Skip to main navigation Skip to search Skip to main content

On sample generation and weight calculation in multiple importance sampling

  • Universidad Carlos III de Madrid
  • University of Helsinki
  • Technical University of Madrid

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Importance sampling is a Monte Carlo technique that approximates moments of target densities by means of weighted samples. These samples are traditionally drawn from a single proposal density. In multiple importance sampling (MIS) a set of different proposal densities is available. In this paper, we propose a formal framework that allows different ways of drawing samples from a set of proposals and different proper weighting functions that can be applied. In particular, we describe three sampling methods and five generic weighting functions. As proper sampling/weighting combinations, six unique MIS schemes (three of them are novel) are discussed throughout the paper. All the methods are analyzed in terms of the variance of the associated estimators, establishing a ranking regarding their performance.

Original languageEnglish
Title of host publicationConference Record of the 49th Asilomar Conference on Signals, Systems and Computers, ACSSC 2015
EditorsMichael B. Matthews
PublisherIEEE Computer Society
Pages1709-1713
Number of pages5
ISBN (Electronic)9781467385763
DOIs
StatePublished - Feb 26 2016
Event49th Asilomar Conference on Signals, Systems and Computers, ACSSC 2015 - Pacific Grove, United States
Duration: Nov 8 2015Nov 11 2015

Publication series

NameConference Record - Asilomar Conference on Signals, Systems and Computers
Volume2016-February
ISSN (Print)1058-6393

Conference

Conference49th Asilomar Conference on Signals, Systems and Computers, ACSSC 2015
Country/TerritoryUnited States
CityPacific Grove
Period11/8/1511/11/15

Keywords

  • Bayesian inference
  • Monte Carlo methods
  • multiple importance sampling

Fingerprint

Dive into the research topics of 'On sample generation and weight calculation in multiple importance sampling'. Together they form a unique fingerprint.

Cite this