arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2008.04214cs.NEnlin.CD
收藏

用Hamilton神经网络掌握高维动力学

Mastering high-dimensional dynamics with Hamiltonian neural networks

Scott T. Miller, John F. Lindner, Anshul Choudhary, Sudeshna Sinha, William L. Ditto

摘要:

我们详细介绍了如何将物理融入神经网络设计可以显着提高学习和预测的动态系统,甚至是非线性系统的许多维度。一个地图建设的角度阐明了优越性的哈密顿神经网络比传统的神经网络。结果阐明了数据、维数和神经网络学习性能之间的关键关系。

英文摘要:

We detail how incorporating physics into neural network design can significantly improve the learning and forecasting of dynamical systems, even nonlinear systems of many dimensions. A map building perspective elucidates the superiority of Hamiltonian neural networks over conventional neural networks. The results clarify the critical relation between data, dimension, and neural network learning performance.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