arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2012.02334cs.LGcs.AIcs.SYeess.SYmath.DS
收藏

面向从数据中学习动力学的能量守恒神经网络基准测试

Benchmarking Energy-Conserving Neural Networks for Learning Dynamics from Data

Siemens Corporation(西门子公司)

Yaofeng Desmond Zhong, Biswadip Dey, Amit Chakraborty

AI总结:

该研究针对从时序数据学习动力学的能量守恒需求,调研10种相关神经网络模型,推导其理论、分析异同并在4个物理系统中对比性能,还提出可借此设计基于能量的控制器。

摘要:

近年来,人们对在深度学习框架中融入物理 informed 归纳偏置的兴趣日益浓厚。具体而言,越来越多的文献在探索利用神经网络从观测时间序列数据中学习动力学时,强制实现能量守恒的方法。本研究中,我们调研了10个近期提出的能量守恒神经网络模型,包括HNN、LNN、DeLaN、SymODEN、CHNN、CLNN及其变体。我们对这些模型背后的理论进行了简明推导,并阐释了它们的异同。我们在4个物理系统中对比了它们的性能,同时指出了利用其中部分能量守恒模型设计基于能量的控制器的可能性。

英文摘要:

The last few years have witnessed an increased interest in incorporating physics-informed inductive bias in deep learning frameworks. In particular, a growing volume of literature has been exploring ways to enforce energy conservation while using neural networks for learning dynamics from observed time-series data. In this work, we survey ten recently proposed energy-conserving neural network models, including HNN, LNN, DeLaN, SymODEN, CHNN, CLNN and their variants. We provide a compact derivation of the theory behind these models and explain their similarities and differences. Their performance are compared in 4 physical systems. We point out the possibility of leveraging some of these energy-conserving models to design energy-based controllers.

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。

↑