arXiv：cs.LG（机器学习，全量分类）· Somyajit Chakraborty, Xizhong Chen·· 1 天前AI 评分27
更低预测误差不等于更懂物理？Sim2Real 神经算子物理诊断研究
Do Better Scores Mean Better Physics? Physics-Grounded Explanations for Sim2Real Neural Operators
AI 导读
研究用 NACA4418 翼型绕流的 CFD 与 PIV 配对数据检验神经算子：在四个神经算子上，移除最活跃 10% 有效观测网格的脉动比等面积随机移除对预测影响更大，但因掩码未匹配脉动能量，该对比只反映敏感性。一个 CNO 速度场误差更低，但两分量脉动能量误差显著高于参考模型，输出衰减压力测试也显示基准误差与域求和脉动能量不一致。
正文
Abstract:Machine-learning surrogates accelerate physical simulation, but lower prediction error need not coincide with lower error in physically relevant flow statistics. We examine this question for flow around a NACA4418 airfoil using paired computational-fluid-dynamics simulations and experimental particle-image-velocimetry measurements. A mean-preserving input intervention removes velocity fluctuations from selected regions of observed flow histories. Across four neural operators, removing fluctuations from the most energetic 10% of valid observed cells changes forecasts more than equal-area random removal. Because the masks are not matched for removed fluctuation energy, this contrast measures sensitivity, not independent evidence of physical importance. Separately, a CNO has lower velocity-field error but substantially higher two-component fluctuation-energy error than the reference on both analysis subsets. An output attenuation stress test also demonstrates disagreement between benchmark errors and domain-summed fluctuation energy. These single-benchmark results motivate reporting complementary physical diagnostics alongside aggregate prediction scores; they do not establish counterfactual physical correctness.
| Comments: | 10 pages, 6 figures. Accepted at the NeurIPS 2026 XAI4Science Workshop, Tiny Paper Track |
| Subjects: | Machine Learning (cs.LG); Fluid Dynamics (physics.flu-dyn) |
| Cite as: | arXiv:2610.00415 [cs.LG] |
| (or arXiv:2610.00415v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.00415 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Somyajit Chakraborty Dr. [view email]
[v1]
Wed, 30 Sep 2026 14:19:15 UTC (603 KB)
来源：arXiv：cs.LG（机器学习，全量分类） · arxiv.org