Name
Shestakov Alexander Valentinovich
Scholastic degree
•
Academic rank
—
Honorary rank
—
Organization, job position
South Federal University
Web site url
—
Articles count: 2
This article considers the organization of neuralnetwork
interactions based on models of cellular
automata. These models are focused on improving
the efficiency of the iterative processes of
functioning of neural networks and their learning.
The models considered in this article are presented in
the form of two-level hierarchical structures. Models
of the lower level are defined as "cellular neural
element". They are based on formal descriptions of
the dynamic neurons with the additional insertion of
the state functions and the special procedures of
formation of the specified function. Also, we have
added special methods for forming patterns of
activation functions. The conception of developed
models is based on the use of the theory of graphs,
theory of neural networks and the mechanism of
cellular automata. These models will be used as the
basis for software modeling
The article discusses the model of a neuron with
memory. The proposed model is formed based on the
model of the dynamic neuron. The proposed solutions
allow to efficiently model dynamic processes and to
provide a two-level scheme for evolution of neural
networks. The first level of evolution is achieved using
traditional neuro-evolution models that allow you to
create or to modify basic network settings. The second
level allows you to track the history of the
development process. The article presents a formal
description of the proposed models and algorithms of
functioning of the network, is the rationale for the use
of the proposed models