Research map / Adversarial robustness and security / Attacks on text and graphs
GAN You See Me? Enhanced Data Reconstruction Attacks against Split Inference
Ziang Li, Mengda Yang, Yaxin Liu, Juan Wang, Hongxin Hu, Wenzhe Yi, Xiaoyang Xu
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 GLASS instead of LM optimization attack | optimization-based DRA · CelebA Block3, no defense | LPIPS: 0.298 -> 0.140 | Improved |
| use GLASS++ instead of IN learning-based attack | learning-based DRA · CelebA Block3, no defense | SSIM: 0.685 -> 0.833 | Improved |
| use GLASS++ instead of IN against NoPeek | learning-based DRA · NoPeek defense Block3 Acc=79.97% | SSIM: 0.376 -> 0.678 | Improved |
| use GLASS instead of rMLE against NoPeek | optimization-based DRA · NoPeek defense Block3 Acc=79.97% | SSIM: 0.006 -> 0.554 | Improved |
| use GLASS++ instead of LM against DISCO (95% channel clipping) | DRA · DISCO Acc=79.34% Block3 | SSIM: 0.620 -> 0.784 | Improved |
| strengthen defense (lower accuracy level) for Noise Mask | GLASS++ · Noise Mask Block3 | SSIM: 0.714 -> 0.524 | Hurt |
| strengthen DISCO defense by relaxing clipping (Acc 79.34% to 79.26%) | GLASS++ · DISCO Block3 | SSIM: 0.784 -> 0.584 | Hurt |
| increase Shredder defense level | GLASS++ · Shredder Block3 | SSIM: 0.695 -> 0.697 | No clear effect |
| train StyleGAN prior on FFHQ instead of CelebA (distribution shift) | GLASS++ · Block3 | SSIM: 0.833 -> 0.727 | Hurt |
| add generator fine-tuning after GLASS++ under distribution shift | GLASS++ with FFHQ prior · Block3 | SSIM: 0.727 -> 0.736 | Improved |
| add generator fine-tuning after GLASS++ under distribution shift | GLASS++ with FFHQ prior · Block3 | LPIPS: 0.211 -> 0.199 | Improved |
| add generator fine-tuning after GLASS++ under distribution shift | GLASS++ with FFHQ prior · Block3 | NIQE: 13.77 -> 14.75 | Hurt |
Closest papers by meaning
- DRAG: Data Reconstruction Attack using Guided Diffusion (ICML 2025 · similarity 0.86)
- InfoDecom: Decomposing Information for Defending Against Privacy Leakage in Split Inference (AAAI 2026 · similarity 0.86)
- A Stealthy Wrongdoer: Feature-Oriented Reconstruction Attack against Split Learning (CVPR 2024 · similarity 0.85)
- Bi-perspective Splitting Defense: Achieving Clean-Seed-Free Backdoor Security (ICML 2025 · similarity 0.83)
- Understanding Reconstruction Attacks with the Neural Tangent Kernel and Dataset Distillation (ICLR 2024 · similarity 0.82)
- Reimagining Mutual Information for Enhanced Defense against Data Leakage in Collaborative Inference (NeurIPS 2024 · similarity 0.82)
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.
