Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!satisfied.elf.com!news.mathworks.com!uunet!spool.mu.edu!news.cs.indiana.edu!blank@cs.indiana.edu
From: "doug blank" <blank@cs.indiana.edu>
Subject: Re: Concept mapping
Message-ID: <1995Feb23.170147.11823@news.cs.indiana.edu>
Organization: Computer Science, Indiana University
References: <3icupo$160o@ns1.CC.Lehigh.EDU>
Date: Thu, 23 Feb 1995 17:01:43 -0500
Lines: 47

In article <3icupo$160o@ns1.CC.Lehigh.EDU>,  <nft0@Lehigh.EDU> wrote:

>Is there a neural net algorithm that could be used to COMPARE a concept map
>(illustrated nodes and links) of an analogy (call this the base) and a
>concept map of what the analogy represents (call this the target).
>
>Understand that the maps of both will inevitably be different.  That is,
>the information contained in the nodes will differ (ie. piston[b] -->
>diaphragm[t] -- a diaphragm is like a piston) and the links may also vary.
>These maps could be represented spatially or hierarchally.  Any
>recommendations out there??

There has been much work on connectionist analogy models. Most of them
are attempting to perform structure-mappings similar to what you have
described. Holyoak and Thagard's ACME model is a type of Boltzmann
machine; it is a constraint-satisfaction, settling network. Each
attribute and relation is a node in the net. A viable mapping is one
whose constraints are consistent, and the network will settle on that
set of mappings. A couple more are Genter, et al's SME, or Keane's
IAM. See Cog Sci #18, 387-438, 1994 for a comparison of all three of
these. 

On a cognitive note, most of these get-a-structure-then-map-it
algorithms suffer from all of the problems that Hofstadter has
attacked with his Copycat and Tabletop programs, and surely can't be
the way humans do it. The problem is always in starting out with rigid
structures.

I have been working on a connectionist network which learns analogies,
but I have had to restrict it to working on very perceptual
analogies. 

-doug blank

>
> -----------------------------------------------------------------------------
> Neil Toporski
> International Multimedia Resource Center
> Lehigh University, Bethlehem, Pennsylvania  18015
> Phone: (610) 758-6067 Fax: (610) 758-6556 Internet: NFT0@LEHIGH.EDU


-- 
=====================================================================
blank@cs.indiana.edu                Douglas Blank, Indiana University
Computer Science                                    Cognitive Science
=====================================================================
