[ABE-L] Seminário UFSCar/ICMC – Sexta 18/09/2015 - 14h00 - na UFSCar

[Rafael Izbicki]

Divulgação da palestra desta semana do Seminário do Programa
Interinstitucional de Pós-graduação em Estatística
<http://www.icmc.usp.br/Portal/conteudo/1096/13/interinstitucional-de-pos-graduacao-em-estatistica>
 (PIPGEs ICMC/USP e UFSCar), São Carlos.

*Data/Horário:*  Sexta-feira 18/09/2015; 14hs.

*Local:* Sala de Seminários do Departamento de Estatística da UFSCar

*Palestrante:* Luca Martino - ICMC, USP.

*Título:* Layered Adaptive Importance Sampling

*Resumo: *Monte Carlo algorithms represent the de facto standard for
approximating complicated integrals involving multidimensional target
distributions. In order to generate random realizations from the target
distribution, Monte Carlo techniques use simpler proposal probability
densities for drawing candidate samples. Performance of any such method is
strictly related to the specification of the proposal distribution, such
that unfortunate choices easily wreak havoc on the resulting estimators. In
this work, we introduce a layered, that is a hierarchical, procedure for
generating samples employed within a Monte Carlo scheme. This approach
ensures that an appropriate equivalent proposal distribution is always
obtained automatically (thus eliminating the risk of a catastrophic
performance), although at the expense of a moderate increase in the
complexity of the resulting algorithm. A hierarchical interpretation of two
well-known methods, such as of the random walk Metropolis-Hastings (MH) and
the Population Monte Carlo (PMC) techniques, is provided. Furthermore, we
provide a general unified importance sampling (IS) framework where multiple
proposal densities are employed, and several IS schemes are introduced
applying the so-called deterministic mixture approach. Finally, given these
schemes, we also propose a novel class of adaptive importance samplers
using a population of proposals, where the adaptation is driven by
independent parallel or interacting Markov Chain Monte Carlo (MCMC) chains.
The resulting algorithms combine efficiently the benefits of both IS and
MCMC methods.


--
Rafael Izbicki
Assistant Professor
Department of Statistics
Federal University of São Carlos (UFSCar)
rizbicki.wordpress.com
-------------- Próxima Parte ----------
Um anexo em HTML foi limpo...
URL: <https://lists.ime.usp.br/archives/abe/attachments/20150913/c710fb64/attachment.html>