Research map / Federated learning and privacy / Differential privacy
Venom: Liquid Diffusion-Guided Gradient Inversion for Breaking Differential Privacy in Federated Learning
Bin Hu, Jingling Yuan, Jiawei Jiang, Chuang Hu
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 liquid Diff-LRN stage on top of SPE with deterministic reconstruction (ablation) | Venom-Diff · MNIST (eps=5) | PSNR (dB): 27.76 -> 28.75 | Improved |
| add liquid Diff-LRN stage (ablation) | Venom-Diff · ImageNet (eps=10) | PSNR (dB): 23.61 -> 24.32 | Improved |
| add liquid Diff-LRN stage (ablation) | Venom-Diff · CIFAR-10 (eps=1) | ASR@0.5: 70% -> 72% | Improved |
| add liquid Diff-LRN stage (ablation) | Venom-Diff · ImageNet (eps=1) | ASR@0.5: 8% -> 10% | Improved |
| add liquid Diff-LRN stage (ablation) | Venom-Diff · MNIST (eps=10) | ASR@0.5: 98% -> 99% | No clear effect |
| add liquid Diff-LRN stage (ablation) | Venom-Diff · all datasets | ASR@0.5: 2-5% ASR improvement | Improved |
| add liquid Diff-LRN stage (ablation) | Venom-Diff · ImageNet runtime (ResNet) | runtime per image (s): 0.88 vs 1.32 (lower is faster) | Hurt |
| replace Mjölnir iterative diffusion with Venom single-step pipeline | Mjölnir · PSNR across datasets | PSNR (dB): 3.4 dB (MNIST, eps = 1) to 11.7 dB (ImageNet, eps = 10) | Improved |
| replace Mjölnir with Venom | Mjölnir · ImageNet (eps=10) | LPIPS: 0.632 -> 0.340 | Improved |
| replace Mjölnir with Venom | Mjölnir · ImageNet (eps=10) | ASR@0.5: 2% -> 45% | Improved |
| replace iterative gradient inversion with analytical SPE plus single-step Diff-LRN | Mjölnir · ResNet ImageNet | runtime: 9.4 hours to 0.88 seconds | Improved |
| replace Mjölnir with Venom | Mjölnir · MNIST (eps=10) | PSNR (dB): 33.09 -> 33.38 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Mjölnir — Mjölnir: Breaking the Shield of Perturbation-Protected Gradients via Adaptive Diffusion (AAAI 2025) · better on 11 of 11 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
- Mjölnir: Breaking the Shield of Perturbation-Protected Gradients via Adaptive Diffusion (AAAI 2025 · similarity 0.89)
- High-Fidelity Gradient Inversion in Distributed Learning (AAAI 2024 · similarity 0.89)
- Concealing Sensitive Samples against Gradient Leakage in Federated Learning (AAAI 2024 · similarity 0.89)
- A New Federated Learning Framework Against Gradient Inversion Attacks (AAAI 2025 · similarity 0.88)
- Gradient Inversion Attacks on Parameter-Efficient Fine-Tuning (CVPR 2025 · similarity 0.87)
- Foreseeing Reconstruction Quality of Gradient Inversion: An Optimization Perspective (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.
