Loopy Belief Propagation

The engine.
Post Reply
Jochen0x90h
Posts: 19
Joined: Thu Jul 11, 2013 1:20 pm

Loopy Belief Propagation

Post by Jochen0x90h »

Hi!
Is it possible to select pure loopy belief propagation as algorithm even if it is not correct, only to see its result?
-Jochen
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: Loopy Belief Propagation

Post by shooltz[BayesFusion] »

Jochen0x90h wrote:Is it possible to select pure loopy belief propagation as algorithm even if it is not correct, only to see its result?
It should be possible - did you try it?
Jochen0x90h
Posts: 19
Joined: Thu Jul 11, 2013 1:20 pm

Re: Loopy Belief Propagation

Post by Jochen0x90h »

In Java the BayesAlgorithmType has a public static final int LBP = 9; which must be Loopy Belief Propagation if you say that it is supported. How do you set iteration count and damping?
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: Loopy Belief Propagation

Post by shooltz[BayesFusion] »

Jochen0x90h wrote:In Java the BayesAlgorithmType has a public static final int LBP = 9; which must be Loopy Belief Propagation if you say that it is supported. How do you set iteration count and damping?
Unfortunately, jSMILE does not expose the DSL_network::SetLBPParams method which controls LBP-specific parameters. However, the sample count is accessible through Network.get/setSampleCount.
Jochen0x90h
Posts: 19
Joined: Thu Jul 11, 2013 1:20 pm

Re: Loopy Belief Propagation

Post by Jochen0x90h »

Have you any experience with residual belief propagation? Does it converge to a solution on networks where synchronous loopy belief propagation starts to oscillate?
marek [BayesFusion]
Site Admin
Posts: 430
Joined: Tue Dec 11, 2007 4:24 pm

Re: Loopy Belief Propagation

Post by marek [BayesFusion] »

As far as I know, we don't have LBP accessible to the users in GeNIe or SMILE. We have it internally as part of the EPIS-BN algorithm. We have Hybrid LBP, which is an experimental algorithm for networks with continuous variables. We have no experience with residual belief propagation. Do you think we should look at it?
Cheers,

Marek
Post Reply