← Back to Sessions

Ambient Noise Full Waveform Inversion With Neural Operators

Description: 

Numerical simulations of seismic wave propagation are central to investigating velocity structures and for improving seismic hazard assessment, but standard methods such as finite element or finite difference are computationally expensive. Recent studies have shown that a new class of machine learning models, called Neural Operators, are able to solve the elastodynamic wave equation orders of magnitude faster than these conventional numerical methods. Full waveform inversion is a prime beneficiary of the accelerated simulations. Neural operators, combined with automatic differentiation, provide an alternative approach to full waveform inversion that does not involve the Born approximation. It thus can potentially overcome some of the cycle skipping problems present in traditional adjoint state formulations. Here we demonstrate the first application of Neural Operators for full waveform inversion on a real seismic dataset: several nodal transects collected across the San Gabriel and San Bernardino Basins in the Los Angeles metropolitan area.

Session: Machine Learning for Full Waveform Inversion: From Hybrid to End-to-End Approaches - I

Type: Oral

Date: 5/3/2024

Presentation Time: 04:45 PM (local time)

Presenting Author: Caifeng

Student Presenter: Yes

Invited Presentation: 

Authors