To implement an existing
algorithm, it is required that the project contains the following component:
·
A
detailed review of the algorithm
·
A
discussion on implementation issues such as data structure, computation
complexity/efficiency, convergence properties, etc.
·
Language
is not specified. However, your source
code should be properly commented, modularized, preferably portable (that is,
reduce platform specific features to minimum).
A make file for different platforms (e.g. HP workstation and PC) so that
the program can be ported easily to different platforms is desirable but not
mandated.
·
A
short tutorial on how to use the program should be included as part of the
report.
·
If
developed on a Unix platform, a man-page should be included. If developed on a
PC platform, it is preferred that program can run on a Window-95/NT
environment.
Suggested Algorithm
Implementation and References.
Textbook, page 226, optimal
brain surgeon algorithm
Textbook, page 243, Nonlinear
Conjugate Gradient Method for supervised training of MLP
Textbook, page 303, An RBF
network training algorithm using adaptation formulas for linear weights and
positions and spreads of centers of RBF network
Textbook, page 658, temporal
back-propagation algorithm
Suggested demonstration
programs implementation
A program using Hopfield
net to solve traveling salesman problem up to 20 cities. It should provide a GUI front end to
illustrate neuron outputs and graphical display of current tour.