6533b827fe1ef96bd1285ddd
RESEARCH PRODUCT
Adaptive Population Importance Samplers: A General Perspective
David LuengoFrancisco LouzadaLuca MartinoVictor Elvirasubject
Computer scienceMatemáticasMonte Carlo methodPopulation02 engineering and technologyMachine learningcomputer.software_genre01 natural sciences010104 statistics & probability[INFO.INFO-TS]Computer Science [cs]/Signal and Image Processing0202 electrical engineering electronic engineering information engineering0101 mathematicseducationComputingMilieux_MISCELLANEOUSeducation.field_of_studybusiness.industryEstimator020206 networking & telecommunicationsStatistical classificationRandom measureMonte Carlo integrationData miningArtificial intelligencebusinessParticle filtercomputer[SPI.SIGNAL]Engineering Sciences [physics]/Signal and Image processingImportance samplingdescription
Importance sampling (IS) is a well-known Monte Carlo method, widely used to approximate a distribution of interest using a random measure composed of a set of weighted samples generated from another proposal density. Since the performance of the algorithm depends on the mismatch between the target and the proposal densities, a set of proposals is often iteratively adapted in order to reduce the variance of the resulting estimator. In this paper, we review several well-known adaptive population importance samplers, providing a unified common framework and classifying them according to the nature of their estimation and adaptive procedures. Furthermore, we interpret the underlying motivation for the different adaptation schemes, opening the door for novel and more efficient algorithms. Finally, we compare the performance of different algorithms available in the literature through a toy example.
| year | journal | country | edition | language |
|---|---|---|---|---|
| 2016-07-10 |