arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.22707cs.CL
收藏

通过排名相关性检测LLM训练数据中的非成员

Detecting Non-Membership in LLM Training Data via Rank Correlations

JPMorgan AI Research(摩根大通人工智能研究院)

Pranav Shetty, Mirazul Haque, Zhiqiang Ma, Xiaomo Liu

AI总结:

研究提出PRISM方法,利用模型logits的排名相关性检测训练数据未被使用,有效排除训练数据中的成员身份,提升模型合规性验证。

摘要:

随着大型语言模型(LLMs)在日益庞大且不透明的文本语料库上进行训练,确定哪些数据参与了训练已成为版权执行、合规审计和用户信任的关键。尽管先前工作关注检测数据集是否被用于训练(成员推断),但验证数据集未被使用的问题却很少被关注。我们通过引入PRISM测试来填补这一空白,该测试仅使用灰盒访问模型logits来检测数据集层面的非成员身份。我们的关键见解是,两个未见过该数据集的模型在归一化token log概率上的排名相关性高于其中一个模型已训练过该数据的情况。基于这一观察,我们构建了一个基于相关性的测试来检测非成员身份。实证结果显示,PRISM在所有测试的数据集上都能可靠地排除训练数据中的成员身份,同时避免假阳性,从而提供了一种验证特定数据集未被包含在LLM训练中的框架。

英文摘要:

As large language models (LLMs) are trained on increasingly vast and opaque text corpora, determining which data contributed to training has become essential for copyright enforcement, compliance auditing, and user trust. While prior work focuses on detecting whether a dataset was used in training (membership inference), the complementary problem -- verifying that a dataset was not used -- has received little attention. We address this gap by introducing PRISM, a test that detects dataset-level non-membership using only grey-box access to model logits. Our key insight is that two models that have not seen a dataset exhibit higher rank correlation in their normalized token log probabilities than when one model has been trained on that data. Using this observation, we construct a correlation-based test that detects non-membership. Empirically, PRISM reliably rules out membership in training data across all datasets tested while avoiding false positives, thus offering a framework for verifying that specific datasets were excluded from LLM training.

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。

↑