Missing values and hidden variables are usually handled either
deterministically with EM or some related maximum likelihood method,
or by stochastic simulation. You might be interested in our paper on
the subject, looking at population-based stochastic search algorithms:
http://ite.gmu.edu/~klaskey/papers/Laskey_Myers_popMCMC.pdf
There are several different ways to represent a noisy-OR as a
hidden-variable model. There were papers on this by Heckerman in UAI
a few years back. I don't have the references, but look in the
online proceedings with author Heckermand and "causal independence"
in the title. You can also look in the lecture notes for my course
on probabilistic modeling:
http://ite.gmu.edu/~klaskey/INFT819/819_Unit2.pdf
Kathy Laskey
>Hi,
> I use Noisy-or belief network in information filtering. But I do
>know how to estimate the paramter (Namely, Cij)from data. Futhmore, there
>are hidden variables in the network, I do not know how to deal with them.
>Maybe EM algorithm is workable.
> I wonder general methods of parameter estimation in belief network
>can apply to Noisy-or model, or there are may-be good methods for
>simplified networks.
> Thanks in advance for any suggestions.
>
>Tong.
>
- --
+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+
Kathryn Blackmond Laskey (703) 993-1644 (voice)
Department of Systems Engineering (703) 993-1521 (fax)
and Operations Research [EMAIL PROTECTED]
Mail Stop 4A6 http://www.ite.gmu.edu/~klaskey/
George Mason University
Fairfax, VA 22030
If a man will begin with certainties, he will end in doubts; but if he
will be content to begin with doubts, he will end in certainties.
Francis Bacon (1561-1626), Advancement of Learning
------- End of Forwarded Message
