SimpleMark: Fast Multi-Bit Text Watermarking under f -Divergence Constraints
SimpleMark embeds multi-bit text watermarks while bounding f-divergence from the base language model.
SimpleMark is a multi-bit text-watermarking framework whose security is defined by f-divergence from the base language-model distribution, including total variation and KL divergence, for every realized key and message. A coding scheme optimally biases next-token distributions under a prescribed divergence budget and characterizes the tradeoff among embedding rate, decoding reliability, and statistical security. Experiments against prior multi-bit schemes on modern language models report substantially lower detectability while keeping message recovery and generation quality competitive.
- Defines per-key, per-message security via f-divergence
- Covers total variation and KL divergence as special cases
- Biases next-token distributions under a divergence budget
- Reports lower detectability with competitive message recovery
From the original article
1 min read at arxiv.orgWe introduce a framework for multi-bit text watermarking with security defined directly through $f$-divergence from the base language model distribution.
By Benjamin D. Kim · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.