questionstorer/sni
收藏Hugging Face2026-05-15 更新2026-05-31 收录
下载链接:
https://hf-mirror.com/datasets/questionstorer/sni
下载链接
链接失效反馈官方服务:
资源简介:
SNI-Dataset是一个用于训练和评估神经算子在几何泛化方面的2D有限元偏微分方程(PDE)解基准数据集。每个样本是一个完整的有限元方法(FEM)问题,定义在非结构化三角形网格上,包括随机多边形几何、边界条件(Dirichlet和/或Neumann),以及可选的系数场或时间步进参数,并配以求解的解场$u$。该数据集伴随Schwarz Neural Inference(SNI)框架,该框架结合局部算子学习和区域分解方法,以在推理时泛化到未见过的复杂几何。数据集包含五种PDE类型:拉普拉斯方程(Dirichlet边界)、拉普拉斯方程(混合边界)、达西流方程、热传导方程和非线性泊松方程,总计约227,830个样本,用于支持PDE求解中几何泛化的研究。
A benchmark of 2D finite-element PDE solutions for training and evaluating neural operators with geometry generalization. Each sample is a complete FEM problem defined on an unstructured triangular mesh—random polygon geometry, boundary conditions (Dirichlet and/or Neumann), and optionally coefficient fields or time-stepping parameters—paired with the solved solution field $u$. The dataset accompanies the Schwarz Neural Inference (SNI) framework, which combines local operator learning with domain decomposition methods to generalize to unseen complex geometries at inference time.
提供机构:
questionstorer


