Accelerating Asymptotically Exact MCMC for Computationally Intensive Models via Local Approximations
Abstract
We construct a new framework for accelerating Markov chain Monte Carlo in posterior sampling problems where standard methods are limited by the computational cost of the likelihood, or of numerical models embedded therein. Our approach introduces local approximations of these models into the Metropolis–Hastings kernel, borrowing ideas from deterministic approximation theory, optimization, and experimental design. Previous efforts at integrating approximate models into inference typically sacrifice either the sampler’s exactness or efficiency; our work seeks to address these limitations by exploiting useful convergence characteristics of local approximations. We prove the ergodicity of our approximate Markov chain, showing that it samples asymptotically from the exact posterior distribution of interest. We describe variations of the algorithm that employ either local polynomial approximations or local Gaussian process regressors. Our theoretical results reinforce the key observation underlying this article: when the likelihood has some local regularity, the number of model evaluations per Markov chain Monte Carlo (MCMC) step can be greatly reduced without biasing the Monte Carlo average. Numerical experiments demonstrate multiple order-of-magnitude reductions in the number of forward model evaluations used in representative ordinary differential equation (ODE) and partial differential equation (PDE) inference problems, with both synthetic and real data. Supplementary materials for thismore »
- Authors:
-
- Massachusetts Inst. of Technology (MIT), Cambridge, MA (United States). Dept. of Aeronautics and Astronautics
- Harvard Univ., Cambridge, MA (United States). Dept. of Statistics
- Univ. of Ottawa, Ottawa (Canada). Dept. of Mathematics and Statistics
- Publication Date:
- Research Org.:
- Massachusetts Inst. of Technology (MIT), Cambridge, MA (United States)
- Sponsoring Org.:
- USDOE Office of Science (SC)
- OSTI Identifier:
- 1535376
- Grant/Contract Number:
- SC0007099
- Resource Type:
- Accepted Manuscript
- Journal Name:
- Journal of the American Statistical Association
- Additional Journal Information:
- Journal Volume: 111; Journal Issue: 516; Journal ID: ISSN 0162-1459
- Publisher:
- Taylor & Francis
- Country of Publication:
- United States
- Language:
- English
- Subject:
- 97 MATHEMATICS AND COMPUTING; Mathematics
Citation Formats
Conrad, Patrick R., Marzouk, Youssef M., Pillai, Natesh S., and Smith, Aaron. Accelerating Asymptotically Exact MCMC for Computationally Intensive Models via Local Approximations. United States: N. p., 2015.
Web. doi:10.1080/01621459.2015.1096787.
Conrad, Patrick R., Marzouk, Youssef M., Pillai, Natesh S., & Smith, Aaron. Accelerating Asymptotically Exact MCMC for Computationally Intensive Models via Local Approximations. United States. https://doi.org/10.1080/01621459.2015.1096787
Conrad, Patrick R., Marzouk, Youssef M., Pillai, Natesh S., and Smith, Aaron. Wed .
"Accelerating Asymptotically Exact MCMC for Computationally Intensive Models via Local Approximations". United States. https://doi.org/10.1080/01621459.2015.1096787. https://www.osti.gov/servlets/purl/1535376.
@article{osti_1535376,
title = {Accelerating Asymptotically Exact MCMC for Computationally Intensive Models via Local Approximations},
author = {Conrad, Patrick R. and Marzouk, Youssef M. and Pillai, Natesh S. and Smith, Aaron},
abstractNote = {We construct a new framework for accelerating Markov chain Monte Carlo in posterior sampling problems where standard methods are limited by the computational cost of the likelihood, or of numerical models embedded therein. Our approach introduces local approximations of these models into the Metropolis–Hastings kernel, borrowing ideas from deterministic approximation theory, optimization, and experimental design. Previous efforts at integrating approximate models into inference typically sacrifice either the sampler’s exactness or efficiency; our work seeks to address these limitations by exploiting useful convergence characteristics of local approximations. We prove the ergodicity of our approximate Markov chain, showing that it samples asymptotically from the exact posterior distribution of interest. We describe variations of the algorithm that employ either local polynomial approximations or local Gaussian process regressors. Our theoretical results reinforce the key observation underlying this article: when the likelihood has some local regularity, the number of model evaluations per Markov chain Monte Carlo (MCMC) step can be greatly reduced without biasing the Monte Carlo average. Numerical experiments demonstrate multiple order-of-magnitude reductions in the number of forward model evaluations used in representative ordinary differential equation (ODE) and partial differential equation (PDE) inference problems, with both synthetic and real data. Supplementary materials for this article are available online.},
doi = {10.1080/01621459.2015.1096787},
journal = {Journal of the American Statistical Association},
number = 516,
volume = 111,
place = {United States},
year = {Wed Oct 21 00:00:00 EDT 2015},
month = {Wed Oct 21 00:00:00 EDT 2015}
}
Free Publicly Available Full Text
Publisher's Version of Record
Other availability
Search WorldCat to find libraries that may hold this journal
Cited by: 63 works
Citation information provided by
Web of Science
Web of Science
Save to My Library
You must Sign In or Create an Account in order to save documents to your library.
