All issues

This paper presents an experimental study of the application of automatic annotation of text data in the question – answer format (QA pairs) under conditions of limited computing resources and data protection requirements. Unlike traditional approaches based on rigid rules or the use of external APIs, we propose using small language models with a small number of parameters that can function locally without a GPU on standard CPU systems. Two models were selected for testing — Gemma-3-4b and Qwen-2.5-3b (quantized 4-bit versions) — and a corpus of documents with a clear structure and a formally rigorous style of presentation was used as source material. An automatic annotation system was developed that implements the full cycle of QA dataset generation: automatic division of the source document into logically connected fragments, formation of “question – answer” pairs using the Gemma-3-4b model, preliminary verification of their correctness using Qwen-2.5-3b based on evidence span from the context and expert quality assessment. The results are exported in JSONL format. Performance evaluation covers the entire QA pair generation system, including fragment processing by the local language model, text preprocessing and postprocessing modules. Performance is measured by the time it takes to generate a single QA pair, the total throughput of the system, RAM usage, and CPU load, which allows for an objective assessment of the computational efficiency of the proposed approach when running on a CPU. An experiment on an extended sample of 12 documents showed that automatic annotation demonstrates stable performance when processing different types of documents, while manual annotation is characterized by significantly higher time costs and high variability. Depending on the type of document, the acceleration of annotation compared to the manual process ranges from 8 to 14 times. Quality analysis showed that most of the generated QA pairs have high semantic consistency with the original context, with only a limited proportion of data requiring expert correction or exception. Although full manual validation of the corpus (the “gold standard”) was not performed as part of this work, the combination of automatic evaluation and selective expert review allows us to consider the resulting quality level acceptable for preliminary automated annotation tasks. Overall, the results confirm the practical applicability of small language models for building autonomous and reproducible automatic text annotation systems under limited computational resources and provide a basis for further research in the field of effective training corpus preparation for natural language processing tasks.

This paper presents a non-iterative method for solving an inverse problem for a parabolictype equation with an unknown time-dependent coefficient at the first spatial derivative. The overdetermination condition is specified as a definite integral of the unknown function with a weighting factor over the spatial domain or its subdomain. The study is motivated by the need to identify dynamic parameters in applied problems, particularly in modeling transport processes in biological fluids, where the flow velocity may vary over time. In contrast to conventional iterative methods that require substantial computational effort and careful selection of regularization parameters, an original approach based on solution decomposition is proposed. At each time layer, the solution is represented as a linear combination of solutions to two auxiliary systems with the same matrix and different right-hand sides, followed by the determination of the unknown coefficient from a discrete analogue of the overdetermination condition. This approach eliminates the need for an iterative procedure. In the presence of inexact overdetermination data, the highest reconstruction accuracy is achieved using a quasi-solution. Numerical experiments on test problems demonstrate high accuracy in reconstructing the unknown functions under small perturbations of the overdetermination condition. The results indicate strong potential for applications in medical diagnostics and other fields requiring rapid processing of experimental data.

The increased potential for jamming satellite navigation signals, which provide the highest positioning accuracy for moving objects, makes the development of alternative navigation systems comparable in accuracy but autonomous extremely important. One of the most effective such approaches is the integration of inertial and optical navigation systems (NS), as they are more resistant to artificial interference. Among NS data, one can distinguish systems that use a velocity field calculation method when processing optical flow, allowing the determination of the linear and angular velocity vectors of an object in the absence of terrain maps and reference points. However, a serious drawback of this method is the high computational cost of determining the velocity field (optical flow parameters), which is difficult to implement onboard an object. In this regard, the article considers an approach that allows for constructing a tightly coupled autonomous inertial-optical navigation scheme for objects moving along known (programmed) trajectories. This scheme utilizes a navigation algorithm that is easily implemented in onboard computers due to the discovered possibility of estimating the navigation vector without preliminary calculation of optical flow parameters, as well as by utilizing functional dependencies of navigation variables arising on analytical (orthodromic) trajectories. To illustrate the generality of the solution, cases of a rigidly mounted video camera on the object and its two-degree stabilization are studied. The navigation algorithm, providing a stochastic estimate of the full vector of linear and angular motion parameters based on measurements of an integrated tightly coupled inertial-optical NS, is built on the basis of an extended Kalman filter for correlated noise of the object and the observer. A numerical experiment illustrating the effectiveness of the proposed approach is conducted.

Pattern formation described by the scalar Fisher–Kolmogorov–Petrovsky–Piscounov equation with nonlocal competition loses and convection linear on coordinates is considered numerically. Initial function localized around a point is shown to transform in a function localized around a ring with symmetrically sited local maxima. The ring radius and number of maxima depend on convection.

The solution of problems of heat conductivity by means of a method of continuous asynchronous cellular automats is considered in the article. Coordination of distribution of temperature in a sample at a given time between cellular automat model and the exact analytical solution of the equation of heattransfer is shown that speaks about expedient use of this method of modelling. Dependence between time of one cellular automatic interaction and dimension of a cellular automatic field is received.

