Newsgroups: comp.ai.shells
Path: cantaloupe.srv.cs.cmu.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!zombie.ncsc.mil!blackbird.afit.af.mil!news
From: ebaenen@afit.af.mil (Eric P. Baenen, Capt, USAF)
Subject: large knowledge bases constructed as bayesian networks
Message-ID: <1994Aug31.211435.12236@afit.af.mil>
Keywords: bayesian networks expert systems graph structure knowledge base
Sender: news@afit.af.mil
Nntp-Posting-Host: tolkien.afit.af.mil
Organization: Air Force Institute of Technology
Date: Wed, 31 Aug 1994 21:14:35 GMT
Lines: 41



Hello,

I am currently doing research with inferencing algorithms for expert  
systems (primarily diagnostic type - backward propogating).

I am trying to find examples of large knowledge bases constructed as  
bayesian networks.  If anyone has an example of such a KB that I could use  
to test/benchmark my algorithm I would sincerely appreciate your help.

Also, I am curious to know if anyone has done any research on the  
structure of large knowledge bases?  The testing of my algorithms with  
large randomly generated graphs (weighted AND/OR directed acyclic graphs -  
converted bayesian networks) have yielded some unexpected trends.

If you have worked with large KBs, can you tell me anything about the  
general structure you encountered; locallity of data, depth of graph or  
network, average number of causal steps or chaining length, etc?

If anyone knows of any recent papers written on the subject of knowledge  
base structure I would also appreciate that information.

Thank you for your time.

Eric Baenen

---

************************************************************
* ERIC P. BAENEN, Capt, USAF
* Air Force Institute of Technology (AFIT)
* Dept. of Computer Engineering
* AFIT Box # 4146
* Wright-Patterson AFB, OH  45433
* Graduate Student: Computer Engineering 
*                      (Artificial Intelligence)
* EMail: ebaenen@afit.af.mil	(NeXTMail preferred)
* Voice Mail: 513-255-3636 Ext 1017
* FAX: 513-476-7204 (comm)   986-7204 (DSN)
************************************************************
