From the preceding introduction we can try to abstract a formal structure which represents - hopefully - all the key properties which are necessary to describe the behavior of self-organizing systems as generally as possible in a way that all the interesting details of concrete adaptive biological systems can be included if needed. This structure is intended as framework for the following more elaborated formalizations of genetic algorithms without and with a phenotype (cf. figure 4.1):
Figure 4.1:
Genetic Loop of Evolutionary Framework of Engineered Intelligent Systems
|
- Environment: All systems
are presupposed to be located in an environment which can change. These changes can be mapped onto a time-line .
- Systems: Every system
is an input-output system with a system function
mapping input
into output
as
. Every system also possesses at least one blueprint ('genotype')
of it's structure, which can be used for the formation of new blueprints.
- Life-Time: A living system
has a finite duration characterized by a start time ('birth') and an end time ('death'). The life-time of a living system is therefore given by the pair
.
- Mating-Function: From a certain point of time during their lifetime
two living systems
can produce a new blueprint ('genotype')
of a new system using their built-in blueprints
by applying a mating function
. Every new blueprint ('genotype') is assumed to be in a storage unit
from which it can be delivered to built up a new system4.1
- Growth-Function, Phenotype: The growth-function can take a blueprint from a storage unit and can built up a new system ('gphenotype')
. The time to grow is an initial part of the life-time of a system represented as
with
). During the growth-time of a system the system is assumed to begin functioning as far as the system-function
can do anything.4.2
- Learning-Capacity: The system-function
of a living system is assumed to be adaptive, i.e. it is assumed that this function can learn. Learning is defined with regard to the observable behavior of the system: If the behavior of a system after some point of time is better than before some point of time (with then the system is classified as having improved and this is assumed to be the result of some learning based on internal changes of the system-function . The usage of the term 'better' presupposes the availability of an operational criterion which can be applied to the observable behavior and which can be measured in an objective way. Example: the measurable time to find food in a certain environment is becoming 'shorter' after time than 'before' .
Gerd Doeben-Henisch
2013-01-14