The output of optical radiation from the source into a fiber optic cable is an important problem in photonic integrated circuits packaging and implementation of hybrid integration technology. Optical radiation output losses are caused by the mismatch of the mode spot of radiation in the optical fiber and in the waveguide as well as by the deviations of optical elements from the optimal mounting position that occur during the assembly of photonic integrated circuits. This work demonstrates the design, fabrication, and investigation of a lensed optical fiber to solve the problem of coordinated output of optical radiation from radiophotonic components into the fiber. A scheme for outputting optical radiation from a semiconductor laser diode included a discrete ball microlens and a collimating lensed optical fiber. The lensed optical fiber with a lens diameter of 250 μm was formed by arc-fusing a segment of FG125LA fiber which was welded to the SMF-28 fiber to form a coreless insert. A model with coreless inserts of different lengths was proposed to determine the optimal geometrical parameters of the lensed optical fiber that provide a collimated beam of radiation at the exit of the lensed optical fiber. The modeling was performed using the Comsol Multiphysics software package. The optical radiation beam type at the output of the formed ball lensed fiber was verified experimentally. The optimal design of the collimating ball lensed fiber was determined. The fabricated experimental sample of collimating ball lensed fiber allowed to realize the optical scheme for optical radiation output and to investigate the efficiency of the scheme. The implemented optical scheme allows to increase the efficiency of optical power transmission from the source to the fiber by a factor of two compared to the butt coupling. In addition, the developed optical radiation output scheme provides the range of acceptable deviation of its elements from the optimal position by at least 12.8 μm. This value is twice as much as the range of permissible deviation of the elements for the butt connection of the radiation source with the optical fiber. The presented scheme of optical radiation output is recommended to be used with an integral beam expander to increase the efficiency of optical power transmission.

The results of studies of fossil resins of different geological ages and geographic origins are presented. A new method of spectral analysis for differentiation of fossil resins by age from the Triassic period to Modernity is developed. Raman spectra of fossil resin of the late Triassic period are obtained for the first time. The possibility of using the method for differentiation of resin from the late Triassic period to the Moderiny is shown. The studies were carried out using Raman spectroscopy. The spectra were obtained using a Renishaw Virsa spectrometer (UK) with an excitation wavelength of 785 nm and a Raport 1064 portable Raman spectrometer (Russia) with an excitation wavelength of 1064 nm. The range of the studied spectra was 400–3200 cm–1. 27 samples of fossil resins from Eurasia, Africa, America, and Australia were studied. Based on the results of studies of fossil resin samples from different geographic locations and ages, differences were found in the values of the ratio of vibrational modes of the valence skeletal vibrations (ν(C=C)) and the deformation vibration of the CH2 bond (σ(CH2)) in the wavenumber ranges of 1650–1600 cm–1 and 1440–1460 cm–1 in fossil resins aged from late Triassic to Modernity. It was found that with increasing age of amber, the degree of their polymerization decreases. For amber of age Triassic, the absence of ν(CH2,CH3) signals in the high-frequency region for the labdanum skeleton of the resin structure was shown, which indicates an extremely low degree of polymerization of it structure. The obtained results potentially allow using the Raman scattering method for additional differentiation of the age of fossil resins, in case of limitation of the application of the radiocarbon analysis method by age (40,000 years). The advantage of the proposed method is the possibility of rapid, with minimal sample preparation, determination of the age of fossil resin of ages Triassic, Cretaceous, Modernity. At the same time, the accuracy of differentiation of the ages of fossil resins in the age range from Cretaceous to Oligocene to Middle Miocene still remains low, which requires additional research. Also, at the current stage of development, the method does not take into account the influence of environmental conditions: climate, fossilization conditions under which the oleoresin was transformed into resin.

