arXiv：cs.LG· Zizhuo Zhang, Xiong Peng, Jingwei Sun, Rong Yao, Shixiong Kai, Mingxuan Yuan, Bo Han·· 3 小时前AI 评分47
面向大语言模型推理的测试时校准学习 TTCL
Test-time Calibration Learning for Large Language Model Reasoning
AI 导读
研究者提出无标签框架 TTCL（Test-Time Calibration Learning），直接在未标注的目标任务数据上联合适配推理准确率与语言化置信度，自监督信号来自模型多次生成的回答。
正文
Abstract:Reliable large language models (LLMs) must not only produce accurate answers but also express confidence that faithfully reflects their probability of being correct. Such calibration is essential for identifying uncertain predictions and supporting reliable decision-making in real-world deployment. Recent studies incorporate calibration learning into reinforcement learning (RL), jointly optimizing answer correctness and verbalized confidence using ground-truth correctness supervision. However, their reliance on labeled data limits their applicability in practical test-time settings, where ground-truth labels are unavailable and calibration may need to adapt to newly encountered target tasks. To address this challenge, we propose Test-Time Calibration Learning (TTCL), a label-free framework that jointly adapts reasoning accuracy and verbalized confidence directly on unlabeled target-task data. Specifically, TTCL derives self-supervision signals for both correctness and calibration from multiple model-generated responses, enabling calibration learning at test time without ground-truth labels. Theoretical analysis further establishes TTCL as a bounded surrogate for the ideal calibration objective. Extensive experiments on mathematical reasoning and factual question answering demonstrate that TTCL consistently improves both accuracy and calibration across diverse models and tasks. On base models, TTCL achieves an average relative accuracy improvement of +40.13% and an ECE reduction of +70.80% across eight benchmarks. Moreover, TTCL can further improve both accuracy and calibration for already calibrated models under domain shift, particularly when source-domain calibration transfers poorly to target tasks. In the math-to-factQA setting, TTCL achieves an average relative accuracy gain of +20.35% and reduces ECE by +53.83%. The source code is released at this https URL.
| Comments: | 34 pages |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.02695 [cs.LG] |
| (or arXiv:2610.02695v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.02695 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Zizhuo Zhang [view email]
[v1]
Fri, 2 Oct 2026 02:17:17 UTC (8,161 KB)
来源：arXiv：cs.LG · arxiv.org