arXiv：cs.AI· Bo Ni·· 23 小时前
FracLearn：让语言模型通过物理经验学习设计
Learning from fracture: Teaching language models to design through physical experience
AI 导读
FracLearn 是一个经验-抽象-泛化框架，在一个受俄罗斯方块启发的断裂游戏中，让语言模型先提出架构并观察模拟断裂响应来积累经验，再抽象为条件性原则并在新架构中组合，最佳综合得分超出经验阶段纪录 11.0%。断裂分析显示性能提升与裂纹偏转、暂时捕获等增韧机制及更高残余承载能力相关，且学习过程不更新模型权重。
正文
Abstract:How can language models learn to optimize physical designs through experience? We introduce FracLearn, an experience-abstraction-generalization framework evaluated through a deliberately unfamiliar, Tetris-inspired fracture game. The model first proposes architectures and examines simulated fracture responses to accumulate experience. Abstraction organizes these observations into conditional principles linking reinforcement placement, crack evolution and load transfer, with supporting evidence and failure conditions. Generalization challenges the model to combine these principles in new architectures and surpass the earlier design record. Campaigns demonstrate progressive refinement and recovery through hypothesis revision, yielding a best composite score 11.0% above the experience-stage record. Fracture analyses connect improved performance to crack deflection and temporary trapping as toughening mechanisms, alongside greater residual load capacity. Learning occurs through external knowledge and decisions, without model-weight updates. This approach offers a route to scientific agents that turn computational experiments into reusable physical knowledge for materials discovery and engineering design.
| Comments: | 21 pages, 7 figures, 3 tables. Includes Supplementary Materials |
| Subjects: | Materials Science (cond-mat.mtrl-sci); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.17804 [cond-mat.mtrl-sci] |
| (or arXiv:2610.17804v1 [cond-mat.mtrl-sci] for this version) | |
| https://doi.org/10.48550/arXiv.2610.17804 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Bo Ni [view email]
[v1]
Tue, 13 Oct 2026 06:30:52 UTC (1,045 KB)
来源：arXiv：cs.AI · arxiv.org