Research map / Federated learning and privacy / Privacy attacks and secure learning
Diffusion to Obfuscation: Time-Adaptive Synthesized Generation Against Gradient Leakage Attacks in Federated Learning
Farchan Raswa, Chun-Shien Lu, Jia-Ching Wang
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 Diff2Obs-I instead of no defense against GI-NAS | Diff2Obs · ImageNet, GI-NAS | LPIPS: 0.292 -> 0.838 | Improved |
| use Diff2Obs instead of CENSOR to keep accuracy | Diff2Obs · CIFAR-10 test accuracy | test accuracy (%): 25.79 -> 75.48 | Improved |
| use Diff2Obs instead of Refiner | Diff2Obs · CIFAR-10 overhead | time to best accuracy (hour): 0.79 -> 0.05 | Improved |
| Diff2Obs-I at larger batch size 4 to 32 (ablation) | Diff2Obs (Obs-I) · ImageNet HFGradInv | LPIPS: 0.834 -> 0.960 | Improved |
| Diff2Obs-I instead of CENSOR at batch size 4 under GIFD | Diff2Obs (Obs-I) · ImageNet GIFD | LPIPS: 0.878 -> 0.883 | Improved |
| adversary adds EoT adaptive attack against Diff2Obs (ablation) | Diff2Obs · ImageNet GIFD | LPIPS: 0.844 -> 0.829 | No clear effect |
| adversary adds EoT adaptive attack against Diff2Obs (ablation) | Diff2Obs · CIFAR-10 GIFD | LPIPS: 0.742 -> 0.761 | No clear effect |
| move from IID to non-IID Dirichlet alpha=1 (ablation) | Diff2Obs · CIFAR-10 | PMM: 98.10% -> 98.86% | No clear effect |
| increase client samples for DDPM from 50 to 1000 (ablation) | Diff2Obs · CIFAR-10 | test accuracy (%): 69.50 -> 75.55 | Improved |
| increase client samples for DDPM from 50 to 1000 (ablation) | Diff2Obs · CIFAR-10 HFGradInv | LPIPS: 0.835 -> 0.842 | No clear effect |
| replace client DDPM by pre-trained Stable Diffusion (ablation) | Diff2Obs · CIFAR-10 | test accuracy (%): 75.48 -> 69.20 | Hurt |
| replace DDPM trained on 500 samples by Stable Diffusion in FID (ablation) | Diff2Obs · CIFAR-10 | FID: 24.30 -> 57.30 | Hurt |
Closest papers by meaning
- Concealing Sensitive Samples against Gradient Leakage in Federated Learning (AAAI 2024 · similarity 0.89)
- Leak and Learn: An Attacker's Cookbook to Train Using Leaked Data from Federated Learning (CVPR 2024 · similarity 0.88)
- Mjölnir: Breaking the Shield of Perturbation-Protected Gradients via Adaptive Diffusion (AAAI 2025 · similarity 0.86)
- DataStealing: Steal Data from Diffusion Models in Federated Learning with Multiple Trojans (NeurIPS 2024 · similarity 0.86)
- Gradient Inversion Attacks on Parameter-Efficient Fine-Tuning (CVPR 2025 · similarity 0.86)
- PID: Prompt-Independent Data Protection Against Latent Diffusion Models (ICML 2024 · similarity 0.86)
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.
