Research map / Large language models / Language, safety and interpretability
PMark: Towards Robust and Distortion-free Semantic-level Watermarking with Channel Constraints
Jiahao Huo, Shuliang Liu, Bin Wang, Junyan Zhang, Yibo Yan, Aiwei Liu, Xuming Hu, Mingxun Zhou
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 multiple channels (b=1 to b=3) to online PMark at N=8 (ablation) | PMark Online · BOOKSUM (OPT-1.3B) | TP@FP=1%: 81.0 -> 98.0 | Improved |
| increase sample budget N from 8 to 16 at b=1 (online) (ablation) | PMark Online · BOOKSUM (OPT-1.3B) | TP@FP=1%: 81.0 -> 84.0 | Improved |
| use offline zero-median prior with small budget N=8 and b=2 (ablation) | PMark Offline · BOOKSUM (OPT-1.3B) | TP@FP=1%: 47.0 at N=8 vs 90.0 at N=16 | Hurt |
| use soft-z-test instead of naive z-test (ablation) | PMark Online · BOOKSUM and C4 (Mistral-7B, attacked) | TP@FP=1%: +11.9% BOOKSUM, +6.4% C4 | Improved |
| semantic-level PMark online vs best SWM baseline SemStamp under GPT paraphrase | Mistral-7B watermarking · BOOKSUM (GPT paraphrase) | TP@FP=1%: 76.0 -> 96.8 | Improved |
| use PMark offline instead of online (ablation) | PMark · Paraphrase attacks | TP@FP=1%: 96.8 -> 94.4 (BOOKSUM GPT, Mistral-7B) | Hurt |
| PMark online vs SemStamp token consumption | SWM · Token efficiency | tokens per sentence: 1694.4 -> 315.8 | Improved |
| PMark vs KGW under Pegasus paraphrase | OPT-1.3B · C4 (Pegasus) | TP@FP=1%: KGW 89.3 vs PMark Online 96.2 (but KGW 97.0 TP@5%) | Mixed |
Closest papers by meaning
- A Semantic Invariant Robust Watermark for Large Language Models (ICLR 2024 · similarity 0.90)
- GaussMark: A Practical Approach for Structural Watermarking of Language Models (ICML 2025 · similarity 0.89)
- Learning to Watermark: A Selective Watermarking Framework for Large Language Models via Multi-Objective Optimization (NeurIPS 2025 · similarity 0.89)
- Theoretically Grounded Framework for LLM Watermarking: A Distribution-Adaptive Approach (NeurIPS 2025 · similarity 0.89)
- Adaptive Text Watermark for Large Language Models (ICML 2024 · similarity 0.89)
- BiMark: Unbiased Multilayer Watermarking for Large Language Models (ICML 2025 · similarity 0.89)
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.
