AUTOMATIC DOCUMENTATION AND MATHEMATICAL LINGUISTICS最新文献

The paper presents a comparative analysis of methods for the representation of three-dimensional objects to perform real-time Boolean operations in the Unity game engine environment. Four main approaches are considered: polygonal representation based on constructive solid geometry (CSG), sign distance functions (SDF), voxel methods, and CAD-systems with boundary representation (B-Rep) and NURBS-surfaces. An experimental study of the performance of polygonal algorithms of Boolean operations and SDF functions based on ray-marching implementation is conducted. It is revealed that polygonal methods are characterized by high initial system construction costs, but they provide stable performance during long operations and preservation of transformation results. SDF functions demonstrate high-speed operations and flexibility in creating smooth transitions between objects but are limited in their application to long-term tasks due to the peculiarities of the computational model. The areas of effective application of each approach are identified: polygonal methods are recommended for tasks that require precise geometric control and integration with traditional graphics pipelines, while SDF functions are optimal for procedural generation, multilayer material rendering, and the creation of dynamic visual effects. The results of this study can be used in the development of interactive simulators, game applications and virtual reality systems.

An information-analytical system (IAS) is presented for high-speed grayscale image segmentation using a modified defuzzification method with triangular membership functions. The aim of this study is to analyze the effects of simplifying the defuzzification formula on the accuracy and contrast of object delineation. The proposed approach includes adaptive learning of the weight coefficient, enabling a dynamic adjustment of the defuzzification process, depending on target values. The basic method of averaging membership values and a modified version that accounts for nonlinear weights are compared. Experiments conducted on 1024 × 720 images demonstrate that the developed IAS provides high segmentation accuracy and improved object contrast with minimal computational costs. The results confirm the superiority of the proposed method over traditional approaches, emphasizing the prospects for applying artificial intelligence to computer vision tasks.

Human pose estimation (HPE) is among the most relevant topics in computer vision research. This technology can be applied in various fields, such as video surveillance, medical care, and sports motion analysis. Due to the increasing demand for HPE, many libraries for this technology have been developed in the last 20 years. Since 2017, many HPE algorithms based on skeletal models have been published and packaged into libraries for easy use by researchers. These libraries are important for researchers who want to integrate them into real-world applications for video surveillance, medical care, and sports motion analysis. This paper investigates the strengths and weaknesses of four popular HPE advanced human pose recognition libraries that can run on mobile devices: Lightweight OpenPose, PoseNet, MoveNet, and Blase Pose.

This article is devoted to the application of machine learning methods to improve the quality of test items. A review of the subject area has been conducted, and two methods for improving quality have been implemented: similar question retrieval and distractor quality assessment. The former involves testing five transformer-based models to generate text embeddings and six clustering algorithms. The latter uses the same transformer models in combination with three classification algorithms. Experimental results demonstrated high effectiveness of the proposed approaches in solving both tasks.

Signature methods are a powerful tool for use in time series analysis, transforming them into a form that is suitable for machine learning tasks. The article examines the fundamental concepts of path signatures, their properties, and their geometric interpretation, as well as computational methods for various types of time series. Examples of signature method applications in different fields, including finance, medicine, and education, are presented, and their advantages over traditional approaches are highlighted. Special attention is given to synthetic data generation based on signatures, which is particularly relevant when working with limited datasets. The experimental results on generating and predicting student digital learning trajectories demonstrate the effectiveness of signature methods for machine learning applications in time series analysis and forecasting.

The metadata of scientific publications are used to build catalogs, determine their citation, and perform other tasks. The automation of metadata extraction from PDF files provides a means of speeding up the execution of the designated tasks, while the possibility of further use of the obtained data depends on the quality of the extraction. Existing software solutions were analyzed, and three were selected: GROBID, CERMINE, and ScientificPdfParser. A procedure for comparing software solutions to recognize texts of scientific publications according to the quality of the metadata extraction is proposed. Based on this procedure, an experiment was conducted to extract four types of metadata (title, abstract, publication date, and author names). To compare software solutions, a dataset of 112 457 publications divided into 23 subject areas formed on the basis of Semantic Scholar data was used. An example of choosing an effective software solution for metadata extraction under the conditions of specified priorities for subject areas and types of metadata using a weighted sum is given. It was determined that for the given example, CERMINE shows efficiency 10.5% higher than GROBID and 9.6% higher than ScientificPdfParser.

The modern information space contains significant amounts of data, but they are often poorly structured, difficult to find, and not always correct. This creates additional difficulties for research, so digital spaces of scientific knowledge are currently being formed, in particular, based on knowledge graphs. To ensure information quality, such graphs are often manually filled with data, which is time-consuming. Therefore, the creation of a tool that provides the ability to automate process of filling a graph with data, as well as ensuring data quality, will simplify and speed up the process of forming digital spaces of scientific knowledge. Methods of automating the filling of the graph with data are proposed, including parallel control of their integrity. Based on the proposed methods, a software module has been developed, and the mechanisms of its functioning and its architecture are described.

The main methods of numerical simulation (the finite difference method, the finite element method, the Monte Carlo method, and the Runge–Kutta method) are presented. The main parameters used to optimize numerical modeling algorithms in terms of code execution time and efficient use of processor resources are considered. The main disadvantages of multithreading related to data synchronization, deadlocks, and race conditions and methods for eliminating them based on the use of mutexes and atomic operations are analyzed, using the Monte Carlo method as an example.

A developed algorithm for the visual perception system for game agents, implemented in the Unity game engine, is presented. The proposed method is based on the comparison of images from two cameras, taking complex visual effects (lighting, shadows, camouflage) into account, supplemented with line-of-sight verification, taking into account object speeds and the mechanics of gradual detection. Testing of the system shows a significant increase in realistic detection relative to traditional methods, while maintaining performance within a small additional processor load. The algorithm was optimized using Unity Job System and dynamic camera activation. The scientific literature on similar solutions has also been analyzed and their strengths and weaknesses have been identified. The results can be applied in video game development to create realistic behavior of non-player characters, especially in games with stealth elements.

This paper proposes a method for estimating the fluctuation characteristics of distributed objects based on the fluctuation analysis and the assumption that the optical flow estimate is equivalent to random-walk increments. The reliability and applicability of the proposed method are evaluated in two computational experiments. The first experiment analyses the Brownian motion of a compact object. The second evaluates the adequacy of the method for estimating the dynamic characteristics of a spatially distributed fluctuating object. In both experiments, the Hurst exponent is validated using detrended fluctuation analysis (DFA). The results obtained indicate the applicability of the method and the need to improve its robustness.