Handwriting can reveal the writer's personality traits, which explains the interest in systems for its automated analysis. Such systems require image collection and preprocessing, feature extraction, and personality classification. Feature extraction is usually done by processing raster images. A few publications deal with vectorized handwriting images, but none use the Potrace algorithm and Bezier curves. We propose a modified Potrace algorithm for computing handwriting line spacing, baseline, and slant. The proposed algorithm is applicable for processing skewed texts and can be used to reduce input data complexity for other feature extraction techniques.
In this work, software is developed for modeling the drying process of hygroscopic materials using cellular automata. The modeling process using cellular automata is described, and the developed transition rules that consider heat and mass transfer not only within the material but also between stacks are provided. The developed software uses a modern multi-tier client-server architecture, including a database for handling large volumes of computational data. Additionally, the results of the modeling, including their validation based on existing experimental data, are presented and analyzed.
In today's construction industry, effective communication is critical to project success. Miscommunication and unclear instructions can lead to delays, cost overruns and safety risks. This paper presents some results of implementing a formalized neural network model for communication analysis in a construction company using advanced text analysis techniques such as Word2Vec. The model is aimed at analyzing the effectiveness of the communication of individual employees by determining the key spheres of their influence on communications. This makes it possible to eliminate or prevent the occurrence of destruction in the process of communication.
This research proposes a new noise-resistant approach for predicting water levels using the capacitive sensor output. The method was tested on data from different sensor configurations. The method performance was compared to a neural network-based (NN) model. General NNs are often used for data linearization and noise compensation in smart sensor systems. However, the domain-aware approach is shown to be better suited for usage on embedded systems in terms of accuracy and memory consumption. The testing on datasets from three different sensor configurations shows 1.3x to 7.7x lower RMSE than the NN models. The proposed model also has fewer parameters, requiring 6x to 9x less memory. This result was validated on a real-time Arm Cortex-M0 embedded device.
The work is devoted to the research of methods of optimizing project management in conditions of limited resources. Emphasis was placed on building a hierarchy of tasks and decision-making processes. Strategies for estimating the minimum project duration and costs are presented. Also, with the help of iterative methods, options for management solutions were developed. After analyzing the constraints and formalizing the tasks, practical solutions are provided to improve the efficiency of the project, ensuring time and financial constraints.
In previous studies, automation tools for constructive-synthesizing modeling of deterministic and stochastic fractal time series were developed with the usage of single-thread and multi-agent approaches. The main goal of this study is to develop an effective system for separate scaling of genetic algorithm operations (crossover and mutation) and model evaluation operations due to significant differences in operations' complexity. The service-oriented approach to the cloud-based system design was used. Each service would be responsible for the corresponding operation and would be scaled independently. To mitigate the load, the sub-population term was introduced and used as the unit of data for the evaluation process. Additionally, the system included the aggregation phase which constructs the new population from sub-parts with the application of the selection operation.
This research is devoted to the neighborhood-based method of optimization of the obtained fast feasible TSP solutions. The experiments were carried out using the well-known problem with the size of 1 million points. The sequential version of the proposed algorithm has been implemented. The efficiency is ensured by solving small sub-problems (taking points that are geometrically close to each other), making problem sizes more manageable and allowing for the tackling of larger problems. The advantage of the algorithm is that its parallel version can be applied to the TSP in 3D space for problems of up to billions of points.
