2026-10-25 · America/Los_Angeles · 论文 · #15

Watermarking Diffusion Language Models with Learned Foresight and Balanced Bands

Green-red watermarks favor tokens from a green list that a secret key derives from the preceding tokens. Diffusion language models, however, often choose a token before its preceding tokens, and the token later shapes its neighbors' green lists. Prior work therefore tilts each choice to raise the expected number of green tokens at its own and its neighbors' positions. Building on this objective, we first learn two components that prior work keeps fixed: vocabulary bands that serve as green lists, separate competing candidates and keep the null distribution exact, and a foresight predictor…

热度 14.0 / 100；排名与评分保留该期记录。