Programmer 7500/MAX_PROGRAMMERS.iso/INFO/NEURLNET/NERVES.ZIP/NS.DOC

Text File | 1991-01-22 | 18.5 KB | 352 lines

NERVOUS SYSTEM CONSTRUCTION KIT FOR IBM COMPATIBLES

VERSION 2.0, 1-22-91

>>>>FOR A COPY OF THE MOST RECENT VERSION OF THE PROGRAM AND TURBO-
C/C++ SOURCE CODE (PLEASE SPECIFY DISK TYPE REQUIRED: 360KB, 1.2MB, OR
720KB), SEND $10.00 (U.S.$15.00 FOR DELIVERY OUTSIDE NORTH AMERICA);
REQUIRES EGA OR VGA; SOURCE CODE CAN BE COMPILED WITH BORLAND'S TURBO
C VERSION 2 OR TURBO C++ VERSION 1<<<<

By Pat and Greg Williams, Rt. 1, Box 302, Gravel Switch, KY 40328,
606-332-7606. Based on the ideas of Randall D. Beer, Department of
Computer Engineering and Science, Case Western Reserve University,
Cleveland, Ohio, as found in his INTELLIGENCE AS ADAPTIVE BEHAVIOR: AN
EXPERIMENT IN COMPUTATIONAL NEUROETHOLOGY, Academic Press, 1990 (ISBN
0-12-084730-2; to order, you can phone A.P. toll-free at 1-800-321-
5068).

READ THIS FIRST: These executable programs, source code files, and
associated documentation are supplied AS IS, WITHOUT WARRANTY. All
consequences of their use (or misuse) are solely the responsibilities
of the users, and in no case shall Pat Williams, Greg Williams,
Randall Beer, or Academic Press be liable for damages arising
therefrom. Modification of the source code is hereby permitted and
encouraged (we do hope you will share your additions and corrections
with us and with others). You may distribute copies of the original
distribution disk as you wish.

FILES ON DISK(S)

EXECUTABLES AND AUXILIARY FILES (IN ROOT DIRECTORY):
  EGAVGA.BGI - Borland graphics file used by NS87.EXE, NS.EXE, and
    NSPLAY.EXE (must be in default directory)
  NS87.EXE - main program for computers with math coprocessors
  NS.EXE - main program for computers without math coprocessors
  NSPLAY.EXE - animation playback program
  NSPRINT.EXE - utility program to print neuron parameters and
    connections
  L3R3.NEU, EDGE3.NEU, APP3.NEU - sample neuron files
  L3R3.ENV, EDGE3.ENV, APP3.ENV - sample environment files
  SINCOS.TAB - data file used by NS.EXE (must be in default directory)
  NS10.SAV - sample animation data file

SOURCE AND AUXILIARY FILES FOR COMPILING NS87.EXE (IN "87" DIRECTORY):
  *.C - C SOURCE FILES
  *.H - C HEADER FILES
  NS87.PRJ - PROJECT FILE FOR TURBO C++
  NS87.PJC - PROJECT FILE FOR TURBO C (MEDIUM MODEL)

SOURCE AND AUXILIARY FILES FOR COMPILING NS.EXE (IN "NO87" DIRECTORY):
  *.C - C SOURCE FILES
  *.H - C HEADER FILES
  NS.PRJ - PROJECT FILE FOR TURBO C++
  NS.PRC - PROJECT FILE FOR TURBO C (MEDIUM MODEL)

SOURCE AND AUXILIARY FILES FOR COMPILING NSPLAY.EXE (IN "PLAY"
  DIRECTORY):
  *.C - C SOURCE FILES
  *.H - C HEADER FILES
  NSPLAY.PRJ - PROJECT FILE FOR TURBO C++
  NSPLAY.PRC - PROJECT FILE FOR TURBO C (LARGE MODEL)

SOURCE AND AUXILIARY FILES FOR COMPILING NSPRINT.EXE (IN "PRINT"
  DIRECTORY):
  *.C - C SOURCE FILES
  DEFN.H - C HEADER FILE (MEDIUM MODEL)