Works referenced in this record:
Local Derivative-Free Approximation of Computationally Expensive Posterior Densities
journal, April 2012
- Bliznyuk, Nikolay; Ruppert, David; Shoemaker, Christine A.
- Journal of Computational and Graphical Statistics, Vol. 21, Issue 2
Markov chain Monte Carlo Using an Approximation
journal, December 2005
- Christen, J. Andrés; Fox, Colin
- Journal of Computational and Graphical Statistics, Vol. 14, Issue 4
Multivariate Adaptive Regression Splines
journal, March 1991
- Friedman, Jerome H.
- The Annals of Statistics, Vol. 19, Issue 1
Bayesian Calibration and Uncertainty Analysis for Computationally Expensive Models Using Optimization and Radial Basis Function Approximation
journal, June 2008
- Bliznyuk, Nikolay; Ruppert, David; Shoemaker, Christine
- Journal of Computational and Graphical Statistics, Vol. 17, Issue 2
Spline-Based Emulators for Radiative Shock Experiments With Measurement Error
journal, June 2013
- Chakraborty, Avishek; Mallick, Bani K.; Mcclarren, Ryan G.
- Journal of the American Statistical Association, Vol. 108, Issue 502
Approximation of Bayesian Inverse Problems for PDEs
journal, January 2010
- Cotter, S. L.; Dashti, M.; Stuart, A. M.
- SIAM Journal on Numerical Analysis, Vol. 48, Issue 1
Bayesian Computation Using Design of Experiments-Based Interpolation Technique
journal, August 2012
- Joseph, V. Roshan
- Technometrics, Vol. 54, Issue 3
Geometric convergence and central limit theorems for multidimensional Hastings and Metropolis algorithms
journal, March 1996
- Roberts, G.
- Biometrika, Vol. 83, Issue 1
MCMC Methods for Functions: Modifying Old Algorithms to Make Them Faster
journal, August 2013
- Cotter, S. L.; Roberts, G. O.; Stuart, A. M.
- Statistical Science, Vol. 28, Issue 3
Smoothing by Local Regression: Principles and Methods
book, January 1996
- Cleveland, William S.; Loader, Clive
- Contributions to Statistics
Introduction to Derivative-Free Optimization
book, January 2009
- Conn, Andrew R.; Scheinberg, Katya; Vicente, Luis N.
Trust Region Methods
book, January 2000
- Conn, Andrew R.; Gould, Nicholas I. M.; Toint, Philippe L.
- MOS-SIAM Series on Optimization
High-Order Collocation Methods for Differential Equations with Random Inputs
journal, January 2005
- Xiu, Dongbin; Hesthaven, Jan S.
- SIAM Journal on Scientific Computing, Vol. 27, Issue 3
Adaptive Construction of Surrogates for the Bayesian Solution of Inverse Problems
journal, January 2014
- Li, Jinglai; Marzouk, Youssef M.
- SIAM Journal on Scientific Computing, Vol. 36, Issue 3
Adaptive Smolyak Pseudospectral Approximations
journal, January 2013
- Conrad, Patrick R.; Marzouk, Youssef M.
- SIAM Journal on Scientific Computing, Vol. 35, Issue 6
Approximate Bayesian computational methods
journal, October 2011
- Marin, Jean-Michel; Pudlo, Pierre; Robert, Christian P.
- Statistics and Computing, Vol. 22, Issue 6
Stochastic spectral methods for efficient Bayesian solution of inverse problems
journal, June 2007
- Marzouk, Youssef M.; Najm, Habib N.; Rahn, Larry A.
- Journal of Computational Physics, Vol. 224, Issue 2
Uncertainty Quantification and Weak Approximation of an Elliptic Inverse Problem
journal, January 2011
- Dashti, M.; Stuart, A. M.
