Geoscience Reference
In-Depth Information
ef
ciently simulate the functioning of A by using models built for some system
operating in the real time. It is necessary to build a sequence of improving models
{A
k
}, where k = 1, 2,
. Cybernetics began to ponder how this could be attained
virtually from the moment
…
the
first, primitive by today
'
s standard, machines
emerged. It was then that the idea of creating arti
cial intelligence and designing
“
appeared, giving rise to the development of robotics. Being
aware that a computer is an obedient executor of some program, cybernetics
doggedly worked on the possibility of imparting to the machine a measure of
unpredictable behavior with some facets of
thinking machines
”
. Treating intelligence as
the ability to correctly respond to a novel situation, scientists came to the conclusion
that machines capable of adapting to the level of individual components and their
structural organization are in fact feasible.
The idea of model-free learning for computer is discussed during last years. The
basic problem is how to teach a computer by changing the structure of relations
between its elements. This analogy with the neural operation of the brain estab-
lished in neurophysiology yielded fruit and helped advance far along the scale of
arti
“
innovation
”
cial intelligence capabilities. Fogel et al. (1966) gave an impetus to an entirely
new cybernetic trend.
Evolutionary modeling on the whole can be represented by a hierarchical two-tier
procedure (Fig.
1.15
). At the
first tier there are two constantly alternating processes
conventionally termed the structural adaptation and utilization processes. At the kth
stage of adaptation at the time of operation of the structural adaptation algorithm,
models of the sequence {A
s,i
}, i=1,
…
,M
s
are synthesized. Special rules form
memory from the most effective models
A
s
; ...;
A
s
[
. At the stage of utilization
following the stage of adaptation, the system selects the most ef
\
cient model.
A schematic diagram of the ith step of kth stage of adaptation of models is represented
in Fig.
1.16
. The
“
object
”
block here denotes that the real-world object is de
ned by
some previous history.
The procedure of the evolutionary selection of models provides for a virtually
time-unlimited operation of the system under irremovable information uncertainty.
Apart from the previous history which, as a rule, does not meet the requirements of
traditional statistical analysis, the researcher has no other available information.
Under such conditions one obviously has to use the maximum of available infor-
mation, in particular, that on the operation of the adaptation and utilization stages.
Fig. 1.15 Schematic representation of the concept of evolutionary modeling (Bukatova et al.
1991)
Search WWH ::
Custom Search