The article discusses the application of neural network autoencoder in the problem of monochrome image colorization. The description of the network architecture, the applied training method and the method of preparing training and validation data is given. A dataset consisting of 540 natural landscape images with a resolution of 256 by 256 pixels was used for training. The results of comparing the quality of the outputs of the obtained model were evaluated and the average coefficients of metrics as well as the mean squared error of the VGG model outputs are presented.

Keywords: neural networks, machine learning, autoencoder, image quality analysis, colorization, CIELAB

Equipping roads with communications is complicated by the almost complete lack of roadside infrastructure, including power lines, as well as difficult terrain. When emergencies occur on this kind of country roads, residents are forced to seek help from nearby settlements that are well-connected. Therefore, providing suburban routes with communications is a key social task. Using an existing base station as an example, this article calculates the attenuation and propagation range of a radio signal for LTE technology and GSM technology, provides a comparative analysis, and uses methods of mathematical modeling and system analysis.

Keywords: LTE, GSM, Okumura-Hata model, Lee model, Longley-Rice model

The article discusses the use of a recurrent neural network in the problem of forecasting pollutants in the air based on actual data in the form of a time series. A description of the network architecture, the training method used, and the method for generating training and testing data is provided. During training, a data set consisting of 126 measurements of various components was used. As a result, the quality of the conclusions of the resulting model was assessed and the averaged coefficients of the MSE metric were calculated.

Keywords: air pollution, forecasting, neural networks, machine learning, recurrent network, time series analysis