>===== Original Message From moq_discuss@moq.org =====
>PATRICK
>In conventional representational neural networks, one node or neuron
>represents a thing. If a dog sees a bunch of hairs, a tail, body, face
>and legs and all (which are assumed to be elements of the concept CAT)
>then a neuron will fire saying 'this is a CAT', which then can be linked
>to representations in memory like 'you can chase CATS'. And then the dog
>might want to choose to chase the cat... in cognitive psychology it was
>(and still is, actually) a hot topic two decades ago of whether there
>exists in the brain of a dog a 'prototype' of a CAT.
ERIN: My understanding of neural networks is a little different. The older
conceptual models had a node equalling a concept but in neural networks the
info is distributed across a set of nodes. (I am not sure but I thought it was
the features that were the nodes.) They thought the older models didn't make
sense with the limited number of neurons a person has available. So saying a
there is a cat neuron seemed silly and thus the neural network interest is hot
but it does not want a node = concept.
>RICK
>Okay. I think I understand this (except maybe the 'prototype' idea).
>
ERIN: An example. A canary is more prototypical bird then an ostrich. It
takes longer for somebody to answer is "is a canary a bird" then "is an
ostrich a bird"
MOQ.ORG - http://www.moq.org
Mail Archive - http://alt.venus.co.uk/hypermail/moq_discuss/
MD Queries - horse@darkstar.uk.net
To unsubscribe from moq_discuss follow the instructions at:
http://www.moq.org/md/subscribe.html
This archive was generated by hypermail 2b30 : Sat Aug 17 2002 - 16:01:57 BST