Re: Grandmothers Galore?

>
> > I think you are doing Healy an injustice by calling his work
> > grandmother cell theory.
>
> Is this an indirect reference to some ongoing argument?

Uh, oh - looks like i wrongly gave the impression that i understood the
Healy article

Very sorry to imply or worse, begin an ongoing argument.

No, i was giving Healy the benefit of the doubt (doubt caused by my lack of
feeling that i quite got the point of his paper) and i focused in my reading
on his proposal for linkage between "concepts" where i think there is some
merit, or at least food for thought (no pun intended).

I meant to suggest that grandmother cell based theories are unlikely to be
fruitful and that there is, OTOH, something of interest in his proposal -
some loose thinking and writing on my part.

While i'm rambling along, i might as well go on to say- i am dubious of any
theory involving a "neural code".  Of course the "neural code" is something
we don't know, but will hopefully understand later. I had a similar response
to William Calvin, who proposed very interesting ideas assuming an unknown
neural code. Perhaps you have read him - i'd love to hear your comments, if
you have.

And i hope to hear your take on Grossberg - like Freud, i've gotten much
more out of his theories expressed second hand by his students and
followers, than directly from his (dare i say, awful) writing. While you're
at it, help me with Edelman, too.

I keep Hebb and Lashley (and D. MacKay) much closer to my heart than
Hopfield and Kohonnen. I try to use connectionist ideas but not much
computational and symbolic in my so-called theorizing. (Hebb and Lashley
sure can write.)

And i apologize to those readers who are rightly wondering why i am wasting
their time in this newsgroup.  I read this and bionet.neuroscience ng and
only rarely see posts that bear on my muddled musings.



> I may need to alter my choice of terms. I was using "grandmother cell"
> to mean a binary unit that completely and only represents one concept.
> Isn't that correct?

I doubt i could describe grandmother cell theory any better.
Thanks for your thoughts, Wendell.


> This is the first time I have seen network representation modeled as
> symbolic and combinatoric rather than subsymbolic and convolutional.
> If I understand G. Heath further above, ANN's were sometimes
> restricted in this way to speed learning, but this approach is no
> longer popular.
>
> I am trying to find to examples of this approach in the connectionist
> literature. Guess the best place to start would be with Grossberg's
> ART, unless someone has a better suggestion.
>
> Cheers,
> Wendell
>
> > > Here is a good one to start with:
> > >
> >
http://cialab.ee.washington.edu/mjhealy/web/Papers/Colimits%20in%20memory.pd
> >