PINNup: Robust Neural Network Wavefield Solutions Using Frequency Upscaling and Neuron Splitting

Huang, Xinquan; Alkhalifah, Tariq

doi:10.1029/2021jb023703

Cited by 43 publications

(9 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For instance, it is possible to perform IFWI on a coarse grid and then reconstruct the subsurface model in a fine grid, or perform IFWI on several discrete models and then generate a continuous subsurface model. In addition, one can implement the inversion in a fully mesh‐free and target‐oriented manner by means of suitable mesh‐free forward modeling algorithms, such as the PINN solver (Huang & Alkhalifah, 2022).…”

Section: Discussionmentioning

confidence: 99%

Implicit Seismic Full Waveform Inversion With Deep Neural Representation

et al. 2023

View full text Add to dashboard Cite

Full waveform inversion (FWI) is arguably the current state‐of‐the‐art amongst methodologies for imaging subsurface structures and physical parameters with seismic data; however, important challenges are faced in its implementation and use. Keys amongst these are (a) building a suitable initial model, from which a local minimum is unlikely to be reached, and (b) availability of tools for evaluation of uncertainty. An algorithm we refer to as implicit full waveform inversion (IFWI), designed using continuously and implicitly defined deep neural representations, appears in principle to address both of these issues. We observe in IFWI, with its random initialization and deep learning optimization, improved convergence relative to standard FWI model initialization and optimization. Models close to the global minimum, capturing relatively high‐resolution subsurface structures, are obtained. In addition, uncertainty analysis, though not solved in IFWI, is meaningfully addressed by approximating Bayesian inference with the addition of dropout neurons. Numerical experimentation with a range of 2D geological models is suggestive that IFWI exhibits a strong capacity for generalization, and is likely well‐suited for multi‐scale joint geophysical inversion.

show abstract

Section: Discussionmentioning

confidence: 99%

Implicit Seismic Full Waveform Inversion With Deep Neural Representation

et al. 2023

View full text Add to dashboard Cite

show abstract

“…However, it is not obvious that the same network will be able to capture the higher wavenumber components of a different velocity model as that is determined by the expressivity of the current architecture. Neuron splitting offers an opportunity to expand the size of the network while utilizing the learned features 56 . It might provide a path for capturing higher resolution information introduced into the updated velocity model.…”

Section: Discussionmentioning

confidence: 99%

A neural network based global traveltime function (GlobeNN)

Taufik

Waheed

Alkhalifah

2023

Sci Rep

View full text Add to dashboard Cite

Global traveltime modeling is an essential component of modern seismological studies with a whole gamut of applications ranging from earthquake source localization to seismic velocity inversion. Emerging acquisition technologies like distributed acoustic sensing (DAS) promise a new era of seismological discovery by allowing a high-density of seismic observations. Conventional traveltime computation algorithms are unable to handle virtually millions of receivers made available by DAS arrays. Therefore, we develop GlobeNN—a neural network based traveltime function that can provide seismic traveltimes obtained from the cached realistic 3-D Earth model. We train a neural network to estimate the traveltime between any two points in the global mantle Earth model by imposing the validity of the eikonal equation through the loss function. The traveltime gradients in the loss function are computed efficiently using automatic differentiation, while the P-wave velocity is obtained from the vertically polarized P-wave velocity of the GLAD-M25 model. The network is trained using a random selection of source and receiver pairs from within the computational domain. Once trained, the neural network produces traveltimes rapidly at the global scale through a single evaluation of the network. As a byproduct of the training process, we obtain a neural network that learns the underlying velocity model and, therefore, can be used as an efficient storage mechanism for the huge 3-D Earth velocity model. These exciting features make our proposed neural network based global traveltime computation method an indispensable tool for the next generation of seismological advances.

show abstract

“…Kaur et al (2021), X. Alkhalifah (2021), andRasht-Behesht et al (2022) improve the efficiency of seismic inversion by accelerating the wave-equation simulation process. Chen and Saygin (2021) propose to utilize latent-space representation of the seismic data compressed by the convolutional autoencoder.…”

Section: Geophysical Inversionmentioning

confidence: 99%

“…Kaur et al. (2021), X. Huang and Alkhalifah (2021), and Rasht‐Behesht et al. (2022) improve the efficiency of seismic inversion by accelerating the wave‐equation simulation process.…”

Section: Highlightsmentioning

confidence: 99%

Machine Learning Developments and Applications in Solid‐Earth Geosciences: Fad or Future?

O’Malley

Beroza

et al. 2023

JGR Solid Earth

View full text Add to dashboard Cite

After decades of low but continuing activity, applications of machine learning (ML) in solid Earth geoscience have exploded in popularity. This special collection provides a snapshot of those applications, which range from data processing to inversion and interpretation, for which ML appears particularly well suited. Inevitably, there are variations in the degree to which these methods have been developed. We hope that the progress seen in some areas will inspire efforts in others. Challenges remain, including the formidable task of how geoscience can keep pace with developments in ML while ensuring the scientific rigor that our field depends on, but with improvements in sensor technology and accelerating rates of data accumulation, the methods of ML seem poised to play an important role for the foreseeable future.

show abstract

PINNup: Robust Neural Network Wavefield Solutions Using Frequency Upscaling and Neuron Splitting

Cited by 43 publications

References 17 publications

Implicit Seismic Full Waveform Inversion With Deep Neural Representation

Implicit Seismic Full Waveform Inversion With Deep Neural Representation

A neural network based global traveltime function (GlobeNN)

Machine Learning Developments and Applications in Solid‐Earth Geosciences: Fad or Future?

Contact Info

Product

Resources

About