Scientific Computing in Machine Learning for Computational Wave Imaging

Speaker：Songting Luo (Iowa State University)

Time：2026-6-25 9:00

Location：Conference Room C503 at Administration Building at Haiyun Campus

Abstract:

Computational wave imaging is vital for uncovering hidden properties in diverse fields of science and engineering, such as seismic imaging and medical imaging. Machine learning has become a prominent method for these inverse problems. But its efficacy relies largely on labeled data, which in turn requires costly experiments and expertise requirements.  Numerical simulation of the related physical models provides an important alternative for data acquisition. But simulating wave propagation is highly nontrivial, especially for high frequency cases and stiff models. In this talk we will report some recent developments for high fidelity wave simulations, where we conduct sophisticated dispersion relation analysis to select appropriate absorbing potentials to restrict the simulation on a bounded domain, and introduce split exponential integrators that combine exponential integration with high-order operator splitting for efficient wave propagation. Equipped with high-order spatial approximations, the proposed methods are shown to be accurate without sacrificing efficiency, allowing for low sapling density and larger CFL numbers. Both stability analysis and numerical experiments justify the high fidelity of the methods.