[Top] [All Lists]

Re: [ontolog-forum] End of the line for triple stores

To: "[ontolog-forum]" <ontolog-forum@xxxxxxxxxxxxxxxx>
From: "John F. Sowa" <sowa@xxxxxxxxxxx>
Date: Mon, 03 Aug 2009 20:48:24 -0400
Message-id: <4A778558.7020406@xxxxxxxxxxx>
Rob and Rich,    (01)

RA> How is associative memory not path-based?    (02)

RC> ... associative memory is memory retrieved by content matching
 > instead of by addressing.  So "select * where a>27.3" is associative
 > retrieval instead of running the array of 1 to a zillion triples
 > to figure out which ones match the specified retrieval expression.
 > The difference is in the conceptual model rather than in the
 > implementation, which may actually be the same one depending on
 > the two designs.    (03)

I agree with Rich.  The SQL select statement is conceptually a
kind of associative retrieval, which is normally implemented by
means of indexes on certain fields of a relation that are defined
as "keys".  So indexing provides a kind of associative retrieval.    (04)

But as Rich said, the implementation might not have indexing
on all fields.  Therefore, some SQL queries will require a
sequential search that is equivalent, in principle, to a
kind of path-based search.    (05)

RA> My current belief is that associative memory in the brain works
 > by content matching (versus addressing) and that the neuronal
 > process the brain uses to perform that matching is 'path-based',
 > or more generally 'arrangement-based' (i.e., the arrangement of
 > the parts of the content (whether visual, aural sequences, ...)
 > do make a difference as to what the 'end point' association is made.    (06)

There are many kinds of guesses (AKA hypotheses) about how the brain
works, and most of the evidence supports some kind of arrangement
based retrieval -- a pattern that is seen or heard somehow activates
similar patterns that are stored in long-term memory.  Nobody knows
exactly how that activation is triggered.  One hypothesis is that
brain waves behave sort of like holograms:  incoming stimulation
causes similar stored patterns to "resonate" when there is a
strong correlation.    (07)

That would be a kind of parallel processing that would allow
a single "query pattern" to activate a resonating response
without doing a sequential search.  There is some evidence
of such resonances in the brain, but nobody knows exactly how
any particular input would stimulate such responses.    (08)

In any case, there are ways of implementing a kind of associative
memory for query graphs:  encode both the graph structure (which
would include cycles, connectivity, and ordering) and the labels
on the graphs (i.e., the ontology) in some numeric form, and
then index the numeric forms.  That is basically what we do
at VivoMind.    (09)

Following are the slides of a talk I presented last week, which
goes into some detail about how such mechanisms work:    (010)

    http://www.jfsowa.com/talks/paradigm.pdf    (011)

See slides 21 and 22, which mention the indexing methods and cite
some references.  The date of these slides is Aug 3, because I
added a couple of extra slides today to cover some of the
discussion during the conference (ICCS '09).    (012)

John Sowa    (013)

Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/  
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/ 
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx    (014)

<Prev in Thread] Current Thread [Next in Thread>