| · · · ·
· · · · |
 |
|
· · · · · ·
· · · · · · · · · · ·
· · · · · |
| Trans generation picture
memory |
A trans generation picture memory builds the basics for the evolutionary
process which will be described subsequently. It has to be recognized,
that this process is a data orientated process which applies picture transforming
methods on existing images.
|
|
· · · · · ·
· · · · · · · · · · ·
· · · · · |
| Initialization process |
The data oriented evolution uses for the initializing step a selection
of m > 1 motives a starting population P(t=0) from the picture memory.
These motives can be basic motives which could be derived directly from
digitized photos or images which underwent a evolutionary process before.
This step is called the initialization process and demands a selection
strategy which is applied by the artist. This could be seen as the first
act of interaction and communication in the framework of understanding
the evolutionary art process as
communication (in german).
In this strategy the number of individuals of a population is settled
and the concrete images are selected. This is normally done by an evaluation
of the artist. In contrast to normal evolutionary algorithms the number
of individuals can be changed from generation to generation.
|
|
· · · · · ·
· · · · · · · · · · ·
· · · · · |
| Initial Population |
Through the initialization the original population is founded.
As the evolutionary process is an iterative process this expression should
be seen further on as synonym for parent population, on which the reproduction
operations are used. |
|
· · · · · ·
· · · · · · · · · · ·
· · · · · |
| Reproduction Strategy |
The reproduction is carried out according to the reproduction
strategy, which has to be defined by the artist as well. The strategy consists
of selection for reproduction, recombination and mutagenation and the number
of the individuals of population. |
|
· · · · · ·
· · · · · · · · · · ·
· · · · · |
| Selection for Reproduction |
The selection of a parent for asexual or for parent individuals for sexual
reproduction is regulated through the strategy of reproduction sel(R),
which is specified by the artist. In evolutionary algorithm the parents
are selected randomly (evolutionary strategy) or they are selected according
their fitness (genetic algorithms). There are two principal classes of
reproduction:
|
|
· · · · · ·
· · · · · · · · · · ·
· · · · · |
|
Asexual Reproduction
Sexual Reproduction
|
Using the asexual reproduction
means, a particular image is selected as parent. On this parent picture
the reproduction operation is applied. In this case this means, the picture
will be copied. To this exact copy one or more transformation operations
are applied which are called "mutational operations" within the evolutionary
algorithms. The way of the transformation and the parameters used are
describe by the artist.
Within the sexual reproduction
at least two pictures as parent pictures are selected, which breed the
offspring. These pictures are selected in the data orientated evolution
segments of the pictures are selected and combined to an inherent picture.
This operation is known as recombination operation. The selection and
the number of the segments are ruled by the recombination strategy.
|
|
· · · · · ·
· · · · · · · · · · ·
· · · · · |
| Reproduction strategy and
the chances for selection |
Which mutation- or recombination operation should be used
has to be decided by the artist in the framework of the reproduction strategy.
All established operations can be used, all of them have a certain chance
of being used. These odds of being used can be determined either from the
artist or by the process itself. |
| Interim Population |
The complete offspring is included into an interim population.
The strength l of the population is laid down before in the reproduction
strategy. |
|
· · · · · ·
· · · · · · · · · · ·
· · · · · |
| Selection for the transfer
into the offspring |
After the goal number of offspring individuals has been generated, the
reproduction process is shut down and the selection process begins. This
is a decision process, in which individuals are selected for the next
generation.
Two principal strategies are available which can be derived from the
different methods of building the interim population:
|
|
· · · · · ·
· · · · · · · · · · ·
· · · · · |
| One generation life span |
1. Only the offspring is transferred into the interim population, the
parent generation is left apart. Hence the lifetime of all individuals
is one generation.
|
|
· · · · · ·
· · · · · · · · · · ·
· · · · · |
| Unknown life span |
2. Parents and offspring are transferred together into the
interim population from which the next generation is formed. In the strategical
class the members of the population have a unknown life span as long no
other limitations are introduced. |
|
· · · · · ·
· · · · · · · · · · ·
· · · · · |
| Basics of sel(N) |
Basics of the selection operation sel(N) is the evaluation
of the individuals by the artist. Two strategies might be applied: |
|
· · · · · ·
· · · · · · · · · · ·
· · · · · |
| Absolute Evaluation |
Every single individual is attached a scalar value which might belong
to the interval [0, 1], hence can be normed.
|
|
· · · · · ·
· · · · · · · · · · ·
· · · · · |
| Relative Evaluation |
All individuals are evaluated again, based solely on the present
individuals. A range order is built up. This strategy lead to deviations
when parent and offspring individuals are integrated in the interim population
because the parents, which were evaluated in the previous generation, are
newly evaluated now. The measure are in individuals in the interim population. |
|
· · · · · ·
· · · · · · · · · · ·
· · · · · |
| Multi objective evaluation |
There exists evaluation possibilities which add more scalars
as evaluation of different attributes (multi objective or vector evaluation).
This evaluation is not discussed here. |