Rethinking Scientific Modeling: Toward Physically Consistent and Simulation-Executable Programmatic Generation *** Conference: The 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining; 2026; Jeju, South Korea CCS: Computing methodologies Artificial intelligence
Yongqing Jiang, Jianze Wang, Zhiqi Shen, Zhenghong Lin, Jiayuan Wang, Yijian Yang, Kaoshan Dai *, Haoran Luo * · 2026
Read Rethinking Scientific Modeling: Toward Physically Consistent and Simulation-Executable Programmatic Generation *** Conference: The 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining; 2026; Jeju, South Korea CCS: Computing methodologies Artificial intelligence by Yongqing Jiang, Jianze Wang, Zhiqi Shen, Zhenghong Lin, Jiayuan Wang, Yijian Yang, Kaoshan Dai *, Haoran Luo * (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-07083.txt.
Description
Yongqing Jiang and seven co-authors rethink how language models make building models. Building models sit at the heart of computational engineering science, where even small physical flaws or spec violations can void later simulations. Language models can make modeling code, yet under strict engineering rules their outputs often fail to run or break physics. A new framework aims to fix that, joining three parts: building domain knowledge, aligning models to constraints, and checking through verification.
CivilInstruct, a domain-specific data set, formalizes building knowledge and constraint reasoning, enabling simulation-ready generation. A two-stage fine-tuning strategy enforces constraint satisfaction and interface compliance, cutting hallucinated and non-conforming outputs sharply. MBEval, a checking benchmark, scores runnability and structural dynamics consistency through closed-loop checks. Tests show steady gains over baselines across rigorous checking metrics.
Catalog record
| CREATOR | Yongqing Jiang, Jianze Wang, Zhiqi Shen, Zhenghong Lin, Jiayuan Wang, Yijian Yang, Kaoshan Dai *, Haoran Luo * |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 29 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-07083