Stability of Gibbs Posteriors from the Wasserstein Loss for Bayesian Full Waveform Inversion

Recently, the Wasserstein loss function has been proven to be effective when applied to deterministic full-waveform inversion (FWI) problems. We consider the application of this loss function in Bayesian FWI so that the uncertainty can be captured in the solution. Other loss functions that are commonly used in practice are also considered for comparison. Existence and stability of the resulting Gibbs posteriors are shown on function space under weak assumptions on the prior and model. In particular, the distribution arising from the Wasserstein loss is shown to be quite stable with respect to high-frequency noise in the data. We then illustrate the difference between the resulting distributions numerically, using Laplace approximations and dimension-robust MCMC to estimate the unknown velocity field and uncertainty associated with the estimates.

[1]  P. Alam.  ‘T’ , 2021, Composites Engineering: An A–Z Guide.

[2]  A. Tarantola.  Inverse problem theory : methods for data fitting and model parameter estimation , 1987 .

[3]  L. Kantorovich.  On the Translocation of Masses , 2006 .

[4]  Brittany D. Froese,et al.  Application of the Wasserstein metric to seismic signals , 2013, 1311.4581.

[5]  Marco A. Iglesias,et al.  A Bayesian Level Set Method for Geometric Inverse Problems , 2015, 1504.00313.

[6]  C. Villani,et al.  Generalization of an Inequality by Talagrand and Links with the Logarithmic Sobolev Inequality , 2000 .

[7]  Andrew M. Stuart,et al.  Geometric MCMC for infinite-dimensional inverse problems , 2016, J. Comput. Phys..

[8]  Andrew J. Majda,et al.  Absorbing Boundary Conditions for Numerical Simulation of Waves , 1977 .

[9]  A. Stuart,et al.  The Bayesian Approach to Inverse Problems , 2013, 1302.6989.

[10]  Jean Virieux,et al.  An overview of full-waveform inversion in exploration geophysics , 2009 .

[11]  T. J. Sullivan,et al.  Well-posed Bayesian inverse problems and heavy-tailed stable quasi-Banach space priors , 2016, 1605.05898.

[12]  Pierre Alquier,et al.  On the properties of variational approximations of Gibbs posteriors , 2015, J. Mach. Learn. Res..

[13]  Romain Brossier,et al.  Regularized seismic full waveform inversion with prior model information , 2013 .

[14]  Brittany D. Froese,et al.  Optimal Transport for Seismic Full Waveform Inversion , 2016, 1602.01540.

[15]  C. Robert,et al.  Inference in generative models using the Wasserstein distance , 2017, 1701.05146.

[16]  Luis Tenorio,et al.  Prior information and uncertainty in inverse problems , 2001 .

[17]  J. Virieux,et al.  Measuring the misfit between seismograms using an optimal transport distance: application to full waveform inversion , 2016 .

[18]  Yunan Yang,et al.  Analysis of optimal transport and related misfit functions in full-waveform inversion , 2017 .

[19]  Rémi Peyre.  Comparison between W2 distance and Ḣ−1 norm, and Localization of Wasserstein distance , 2011, ESAIM: Control, Optimisation and Calculus of Variations.

[20]  J. Scales,et al.  Bayesian seismic waveform inversion: Parameter estimation and uncertainty analysis , 1998 .

[21]  O. Ghattas,et al.  A Bayesian approach to estimate uncertainty for full-waveform inversion using a priori information from depth migration , 2016 .

[22]  Georg Stadler,et al.  Scalable and efficient algorithms for the propagation of uncertainty from data through inference to prediction for large-scale problems, with application to flow of the Antarctic ice sheet , 2014, J. Comput. Phys..

[23]  Jonas Latz,et al.  On the Well-posedness of Bayesian Inverse Problems , 2019, SIAM/ASA J. Uncertain. Quantification.

[24]  Marco A. Iglesias,et al.  Hierarchical Bayesian level set inversion , 2016, Statistics and Computing.

[25]  The quadratic Wasserstein metric for inverse data matching , 2019, ArXiv.

[26]  Yunan Yang,et al.  Analysis and Application of Optimal Transport For Challenging Seismic Inverse Problems , 2019, 1902.01226.

[27]  Ryan Martin,et al.  Calibrating general posterior credible regions , 2015, Biometrika.

[28]  Peter Mora,et al.  Inversion = Migration + Tomography , 1988, Shell Conference.

[29]  Bjorn Sprungk,et al.  On the convergence of the Laplace approximation and noise-level-robustness of Laplace-based Monte Carlo methods for Bayesian inverse problems , 2019, Numerische Mathematik.

[30]  M. Sacchi,et al.  Simultaneous seismic data denoising and reconstruction via multichannel singular spectrum analysis , 2011 .

[31]  C. Villani.  Topics in Optimal Transportation , 2003 .

[32]  Youssef M. Marzouk,et al.  Bayesian inference with optimal maps , 2011, J. Comput. Phys..

[33]  Andreas Fichtner,et al.  Resolution analysis in full waveform inversion , 2011 .

[34]  Andrew M. Stuart,et al.  Reconciling Bayesian and Total Variation Methods for Binary Inversion , 2017 .

[35]  Bjorn Engquist,et al.  Application of optimal transport and the quadratic Wasserstein metric to full-waveform inversion , 2016, 1612.05075.

[36]  Giuseppe Savaré,et al.  A new class of transport distances between measures , 2008, 0803.1235.

[37]  Mohammad Motamed,et al.  WASSERSTEIN METRIC-DRIVEN BAYESIAN INVERSION WITH APPLICATIONS TO SIGNAL PROCESSING , 2018, International Journal for Uncertainty Quantification.

[38]  Matthew M. Dunlop.  Multiplicative noise in Bayesian inverse problems: Well-posedness and consistency of MAP estimators , 2019, 1910.14632.

[39]  Marco Iglesias,et al.  Bayesian inversion in resin transfer molding , 2017, Inverse Problems.

[40]  Tristan van Leeuwen,et al.  Bayesian uncertainty estimation for full waveform inversion: A numerical study , 2019, SEG Technical Program Expanded Abstracts 2019.

[41]  Max Sommerfeld,et al.  Inference for empirical Wasserstein distances on finite spaces , 2016, 1610.03287.