Research map / Federated learning and privacy / Privacy attacks and secure learning
"Veil Privacy on Visual Data: Concealing Privacy for Humans, Unveiling for DNNs"
Shuchao Pang, Ruhao Ma, Bing Li, Yongbin Zhou, Yazhou Yao
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 |
|---|---|---|---|
| train on veiled data instead of original data, then test on original data | Veil Privacy · CIFAR10 (AlexNet) | accuracy: 85.99% -> 81.73% | Hurt |
| train on veiled data, test on original data | Veil Privacy · CIFAR100 (ViT) | accuracy: 87.35% -> 81.86% | Hurt |
| train on veiled data, test on original data | Veil Privacy · ORL (LeNet-5) | accuracy: 96.97% -> 96.97% | No clear effect |
| use Veil Privacy instead of dataset distillation for privacy-preserving surrogate data | CIFAR10 AlexNet · CIFAR10 | accuracy on original data: 66.3±0.2% -> 81.73% | Improved |
| use random pixel flipping instead of random noise initialization (ablation) | Veil Privacy · ORL | accuracy on original data: 55.76% -> 96.97% | Improved |
| increase flipping ratio p from 0.3 to 0.5 (ablation) | Veil Privacy · ORL | SSIM veiled vs original (lower is more concealing): 0.14702 -> 0.03130 | Improved |
| choose last FC layer instead of conv layer for gradient iteration (ablation) | Veil Privacy · ORL | SSIM veiled vs original (lower is more concealing): 0.10403 -> 0.03130 (Conv vs Fc3) | Improved |
| add gradient iteration algorithm after random pixel flipping (ablation) | Veil Privacy · ORL | data utility | Improved |
| GPEN restoration on veiled versus noised data | security evaluation · BioID | SSIM restored vs original: 0.33262 (noised, restored) vs 0.06315 (veiled, restored) | Improved |
Closest papers by meaning
- Balancing Privacy and Performance: A Many-in-One Approach for Image Anonymization (AAAI 2025 · similarity 0.88)
- PID: Prompt-Independent Data Protection Against Latent Diffusion Models (ICML 2024 · similarity 0.87)
- ViP: A Differentially Private Foundation Model for Computer Vision (ICML 2024 · similarity 0.86)
- Validating Privacy-Preserving Face Recognition under a Minimum Assumption (CVPR 2024 · similarity 0.86)
- Private Attribute Inference from Images with Vision-Language Models (NeurIPS 2024 · similarity 0.86)
- MultiPriv: Benchmarking Individual-Level Privacy Reasoning in Vision-Language Models (ICML 2026 · 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.
