Research map / Federated learning and privacy / Privacy attacks and secure learning
GIFD: A Generative Gradient Inversion Method with Feature Domain Optimization
Hao Fang, Bin Chen, Xuan Wang, Zhi Wang, Shu-Tao Xia
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 |
|---|---|---|---|
| Search intermediate feature domains after the latent space instead of latent space only (ablation) | GIFD · ImageNet (BigGAN) | PSNR: 13.9451 -> 18.6457 | Improved |
| Select output from the layer with least gradient matching loss instead of last searched layer (ablation) | GIFD · ImageNet (BigGAN) | PSNR: 18.6457 -> 19.4662 | Improved |
| Add l1 ball constraint on intermediate features (ablation) | GIFD · ImageNet (BigGAN) | PSNR: 19.4662 -> 20.0534 | Improved |
| Add l1 ball constraint on intermediate features (ablation) | GIFD · FFHQ (StyleGAN2) | PSNR: 20.5839 -> 21.3368 | Improved |
| Optimize intermediate features rather than GAN parameters (GIAS) | GAN-based gradient inversion · ImageNet (BigGAN) | PSNR: 17.4923 -> 20.0534 | Improved |
| Latent-only GAN search (GGL) versus GAN-free methods on ImageNet | GGL · ImageNet (BigGAN) | PSNR: 13.3885 vs 17.0756 (IG) | Hurt |
| Use GIFD under Gaussian noise defense on FFHQ | GIFD · FFHQ, Gaussian noise std 0.1 | PSNR: 13.7118 vs GGL 14.8982 | Hurt |
| Use GIFD on out-of-distribution data versus GIAS | GIFD · ImageNet* Cartoon | PSNR: 19.0247 -> 22.8055 | Improved |
| Increase batch size from 1 to 32 (ablation) | GIFD · ImageNet | PSNR: 20.6217 -> 13.8106 | Hurt |
| Use GIFD instead of GIAS for speed | GIFD · ImageNet, B=4 | convergence time (s): 1360.7808 -> 907.727 | Improved |
Later papers that compared with this method
1 later accepted paper reports GIFD as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Enhanced Privacy Leakage from Noise-Perturbed Gradients via Gradient-Guided Conditional Diffusion Models (AAAI 2026 · better on 2 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Gradient Inversion Attacks on Parameter-Efficient Fine-Tuning (CVPR 2025 · similarity 0.87)
- SPEAR: Exact Gradient Inversion of Batches in Federated Learning (NeurIPS 2024 · similarity 0.87)
- GRAIN: Exact Graph Reconstruction from Gradients (ICLR 2025 · similarity 0.86)
- Concealing Sensitive Samples against Gradient Leakage in Federated Learning (AAAI 2024 · similarity 0.86)
- Gradient Inversion of Multimodal Models (ICML 2025 · similarity 0.86)
- High-Fidelity Gradient Inversion in Distributed Learning (AAAI 2024 · similarity 0.85)
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.
