Research map / Federated learning and privacy / Privacy attacks and secure learning
Gradient Inversion with Generative Image Prior
Jinwoo Jeon, Jaechang Kim, Kangwook Lee, Sewoong Oh, Jungseul Ok
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 |
|---|---|---|---|
| add generative prior search (GI-x to GI-z/w) | GI-x · ImageNet 64x64 val, ResNet18, B=4 | PSNR: 13.78 -> 15.58 | Improved |
| add generative prior search (GI-x+BN to GI-z/w+BN) | GI-x+BN · ImageNet 64x64 val, ResNet18, B=4 | PSNR: 15.52 -> 16.31 | Improved |
| add generative prior search (GI-x to GI-z/w) | GI-x · ImageNet 64x64 val, ResNet18, B=4 | LPIPS: 0.4376 -> 0.3023 | Improved |
| add parameter space search after latent search (GI-z to GI-z/w) (ablation) | GIAS · ImageNet 64x64 val, ResNet18, B=4 | PSNR: 14.27 -> 15.58 | Improved |
| add latent space search before parameter search (GI-w to GI-z/w) (ablation) | GIAS · ImageNet 64x64 val, ResNet18, B=4 | PSNR: 14.70 -> 15.58 | Improved |
| optimize only parameters of untrained generator instead of pixels (GI-x to GI-w) (ablation) | GI-x · ImageNet 64x64 val, ResNet18, B=4 | PSNR: 13.78 -> 14.70 | Improved |
| optimize only parameters of untrained generator instead of pixels (GI-x to GI-w) (ablation) | GI-x · ImageNet 64x64 val, ResNet18, B=4 | LPIPS: 0.4376 -> 0.5121 | Hurt |
| GIAS without BN versus GI-x with BN statistics | GIAS · ImageNet 64x64 val, ResNet18, B=4 | PSNR: 15.58 vs 15.52 | No clear effect |
| use latent search then parameter search instead of latent then input pixels (GI-z/x) (ablation) | GI-z · ImageNet 64x64 val, ResNet18, B=4 | visual quality | Improved |
| add Gaussian gradient noise 0.01 | GIAS · ImageNet 64x64 val, ResNet18, B=4 | PSNR | No clear effect |
| add large Gaussian gradient noise 0.1 | GIAS · ImageNet 64x64 val, ResNet18, B=4 | PSNR | Hurt |
| increase batch size to 32 | GIAS · ImageNet 64x64 val, ResNet18, B=4 | PSNR | Hurt |
Closest papers by meaning
- A New Federated Learning Framework Against Gradient Inversion Attacks (AAAI 2025 · similarity 0.89)
- Gradient Inversion Attacks on Parameter-Efficient Fine-Tuning (CVPR 2025 · similarity 0.89)
- Concealing Sensitive Samples against Gradient Leakage in Federated Learning (AAAI 2024 · similarity 0.88)
- Gradient Inversion of Multimodal Models (ICML 2025 · similarity 0.88)
- Federated Learning via Input-Output Collaborative Distillation (AAAI 2024 · similarity 0.87)
- High-Fidelity Gradient Inversion in Distributed Learning (AAAI 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.
