BibTeX
@ARTICLE{
         Liu2025ADB,
       author = "Liu, Feng and Li, Haipeng and Zou, Guangyuan and Li, Junlun",
       title = "Automatic Differentiation-Based Full Waveform Inversion With Flexible Workflows",
       journal = "Journal of Geophysical Research: Machine Learning and Computation",
       volume = "2",
       number = "1",
       pages = "e2024JH000542",
       keywords = "automatic differentiation, full waveform inversion, wave equation, deep prior,
         uncertainty estimation with dropout, flexible workflow",
       doi = "10.1029/2024JH000542",
       abstract = "Abstract Full waveform inversion (FWI) is able to construct high-resolution
         subsurface models by iteratively minimizing discrepancies between observed and simulated seismic
         data. However, its implementation can be rather involved for complex wave equations, objective
         functions, or regularization. Recently, automatic differentiation (AD) has proven to be effective in
         simplifying solutions of various inverse problems, including FWI. In this study, we present an
         open-source AD-based FWI framework (ADFWI), which is designed to simplify the design, development,
         and evaluation of novel approaches in FWI with flexibility. The AD-based framework not only includes
         forword modeling and associated gradient computations for wave equations in various types of media
         from isotropic acoustic to vertically or horizontally transverse isotropic elastic, but also
         incorporates a suite of objective functions, regularization techniques, and optimization algorithms.
         By leveraging state-of-the-art AD, objective functions such as soft dynamic time warping and
         Wasserstein distance, which are difficult to apply in traditional FWI are also easily integrated
         into ADFWI. In addition, ADFWI is integrated with deep learning for implicit model
         reparameterization via neural networks, which not only introduces learned regularization but also
         allows rapid estimation of uncertainty through dropout. To manage high memory demands in large-scale
         inversion associated with AD, the proposed framework adopts strategies such as mini-batch and
         checkpointing. Through tests on synthetic and field data, we demonstrate the novelty, practicality
         and robustness of ADFWI, which can be used to address challenges in FWI and as a workbench for
         prompt experiments and development of new inversion strategies.",
       year = "2025",
       ad_area = "Geophysics",
       ad_tools = "ADFWI"
}