Performed some algorithms consideration of data analysis, that’s shown their base
simplicity. Genetic optimization were transformed to two – step version of stochastic search,
whose steps are preliminary mixing of primary search results (interpreted as crossing) and
secondary stochastic search (interpreted as mutation). Potential function method allowed
implementing the simple procedure of clasterization without any additional requirements to
input sample. Learning algorithm of perceptron’s classifier was used the preliminary
averaging in secondary neurons with any constant subtraction. Additional adaptive
coefficients normalizing do it insufficient at maximization used as decisive function. Fuzzy
control learning were developed that’s based on control transactions frequencies equalization
at equidistant sample of input states.
1. Statistical methods for digital computers, ed. by Enslein K. ea, NY.,
Wiley, 1977, 464 p.
2. Rutkowskaya D. ea Neyronnye seti, geneticheskiye algoritmy i nechetkiye
sistemy (Neuron’s nets, genetic algorithms and fuzzy systems), publisher
“Telecom”, 2006, 452 p.
3. Rotshtein A. Intellektualnye technologii identifikatsii, Vinnitsa, publisher
“Universum”, 1999, 320 p.
4. Aizerman M. ea Metod potentsialnych functsij v teorii obucheniya mashin
(Potential functions method in mashine learning theory), M., publisher
“Nauka”, 1970, 384 p.
5. Vapnik V. ea Teoriya raspoznavaniya obrazov (statisticheskiye problem
obucheniya), (Image recognition theory (statistical learning problems)),
M., publisher “Nauka”, 1974, 416 p
6. Tsypkin J. Adaptatsiya i obucheniye v avtimaticheskich sistemach (Adaptation
and learning in control systems), M., publisher “Nauka”, 1968, 400 p
Moiseev Alexander Alexandrovich
Candidate of Technical Sciences
State Research Institute of Chimmotology
Moscow, Russian Federation
