arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2407.21248cs.CLcs.CRcs.LG
收藏

通过意外词元实现大语言模型的自适应预训练数据检测

Adaptive Pre-training Data Detection for Large Language Models via Surprising Tokens

Anqi Zhang, Chaofeng Wu

AI总结:

本文提出一种自适应预训练数据检测方法,通过定位并利用“意外词元”的预测概率来识别大语言模型是否见过某段数据,无需访问训练语料或额外训练,在多个基准上最大提升29.5%,并引入新基准Dolma-Book。

摘要:

尽管大语言模型(LLMs)被广泛使用,但由于其训练数据不透明,引发了关于隐私、安全和版权的日益关注,这使得预训练数据检测问题被提上日程。当前针对该问题的解决方案利用了机器学习隐私领域探索的技术,例如成员推理攻击(MIAs),这些技术严重依赖大语言模型的逐字记忆能力。然而,这种依赖带来了挑战,尤其是在训练数据量巨大且有效训练轮数受限的情况下。在本文中,我们提出了一种自适应预训练数据检测方法,该方法缓解了这种依赖并有效放大了识别能力。我们的方法自适应地定位输入中的“意外词元”。如果大语言模型对某个词元的预测“确定但错误”,即概率分布的香农熵较低且真实词元的概率同时较低,那么该词元对该大语言模型而言就是意外的。通过使用意外词元的预测概率来衡量“意外程度”,该检测方法基于一个简单假设实现:与看到未见过的数据相比,模型看到已见过的数据时意外程度更低。该方法无需访问预训练数据语料库,也无需参考模型等额外训练即可应用。在各种基准和模型上进行的大量实验中,我们的方法与现有方法相比表现出一致的提升,最大改进幅度达到29.5%。我们还引入了一个基于新框架开发的新基准Dolma-Book,该框架利用模型训练前后收集的书籍数据来提供进一步评估。

英文摘要:

While large language models (LLMs) are extensively used, there are raising concerns regarding privacy, security, and copyright due to their opaque training data, which brings the problem of detecting pre-training data on the table. Current solutions to this problem leverage techniques explored in machine learning privacy such as Membership Inference Attacks (MIAs), which heavily depend on LLMs' capability of verbatim memorization. However, this reliance presents challenges, especially given the vast amount of training data and the restricted number of effective training epochs. In this paper, we propose an adaptive pre-training data detection method which alleviates this reliance and effectively amplify the identification. Our method adaptively locates \textit{surprising tokens} of the input. A token is surprising to a LLM if the prediction on the token is "certain but wrong", which refers to low Shannon entropy of the probability distribution and low probability of the ground truth token at the same time. By using the prediction probability of surprising tokens to measure \textit{surprising}, the detection method is achieved based on the simple hypothesis that seeing seen data is less surprising for the model compared with seeing unseen data. The method can be applied without any access to the the pre-training data corpus or additional training like reference models. Our approach exhibits a consistent enhancement compared to existing methods in diverse experiments conducted on various benchmarks and models, achieving a maximum improvement of 29.5\%. We also introduce a new benchmark Dolma-Book developed upon a novel framework, which employs book data collected both before and after model training to provide further evaluation.

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。

↑