The results of geometry optimization of the two-dimensional photonic crystal waveguide for minimization of optical losses and stabilization of waveguide modes are presented. The main factor (not including absorption) is the appearance of Bragg reflection. Bragg reflection can be decreased by the decrease of the regions of overlaps with high refractive index in photonic crystal. For this purpose, the holes in photonic crystal can be fabricated not as the whole holes but as the parts of the holes. For waveguide modes stabilization the varying of waveguide width was performed. Computer simulation was performed in Comsol Multiphisics 5.5. Energy zone structure of photonic crystal surrounding waveguide was computed by Bloch functions method. In modeling, the free-triangle grid with quality “extremely fine” was used. The frequency near which eigenvalues were looked for has range of 190–200 THz. For the solving of the problems procedures ARPACK FORTRAN were used which work on base of iteration of Arnoldi (IRAM). Modeling have shown that the used geometrical optimization makes possible to decrease the Bragg reflection by 1.75 times. It was established that the losses of photon crystal waveguide in this case do not exceed 0.4 dB/cm. It was shown that the the decrease in the photonic crystal holes diameter with a constant period of the photonic crystal lattice leads to a decrease in the width of the photonic forbidden bandgap. It was shown also that in the waveguide with optimum width the modulation of waveguide mode is maintained but its amplitude decreases significantly. The obtained results can be used in the development of integrated-optical devices for telecommunications and sensorics with low optical losses.

The results of the investigation and develop of the suppressing additive noise methods in fiber-optic interferometric sensors, in which optical scheme is based on a two-beam Michelson interferometer, are presented. The proposed solutions are designed to reduce the level of additive noise and to increase the signal-to-noise ratio in the fiber-optic interferometric sensor phase signal. The first proposed method assumes the insertion of an additional photodetector into the optical scheme of the sensor, which allows simultaneous registration of the same interference signal twice. In the optical scheme of the second proposed method, an additional fiber delay line is included, which allows to interrogate the same sensor twice. Mathematical analysis based on the coherent averaging of signals is applied for the suppressing of the additive noises. Coherent averaging of signals allows, without changing the useful component of the phase signal, to reduce the additive noise level by the value which is proportional to the square root of the averaged signals number. The experimental study of the proposed suppressing additive noise methods presented a decrease of the noise level by factor of 1.4 and an increase of the signal-to-noise ratio on an average by 2.87 dB in the frequency range from 250 to 2250 Hz in the phase signal of the fiber-optic interferometric sensor. The proposed methods of the suppressing additive noise can be used to increase the signal-to-noise ratio in fiber-optic measuring systems based on the interferometric sensors arrays, such as fiber-optic towed and bottom seismic streamers, perimeter security systems, fiber-optic navigation systems and complexes.

A method for compensating for the constant noise component of reflectograms for a coherent optical backscatter reflectometer in the time domain is proposed. The presented solution ensures correct visualization of reflectograms, improves their readability and allows recording weak reflected signals from a fiber communication line. The experimental testing of the method was carried out on a working sample of a coherent optical reflectometer. To compensate for the constant component of reflectogram noise, it is proposed to record the noise signal from the input path of the reflectometer immediately before sending the optical probing signal to the communication line. The data obtained in this way do not contain a useful signal from the connection line. In this case, the data can be used to determine the constant component of the noise level by calculating its root mean square value. Compensation for the constant noise level is performed by subtracting the constant component of the noise from the data of the entire reflectogram. The described method was tested on a working sample of a coherent optical reflectometer developed at the Light-Guided Photonics Research Center of ITMO University. The technique was tested on two test fiber lines: a 200-km-long optical fiber and a 300-km-long line with three optical amplifiers. It was shown that the application of the technique allows to significantly expand the dynamic range of signals presented on reflectograms by more than 10 dB and to increase the distinguishability of weak signals at the noise level of the device. The practical significance of the work lies in the possibility of compensating for the constant component of noise in the reflectogram of a fiber-optic communication line with optical amplifiers without calibrations and preliminary settings of the coherent optical reflectometer. 

