The blocking Gibbs sampler samples a set of nodes given the Markov blanket of
this set. Thus, the blocking Gibbs sampler is a generalization of the "usual"
Gibbs sampler that samples a single variable at a time given its Markov
blanket. The blocking Gibbs sampler thus combines the benefits of exact local
computations and the Gibbs sampler, allowing approximate inference in very
large Bayesian networks with a much higher degree of mixing than the
single-site Gibbs sampler. I can provide references to those interested.
Greetings,
Claus Skaanning
mengshoe@odysseus.ai.uiuc.edu wrote:
> I'm interested in the concept of Markov blanket in Bayesian networks (BNs).
>
> The definitions of Markov blanket I've come across so far deal with the
> Markov blanket of one node. Has there been any work on generalizing and
> using
> the Markov blanket of several nodes? I'm interested in this primarily from
> the
> perspective of inference and learning, but any pointers would be
> interesting ...
>
> Thanks in advance,
>
> Ole J. Mengshoel.