CIS 479 Assignment 4
Winter 1997
Your final assignment will be a neural network project. In
this assignment you are to develop a neural network which handles
some classification, diagnostic, or predictive task. The number
of neurons involved will vary according to your choice of
application. I would suggest that you try to identify an
application with 6 to 10 variables, has around 30 training cases,
and 5 or so test cases. You may work in groups or 2 or 3, if you
select a sufficiently rich application. If you work in a group
each person is to be assigned a percentage grade (indicating
contribution to total project by each group member).
Examples, might be computer system selection based on
desired component attributes or diagnosing common microcomputer
system problems. You might decide to examine your MP4 expert
system project or try to predict the NBA playoffs (or the outcome
of the recent NCAA tournament). You might try to diagnose
diseases, recognize characters, or select game moves.
You may implement your system using Brainmaker, Neuralworks,
or any other neural net tool you may have access to. You may
develop a system from scratch using Lisp (C or Pascal), or some
other language. I would suggest using Netmaker to build your
initial network definition file and then use Brainmaker to train
and test your network.
Chapter 14 in the Tanimoto text contains some comments on
the theoretical basis for neural networks. The documentation for
Brainmaker (and Netmaker) is in the CW Micro Lab. Sample network
definition files are on the Novell network. Prior to beginning
the design of a neural network, you should identify a domain
expert or some printed source for the facts used to train and
test your network. You will need to provide some evidence that
your network does in fact perform correctly.
To be sure things don't get stuck in the problem selection
phase, I would like you to give me a written note 1 week from
today describing your problem and indicating the members of your
group (if any). You are free to change problems, but you will
need to have me approve your new problem. You may not to drop out
of a group once you make the commitment to work with other
people.
You will need to turn in an annotated file listings for your
network, your training fact sets, and your test facts. I will
need to see whatever hardcopy you are able to generate
documenting the results of your training activities and the test
data runs. I will also need a 3 to 4 page memo discussing the
organization of your network, the variables involved, the source
of the facts processed by your network, and the length of the
training process. Your work is due at our last class period on
Tuesday, April 22nd. No late work will be accepted after our
final exam on April 24th.