Applied Math Seminar-Neural Network-assisted methods for inverse random source problems

Time: Friday, Nov. 11th. 3:30-4:30 pm

Place: Zoom

Zoom link: https://gwu-edu.zoom.us/j/95630895609

Speaker: Ying Liang, Purdue University

Title:  Neural Network-assisted methods for inverse random source problems

Abstract: We propose a data-assisted approach for recovering the statistical properties of the random source in the source scattering problems for acoustic wave propagation, using boundary measurement data of random radiated wave fields.  We first implement a regularized Kaczmarz method with the multifrequency scattering data on the boundary to construct approximations of the source profile. Then we compare the performance of different data-driven algorithms to enhance the approximation.  Such a data-assisted approach provides satisfying reconstruction with much fewer realizations than traditional methods.  Numerical experiments are presented to compare the performance of different Image-to-Image translation algorithms and demonstrate the efficiency of the proposed framework. The resulting reconstruction is shown to be stable with respect to the change of the noise in the observation data.