All issues

This paper investigates the effectiveness of Retrieval-Augmented Generation (RAG) combined with various Large Language Models (LLMs) for document retrieval and information access in corporate information systems. We survey typical use-cases of LLMs in enterprise environments, outline the RAG architecture, and discuss the major challenges that arise when integrating LLMs into a RAG pipeline. A system architecture is proposed that couples a text-vector encoder with an LLM. The encoder builds a vector database that indexes a library of corporate documents. For every user query, relevant contextual fragments are retrieved from this library via the FAISS engine and appended to the prompt given to the LLM. The LLM then generates an answer grounded in the supplied context. The overall structure and workflow of the proposed RAG solution are described in detail. To justify the choice of the generative component, we benchmark a set of widely used LLMs — ChatGPT, GigaChat, YandexGPT, Llama, Mistral, Qwen, and others — when employed as the answer-generation module. Using an expert-annotated test set of queries, we evaluate the accuracy, completeness, linguistic quality, and conciseness of the responses. Model-specific characteristics and average response latencies are analysed; the study highlights the significant influence of available GPU memory on the throughput of local LLM deployments. An overall ranking of the models is derived from an aggregated quality metric. The results confirm that the proposed RAG architecture provides efficient document retrieval and information delivery in corporate environments. Future research directions include richer context augmentation techniques and a transition toward agent-based LLM architectures. The paper concludes with practical recommendations on selecting an optimal RAG–LLM configuration to ensure fast and precise access to enterprise knowledge assets.

The article describes a mathematical model that simulates performance of a hierarchical organization during an early stage of a crisis. A distinguished feature of this stage of crisis is presence of so called early warning signals containing information on the approaching event. Employees are capable of catching the early warnings and of preparing the organization for the crisis based on the signals’ meaning. The efficiency of the preparation depends on both parameters of the organization and parameters of the crisis. The proposed simulation agentbased model is implemented on Java programming language and is used for conducting experiments via Monte- Carlo method. The goal of the experiments is to compare how centralized and decentralized organizational structures perform during sudden and smoldering crises. By centralized organizations we assume structures with high number of hierarchy levels and low number of direct reports of every manager, while decentralized organizations mean structures with low number of hierarchy levels and high number of direct reports of every manager. Sudden crises are distinguished by short early stage and low number of warning signals, while smoldering crises are defined as crises with long lasting early stage and high number of warning signals not necessary containing important information. Efficiency of the organizational performance during early stage of a crisis is measured by two parameters: percentage of early warnings which have been acted upon in order to prepare organization for the crisis, and time spent by top-manager on working with early warnings. As a result, we show that during early stage of smoldering crises centralized organizations process signals more efficiently than decentralized organizations, while decentralized organizations handle early warning signals more efficiently during early stage of sudden crises. However, occupation of top-managers during sudden crises is higher in decentralized organizations and it is higher in centralized organizations during smoldering crises. Thus, neither of the two classes of organizational structures is more efficient by the two parameters simultaneously. Finally, we conduct sensitivity analysis to verify the obtained results.

Considered the application of physics-informed neural networks using multi layer perceptrons to solve Cauchy initial value problems in which the right-hand sides of the equation are continuous monotonically increasing, decreasing or oscillating functions. With the use of the computational experiments the influence of the construction of the approximate neural network solution, neural network structure, optimization algorithm and software implementation means on the learning process and the accuracy of the obtained solution is studied. The analysis of the efficiency of the most frequently used machine learning frameworks in software development with the programming languages Python and C# is carried out. It is shown that the use of C# language allows to reduce the time of neural networks training by 20–40%. The choice of different activation functions affects the learning process and the accuracy of the approximate solution. The most effective functions in the considered problems are sigmoid and hyperbolic tangent. The minimum of the loss function is achieved at the certain number of neurons of the hidden layer of a single-layer neural network for a fixed training time of the neural network model. It’s also mentioned that the complication of the network structure increasing the number of neurons does not improve the training results. At the same time, the size of the grid step between the points of the training sample, providing a minimum of the loss function, is almost the same for the considered Cauchy problems. Training single-layer neural networks, the Adam method and its modifications are the most effective to solve the optimization problems. Additionally, the application of twoand three-layer neural networks is considered. It is shown that in these cases it is reasonable to use the LBFGS algorithm, which, in comparison with the Adam method, in some cases requires much shorter training time achieving the same solution accuracy. The specificity of neural network training for Cauchy problems in which the solution is an oscillating function with monotonically decreasing amplitude is also investigated. For these problems, it is necessary to construct a neural network solution with variable weight coefficient rather than with constant one, which improves the solution in the grid cells located near by the end point of the solution interval.

