arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2404.02655cs.CL
收藏

通过激发保真度校准大语言模型的置信度

Calibrating the Confidence of Large Language Models by Eliciting Fidelity

School of Computer Science, Fudan University(复旦大学计算机科学学院);Meituan(美团)

Mozhi Zhang, Mianqiu Huang, Rundong Shi, Linsen Guo, Chong Peng, Peng Yan, Yaqian Zhou, Xipeng Qiu

AI总结:

针对对齐后大语言模型过度自信、置信度与正确率校准不佳的问题,将置信度分解为问题不确定性与答案保真度,提出即插即用的置信度估计方法,经多模型多数据集验证有效,还提出新校准指标并探讨真正良好校准的置信度,可作为该领域基线。

摘要:

采用RLHF(基于人类反馈的强化学习)等技术优化的大语言模型在有用性和无害性方面已实现良好的对齐。然而,对齐后的这类语言模型往往表现出过度自信的问题,即其表达出的置信度与自身正确率无法实现准确校准。本文将语言模型置信度分解为对问题的“不确定性”以及对语言模型生成答案的“保真度”,进而提出一种即插即用的方法来估计语言模型的置信度。我们在4个MCQA(多项选择题问答)数据集上对6个RLHF-LM(基于RLHF优化的语言模型)开展实验,结果表明该方法具备良好的校准性能。此外,我们提出了IPR和CE这两个全新指标用于评估模型的校准效果,并围绕“真正校准良好的置信度”展开了详细探讨。本方法可作为强力基线,我们也希望这项工作能为模型置信度校准研究提供一些启发。

英文摘要:

Large language models optimized with techniques like RLHF have achieved good alignment in being helpful and harmless. However, post-alignment, these language models often exhibit overconfidence, where the expressed confidence does not accurately calibrate with their correctness rate. In this paper, we decompose the language model confidence into the \textit{Uncertainty} about the question and the \textit{Fidelity} to the answer generated by language models. Then, we propose a plug-and-play method to estimate the confidence of language models. Our method has shown good calibration performance by conducting experiments with 6 RLHF-LMs on four MCQA datasets. Moreover, we propose two novel metrics, IPR and CE, to evaluate the calibration of the model, and we have conducted a detailed discussion on \textit{Truly Well-Calibrated Confidence}. Our method could serve as a strong baseline, and we hope that this work will provide some insights into the model confidence calibration.

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。

↑