The panel basically discussed three topics:
1. BNIF - so what?
2. Why standards
2. XML - let a thousand flowers bloom
I began with a review of why, in my opinion, the Bayes net interchange
format wasn't really happening, despite some use (basically, not many
people really care, and there is too much information loss in
translating into BNIF).
Tod Levitt went next and discussed why one might want a standard, from
the perspective of International standards organizations
Fabio Cozman and David Hovell then presented an alternative to a
"standard": the use of XML to publish self-defining formats.
The idea, as I understand it, is that a repository will be created
which will hold contributed XML formats, networks in those formats,
and pointers to tools (inference, learning, etc) that can utilize one
or more of the formats. It is hoped that such a public repository
will result in "emerging standardization", that is, a process in which
everyone steals others good ideas and a consensus gradually evolves
which is responsive to growing expressivity needs (DPNs, results,
learning data sets, chain graphs, local structure, object and
frame-based models, .....)
IET and Microsoft will work together to establish the repository, at
least I think that is what I remember as the outcome.
Anyone who was there have additions/corrections?
tnx - bruce
dambrosi@iet.com
dambrosi@cs.orst.edu