Re: intellignece and conciousness, the one implies the other?

"JGCASEY" <jgkjcasey@xxxxxxxxxxxx> wrote:
> Curt Welch wrote:
>
> > Whatever we have in the way of a mind, or a soul,
> > or consciousness is just a product of what the
> > body does.
>
> Or whatever we have in the way of a body is just
> a product of what the mind, or soul, etc does :)
>
> As an aside have you been working on your net?

Haven't had time to do anything with them in a long time now.

> The last thing I did was feed it a 64 bit pattern
> of pulses and then after a while feed it a different
> 64 bit pattern of pulses. The effect seemed to be
> the second pattern shuffled the top nodes around a
> bit but essentially joined up with the old pathways
> of the previous pattern. That is the output was the
> same or similar as the first output.

What do you mean by that? Were you feeding it one input which had a 64
pulse pattern of different spacing that repeated over and over?

Or were you feeding it 64 inputs where each input represented a different
bit and the bits were somehow encoded with pulses? If so, how were the
bits encoded?

> This is what you might expect from streams of water
> comming down from a mountain range and joining up
> into a few rivers down the bottom. Change the stream
> sources high in the mountains and they will still
> tend to end up comming out the same way.

It should show different activity patterns for those different inputs.
They shouldn't look the same.

To make it work correctly however you must include the code which adjusts
the sorting gap (threshold as you call it) to create the fixed output
ratio. And, very important, the adjustment must be slow enough that it
doesn't rebalance the network for each pattern. The point of the
adjustment is to create an average over a very long period (one which must
include both your two inputs).

What you should see if it's coded correctly, is that the different input
patterns will create unique patterns of activity in the top nodes, and
then, if the patterns are simple enough, the bottom nodes will look
somewhat random with fairly equal distribution of activity. You will be
able to spot which input pattern is beeing feed to the network by the
pattern of activity it creates in the top nodes.

If your adjustment code however is adjusting too quickly, the net will
completly forget the previous pattern and re-tune the network for the
current pattern after each pattern switch. You don't want it doing that
because it will not correctly create different internal activty patterns
for each input.


Also, you have to remeber that this is a temporal network. If you send a
sequential pattern (which I couldn't tell from your description above),
then you are not sending it just one pattern, you are sending it many
patterns.

Meaning, if you send a stream of pulses with the spacing 1 2 3 over and
over, the input looks like this:

123123123123 ...

But that's not one pattern, that's 3 patterns:

123123123
231231231
312312312

So for each input pulse, the temporal pattern is defined by the past
history of pulses. And for each pulse in that 3 pulse sequence, there are
3 different past histories which the net will see.

And if you switch to the pattern: 456 repeating, then that's 3 more input
history patterns the net will get to see. But not only that, there's a
large history of patterns that happens at the transistion:

1231234
12312345
123123456
1231234564
12312345645
123123456456

So for the the first 6 pulses in the 456 sequence, you get 6 different
temporal pulse histories defined by your input stream.

If you send it 123 10 times in a row, followed by 456 10 times in a row,
and repeat, you are not sending the net 2 different patterns, you are in
fact sending it 60 different temporal patterns. And the network will, if
it's large enough, create a different pattern of internal activity for each
of these 60 different input "states".

--
Curt Welch http://CurtWelch.Com/
curt@xxxxxxxx http://NewsReader.Com/
.

Relevant Pages

  • Re: Can operant conditioning account for all learning?
    ... pattern and s,t and u are examples of the cat pattern. ... "temporal" gates wired up in what topology. ... The pulse routing network should scale correctly as long as it doesn't have ...
    (comp.ai.philosophy)
  • Re: Jeff Hawkins Q&A
    ... A signal is just a pulse or spike ... the way they, the group of neurons, are connected ... Thus a group of letters have no meaning themselves ... By "meaning" I refer to the effect that pattern ...
    (comp.ai.philosophy)
  • Re: I Need help from all the group participants
    ... the poster needs. ... Suppose you want to find out what is someone's pulse/heart rate, ... to find out someone's pulse is to ask: "What is John's pulse", ... have different pattern, e.g. "Please, tell me John's pulse immediatelly" ...
    (comp.lang.python)
  • Re: intelligence
    ... Morse code program gave me any insight I didn't already ... pattern problems better than the way it is already done. ... network correctly solves, and which it is yet to solve, you don't seem to ... The pulse sorting network you know about creates a hierarchy of pulse ...
    (comp.ai.philosophy)
  • Re: Can operant conditioning account for all learning?
    ... How the brain deals with asynchronous data is a question ... You liked to talk about pulses of infinite small widths ... network that are too close together they blend into one ... pattern and s,t and u are examples of the cat pattern. ...
    (comp.ai.philosophy)