нейронна мережа

The article describes a study of identification of fake news based on natural language processing, big data analysis and deep learning technology. The developed system automatically checks the news for signs of fake news, such as the use of manipulative language, unverified sources and unreliable information. Data visualization is implemented on the basis of a friendly user interface that displays the results of news analysis in a convenient and understandable format.

The paper considers a method of generating fractal camouflage structures (grids) using a randomized system of iterative functions. This method allows for changing the base structure (type of mesh), which in turn makes it possible to determine the parameters by which the object can be identified as a fractal camouflage mesh. In the mathematical description of the improved RSIF, the color range parameters (set of colors) are introduced, allowing the fractal structure to be adjusted to the colors of the landscape where the camouflage net will be applied.

This paper investigates the impact of the number of hidden layers, the number of neurons in these layers, and the types of activation functions on the accuracy of classifying projectiles of six types (A – (artillery); A/M – (artillery/missile); A/R – (armor-piercing); A/RC – (armor-piercing- incendiary); M – (missile); R – (armor-piercing shells)) using a multi-layer neural network, evaluated by a confusion matrix.

A system for recognizing objects that are captured in real time on a video camera in a noisy environment that changes to the surrounding conditions has been built. The method of filling the database for mobile military objects was studied. For object recognition, the YOLO v8 neural network is used, which allows you to track moving and identify objects that fall into the video from the video camera. This neural network makes it possible to track objects with a change in scale, during movement with obstacles.

The relevance of creating a comprehensive system for meat control and identification to determine its freshness level has been demonstrated in the study. The drawbacks of traditional organoleptic and laboratory methods commonly used for meat inspection were analyzed. The authors presented the advantages and challenges of employing an electronic nose. A design for a meat control and identification system was proposed, which includes an Arduino Uno microcontroller, Raspberry Pi, USB to TTL adapter, gas sensors, color sensor, thermal camera, and image sensor.

This paper investigates the problem of embedding features used in datasets for training neural networks. The use of embeddings increases the performance of neural networks, and therefore is an important part of data preparation for deep learning methods. Such a process is based on semantic metrics. It is proposed to use ontologies of the subject areas to which the corresponding feature belongs for embedding. This work developed such a method and investigated its use for the task of categorizing text documents. The research results showed the advantage of the developed method.

 Features of designing a system of automated generation of poetic works, which opens up new opportunities for artistic speech and show business, especially the preparation of poems and songs have been considered. Quite often lyrics without special content become successful due to the lack of complex plots, as well as due to the unobtrusiveness and ease of perception by listeners. Well-known literature sources and available software products that can generate poetic works by combining different methods and algorithms are analyzed.

The requirements are formed, the method is chosen and the main stages of development of mobile means of neuro-like cryptographic encryption and real-time data decryption are considered. It is shown that the development of mobile means of neuro-like cryptographic encryption and decryption of real-time data with high efficiency of equipment is reduced to minimize hardware costs while providing a variety of requirements, characteristics and limitations. The tabular-algorithmic method of calculating the scalar product has been improved.

In this paper, we propose a method for optimizing the structure of a multilayer neural network based on minimizing nonlinear generalized error, which is based on the principle of minimum length of description. According to this principle, the generalized error is determined by the error in the description of the model and the error in the approximation of the data by the neural network in the nonlinear approximation.

A model of parallel sorting neural network of discrete-time has been proposed. The model is described by system of difference equations and by step functions. The model is based on simplified neural circuit of discrete-time that identifies maximal/minimal values of input data and is described by difference equation and by step functions. A bound from above on a number of iterations required for reaching convergence of search process to steady state is determined. The model does not need a knowledge of change range of input data.