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

From: Wolf Kirchmeir (wwolfkir_at_sympatico.ca)
Date: 09/26/04 

Date: Sun, 26 Sep 2004 08:28:20 -0400

ray scanlon wrote:
> Wolf Kirchmeir writes:
>
>
>>The phrase may in fact mislead, since it implies a comparison (and the
>>comparison has been explicitly invoked by some contributors to this
>>thread.) But I don't see any comparison happening - there is AFAIK no
>>"comparator module" that takes input from the "perception module" and
>>the "face recognition module" and outputs "Yup, that's Johnny." The fact
>>is that information about edges, etc is produced in the retinal layer,
>>that this information passes to various NNs whose output goes to other
>>NNs, etc until the final output is "That's Johnny." At each layer, the
>>information is reorganised - that is, the inputs to the next layer have
>>a different pattern. Note that the face recognition module is not as
>>simple as the name impies - some people who have had damage in that area
>>can still tell they are looking at a face, they just can't tell who it
>>is. Others can't even tell they are looking at a face. Since the damage
>>is to the NN, the "reference state" can only be the organisation of the
>>NN, as I said above. Again, now what? The organisation of the NN
>>determines how it operates, is all. Calling it a reference state doesn't
>>help us understand how it's organised.
>
>
> If we are to talk about electronic devices, it might help to think of
> the brain as a PAL (programmable array logic} rather than a Von
> Neumann computer. Then the signal energy comes in through the sensory
> neurons, filters through the interneurons, and comes out as a motor
> act. (Majendie's Law) No magic, just circuitry.

That looks like a Better Idea, all right. Just how complexly
interconnected can one make the elements of a PLA? Does each element
have a single input and a single output? If so, it's not IMO complex
enough to do what biological NNs do. See below.

> Now we have to think how we should abstract groups from this amorphous
> mass of interneurons. We can talk about motor program generators,
> initiators, and controllers--not word processing. What circuitry do we
> need so that a motor program generator can be modified by experience?

Feedback _networks_ (not simple loops.)

> What circuitry to throttle the signal energy temporarily as it rushes
> through one path so that it also has time to run through an
> alternative path?

A simple loop NN can hold a signal spike indefinitely: Ni -->Nx --> Ny
--> Nx will hold the input from Ni as long as the NN runs. But we also
need some Nx -->Ny so that the signal can eventaully go someplace else.
This means that any neuron can have more than one output. That's the key
IMO.

NB that this is the only way a NN can store data, and then storage is in
fact a constant cycle through the same sequence, which is not what
storage means in ref. to computers. RAM is refreshed, but that's not the
same process. In actual biological NNs the signal eventually decays.

> How does the signal energy actuate the throttle? How
> does it disable the throttle?

Use inhibiting signals as well activating ones. Each neuron has one
input, which requires a certain activation strength before it fires. Use
a few holding loops to accumulate signals until there are enough to fire
some downstream neuron Nk. Use inhibiting feedback from Nk to the
holding loops to switch them off so that a new signal can be held. NB,
again, that Nk must have at least two outputs. In general any given
neuron will have a single input and one or more outputs. It's the
multiple outputs that enable the topology you're looking for. In
biological NNs, most neurons have two or more outputs.

Problem is, AFAIK there's no simple way to describe such networks. Graph
theory can characterise the topology of networks (by type of
connectivity, for example), but AFAICT it can't handle the actual
topology of a NN that's complex enough to do interesting things.

> ray

The above ideas are _not_ original with me. I first came across them in
the early to mid-60s in a book title The Minds of Robots, which
disappeared from the university library shortly after I returned it, so
I never found it again. Can't recall the author, but do remember the
book was published in Bloomington.

Relevant Pages

  • Re: Neural netss (was Re: death of the mind.)
    ... >> This means that any neuron can have more than one output. ... >> Use inhibiting signals as well activating ones. ... >> neuron will have a single input and one or more outputs. ... >> multiple outputs that enable the topology you're looking for. ...
    (sci.cognitive)
  • Re: Neural netss (was Re: death of the mind.)
    ... We can talk about motor program generators, ... > This means that any neuron can have more than one output. ... > Use inhibiting signals as well activating ones. ... > neuron will have a single input and one or more outputs. ...
    (sci.cognitive)
  • Re: consciousness
    ... brain, but the thing we are sensing normally starts "out there". ... the signals is irrelevant to our conscious awareness. ... If you have two electrons in a wire, and one moves, it can cause the one ... But you haven't explained what happens if you take that neuron out of the ...
    (comp.ai.philosophy)
  • Re: Symbolic AI: Why Marvin Minsky and Curt Welch Are Out to Lunch
    ... a pyramidal neuron in the sensory cortex is strongly increased if they ... fire about 10 ms before the target neuron fires. ... significantly correlated signals from two related streams. ...
    (comp.ai.philosophy)
  • Re: Access bus signal names in S-function
    ... It may turn out easier/less messy to call the m-function ... a single input. ... access the signal names if I have a bus of many signals ... I can access the InputSignalNames from the originating ...
    (comp.soft-sys.matlab)