Research map / Large language models / Language, safety and interpretability
IPMark: A Sentence-Level Watermark for LLMs with Hierarchical Personalization and Efficient Detection
Wenbo An, Lianwei Wu, Zehao 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 quality assessment module (ablation) | IPMark · C4 (Llama-3.2-3B) | perplexity: 3.012 -> 2.779 | Improved |
| add quality assessment module (ablation) | IPMark · C4 (Llama-3.2-3B) | AUC-ROC: 0.997 -> 0.999 | Improved |
| add syntactic evaluation structure (ablation) | IPMark · C4 (Llama-3.2-3B) | AUC-ROC: 0.989 -> 0.999 | Improved |
| add syntactic evaluation structure (ablation) | IPMark · C4 (Llama-3.2-3B) | perplexity: 2.879 -> 2.779 | Improved |
| add both quality assessment and syntactic structure (ablation) | IPMark · C4 (Llama-3.2-3B) | Best F1: 0.774 -> 0.985 | Improved |
| add both quality assessment and syntactic structure (ablation) | IPMark · C4 (Llama-3.2-3B) | perplexity: 3.340 -> 2.779 | Improved |
| use hierarchical retrieval instead of linear scan | IPMark detection · 100,000 users | average latency (s): 72.677 -> 5.020 | Improved |
| use IPMark instead of Unigram | watermark method · C4 (Llama-3.2-3B) | perplexity: 2.892 -> 2.779 | Improved |
| use IPMark instead of KGW under Dipper rewrite | watermark method · C4 Dipper Rewrite | AUC-ROC: 0.764 -> 0.773 | Improved |
| use IPMark instead of PF under back-translation | watermark method · C4 back-translation | AUC-ROC: 0.922 -> 0.840 | Hurt |
| use IPMark instead of EXP under copy-paste-50 | watermark method · C4 Copy-Paste-50 | AUC-ROC: 0.983 -> 0.871 | Hurt |
| use IPMark instead of PF in text quality | watermark method · C4 (Llama-3.2-3B) | perplexity: 8.056 -> 2.779 | Improved |
Closest papers by meaning
- Learning to Watermark: A Selective Watermarking Framework for Large Language Models via Multi-Objective Optimization (NeurIPS 2025 · similarity 0.90)
- SAEMark: Steering Personalized Multilingual LLM Watermarks with Sparse Autoencoders (NeurIPS 2025 · similarity 0.90)
- Theoretically Grounded Framework for LLM Watermarking: A Distribution-Adaptive Approach (NeurIPS 2025 · similarity 0.89)
- PRO: Enabling Precise and Robust Text Watermark for Open-Source LLMs (NeurIPS 2026 · similarity 0.89)
- Black-Box Detection of Language Model Watermarks (ICLR 2025 · similarity 0.89)
- A Semantic Invariant Robust Watermark for Large Language Models (ICLR 2024 · 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.
