Informatics and Automation

Adaptive Regression Model Construction Based on the Functional Quality Analysis of the Sequence Segment Processing

Ilya Lebedev — Tue, 01 Apr 2025 00:00:00 +0000

The article considers the problem of constructing an adaptive model aimed at improving the quality indicators of processing information sequences. In data processing techniques that have found application in many application areas, the applied analysis of observation objects is computationally resource-intensive and requires many iterations in case of changes in data properties. The article proposes a technique for selecting segments of an information sequence obtained in different ways, which differs in the use of the quality functional of regression models for processing subsequences. The sequences of observation objects received at the input of the model are divided by various specified segmentation algorithms. Pre-selected regression models are trained on each obtained segment and, depending on the obtained values of the calculated quality functional, the best models in terms of quality indicators are assigned to the segments. This allows us to form an aggregation model for data processing. Based on the experiment on model data and samples, the proposed technique is assessed. The values of the quality indicator MSE and MAE are obtained for different processing algorithms and with a different number of segments. The proposed method makes it possible to increase the MSE and MAE indicators by segmentation and assignment of regression models that have the best indicators on individual segments. The proposed solution is aimed at further improvement of ensemble methods. Its application allows to increase the efficiency of setting up basic algorithms in case of data property transformation and to improve the interpretability of results. The method can be used in developing models and methods for processing information sequences.

Two-Level Optimization of Task Distribution into Batches and Scheduling Their Execution in Pipeline Systems with Limited Buffers

Kirill Krotov — Tue, 01 Apr 2025 00:00:00 +0000

Currently, existing mathematical models and algorithms provide optimization of schedules for the execution of single tasks or fixed task packages on devices of conveyor systems containing buffers of limited sizes. These models and algorithms do not allow searching for optimal solutions for grouping the same type of tasks into packages and by sequence of packages to implement operations with them on devices of conveyor systems. Increasing the efficiency of using the resources of conveyor systems is achieved by optimizing solutions for grouping the same type of tasks into packages and by sequences of packages for performing operations with them. The solution to this problem is carried out in the work by using an approach that implements two-level optimization, which allows you to form a hierarchy of subtasks for finding effective solutions. The involvement of the mentioned approach involves the development of mathematical models of hierarchical games that allow identifying effective solutions of the type under consideration. Two mathematical models of hierarchical games have been constructed, the use of which makes it possible to optimize package compositions at the upper level by the leading player and optimize package execution schedules in pipeline systems at the lower level by the slave player. The method of determining the optimal solutions for each of the players provides for the order of moves set in the game and the exchange of solutions between them during the game. The first mathematical model of the hierarchical game implements the definition of effective solutions when taking into account the downtime of processing devices in the process of implementing operations with packages. The second mathematical model of the game implements the definition of effective solutions, taking into account the total waiting time for buffers to place tasks in them, with which operations on previous devices were completed. To do this, expressions have been formed that allow you to determine buffer downtime while waiting for tasks from packages to be ready for placement based on the time characteristics of the processes of performing operations with them on the devices of the systems under consideration. The algorithm for determining optimal solutions according to the order of operations with packages at the lower level in each of the hierarchical games is based on a developed mathematical model of the processes of implementing actions with packages in these systems and the corresponding modeling algorithm. The implementation of the optimization approach under consideration allowed us to obtain results that showed that the use of buffers can significantly increase the efficiency of the processes of performing operations with packets on the devices of the systems under consideration; increasing the size of intermediate buffers allows us to increase the efficiency of these processes to a greater extent with significant heterogeneities in the values of time parameters characterizing them; using the first model of a hierarchical game allows us to achieve a greater increase in the efficiency of processes in comparison with the second model.