- SIAM Journal on Numerical Analysis, Vol. 49, Issue 6
Preconditioning Markov Chain Monte Carlo Simulations Using Coarse-Scale Models
journal, January 2006
- Efendiev, Y.; Hou, T.; Luo, W.
- SIAM Journal on Scientific Computing, Vol. 28, Issue 2
Local Gaussian Process Approximation for Large Computer Experiments
journal, April 2015
- Gramacy, Robert B.; Apley, Daniel W.
- Journal of Computational and Graphical Statistics, Vol. 24, Issue 2
The pseudo-marginal approach for efficient Monte Carlo computations
journal, April 2009
- Andrieu, Christophe; Roberts, Gareth O.
- The Annals of Statistics, Vol. 37, Issue 2
Sparse pseudospectral approximation method
journal, July 2012
- Constantine, Paul G.; Eldred, Michael S.; Phipps, Eric T.
- Computer Methods in Applied Mechanics and Engineering, Vol. 229-232
A Stochastic Collocation Approach to Bayesian Inference in Inverse Problems
journal, January 2009
- Marzouk, Youssef; Xiu, Dongbin
- Communications in Computational Physics, Vol. 6, Issue 4
Robust Locally Weighted Regression and Smoothing Scatterplots
journal, December 1979
- Cleveland, William S.
- Journal of the American Statistical Association, Vol. 74, Issue 368
The Design and Analysis of Computer Experiments
book, January 2003
- Santner, Thomas J.; Williams, Brian J.; Notz, William I.
- Springer Series in Statistics
Bayesian calibration of computer models
journal, August 2001
- Kennedy, Marc C.; O'Hagan, Anthony
- Journal of the Royal Statistical Society: Series B (Statistical Methodology), Vol. 63, Issue 3
Bayesian inverse problems with Monte Carlo forward models
journal, February 2013
- Marzouk, Youssef; Langmore, Ian; Bal, Guillaume
- Inverse Problems and Imaging, Vol. 7, Issue 1
A Sparse Grid Stochastic Collocation Method for Partial Differential Equations with Random Input Data
journal, January 2008
- Nobile, F.; Tempone, R.; Webster, C. G.
- SIAM Journal on Numerical Analysis, Vol. 46, Issue 5
Approximating likelihoods for large spatial data sets
journal, May 2004
- Stein, Michael L.; Chi, Zhiyi; Welty, Leah J.
- Journal of the Royal Statistical Society: Series B (Statistical Methodology), Vol. 66, Issue 2
Efficient MCMC Schemes for Computationally Expensive Posterior Distributions
journal, February 2011
- Fielding, Mark; Nott, David J.; Liong, Shie-Yui
- Technometrics, Vol. 53, Issue 1
Convergence of adaptive and interacting Markov chain Monte Carlo algorithms
journal, December 2011
- Fort, G.; Moulines, E.; Priouret, P.
- The Annals of Statistics, Vol. 39, Issue 6
Construction of a genetic toggle switch in Escherichia coli
journal, January 2000
- Gardner, Timothy S.; Cantor, Charles R.; Collins, James J.
- Nature, Vol. 403, Issue 6767
Design and Analysis of Computer Experiments
journal, November 1989
- Sacks, Jerome; Welch, William J.; Mitchell, Toby J.
- Statistical Science, Vol. 4, Issue 4
An Adaptive Metropolis Algorithm
journal, April 2001
- Haario, Heikki; Saksman, Eero; Tamminen, Johanna
- Bernoulli, Vol. 7, Issue 2
Bayesian calibration of a large-scale geothermal reservoir model by a new adaptive delayed acceptance Metropolis Hastings algorithm: ADAPTIVE DELAYED ACCEPTANCE METROPOLIS-HASTINGS ALGORITHM
journal, October 2011
- Cui, T.; Fox, C.; O'Sullivan, M. J.
- Water Resources Research, Vol. 47, Issue 10
Parameter and State Model Reduction for Large-Scale Statistical Inverse Problems
journal, January 2010
- Lieberman, Chad; Willcox, Karen; Ghattas, Omar
- SIAM Journal on Scientific Computing, Vol. 32, Issue 5
General state space Markov chains and MCMC algorithms
journal, January 2004
- Roberts, Gareth O.; Rosenthal, Jeffrey S.
