Cognitron simulation program---written by

	Duncan Thomson
	Signal Processing Division
	Department of Electric and Electronic Engineering
	University of Strathclyde
	204 George Street
	Glasgow G1 1XW
	UK
	email: CNBR50@vaxa.strathclyde.ac.uk

	basic routines and data structures

adapted by

	P. Patrick van der Smagt
	University of Amsterdam
	Department of Computer Systems
	Kruislaan 403
	1098 SJ  Amsterdam
	THE NETHERLANDS
	email: smagt@fwi.uva.nl

	speeded up, improved I/O, added feedback routines, dynamic
	allocation


I have compiled cog under Turbo C (2.0) and SunOS 4.x.  For the
former, use makefile.tcc; for the latter, Makefile.


cog should be started up by
	cog [-v] [-w <wgt-file>] <in-file> [<out-file>]

The input file <in-file> should look like:

# iterations
# layers
SIZE (both width and height) per layer
S_SIZE = size of connectable area for excitatory and inhibitory input
O_SIZE = vicinity in which a neuron is said to win (O_SIZE x O_SIZE)
H_SIZE = area for lateral inhibition
q0 q1  = learning rates
# patterns
pat0
pat1
pat2
...
patn


For example, save the following file in "line5" and run "cog line5":
20
4
5 3 2 4
2 6
4
0x0,0x0,0x1F,0x0,0x0
0x4,0x4,0x4,0x4,0x4
0x1,0x2,0x4,0x8,0x10
0x10,0x8,0x4,0x2,0x1

Now set #iterations to 0, and run "cog -w line5.wts line5".


I would appreciate any suggestions or improvements.

					Patrick van der Smagt
