Scalable Optimization-Based Sampling on Function Space

Optimization-based samplers such as randomize-then-optimize (RTO) [2] provide an efficient and parallellizable approach to solving large-scale Bayesian inverse problems. These methods solve randomly perturbed optimization problems to draw samples from an approximate posterior distribution. "Correcting" these samples, either by Metropolization or importance sampling, enables characterization of the original posterior distribution. This paper focuses on the scalability of RTO to problems with high- or infinite-dimensional parameters. We introduce a new subspace acceleration strategy that makes the computational complexity of RTO scale linearly with the parameter dimension. This subspace perspective suggests a natural extension of RTO to a function space setting. We thus formalize a function space version of RTO and establish sufficient conditions for it to produce a valid Metropolis-Hastings proposal, yielding dimension-independent sampling performance. Numerical examples corroborate the dimension-independence of RTO and demonstrate sampling performance that is also robust to small observational noise.

[1]  T. Coleman,et al.  On the Convergence of Reflective Newton Methods for Large-scale Nonlinear Minimization Subject to Bounds , 1992 .

[2]  R. Tweedie,et al.  Exponential convergence of Langevin distributions and their discrete approximations , 1996 .

[3]  Andrew M. Stuart,et al.  Inverse problems: A Bayesian perspective , 2010, Acta Numerica.

[4]  Tiangang Cui,et al.  Likelihood-informed dimension reduction for nonlinear inverse problems , 2014, 1403.4680.

[5]  Zheng Wang,et al.  Bayesian Inverse Problems with l1 Priors: A Randomize-Then-Optimize Approach , 2016, SIAM J. Sci. Comput..

[6]  Stephen J. Wright,et al.  Numerical Optimization , 2018, Fundamental Statistical Inference.

[7]  Nathan Halko,et al.  Finding Structure with Randomness: Probabilistic Algorithms for Constructing Approximate Matrix Decompositions , 2009, SIAM Rev..

[8]  A. Gelman,et al.  Weak convergence and optimal scaling of random walk Metropolis algorithms , 1997 .

[9]  G. Roberts,et al.  MCMC Methods for Functions: ModifyingOld Algorithms to Make Them Faster , 2012, 1202.0709.

[10]  E. Somersalo,et al.  Statistical and computational inverse problems , 2004 .

[11]  T. J. Dodwell,et al.  A Hierarchical Multilevel Markov Chain Monte Carlo Algorithm with Applications to Uncertainty Quantification in Subsurface Flow , 2013, SIAM/ASA J. Uncertain. Quantification.

[12]  M. Girolami,et al.  Riemann manifold Langevin and Hamiltonian Monte Carlo methods , 2011, Journal of the Royal Statistical Society: Series B (Statistical Methodology).

[13]  Tiangang Cui,et al.  Dimension-independent likelihood-informed MCMC , 2014, J. Comput. Phys..

[14]  Dean S. Oliver,et al.  Metropolized Randomized Maximum Likelihood for Improved Sampling from Multimodal Distributions , 2015, SIAM/ASA J. Uncertain. Quantification.

[15]  Jonathan C. Mattingly,et al.  SPDE limits of the random walk Metropolis algorithm in high dimensions , 2009 .

[16]  G. Prato.  An Introduction to Infinite-Dimensional Analysis , 2006 .

[17]  Tiangang Cui,et al.  Scalable posterior approximations for large-scale Bayesian inverse problems via likelihood-informed parameter and state reduction , 2015, J. Comput. Phys..

[18]  R. Tweedie,et al.  Rates of convergence of the Hastings and Metropolis algorithms , 1996 .

[19]  Andrew M. Stuart,et al.  Complexity analysis of accelerated MCMC methods for Bayesian inversion , 2012, 1207.2411.

[20]  L. Tierney.  A note on Metropolis-Hastings kernels for general state spaces , 1998 .

[21]  Y. Marzouk,et al.  An introduction to sampling via measure transport , 2016, 1602.05023.

[22]  Youssef Marzouk,et al.  Transport Map Accelerated Markov Chain Monte Carlo , 2014, SIAM/ASA J. Uncertain. Quantification.

[23]  Tiangang Cui,et al.  Optimal Low-rank Approximations of Bayesian Linear Inverse Problems , 2014, SIAM J. Sci. Comput..

[24]  G. Roberts,et al.  MCMC methods for diffusion bridges , 2008 .

[25]  Thomas F. Coleman,et al.  An Interior Trust Region Approach for Nonlinear Minimization Subject to Bounds , 1993, SIAM J. Optim..

[26]  Benjamin Peherstorfer,et al.  Optimal Model Management for Multifidelity Monte Carlo Estimation , 2016, SIAM J. Sci. Comput..

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

[28]  Andrew M. Stuart,et al.  Importance Sampling: Computational Complexity and Intrinsic Dimension , 2015 .

[29]  Omar Ghattas,et al.  A Randomized Maximum A Posteriori Method for Posterior Sampling of High Dimensional Nonlinear Bayesian Inverse Problems , 2016, SIAM J. Sci. Comput..

[30]  Jonathan C. Mattingly,et al.  Diffusion limits of the random walk metropolis algorithm in high dimensions , 2010, 1003.4306.

[31]  Aaron Smith,et al.  Parallel Local Approximation MCMC for Expensive Models , 2016, SIAM/ASA J. Uncertain. Quantification.

[32]  Daniel Rudolf,et al.  On a Generalization of the Preconditioned Crank–Nicolson Metropolis Algorithm , 2015, Found. Comput. Math..

[33]  Ben Calderhead,et al.  A general construction for parallelizing Metropolis−Hastings algorithms , 2014, Proceedings of the National Academy of Sciences.

[34]  Gene H. Golub,et al.  Matrix computations , 1983 .

[35]  Matthias Morzfeld,et al.  Implicit particle filters for data assimilation , 2010, 1005.4002.

[36]  Stefan Heinrich,et al.  Multilevel Monte Carlo Methods , 2001, LSSC.

[37]  Heikki Haario,et al.  Randomize-Then-Optimize: A Method for Sampling from Posterior Distributions in Nonlinear Inverse Problems , 2014, SIAM J. Sci. Comput..

[38]  Matthias Morzfeld,et al.  A random map implementation of implicit filters , 2011, J. Comput. Phys..

[39]  Michael B. Giles,et al.  Multilevel Monte Carlo Path Simulation , 2008, Oper. Res..

[40]  James O. Berger,et al.  Markov chain Monte Carlo-based approaches for inference in computationally intensive inverse problems , 2003 .