Adaptive Resonance Theory .*;


The VB.Net example uses two buttons: The first to run the initial epoch, the second to run all later epochs.

Example 2: Font Recognition

This example best illustrates how the weights organize into "maps". The weight values start to look a little like the letters, but with a few missing pieces (depending on the level of constraint used).

The results are expressed in a more simplified manner than Example 1, though they still take some interpreting.

Example Results 1

With a vigilance parameter of 0.3 and 10 cluster units available, the weights will stabilize after two epochs. The final weight values are only shown for those clusters that were used:

Input vectors assigned to each cluster:

Cluster # 0: 2, 
Cluster # 1: 12, 
Cluster # 2: 0, 7, 
Cluster # 3: 8, 10, 11, 13, 
Cluster # 4: 14, 
Cluster # 5: 5, 9, 16, 19, 
Cluster # 6: 1, 3, 4, 6, 15, 17, 18, 20, 
Cluster # 7: 
Cluster # 8: 
Cluster # 9: 

Final weight values for each cluster:

..##...
.......
.......
.......
.......
.......
.......
.#.....
..#.##.

.......
.......
.......
.......
.......
.......
.......
.#.....
..###..

.......
.......
.......
.......
..#....
.#.....
.#.....
.#...#.
.#...#.

#......
#......
#......
#......
#......
#......
#......
#......
#......

...#...
.......
.......
..#.#..
.#.....
.#.....
.......
......#
##...#.

....#..
.....#.
.......
.......
.......
.......
.......
.#...#.
..###..

###....
.#.....
.#.....
.#.....
.#.....
.#.....
.#.....
.#.....
###....

 

Example Results 2

With a vigilance parameter of 0.7 and 15 cluster units available, the weights will still stabilize after two epochs. However, since the degree of similarity (vigilance) was increased, the algorithm became more "picky" about which cluster to place each input pattern. Therefore, more clusters were used.

Again, the final weight values are only shown for those clusters that were used:

Input vectors assigned to each cluster:

Cluster # 0: 7, 
Cluster # 1: 1, 3, 
Cluster # 2: 2, 9, 
Cluster # 3: 4, 6, 20, 
Cluster # 4: 5, 12, 19, 
Cluster # 5: 8, 10, 
Cluster # 6: 11, 
Cluster # 7: 13, 
Cluster # 8: 14, 
Cluster # 9: 17, 
Cluster # 10: 16, 
Cluster # 11: 15, 18, 
Cluster # 12: 0, 
Cluster # 13: 
Cluster # 14: 

Final weight values for each cluster:

...#...
...#...
...#...
..#.#..
..#.#..
.#...#.
.#...#.
.#...#.
.#...#.

#####..
.#.....
.#....#
.#....#
.#.....
.#....#
.#....#
.#.....
#####..

..###..
.#.....
#......
#......
#......
#......
#......
.#.....
..###..

###..##
.#.....
.#.....
.#.....
.##....
.#.#...
.#.....
.#.....
###..##

.....#.
.....#.
.....#.
.....#.
.....#.
.....#.
.....#.
.#...#.
..###..

#####..
#......
#.....#
#.....#
#......
#.....#
#.....#
#......
#####..

#######
#......
#......
#......
#####..
#......
#......
#......
#######

#....#.
#...#..
#..#...
#.#....
##.....
#.#....
#..#...
#...#..
#....#.

...#...
...#...
..#.#..
..#.#..
.#...#.
.#####.
#.....#
#.....#
##...##

####...
.#.....
.#.....
.#...#.
.#.....
.#.....
.#.....
.#.....
####...

..###.#
.#...##
#.....#
#......
#......
#......
#.....#
.#...#.
..###..

######.
.#....#
.#.....
.####..
.#.....
.#.....
.#.....
.#....#
######.

..##...
...#...
...#...
..#.#..
..#.#..
.#####.
.#...#.
.#...#.
###.###

 

public void footer() {
About | Contact | Privacy Policy | Terms of Service | Site Map
Copyright© 2009-2012 John McCullock. All Rights Reserved.
}