Weight sensors are widely used in the freight transportation industry. In the systems for weighing vehicles while moving, ceramic, polymer, quartz piezoelectric sensors, load or hydraulic cells, strain gauges are used as sensitive elements. However, most electric sensors are susceptible to electromagnetic interference. Currently, fiber-optic sensors are most actively developed and put into operation due to their relatively low cost, small weight and size parameters, high measurement accuracy and complete passivity to electromagnetic disturbances. Fiber-optic sensors are most often implemented using fiber Bragg gratings, taking into account the convenience of their multiplexing. Mechanical deformations lead to a shift in the wavelength of the Bragg resonance of the grating. At the same time, the problem of using fiber gratings is associated with their sensitivity to temperature. To achieve high accuracy in measuring the deformation value, and accordingly, the weight characteristics of the object, it is necessary to eliminate or compensate for the effect of the sensor temperature on its readings. Most modern studies describe sensors that either operate in laboratory conditions or involve the use of an additional sensor that complicates the circuit for measuring temperature. The paper proposes a method for solving the problem of cross-sensitivity of a fiber diffraction structure to temperature and deformation. The method is based on the use of a pair of closely spaced gratings in the sensitive element of the sensor. One of the gratings has a constant period along the length, and the other has a variable one. The design of the sensitive element ensures the transfer of mechanical load only to the fiber grating with a constant period, and the temperature change equally affects both diffraction structures. A design solution for the sensitive element is proposed that allows for temperature effects compensation without using additional elements. A mathematical model of temperature effects is presented, allowing estimating the dependence of the temperature gradient on time for different thicknesses of the sensitive element. Modeling has showed that for a sample 0.95 cm thick, the temperature gradient inside the substrate is insignificant. With a sharp change in temperature, the equalization of the temperature field within the substrate at a level of 90 % occurs in no more than 2.5 s. The mechanical load on the sensitive element can pass relative to the fiber grating at different angles in connection with which the value of the shift of the central wavelength of the Bragg resonance was studied in detail depending on the point of application and direction of the load. The proposed technology may be of interest in the development and operation of automatic weight and size control systems with temperature compensation without the use of additional sensors. The proposed system is easy to operate and it has a low cost.

The article is devoted to the study and systematic generalization of the existing experience in the field of determination and control of geometric parameters of various objects using optical methods. When searching for literary sources on the work subject, open international bibliographic databases and search engines were used. Scientific articles devoted to the description of hardware and software for contactless geometric measurements and/or restoration of the threedimensional surface shape of material objects constructed on the basis of optical methods as well as examples of their application to solve practical problems were selected for consideration. The selection criterion for the works under consideration corresponded to the set of keywords and publication in highly rated domestic and foreign publications no older than 2010. A systematic classification of optical methods and hardware and software for contactless geometric measurements and restoration of the three-dimensional surface shape of objects described in peer-reviewed scientific publications is proposed, a comparative qualitative assessment is performed. The most effective methods for solving individual practical problems are identified. The main limitations of the considered methods and means are indicated. The main trends in the development of the considered methods associated with miniaturization and development of electronic component manufacturing technologies, increased sensitivity, spatial and temporal resolution of detecting elements, expanded range and functionality of radiation sources, and the development of automated data processing capabilities are highlighted. The article is a systematic review that can be used to select an optical method that is optimal for solving practical problems in such areas as non-destructive testing and minimally invasive diagnostics, navigation of robotic systems, and creation of digital copies of material objects. In addition, the presented article can be useful for students of specialized specialties of technical educational institutions to familiarize themselves with the current crosssection of modern methodological and hardware-software tools.

This work demonstrates for the first time the selectivity of silver molecular clusters luminescence in silicate glass formed by the ion exchange method from a salt melt containing 0.1 mol.% silver nitrate (AgNO3) and 99.9 mol.% sodium nitrate (NaNO3). Commercial silicate microscope slides of the following system were used: SiO2-Na2O-K2O-CaO-MgO-Al2O3 with Fe2O3-SO3 impurities. Molecular clusters were obtained by low-temperature ion exchange in a melt of 0.1 % AgNO3/99.9 % NaNO3 mol.% for 10 and 15 min at 320 °C. The luminescent properties of silver molecular clusters in the ion-exchange layer of microscope slides were studied. Bands of silver clusters of different sizes (Ag1-5) were found in the luminescence spectra. In this case, clusters Ag1-3 are excited by shorter wavelengths, and clusters Ag4-5 only by far ultraviolet and visible radiation up to 500 nm. In the process of ion exchange lasting up to 10 minutes, the appearance of luminescence selectivity was revealed, which occurs due to the presence of a low concentration of silver clusters of different sizes Ag1-5 in the ion-exchange layer. The obtained results can be used in the development of a photosensitive element for a selective ultraviolet radiation detector.

