arXiv：cs.LG（机器学习，全量分类）· Aadi Dash, Lennon J. Shikhman, Michael Galarnyk·· 5 小时前AI 评分28
OperatorCLIP 诊断：文本真能引导神经 PDE 代理模型吗？
Does Text Steer Neural PDE Surrogates? A Controlled Diagnostic with OperatorCLIP
AI 导读
研究用 OperatorCLIP 检验文本条件神经代理模型是否真正利用文本语义，对比无条件的 FNO、固定句子 FiLM 控制和经对比对齐训练的固定任务描述。在 Darcy2D、ShallowWater2D 和 CNS3D 三个任务的三种子实验中，固定条件控制在两个 2D 任务上平均测试误差更低，提示词干预未显示可靠的语义排序。
正文
Abstract:Lower error from a text-conditioned neural surrogate does not, by itself, show that the model uses the meaning of the text. We examine this attribution problem with OperatorCLIP, comparing an unconditioned FNO, a constant-sentence FiLM control, and a fixed task description trained with contrastive alignment. Three-seed experiments cover Darcy2D, ShallowWater2D, and three-dimensional compressible Navier-Stokes (CNS3D). Constant conditioning has lower mean test error on both 2D tasks. Relative to this control, task text plus alignment has a similar mean on ShallowWater2D and CNS3D and a higher mean on Darcy2D; these descriptive comparisons have substantial seed uncertainty. The latter comparison changes both prompt content and loss, so it isolates neither effect. The text encoder is trained from scratch, and each conditioned model sees only one description during training. In this regime, pairwise InfoNCE cannot identify matched pairs and has minimum $\log B$. Prompt interventions show no reliable semantic ordering. This methodological caution demonstrates why pathway controls are needed; it neither establishes semantic competence of the encoder nor tests the effectiveness of text under varying physical context.
| Comments: | 8 pages, 2 figures, 2 tables. Accepted to NeurIPS 2026 Workshop on Representation for the Physical Sciences |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2609.38517 [cs.LG] |
| (or arXiv:2609.38517v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2609.38517 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Lennon Shikhman [view email]
[v1]
Tue, 29 Sep 2026 20:37:15 UTC (46 KB)
来源：arXiv：cs.LG（机器学习，全量分类） · arxiv.org