An AudioCodec Based on the Perceptual Equality between the Original and Restored Audio Signals

Ilya Chizhov — Tue, 01 Apr 2025 00:00:00 +0000

A method for lossy audio data compression (AudioCodec) is presented. It allows for improving objective quality of the restored audio signal by 25% at a bitrate of 390 kbps and 55% at a bitrate of 64 kbps compared to the AAC MPEG-4 format. The proposed method of audio data compression is based on an advanced theory of lossy audio data compression (TLAC), which is also introduced in the article. The improvement in the objective quality of the reconstructed audio signal (according to the standardized PEAQ measure) is achieved because the TLAC overcomes issues in modern lossy audio data compression methods related to the use of psychoacoustic principles of human sound perception, including after overcoming the "psychoacoustic compression limit" of the audio signal (i.e. the moment in perceptual coding when the available bit budget is insufficient to encode all spectral components with the accuracy required from a psychoacoustic perspective). This allows for achieving perceptual equality between the original and reconstructed audio signals. As an analysis of the state of the art, solutions for both lossless and lossy audio data compression, as well as those using artificial intelligence, are considered. In all modern lossy audio data compression methods, the procedure for selecting the spectral components to be preserved, as well as the permissible quantization error, is carried out through a series of highly complex procedures collectively referred to as the "psychoacoustic model of the lossy audio compression method". In a strict sense, perceptual equality between the spectra of the original and restored signals has not been proven by any research group and, therefore, cannot be guaranteed by them. Independent experts regularly publish tests demonstrating that modern audio codecs have issues with certain audio signals. The article proposes an AudioCodec based on the perceptual equality between the original and restored audio signals, which is based on the new ideas of the theory of lossy audio compression (TLAC). These ideas guarantee the achievement of perceptual equality between the original and restored audio signals at different bitrates, therefore, the AudioCodec built on its basis is free from the above-mentioned issues and, as a result, significantly outperforms modern AudioCodecs in terms of the objective quality of the restored audio signal, as measured by PEAQ.

Solving Multi-Objective Rational Placement of Load-Bearing Walls Problem via Genetic Algorithm

Vladislav Zinov, Vadim Kartak, Yulia Valiakhmetova — Tue, 01 Apr 2025 00:00:00 +0000

The rational placement of load-bearing walls remains a complex and poorly studied problem, despite the existence of numerous algorithms and models for solving the similar problem of column placement. The main complexity factors are the large number of alternative solutions, the significant time required to calculate deformations for a given wall placement, and the multi-objective nature of the problem. In addition to the nonlinear criterion for estimating deformations, it is necessary to minimize the length of load-bearing walls and the number of their unique lengths. A model for the rational placement of load-bearing walls is proposed, which divides the walls into functional parts with a specific step and considers all the required target criteria. Adjacent wall parts with the same functionality are combined into segments. The combinatorial formulation applied in the model of the problem allows the use of genetic algorithms as a solution tool. Therefore, a new approach to multi-objective genetic algorithm is proposed, containing metrics for calculating population diversity at the phenotype and genotype levels. Modifications of crossover, mutation, and selection operators, considering the segmental structure of the wall's genotype, are presented. A comparative analysis of the developed algorithm with other known multi-objective genetic algorithms showed that the developed algorithm finds, on average, three times more non-dominated solutions, particularly more plans with a lower deformation estimates, despite the twice-longer execution time. The proposed model differs significantly from previous models in terms of handling deformations in slab-support systems, comparing placement plans with each other rather than calculating precise reinforcement estimates, which is often unnecessary at the early stages. The proposed genetic algorithm scheme increases the number of found nondominated solutions without losing their diversity, and can be used to solve other multi-objective problems, taking into account the specified features. The developed algorithm was easily integrated into the CAD-based decision support software and can be used in practice by building designers.

