Research map / Federated learning and privacy / Privacy attacks and secure learning
Inverting Gradients - How easy is it to break privacy 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 |
|---|---|---|---|
| Use cosine similarity loss and signed Adam instead of Euclidean loss with L-BFGS | Image gradient inversion · CIFAR-10 validation, trained ResNet20-4 | PSNR: 6.90 ± 2.80 -> 13.95± 3.38 | Helped |
| Use cosine similarity loss and signed Adam instead of Euclidean loss with L-BFGS | Image gradient inversion · CIFAR-10 validation, untrained LeNet (Zhu) | PSNR: 46.25 ± 12.66 -> 18.00 ± 3.33 | Hurt |
| Replace cosine similarity with L2 loss (ablation) | Proposed image reconstruction attack · CIFAR-10 validation, trained ResNet-18 | PSNR: 20.12 dB (± 1.02) -> 15.13 dB (± 0.70) | Hurt |
| Remove total variation (ablation) | Proposed image reconstruction attack · CIFAR-10 validation, trained ResNet-18 | PSNR: 20.12 dB (± 1.02) -> 19.96 dB (± 0.75) | No clear effect |
| Replace Adam with L-BFGS (ablation) | Proposed image reconstruction attack · CIFAR-10 validation, trained ResNet-18 | PSNR: 20.12 dB (± 1.02) -> 5.13 dB (± 0.50) | Hurt |
| Increase local updates to 100 steps | Federated averaging reconstruction attack · Single image, batch size 1 | PSNR | No clear effect |
| Use a learning rate of 1e-1 for local federated updates | Federated averaging reconstruction attack · Single image, batch size 1 | PSNR | Hurt |
| Average gradients over multiple images | Federated averaging · CIFAR-100, batch of 100 images | image recoverability | Mixed |
| Increase network width from 16 to 128 channels | Image gradient inversion · CIFAR-10 | PSNR: 19 to almost 23 | Helped |
| Use circular padding instead of zero-padding | Convolutional neural network · Image gradient inversion | object localization | Hurt |
Closest papers by meaning
- Gradient Inversion Attacks on Parameter-Efficient Fine-Tuning (CVPR 2025 · similarity 0.90)
- A New Federated Learning Framework Against Gradient Inversion Attacks (AAAI 2025 · similarity 0.89)
- Concealing Sensitive Samples against Gradient Leakage in Federated Learning (AAAI 2024 · similarity 0.89)
- High-Fidelity Gradient Inversion in Distributed Learning (AAAI 2024 · similarity 0.88)
- Leak and Learn: An Attacker's Cookbook to Train Using Leaked Data from Federated Learning (CVPR 2024 · similarity 0.88)
- Foreseeing Reconstruction Quality of Gradient Inversion: An Optimization Perspective (AAAI 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.