WaterDrum: Watermarking for Data-centric Unlearning Metric
Xinyang Lu, Xinyuan Niu, Gregory Kang Ruey Lau, Bui Thi Cam Nhung, Rachael Hwee Ling Sim, Fanyu Wen, Chuan-Sheng Foo, See-Kiong Ng, Bryan Kian Hsiang Low · 2025
Read WaterDrum: Watermarking for Data-centric Unlearning Metric by Xinyang Lu, Xinyuan Niu, Gregory Kang Ruey Lau, Bui Thi Cam Nhung, Rachael Hwee Ling Sim, Fanyu Wen, Chuan-Sheng Foo, See-Kiong Ng, Bryan Kian Hsiang Low (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-05064.txt.
Description
Xinyang Lu and a team of co-authors present WaterDrum in two thousand twenty-five. WaterDrum grades how well a language model has unlearned data. Utility based grades can fail when kept and dropped data read alike. Retraining from scratch to check is often out of reach as well.
Their metric is data centric and rests on robust text watermarking. Marks in the data reveal if its sway truly left the model. New benchmark sets vary how alike the kept and dropped parts are. That lets unlearning methods face a strict, fair test. Code and data sets are shared in public.
Catalog record
| CREATOR | Xinyang Lu, Xinyuan Niu, Gregory Kang Ruey Lau, Bui Thi Cam Nhung, Rachael Hwee Ling Sim, Fanyu Wen, Chuan-Sheng Foo, See-Kiong Ng, Bryan Kian Hsiang Low |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 88 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-05064