This report presents the results of an experiment to obtain and characterization of quasi-bulk monocrystal epitaxial ε-Ga2O3/GaN heterostructures which contain V-defects at the interface. The significance of this work is due to an active search for ways to develop device-heterostructures for deep-ultraviolet optoelectronics and for HEMT transistors in high-frequency and high-power electronics. One such solution is epitaxial growth of a thick ε-Ga2O3 layer on templates with GaN epitaxial layer to form ε-Ga2O3/GaN heterostructure. The ε-Ga2O3 is characterized by a wide band gap and high spontaneous polarization. The Ga2O3 layer was grown by chloride-hydride vapor-phase epitaxy on pre-prepared GaN/AlN/3C-SiC/Si structures. As the reactor was cooled to room temperature, the Ga2O3/GaN heterostructure (with an AlN buffer layer) spontaneously detached from the SiC/Si substrate due to the balance of the layer thermal expansion coefficients. The surface morphology of the gallium oxide layer and the cross-section of the interface in the Ga2O3/GaN heterostructures were studied using scanning electron microscopy with TESCAN MIRA 3. The phase composition and crystal quality of the Ga2O3/GaN heterostructures were studied by X-ray diffraction on the DRON-8 diffractometer of JSC Bourevestnik. Transmission spectra of the heterostructures were obtained by spectrophotometry using an integrating sphere. This paper demonstrates the epitaxial growth of Ga2O3/GaN heterostructures through chloride-hydride vapor phase epitaxy and the possibility of detaching epitaxial layers from the template. Scanning electron microscopy studies of cross-section have shown that the Ga2O3/GaN heterostructures contain a quasi-bulk gallium oxide with a thickness of about 100 microns. Additionally, it was also shown that the heterostructures contained V-defects on the GaN growth surface, with the Ga2O3 layer grow over V-defects. X-ray diffraction study revealed that the formed heterostructure contains ε-Ga2O3 monocrystal layer, and the fell width at half maximum intensity of the diffraction reflection curve 0,0,10 is 1.8 degrees, which indicates satisfactory quality for thick epitaxial layer. The results of the study openup the possibility to develop templates and substrates for epitaxial growth of wide-bandgap semiconductor materials. Optimization of the growth modes and geometry of such heterostructures with a thick ε-Ga2O3 layer is a promising area for further research.

Evidence-based technologies for product quality have a positive impact on a wide range of social and economic processes. One of the immanent problems of implementing such technologies is determined by the contradiction between the need to ensure open access to information about the stages of the technological process and the confidentiality of some of such data. The use of strict cryptographic procedures to resolve this contradiction is often impossible due to resource constraints, in particular, the lack of continuous telecommunications between the parties involved. The results obtained are aimed at ensuring the feasibility of product quality traceability systems under resource constraints. They are based on a new architectural solution and the integration of classical methods and tools to ensure information security. The paper proposes a three-level model of a product quality traceability system with controlled quality degradation and scenarios for ensuring the continuity of its safe operation. The basic concepts of the proposed solution are: separation of stored data into publicly available and confidential; procedures for “deferred” trusted access in conditions where direct communication with one of the data owners is impossible; data separation into shards — functionally or geographically localized data warehouses; immanent properties of distributed registry systems in terms of ensuring data integrity and availability, non-repudiation of operations. The paper introduces typical scenarios for the use of a hierarchical product quality tracking system, sets and proposes a solution to the problem of ensuring information security of their implementation. The approach to reducing the level of information security of specific implementations in conditions of resource constraints is justified by taking into account the specifics of the functioning of application systems. The information security of the new results is confirmed by computer modeling using specialized protocol security analysis tools. Unlike well-known models focused on the use of stable communication channels, centralized data models, strict cryptographic algorithms and significant computing resources that do not involve accessing data in the absence of communication with their owner, the proposed solution provides authenticated controlled access to the requested confidential data and in the absence of communication with their owner. An immanent disadvantage of the implementation of the discussed scenarios is a certain decrease in the level of information security associated with delegating trust to a third party as well as simplifying the compromise of distributed registry shard nodes.