- Probability Surveys, Vol. 1, Issue 0
Data Assimilation: A Mathematical Introduction
book, January 2015
- Law, Kody; Stuart, Andrew; Zygalakis, Konstantinos
- Texts in Applied Mathematics
Statistical inverse problems: Discretization, model reduction and inverse crimes
journal, January 2007
- Kaipio, Jari; Somersalo, Erkki
- Journal of Computational and Applied Mathematics, Vol. 198, Issue 2
Design and analysis of computer experiments
conference, September 1998
- Booker, Andrew
- 7th AIAA/USAF/NASA/ISSMO Symposium on Multidisciplinary Analysis and Optimization
The Design and Analysis of Computer Experiments
book, January 2018
- Santner, Thomas J.; Williams, Brian J.; Notz, William I.
- Springer Series in Statistics
Design and analysis of computer experiments
journal, December 2010
- Kuhnt, Sonja; Steinberg, David M.
- AStA Advances in Statistical Analysis, Vol. 94, Issue 4
Local Gaussian Process Approximation for Large Computer Experiments
text, January 2015
- Gramacy, Robert B.; Apley, Daniel W.
- Taylor & Francis
Local Gaussian process approximation for large computer experiments
text, January 2015
- Gramacy, Robert B.; Apley, Daniel W.
- Taylor & Francis
Stochastic spectral methods for efficient Bayesian solution of inverse problems
conference, January 2005
- Marzouk, Youssef M.
- BAYESIAN INFERENCE AND MAXIMUM ENTROPY METHODS IN SCIENCE AND ENGINEERING: 25th International Workshop on Bayesian Inference and Maximum Entropy Methods in Science and Engineering, AIP Conference Proceedings
The pseudo-marginal approach for efficient Monte Carlo computations
text, January 2009
- Andrieu, Christophe; Roberts, Gareth O.
- arXiv
MCMC Methods for Functions: Modifying Old Algorithms to Make Them Faster
text, January 2012
- Cotter, S. L.; Roberts, G. O.; Stuart, A. M.
- arXiv
Adaptive Smolyak Pseudospectral Approximations
preprint, January 2012
- Conrad, Patrick R.; Marzouk, Youssef M.
- arXiv
Bayesian treed Gaussian process models with an application to computer modeling
preprint, January 2007
- Gramacy, Robert B.; Lee, Herbert K. H.
- arXiv
Uncertainty quantification and weak approximation of an elliptic inverse problem
preprint, January 2011
- Dashti, Masoumeh; Stuart, Andrew M.
- arXiv
Austerity in MCMC Land: Cutting the Metropolis-Hastings Budget
preprint, January 2013
- Korattikara, Anoop; Chen, Yutian; Welling, Max
- arXiv
Works referencing / citing this record:
A transport-based multifidelity preconditioner for Markov chain Monte Carlo
journal, November 2019
- Peherstorfer, Benjamin; Marzouk, Youssef
- Advances in Computational Mathematics
Deterministic Sampling of Expensive Posteriors Using Minimum Energy Designs
text, January 2019
- Joseph, V. Roshan; Wang, Dianpeng; Gu, Li
- Taylor & Francis
Deterministic Sampling of Expensive Posteriors Using Minimum Energy Designs
text, January 2019
- Joseph, V. Roshan; Wang, Dianpeng; Gu, Li
- Taylor & Francis
Deterministic Sampling of Expensive Posteriors Using Minimum Energy Designs
journal, March 2019
- Joseph, V. Roshan; Wang, Dianpeng; Gu, Li
- Technometrics, Vol. 61, Issue 3
Bayesian Estimation for Stochastic Gene Expression Using Multifidelity Models
journal, February 2019
- Vo, Huy D.; Fox, Zachary; Baetica, Ania
- The Journal of Physical Chemistry B, Vol. 123, Issue 10
Deterministic Sampling of Expensive Posteriors Using Minimum Energy Designs
text, January 2018
- Joseph, V. Roshan; Wang, Dianpeng; Gu, Li
- Taylor & Francis
Surrogate-Based Bayesian Inverse Modeling of the Hydrological System: An Adaptive Approach Considering Surrogate Approximation Error
text, January 2018
- Zhang, Jiangjiang; Zheng, Qiang; Chen, Dingjiang
- arXiv
A transport-based multifidelity preconditioner for Markov chain Monte Carlo
preprint, January 2018
- Peherstorfer, Benjamin; Marzouk, Youssef
- arXiv