How to combine the prior knowledge during learning

The engine.
Post Reply
musicpxg
Posts: 19
Joined: Sat Feb 16, 2013 6:24 am

How to combine the prior knowledge during learning

Post by musicpxg »

Hi,

I already designed the structure of a network. I want to learn the parameters. There are questions as follows:

There are no training data for a root node in my network. I have prior experience to assign its value. So, how can I hand this problem when I use the EM algorithm? Is it a good idea to generate data accruing to the prior knowledge and then add the data to the missing column in the training data set?

Another question is that, there are some unobservable node (hidden node) in the network. Can I still employee EM to learn the parameters?

Thank you so much!

Regards,
Yang
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: How to combine the prior knowledge during learning

Post by shooltz[BayesFusion] »

There are no training data for a root node in my network. I have prior experience to assign its value. So, how can I hand this problem when I use the EM algorithm?
If you want EM to keep the node's parameters unchanged, include it in the fixed nodes set (it's one of the parameters passed to the learn method as vector or array, depending on the language you're using).
Another question is that, there are some unobservable node (hidden node) in the network. Can I still employee EM to learn the parameters?
The parameters will be learned even if no data is associated with the node (unless you explicitly assign the node to the fixed node set).
Post Reply