Slice Sampling with Adaptive Multivariate Steps: The Shrinking-Rank Method - Statistics > ComputationReportar como inadecuado




Slice Sampling with Adaptive Multivariate Steps: The Shrinking-Rank Method - Statistics > Computation - Descarga este documento en PDF. Documentación en PDF para descargar gratis. Disponible también para leer online.

Abstract: The shrinking rank method is a variation of slice sampling that is efficientat sampling from multivariate distributions with highly correlated parameters.It requires that the gradient of the log-density be computable. At eachindividual step, it approximates the current slice with a Gaussian occupying ashrinking-dimension subspace. The dimension of the approximation is shrunkorthogonally to the gradient at rejected proposals, since the gradients atpoints outside the current slice tend to point towards the slice. This causesthe proposal distribution to converge rapidly to an estimate of the longestaxis of the slice, resulting in states that are less correlated than thosegenerated by related methods. After describing the method, we compare it to twoother methods on several distributions and obtain favorable results.



Autor: Madeleine B. Thompson, Radford M. Neal

Fuente: https://arxiv.org/







Documentos relacionados