Monday, November 30, 2026 · 9:00 AM – 10:00 AM
Add to calendarFull waveform inversion (FWI) is a powerful technique for estimating subsurface physical properties from seismic data. However, its nonlinear nature and the limited availability of low-frequency seismic data make it susceptible to cycle skipping, often preventing convergence toward an accurate subsurface model.
In this work, Hamad explores two complementary machine learning approaches to address these challenges. The first investigates the use of generative diffusion models to extrapolate missing low-frequency content in seismic data. The second introduces a neural network-based approach to mitigate cycle skipping through the alignment of observed and modeled seismic waveforms during inversion.
Together, these approaches explore how machine learning can complement physics-based inversion and improve the robustness of FWI.
Mitchell Earth Sciences 397 Panama Mall, Stanford, CA 94305 Room 350/372
Monday, November 30, 2026 · 9:00 AM – 10:00 AM