Abstract:The popularity of Adaptive MCMC has been fueled on the one hand by its success in applications, and on the other hand, by mathematically appealing and computationally straightforward optimisation criteria for the Metropolis algorithm acceptance rate (and, equivalently, proposal scale). Similarly principled and operational criteria for optimising the selection probabilities of the Random Scan Gibbs Sampler have not been devised to date. In the present work, we close this gap and develop a general purpose Adaptive Random Scan Gibbs Sampler that adapts the selection probabilities. The adaptation is guided by optimising the $L_2-$spectral gap for the target's Gaussian analogue, gradually, as target's global covariance is learned by the sampler. The additional computational cost of the adaptation represents a small fraction of the total simulation effort. ` We present a number of moderately- and high-dimensional examples, including truncated Gaussians, Bayesian Hierarchical Models and Hidden Markov Models, where significant computational gains are empirically observed for both, Adaptive Gibbs, and Adaptive Metropolis within Adaptive Gibbs version of the algorithm. We argue that Adaptive Random Scan Gibbs Samplers can be routinely implemented and substantial computational gains will be observed across many typical Gibbs sampling problems. We shall give conditions under which ergodicity of the adaptive algorithms can be established.

Geometric ergodicity of Gibbs samplers for Bayesian error-in-variable regression

A Short Note on the Geometric Ergodicity of Markov Chains for Bayesian Linear Regression Models with Heavy-Tailed Errors

Gibbs sampling for a Bayesian hierarchical general linear model

A Short Note on the Efficiency of Markov Chains for Bayesian Linear Regression Models with Heavy-Tailed Errors

Convergence Analysis of MCMC Algorithms for Bayesian Multivariate Linear Regression with Non‐Gaussian Errors

Bayesian variable selection in linear regression models with instrumental variables

Inference from Iterative Simulation Using Multiple Sequences

A uniformly ergodic Gibbs sampler for Bayesian survival analysis

Nonreversible Langevin Samplers: Splitting Schemes, Analysis and Implementation

A variational approach to sampling in diffusion processes

Insufficient Gibbs sampling

Geometric ergodicity of trans-dimensional Markov chain Monte Carlo algorithms

A non-asymptotic error analysis for parallel Monte Carlo estimation from many short Markov chains

Gaussian Processes with Errors in Variables: Theory and Computation

Ergodicity of Langevin Dynamics and its Discretizations for Non-smooth Potentials

Subsampling Error in Stochastic Gradient Langevin Diffusions

On the properties of variational approximations of Gibbs posteriors

Convergence analysis of data augmentation algorithms for Bayesian robust multivariate linear regression with incomplete data

Adapting The Gibbs Sampler

Information-theoretic Analysis of the Gibbs Algorithm: An Individual Sample Approach