Continuous variables

The front end.
mark
Posts: 179
Joined: Tue Nov 27, 2007 4:02 pm

Post by mark »

What is a fuzzy Bayesian network?

Hard evidence is observing the state of a node with probability 1. Soft evidence is observing a distribution over the states of a node.
ben_mrad
Posts: 10
Joined: Sat Mar 27, 2010 2:53 pm

Soft Evidence and Fuzzy Bayesian Network

Post by ben_mrad »

A FBN is just a Bayesian network with variables which have fuzzy states.These states are simultaneously uncertain as well.Such a network combines the advantages of a fuzzy representation and Bayesian networks.

For example, in a node "height" which contains the stats small, average and big, an individual of size of 1.65 is in the same time small and average with the percentages 25 % and 75 %.
Thus if we observe the height 1.65, stats take as values in soft evidence 0.25 and 0.75 successively.

My probléme maintaining is how to calculate the values of stats of the observed node in soft evidence?

thank you
mark
Posts: 179
Joined: Tue Nov 27, 2007 4:02 pm

Post by mark »

No, I believe this is still quite different. In case of Bayesian networks only one state can be true at a time but you may not know which one, hence the distributions. In case of a fuzzy Bayesian networks multiple states may be true to certain degrees at the same time, which is conceptually different.
ben_mrad
Posts: 10
Joined: Sat Mar 27, 2010 2:53 pm

Soft Evidence and Fuzzy Bayesian Network

Post by ben_mrad »

How to calculate the values of stats of the observed node in soft evidence?
mark
Posts: 179
Joined: Tue Nov 27, 2007 4:02 pm

Post by mark »

Since they are observed, they don't change when inference is performed. Beliefs of neighboring nodes will be updated.
ben_mrad
Posts: 10
Joined: Sat Mar 27, 2010 2:53 pm

Soft Evidence and Fuzzy Bayesian Network

Post by ben_mrad »

Effectively, the evidence does not change but I find that value propagated in the inference are not the observed value (soft evidence) but rather of the other values.
For example, in Network "Asia" if node "Visit To Asia" with soft evidence is 0.2 in stat0 and 0.8 in stat1 the values of the stats of this node which will be propagated are 0.961 for stat0 and 0.039 for stat1.
How to calculate these values (0.961 and 0.039) ?
marek [BayesFusion]
Site Admin
Posts: 430
Joined: Tue Dec 11, 2007 4:24 pm

Soft vs. virtual evidence

Post by marek [BayesFusion] »

GeNIe implements what is called "virtual evidence," which is uncertain observation that will propagate through the network. Virtual evidence can be (and actually is in SMILE and GeNIe) implemented by just adding a dummy child node and making sure that the CPT of that dummy node results in sending uncertain evidence. You are referring to soft evidence and I'm afraid we don't have that :-(. We will have it one day, so please stay tuned.
Cheers,

Marek
ben_mrad
Posts: 10
Joined: Sat Mar 27, 2010 2:53 pm

Soft Evidence

Post by ben_mrad »

Thank you very much for all this information...
I think that I understood the method of calculate values of the stats of nodes after the update by (soft Evidence). Made, always in the example ' Asia ' if soft evidence in node "Visit To Asia" is 0.2 in stat0 and 0.8 in stat1, the values of the stats of this node which will be propagated are 0.961 for stat0 and 0.039 for stat1.
Then, we make the first update by observing stat0 (we obtain val1) then we make the second update by observing stat1 (we obtain val2).
Finally, the values found for states of every node are obtaining by this manner (val1 * 0.961 + val2 * 0.039).
My question is how to calculate these values (0.961 and 0.039) of the observed node ("Visit To Asia" )?
ben_mrad
Posts: 10
Joined: Sat Mar 27, 2010 2:53 pm

Soft Evidence

Post by ben_mrad »

I think that I have found the answer to my question concerning "soft evidence" in the article "The Adaptive Safety Analysis and Monitoring System".
In fact, the values propageted are not the values observed but the quantified values.
My question is why the propagated values are quantified values and not observed values?
Thank you very much
Post Reply