Research map / Federated learning and privacy / Privacy attacks and secure learning
Geminio: Language-Guided Gradient Inversion Attacks in Federated Learning
Junjie Shan, Ziqi Zhao, Jialin Lu, Rui Zhang, Siu Ming Yiu, Ka-Ho Chow
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 |
|---|---|---|---|
| use 5% instead of 20% auxiliary data (ablation) | Geminio · CIFAR-20 | attack F-1 (%): 68.13 -> 60.37 | Hurt |
| use ImageNet as auxiliary data instead of CIFAR-20 train subset (ablation) | Geminio · CIFAR-20 | attack F-1 (%): 68.13 -> 64.37 | Hurt |
| use Caltech256 as auxiliary data instead of CIFAR-20 train subset (ablation) | Geminio · CIFAR-20 | attack F-1 (%): 68.13 -> 50.48 | Hurt |
| use SigLIP instead of CLIP (ablation) | Geminio · CIFAR-20 | attack F-1 (%): 68.13 -> 69.54 | No clear effect |
| add Geminio malicious model to DLG/InvertingGrad/HFGradInv reconstruction | gradient inversion attacks · ImageNet batch 64, query 'Any weapon?' | LPIPS | Improved |
| add Geminio to label inference attacks | five breaching label inference attacks · CIFAR-20 | label inference accuracy | Improved |
| apply gradient pruning defense up to 95% (ablation) | Geminio · CIFAR-20 | LPIPS | No clear effect |
| apply gradient pruning at 99% (ablation) | Geminio · CIFAR-20 | LPIPS | Hurt |
| add Laplacian noise scale 0.10 (ablation) | Geminio · CIFAR-20 | LPIPS | No clear effect |
| add Laplacian noise scale 0.50 (ablation) | Geminio · CIFAR-20 | LPIPS | Hurt |
| more local epochs under FedAvg (ablation) | Geminio · ImageNet FedAvg | attack recall | Hurt |
| use batch size 8 instead of 64 for training malicious model (ablation) | Geminio · CIFAR-20 | attack recall | Hurt |
Closest papers by meaning
- Gradient Inversion of Multimodal Models (ICML 2025 · similarity 0.88)
- DAGER: Exact Gradient Inversion for Large Language Models (NeurIPS 2024 · similarity 0.88)
- Leveraging Soft Prompts for Privacy Attacks in Federated Prompt Tuning (NeurIPS 2026 · similarity 0.87)
- High-Fidelity Gradient Inversion in Distributed Learning (AAAI 2024 · similarity 0.87)
- Leak and Learn: An Attacker's Cookbook to Train Using Leaked Data from Federated Learning (CVPR 2024 · similarity 0.87)
- Concealing Sensitive Samples against Gradient Leakage in Federated 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.
