arXiv cs.AI· Ruo-Qian Wang·· 12 小时前AI 评分31
大语言模型何时是好的校准智能体？物理标注提示用于有预算约束的工程仿真器校准
When is a large language model a good calibration agent? Physics-labeled prompting for budgeted engineering simulator calibration
AI 导读
研究测试了 LLM 在物理诊断描述条件下能否比无导数优化器更高效地提出仿真器参数更新。在圆柱尾流 LES 与阻尼振子逆问题上，首个物理提示的 LLM 提议在 25 次试验中有 23 次改进初始设计，5 次评估后中位校准误差比无物理对照低 21.9 倍，但该优势在 15 次评估内消失，且 LLM 解码器波动带来 9.8 倍性能差异。
正文 · 原文
Abstract:Calibrating expensive engineering simulators requires tuning model parameters to match reference data, a process heavily constrained by computational costs. This study investigates whether Large Language Models (LLMs), conditioned on physical diagnostic descriptions of simulation errors, can propose informative parameter updates more efficiently than conventional derivative-free optimizers. We test this on a cylinder-wake Large-Eddy Simulation and a damped-oscillator inverse problem. Physics-labeled prompts are evaluated against four text ablations (removed, anonymized, padded, misdirected) and standard optimizers (Nelder-Mead, Bayesian optimization, CMA-ES, and random search). Results demonstrate a strong early semantic advantage: the first physics-informed LLM proposal improves the initial design in 23 of 25 trials. After five evaluations, the median calibration error is 21.9 times lower than physics-free controls. However, this advantage is transient - disappearing by fifteen evaluations - and inherent LLM decoder variability yields a 9.8-fold performance spread. To address this, we introduce a hybrid workflow that uses the LLM as a warm-start, passing accumulated evaluations to a conventional optimizer. This two-stage numerical continuation policy reaches target accuracy in just 9 simulator runs, outperforming the LLM alone (12 runs), Nelder-Mead (17 runs), and Bayesian optimization (>60 runs). Supported by sensitivity analysis validating the test cases, this hybrid physics-labeled LLM scheme proves highly effective as a calibration agent under strict evaluation budgets.
| Comments: | Submitted to Engineering Application of Artificial Intelligence |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.22379 [cs.LG] |
| (or arXiv:2610.22379v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.22379 arXiv-issued DOI via DataCite |
Submission history
From: Ruo-Qian Wang [view email]
[v1]
Thu, 1 Oct 2026 16:08:53 UTC (1,305 KB)
来源：arXiv cs.AI · arxiv.org