No abstract

The most common approaches to assessing the quality of training neural networks in the context of the problem of "small training sets" are analyzed. A review of the code implementation of the most universal approaches and ways of extending training/testing samples is carried out. The logic of the work of STN-module is analyzed. It can be inserted into existing convolutional architectures, giving neural networks the ability to actively spatially transform feature maps, conditional on the feature map itself, without any extra training supervision or modification to the optimization process.

No abstract

Context. Information monitoring technology is used to reduce information uncertainty about the regularity of air temperature changes during managing work in hard-to-reach places [1]. The task was to create a method for modelling one of the climatic indicators, air temperature, in the given territories in the information monitoring technology structure. Climate models are the main tools for studying the response of the ecological system to external and internal influences. The problem of reducing information uncertainty in making managerial decisions is eliminated by predicting the consequences of using planned control actions using climate modelling methods in information monitoring technology. The information technology of climate monitoring combines satellite observation methods and observations on climate stations, taking into account the spatial and temporal characteristics, to form an array of input data. It was made with the methods for synthesizing models of monitoring information systems [1] and methods of forming multilevel model structures of the monitoring information systems [1] for converting observation results into knowledge, and with the rules for interpreting obtained results for calculating the temperature value in the uncontrolled territories. Objective of the work is to solve the problem of identifying the functional dependence of the air temperature in a given uncontrolled territory on the results of observations of the climate characteristics by meteorological stations in the information technology of climate monitoring structure. Method. The methodology for creating information technologies for monitoring has been improved to expand its capabilities to perform new tasks of forecasting temperature using data from thermal imaging satellites and weather stations by using a new method of climate modelling. A systematic approach to the process of climate modelling and the group method of data handling were used for solving problems of functional dependence identification, methods of mathematical statistics for evaluating models. Results. The deviation of the calculated temperature values with the synthesized monitoring information systems models from the actual values obtained from the results of observations by artificial earth satellites does not, on average, exceed 2.5°С. Temperature traces obtained from satellite images and weather stations at similar points show similar dynamics. Conclusions. The problem of the functional dependence identification of air temperature in uncontrolled territories on the results of observations at meteorological stations is solved. The obtained results were used in the process of creating a new method of climate modelling within information technology of climate monitoring. Experimental confirmation of the hypothesis about the possibility of using satellite images in regional models of temperature prediction has been obtained. The effectiveness of the application of the methodology for the creation of monitoring information technologies during the implementation of the tasks of reducing uncertainty for management decisions during works in non-controlled territories has been proven.

No abstract