Localization for MCMC: sampling high-dimensional posterior distributions with local structure
From MaRDI portal
Publication:2214525
Abstract: We investigate how ideas from covariance localization in numerical weather prediction can be used in Markov chain Monte Carlo (MCMC) sampling of high-dimensional posterior distributions arising in Bayesian inverse problems. To localize an inverse problem is to enforce an anticipated "local" structure by (i) neglecting small off-diagonal elements of the prior precision and covariance matrices; and (ii) restricting the influence of observations to their neighborhood. For linear problems we can specify the conditions under which posterior moments of the localized problem are close to those of the original problem. We explain physical interpretations of our assumptions about local structure and discuss the notion of high dimensionality in local problems, which is different from the usual notion of high dimensionality in function space MCMC. The Gibbs sampler is a natural choice of MCMC algorithm for localized inverse problems and we demonstrate that its convergence rate is independent of dimension for localized linear problems. Nonlinear problems can also be tackled efficiently by localization and, as a simple illustration of these ideas, we present a localized Metropolis-within-Gibbs sampler. Several linear and nonlinear numerical examples illustrate localization in the context of MCMC samplers for inverse problems.
Recommendations
- Dimension-independent likelihood-informed MCMC
- Low-rank independence samplers in hierarchical Bayesian inverse problems
- Parallel local approximation MCMC for expensive models
- Randomize-then-optimize: a method for sampling from posterior distributions in nonlinear inverse problems
- Ensemble sampler for infinite-dimensional inverse problems
Cites work
- A computational framework for infinite-dimensional Bayesian inverse problems. I: The linearized case, with application to global seismic inversion
- A computational framework for infinite-dimensional Bayesian inverse problems. II: stochastic Newton MCMC with application to ice sheet flow inverse problems
- A general purpose sampling algorithm for continuous distributions (the t-walk)
- A nonparametric ensemble transform method for Bayesian inference
- A stochastic Newton MCMC method for large-scale statistical inverse problems with application to seismic inversion
- Accelerated Gibbs sampling of normal distributions using matrix splittings and polynomials
- An explicit link between Gaussian fields and Gaussian Markov random fields: the stochastic partial differential equation approach
- Approximating the inverse of banded matrices by banded matrices with applications to probability and statistics
- Can local particle filters beat the curse of dimensionality?
- Dimension-independent likelihood-informed MCMC
- Ensemble samplers with affine invariance
- Fast algorithms for Bayesian uncertainty quantification in large-scale linear inverse problems based on low-rank partial Hessian approximations
- Fast sampling in a linear-Gaussian inverse problem
- Gaussian Markov Random Fields
- scientific article; zbMATH DE number 4070125 (Why is no real title available?)
- scientific article; zbMATH DE number 1222285 (Why is no real title available?)
- scientific article; zbMATH DE number 1085989 (Why is no real title available?)
- scientific article; zbMATH DE number 849920 (Why is no real title available?)
- Importance sampling: intrinsic dimension and computational cost
- Inverse problems: a Bayesian perspective
- Likelihood-informed dimension reduction for nonlinear inverse problems
- MCMC methods for functions: modifying old algorithms to make them faster
- MCMC using Hamiltonian dynamics
- MCMC-based image reconstruction with uncertainty quantification
- Monte Carlo errors with less errors
- Nonlinear data assimilation
- Optimal low-rank approximations of Bayesian linear inverse problems
- Optimal Scaling of Discrete Approximations to Langevin Diffusions
- Optimal scalings for local Metropolis-Hastings chains on nonproduct targets in high dimensions
- Optimal tuning of the hybrid Monte Carlo algorithm
- Rate of convergence of the Gibbs sampler in the Gaussian case
- Regularized estimation of large covariance matrices
- Riemann manifold Langevin and Hamiltonian Monte Carlo methods. With discussion and authors' reply
- Scalable posterior approximations for large-scale Bayesian inverse problems via likelihood-informed parameter and state reduction
- Scaling Limits for the Transient Phase of Local Metropolis–Hastings Algorithms
- Spectral gaps for a Metropolis-Hastings algorithm in infinite dimensions
- State estimation and prediction using clustered particle filters
- Weak convergence and optimal scaling of random walk Metropolis algorithms
Cited in
(21)- Spectral gap of replica exchange Langevin diffusion on mixture distributions
- Rate-optimal refinement strategies for local approximation MCMC
- Accelerating Metropolis-within-Gibbs sampler with localized computations of differential equations
- Spatial localization for nonlinear dynamical stochastic models for excitable media
- Dimension-independent likelihood-informed MCMC
- Sparse approximation of triangular transports. I: The finite-dimensional case
- A unified performance analysis of likelihood-informed subspace methods
- MALA-within-Gibbs samplers for high-dimensional distributions with sparse conditional structure
- Parallel local approximation MCMC for expensive models
- Bayesian Inference on Local Distributions of Functions and Multidimensional Curves with Spherical HMC Sampling
- A blocking scheme for dimension-robust Gibbs sampling in large-scale image deblurring
- Rank bounds for approximating Gaussian densities in the tensor-train format
- Convergence acceleration of ensemble Kalman inversion in nonlinear settings
- Analysis of a localised nonlinear ensemble Kalman Bucy filter with complete and accurate observations
- Replica exchange for non-convex optimization
- Localized ensemble Kalman inversion
- Canonical Monte Carlo multispin cluster method
- Chilled sampling for uncertainty quantification: a motivation from a meteorological inverse problem *
- Localized Schrödinger bridge sampler
- Local MALA-within-Gibbs for Bayesian image deblurring with total variation prior
- A localized consensus-based sampling algorithm
This page was built for publication: Localization for MCMC: sampling high-dimensional posterior distributions with local structure
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2214525)