Routing of Autonomous Devices in Three-Dimensional Space

Vladimir Kureychik, Vladislav Danilchenko, Evgeniya Danilchenko — Tue, 01 Apr 2025 00:00:00 +0000

The article addresses the problem of routing autonomous devices in three-dimensional space, which is a relevant task for intelligent control. The three-dimensional space is characterized by a high degree of freedom, complex topology, and dynamic environmental changes, which significantly complicate the task of effective trajectory planning. The development of routing methods that ensure safety, energy efficiency, and computational efficiency is crucial for improving the performance of autonomous systems. The paper presents a comprehensive routing system based on a hybrid approach that combines high-level modeling of the working space with metaheuristic optimization methods. Hierarchical data structures, such as octrees, are used to represent the three-dimensional environment, providing compactness and flexibility for spatial models. These models are transformed into graph structures, allowing the routing problem to be described as an optimization problem on graphs. A modified metaheuristic ant colony optimization algorithm, belonging to the class of swarm optimization methods, is proposed. The algorithm is designed to build safe and energy-efficient routes, as well as to solve problems related to finding the shortest Hamiltonian cycles and dynamically reconfiguring routes in a changing external environment. The paper presents the results of computational experiments, including algorithm testing in three-dimensional space and a comparative analysis with other routing algorithms. The computational experiment confirmed the effectiveness of the developed routing algorithm, including reduced computation time and improved energy efficiency of autonomous devices. The prospects for further research include integrating the proposed system into a wide range of applications for autonomous devices aimed at optimizing control processes and enhancing performance in a dynamically changing external environment. It is worth noting that the developed algorithm can be adapted to solve complex tasks where routing and wind generator placement on a plane are interrelated. The placement problem is directly connected to route construction for servicing these objects, which requires a comprehensive approach for an efficient solution. This will be part of a decision support system designed for the planning and servicing of wind power complexes, ensuring their effective operation and resource management.

Invasive Approach to Verification of Functional and Structural Specifications Implemented in Custom Integrated Circuits

Tue, 01 Apr 2025 00:00:00 +0000

An approach to verification of functional and structural specifications implemented in custom integrated circuits based on invasive research methods is presented. The relevance of this research is determined by the necessity of verification of functional-structural specifications supplied by third-party implementers of hardware implementations of information security algorithms, the difficulty of detecting modifications of these algorithms and undocumented capabilities implemented at the hardware level, and the lack of uniform, universal or standardized methods for solving this problem. The mathematical formulation of the research problem is specified; its essence is to verify the equality of the values of the declared specification parameters and their values restored by the reverse engineering method. The results of the application of the verification technique of functional and structural specifications are presented using examples of its adaptation to the study of hardware-implemented DES and AES encryption algorithms. The restored functional and structural blocks of the algorithms (in particular, the substitution block) were successfully verified.

Building Predictive Smell Models for Virtual Reality Environments

Tue, 01 Apr 2025 00:00:00 +0000

In a sensory-rich environment, human experiences are shaped by the complex interplay of multiple senses. However, digital interactions predominantly engage visual and auditory modalities, leaving other sensory channels, such as olfaction, largely unutilized. Virtual Reality (VR) technology holds significant potential for addressing this limitation by incorporating a wider range of sensory inputs to create more immersive experiences. This study introduces a novel approach for integrating olfactory stimuli into VR environments through the development of predictive odor models, termed SPRF (Sensory Predictive Response Framework). The objective is to enhance the sensory dimension of VR by tailoring scent stimuli to specific content and context with the collection of information about the location of scent sources and their identification through features to serve to reproduce them in the space of the VR environment, thereby enriching user engagement and immersion. Additionally, the research investigates the influence of various scent-related factors on user perception and behavior in VR, aiming to develop predictive models optimized for olfactory integration. Empirical evaluations demonstrate that the SPRF model achieves superior performance, with an accuracy of 98.13%, significantly outperforming conventional models such as Convolutional Neural Networks (CNN, 79.46%), Long Short-Term Memory (LSTM, 80.37%), and Support Vector Machines (SVM, 85.24%). Additionally, SPRF delivers notable improvements in F1-scores (13.05%-21.38%) and accuracy (12.89%-18.67%) compared to these alternatives. These findings highlight the efficacy of SPRF in advancing olfactory integration within VR, offering actionable insights for the design of multisensory digital environments.

