11.07.2015 Views

Untitled - UFRJ

Untitled - UFRJ

Untitled - UFRJ

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

An Efficient Split-Merge MCMC Algorithm for MixtureModels with an Unknown Number of ComponentsErlandson Ferreira Saraiva ∗Departamento de Estatística, Universidade Federal de São CarlosLuís Aparecido MilanDepartamento de Estatística, Universidade Federal de São CarlosFrancisco Louzada-NetoDepartamento de Estatística, Universidade Federal de São CarlosWe propose a new split-merge MCMC algorithm for estimation of mixture models with an unknownnumber of components. The strategy for splitting is based on data and posterior distribution. Allocationprobabilities are calculated based on component parameters which are generated from the posterior distributiongiven the previously allocated observations. The split-merge proposals allows a major changein configuration of the latent variables in a single iteration of the algorithm, avoiding possible localmodes, and are accepted according to the Metropolis-Hastings probability. As advantages, our approachdetermines a quick split proposal, in contrary to former split procedures which require substantial computationaleffort, and remove the need to specify a transition function and consequently the calculate ofthe jacobian, turning the probability of acceptance easier of compute and simplifying the implementation.The performance of the method is verified using an artificial data set and two real data sets. Thefirst real data set consist of benchmark data of velocities from distant galaxies diverging from our ownwhile the second is Escherichia Coli bacterium gene expression data.∗ Apresentador/Speaker35

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!