Another Mitchell/Ware Release

Neural Net Simulator (c) 1987 by Fred Mitchell (M/W)

This software follows the same conventions of all Mitchell/Ware.
It may FREELY be distributed in any machine-readable form,
unmodified, in its entirety. Questions? Suggestions? Feel
free to call at 638-4936 12pm-8pm Mon-Fri, 2pm-8pm Sat and
Sun. Donations? Feel Free to send as much as you want to: 670
Magnolia Court, Bensalem, Pa 19020. Make your check payable to
Fred Mitchell. He will be more than glad to know that he
has your support, whether verbal or monetary.


NNS

I will assume that you have some working knowledge of the nature
of neurons. (Otherwise, why would you spend 4-8 minutes
downloading this package?) 256 neurons are simulated, each with
16 inputs (which I shall refer to as dentrites from now on) from
other neurons. Some can be inhibitory dentrites,
depending on the inhibitory selection, which is made by random
chance, to an averave percentage specified by the "% Inhibitory"
input. The wiring is also done automatically. And you can also set
the stimulus, and observe the response.

WALKTHROUGH

Now, let's explain how to run the program. After you have invoked
the program, you will see the initial message. When you're finish
reading it, hit RETURN and BE PATIENT!

Next, it will ask you to set the stimulus. This will be a 4X4 grid
to the left. Use the function keys to turn on or off indivual
'pixels'. Hit RETURN when done.

LEARNING RATE - sets the rate at which the threshold firing levels
of the neurons will be affected. If a neuron is firing, its
threshold may be raised. Conversely, if it is not firing, it may
be lowered. The rate is the percentage probability that any given
neuron will be adjusted.

SET THRESHOLD - sets the initial threshold of all neurons.
Remember that all neurons have 16 dentrites. The number should be
low - 1 to 6. Obviously, neurons will not fire above a 16 setting.
But, the thresholds (on an individual basis) will be lowered until
firing begins.

% INHIBITORY - sets the percentage of inhibitory dentrites that
will be present. As to which ones will be made inhibitory will be
up to chance. Experiment with any range from 0 to 100 percent.
Best results may be obtained around 10-20%, but try anything.

NEXT, you will see something like this appear:

.... ................ ....
.... ................ ....
.... ................ ....
.... ................ ....
................
................
................
................
................
................
................
................
................
................
................
................

(S)timulus (T)hreshold (I)nhibit (R)un continuously
(C)lear (L)earn SPACE - step ESC - escape

The "S", "T", "I", and "L" commands we have already gone over. You
can change anyone of them throughout the course of running. The
"R" command lets the net run continuously, and the "C" command
clears the network. SPACE allows you to single-step, and ESC
returns you to DOS.


LAST WORDS

This is not an exact simulation, of course, but it may give some
insight into the possibilities of using electronic neurons as
computers. By the way, the 16X16 grid in the middle represents the
states of all the neurons. The 4X4 to the right is the response.
Later, time (and God) permitting, I will do a much more accurate
model of the neuron, or do a Perpectron simulation. Any
suggestions or questions should be directed to Fred Mitchell at
(215) 638-4936, or by leaving a response on the BBS you found this
program to my pseudonym, PAUL ATREIDES. (I know, but there is a
reason that I am using a pseudonym.)