Dale Patterson
dpatte3@cs.umbc.edu
Description
This page lists resources and results from a project in Artificial Neural Networks.
The objective of this project assignment is an exercise of Kohonen Self-Organizing-Map
(SOM). First, generate 60 2-dimensional, unit length input vectors, then construct a ring
SOM of 12 nodes according to the algorithm in Section 5.5 of the textbook, "Elements of
Artificial Neural Networks".
Source Code
som.zip includes the following programs.
clustered.m
distance.m
load_som.m
norm_circle.m
project2.m
som.m
som_animate.m
som_movie.m
som_plot.m
som_plot_stats.m
som_statistics.m
Results
The project report discusses executing the som algorithm
on ten sample sets and includes plotting of the SOM at various times for sample set six.
The following .avi files show the progress of SOM from the initial state until termination
som6_1.avi The first 30 epochs
som6_2.avi The second 30 epochs
som6_3.avi The third 30 epochs
som6_4.avi The final 30 epochs
Each is two minutes long and due to space constraints in not high quality.
The data recorded from the ten samples sets is found can be downloaded in
here. Each weight file is a binary file that can be used
by "load_som.m", "som_animate.m", "som_plot.m", "som_movie.m", "som_plot_stats.m",
and "som_statistics.m" to visualize the topological structure of the som at different
times for the different samples.
Below, I've included plots for each sample set for initial state, after 3 epochs, after
10 epochs and the final configuration.
Sample Set 1
Sample Set 2
Sample Set 3
Sample Set 4
Sample Set 5
Sample Set 6
Sample Set 7
Sample Set 8
Sample Set 9
Sample Set 10