This package contains the implementation of the Cognon Simulator.

It implements a model of spike-based neuron processing discovered by 
David H. Staelin and Carl H. Staelin and described in ...

This simulator was written in C++ and developed under Linux.  It is 
publicly released under the GNU GPL version 2.0.

To build and run the simulator requires at least the following packages
to be installed:

    GNU C++ version 4.0 or later
    GNU Make
    Boost libraries

To build and run the program:

[1] Download the package from cognon.net

[2] Unpack the zip or tar file

[3] In a command shell run:

    $ cd cognon-1.0
    $ make all

[4] To reproduce the data tables shown in the paper run:

    $ make tables

[5] To run the sample configurations:

    $ ./cognon sample-input.spec

    This simply runs the configurations and outputs a CSV data file
    giving the learning performance for those configurations.

[6] To run a sample set of configurations which are optimized by
    a search process:

    $ ./cognon -c sample-optimize.spec

    This gives the system a base configuration from which it searches
    for the optimal values of W, H, Q, and G_m.

[7] Alternatively, you can build your own version of the simulator to
    develop a better search strategy, build multi-neuron simulations,
    explore alternative learning strategies, or pursue other interests.

Tests:

    There are a number of standalone unit tests for aspects of the
    Cognon simulator.  There are not enough of them, but there is 
    a start.  They are:

    alice_test
    bob_test
    cognon_test
    neuron_test
    wordset_test

Tables:

    You can replicate our experiments simply by typing:

    $ make tables

    which runs cognon on these files:

    table-2.1.spec
    table-2.3.spec
    table-2.4.spec
    table-3.2.spec
    table-3.3.spec
    table-3.4.spec

    There are a few standalone programs that perform the expensive search
    process to find the best configurations for the tables from the monograph.
    They can take a long time to run, but you may run the programs using:

    $ table-2.1
    $ table-2.3
    $ table-2.4
    $ table-3.3

    You can run all of them by typing:

    $ make searches

Graphs:

    You can replicate the experimental data for our graphs by typing:

    $ make graphs

    This will generate the Mathematica text input files containing the
    data used to create Graphs 2.1, 2.2, and 2.3 from "Models for Neural
    Spike Computation and Cognition":

    graph-2.1.data
    graph-2.2.data
    graph-2.3.data
    predict-2.1.data
    predict-2.3.data

This package contains the following files:

COPYING
Makefile
README
alice.cc
alice.h
alice_test.cc
bob.cc
bob.h
bob_test.cc
cognon-orig.h
cognon.cc
cognon.h
cognon_main.cc
cognon_stats.cc
cognon_test.cc
compat.cc
compat.h
monograph.cc
monograph.h
mtrand.h
neuron.cc
neuron.h
sample-input.spec
sample-input.spec
sample-optimize.spec
sample-optimize.spec
table-2.1.cc
table-2.1.spec
table-2.3.cc
table-2.3.spec
table-2.4.cc
table-2.4.spec
table-3.2.spec
table-3.3.cc
table-3.3.spec
table-3.4.spec
wordset.cc
wordset.h
wordset_test.cc