We investigated numerically the dynamics of kinks of modified sine-Gordon equation in the model with localized spatial modulation of a periodic potential (or impurity). We considered the case of two identical impurities. We showed the possibility of collective effects of the influence of impurities, which are heavily dependent on the distance between them. We demonstrated the existence of a certain critical value of the distance between impurities, which has two qualitatively different scenarios of the dynamic behavior of kink.

The goal of this work is to generalize second order mathematical models for automotive flow using algorithm for building state equation — the dependency of pressure on traffic density — which is adequate with regard to real world data. The form of state equation, which closes the system of model equations, is obtained from experimental form of fundamental diagram — the dependency of traffic flow intensity on its density, and completely defines all properties of any phenomenological model. The proposed approach was verified using numerical experiments on typical traffic data, obtained from PeMS system (http://pems.dot.ca.gov/), using segment of I-507 highway in California, USA as model system.

The present article sets out the scientific approach of Dmitry Sergeevich Chernavskii to the modelling of economic processes. It recounts the history of works of Dmitry Sergeyevich on the economic front, its milestones and achievements. One of the most important advances in the economic analysis was the prediction by a team of scientists headed by D. S. Chernavskii, the major crises that have occurred in our country over the last 20 years, namely, the default of 1998, the crisis of industrial production in the second half of the 2000s, the 2008 crisis and the ensuing recession. As an example, the dynamic analysis of the global macroeconomic processes shows the model of functioning of the dollar as the world currency. On this particular example shows the possibility of seigniorage due to the issue of the dollar and the calculated “window of opportunity” that allows you to issue dollars as the global currency, without prejudice to its own economy.

A model for the development of a closed society (without external economic relations) in the one-product approach is considered as an example of dynamic analysis of the economy of a separate state. The model is based on the principles of market economy, i.e. the dynamics of prices is determined by the balance of supply and demand. It is shown that in the general case, the state of market equilibrium is not unique. Several steady states with different levels of production and consumption are possible. Effect of addressed emission of money in underproductive state is considered. It is shown that, depending on its size it can lead to the transition to a highly productive condition, and just cause inflation without transition. The relationship of these results with the “Keynesian” and “monetarist” approaches is discussed.

The system of linear hyperbolic kinetic equations with the relaxation term of Bhatnagar–Gross–Krook type for modelling of linear diffusion processes by the lattice Boltzmann method is considered. The coefficients of the equations depend on the discrete velocities from the pattern in velocity space. The system may be considered as an alternative mathematical model of the linear diffusion process. The cases of widely-used patterns on speed variables are considered. The case of parametric coefficients takes into account. By application of the method of Chapman–Enskog asymptotic expansion it is obtained, that the system may be reduced to the linear diffusion equation. The expression of the diffusion coefficient is obtained. As a result of the analysis of this expression, the existence of numerical diffusion in solutions obtained by application of lattice Boltzmann equations is demonstrated. Stability analysis is based on the investigation of wave modes defined by the solutions of hyperbolic system. In the cases of some one-dimensional patterns stability analysis may be realized analytically. In other cases the algorithm of numerical stability investigation is proposed. As a result of the numerical investigation stability of the solutions is shown for a wide range of input parameters. The sufficiency of the positivity of the relaxation parameter for the stability of solutions is demonstrated. The dispersion of the solutions, which is not realized for a linear diffusion equation, is demonstrated analytically and numerically for a wide range of the parameters. But the dispersive wave modes can be damped as an asymptotically stable solutions and the behavior of the solution is similar to the solution of linear diffusion equation. Numerical schemes, obtained from the proposed systems by various discretization techniques may be considered as a tool for computer modelling of diffusion processes, or as a solver for stationary problems and in applications of the splitting lattice Boltzmann method. Obtained results may be used for the comparison of the theoretical properties of the difference schemes of the lattice Boltzmann method for modelling of linear diffusion.

The application of a conservative numerical method of fluxes is examined for studying the vortex structures in the massive, fast-turned compact astrophysical objects, which are in self-gravity conditions. The simulation is accomplished for the objects with different mass and rotational speed. The pictures of the vortex structure of objects are visualized. In the calculations the gas-dynamic model is used, in which gas is accepted perfected and nonviscous. Numerical procedure is based on the finite-difference approximation of the conservation laws of the additive characteristics of medium for the finite volume. The “upwind” approximations of the densities of distribution of mass, components of momentum and total energy are applied. For the simulation of the objects, which possess fast-spin motion, the control of conservation for the component of moment of momentun is carried out during calculation. Evolutionary calculation is carried out on the basis of the parallel algorithms, realized on the computer complex of cluster architecture. Algorithms are based on the standardized system of message transfer Message Passing Interface (MPI). The blocking procedures of exchange and non-blocking procedures of exchange with control of the completion of operation are used. The parallelization on the space in two or three directions is carried out depending on the size of integration area and parameters of computational grid. For each subarea the parallelization based on the physical factors is carried out also: the calculations of gas dynamics part and gravitational forces are realized on the different processors, that allows to raise the efficiency of algorithms. The real possibility of the direct calculation of gravitational forces by means of the summation of interaction between all finite volumes in the integration area is shown. For the finite volume methods this approach seems to more consecutive than the solution of Poisson’s equation for the gravitational potential. Numerical calculations were carried out on the computer complex of cluster architecture with the peak productivity 523 TFlops. In the calculations up to thousand processors was used.