Re: Neural netss (was Re: death of the mind.)

From: Eray Ozkural exa (erayo_at_bilkent.edu.tr)
Date: 08/30/04 

Date: 30 Aug 2004 14:58:07 -0700

"Rick Craik" <rick@@icebergideas..com> wrote in message news:<ycGYc.6188$7i2.301388@news20.bellglobal.com>...
> "Glen Foy" <spam33@butter.toast.net> wrote in message
> news:41320f8d@news03.toast.net...
>
> [snip]
>
> >
> > Neural nets have both excitatory and inhibitory inputs, so controlling
> > feedback is definitely possible. And here is the interesting point.
> > SUSTAIN may be an essential part of neural net architecture. Of all
> > activated concepts, the concept that SUSTAINS THE LONGEST may
> > be the concept which is eventually selected.
> >
> > -Glen
>
> I would still speculate that the activated parallel concepts need time,
> and the selection is based upon the slightly quicker. I think this
> would allow the slightly quicker concepts to control the slower ones.
> This would seem to contradict your interesting point; where
> "sustains the longest" would then mean unselected concepts.
> For example, the leading concept could also sustain any
> trailing related parallel concepts. If the leader falters,
> the trailers would indeed be sustained the longest, but
> the parallel processing of trailers may continue after the
> leader has responded, making them the longest sustained
> concepts.
>
> (I'd be more inclined to be interested in an increase of frequency,
> like arriving at a melody where "sustains the longest" might liken
> to a bass note. Do we not sustain a 7th note as a lead in to a
> bass change to resolve it?)

Intriguing, your idea seems to employ Speed Prior for an inductive
bias. Have you implemented such a learner?

Regards, 

--
Eray Ozkural