摘要：Our conditionally conjugate split-merge technique belongs to the family of
trans-dimensional MCMC algorithms, which includes, for example, reversible-jump
MCMC (Green 1995), birth-death MCMC (Stephens 2000a), and split-merge MCMC
(Jain and Neal 2004), (Dahl 2003). Trans-dimensional MCMC algorithms construct
Markov chain transitions between states that vary in dimension. For Dirichlet process
mixture models, this involves the creation or deletion of mixture components.
