SURF
收藏资源简介:
SURF数据集由苏黎世联邦理工学院创建,旨在评估图神经网络在流体动力学模拟中的泛化能力。该数据集包含7个大规模的2D流体流动模拟数据集,每个数据集至少包含1200个独立数据点。SURF数据集设计用于测试模型在不同环境设置下的适应性,如网格分辨率、参数范围、拓扑结构和动态模拟环境。通过这些数据集,研究者可以评估和比较不同模型在流体动力学预测中的性能和泛化能力,从而推动机器学习在流体模拟领域的应用。
The SURF dataset was created by ETH Zurich to evaluate the generalization capability of graph neural networks (GNNs) in fluid dynamics simulations. This dataset comprises 7 large-scale 2D fluid flow simulation datasets, with each containing no fewer than 1200 independent data points. The SURF dataset is designed to test model adaptability across diverse environmental configurations, including grid resolution, parameter ranges, topological structures and dynamic simulation environments. Using these datasets, researchers can assess and compare the performance and generalization abilities of different models in fluid dynamics prediction, thereby advancing the application of machine learning in the field of fluid simulation.

- 1SURF: A Generalization Benchmark for GNNs Predicting Fluid Dynamics苏黎世联邦理工学院 · 2023年



