Research map / Federated learning and privacy / Privacy attacks and secure learning
FedOT: Ownership Verification and Leakage Tracing via Watermarks for Federated LDMs
Wenlong Cheng, Yuan Gan, Yunqiu Xu, Jiaxu Miao
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 negative Latent Vector Transformation (ablation) | FedOT w/o LVT · LAION-10K (SD2.1 federated, after VAE replacement attack) | FID after VAE replacement: 17.054 -> 40.537 | Improved |
| add translation Latent Vector Transformation (c=11) (ablation) | FedOT w/o LVT · LAION-10K (SD2.1 federated, after VAE replacement attack) | FID after VAE replacement: 17.054 -> 92.497 | Improved |
| add mirror Latent Vector Transformation (ablation) | FedOT w/o LVT · LAION-10K (SD2.1 federated, after VAE replacement attack) | FID after VAE replacement: 17.054 -> 70.622 | Improved |
| use deterministic negative transformation instead of random Gaussian noise (ablation) | FedOT · LAION-10K (SD2.1 federated, no attack) | FID: 35.585 -> 20.367 | Improved |
| add chunked watermark with client suffix bits | Stable Signature* · LAION-10K client tracing | Detection rate: -- -> 0.993 | Improved |
| add negative LVT to FedOT (ablation) | FedOT w/o LVT · LAION-10K ownership verification (first 16 bits) | Detection rate: 0.996 -> 0.960 | Hurt |
| add negative LVT to FedOT (ablation) | FedOT w/o LVT · LAION-10K (SD2.1 federated, no attack) | FID: 16.687 -> 20.367 | Hurt |
| increase translation coefficient c from 2 to 11 (ablation) | FedOT_tran · LAION-10K (SD2.1 federated, after VAE replacement attack) | FID increase under replacement attack: -1.424 -> +70.070 | Improved |
| increase translation coefficient c from 5 to 11 (ablation) | FedOT_tran · LAION-10K (SD2.1 federated, no attack) | FID: 19.684 -> 22.427 | Hurt |
| decrease lambda_i from 0.8 to 0.05 (ablation) | watermarked VAE decoder · VAE decoder watermark fine-tuning | PSNR vs bit accuracy: PSNR 34.793 -> 29.842; Bit Acc 0.906 -> 0.965 | Mixed |
| use r=16 instead of r=8 verification bits (ablation) | FedOT_neg · LAION-10K ownership verification (first 16 bits) | Detection rate: 0.944 -> 0.960 | Improved |
| use r=16 instead of r=24 verification bits (ablation) | FedOT_neg · LAION-10K client tracing (last 32 bits) | Bit accuracy: 0.899 -> 0.910 | Improved |
Closest papers by meaning
- SleeperMark: Towards Robust Watermark against Fine-Tuning Text-to-image Diffusion Models (CVPR 2025 · similarity 0.87)
- FedGMark: Certifiably Robust Watermarking for Federated Graph Learning (NeurIPS 2024 · similarity 0.87)
- MFL-Owner: Ownership Protection for Multi-modal Federated Learning via Orthogonal Transform Watermark (AAAI 2025 · similarity 0.87)
- StableGuard: Towards Unified Copyright Protection and Tamper Localization in Latent Diffusion Models (NeurIPS 2025 · similarity 0.86)
- Harnessing Frequency Spectrum Insights for Image Copyright Protection Against Diffusion Models (CVPR 2025 · similarity 0.86)
- Black-Box Forgery Attacks on Semantic Watermarks for Diffusion Models (CVPR 2025 · 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.
