Group
Method of Data Handling is the realization of inductive approach
for mathematical modeling of complex systems. The contents of the
"Soviet Journal of Automation and Information Sciences" (previously
"Soviet Automatic Control", and then "Journal of Automation and
Information Sciences" - full translation of the ukrainian journal
"Avtomatica") reflect to some extent the history and facilitate
the tracking of the experimental and analytical development of the
GMDH technique.
The
first investigation of GMDH abroad had been made by R.Shankar in
1972 [39]. Later different variants were published by japanese [16]
and polish scientists. Their conclusion was following: GMDH is the
best method for solving of the AI problems - identification, short-term
and long-term forecast of random processes and pattern recognition
in complex systems. Mathematical GMDH theory showed that regression
analysis can be described as the particular case of GMDH [25].
Many
scientists are join to development of GMDH theory and applications
now. It is property not of the author
Prof. A.G.Ivakhnenko, but many other. This method was developed
since 1968 in the Combined Control Systems (CCS) group of the Institute
of Cybernetics in Kyiv (Ukraine). Contributions to the field have
come from many research areas of different disciplines. More than
220 candidate (Ph.D.) and 27 doctoral dissertations were defended
in this approach by scientists from several countries. Different
19 monographs devoted to this method have been published in 10 countries.
Period 1968-1971 is characterized by application of one regularity
criterion for solving of the problems of identification, pattern
recognition and short-term forecasting. As reference functions polynomials,
logical nets, fuzzy Zadeh sets and Bayes probability formulas were
used. Authors were staggered by high accuracy of forecastings. Noiseimmunity
was not investigated.
Period 1972-1975. The problem of modeling of noised data and
with incomplete information basis was solved. Multicriteria selection
and utilization of additional apriory information for noiseimmunity
increasing were proposed. Best experiments showed that with extended
definition of the optimal model by additional criterion noise level
can be ten times more than signal. Then it was improved using Shennon's
theorem of General Communication theory.
Period 1976-1979. The convergence of multilayered GMDH algorithms
was investigated. It was shown that some multilayered algorithms
have "multilayerness error" - analogical to static error of control
systems. In 1977 solution of objective systems analysis problems
by multilayered GMDH algorithms was proposed. It turned out that
sorting-out by criteria ensemble allow to choose the only optimal
system of equations and therefore to show complex object elements,
their main input and output variables.
Period 1980-1988. Many important theoretical
results were received. It became clear that full (contentative)
physical models cannot be used for long-term forecasting. It was
proved, that non-physical models of GMDH are more accurate for approximation
and forecast than physical models of regression analysis. Two-level
algorithms which use two different time scales for modeling were
developed.
Period 1989-to present time. New algorithms (AC,
OCC,PF) for non-parametric
modeling of fuzzy objects and SLP
for expert systems were developed and investigated. Present stage
of work is devoted to development and implementation, mainly into
economical systems, of twice-multilayered neuronets, which open
a new solution to the problem of self-organization of artificial
neuronets - models of human brains. Among the last it is possible
to point out work with
title "Inductive computer advisor for current forecasting of ukrainian
macroeconomy".
At
present time it can be noted that further increase of accuracy of
artificial intelligence problems may need application of preliminary
working up of experimental data by the Objective Computer Clusterization
(OCC) algorithm. Then the
main Combinatorial (COMBI)
algorithm should be used. Further increase of accuracy can be finished
by implementation at the output of processing system of the neural
nets with active neurons (TMNN),
which is found their input connections by self-organization themselves.
Glushkov Institute of Cybernetics is one of the principal centers
of research in the field of control theory and applications in exUSSR.
There was created the first on the Europe continent electronic computer
MESM in 1950.
|