In biology, information about interactions between the proteins or genes under study can be represented as a biological graph. A connected subgraph, whose vertices perform a common biological function, is called an active module. The Markov Chain Monte Carlo (MCMC) algorithm is an effective method for identifying active modules in biological graphs. In the context of protein-protein interactions, accurately identifying the active module allows for determining which protein function disruption leads to certain changes (e.g., diseases) in a biological system (cell/organism). This study demonstrates that applying MCMC in combination with models (that take graph topology into account) provides higher accuracy in identifying the active module. This study independently utilizes a protein-protein interaction graph (InWebIM) and the GeneMANIA functional association network for training the model and comparing it with the known MCMC-based method. To search for the active module, a combination of MCMC and a machine learning method, gradient boosting (xgboost), was employed. The combined use of the MCMC-based method and gradient boosting improves the accuracy of active module identification compared to the MCMC-based method alone on simulated data. Improving the accuracy of active module identification is crucial for studying the biological mechanisms of diseases and discovering individual proteins functionally associated with the development of diseases.

This study presents an algorithm for the problem of detecting defects on hard surfaces when trained with zero or a small number of examples, addressing the challenge of limited data availability. The existing defect detection methodology using machine vision is enhanced. A hybrid approach is proposed, combining the strengths of the SSD detector and Siamese Neural Networks (SNN). The SSD detector extracts feature vector representations from images, while the SNNs are used to construct the feature space. The new approach demonstrates high accuracy in detecting both known and previously unseen defects in the training dataset. Based on testing across seven different datasets, the model showed good performance in scenarios with a limited number of training examples. A comparative analysis with existing models highlights the high performance of the proposed approach and its potential as an innovative and effective solution for the universal detection of defects on hard surfaces.

The increasing volume of user-generated content on social media platforms necessitates effective tools for understanding public sentiment. This study presents an approach to sentiment analysis of Arabic tweets using supervised machine learning techniques. We explored the performance of three popular algorithms — Support Vector Machines (SVM), Naive Bayes (NB), and Logistic Regression (LR) — on two distinct corpora: the Arabic Sentiment Text Corpus (ASTC) and a dataset of Arabic tweets. Our methodology involved four tests assessing the impact of corpus characteristics, preprocessing techniques, weighting methods, and the use of N-grams on classification accuracy. The first test established that the choice of corpus significantly influences model performance, with SVM showing superior accuracy on the structured ASTC, while NB excelled with the informal Arabic tweets. In the second test, preprocessing steps, including the removal of punctuation and stop-words, led to a noticeable improvement in classification accuracy for the Arabic tweets but had minimal or even negative effects on the ASTC. The third test indicated that incorporating N-grams yielded modest improvements for NB and LR in more structured texts, while its impact on tweets was negligible. Finally, the fourth test compared different weighting techniques, revealing that SVM benefitted from the Term Frequency-Inverse Document Frequency weighting method, while NB performance remained stable regardless of the weighting approach. These findings underscore the importance of tailoring preprocessing and feature extraction strategies to the specific characteristics of the dataset, ultimately enhancing the accuracy of sentiment analysis in Arabic language contexts

The main function of large language models is to simulate the behavior of native speakers in the most correct way. Hence, it is necessary to have assessment datasets to track progress in solving this problem as well as regularly compare competing models with each other. There are some datasets of this type, the so-called linguistic acceptability corpora. The hypothesis that underlies the creation of these corpora assumes that large language models, like native speakers, should be able to distinguish correct, grammatical sentences from the ungrammatical ones that violate the grammar of the target language. The paper presents the parametric corpus for Russian, RuParam. Our corpus contains 9.5 thousand minimal pairs of sentences that differ in grammaticality — each correct sentence corresponds to a minimally different erroneous one. The source of ungrammaticality in each pair is supplied with the linguistic markup provided by experts. RuParam consists of two parts: the first part uses a totally new data source for the task of testing large language models — lexical and grammatical tests on Russian as a foreign language. The second part consists of (modified and tagged) examples from real texts that represent grammatical phenomena, not included in the RFL teaching program due to their complexity. As have shown our experiments with different Large Language Models, the highest results are achieved by those models that have been trained on Russian most carefully at all stages, from data preparation and tokenization to writing instructions and reinforcement learning (these are first of all YandexGPT and GigaChat). Multilingual models, which usually receive little or no emphasis on Russian, showed significantly lower results. Still, even the best models results are far from the assessors who completed the task with almost 100 % accuracy. The models ranking obtained during the experiment shows that our corpus reflects actual degree of proficiency in Russian. The resulting rating can be helpful when choosing a model for natural language processing task requiring grammar knowledge: for example, building morphological and syntactic parsers. Thereafter, the proposed corpus can be used to test your own models.

