 32 | avg LPIPS: 0.45 -> 0.69 | Helped |
| combine gradient pruning p=0.9 with Intra-InstaHide k=4 | ResNet-18 · CIFAR-10 attack bs 32 | avg LPIPS: 0.45 -> 0.73 | Helped |
| combine gradient pruning p=0.9 with Intra-InstaHide k=4 | ResNet-18 · CIFAR-10 | test accuracy: 93.37 -> 86.10 | Hurt |
| hide BatchNorm statistics from attacker | gradient inversion attack (Geiping et al.) · CIFAR-10 attack bs 32, no defense | avg LPIPS: 0.45 -> 0.48 | Helped |
| larger attack batch size (1 -> 32) | no defense · CIFAR-10 gradient inversion | avg LPIPS: 0.19 -> 0.45 | Helped |
| Intra-InstaHide encoding | FL pipeline · end-to-end single image recovery, N=5,000 | attacker GPU hours: 0.25 -> 934.48 | Helped |
Closest papers by meaning
- Foreseeing Reconstruction Quality of Gradient Inversion: An Optimization Perspective (AAAI 2024 · similarity 0.92)
- Leak and Learn: An Attacker's Cookbook to Train Using Leaked Data from Federated Learning (CVPR 2024 · similarity 0.91)
- Concealing Sensitive Samples against Gradient Leakage in Federated Learning (AAAI 2024 · similarity 0.91)
- A New Federated Learning Framework Against Gradient Inversion Attacks (AAAI 2025 · similarity 0.90)
- Gradient Inversion Attacks on Parameter-Efficient Fine-Tuning (CVPR 2025 · similarity 0.90)
- Recovering Labels from Local Updates in Federated Learning (ICML 2024 · 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.