SupportCal：通过参考支持与佐证对训练后大语言模型进行无标签校准
SupportCal: Label-Free Calibration of Post-Trained LLMs via Reference Support and Corroboration
浏览论文内容
中文总结 AI 辅助
针对训练后模型过度自信问题，提出无标签校准方法SupportCal，利用预训练参考模型的支持与佐证对不一致样本加权，在多个基准上降低期望校准误差。
中文摘要 AI 辅助
训练后阶段通常能提升任务性能，但可能损害置信度校准，导致训练后语言模型（PoLMs）比其对应的预训练语言模型（PLMs）更加过度自信。由于任务特定的有标签校准数据可能成本高昂或不可用，对应的预训练PLM为事后校准提供了天然的无标签参考。先前的基于一致性门控的PLM参考校准方法仅使用PoLM与其PLM参考一致的样本来拟合标量温度，排除不一致样本，因为直接对齐可能使拟合温度过高并导致欠自信。我们重新审视这种二元处理方式。一项受控的重新引入诊断揭示了非单调的总体效应：纳入中等比例的不一致样本可以改善校准，而当单位权重纳入接近完整不一致集时，益处逐渐减弱。我们提出SupportCal，一种无标签的事后方法，该方法以单位权重保留一致样本，并根据从规模兼容的候选池中选取的预训练参考的自身基础PLM相对支持与佐证，为不一致样本分配连续权重。我们进一步刻画了所得加权目标何时允许有限最优温度。在MedMCQA和MathQA上，对于几乎所有评估的目标模型配置，SupportCal比仅一致基线实现了更低的ECE；补充的TweetEval情感分析结果在固定标签分类任务上显示出相同模式。
英文摘要
Post-training often improves task performance but can degrade confidence calibration, leaving post-trained language models (PoLMs) more overconfident than their corresponding pretrained language models (PLMs). Because task-specific labeled calibration data can be costly or unavailable, the corresponding PLM provides a natural label-free reference for post-hoc calibration. Prior agreement-gated PLM-referenced calibration fits a scalar temperature using only examples on which the PoLM and its PLM reference agree, excluding disagreement examples because direct alignment can drive the fitted temperature excessively high and induce under-confidence. We revisit this binary treatment. A controlled reintroduction diagnostic reveals a non monotonic aggregate effect: admitting a moderate fraction of disagreement examples can improve calibration, whereas the benefit diminishes as unit weight inclusion approaches the full disagreement set. We introduce SupportCal, a label-free post-hoc method that retains agreement examples at unit weight and assigns disagreement examples continuous weights based on the own-base PLM's relative support and corroboration from pretrained references selected from a size-compatible candidate pool. We further characterize when the resulting weighted objective admits a finite optimal temperature. Across MedMCQA and MathQA, SupportCal yields lower mean ECE than the agreement-only baseline for nearly all evaluated target-model configurations; supplementary TweetEval Sentiment results show the same pattern on a fixed-label classification task.
发表机构
- The University of Sydney（悉尼大学）
- University of Cambridge（剑桥大学）
- The University of Adelaide（阿德莱德大学）
机构由 AI 辅助整理，请以论文原文为准。
