Chapter 4 . Applications of Evolutionary Computation
53
problems (such as vision, language and cooperation), there was a clear attraction
towards the use of these principles to construct information systems that could be
used for modelling, discovery of patterns, construction of artificial systems,
design work and optimization.
Evolution in real-world systems can best be described as a two-step process of
heritable variation and selection. Variation occurs in the variety of behaviors that
are exhibited by individuals of organisms interacting in communities, populations,
and environments. The individual behavior (phenotype) is the product of a genetic
composition (genotype) and the interaction of that genotype with the cellular
environment. However, nature only measures the worth (fitness) of any individual
at the level of phenotype. Selection removes those individuals from the population
that do not have an appropriate fitness leaving behind those organisms with
sufficient fitness to pass their genotype to the subsequent generation. During this
process of heredity, variation to the genotype can occur, which may or may not
lead to alternative behavior in the progeny. The process of selection repeats itself
on the second generation of individuals, culling those with insufficient fitness.
Variation in the reproductive process is the source of change at the genetic level,
wh ich may translate into new innovation at the phenotypic level. Selection serves
as a filtering mechanism to ensure that individuals of low fitness are removed
along the way. Evolution, then, is the coupling of these two processes over time.
Sewall Wright (1932) offered the concept of an adaptive landscape as a means
to describe the manner in which evolution may proceed into novel adaptive zones.
Individual genotypes can be mapped into their respective phenotypes, which are
in turn mapped onto the surface of an adaptive topography. Each peak on this
topography represents a phenotype of high fitness (and, therefore, one or more
optimized genotypes). Evolution proceeds up the slopes of these peaks towards
solutions of increasing fitness as the selective mechanism culls inappropriate
phenotypic variants. However, this is an admiUedly idealized concept. In reality,
the adaptive topography changes with time as a function of the environment and
organism-environment interactions. Simulation of evolution in a computer can
demonstrate these same phenomena and can be used to search both static and
temporal fitness landscapes for regions of high fitness.
Within an engineering context, search algorithms define a problem in terms of
a search-space (the space of all possible solutions). Individual points in this
search-space represent solutions to the problem at hand. The goal is to find useful
solutions by traversing this search space in an efficient manner. However in many
engineering problems, the number of potential solutions is astronomical and an
exhaustive search of all solutions is infeasible in real time. Evolutionary
algorithms have proven to be successful at searching complex nonlinear adaptive
topographies (fitness functions) to return a near-optimal (or optimal) solution in
real time. Evolutionary algorithms are an extremely successful approach to
problems that can be framed as a search for a set of particular values, conditions
or structures. A requirement for such a system is the ability to measure a relative
fitness between individuals. Through the use of evolutionary algorithms, complex,
53
problems (such as vision, language and cooperation), there was a clear attraction
towards the use of these principles to construct information systems that could be
used for modelling, discovery of patterns, construction of artificial systems,
design work and optimization.
Evolution in real-world systems can best be described as a two-step process of
heritable variation and selection. Variation occurs in the variety of behaviors that
are exhibited by individuals of organisms interacting in communities, populations,
and environments. The individual behavior (phenotype) is the product of a genetic
composition (genotype) and the interaction of that genotype with the cellular
environment. However, nature only measures the worth (fitness) of any individual
at the level of phenotype. Selection removes those individuals from the population
that do not have an appropriate fitness leaving behind those organisms with
sufficient fitness to pass their genotype to the subsequent generation. During this
process of heredity, variation to the genotype can occur, which may or may not
lead to alternative behavior in the progeny. The process of selection repeats itself
on the second generation of individuals, culling those with insufficient fitness.
Variation in the reproductive process is the source of change at the genetic level,
wh ich may translate into new innovation at the phenotypic level. Selection serves
as a filtering mechanism to ensure that individuals of low fitness are removed
along the way. Evolution, then, is the coupling of these two processes over time.
Sewall Wright (1932) offered the concept of an adaptive landscape as a means
to describe the manner in which evolution may proceed into novel adaptive zones.
Individual genotypes can be mapped into their respective phenotypes, which are
in turn mapped onto the surface of an adaptive topography. Each peak on this
topography represents a phenotype of high fitness (and, therefore, one or more
optimized genotypes). Evolution proceeds up the slopes of these peaks towards
solutions of increasing fitness as the selective mechanism culls inappropriate
phenotypic variants. However, this is an admiUedly idealized concept. In reality,
the adaptive topography changes with time as a function of the environment and
organism-environment interactions. Simulation of evolution in a computer can
demonstrate these same phenomena and can be used to search both static and
temporal fitness landscapes for regions of high fitness.
Within an engineering context, search algorithms define a problem in terms of
a search-space (the space of all possible solutions). Individual points in this
search-space represent solutions to the problem at hand. The goal is to find useful
solutions by traversing this search space in an efficient manner. However in many
engineering problems, the number of potential solutions is astronomical and an
exhaustive search of all solutions is infeasible in real time. Evolutionary
algorithms have proven to be successful at searching complex nonlinear adaptive
topographies (fitness functions) to return a near-optimal (or optimal) solution in
real time. Evolutionary algorithms are an extremely successful approach to
problems that can be framed as a search for a set of particular values, conditions
or structures. A requirement for such a system is the ability to measure a relative
fitness between individuals. Through the use of evolutionary algorithms, complex,