Generative artificial intelligence systems have a significant impact on tasks related to natural language processing: machine translation, sentiment analysis, text generation, and summarisation, etc. The aim of the presented work was to determine the features of automatically generated academic texts in comparison with texts created by authors, and to evaluate the capabilities of different methods in relation to the task of their classification. The paper analyses two types of abstracts: collected from academic journals on computational linguistics and Germanic studies and generated from the titles of the corresponding articles using ChatGPT-4o mini. The total amount of data was 60 items. The choice of article topics is due to the fact that the texts belong to the same subject area but differ in their structure. The first group which contains original texts on computational linguistics, is similar to the abstracts of academic articles on computer science, and contains a large amount of English terminology. The second group contains texts on Germanic studies and is more descriptive-narrative in their nature. We analyzed the differences between the two types of abstracts and classified them into two categories with the help of experts, three detector systems to determine the involvement of artificial intelligence in the creation of texts (Smodin, ZeroGPT and GPTZero), as well as the ChatGPT system itself. The analysis showed that the generated texts are characterized by a clear formal structure and adherence to the rules of academic text construction in accordance with IMRAD (Introduction, Methods, Results and Discussion). They are superficial in content and they do not always follow the scientific style; there are repetitions of constructions and paraphrasing of article titles, which is not found in the abstracts written by the authors without artificial intelligence. Automatically generated abstracts need not only further editing (because in some cases lexical and syntactic coherence is broken and ambiguity is present), but also verification of the facts and terms mentioned. Among the detector systems, the highest scores in Precision, Accuracy and F1-score are achieved by Smodin tools, while the best results in Recall are achieved by ZeroGPT. The lowest results in abstract evaluation when compared with other tools were achieved by the ChatGPT system itself. Expert-assisted classification showed the highest results in the case of Germanic abstracts. The results may be useful for researchers when working with academic texts on linguistics as well as for further finetuning of neural network models.

Kubernetes has become a cornerstone of modern software development enabling scalable and efficient deployment of microservices. However, this scalability comes with significant security challenges, particularly in detecting specific attack types within dynamic and ephemeral environments. This study presents a focused application of Machine Learning (ML) techniques to enhance security in Kubernetes by detecting Denial of Service (DoS) attacks and differentiating between DoS attacks, resource overload caused by attacks, and natural resource overloads. We developed a custom monitoring agent that collects telemetry data from various sources, including real-world workloads, actual attack scenarios, simulated hacking attempts, and induced overloading on containers and pods, ensuring comprehensive coverage. The dataset comprising these diverse sources was meticulously labeled and preprocessed, including normalization and temporal analysis. We employed and evaluated various ML classifiers, with Random Forest and AdaBoost emerging as the top performers, achieving F1 macro scores of 0.9990 ± 0.0006 and 0.9990 ± 0.0003, respectively. The novelty of our approach lies in its ability to accurately distinguish between different types of resource overloads and provide robust detection of DoS attacks within Kubernetes environments. These models demonstrated a high degree of accuracy in detecting security incidents, significantly reducing false positives and false negatives. Our findings highlight the potential of ML models to provide a targeted, proactive security framework for Kubernetes, offering robust protection against specific attack vectors while maintaining system reliability.

