Digital Signal Processing Reference
In-Depth Information
Algorithm 8.5
: The opt-aiNet algorithm
1. (
Initialization
): randomly initialize a population with a small number
of individuals;
2. While the stopping criterion is not met do:
(a) (
Fitness evaluation
) Determine the fitness of each individual of
the population and normalize the fitness vector;
(b) (
Replication
) Generate a number of copies (offspring) of each
individual;
(c) (
Mutation
) Mutate each of these copies in a manner that be
inversely proportional to the fitness of its parent cell, and also
keep the parent cell. The mutation is given by
c
=
+
(
)
c
α
N
0, 1
(8.5)
1
β
exp
fit
∗
)
α
=
(
−
where
c
is a mutated version of
c
,
N
is a zero-mean Gaussian
random vector with uncorrelated unit variance elements
β is a parameter that controls the decay of an inverse expo-
nential function
fit
∗
is the fitness of an individual normalized to lie in the
interval
(
0, 1
)
[
0, 1
]
A mutation is only accepted if the mutated individual
c
is
within the proper domain;
(d) (
Fitness evaluation
) Determine the fitness of all new (mutated)
individuals of the population;
(e) (
Selection
) For each clone group formed by the parent individual
and its mutated offspring, select the individual with highest fit-
ness and calculate the average fitness of the selected population;
(f) (
Local convergence
) If the average fitness of the population is not
significantly different from that at the previous iteration, then
continue; otherwise return to step 2(a);
(g) (
Network interactions
) Determine the affinity (degree of similarity
measured via the Euclidean distance) of all individuals of the
population. Suppress all those individuals whose affinities are
less than a suppression threshold σ
s
, except the best of them,
and determine the number of network individuals, termed
memory cells, after suppression;
(h) (
Diversity introduction
) Introduce a percentage of randomly
generated individuals and return to step 2