I. INTRODUCTION

When we came upon Dr. Beer's beautifully done book (a revision of his
1989 dissertation), we immediately decided to attempt a PC-replication
of his computer simulation of the real-time behavior of a simplified
cockroach. The simulation is the first we've seen which starts at the
level of (reasonably realistically modeled) individual neurons and
ends up with whole-organism behavior. Dr. Beer's hexapod bug walks
with various gaits, wanders, follows edges, moves toward "food" when
its "energy" supply is low, "eats," and manages to avoid conflicts
among its various types of behavior. As programmed by Dr. Beer, the
bug has about 80 neurons, with about 150 connections and about 500
user-settable parameters. The program allows arbitrary connection of
neurons (and, via modifications to the source code [Turbo C compiler
needed for this], arbitrary specification of organism and environment
models), making it a general "construction kit" for small artificial
neural networks. Note that these networks are NOT of the generalized
type beloved by "connectionism," but rather are designed to perform
specific functions within the context of an organism's ecological
niche. Psychological control theory enthusiasts should note that Dr.
Beer's bug's nervous system isn't organized as a Powersian
hierarchical control system (see William T. Powers, BEHAVIOR: THE
CONTROL OF PERCEPTION, Aldine, Chicago, 1973), but networks organized
in that way certainly can be constructed using the program. TO
UNDERSTAND THE PROGRAM, YOU WILL NEED TO READ (AT LEAST PARTS OF)
INTELLIGENCE AS ADAPTIVE BEHAVIOR.

Suggestions, questions, modifications, etc., are welcome. Let us know
your thoughts about this stuff... down the road possibilities include
simulations of nematodes (having only a few hundred neurons in their
entire nervous system), Aplysia (modifiable synapses), spiders (prey-
catching, web building?), and op-amp-circuit realization of neurons
(for speedier computations in parallel).

We recommend INTELLIGENCE AS ADAPTIVE BEHAVIOR not only for its
development of the bug model, but for its extended critique of
traditional approaches in Artificial Intelligence. Here's the
foundation for a new field, folks: non-verbal AI. Thank you, Dr. Beer!

II. RUNNING THE PROGRAM

The executable files are NS.EXE (if you don't have a math coprocessor)
and NS87.EXE (if you have a coprocessor). Results of simulations with
the two files are quite similar BUT NOT IDENTICAL, because of round-
off and other differences between integer and long calculations in the
former and float and double calculations in the latter. Actually, you
can run either file on computers with or without coprocessors, but
NS.EXE runs about half as fast as NS87.EXE with a coprocessor, and
NS87.EXE runs about five times slower than NS.EXE without a
coprocessor.

With one of the executable files (and the Borland graphics file
EGAVGA.BGI, included on the distribution disk) on the default drive,
at the DOS prompt, type NS or NS87, as appropriate, and press ENTER.
Within a short time, you should see a menu of eight possibilities,
each of which can be accessed by pressing the key corresponding to its
first letter. For example, to quit the program, press Q.

There are two types of data files used by NS.EXE or NS87.EXE: neuron
files, which provide descriptions of particular nervous system
designs, and environment files, which provide descriptions of
particular environment ("world") configurations and initial
conditions. Three sets of neuron and environment files are included:
WANDER.NEU and WANDER.ENV, EDGE3.NEU and EDGE3.ENV, and APP3.NEU and
APP3.ENV. You can recall these files, and you can save and recall new
neuron and environment files.

To see the program in action, press R (for "Read files") and then, at
the prompt for a neuron file spec, key in WANDER.NEU and press ENTER.
At the prompt for an environment file spec, key in WANDER.ENV and
press ENTER. The current neuron and environment files appear at the
bottom of the menu screen. Now press S (for "Simulate"); you should
see an EGA-graphics-mode screen with a representation of a bug near
the lower left corner. To begin the simulation, press ENTER. The bug
will begin to move, and plots will be drawn at the right side of the
screen. Each of the plots except the bottom one shows the firing
frequency of a named neuron in the bug's nervous system (i.e., neuron
RBR) vs. (simulated, not real) time. The bottom plot shows the gait
pattern of the bug's legs -- a line means the foot is up (swing mode),
while no line means the foot is down (stance mode).

As shown at the bottom of the simulation screen, you can press SPACE
to enter single-stepping mode; each time you subsequently press SPACE,
another point will appear on each plot (corresponding to an increment
of 0.010 seconds of simulated time; actually, this increment
corresponds to two "internal" integration intervals, each 0.005
simulated seconds long). To return to continuous simulation, press
ENTER.

You can get back to the menu screen at any time by pressing Esc, but
don't do that just yet. First, note the "D:2" message at the bottom of
the screen. This means that the bug is being redrawn after every
second simulation step (that is, every 0.020 seconds of simulated
time). You can speed up the simulation a little by pressing D while in
the single-step mode, so that the message reads "D:4" (and the bug is
redrawn after every fourth simulation step, or every 0.040 seconds).
Subsequent presses of D result in redrawing of the bug every eighth
step, every 16th step, every 32nd step, every 64th step (giving a very
"jerky" bug, indeed!), and (back to) every second step.

