James Cussens wrote:
>
> I've just finished giving a few lectures on join tree propagation, and
> it struck me that some sort of animated demonstration of how it works
> would be a lot more illuminating than the static descriptions I have
> been using. I was thinking of something that eg showed the messages
> being constructed and absorbed. Most packages that I have looked at
> quite naturally hide this stuff away - but for teaching you want it
> made explicit.
>
> Before I recruit some unsuspecting student to produce said software as
> a summer project, I though I had better check that there is not
> something suitable already.
>
We have some AI teaching demos in Java at:
http://www.cs.ubc.ca/labs/lci/CIspace/
These can be run from a web browser.
One of these is a Bayes net evaluation engine that shows variable
elimination (or bucket elimination). You can see the effect of
eliminating variables when querying a node, inspect the messages
(resulting factors), see how evidence is incorporated, etc.
Variable elimination is a good place to start teaching about Bayes net
evaluation because it's so simple. You can them motivate the join-tree
propagation as precomputing values and storing the important parts of
messages, so you can compute the posterior on all variables with just
two passes.
We would appreciate any feedback on this or the other tools. So far we
also have a tool showing various search algortihms and one for
constraint satisfaction. We are developing a neural-network tool, and
are thinking about others to add.
David
--
David Poole, Office: +1 (604) 822-6254
Department of Computer Science, [EMAIL PROTECTED]
University of British Columbia, http://www.cs.ubc.ca/spider/poole