Enhanced People Re-identification in CCTV Surveillance Using Deep Learning: A Framework for Real-World Applications

Mossaab Idrissi Alami, Abderrahmane Ez-zahout, Fouzia Omary — Tue, 01 Apr 2025 00:00:00 +0000

People re-identification (ReID) plays a pivotal role in modern surveillance, enabling continuous tracking of individuals across various CCTV cameras and enhancing the effectiveness of public security systems. However, ReID in real-world CCTV footage presents challenges, including changes in camera angles, variations in lighting, partial occlusions, and similar appearances among individuals. In this paper, we propose a robust deep learning framework that leverages convolutional neural networks (CNNs) with a customized triplet loss function to overcome these obstacles and improve re-identification accuracy. The framework is designed to generate unique feature embeddings for individuals, allowing precise differentiation even under complex environmental conditions. To validate our approach, we perform extensive evaluations on benchmark ReID datasets, achieving state-of-the-art results in terms of both accuracy and processing speed. Our model's performance is assessed using key metrics, including Cumulative Matching Characteristic (CMC) and mean Average Precision (mAP), demonstrating its robustness in diverse surveillance scenarios. Compared to existing methods, our approach consistently outperforms in both accuracy and scalability, making it suitable for integration into large-scale CCTV systems. Furthermore, we discuss practical considerations for deploying AI-based ReID models in surveillance infrastructure, including system scalability, real-time capabilities, and privacy concerns. By advancing techniques for re-identifying people, this work not only contributes to the field of intelligent surveillance but also provides a framework for enhancing public safety in real-world applications through automated and reliable tracking capabilities.

Use of Pre-Trained Multilingual Models for Karelian Speech Recognition

Irina Kipyatkova, Ildar Kagirov, Mikhail Dolgushin — Tue, 01 Apr 2025 00:00:00 +0000

This paper presents an experimental study aimed at solving the problem of training speech recognition models under conditions of limited available speech and text data. Current approaches to this issue are discussed in detail, particularly the use of pre-trained multilingual models and data augmentation techniques. As part of this study, multilingual models based on Wav2Vec and Whisper were adapted to the Livvi dialect of the Karelian language, and an investigation into the use of an external language model to enhance recognition accuracy was conducted. The paper also describes a specially collected and prepared speech database and a basic recognition system developed using the Kaldi toolkit. Quantitative test results are provided as well, demonstrating the effectiveness of the chosen methods. For instance, Transformer-based models, particularly Wav2Vec, outperformed the baseline models trained using Kaldi software tools. Fine-tuning the Wav2Vec models reduced the word error rate to 24.73% on the validation set and 25.25% on the test set, while a combination of the Wav2Vec-BERT 2.0-based model with an external language model further reduced errors to 17.12% and 17.72%, respectively. This paper is primarily aimed at specialists in the field of automatic speech recognition for low-resource and Balto-Finnic languages. Additionally, the results of this work can be practically applied in field research involving Karelian text transcription. Future work includes expanding the database to improve model adaptation and enhance performance in real-world scenarios.

Detection of Student Engagement via Transformer-Enhanced Feature Pyramid Networks on Channel-Spatial Attention

