Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.3 4.3bsd-beta 6/6/85; site ucbvax.BERKELEY.EDU
Path: utzoo!decvax!bellcore!ulysses!ucbvax!gte-labs.CSNET!rich
From: rich@gte-labs.CSNET (Rich Sutton)
Newsgroups: mod.ai
Subject: Seminar - Connectionist Expert Systems (GTE)
Message-ID: <8606050424.AA02862@ucbvax.Berkeley.EDU>
Date: Wed, 4-Jun-86 11:18:18 EDT
Article-I.D.: ucbvax.8606050424.AA02862
Posted: Wed Jun  4 11:18:18 1986
Date-Received: Thu, 5-Jun-86 09:44:03 EDT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The ARPA Internet
Lines: 35
Approved: ailist@sri-ai.arpa


           "Connectionist Expert Systems in a Noisy World"
	 	       by Stephen I. Gallant 
 
This talk will describe a model for connectionist expert systems
(MACIE) and show how it is well suited to noisy and redundant
environments.
 
     Connectionist expert systems are diagnostic expert systems
based upon a connectionist model with several interesting features:
 
     -- They can be generated from training examples (and/or rules)
 
     -- They perform forward chaining to make conclusions and 
        backward chaining to elicit additional information
 
     -- They give IF-THEN rules to justify their inferences, even
        though their knowledge base contains no such rules
 
     -- They are arguably less prone to brittle behavior than
        traditional expert systems.
 
     In the talk it will be shown how an expert system for a noisy
and redundant problem can be constructed from:  (1) a noise-free 
model of an underlying process (perhaps a traditional expert 
system) and (2) a model for the noise involved.  System generation 
is entirely automated.
 
Where:	 GTE Labs, Waltham, MA
When:	 June 11th, 9:30 am
Contact: Rich Sutton, rich@gte-labs.csnet, 617-466-4133 (or 466-4207)
Net address of speaker: sig@northeastern
Also:	 That afternoon we will have an informal research meeting of
	 connectionists from GTE, UMass, and Northeastern
Welcome: Visitors are welcome!