|
<![CDATA[
To: Frank Schuurmans <frank@bio.vu.nl>
Hi Frank,
>Because of the shape of the firing in time the neuron has a 'memory'
>My guess is that firing frequencies are very important in the functioning
of
>the brain.
I do not know as much about the brain as you, but I certainly agree that the
firing frequency must be very important. This could be modelled by a
webaxon containing a positive integer say 0-255 or whatever.
>This could be modelled into webneurons by storing the times at which
>a webneuron has been fired. At least this data doesn't need resetting like
>storing which neuron called.
This sounds interesting.
This topic is difficult. The problem is that real neurons are active (or fired
up) all the time and always monitoring their inputs through their
continually running "script". Each real neuron is like a small computer in
itself. Until we have a microprocessor in each webneuron I don't think we
can do it 100% properly.
The closest we can get is to cancel the FIRE command and just let the
CPU keep cycling through the script of every webneuron on its computer.
Every webneuron should therefore be held in RAM at all times. This also
has the advantage of simplifying that horrible multi-tasking STACK, since
every webneuron is now in the "whole stack" at all times, and there is no
need to move them in and out.
In most cases the CPU would just dip into the webneuron script and then
out again after finding that nothing needed doing. However, the CPU will
obviously be slowed down a lot by this strategy.
I am very excited by this development. The only problem seems speed.
This can always be overcome later, perhaps by changing the computer
hardware. With this model it may easily be possible to build multiple
processor computers. Each processor could be considered a "server" of the
webneurons in its (non-shared) memory area. Perhaps each processor
could have its own IP address!
For the moment it will not be a problem to follow the "whole stack" option
providing we keep the number of webneurons to a reasonable number.
Modern CPU's could probably cope with hundreds. As a demo it is
attractive because it will be much simpler to implement. I recommend
trying this if you agree?
One thought occurs - are we losing the ability to duplicate conventional
programming tasks by moving so close to how the brain works? On the
other hand let's just try it and see what happens.
Having the FIRE command takes this burden off the CPU, but adds
programming complexity, and our main goal is to simplify. In any case the
modern CPU does not match the way the brain works at all. It is purely
central and totally undistributed. We need hardware redesign. The whole
Von Neumann thing is wrong, and has also misled software development.
Probably the best they could do at the time though.
]]>
|
John's Emails