Using modern object-oriented programming language C# a program for simulation of operation of the conveyor for flexible automated assembly of PC was developed. Class diagram of the simulation model of a flexible automated assembly line for PC assembly in mass production mode is presented. Simulation results analysis is presented.

This paper formulates and solves a practical problem of data recovery regarding the distribution of languages on regional level in context of China. The necessity of this recovery is related to the problem of the determination of the linguistic diversity indices, which, in turn, are used to analyze empirically and to predict sources of social and economic development as well as to indicate potential conflicts at regional level. We use Ethnologue database and China census as the initial data sources. For every language spoken in China, the data contains (a) an estimate of China residents who claim this language to be their mother tongue, and (b) indicators of the presence of such residents in China provinces. For each pair language/province, we aim to estimate the number of the province inhabitants that claim the language to be their mother tongue. This base problem is reduced to solving an undetermined system of algebraic equations. Given additional restriction that Ethnologue database introduces data collected at different time moments because of gaps in Ethnologue language surveys and accompanying data collection expenses, we relate those data to a single time moment, that turns the initial task to an ’ill-posed’ system of algebraic equations with imprecisely determined right hand side. Therefore, we are looking for an approximate solution characterized by a minimal discrepancy of the system. Since some languages are much less distributed than the others, we minimize the weighted discrepancy, introducing weights that are inverse to the right hand side elements of the equations. This definition of discrepancy allows to recover the required variables. More than 92% of the recovered variables are robust to probabilistic modelling procedure for potential errors in initial data.

The paper is devoted to the secondary use of spectrum in telecommunication networks. It is emphasized that one of the solutions to this problem is the use of cognitive radio technologies and dynamic spectrum access for the successful functioning of which a large amount of information is required, including the parameters of base stations and network subscribers. Storage and processing of information should be carried out using a radio environment map, which is a spatio-temporal database of all activity in the network and allows you to determine the frequencies available for use at a given time. The paper presents a two-level model for forming a map of the radio environment of a cellular communication system LTE, in which the local and global levels are highlighted, which is described by the following parameters: a set of frequencies, signal attenuation, signal propagation map, grid step, current time count. The key objects of the model are the base station and the subscriber unit. The main parameters of the base station include: name, identifier, cell coordinates, range number, radiation power, numbers of connected subscriber devices, dedicated resource blocks. For subscriber devices, the following parameters are used: name, identifier, location, current coordinates of the device cell, base station identifier, frequency range, numbers of resource blocks for communication with the station, radiation power, data transmission status, list of numbers of the nearest stations, schedules movement and communication sessions of devices. An algorithm for the implementation of the model is presented, taking into account the scenarios of movement and communication sessions of subscriber devices. A method for calculating a map of the radio environment at a point on a coordinate grid, taking into account losses during the propagation of radio signals from emitting devices, is presented. The software implementation of the model is performed using the MatLab package. The approaches are described that allow to increase the speed of its work. In the simulation, the choice of parameters was carried out taking into account the data of the existing communication systems and the economy of computing resources. The experimental results of the algorithm for the formation of a radio environment map are demonstrated, confirming the correctness of the developed model.

