Re: What evidence is there that a neuron is digital computer ?

On Jul 25, 7:29 pm, c...@xxxxxxxx (Curt Welch) wrote:

John, you clearly don't understand something because what
you write doesn't make sense.

Because you fail to read it properly and in the context of
your claims about the *use* of the *resolution* to bring
about some outcome which is separate to temporal coding at
a usable resolution. In other posts I made clear what I
was talking about and it amounts to what you write here:

Of course in real life, there are many forces at work
controlling how close it is to being "too far" and as
such, we don't have the power to intentionally adjust
the position of the stick beyond a certain limit while
keeping it from falling over.

And I made that clearer in other posts when I wrote about
the source of chaos being these infinite resolutions and
made my statement clearer again with the question: Did the
neuron *make full use of this analog temporal precision*.
Below you claim it does make FULL use of this resolution.

My challenge to you was how you, or a neural network, would
make practical use of this *precision* (not about wether
the precision existed or if temporal coding was used).

You wrote:
If that difference falls across the detection threshold,
of course it [the neuron] will detect it.

Now let me be clear what I mean by "detect it". It means it
has a specific measurable effect, immune from random thermal
noise, in bringing about some useful outcome.

There are issues such as when a pulse hits the neuron what
form does it take to represent this infinite resolution?
Do you understand the molecular way the analog value exists
in the neuron so that it can hold or use this infinite or
high resolution temporal difference? How about internal
noise or residue from previous inputs effecting the actual
resolution possible.

It is not about if the event occurs (an extra pulse every
so often) but how this extra pulse, representing this fine
temporal resolution, is actually used. And also importantly
not accumulated to cause a chaotic system to develop as
it does with a weather system.

You wrote:
... after the recharge period, the full analog temporal
resolution of the pulses received is made use of.

Explain or show how this *full* analog temporal resolution
is made use of (not if it exists "out there" which I have
never had an issue with).

For example consider a simple animal with two light sensitive
neurons on either side of its head and each connected to a
motor neuron on the opposite side of the body. The behavior
of this animal would be to move toward a light source. Now
the "time of arrival" of photons would be random although the
average number of photons per unit of time would depend on
the light intensity. Let us say a pulse is produced for
every photon. Now is the full analog temporal resolution of
the pulses received *being made use of* in the neurons of
this animal?

If this animal is not making use of the temporal analog
resolution then show me a neural net that does make full
use of the analog temporal high resolution as you claim
neurons in the brain do.

And this is not about this resolution existing it is about
*using* this resolution. I know how you talk tricky.

So how is this resolution *being made use of* so it can do
something it couldn't do if it didn't have this resolution.

In theory you should be able to store an infinite set of
values between two pulse events. A value for every possible
temporal position between those two events. But how would
you generate it and how would you "read" out that value?
For that amounts to the claim you make with neurons.

casey wrote:
On Jul 23, Don Geddis wrote:
[...]
But the output frequency of pulses is an analog computation
on the input frequency of pulses.

Ok. You are saying the neurons are using frequency coding?


"frequency coding" is just another way of saying "sensitive
to time". It's a way I believe Don was hoping would make
you understand what you seemed to be missing.

Well I have used frequency coding and I think I understand
clearly what is involved but there are reasons to believe it
is not the only kind of neural code used in the brain. In fact
I have pointed out to you that the audio input to the brain
uses frequency pulse code for amplitude and place code for the
audio frequency. But that is just the input stage.


What do you think a neuron computes?


I believe it's best understood at a temporal pattern recognizer
(or temporal pattern classifier). The patterns real neurons
recognize are too complex to be simplified as "frequency coding"
but that view helps people get into the right mind set I believe.

Or lead them up the garden path.

The question however is not about the "right frame of mind" but
what a particular neuron *actually does* in terms of pulses in
and pulses out keeping in mind there are different neurons and
they behave differently.

JC

.

Relevant Pages

  • Dubious analogs (was Symbol Grounding Problem)
    ... The paper didn't discuss real neurons but rather model neurons. ... resolution in space and time against resolution in intensity. ... You mentioned analog devices. ...
    (comp.ai.philosophy)
  • Re: What evidence is there that a neuron is digital computer ?
    ... inputs from all the source neurons, ... The question is, does it digitize the input first, and only then process ... and a sequence of these frames is what we think of as video. ... neuron at ANY time and that the neuron makes use of that resolution. ...
    (comp.ai.philosophy)
  • Re: What evidence is there that a neuron is digital computer ?
    ... Infinite temporal resolution is used the sensors. ... in the past about how your pulse sorting units could ... If you simply make an analog gate to detect ...
    (comp.ai.philosophy)
  • Re: What evidence is there that a neuron is digital computer ?
    ... You said that neurons working on the frequency ... temporal processing beyond just "frequency encoding" you did a very bad job ... I was also working on your pulse sorting networks and all ... function computing neural networks which had no ability to use or respond ...
    (comp.ai.philosophy)
  • Re: What evidence is there that a neuron is digital computer ?
    ... This matters if there are single inputs to neurons. ... arrival times of different pulse trains at the same ... I did not switch to the idea of exploring designs with async high ... Until you can show me a good a real reasons to go back to what I've already ...
    (comp.ai.philosophy)