About Speakers Schedule INS
2024年人工智能与计算数学会议
Artificial Intelligence and Computational Mathematics Conference

NSNO: Neumann Series Neural Operator for Solving Helmholtz Equations in Inhomogeneous Medium

Speaker

史作强 Zuoqiang Shi , 清华大学 Tsinghua University

Time

17 Mar, 09:15 - 09:45

Abstract

We introduce a novel neural operator based on Neumann series to approximate the map from inhomogeneous coefficients to solutions in Helmholtz equation. In the architecture of the neural network, we combine the idea Neumann series, FNO and U-net such that our network is capable to capture high frequency and multiscale feature. NSNO achieves higher accuracy than FNO in the tests of Helmholtz equation and also gives good reconstruction in inverse scattering problem.