Graph Based Evolutionary Algorithms - Advancing Artificial Intelligence Through Biological Process Applications - page 327

CONCLUSION

the problem. This would give a designer the ability
to verify the number of distinctly different solu-
tions in an evolving population to ensure that the
desired number of solutions is being delivered. In
addition, a designer with knowledge of the make-
up of the solutions occupying the search space
would be able to incorporate domain knowledge
into the evolving designs. This may be done to
speed convergence by introducing known solution
elements or to encourage exploration by manual
introduction of diversity.
Work continues applying GBEAs to engineer-
ing and scientific applications. Some of these
projects include responsible antibiotic use in
both production animals and humans, interactive
optimization of thermal in profiles in a virtual
environment, and ideal use of fuel resources for
the power industry. As information from these
and other problems become available, it will be
added to the developing taxonomy of evolution-
ary computation problems. The theories derived
from this work can be tested against newer test
problems, with an eventual goal of establishing
guidelines that can be used to select a graph
and other parameters a priori that achieve near
maximum benefits.

Graph Based Evolutionary Algorithms provide a
simple-to-use tool with modest computational cost
which allows the user to achieve a desired level
of diversity in an evolving population. This tool
can be used on many types of evolutionary algo-
rithm problems beyond the bit string, real string,
and genetic programming problems used here as
examples. The preferred graph is problem spe-
cific, although for the more difficult and deceptive
test problems studied here a diversity preserving
graph performed best. In addition, GBEAs can
be used to decrease the necessary population size
to find a solution and to decrease time to solution.
They can also be used to find multiple solutions
if such are desirable. The results of the applied
problems where GBEAs were used indicate that,
at a minimum, modest amounts of diversity pres-
ervation yield a higher quality solution. For the
stove design problem this yielded a large saving
of computational and wall times by decreasing the
number of mating events, each of which requires
lengthy fitness evaluations. Preliminary results
for the hydraulic mixing nozzle problem show a
single solution found using a highly connected
graph and three distinctly dissimilar solutions
when using a diversity preserving graph. In
addition, GBEAs can be used to classify both
evolutionary computation problems and evolu-
tionary computation methods. By classifying
evolutionary computation problems, it is possible
to select an algorithm or method a priori that
will give the desired result more rapidly. Taxo-
nomic classification of evolutionary computation
problems, enabled by GBEAs, permits objective
evaluation of new techniques.

REFERENCES

Ashlock, D. & Lathrop, J. I. (1998). A Fully Char-
acterized Test Suite for Genetic Programming. In
Evolutionary Programming VII , (pp. 537-546).
New York, NY: Springer-Verlag.

Ashlock, D., Guo, L., & Qiu, F. (2002). Greedy
Closure Genetic Algorithms. In Proceedings of
the 2002 Congress on Evolutionary Computation
(pp. 1296-1301). Piscataway, NJ: IEEE Press.

FUTURE RESEARCH DIRECTIONS

Ashlock, D. & Schonfeld, J. (2005). Nonlinear
Projection for the Display of High Dimensional
Distance Data. In Proceedings of the 2005 Con-
gress on Evolutionary Computation , vol. 3 (pp.
2776-2783). Piscataway, NJ: IEEE Press.

One useful tool would be the inclusion of a method
to discern the level of population diversity while
the graph based evolutionary algorithm is solving