Posterior computation with the Gibbs zig-zag sampler

Matthias Sachs; Deborshee Sen; Jianfeng Lu; David Dunson

doi:10.1214/22-BA1319

Posterior computation with the Gibbs zig-zag sampler

Matthias Sachs^*, Deborshee Sen, Jianfeng Lu, David Dunson

^*Corresponding author for this work

Mathematics

Research output: Contribution to journal › Article › peer-review

1 Citation (Scopus)

40 Downloads (Pure)

Abstract

An intriguing new class of piecewise deterministic Markov processes (PDMPs) has recently been proposed as an alternative to Markov chain Monte Carlo (MCMC). We propose a new class of PDMPs termed Gibbs zig-zag samplers, which allow parameters to be updated in blocks with a zig-zag sampler applied to certain parameters and traditional MCMC-style updates to others. We demonstrate the flexibility of this framework on posterior sampling for logistic models with shrinkage priors for high-dimensional regression and random effects, and provide conditions for geometric ergodicity and the validity of a central limit theorem.

Original language	English
Pages (from-to)	909-927
Number of pages	19
Journal	Bayesian Analysis
Volume	18
Issue number	3
Early online date	14 Sept 2022
DOIs	https://doi.org/10.1214/22-BA1319
Publication status	Published - Sept 2023

Bibliographical note

Funding Information:
DS and DD acknowledge support from National Science Foundation grant 1546130. MS and DS acknowledge support from grant DMS-1638521 from SAMSI. The work of JL is supported in part by the National Science Foundation via grants DMS-1454939 and CCF-1934964 (Duke TRIPODS).

Publisher Copyright:
© 2023 International Society for Bayesian Analysis

Keywords

Gibbs sampler
Markov chain Monte Carlo
non-reversible
piecewise deterministic Markov process
sub-sampling

ASJC Scopus subject areas

Statistics and Probability
Applied Mathematics

Access to Document

10.1214/22-BA1319Licence: Creative Commons: Attribution (CC BY)

SachsM2022PosteriorFinal published version, 1.03 MBLicence: Creative Commons: Attribution (CC BY)

Cite this

@article{3334505de78f4b36a897538d630e10fc,

title = "Posterior computation with the Gibbs zig-zag sampler",

abstract = "An intriguing new class of piecewise deterministic Markov processes (PDMPs) has recently been proposed as an alternative to Markov chain Monte Carlo (MCMC). We propose a new class of PDMPs termed Gibbs zig-zag samplers, which allow parameters to be updated in blocks with a zig-zag sampler applied to certain parameters and traditional MCMC-style updates to others. We demonstrate the flexibility of this framework on posterior sampling for logistic models with shrinkage priors for high-dimensional regression and random effects, and provide conditions for geometric ergodicity and the validity of a central limit theorem.",

keywords = "Gibbs sampler, Markov chain Monte Carlo, non-reversible, piecewise deterministic Markov process, sub-sampling",

author = "Matthias Sachs and Deborshee Sen and Jianfeng Lu and David Dunson",

note = "Funding Information: DS and DD acknowledge support from National Science Foundation grant 1546130. MS and DS acknowledge support from grant DMS-1638521 from SAMSI. The work of JL is supported in part by the National Science Foundation via grants DMS-1454939 and CCF-1934964 (Duke TRIPODS). Publisher Copyright: {\textcopyright} 2023 International Society for Bayesian Analysis",

year = "2023",

month = sep,

doi = "10.1214/22-BA1319",

language = "English",

volume = "18",

pages = "909--927",

journal = "Bayesian Analysis",

issn = "1936-0975",

publisher = "Carnegie Mellon University",

number = "3",

}

TY - JOUR

T1 - Posterior computation with the Gibbs zig-zag sampler

AU - Sachs, Matthias

AU - Sen, Deborshee

AU - Lu, Jianfeng

AU - Dunson, David

N1 - Funding Information: DS and DD acknowledge support from National Science Foundation grant 1546130. MS and DS acknowledge support from grant DMS-1638521 from SAMSI. The work of JL is supported in part by the National Science Foundation via grants DMS-1454939 and CCF-1934964 (Duke TRIPODS). Publisher Copyright: © 2023 International Society for Bayesian Analysis

PY - 2023/9

Y1 - 2023/9

N2 - An intriguing new class of piecewise deterministic Markov processes (PDMPs) has recently been proposed as an alternative to Markov chain Monte Carlo (MCMC). We propose a new class of PDMPs termed Gibbs zig-zag samplers, which allow parameters to be updated in blocks with a zig-zag sampler applied to certain parameters and traditional MCMC-style updates to others. We demonstrate the flexibility of this framework on posterior sampling for logistic models with shrinkage priors for high-dimensional regression and random effects, and provide conditions for geometric ergodicity and the validity of a central limit theorem.

AB - An intriguing new class of piecewise deterministic Markov processes (PDMPs) has recently been proposed as an alternative to Markov chain Monte Carlo (MCMC). We propose a new class of PDMPs termed Gibbs zig-zag samplers, which allow parameters to be updated in blocks with a zig-zag sampler applied to certain parameters and traditional MCMC-style updates to others. We demonstrate the flexibility of this framework on posterior sampling for logistic models with shrinkage priors for high-dimensional regression and random effects, and provide conditions for geometric ergodicity and the validity of a central limit theorem.

KW - Gibbs sampler

KW - Markov chain Monte Carlo

KW - non-reversible

KW - piecewise deterministic Markov process

KW - sub-sampling

UR - http://www.scopus.com/inward/record.url?scp=85170277523&partnerID=8YFLogxK

U2 - 10.1214/22-BA1319

DO - 10.1214/22-BA1319

M3 - Article

SN - 1936-0975

VL - 18

SP - 909

EP - 927

JO - Bayesian Analysis

JF - Bayesian Analysis

IS - 3

ER -

Posterior computation with the Gibbs zig-zag sampler

Abstract

Bibliographical note

Keywords

ASJC Scopus subject areas

Access to Document

Fingerprint

Cite this