In the dynamic landscape of the software and information technology industry, a plethora of product innovations emerge daily. The insatiable demand for novel business and consumer software solutions continues unabated. Organizations are compelled to navigate the complex interplay between market demands, competitive positioning, and operational efficiency while maintaining robust security protocols. This intricate balance necessitates the development of resilient products, robust go-to-market strategies, and a continuously expanding customer base. Strategic identification of high-potential prospects, deployment of cutting-edge operational tools and technologies, and relentless innovation for market sustainment constitute the primary focus areas for organizational investment. To bolster business operations and enhance customer relationships, tools such as Customer Relationship Management (CRM) systems play a pivotal role in orchestrating the customer lifecycle journey. This paper delineates a comprehensive framework for efficiently architecting and designing CRM platforms tailored for enterprise SaaS businesses. It proposes an adaptable design paradigm that companies can adopt while customizing and configuring their CRM systems, alleviating concerns about technical intricacies and platform constraints, thus enabling a laser focus on business objectives and customer-centricity. Furthermore, the paper elucidates a pragmatic example showcasing the synergistic integration of business communications and backend system operations. The study culminates in proposing best practices for the CRM software development lifecycle, encompassing crucial security and compliance considerations in this era of heightened data privacy awareness.
The application of Artificial Intelligence (AI) in the higher education system creates many opportunities for improving the learning process, research activities, managing and supporting structures. The methods of analysis of the resources, surveys and questionnaires, observation, experimental studies and statistical analysis are used in the research. The proposed technology of application of AI in the higher education system focuses on its impact on the quality of education and organizational processes, it covers the aspects of personalized learning, automation of administrative processes, improving the learning and assessment process, development of distance learning and research. It consists of the six stages: analysis of needs and opportunities, implementation strategy planning, data collection and processing, implementation of algorithms, testing and evaluation of results, continuous improvement. There was conducted and statistically checked the experimental research of the implementation of AI into the higher education system. On the basis of the comparison values before and after the experiment it can be concluded that the application of artificial intelligence in higher education system is effective and had a significant impact on the various aspects of educational process.
In this study, the daily returns of the ten largest-cap cryptocurrencies have been analyzed. It is shown that the returns of cryptocurrencies most often follow a Laplace distribution. Analytical expressions have been obtained for assessing the investment risk of cryptocurrencies using the VaR method. Numerical risk estimates have been calculated using VaR and CVaR methods. Our analysis indicates that the Conditional Value at Risk (CVaR) is 40-45% higher than the Value at Risk (VaR), suggesting that the VaR method underestimates potential losses.
Automation and optimization of recruitment processes are crucial components of company operations. One of the key tasks in recruitment automation is the analysis of resumes, which often have a complex structure, complicating their processing by traditional artificial intelligence methods. This study focuses on the development and evaluation of methods for converting resumes from PDF format to text format for further use in artificial intelligence systems. The paper analyzes the basic problems associated with preserving structural and contextual information during the processing of such documents. The use of the multimodal GPT-4 Vision model for reading and converting text from PDF documents is proposed. A comparison is made between the results obtained using GPT-4 Vision and traditional OCR Tesseract. BLEU, ROUGE-1, and ROUGE-L metrics were used to evaluate accuracy. A total of 157 resumes were collected and analyzed, revealing an increase in discrepancies between texts obtained using GPT-4 Vision and OCR as the volume of text increases.
This study compares the effectiveness of various Generative Adversarial Network architectures, including WGAN and WGAN-GP, in data clustering using the Iris dataset. Performance was evaluated with metrics such as Silhouette Score, Davies-Bouldin Index, Adjusted Rand Index, and Mutual Information. WGANs demonstrated the best clustering performance, while GANs showed promising results but were outperformed by WGAN. Contrary to expectations, WGAN-GP performed the worst. The research underscores the potential of GANs in clustering and highlights the need for further optimization based on specific datasets.
This paper introduces an advanced Top-K hierarchical classification model specifically designed for technical Ukrainian texts within the automotive industry. Addressing the unique challenges of complex terminology, mixed-language usage, and frequent data inconsistencies, the study enhances Natural Language Processing (NLP) techniques through a robust classification framework. The Top-K approach allows the model to generate multiple high-confidence class predictions, improving accuracy and decision-making processes, particularly in scenarios involving ambiguous or unlabeled data. Coupled with advanced data preprocessing methods, including language identification, tokenization, and lemmatization, the model achieves significant improvements in classification accuracy, ranging from 85% to 98% across various datasets. This Top-K hierarchical model also supports real-time applications, making it highly effective for use in automotive diagnostics, repair scheduling, and other industry-specific tasks. The research underscores the value of tailored NLP solutions for specialized domains and highlights the potential for broader application of the Top-K classification approach in technical fields. Future work will explore further enhancements and applications of this model to complex data extraction and classification tasks within the automotive industry and beyond.
