ABSTRACTS OF ARTICLES OF THE JOURNAL "INFORMATION TECHNOLOGIES".
No. 7. Vol. 25. 2019

DOI: 10.17587/it.25.441-448

À. B. Barsky, D. Tech. Sc., Professor, e-mail: arkbarsk@mail.ru, D. I. Melnik, Ph. D., Senior Researcher, e-mail: mdi_dim@mail.ru, Scientific Research Institute (Moscow) Central Research Institute VVKO, Ministry of Defense of Russia

Neural Network Target Distribution Model for Computing System of Data Flow Architecture

A method of applying a logical neural network for solving the task of target distribution is proposed. The trajectory of the target for each coordinate is described by polynomials. It is assumed that, using a model of the object to be defended, the shooting complexes were fired at reference trajectories, and a knowledge base representing the logical neural network was built. It connects the values of the coefficients of the reference trajectories and the readiness values of the firing complexes with the decisions on their appointment to hit the target. To perform the associative sampling procedure, the receptor excitation values are set. The most "energized" neuron as a result of the activation function count indicates the selected shooting complex. The knowledge base is developed in the process of successful operation of the system. In the commands of the computing system of the data flow architecture, a program of the traditional type is given, according to which switching of actuators is carried out to perform the NEUROCOMPUTER procedure. A method is proposed for reducing the complexity of the scalar product of highly rarefied (by zeros) vectors, which is the basic operation for counting the values of the neuron activation function. To do this, use the CYCLE command with the listed parameter values, as was indicated in the original description of the ALGOL language.
Keywords: target distribution, reference trajectory, shooting complex, logical neural network, scalar product of vectors, data flow architecture

P. 441–448

To the contents