The exponential growth of digital information necessitates the development of robust text retrieval methods since most of the methods are domain or task-specific which limits their implementation. In this case multi-task learning is a promising alternative as it helps a model to have more meaningful embeddings; however such cases require usage of task separation methods. Many studies explore multi-task learning to improve generalization but tend to focus on large models. However, in real-world, speech analytics tasks that require searching through hundreds of millions of vectors in real-time, smaller models become more appropriate. This paper presents a novel approach to enhance the robustness of multi-task text retrieval models through the use of prompts. We use contrastive learning to train encoder models both in single-task and multi-task configurations and compare their performances as well as analyze the efficiency of different prompt usage strategies including hard prompts represented by explicit natural language instructions and soft prompts of varying lengths represented by model special tokens. Experiments are conducted by applying prompts to both the query and candidate document as well as to queries only keeping the candidate without prompts to reuse pre-encoded candidates in multi-task retrieval without significant quality loss. The obtained results are compared using R@1, R@5, and MRR metrics which are most applicable for evaluating in-domain and out-of-domain search. Single-task models show better performance on in-domain training data, while multi-task models demonstrate superior performance on out-of-domain data highlighting their increased robustness to domain shifts. Applying prompts to both elements–query and document–yields better performance than applying them solely to the query. Soft prompts are found to be preferable to hard as they better adapt the model to different domains. The findings of this study can be useful for improving text retrieval models, especially in scenarios involving multi-task systems where high adaptability and performance on new data are required. Trainable prompts could be an effective tool for enhancing the flexibility of models in various applications, such as information retrieval and question-answering systems.

Today, Community Question Answering (CQA) forums such as Stack Overflow are becoming an irreplaceable tool for software developers, providing fast and efficient solution search and prompt community response. Although modern Pretrained Language Models (PLMs), also trained including on data from such forums, have the potential to automate answering of domain-specific questions, they often show significant limitations in complex domains such as programming due to the heterogeneity of the domain and variety in contexts of the questions being asked. In our study, we propose an approach to solving this problem based on structuring data in a complex domain. The first stage includes decomposing available forum data with the selection of thematic subsets. Next, for individual topics, models are finetuned using Reinforcement Learning with Human Feedback (RLHF) using the voting available in the forum data. Finally, to manage the ensemble of finetuned models, question classification is used with subsequent selection of the appropriate model. Experimental studies were conducted on a subset of Python-related questions from Stack Overflow, using the Llama 7B model as the base language model. Experimental studies were conducted on a subset of Python-related questions from Stack Overflow forum using the Llama 7B model as a base PLM. The results of the studies showed that by classifying questions we can improve the model performance up to +22.5 % on the Rouge metric. Moreover, the inclusion of RLHF resulted in an additional improvement of up to +11.2 %. To validate these results, we performed human evaluation of the generated responses, which confirmed the effectiveness of our approach. This study shows that by structuring community data and processing implicit feedback, we can significantly improve PLM performance in CQA tasks in complex domains characterized by high heterogeneity, such as software development.

Automata-based programming is a programming paradigm that has been successfully used in the development of reactive systems, distributed control systems, and various mission-critical applications where the ability to verify the compliance of a real system with its model given in the form of specifications is critical. The traditional testing of such systems can be difficult; thus, more advanced verification tools are required to increase confidence in the reliability of real systems. The previously proposed language for the specification of the Cooperative Interaction of Automata-based Objects (CIAO) was successfully used to develop several different reactive systems as a result of which a number of shortcomings were identified and eliminated in the new version of CIAO v.3. This new version of the language was developed for the automatic verification of automata-based programs according to the formal specifications of a certain class of real-time systems. Three innovations distinguish CIAO v.3 from previous versions. First, an explicit distinction between automata classes and automaton objects as instances of these classes. Second, we specify the binding of automaton objects through interfaces using a connection scheme. Third, we describe the semantics of the behavior of a system of interacting automaton objects using a semantic graph. This paper presents the main concepts of the new language version including the abstract syntax, operational semantics, and metamodel. The third version of the CIAO language naturally includes almost all the advantages of object-oriented programming into the paradigm of automata programming. The connection of automaton objects through the corresponding interfaces is arbitrarily reflected by the connection scheme. A semantic graph describing the semantics of the behavior of the automata-based program is used to implement automatic verification with respect to formal specifications.

