Seminar, Hosted by Stanford University Department of Geophysics
Machine Learning-Assisted Full Waveform Inversion: Low-Frequency Extrapolation and Cycle-Skipping Mitigation
Abstract
Full waveform inversion (FWI) estimates subsurface physical properties from seismic data, but its nonlinearity and the scarcity of low-frequency data make it prone to cycle skipping, which often prevents convergence to an accurate subsurface model. In this master's presentation Hamad Alswaidan explores two complementary machine-learning approaches: generative diffusion models that extrapolate the missing low-frequency content of seismic data, and a neural-network method that mitigates cycle skipping by aligning observed and modelled waveforms during inversion, showing how machine learning can complement physics-based inversion and make FWI more robust.
Topics
- Full Waveform Inversion, search World Wide
- seismic data, search World Wide
- Low-Frequency Extrapolation, search World Wide
- cycle skipping, search World Wide
- generative diffusion models, broader topic: diffusion models
- subsurface physical properties, search World Wide
- physics-based inversion, search World Wide