Test correct ave rho overlapping rectangular distributions ave num
B subnetworks. Notice that this requirement is more demanding than is required for the conclusion of the Two-pass Learning theorem.
|
||
---|---|---|
|
||
|
||
|
||
0 | 0.2 | 0.4 | 0.6 | 0.8 |
---|
Rho
(b) Overlapping Rectangular Distributions
400 | 0 | 0.4 | 0.6 | 0.8 |
|
|||
---|---|---|---|---|---|---|---|---|
350 | ||||||||
300 | ||||||||
250 | ||||||||
200 | SD | |||||||
150 | ||||||||
100 | ||||||||
50 | ||||||||
0 |
Rho
subnetwork. The vigilance setting for the B subnetwork was fixed at 1.0. This is standard for classification problems, since binary coded class labels are used as inputs to the B subnetwork. Finally, a Bayesian classifier was applied to the testing data and performance was calculated, giving a basis for comparison.
900 | ||
---|---|---|
800 | ||
700 | ||
600 |
|
|
500 | ||
400 | ||
300 | ||
200 | ||
100 |
0
0 | 0.2 | 0.4 | 0.6 | 0.8 |
---|
|
350 | 0 | Ave |
|
0.4 | 0.6 | 0.8 | |
---|---|---|---|---|---|---|---|---|
300 | ||||||||
250 | ||||||||
200 | ||||||||
150 | SD | |||||||
100 | ||||||||
50 | ||||||||
0 |