Also, you can inject current into one of the bug's neurons, provided
that a variable of that neuron is being plotted. When in the single-
step mode, type N=I and then press ENTER, where the number N
designates the position (1=top, 5=bottom) of the plot for the current-
injected neuron and I designates the injected current in nanoamps
(between 0.1 and 99.9, positive or negative). The injected current is
shown at the bottom right of the screen. For example, to inject -15
nanoamps into neuron RBL (with its frequency plot in position 3), type
3=-15 and press ENTER. Then press SPACE for a single step or ENTER for
continuous simulation, with the current being injected.

The number between the "D:2" message and (when present) the indication
of injected current is the current energy of the bug. The maximum
energy is 999 units; during simulation, the energy decreases by one
unit after every eight steps (that is, after every 0.080 seconds of
simulated time); when the energy reaches zero units, "RIP" appears and
the simulation stops.

When the single-step mode is on, you can toggle (by pressing TAB)
between plots and a list of the names of the neurons (up to as many as
120) in the nervous system, color-coded according to their firing
frequency (normalized between 0 and 1). Try it. In the single-step
mode, press TAB, then press SPACE or ENTER to see the neuron names.
The colors of the names will change as the simulation runs, according
to the color key at the bottom right. (Names of neurons with firing
frequency = 0 are shown in the same color as the "0" in the key; names
of neurons with firing frequency between 0 and 1/3 are shown in the
same color as the block between "0" and "1/3" in the key; names of
neurons with firing frequency between 1/3 and 2/3 are shown in the
same color as the block between "1/3" and "2/3" in the key; and names
of neurons with firing frequency between 2/3 and 1 are shown in the
same color as the block between "2/3" and "1" in the key.

When in the single-step mode, you can initiate recording of the
simulation by pressing a function key (F1 through F10). This causes an
animation data file to be written to disk as the simulation proceeds;
later, you can play back the simulation at higher speeds by using
NSPLAY.EXE. The animation data file is written to the default
directory unless you previously specified a recording path by pressing
"P" (for "Path specification for recording") while at the menu screen.
The name of the animation data file is "NSX.SAV" with X = 1, 2,..., or
10, corresponding to the number of the function key pressed to
initiate recording. Note that there is no warning given about
overwriting an old file by a new one with the same name! Also note
that disk errors are handled via DOS -- if you must "abort," then the
program will terminate and you will be back at the DOS prompt! Once
recording has been initiated, a message appears at the bottom of the
screen: "RECORDING=X" (with X = 1, 2,..., or 10). To quit recording at
any time, press Esc. Other keys work as usual during simulation.

Now press Esc to return to the menu screen. Neuron and environment
files can be constructed, modified, and saved for later recall by
operations accessed via the menu.

Press W (for "Write files"). You are asked whether you want to write a
neuron file. Press Y (for "Yes"). Press N (for "No") when asked
whether you want to use the file spec WANDER.NEU. Then, in response to
the query for a neuron file name, type in TEST.NEU (we recommend using
the NEU extension for all neuron files) and press ENTER. Press "Y"
when asked whether you want to write an environment file, press "N"
when asked if you want to use the file spec WANDER.ENV, and then type
in TEST.ENV (we recommend using the ENV extension for all environment
files) and press ENTER. Note that TEST.NEU and TEST.ENV are now the
current files, as shown at the bottom of the screen. You are going to
modify these files.

