Saturday, August 10, 2002


"RISO: distributed, heterogeneous belief networks. A belief network is a probability model defined on an acyclic directed graph; distributed means nodes can be on different hosts, and heterogeneous means allowing different conditional distributions."

From the thesis:
"RISO is a system to aid reasoning in spatially and temporally extended problems, which implements a class of graphical probability models called distributed
belief networks."

"As the number of sites is increasing 400% per year, we can expect RISO installations to soon outnumber elementary particles, not to mention available IP

"It is natural to represent each geographical unit with a belief network, and to represent the flow of information from one locale to another by edges connecting variables in
separate belief networks; different kinds of messages travel with the arrows and against the arrows."

"The RISO inference algorithm is based on the polytree algorithm for belief network inference, in which “messages” (predictive distributions called 1/-messages and likelihood functions called ¸-messages) are computed."

When we create this distributed belief system we can finally find Elvis (25 years dead this month).

I've been able to get the Java code going with a little juggling. Wrote an Ant script which helps. It still needs some work and the initial instructions are wrong. Cutting and pasting from PDF is fun too: I think "inference" came out with a registered symbol (®), 1/- is pi and ,- is rho (or something).
Post a Comment