🚀 go-pugleaf

[Please note follow-ups]

In article <34446D77.F829445C@qi3.com>, Steven Packham
<packhste@qi3.com> wrote:

>Erik Vick wrote:
>> 
>> Louis Savain wrote:
>> >
>> > In article <343AAFF1.6D35@dial.pipex.com>, Howard Snart
>> > <aac88@dial.pipex.com> wrote:
>> >
>> > >[...]
>> > >This approach stems from several misgivings I have over the existing
>> > >methods of teaching neural nets... most significantly
>> > >backtracking.(which I see as un-scalable, and fundementally wrong... you
>> > >force the outputs to conform, whereas the problem may lie far from the
>> > >outputs)
>> >
>> >   I agree.  Traditional neural nets are not biologically plausible,
>> > nor do they do the kind of learning that is needed to gain common
>> > sense.
>> >
>
>	I disagree. I feel that the formation of a topographical map
>of the environment is obvious starting in Striata cortex area 17 (field
>of vision division, edge detection), 18 and 19 (starting three
>dimensional mappings with stereoscopic details. Where the corpus
>colosum begins to wire both hemisphere together).
>
>	All of this structure and organization within the brain can be
>explained with mechanism which the neuron does posses. The mechanism
>seem to contain some very simple functionality which parallels the
>learning process that back-prop engages. This is only one very
>insignificant aspect of computing ability which neurons do posses. In
>fact Back-Prop can be broken down into a very basic model that neuron
>could employ quite easily.

>We are not ignoring this are we?

  Well, there is no evidence, AFAIK, that biological neural nets use
back propagation.  The only thing that could come remotely close to
back propagation is what is called "antidromic or retrograde spikes".
These are known to affect computation in a neuron's dendrite and, as
suggested by the context of this thread, they are observed to greatly
modify synaptic efficacy depending on the temporal order between
pre-synaptic and post synaptic potentials (Henry Markram, 1-10-97,
Science).  I personally feel that back-propagation is the least
biologically plausible aspect of conventional NNs.  Indeed, I see
back-prop as a dead end.  Just one man's opinion.

>[...]
>Actually the Back Prop model is a very simplified model of a neuron.

  I don't see how the back-prop model can be a model of a neuron in
the light of what I just wrote.

>[...]
>I am actually very surprised how obvious the computational mechanism
>of the brain are.

  There are many, including yours truly, who will disagree with this.
  
>The problem I see is there vast variations. The many different
>structures are enough to take quite a while to familiarize ones self
>with.

 I disagree with the above although there are many who will agree with
you.  I see little variation in fundamental operating principles from
one neuron to the next.

Regards,

Louis Savain

President, Marengo Media, Inc.