A demonstrational implementation of a simple multilayered perceptron
(A fully connected feed-forward artificial neural network) with
steepest descent training, two faster heuristic learning algorithms,
and a possibility of integrating the non-smooth local optimizer
Solvopt (by other authors). 


The novelty in this package is the possibility of using non-smooth
cost functions for training (requires some non-smooth optimizer).

Contents:

  m/          - Matlab implementation of the 'MLP engine'

  mutil/      - Matlab tools required for research code and demos

  demo/       - Demonstration and research applications

  courseware  - A stand-alone version for class-room use (in Matlab
                and also a separate implementation in Java)

  externals   - A placeholder for third-party open-source tools and
                open-content data and a bash-script to fetch these
                over the internet.

Most of the code was created as part of a EU-funded (European Regional
Development Fund via Tekes) project called RISC-PROS during
2008-2010. Licensed under the MIT License as stated in the original
project agreement (LICENSE.txt in each subdirectory is the actual
license text).

We do not distribute software or data from external sources, even
though some are required for our demonstrations and research code to
work. The directory ''externals/'' contains a bash-script that can
(try to) automatically fetch the required bits using an internet
connection and standard unix tools. On a Windows system (without
MinGW, Cygwin, or similar unix-like environment), or in case of
problems with the script, you may manually download and unpack the
stuff that the script would. Installation of the unix-like tools is of
course recommended.

No warranty and no promises of support of any kind. But you may
contact the author for any questions, improvements, or issues
regarding this software.

Author: Paavo Nieminen <paavo.j.nieminen@jyu.fi>

Copyright (for most parts): University of Jyvaskyla, 2008-2011



[Note: The 'a' letters in Jyvaskyla should be with two dots as in
LaTeX 'Jyv\"{a}skyl\"{a}' but I didn't want to use other than ASCII
coding for this README.]