Research map / Optimization / Optimizers
Gradient Inversion Attacks Beyond SGD
Guangnian Wan, Gongfan Fang, Xinyin Ma, Xinchao 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 |
|---|---|---|---|
| add analytical label recovery from optimizer updates (ablation) | update-matching attack · ImageNet (Adam) | PSNR (dB): 15.3358 -> 19.6252 | Improved |
| add the proposed update-matching objective (ablation) | label-recovery attack · ImageNet (Adam) | PSNR (dB): 11.4221 -> 19.6252 | Improved |
| add analytical label recovery (ablation) | update-matching attack · ImageNet (Adam) | LPIPS: 0.7175 -> 0.4614 | Improved |
| add update-matching objective (ablation) | label-recovery attack · ImageNet (Adam) | LPIPS: 0.8413 -> 0.4614 | Improved |
| use the proposed attack instead of HFGI when only Adam updates are visible | HFGI · ImageNet (Adam updates) | PSNR (dB): 13.3472 -> 20.5985 | Improved |
| use the proposed attack instead of HFGI when only Adam updates are visible | HFGI · PACS (Adam updates) | PSNR (dB): 12.4712 -> 19.0078 | Improved |
| use the update-matching attack instead of HFGI with raw SGD gradients | HFGI · ImageNet (SGD) | PSNR (dB): 20.7997 -> 23.9195 | Improved |
| use the update-matching attack instead of HFGI with raw SGD gradients | HFGI · PACS (SGD) | PSNR (dB): 19.2953 -> 24.8643 | Improved |
| use the proposed attack instead of IG under RMSProp | IG · ImageNet (RMSProp) | PSNR (dB): 11.8515 -> 20.4626 | Improved |
| use the proposed attack instead of IG under AdaGrad | IG · ImageNet (AdaGrad) | PSNR (dB): 9.6361 -> 18.7233 | Improved |
| use the proposed attack instead of IG under SGD with momentum | IG · ImageNet (Momentum) | PSNR (dB): 10.6764 -> 17.0135 | Improved |
| pair update-matching with existing label inference | update-matching objective · ImageNet (Adam) | reconstruction quality | Hurt |
Closest papers by meaning
- Foreseeing Reconstruction Quality of Gradient Inversion: An Optimization Perspective (AAAI 2024 · similarity 0.89)
- Recovering Labels from Local Updates in Federated Learning (ICML 2024 · similarity 0.87)
- High-Fidelity Gradient Inversion in Distributed Learning (AAAI 2024 · similarity 0.87)
- Gradient Inversion Attacks on Parameter-Efficient Fine-Tuning (CVPR 2025 · similarity 0.86)
- Towards Eliminating Hard Label Constraints in Gradient Inversion Attacks (ICLR 2024 · similarity 0.86)
- Concealing Sensitive Samples against Gradient Leakage in Federated 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.