The article considers the structures of representation of the graph of inter-agent connections for increasing the efficiency of agent interaction in cooperative competitive games using graph neural networks. A comparative assessment of metrics and adjacency matrices for graphs of connections defined using geometric and semantic metrics of proximity is performed. It is shown that semantic proximity is more effective in constructing a graph of inter-agent connections, and the use of oriented graphs ensures flexible management of information flows. The proposed patterns are important to consider when organizing multi-agent reinforcement learning in a wide range of application areas.

At present, the most common technology for hydrogen production is steam methane reforming. Its key disadvantage is significant emissions of carbon dioxide into the atmosphere due to the presence of natural gas combustion in the air in the reformer furnace. This problem can be solved by switching to oxygen combustion of organic fuel. This paper presents the results of developing a new process flow diagram for a steam methane reforming and a comparative analysis of its energy and environmental characteristics with the closest analogue: steam methane reforming with monoethanolamine cleaning of exhaust gases. To perform a thermodynamic analysis of process flow diagram options using the Aspen Plus software package, mathematical models have been developed that include sequentially solved equations for the processes of oxygen combustion of fuel, steam reforming reaction, steam shift reaction and monoethanolamine absorption reaction at variable pressure. In addition, the modeling took into account the possibility of two side reactions: steam reforming of carbon monoxide and carbon dioxide reforming of methane. The NIST REFPROP database was used to determine the thermodynamic properties of the substances. The thermodynamic analysis showed that for the proposed flow chart of the oxygen-fired methane steam methane reforming, an increase in temperature from 850 to 1050 °C results in a 14.4 % decrease in the mass flow rate of natural gas. At the same time, the thermodynamically optimal temperature in the reformer, equal to 950 °C, provides the possibility of achieving the fuel HUF value of 79.2 %. In turn, the comparison of the energy and environmental characteristics of the two considered steam methane reforming units allowed us to conclude that the proposed flow chart with oxygen-fired fuel has two advantages over the flow chart with CO2 capture by absorption in monoethanolamine: higher energy efficiency (net efficiency is 2.12 % higher) and lower greenhouse gas emissions (carbon dioxide emissions are 14.5 times lower). The proposed process flow diagram, as well as the developed mathematical models, can be used in the development of highly efficient steam methane conversion plants with minimal emissions of harmful substances into the atmosphere. 

The problem of stability assessment of nanoelectronic structures including hybrid transistor-memristor non-volatile memory is considered. The results of the study of processes in nanoelectronic structures using memristors indicate that in addition to the usual parameter drift inherent in semiconductor devices, new unique effects arise in them, in particular, such effects that lead to uncertainty in the evaluation of the state of memristor memory cells. The study of such effects is in its infancy, in part due to the lack of models that allow full investigation of parameter variability and state drift of memristors. In this regard, we propose to use the metaheuristic particle swarm method which allows us to evaluate the stability of hybrid transistor-memristor memory. The methods of topological and parametric analysis of nanoelectronic structures with memristors, the method of interval analysis of similar structures, the method of particle swarm optimization for solving interval algebraic and differential equations are used in this work. A structuralparametric model of a hybrid memristor-based memory device is proposed, taking into account finite increments of their parameters caused by the influence of external and internal factors. An algorithm for estimating the parameters of a hybrid memristor-based memory device using a modified particle swarm optimization method is developed. Interval mathematical models serve as a basis for the development of new principles of organization of ultra-dense nonvolatile memory and create prerequisites for new approaches to the organization of computations in memory. The computational algorithm based on the method of particle swarm optimization allows us to evaluate the performance of hybrid metaloxide-semiconductor structures (MOS structures) with memristors under real operating conditions, resulting in the possibility to expand the scope of application of devices using quantum effects in various technical application

The analysis of formalized conditions for creating universal images falsely classified by computer vision algorithms, called adversarial examples, on YOLO neural network models is presented. The pattern of successful creation of a universal destructive image depending on the generated dataset on which neural networks were trained using the Fast Sign Gradient Method attack is identified and studied. The specified pattern is demonstrated for YOLO8, YOLO9, YOLO10, YOLO11 classifier models trained on the standard COCO dataset.