Wasserstein de-initialization for Markov chains
En palabras de los autores
This article generalizes the de-initialization framework, proposed by (Roberts and Rosenthal, 2001) for total variation distance, to Wasserstein distances. In essence, de-initialization captures the phenomenon that in the presence of certain structural features, the convergence behaviour of a Markov chain can be reduced to that of a "simpler" stochastic process. Our results allow to measure this in terms of a general Wasserstein distance. We apply them to analyse the Wasserstein convergence behaviour of Gibbs sampling, linchpin variable sampling and slice sampling.
Resultado principalEl resumen no menciona limitaciones.
Apareció: jueves, 8 de octubre. arXiv. Preprint, todavía sin revisión por pares.
Comentario de los autores: 36 pages, 4 figures