Database Reference
In-Depth Information
1. Set t, the generation count, to 0.
2. For each species k,
• create an initial population, Pop
k
(t), randomly.
3. For each species k,
• evaluate the fitness of individuals in Pop
k
(t).
4. Compose the collaborative structure S by combining the best individual from each
species.
5. While the termination criterion is not matched:
•
For each species k,
•
evaluate the fitness values of individuals in Pop
k
(t) with respect to the col-
laborative structure S.
•
select individuals for reproduction according to their fitness values.
•
apply genetic operators to produce offspring.
•
evaluate the fitness values of the offspring with respect to the collaborative
structure S.
•
replace members in Pop
k
(t) with offspring, which gives the new population
Pop
k
(t + 1).
•
Update S.
Fig. 6.3.
Cooperativecoevolution algorithm.
evolution is applied in learning neural networks [6.6] and in learning sequen-
tial decision rules [6.7].
6.3 Learning Using Evolutionary Computation
Recently, a few attempts [6.24], [6.9] were made that apply evolutionary
computation to tackle the problem of learning Bayesian networks using the
search-and-scoring approach. In [6.24], genetic algorithms (GAs) are used,
but [6.9] uses evolutionary programming (EP).
6.3.1 Using GA
Larranaga et al. [6.24] proposed using genetic algorithms [6.30], [6.31] to
search for the optimal Bayesian network structure. In their research, the
network structure (composed of n nodes) is represented by an n × n connec-
tivity matrix C which is, in effect, the transpose of the adjacency matrix.
Each element C
ij
inthematrixisdefinedas:
C
ij
=
1,
if node j is a parent of node i
0,
otherwise.
With this representation, the ith row in the matrix encodes the parent set of
node N
i
(i.e., Π
N
i
). An illustration is given in Fig. 6.4.
By flattening the matrix, the bit-string representation is obtained:
C
11
C
21
C
31
...C
n1
C
21
C
22
...C
nn
.
Search WWH ::
Custom Search