Digital Signal Processing Reference
In-Depth Information
Fig. 13.8 Flow chart of the
GA procedure
Program
parameters
INITIAL
POPULATION
TRAINING &
GRADING
CREATING
INTERMEDIATE
POPULATION
GENETIC
MODIFICATIONS
- NEXT
GENERATION
TRAINING
GRADING
No
Yes
SELECTION
CRITERION
FULFILLED?
No
Yes
MAX. NUMBER
OF GENERATIONS
REACHED?
BEST NET
TOPOLOGY
After selection of ANNs for the intermediate population is done, several genetic
operations can be applied to create individuals of the next population. The most
important one is the crossover of two parent individuals (Fig.
13.10
) to produce
new descendants. A crossover can be an exchange of single neurons, groups of
neurons or whole layers between two ANNs. Crossovers are very important and
frequent at the beginning of a GA so that a wide variety of different individuals can
be produced. With the increasing number of generations the probability of
crossovers reduces since later in the process the displacement of weaker indi-
viduals by the successful one becomes more important than the creation of various
new individuals. Furthermore, mutations can take place which change the network
topology randomly by adding or removing neurons. Mutations are used to avoid
Search WWH ::
Custom Search