(MX)
download PDF
with first 40 pages
from my latest eBook
if you need more operators
click here
Keywords
adaptive,
fitness driven crossover,
schema preservation,
epistasis
Motivation
●
Protecting good schemata from destruction.
●
Searching through the solution space in promising directions
depending on fitness.
Source text
●
Louis
S.J.,
Rawlins G.J.E. (1991),
Designer Genetic Algorithms: Genetic Algorithms in Structure Design, in
Proceedings of the Fourth International
Conference on genetic Algorithms,
Morgan Kaufman, pp.5360
WEB:
http://citeseer.ifi.unizh.ch/louis91designer.html
http://citeseer.ist.psu.edu/louis91designer.html
Read also
●
Maini H.,
Mehrotra K.,
Mohan Ch.,
Ranka S. (1994), KnowledgeBased Nonuniform Crossover, in Proceedings of
IEEE International Conference on Evolutionary Computation,
Orlando
WEB:
http://intl.ieeexplore.ieee.org/xpl/abs_free.jsp?arNumber=350048
●
Vekaria
K.,
Clack C. (1999), Biases Introduced by Adaptive Recombination Operators,
in Proceedings of the Genetic and Evolutionary Computation Conference,
pp. 670677
WEB:
http://citeseer.ifi.unizh.ch/vekaria99biases.html
http://citeseer.ist.psu.edu/vekaria99biases.html
●
Chou
Ch.H.,
Chen J.N. (2000),
Genetic Algorithms: initialization schemes and genes extraction, in
Proceedings of The Ninth IEEE International Conference on Fuzzy Systems,
pp. 965  968
WEB:
http://intl.ieeexplore.ieee.org/xpl/abs_free.jsp?arNumber=839167
See also
●
KnowledgeBased Nonuniform Crossover
Algorithm
1.
select two parents A^{(t)} and B^{(t)}
from a parent pool
2.
create two offspring C^{(t+1)} and D^{(t+1)}
as follows:
3.
for i = 1 to n do
4.
c_{i}^{(t+1)}=a_{i}^{(t)}
5.
d_{i}^{(t+1)}=b_{i}^{(t)}
6.
end do
7.
for i = 1 to n do
8.
if p_{i}^{B}=1
and p_{i}^{A}=0 then
9.
c_{i}^{(t+1)}=b_{i}^{(t)}
10.
end if
11.
if
p_{i}^{B}=0 and p_{i}^{A}=1
then
12.
d_{i}^{(t+1)}=a_{i}^{(t)}
13.
end if
14.
end do
where:
P^{A}=(p_{1}^{A},...,p_{n}^{A})
– crossover mask of the parent A^{(t)}
P^{B}=(p_{1}^{B},...,p_{n}^{B})
– crossover mask of the parent B^{(t)}
Comments
●
The MX operator uses a mask vector to determine which bits of which
parent are inherited by the offspring. The first step is the duplication
of the bits of the parents. The bits of the first parent are copied to
the first offspring and, accordingly, of the second parent to the second
offspring (rows: 36). In the second step, the offspring exchange bits
among each other (rows: 9 and 12) at those positions where the mask
vectors of the parent were equal to 1, indicated domination of that
parent at that position and the mask vectors of the other parent were
equal to 0 (rows: 8 and 11).
●
The mask vectors are initiated in P(0) randomly.
During every GA iteration, the mask vectors are inherited by each
offspring from its parent. Then the mask vectors of the offspring as
well as the parents undergo modification. The modification process
(not described here) is based on the comparison of fitness of the
offspring and the parents. If good offspring were created, the
masks of the parents do not need to be modified and the masks of the
offspring may be very similar to those of the parents. In a situation
where bad offspring were created the masks of the parents as well
as of the offspring need to be modified.
Experiment domains
●
nbit parity checker
●
nbit adder
Compared to
●
1Point Crossover
