Advanced search
1 file | 11.29 MB Add to list

Deep generative models in inversion : the impact of the generator's nonlinearity and development of a new approach based on a variational autoencoder

Author
Organization
Abstract
When solving inverse problems in geophysical imaging, deep generative models (DGMs) may be used to enforce the solution to display highly structured spatial patterns which are supported by independent information (e.g. the geological setting) of the subsurface. In such case, inversion may be formulated in a latent space where a lowdimensional parameterization of the patterns is defined and where Markov chain Monte Carlo or gradient-based methods may be applied. However, the generative mapping between the latent and the original (pixel) representations is usually highly nonlinear which may cause some difficulties for inversion, especially for gradientbased methods. In this contribution we review the conceptual framework of inversion with DGMs and propose that this nonlinearity is caused mainly by changes in topology and curvature induced by the generative function. As a result, we identify a conflict between two goals: the accuracy of the generated patterns and the feasibility of gradient-based inversion. In addition, we show how some of the training parameters of a variational autoencoder, which is a particular instance of a DGM, may be chosen so that a tradeoff between these two goals is achieved and acceptable inversion results are obtained with a stochastic gradient-descent scheme. A series of test cases using synthetic models with channel patterns of different complexity and cross-borehole traveltime tomographic data involving both a linear and a nonlinear forward operator show that the proposed method provides useful results and performs better compared to previous approaches using DGMs with gradient-based inversion.
Keywords
Computers in Earth Sciences, Information Systems, Log imputation, Random forest, Feature importance, Prediction interval, SEISMIC DATA, PREDICTION, CLASSIFICATION, RESERVOIR, VELOCITY, Deep generative model, Geophysical inversion, Geological prior information, Variational autoencoder, Stochastic gradient descent, Deep learning, SIMULATION

Downloads

  • (...).pdf
    • full text (Published version)
    • |
    • UGent only
    • |
    • PDF
    • |
    • 11.29 MB

Citation

Please use this url to cite or link to this publication:

MLA
Lopez Alvis, Jorge, et al. “Deep Generative Models in Inversion : The Impact of the Generator’s Nonlinearity and Development of a New Approach Based on a Variational Autoencoder.” COMPUTERS & GEOSCIENCES, vol. 152, 2021, doi:10.1016/j.cageo.2021.104762.
APA
Lopez Alvis, J., Laloy, E., Nguyen, F., & Hermans, T. (2021). Deep generative models in inversion : the impact of the generator’s nonlinearity and development of a new approach based on a variational autoencoder. COMPUTERS & GEOSCIENCES, 152. https://doi.org/10.1016/j.cageo.2021.104762
Chicago author-date
Lopez Alvis, Jorge, Eric Laloy, Frédéric Nguyen, and Thomas Hermans. 2021. “Deep Generative Models in Inversion : The Impact of the Generator’s Nonlinearity and Development of a New Approach Based on a Variational Autoencoder.” COMPUTERS & GEOSCIENCES 152. https://doi.org/10.1016/j.cageo.2021.104762.
Chicago author-date (all authors)
Lopez Alvis, Jorge, Eric Laloy, Frédéric Nguyen, and Thomas Hermans. 2021. “Deep Generative Models in Inversion : The Impact of the Generator’s Nonlinearity and Development of a New Approach Based on a Variational Autoencoder.” COMPUTERS & GEOSCIENCES 152. doi:10.1016/j.cageo.2021.104762.
Vancouver
1.
Lopez Alvis J, Laloy E, Nguyen F, Hermans T. Deep generative models in inversion : the impact of the generator’s nonlinearity and development of a new approach based on a variational autoencoder. COMPUTERS & GEOSCIENCES. 2021;152.
IEEE
[1]
J. Lopez Alvis, E. Laloy, F. Nguyen, and T. Hermans, “Deep generative models in inversion : the impact of the generator’s nonlinearity and development of a new approach based on a variational autoencoder,” COMPUTERS & GEOSCIENCES, vol. 152, 2021.
@article{8703586,
  abstract     = {{When solving inverse problems in geophysical imaging, deep generative models (DGMs) may be used to enforce the solution to display highly structured spatial patterns which are supported by independent information (e.g. the geological setting) of the subsurface. In such case, inversion may be formulated in a latent space where a lowdimensional parameterization of the patterns is defined and where Markov chain Monte Carlo or gradient-based methods may be applied. However, the generative mapping between the latent and the original (pixel) representations is usually highly nonlinear which may cause some difficulties for inversion, especially for gradientbased methods. In this contribution we review the conceptual framework of inversion with DGMs and propose that this nonlinearity is caused mainly by changes in topology and curvature induced by the generative function. As a result, we identify a conflict between two goals: the accuracy of the generated patterns and the feasibility of gradient-based inversion. In addition, we show how some of the training parameters of a variational autoencoder, which is a particular instance of a DGM, may be chosen so that a tradeoff between these two goals is achieved and acceptable inversion results are obtained with a stochastic gradient-descent scheme. A series of test cases using synthetic models with channel patterns of different complexity and cross-borehole traveltime tomographic data involving both a linear and a nonlinear forward operator show that the proposed method provides useful results and performs better compared to previous approaches using DGMs with gradient-based inversion.}},
  articleno    = {{104762}},
  author       = {{Lopez Alvis, Jorge and Laloy, Eric and Nguyen, Frédéric and Hermans, Thomas}},
  issn         = {{0098-3004}},
  journal      = {{COMPUTERS & GEOSCIENCES}},
  keywords     = {{Computers in Earth Sciences,Information Systems,Log imputation,Random forest,Feature importance,Prediction interval,SEISMIC DATA,PREDICTION,CLASSIFICATION,RESERVOIR,VELOCITY,Deep generative model,Geophysical inversion,Geological prior information,Variational autoencoder,Stochastic gradient descent,Deep learning,SIMULATION}},
  language     = {{eng}},
  pages        = {{18}},
  title        = {{Deep generative models in inversion : the impact of the generator's nonlinearity and development of a new approach based on a variational autoencoder}},
  url          = {{http://doi.org/10.1016/j.cageo.2021.104762}},
  volume       = {{152}},
  year         = {{2021}},
}

Altmetric
View in Altmetric
Web of Science
Times cited: