Vol 32, No 1 (2026)

Two strategies for filling hash tables consisting of several subtables are considered, without displacing previously inserted elements, one of which is based on sequential, the other on uniform filling of subtables. An analytical model is proposed for estimating the parameters of filling in the subtables and the combined hash table for the two strategies under consideration. Computer modeling was carried out, which confirmed the high accuracy of the proposed model and the advantage of a strategy based on sequential filling of subtables.

The problem of constructing benchmarks for assessing the efficiency of feature selection algorithms used in regression problems is considered. The developed benchmark includes synthetic and real dataframes that differ in various data parameters, such as dimensionality, correlation between features, noise level, and the ratio between numerical and categorical variables. А generation method using various distributions and control parameters is proposed for synthetic data. А computational experiment using a feature selection method based on the partial least squares method was conducted, demonstrating the efficiency of the proposed benchmark for objective comparison of algorithms. The ways of further development of the benchmark are outlined, including expanding the set of parameters and using new data types.

The relevance of the study is due to the need to automate the analysis of text documents containing descriptions of physical and technical effects. In the context of modern development of science and technology, the volume of scientific articles, patent documents and grant reports is rapidly increasing, which requires effective methods for extracting and analyzing such key data. The theoretical significance of the work lies in the development of a new method for automatic extraction of physical and technical data in the form of keyphrases from natural-language text documents, ensuring the cooperation of deep learning technologies and methods of semantic-ontological text analysis. The practical significance of the work lies in the creation of a software for automatic extraction of elements of physical and technical effects from natural-language texts. The corpora of sentences (more than 4.3 thousand) was formed from the texts of patents containing physical and technical structured information in the form of descriptions of physical effects, solved technical problems. Neural network models keyT5, T5 and Bert were trained to extract physical and technical information. The T5 and KeyT5 models demonstrated high results in extracting keyphrases in the form of elements of descriptions of physical and technical effects (precision over 0.94, recall over 0.95).

The wild animal images captured by camera traps often have different quality due to such artifacts as low lighting conditions, complex background, meteorological conditions, the use of low-resolution video cameras, etc. А modern solution to the problem of recognizing wild animals is the use of deep learning models that need to be trained on "good" examples. Thus, assessing the informative content of such images is in the scope of interest. Image quality factors (brightness, contrast, blurriness and weather conditions), as well as the shape and position of the animal relative to the camera trap, are taken into account. Production rules have been developed for making decisions about dividing images into classes of varying informative degrees of information content. The experiments were carried out using a data set collected in the Ergaki Natural Park, Krasnoyarskiy Kray, in 2012-2021. The average error value of the proposed method for all classes is 6.4 % relative to the expert assessment.