本文为EPFL（作者：Luca Zampieri）的硕士论文，共67页。

本硕士论文以Navier-Stokes方程为重点，探讨如何将几何深度学习应用于数值模拟领域。随着最近深度学习的成功应用，在流体模拟领域也具有了实验的空间。在这里我们设计这样一个实验，我们提出了一种端到端可微的体系结构，允许流体模拟的对象到网格预测。我们提供了与三个不同数据集的基线和视觉结果的比较：机翼、后向台阶和无人机。

This master thesis explores ways to applygeometric deep learning to the field of numerical simulations with an emphasison the Navier-Stokes equations. With the recent success of Deep Learning, thereshould be room for experimentation also in the field of fluid simulations. Herewe lead such an experiment. We propose an end-to-end differentiablearchitecture that allow object-to-mesh predictions of fluid simulations. Weprovide a comparison with a baseline and visual results on three differentdatasets: airfoils, backward facing steps and winged drones.

1. 引言

2. 数字模拟与图深度学习

3. 体系结构

4. 实验

5. 结论与未来工作展望

更多精彩文章请关注公众号：