Biomedical Engineering Reference
In-Depth Information
Figure 4. Line formula used by the time decreased
activation
selected is directly proportional to its adjustment
value ruleta, and a tournament among individuals
technique. This technique is a variation of random
selection, in which first one chooses at random
a certain number of individuals from which the
best-adapted ones are selected.
The mutation operator used does not differ
from the one proposed by Holland. One individual
is chosen at random and then one of the weights
or line gradients in it is selected. A new value is
generated at random within the range adequate to
the parameter, and the individual is incorporated.
This individual is reevaluated and placed back
into the population depending on its new level of
adaptation to the problem (Figure 8).
The crossover operator must be modified, since
it must be applied to both parts of the individual:
the weight part and the activation functions part
(Figure 5). The crossover operator has been
designed as if both parts were two independent
y
=
mx
+
b
applying the genetic operators (Figure 7). Once
the population has been initialized, the genetic
operators of selection, crossover and mutation of
individuals start to be applied.
In order to measure the fitness of each indi-
vidual, as the classical training algorithms, it is
used the mean square error (MSE) between the
outputs of each ANN and the optimal outputs.
The selection of individuals to which the
crossover operator is going to be applied is made
first. Three values common in GA have been se-
lected for this operator: random selection within a
population of individuals, the famous Montecarlo
Technique, in which an individual's chance to be
Figure 5. Codification of the ANN. (a) Weights (b) line gradients
Input Neuron
Input Neuron
a
1
1
2
2
3
3
4
4
5
5
7
7
6
6
Output Neuron
1
2
3
4
5
6
7
1
2
3
4
5
6
7
3
1
4
7
b)
Input Neuron
Input Neuron
5
1
1
2
2
3
3
4
4
5
5
6
6
7
7
2
1
2
3
4
5
6
7
1
2
3
4
5
6
7
6
Search WWH ::
Custom Search