SMC with Adaptive Weights for Approximate Bayesian Computation

Authors: 
Fernando Bonassi , Mike West
Duke University

Jun 11 2012

Methods of Approximate Bayesian Computation (ABC) are becoming increasingly popular for analysis of complex problems in which traditional MCMC and other approaches are not easy to apply. A major issue facing ABC is that the approaches inherently face high rejection rates. One strategy to deal with this is to modify the basic ABC sampling strategy based on ideas of sequential Monte Carlo (ABC SMC). We do this here, introducing an ABC SMC sampler with adaptive weights. Motivated by the idea of mixture model approximation to the joint prior distribution of data and parameters, this method integrates data-based information into the computation of Monte Carlo weights, and can yield substantial improvements in efficiency in terms of overall rejection rates. This is illustrated in examples with simulated data and real data in several model contexts.

Published version:   Bayesian Analysis, 2014.


This work was partly supported by grant DMS-1106516 from the U.S.National Science Foundation (NSF), and grants P50-GM081883 and RC1-AI086032 of the U.S National Institutes of Health. Any opinions, findings and conclusions or recommendations expressed in this work are those of the author and do not necessarily reflect the views of the NSF or the NIH.

Manuscript: 

PDF icon 2012-08.pdf

BibTeX Citation: 

@ARTICLE{BonassiSMCABC2012,
  author = {F. V. Bonassi and M. West},
  title = {Sequential {M}onte {C}arlo with adaptive weights for approximate {B}ayesian computation},
  journal = {Bayesian Analysis},
  year = {2014},
  note = {First published online: September 25, 2014},
  doi={10.1214/14-BA891}, 
  url = {https://stat.duke.edu/research/papers/2012-08}
}