Another general approach to parallelizm, is conditioning:
Select a subset of variables and let each processor compute
the belief (or mpe) conditioned on a single instantiation of this
subset. (Of course, if you dont have enough processors assign each
a subset of assignments to the conditioning set)
This, autmatically partitions table multiplication between
processors (the computation in each processor can be by join-tree
clustering or any other manner).
------Rina.
>
>Anders -
>
>In UAI-92 (I think) Zhaoyu Li, Tony Fountain, and I (Bruce D'Ambrosio)
>published a paper on parallelizing SPI, where the main source of
>parallelism was distributing a single conformal product over multiple
>processors.
>
>tnx Bruce
>dambrosi@cs.orst.edu
>
>In message <364DB873.CCBB1716@cs.auc.dk>, Anders L Madsen writes:
>>Hi All,
>>
>>As has been mentioned by many researchers, a good way of exploiting
>>parallelisme in multiprocessor computers when performing Bayesian
>>inference is to have the table operations split up in independent
>>subtasks.
>>
>>Does anyone know of any algorithms developed for performing
>>multiplication, division, and marginalization of multiple dimensioned
>>probability potentials or more general multiple dimensioned structures?
>>
>>Any replies will be appreciated.
>>
>>Thanks
>>Anders L Madsen
>>
>>*--------------------------------------*
>>| PhD-Student |
>>| Anders L Madsen |
>>*--------------------------------------*
>>| Aalborg University |
>>| Department of Computer Sience |
>>| Frederik Bajers Vej 7C |
>>| DK-9220 Aalborg |
>>| Denmark |
>>| Email : anders@cs.auc.dk |
>>| Http : http://www.cs.auc.dk/~anders |
>>*--------------------------------------*
>>
>>