Re: it's not just the box dummy - it's just a houseofcards.-Reply -Reply -Reply -- MIDRANGE-L

Note to the list:  I took this post off line to get Rob's article & have
traded some Emails since, but now that the list seems to have picked
up on the I'll bring our conversation back to the list (saves typing :))

So Rob, if I follow what your're saying, then the NDB somehow knows
or guesses what connections might be contextually useful for a given
query or operation.  What if it guesses wrong, as human often do,
especially when exceptions to general rules pop up like most birds
fly but some don't?  Is there some feedback mechanism to allow the
NDB to generate connections "on the fly" when it discovers via a
longer route (e.g. Siamese cats to cats to felines to mammels) a
contextually useful connection (e.g. Siamese cats are mammels)?  In
other words, does the NDB "learn" over time?

>> "Rob Dixon" <rob.dixon@erros.co.uk> 07/27/98 11:13am >>

<snip>

Personally I think we have some way to go before we can create
systems that can learn by example and which are suitable for
business applications and are scalable.  I believe that the first step in
this process must be to create a database mechanism that can be used
to store AND retrieve real data with all its complexity, and the
Neural Database concept may be the first step towards that.

<snip - Scott's original question>
< If the vast number of connections make the neural
> database so powerful, then doesn't each subsequent addition to the
> database require more and more processing to define meaningful
> connections to all the existing rules/data, eventually bogging
> things down?  

This is what you might expect but -  NO.

<snip>

The NDB is not yet a fully intelligent database and the search
mechanism only looks where it is instructed by the NDB - this stores
connections about other connections.  A human being, when asked
about birds, might say "Birds fly".  As a generalisation, this is true,
and in many contexts would be an adequate answer, but, when
pressed, we might remember species of birds that do not fly. In
other words, that initial response was not as a result of considering
all connections about birds.  

<snip>

Taking your example of the Siamese Cat, all those connections may be
there but I do not necessarily need to consider them all together - it
depends on my task (which might considered as another word for
context).  Once you have defined the properties of a "cat" (feline,
mammal etc.,) in the NDB, if you then put in "siamese Cat", it will
automatically inherit the properties of "cat", as will "Persian Cat" if
you put that up later. 

+---
| This is the Midrange System Mailing List!
| To submit a new message, send your mail to MIDRANGE-L@midrange.com.
| To subscribe to this list send email to MIDRANGE-L-SUB@midrange.com.
| To unsubscribe from this list send email to MIDRANGE-L-UNSUB@midrange.com.
| Questions should be directed to the list owner/operator: david@midrange.com
+---