EXACT SAMPLING USING BRANCHING PARTICLE SIMULATION
Abstract
Particle methods, also known as Sequential Monte Carlo methods, are a popular
set of computational tools used to sample approximately from non-standard probability distri-
butions. A variety of convergence results ensure that, under weak assumptions, the distribution
of the particles converges to the target probability distribution of interest as the number of
particles increases to infinity. Unfortunately it can be difficult to determine practically how
large this number needs to be to obtain a reliable approximation. We propose here a procedure
which allows us to return exact samples. The proposed algorithm relies on the combination of
an original branching variant of particle Markov chain Monte Carlo methods and dominated
coupling from the past.
Origin | Files produced by the author(s) |
---|