Research map / Adversarial robustness and security / Attacks on text and graphs
PLMmark: A Secure and Robust Black-Box Watermarking Framework for Pre-trained Language Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 supervised contrastive watermark embedding rather than the strongest reported baseline | BERT · SST-2 | WACC: 84.01 -> 99.61 | Helped |
| Use supervised contrastive watermark embedding rather than the strongest reported baseline | BERT · SST-5 | WACC: 84.74 -> 99.89 | Helped |
| Use supervised contrastive watermark embedding rather than the strongest reported baseline | BERT · Offenseval | WACC: 96.90 -> 99.89 | Helped |
| Use supervised contrastive watermark embedding rather than the strongest reported baseline | BERT · Lingspam | WACC: 69.45 -> 98.82 | Helped |
| Use supervised contrastive watermark embedding rather than the strongest reported baseline | BERT · AGNews | WACC: 32.93 -> 91.76 | Helped |
| Use supervised contrastive watermark embedding rather than the strongest reported baseline | RoBERTa · SST-2 | WACC: 62.68 -> 95.03 | Helped |
| Use supervised contrastive watermark embedding rather than the strongest reported baseline | RoBERTa · SST-5 | WACC: 76.19 -> 86.11 | Helped |
| Use supervised contrastive watermark embedding rather than the strongest reported baseline | RoBERTa · Offenseval | WACC: 92.40 -> 99.96 | Helped |
| Use supervised contrastive watermark embedding rather than the strongest reported baseline | RoBERTa · Lingspam | WACC: 55.43 -> 70.06 | Helped |
| Use supervised contrastive watermark embedding rather than the strongest reported baseline | RoBERTa · AGNews | WACC: 44.91 -> 68.36 | Helped |
| Increase trigger insertion times from k=1 to k=5 (ablation) | PLMmark · AGNews | WACC: 74.08 -> 91.76 | Helped |
| Use PLMmark rather than a clean model for ownership verification | BERT watermarking framework · SST-2 | WACC: 10.21 -> 99.61 | Helped |
Closest papers by meaning
- Task-Agnostic Language Model Watermarking via High Entropy Passthrough Layers (AAAI 2025 · similarity 0.88)
- Black-Box Detection of Language Model Watermarks (ICLR 2025 · similarity 0.88)
- Watermarks in the Sand: Impossibility of Strong Watermarking for Language Models (ICML 2024 · similarity 0.87)
- PRO: Enabling Precise and Robust Text Watermark for Open-Source LLMs (NeurIPS 2026 · similarity 0.87)
- A Semantic Invariant Robust Watermark for Large Language Models (ICLR 2024 · similarity 0.87)
- Lightweight-Mark: Rethinking Deep Learning-Based Watermarking (ICML 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.