Adaptive Gibbs samplers and related MCMC methods
arXiv:1101.5838 · doi:10.1214/11-AAP806
Abstract
We consider various versions of adaptive Gibbs and Metropolis-within-Gibbs samplers, which update their selection probabilities (and perhaps also their proposal distributions) on the fly during a run by learning as they go in an attempt to optimize the algorithm. We present a cautionary example of how even a simple-seeming adaptive Gibbs sampler may fail to converge. We then present various positive results guaranteeing convergence of adaptive Gibbs samplers under certain conditions.
Published in at http://dx.doi.org/10.1214/11-AAP806 the Annals of Applied Probability (http://www.imstat.org/aap/) by the Institute of Mathematical Statistics (http://www.imstat.org). arXiv admin note: substantial text overlap with arXiv:1001.2797
References in corpus (7)
- On the ergodicity properties of some adaptive MCMC algorithms
- On the ergodicity of the adaptive Metropolis algorithm on unbounded domains
- Weak convergence of Metropolis algorithms for non-i.i.d. target distributions
- On the stability and ergodicity of adaptive scaling Metropolis algorithms
- A Mixture-Based Approach to Regional Adaptation for MCMC
- Regeneration and Fixed-Width Analysis of Markov Chain Monte Carlo Algorithms
- Gibbs Sampling, Exponential Families and Orthogonal Polynomials
Cited by in corpus (9)
- A Survey of Stochastic Simulation and Optimization Methods in Signal Processing
- Component-Wise Markov Chain Monte Carlo: Uniform and Geometric Ergodicity under Mixing and Composition
- The symplectic geometry of closed equilateral random walks in 3-space
- Policy Guided Monte Carlo: Reinforcement Learning Markov Chain Dynamics
- Gradient Scan Gibbs Sampler: an efficient algorithm for high-dimensional Gaussian distributions
- Fast Gibbs sampling for high-dimensional Bayesian inversion
- Stability of adversarial Markov chains, with an application to adaptive MCMC algorithms
- Policy-guided Monte Carlo on general state spaces: Application to glass-forming mixtures
- A Point Mass Proposal Method for Bayesian State-Space Model Fitting