Press N (for "Neuron modification"). When prompted for a neuron name,
press ? and then ENTER to see a listing of all neurons in the nervous
system specified by file TEST.NEU (which, at this point, is just a
duplicate of WANDER.NEU, with a different name). Now type LC (capital
letters, please; neuron names are case-sensitive) and press ENTER.
Near the top of the screen are several user-settable parameters for
neuron LC. Use the up and down arrow keys to move the cursor, then
type in new values. Move the cursor to the Gain value (1.00) and key
in 0.5, then press ENTER. Now the cursor is on the Intrinsic current
line; press ENTER to switch to NON-RANDOM, press ENTER again to
switch to RANDOM, and press ENTER a third time to switch back to NONE.
You see that there are three possibilities, or cases, for the
Intrinsic current. Each case is described in detail in Dr. Beer's
book. For the NON-RANDOM and RANDOM cases, you can specify several
parameter values. Sensory current function has seven cases, accessed
in turn by successively pressing ENTER. Similarly, Motor output type
has three cases. Now move the cursor to the Connections line and press
ENTER; a new screen appears, allowing specification of the neurons
which synapse on LC; synaptic connections can be DIRECT, or GATED or
MODULATED by a third neuron's synapse (see Dr. Beer's book for
details). Both direct and indirect current factors can be specified.
To delete a connection, move the cursor to the proper line and press
Control and D simultaneously. To return to the neuron parameters
screen, press Esc. You can alter the name of the neuron by editing the
top line (the "old" neuron is NOT deleted; this allows quick
duplication of neurons); you can delete the neuron by pressing Control
and D simultaneously. When you are done making modifications, press
Control and S simultaneously to save the modifications (otherwise they
will be lost!), then press Esc to return to the neuron name screen. At
the neuron name screen, you can key in another neuron name for
modification and press ENTER (remember, to see a list of all neurons,
press ? and then ENTER). And, by pressing Control and S
simultaneously, you can save the modified neuron file (you'll be
prompted for a file spec). Or you can press Esc to return to the menu
screen. Up to 200 neurons can be specified in a neuron file for use in
NS87.EXE; up to 150 neurons can be specified

At the menu screen, press E (for "Environment specification") to enter
the number of food patches, the coordinates of the centers of the food
patches, the sizes (in energy units) of the food patches, the number
of obstacle blocks, and the coordinates of the upper left corners of
the blocks. Press Esc before answering the final prompt to stop the
succession of prompts (retaining parameter values specified to that
point).

At the menu screen, press I (for "Initialize conditions" -- although
this can be done in the middle of a simulation, if you want) to enter
the coordinates of the bug, the angle of the bug, and the energy of
the bug. Press Esc to stop the succession of prompts (retaining
parameter values specified to that point).

At the menu screen, press G (for "Graph specification") to set up the
plotted graphs (1=top through 5=bottom) at the right of the simulation
screen. You are prompted for the type (none, neuron variable, or gait
pattern) for each graph number. If you choose to plot a neuron
variable, then you are prompted for neuron name and variable type
(voltage, frequency, or current); if you choose to plot voltage or
current, you are prompted for the axis scale (in millivolts or
picoamps, respectively). Press Esc to stop the succession of prompts
(retaining parameter values specified to that point).

At the menu screen, press W (for "Write files") to save neuron and/or
environment files with the current parameter values.

III. SIMULATION FILES

The three sets of ready-to-run simulation files included with the
program can be modified as desired.

WANDER.NEU and WANDER.ENV: Wandering controller neural circuits only.

EDGE3.NEU and EDGE3.ENV: Wandering controller and edge-following
neural circuits only; obstacle blocks.

APP3.NEU and APP3.ENV: Wandering controller, edge-following,
appetitive, and consummatory neural circuits; obstacle blocks and food
patch.

IV. ANIMATION PLAYBACK PROGRAM

The program NSPLAY.EXE runs animation data files previously written
during simulations. Animation playback speeds can be much higher than
the speed of the original simulation in NS.EXE or NS87.EXE (for
example, faster than real-time on our 16MHz 386SX with no math
coprocessor). At the DOS prompt, with NSPLAY.EXE in the current
directory, type:

  NSPLAY [PATH, IF NECESSARY]NSX.SAV

(where X = 1, 2,..., or 3, corresponding to an existing animation data
file) and press ENTER.

The distribution disk includes the sample animation data file
NS10.SAV; with both NSPLAY.EXE and NS10.SAV in the default directory,
type:

  NSPLAY NS10.SAV

and press ENTER. Assuming that sufficient memory is available, the
animation file is read (use a RAM disk for highest speed!) and the
simulation screen appears. Press ENTER to begin the animation at the
highest possible speed. Press SPACE to single-step. In single-step
mode, press - (repeatedly if desired) to slow the animation; press +
(repeatedly if desired) to speed it up. When the recorded simulation
is complete, the animation file is read again and the process repeats.
Press Esc at any time to return to DOS.

V. NEURON PRINT PROGRAM

NSPRINT.EXE prints out a listing of all the neurons (with parameters
and connections) in any neuron file. To run this utility, at the DOS
prompt type:

  NSPRINT filename

and press ENTER. The filename includes the drive, path, and extension
if necessary. For example, you might type NSPRINT C:WANDER.NEU, then
press ENTER. Then follow instructions given by the program.