A. Naveen, I. Jeena Jacob, Ajay Kumar Mandava — Tue, 01 Apr 2025 00:00:00 +0000

One of the most important aspects of contemporary educational systems is student engagement detection, which involves determining how involved, attentive, and active students are in class activities. For educators, this approach is essential as it provides insights into students' learning experiences, enabling tailored interventions and instructional enhancements. Traditional techniques for evaluating student engagement are often time-consuming and subjective. This study proposes a novel real-time detection framework that leverages Transformer-enhanced Feature Pyramid Networks (FPN) with Channel-Spatial Attention (CSA), referred to as BiusFPN_CSA. The proposed approach automatically analyses student engagement patterns, such as body posture, eye contact, and head position, from visual data streams by integrating cutting-edge deep learning and computer vision techniques. By integrating the attention mechanism of CSA with the hierarchical feature representation capabilities of FPN, the model can accurately detect student engagement levels by capturing contextual and spatial information in the input data. Additionally, by incorporating the Transformer architecture, the model achieves better overall performance by effectively capturing long-range dependencies and semantic relationships within the input sequences. Evaluation using the WACV dataset demonstrates that the proposed model outperforms baseline techniques in terms of accuracy. Specifically, in terms of accuracy, the FPN_CSA_Trans_EH variant of the proposed model outperforms FPN_CSA by 3.28% and 4.98%, respectively. These findings underscore the efficacy of the BiusFPN_CSA framework in real-time student engagement detection, offering educators a valuable tool for enhancing instructional quality, fostering active learning environments, and ultimately improving student outcomes.

Factoring Decision Support System Based on Optimized Quantum Algorithms QMC

Aleksandr Chuvakov, Rodion Boryaev — Tue, 01 Apr 2025 00:00:00 +0000

The continuous growth of financial markets dictates the need for its participants to seek new approaches to financial analysis to gain competitive advantages, including through the use of new approaches in the field of computing. Quantum computing can be used as a tool for obtaining these advantages over competitors. In particular, Monte Carlo modeling, although widely used in financial risk management, requires significant computing resources due to the large number of scenarios required to obtain an accurate result. To optimize this approach, quantum amplitude estimation algorithms are used, which accelerate this process if pre-calculated probability distributions are used to initialize input quantum states. However, in the absence of these distributions in existing approaches to this topic, they are generated numerically using classical computing, which completely negates the advantage of the quantum approach. This article proposes a solution to this problem by using quantum computing, including for the generation of probability distributions. The article discusses the creation of quantum circuits for modeling the evolution of risk factors over time for capital flows, interest rates, and credit risks, and presents the combination of these models with quantum amplitude estimation algorithms as an example of using the obtained algorithms for credit risk management. In conclusion, the article analyzes the possibility of using the obtained circuits in financial analysis.

An Assembled Model of Multilayer Geoinformation Space-Time

Anton Ivaschenko, Oleg Golovnin, Anastasia Golovnina, Evgeniya Dodonova — Tue, 01 Apr 2025 00:00:00 +0000

Modern organizational management technologies involve collecting and processing large amounts of data to calculate the parameters of the functioning of the objects and processes under study. Since the main feature of the collected parameters is their binding to territories, on the one hand, and attribution to time periods, on the other hand, the use of geographic information systems and technologies is required. Despite the development of modern geographic information technologies, the issues of their practical application to support decision-making, taking into account the combined influence of spatial and temporal factors, have not been fully resolved. The article proposes an assembled model of geoinformation multilayer space-time, which is a graph whose vertices are the parameter values ordered by layers with the placement of time marks in the time layer, and the arcs describe the relations between them that are divided into three types: topological, semantic and chronological. Conjugation and ordering of parameters, according to the proposed model, allows you to correctly pose and solve the optimization problem, and, consequently, eliminate the problem of the practical use of accumulated analytics in the processes of supporting management decision-making. The proposed model is used in the digital platform of integral monitoring for the digital transformation of the processes of collecting, analyzing and visualizing utility resource data. The general management task is considered, and a specific example is given for one of the urgent tasks of regional management, i.e. social gasification, in which the optimization of the process of processing applications for connecting residential buildings to the gas supply system within the boundaries of the selected region is carried out. The assembled model of geoinformation multilayer space-time allows formulating universal statements of decision support problems for various geoinformatics applications in logistics, transport resource management, as well as in situational centers for enterprise and regional management, business analytics systems and organizational systems management.