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 2404.04689stat.MLcs.CLcs.LG
收藏

面向大语言模型置信度评分的多重校准

Multicalibration for Confidence Scoring in LLMs

AWS AI(亚马逊云科技人工智能部门);University of Pennsylvania(宾夕法尼亚大学)

Gianluca Detommaso, Martin Bertran, Riccardo Fogliato, Aaron Roth

AI总结:

该研究利用嵌入聚类与自标注构建交叉分组,并提出降低过拟合的多重校准算法,从而提升 LLM 输出置信度在校准和准确率细粒度指标上的表现。

摘要:

本文提出使用“多重校准”(multicalibration),为大语言模型(LLM)生成的输出产生可解释且可靠的置信度分数。多重校准不仅要求边际意义上的校准,还要求在数据的各种交叉分组上同时实现校准。我们展示了如何通过两种技术为提示/补全对构建与正确概率相关的分组:在嵌入空间内进行聚类,以及“自标注”——通过向 LLM 提出关于提示的各种是非问题来查询该模型。我们还开发了多重校准算法的新型变体,通过降低其过拟合倾向来提升性能。通过在多个问答数据集和 LLM 上进行系统基准测试,我们表明,与现有方法相比,所提技术能够产生置信度分数,并在校准和准确率的细粒度指标上带来显著改进。

英文摘要:

This paper proposes the use of "multicalibration" to yield interpretable and reliable confidence scores for outputs generated by large language models (LLMs). Multicalibration asks for calibration not just marginally, but simultaneously across various intersecting groupings of the data. We show how to form groupings for prompt/completion pairs that are correlated with the probability of correctness via two techniques: clustering within an embedding space, and "self-annotation" - querying the LLM by asking it various yes-or-no questions about the prompt. We also develop novel variants of multicalibration algorithms that offer performance improvements by reducing their tendency to overfit. Through systematic benchmarking across various question answering datasets and LLMs, we show how our techniques can yield confidence scores that provide substantial improvements in fine-grained measures of both calibration and accuracy compared to existing methods.

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。

↑