arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.17608cs.LGcs.AIstat.ML
收藏

面向随时验证的统计水印技术

Towards Anytime-Valid Statistical Watermarking

Department of XXX, University of YYY, Location, Country(YYY大学XXX系);School of ZZZ, Institute of WWW, Location, Country(WWW研究所ZZZ学院)

Baihe Huang, Eric Xu, Kannan Ramchandran, Jiantao Jiao, Michael I. Jordan

AI总结:

本文提出Anchored E-Watermarking框架,通过构建测试超级鞅实现随时验证的统计水印技术,提升样本效率13-15%。

摘要:

大语言模型(LLMs)的普及需要高效的机制来区分机器生成内容与人类文本。虽然统计水印技术已作为一种有前景的解决方案出现,但现有方法存在两个关键限制:缺乏选择采样分布的系统方法以及依赖固定地平线假设检验,这阻止了有效的提前停止。在本文中,我们通过开发第一个基于e值的水印框架,即Anchored E-Watermarking,来弥合这一差距,该框架统一了最优采样与随时验证的推理。与传统方法不同,我们的框架通过为检测过程构建测试超级鞅,实现了有效的随时推理。通过利用锚定分布来近似目标模型,我们以最坏情况下的对数增长率为基准,表征最优e值,并推导出最优的预期停止时间。我们的理论主张通过模拟和在已建立的基准上的评估得到验证,显示我们的框架可以显著提高样本效率,将检测所需的平均标记预算减少13-15%,相对于最先进的基线。

英文摘要:

The proliferation of Large Language Models (LLMs) necessitates efficient mechanisms to distinguish machine-generated content from human text. While statistical watermarking has emerged as a promising solution, existing methods suffer from two critical limitations: the lack of a principled approach for selecting sampling distributions and the reliance on fixed-horizon hypothesis testing, which precludes valid early stopping. In this paper, we bridge this gap by developing the first e-value-based watermarking framework, Anchored E-Watermarking, that unifies optimal sampling with anytime-valid inference. Unlike traditional approaches where optional stopping invalidates Type-I error guarantees, our framework enables valid, anytime-inference by constructing a test supermartingale for the detection process. By leveraging an anchor distribution to approximate the target model, we characterize the optimal e-value with respect to the worst-case log-growth rate and derive the optimal expected stopping time. Our theoretical claims are substantiated by simulations and evaluations on established benchmarks, showing that our framework can significantly enhance sample efficiency, reducing the average token budget required for detection by 13-15% relative to state-of-the-art baselines.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