University of Texas at Austin

Upcoming Event: PhD Dissertation Defense

Hybrid Misfit Functions and Diffusion Priors for Multi-parameter Seismic Full Waveform Inversion

Yiran Shen, CSEM Ph.D. Candidate

3 – 5PM
Tuesday Apr 14, 2026

POB 4.304

Abstract

This dissertation studies multi-parameter full waveform inversion (FWI), a PDE-constrained optimization framework for recovering subsurface properties from seismic wave measurements. It addresses three central challenges in seismic inversion: cycle-skipping caused by phase mismatch, cross-talk between velocity and impedance parameters, and ill-posedness arising from limited data.
 
To address these challenges, the work integrates sensitivity analysis, optimal transport-based misfit design, operator-splitting methods, and diffusion-based generative priors to develop a more stable and physically realistic inversion framework. These components improve phase alignment, better separate parameter sensitivities, preserve amplitude information, and incorporate geological prior knowledge into the reconstruction process.

Biography

Yiran Shen is a Ph.D. candidate in Computational Science, Engineering, and Mathematics (CSEM) at The University of Texas at Austin, working under the supervision of Prof. Björn Engquist. Her research focuses on full waveform inversion (FWI), PDE-constrained inverse problems, and the integration of modern generative models—particularly diffusion and score-based models—as learned priors and regularizers for ill-posed geophysical imaging problems. She is especially interested in multiscale wave phenomena, optimal transport–based misfit functions, and manifold-based formulations of inverse problems. Yiran’s work bridges applied mathematics, computational geophysics, and scientific machine learning, with applications to seismic imaging and subsurface characterization.

Hybrid Misfit Functions and Diffusion Priors for Multi-parameter Seismic Full Waveform Inversion

Event information

Date
3 – 5PM
Tuesday Apr 14, 2026
Location POB 4.304
Hosted by Bjorn Engquist