MALA-within-Gibbs Samplers for High-Dimensional Distributions with Sparse Conditional Structure

Markov chain Monte Carlo (MCMC) samplers are numerical methods for drawing samples from a given target probability distribution. We discuss one particular MCMC sampler, the MALA-within-Gibbs sampler, from the theoretical and practical perspectives. We first show that the acceptance ratio and step size of this sampler are independent of the overall problem dimension when (i) the target distribution has sparse conditional structure, and (ii) this structure is reflected in the partial updating strategy of MALA-within-Gibbs. If, in addition, the target density is block-wise log-concave, then the sampler's convergence rate is independent of dimension. From a practical perspective, we expect that MALA-within-Gibbs is useful for solving high-dimensional Bayesian inference problems where the posterior exhibits sparse conditional structure at least approximately. In this context, a partitioning of the state that correctly reflects the sparse conditional structure must be found, and we illustrate this process in two numerical examples. We also discuss trade-offs between the block size used for partial updating and computational requirements that may increase with the number of blocks.

[1]  A. Chorin,et al.  Stochastic Tools in Mathematics and Science , 2005 .

[2]  Jonathan Goodman,et al.  Random-walk interpretations of classical iteration methods , 1995 .

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

[4]  Andrew M. Stuart,et al.  Uncertainty Quantification and Weak Approximation of an Elliptic Inverse Problem , 2011, SIAM J. Numer. Anal..

[5]  Gareth Roberts,et al.  Optimal scalings for local Metropolis--Hastings chains on nonproduct targets in high dimensions , 2009, 0908.0865.

[6]  A. Stuart,et al.  Spectral gaps for a Metropolis–Hastings algorithm in infinite dimensions , 2011, 1112.1392.

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

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

[9]  Shiwei Lan,et al.  Adaptive dimension reduction to accelerate infinite-dimensional geometric Markov Chain Monte Carlo , 2018, J. Comput. Phys..

[10]  Tiangang Cui,et al.  Scalable Optimization-Based Sampling on Function Space , 2019, SIAM J. Sci. Comput..

[11]  A. Sokal,et al.  Multi-grid Monte Carlo (II). Two-dimensional XY model , 1991 .

[12]  Ulli Wolff,et al.  Monte Carlo errors with less errors , 2004 .

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

[14]  Nan Chen,et al.  Rigorous Analysis for Efficient Statistically Accurate Algorithms for Solving Fokker-Planck Equations in Large Dimensions , 2017, SIAM/ASA J. Uncertain. Quantification.

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

[16]  Yuxin Chen,et al.  Accelerated Dimension-Independent Adaptive Metropolis , 2015, SIAM J. Sci. Comput..

[17]  James Martin,et al.  A Computational Framework for Infinite-Dimensional Bayesian Inverse Problems Part I: The Linearized Case, with Application to Global Seismic Inversion , 2013, SIAM J. Sci. Comput..

[18]  J. Bear,et al.  Modeling groundwater flow and pollution , 1987 .

[19]  J. Rosenthal,et al.  Optimal scaling of discrete approximations to Langevin diffusions , 1998 .

[20]  Nan Chen,et al.  Filtering Nonlinear Turbulent Dynamical Systems through Conditional Gaussian Statistics , 2016 .

[21]  Tiangang Cui,et al.  Certified dimension reduction in nonlinear Bayesian inverse problems , 2018, Math. Comput..

[22]  Goodman,et al.  Multigrid Monte Carlo method for lattice field theories. , 1986, Physical review letters.

[23]  J. M. Sanz-Serna,et al.  Optimal tuning of the hybrid Monte Carlo algorithm , 2010, 1001.4460.

[24]  Alexandros Beskos.  A stable manifold MCMC method for high dimensions , 2014 .

[25]  A. Sokal.  Monte Carlo Methods in Statistical Mechanics: Foundations and New Algorithms , 1997 .

[26]  N. Pillai,et al.  A Function Space HMC Algorithm With Second Order Langevin Diffusion Limit , 2013, 1308.0543.

[27]  Matthias Morzfeld,et al.  Localization for MCMC: sampling high-dimensional posterior distributions with local structure , 2017, J. Comput. Phys..

[28]  Goodman,et al.  Multigrid Monte Carlo method. Conceptual foundations. , 1989, Physical review. D, Particles and fields.

[29]  Colin Fox,et al.  Fast Sampling in a Linear-Gaussian Inverse Problem , 2015, SIAM/ASA J. Uncertain. Quantification.

[30]  Colin J. Cotter,et al.  Probabilistic Forecasting and Bayesian Data Assimilation , 2015 .

[31]  Alicia A. Johnson,et al.  Component-Wise Markov Chain Monte Carlo: Uniform and Geometric Ergodicity under Mixing and Composition , 2009, 0903.0664.

[32]  C. Fox,et al.  Accelerated Gibbs sampling of normal distributions using matrix splittings and polynomials , 2015, 1505.03512.

[33]  Mylène Bédard,et al.  Hierarchical models: Local proposal variances for RWM-within-Gibbs and MALA-within-Gibbs , 2017, Comput. Stat. Data Anal..

[34]  A. Weerts,et al.  Data assimilation methods for Delft-FEWS , 2003 .

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

[36]  Nan Chen,et al.  Conditional Gaussian Systems for Multiscale Nonlinear Stochastic Systems: Prediction, State Estimation and Uncertainty Quantification , 2018, Entropy.

[37]  Sebastian J. Vollmer,et al.  Dimension-Independent MCMC Sampling for Inverse Problems with Non-Gaussian Priors , 2013, SIAM/ASA J. Uncertain. Quantification.

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

[39]  G. Roberts,et al.  OPTIMAL SCALING FOR PARTIALLY UPDATING MCMC ALGORITHMS , 2006, math/0607054.

[40]  J. Rosenthal,et al.  Scaling limits for the transient phase of local Metropolis–Hastings algorithms , 2005 .

[41]  Youssef M. Marzouk,et al.  Inference via Low-Dimensional Couplings , 2017, J. Mach. Learn. Res..

[42]  J. Møller,et al.  Log Gaussian Cox Processes , 1998 .

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

[44]  James Martin,et al.  A Computational Framework for Infinite-Dimensional Bayesian Inverse Problems, Part II: Stochastic Newton MCMC with Application to Ice Sheet Flow Inverse Problems , 2013, SIAM J. Sci. Comput..

[45]  Ning Liu,et al.  Inverse Theory for Petroleum Reservoir Characterization and History Matching , 2008 .

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

[47]  Matthias Morzfeld,et al.  Parameter estimation by implicit sampling , 2013, 1308.4640.

[48]  Xin T. Tong,et al.  Spatial Localization for Nonlinear Dynamical Stochastic Models for Excitable Media , 2019, Chinese Annals of Mathematics, Series B.