Descripción
|
|
---|---|
Markov Chain Monte Carlo (MCMC) methods, such as the Metropolis-Hastings (MH) algorithm, are widely used for Bayesian inference. One of the most important challenges for anyMCMC method is speeding up the convergence of the Markov chain, which depends crucially on a suitable choice of the proposal density. Adaptive Rejection Metropolis Sampling (ARMS) is a well-known MH scheme that generates samples from one-dimensional target densities by making use of adaptive piecewise linear proposals constructed using support points taken from rejected samples. The ARMS algorithm is often applied within a Gibbs sampler, where the reduction of the burn-in period is crucial. In this work, we point out a critical drawback in the adaptive structure of ARMS and propose an alternative scheme (A2RMS) in order to speed up the convergence of the chain to the target distribution. With the A2RMS algorithm, the sequence of proposals densities converges to the true shape of the target, allowing us to perform virtually exact sampling, since the correlation among the samples vanishes quickly to zero. Moreover, at the same time, the computational cost is kept bounded. Since the novel scheme also allows us to simplify the construction of the sequence of proposal distributions w.r.t. to the technique described in [Gilks et al. (1995)], then we also provide different simplified procedures to build the proposal. Numerical results show that the new algorithm outperforms the standard ARMS and other techniques in terms of estimation accuracy and reduced correlation among the generated samples. | |
Internacional
|
Si |
Nombre congreso
|
European Meeting of Statisticians (EMS) |
Tipo de participación
|
960 |
Lugar del congreso
|
Budapest (Hungría) |
Revisores
|
No |
ISBN o ISSN
|
|
DOI
|
|
Fecha inicio congreso
|
20/07/2013 |
Fecha fin congreso
|
25/07/2013 |
Desde la página
|
|
Hasta la página
|
|
Título de las actas
|
Abstracts of the 29-th European Meeting of Statisticians |