TL;DR本研究利用 Hamilton 力学来为神经网络提供更好的归纳偏差,使其能够在自我监督的状态下学习并遵守物理中的守恒定律;研究表明我们的模型在能量守恒等问题上具有更快的训练速度和更好的泛化性能,并且是一个完全可逆的时间模型。
Abstract
Even though neural networks enjoy widespread use, they still struggle to
learn the basic laws of physics. How might we endow them with better inductive
biases? In this paper, we draw inspiration from hamiltonian mechani