skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Gradient-free MCMC methods for dynamic causal modelling

Abstract

Here, we compare the performance of four gradient-free MCMC samplers (random walk Metropolis sampling, slice-sampling, adaptive MCMC sampling and population-based MCMC sampling with tempering) in terms of the number of independent samples they can produce per unit computational time. For the Bayesian inversion of a single-node neural mass model, both adaptive and population-based samplers are more efficient compared with random walk Metropolis sampler or slice-sampling; yet adaptive MCMC sampling is more promising in terms of compute time. Slice-sampling yields the highest number of independent samples from the target density -- albeit at almost 1000% increase in computational time, in comparison to the most efficient algorithm (i.e., the adaptive MCMC sampler).

Authors:
 [1];  [1];  [1]
  1. Univ. College London, London (United Kingdom)
Publication Date:
Research Org.:
Univ. College London (United Kingdom)
Sponsoring Org.:
USDOE Office of Science (SC)
OSTI Identifier:
1241923
Alternate Identifier(s):
OSTI ID: 1344386
Grant/Contract Number:
AC05-00OR22725
Resource Type:
Journal Article: Published Article
Journal Name:
NeuroImage
Additional Journal Information:
Journal Volume: 112; Journal ID: ISSN 1053-8119
Publisher:
Elsevier
Country of Publication:
United States
Language:
English
Subject:
74 ATOMIC AND MOLECULAR PHYSICS

Citation Formats

Sengupta, Biswa, Friston, Karl J., and Penny, Will D. Gradient-free MCMC methods for dynamic causal modelling. United States: N. p., 2015. Web. doi:10.1016/j.neuroimage.2015.03.008.
Sengupta, Biswa, Friston, Karl J., & Penny, Will D. Gradient-free MCMC methods for dynamic causal modelling. United States. doi:10.1016/j.neuroimage.2015.03.008.
Sengupta, Biswa, Friston, Karl J., and Penny, Will D. Sat . "Gradient-free MCMC methods for dynamic causal modelling". United States. doi:10.1016/j.neuroimage.2015.03.008.
@article{osti_1241923,
title = {Gradient-free MCMC methods for dynamic causal modelling},
author = {Sengupta, Biswa and Friston, Karl J. and Penny, Will D.},
abstractNote = {Here, we compare the performance of four gradient-free MCMC samplers (random walk Metropolis sampling, slice-sampling, adaptive MCMC sampling and population-based MCMC sampling with tempering) in terms of the number of independent samples they can produce per unit computational time. For the Bayesian inversion of a single-node neural mass model, both adaptive and population-based samplers are more efficient compared with random walk Metropolis sampler or slice-sampling; yet adaptive MCMC sampling is more promising in terms of compute time. Slice-sampling yields the highest number of independent samples from the target density -- albeit at almost 1000% increase in computational time, in comparison to the most efficient algorithm (i.e., the adaptive MCMC sampler).},
doi = {10.1016/j.neuroimage.2015.03.008},
journal = {NeuroImage},
number = ,
volume = 112,
place = {United States},
year = {Sat Mar 14 00:00:00 EDT 2015},
month = {Sat Mar 14 00:00:00 EDT 2015}
}

Journal Article:
Free Publicly Available Full Text
Publisher's Version of Record at 10.1016/j.neuroimage.2015.03.008

Citation Metrics:
Cited by: 9works
Citation information provided by
Web of Science

Save / Share:
  • Here, we compare the performance of four gradient-free MCMC samplers (random walk Metropolis sampling, slice-sampling, adaptive MCMC sampling and population-based MCMC sampling with tempering) in terms of the number of independent samples they can produce per unit computational time. For the Bayesian inversion of a single-node neural mass model, both adaptive and population-based samplers are more efficient compared with random walk Metropolis sampler or slice-sampling; yet adaptive MCMC sampling is more promising in terms of compute time. Slice-sampling yields the highest number of independent samples from the target density -- albeit at almost 1000% increase in computational time, in comparisonmore » to the most efficient algorithm (i.e., the adaptive MCMC sampler).« less
  • Recent air quality studies have highlighted that important differences in pollutant concentrations can occur over the day and between different locations. Traditional exposure analyses, however, assume that people are only exposed to pollution at their place of residence. Activity-based models, which recently have emerged from the field of transportation research, offer a technique to micro-simulate activity patterns of a population with a high resolution in space and time. Due to their characteristics, this model can be applied to establish a dynamic exposure assessment to air pollution. This paper presents a new exposure methodology, using a micro-simulator of activity-travel behaviour, tomore » develop a dynamic exposure assessment. The methodology is applied to a Dutch urban area to demonstrate the advantages of the approach for exposure analysis. The results for the exposure to PM{sub 10} and PM{sub 2.5}, air pollutants considered as hazardous for human health, reveal large differences between the static and the dynamic approach, mainly due to an underestimation of the number of hours spent in the urban region by the static method. We can conclude that this dynamic population modelling approach is an important improvement over traditional methods and offers a new and more sensitive way for estimating population exposure to air pollution. In the light of the new European directive, aimed at reducing the exposure of the population to PM{sub 2.5}, this new approach contributes to a much more accurate exposure assessment that helps evaluate policies to reduce public exposure to air pollution.« less
  • Parallel processing is a promising technology for the speedup of the dynamic simulations required in transient stability analysis. In this paper, three methods for dynamic simulation on parallel computers are described and compared. The methods are based on the concepts of spatial and/or time parallelization. In all of them, sets of linear algebraic equations are solved using different versions of Conjugate Gradient methods which have been successfully applied in other scientific and engineering applications. The algorithms presented in the paper were tested in a commercially available parallel computer using an actual large power system model. The results obtained in themore » tests showed a considerable reduction in computation time.« less
  • A technique for detecting the causal relationship between fluctuating signals is used to investigate the relation between flux and gradient in fusion plasmas. Both a resistive pressure gradient driven turbulence model and experimental Langmuir probe data from the TJ-II stellarator are studied. It is found that the maximum influence occurs at a finite time lag (non-instantaneous response) and that quasi-periodicities exist. Furthermore, the model results show very long range radial influences, extending over most of the investigated regions, possibly related to coupling effects associated with plasma self-organization. These results clearly show that transport in fusion plasmas is not local andmore » instantaneous, as is sometimes assumed.« less