Although human skills are heavily involved in the Requirements Engineering process, in particular, in requirements elicitation, analysis and specification, still methodology and formalism play a determining role in providing clarity and enabling analysis. In this paper, we propose a method for deriving formal specifications, which are applicable to dependable software systems. First, we clarify what the method itself is. Computer science has a proliferation of languages and methods, but the difference between the two is not always clear. This is a conceptual contribution. Furthermore, we propose the idea of Layered Fault Tolerant Specification (LFTS). The principle consists in layering specifications in (at least) two different layers: one for normal behaviors and others (if more than one) for abnormal behaviors. Abnormal behaviors are described in terms of an Error Injector (EI), which represent a model of the expected erroneous interference coming from the environment. This structure has been inspired by the notion of an idealized Fault Tolerant component, but the combination of LFTS and EI using rely guarantee thinking to describe interference is our second contribution. The overall result is the definition of a method for the specification of systems that do not run in isolation but in the real, physical world. We propose an approach that is pragmatic to its target audience: techniques must scale and be usable by non-experts, if they are to make it into an industrial setting. This article is making tentative steps, but the recent trends in Software Engineering such as Microservices, smart and software-defined buildings, M2M micropayments and Devops are relevant fields continue the investigation concerning dependability and rely guarantee thinking.

In this paper we consider the structure of site percolation models on three-dimensional square lattices with various shapes of (1,π)-neighborhood. For these models, are proposed iso- and anisotropic modifications of the invasion percolation algorithm with (1,0)- and (1,π)-neighborhoods. All the above algorithms are special cases of the anisotropic invasion percolation algorithm on the n-dimensional lattice with a (1,π)-neighborhood. This algorithm is the basis for the package SPSL, released under GNU GPL-3 using the free programming language R.

The volume and structure of accumulated credit debt to the banking system depends on many factors, the most important of which is the level of interest rates. The correct assessment of borrowers’ reaction to the changes in the monetary policy allows to develop econometric models, representing the structure of the credit portfolio in the banking system by terms of lending. These models help to calculate indicators characterizing the level of interest rate risk in the whole system. In the study, we carried out the identification of four types of models: discrete linear model based on transfer functions; the state-space model; the classical econometric model ARMAX, and a nonlinear Hammerstein –Wiener model. To describe them, we employed the formal language of automatic control theory; to identify the model, we used the MATLAB software pack-age. The study revealed that the discrete linear state-space model is most suitable for short-term forecasting of both the volume and the structure of credit debt, which in turn allows to predict trends in the structure of accumulated credit debt on the forecasting horizon of 1 year. The model based on the real data has shown a high sensitivity of the structure of credit debt by pay back periods reaction to the changes in the Ñentral Bank monetary policy. Thus, a sharp increase in interest rates in response to external market shocks leads to shortening of credit terms by borrowers, at the same time the overall level of debt rises, primarily due to the increasing revaluation of nominal debt. During the stable falling trend of interest rates, the structure shifts toward long-term debts.

Social media is a crucial indicator of the position of assets in the financial market. The paper describes the rigid solution for the classification problem to determine the influence of social media activity on financial market movements. Reputable crypto traders influencers are selected. Twitter posts packages are used as data. The methods of text, which are characterized by the numerous use of slang words and abbreviations, and preprocessing consist in lemmatization of Stanza and the use of regular expressions. A word is considered as an element of a vector of a data unit in the course of solving the problem of binary classification. The best markup parameters for processing Binance candles are searched for. Methods of feature selection, which is necessary for a precise description of text data and the subsequent process of establishing dependence, are represented by machine learning and statistical analysis. First, the feature selection is used based on the information criterion. This approach is implemented in a random forest model and is relevant for the task of feature selection for splitting nodes in a decision tree. The second one is based on the rigid compilation of a binary vector during a rough check of the presence or absence of a word in the package and counting the sum of the elements of this vector. Then a decision is made depending on the superiority of this sum over the threshold value that is predetermined previously by analyzing the frequency distribution of mentions of the word. The algorithm used to solve the problem was named benchmark and analyzed as a tool. Similar algorithms are often used in automated trading strategies. In the course of the study, observations of the influence of frequently occurring words, which are used as a basis of dimension 2 and 3 in vectorization, are described as well.