Information Technology Reference
In-Depth Information
world problems, as in nature, there is no such thing as perfection and minor
improvements are always possible. The disadvantages of such an evolution-
ary strategy, however, become evident when average fitness reaches best
fitness before a perfect or good solution is found. Figure 12.8 shows such a
case where the population stabilized on a mediocre solution. In this particu-
lar case, after generation 86 adaptation became impossible because all indi-
viduals share the same genetic makeup. Indeed, the small success rates typi-
cal of populations undergoing recombination alone (see, for instance, Figure
12.1 and Table 12.2) indicate that, most of the times, homogenizing
populations converge before finding a good solution because they become
irrevocably stuck in some local point, not necessarily optimal.
200
Best fitness
Avg fitness
150
100
50
0
0 0 0 0 0 0 0 0 0 0 0
Generations
Figure 12.8.
Convergence on a mediocre solution in homogenizing populations.
The parameters used in this experiment are exactly the same as in Figure 12.7.
It is worth noticing that in the experiments summarized in Table 12.2, totally
random initial populations were used and, therefore, the number of viable
individuals in those initial populations was not controlled. In the next section I
will show how the number of viable individuals in initial populations can be
rigorously controlled in order to analyze the founder effect in artificial evolu-
tionary systems.
12.2.2 The Founder Effect in Simulated Evolutionary Processes
For this analysis, a smaller, totally random “initial” population (founder popu-
lation) composed of a certain number of viable individuals is created. That
is, the run only starts when all the members of the founder population are
viable or, in other words, have positive fitness. These founder individuals
