Title: | Взаимосвязь проблем распознавания образов, машинного мышления и обучения. Published in Russian: Взаимосвязь проблем распознавания образов, машинного мышления и обучения. |

Author(s): | Гриценко В.И., Шлезингер М.И. |

Source: |
Международный научно-технический журнал «Проблемы управления и информатики» .- 2020.- № 3 .- с. 108-136 . |

Abstract: | INTERRELATION OF PATTERN RECOGNITION, MACHINE THINKING AND LEARNING PROBLEMS.
The paper reviews the state-of-the-art in structural recognition, a research area in modern pattern recognition theory. The paper shows that basic problems of structural recognition go beyond machine learning theory and in such way slightly moderates the revived idea that the pattern recognition problem is en-tirely exhausted with machine learning. At the same time, the paper shows that main concepts and problems of structural recognition form a base for appropri-ate formalization of particular type of thought processes, which differ from learning and are called imaginative thinking. The main idea of this formaliza-tion relies on classical theory of Constraint Satisfaction Problem, one of the acknowledged paradigms of machine thinking. However the binding of this theory to real recognition tasks forces to generalize the theory itself and in such way to specify and refine the concept of machine thinking, for formalization of which the theory was intended. A generalized problem of structural recognition and imaginative thinking is formulated in the paper, classical Constraint Satis-faction Problem being its special case as well as its stochastic and optimization modifications, appropriate with regard to realistic recognition problems. For the Gibbs’ statistical model of recognized object, it is shown how recognition of such objects is reduced to such or other special case of generalized structural recognition problem. For more general statistical models, not necessarily Gibbs’ models, the application of known learning procedures to fixed size learning samples is analyzed. The flaw known as the short sample effect is ex-plored, its deep-rooted causes are determined as well as a way to overcome them. |

Files: | [pdf, Russian] |