The polymorphic RBF network was proposed, which demonstrates the required precision in approximating nonlinear functions, as shown by its success in optimizing helicopter turboshaft engines' performance parameters. It trains quickly with minimal configuration and produces smoother fitting curves than other neural networks, aiding in trend analysis. Results indicate that the diagnostic neural model for engines with power loss shows a sharp decline to minimal values, suggesting a potential malfunction. In contrast, the model for a functioning engine shows smooth parameter variations throughout the observation period.
This paper describes the application of the statistical framework within the TextAttributor 1.0 system for automatic authorship attribution. It is experimentally demonstrated that comparing fifteen stylometric features of short Ukrainian-language texts using Euclidean distance metrics and cluster analysis can serve as a basis for determining authorship.
Refactoring is a process of transforming software code to improve its quality while preserving its functionality. The goal of this research is to develop methods for refactoring SQL queries. Changing the text of queries does not affect the schema of the existing database. The goal of the refactoring methods themselves is to make the query easier to understand and improve its readability. Readability assessments of query texts are based on the Automated Readability Index. Methods aimed at simplifying the structure of queries have shown improvements in the readability index of queries up to 20%.
Identifying anomalies in multivariate time series data for modern industrial production processes is becoming increasingly essential to prevent system failures, reduce downtimes, and enhance safety. The advent of the Internet of Things has enabled the collection of vast amounts of data from industrial machinery, providing a rich source of information for anomaly detection. In this context, we propose a Transformer-based reconstruction approach with additional enhancements, including projection and patching, as well as a transposed Convolutional Neural Network for reconstruction. Our approach significantly outperforms existing non-Transformer-based models on the Tennessee Eastman Process dataset.
With the development of the Internet and e-commerce, the number of online reviews has increased, creating a large array of data for user decision-making. However, traditional recommender systems often do not take into account the emotional preferences of users, and collaborative filtering methods face the problems of data sparsity and cold start. This paper proposes a recommendation algorithm that combines emotional color analysis of online reviews with collaborative filtering to improve recommendation accuracy. The algorithm analyzes the emotions in user feedback, determines their preferences, and uses this information to generate personalized recommendations. Experiments have shown that the proposed approach improves the accuracy, completeness, and coverage of recommendations compared to traditional methods, especially in the category of similar product recommendations.
A multilayer perceptron training method based on the backpropagation algorithm has been developed to enhance the helicopter turboshaft engines' residual life-predicting accuracy by integrating the initial parameter into the output layer. Using the Adam optimizer for adaptive learning rates achieves high accuracy (up to 99.3%) in determining the residual life of compressor turbine blades. Experiments show that this method, with 160 epochs, reaches 99.3% accuracy and reduces losses to 0.5% in solving the residual life estimation task.
The developments proposed in the article are aimed at examining the essence of project-oriented management of the development of a seaport as an element of the transport system and proposing on this basis a software product that will allow managing risks in the port during the implementation of investment development projects, taking into account the use of modern information technologies. The work presents a model of the life cycle of the project for the implementation of a software product for risk management in the port, which allows the distribution of the process of creating a software product into stages with risk identification during such development. The study also touches upon the development of a model for determining risks between the customer and the developer of the software product during the implementation of investment development projects in seaports, which allows for establishing a balanced relationship between the risks of the project stakeholders and proposing ways to eliminate them by identifying the critical area of risk and determining the probability of compliance of the software product with the technical task.
