All issues

We present a novel model for the dynamical trap of the stimulus – response type that mimics human control over dynamic systems when the bounded capacity of human cognition is a crucial factor. Our focus lies on scenarios where the subject modulates a control variable in response to a certain stimulus. In this context, the bounded capacity of human cognition manifests in the uncertainty of stimulus perception and the subsequent actions of the subject. The model suggests that when the stimulus intensity falls below the (blurred) threshold of stimulus perception, the subject suspends the control and maintains the control variable near zero with accuracy determined by the control uncertainty. As the stimulus intensity grows above the perception uncertainty and becomes accessible to human cognition, the subject activates control. Consequently, the system dynamics can be conceptualized as an alternating sequence of passive and active modes of control with probabilistic transitions between them. Moreover, these transitions are expected to display hysteresis due to decision-making inertia.

Generally, the passive and active modes of human control are governed by different mechanisms, posing challenges in developing efficient algorithms for their description and numerical simulation. The proposed model overcomes this problem by introducing the dynamical trap of the stimulus-response type, which has a complex structure. The dynamical trap region includes two subregions: the stagnation region and the hysteresis region. The model is based on the formalism of stochastic differential equations, capturing both probabilistic transitions between control suspension and activation as well as the internal dynamics of these modes within a unified framework. It reproduces the expected properties in control suspension and activation, probabilistic transitions between them, and hysteresis near the perception threshold. Additionally, in a limiting case, the model demonstrates the capability of mimicking a similar subject’s behavior when (1) the active mode represents an open-loop implementation of locally planned actions and (2) the control activation occurs only when the stimulus intensity grows substantially and the risk of the subject losing the control over the system dynamics becomes essential.

The work is devoted to the description of the author’s formal statements of the clustering problem for a given number of clusters, algorithms for their solution, as well as the results of using this toolkit in medicine.

The solution of the formulated problems by exact algorithms of implementations of even relatively low dimensions before proving optimality is impossible in a finite time due to their belonging to the NP class.

In this regard, we have proposed a hybrid algorithm that combines the advantages of precise methods based on clustering in paired distances at the initial stage with the speed of methods for solving simplified problems of splitting by cluster centers at the final stage. In the development of this direction, a sequential hybrid clustering algorithm using random search in the paradigm of swarm intelligence has been developed. The article describes it and presents the results of calculations of applied clustering problems.

To determine the effectiveness of the developed tools for optimal clustering of multidimensional objects according to a variety of heterogeneous indicators, a number of computational experiments were performed using data sets including socio-demographic, clinical anamnestic, electroencephalographic and psychometric data on the cognitive status of patients of the cardiology clinic. An experimental proof of the effectiveness of using local search algorithms in the paradigm of swarm intelligence within the framework of a hybrid algorithm for solving optimal clustering problems has been obtained.

The results of the calculations indicate the actual resolution of the main problem of using the discrete optimization apparatus — limiting the available dimensions of task implementations. We have shown that this problem is eliminated while maintaining an acceptable proximity of the clustering results to the optimal ones. The applied significance of the obtained clustering results is also due to the fact that the developed optimal clustering toolkit is supplemented by an assessment of the stability of the formed clusters, which allows for known factors (the presence of stenosis or older age) to additionally identify those patients whose cognitive resources are insufficient to overcome the influence of surgical anesthesia, as a result of which there is a unidirectional effect of postoperative deterioration of complex visual-motor reaction, attention and memory. This effect indicates the possibility of differentiating the classification of patients using the proposed tools.

We present an inhomogeneous two-dimensional network model of two-phase flow in porous media. The edges of the network are assumed to be capillary tubes of different radii. We propose a new algorithm for handling phase fluxes at the nodes of this network model. We perform two test problems and show that the two-phase flow in this inhomogeneous network model demonstrates properties that are analogous to those of real porous media: capillary imbibition, dependence of capillary pressure on saturation and effect of capillary forces in two-phase displacement. The two test problems are: the counter-current imbibition and the twophase displacement in a periodically inhomogeneous porous medium. In the former problem, we implement a network consisting of two regions: a region of low-permeability with thin capillaries surrounded by a region of high-permeability with thick capillaries, initially saturated with wetting and nonwetting incompressible fluids, respectively. Capillary equilibrium is established due to counter-current imbibition by a region. We examine the dependence: of saturation of the wetting fluid with respect to time in the regions, and of capillary pressure on the current saturation. We have obtained a qualitative agreement with the known experimental and theoretical results, which will further allow us to use this network model to verify homogenized models of capillary nonequilibrium. In the latter problem, we consider the two-phase displacement, where the network is initially saturated with nonwetting fluid. Then wetting fluid is injected through a boundary at a constant rate. We analyze the saturation with respect to the axis which is along the applied pressure gradient for various moments in time with various values of coefficients of surface tension. The results show that for lower values of coefficient of surface tension, the wetting fluid prefers to invade through the thicker tubes, and in the case of higher values, through thinner tubes.

In today’s technology-driven world, lite devices like Internet of Things (IoT) devices and microcontrollers (MCUs) are becoming increasingly common. These devices are more energyefficient and affordable, often with reduced features compared to the standard versions such as very limited memory and processing power for typical machine learning models. However, modern machine learning models can have millions of parameters, resulting in a large memory footprint. This complexity not only makes it difficult to deploy these large models on resource constrained devices but also increases the risk of latency and inefficiency in processing, which is crucial in some cases where real-time responses are required such as autonomous driving and medical diagnostics. In recent years, neural networks have seen significant advancements in model optimization techniques that help deployment and inference on these small devices. This narrative review offers a thorough examination of the progression and latest developments in neural network optimization, focusing on key areas such as quantization, pruning, knowledge distillation, and neural architecture search. It examines how these algorithmic solutions have progressed and how new approaches have improved upon the existing techniques making neural networks more efficient. This review is designed for machine learning researchers, practitioners, and engineers who may be unfamiliar with these methods but wish to explore the available techniques. It highlights ongoing research in optimizing networks for achieving better performance, lowering energy consumption, and enabling faster training times, all of which play an important role in the continued scalability of neural networks. Additionally, it identifies gaps in current research and provides a foundation for future studies, aiming to enhance the applicability and effectiveness of existing optimization strategies.

The efficiency of mobile robotic systems (MRS) that monitor the traffic situation, urban infrastructure, consequences of emergency situations, etc., directly depends on the quality of vision systems, which are the most important part of MRS. In turn, the accuracy of image processing in vision systems depends to a great extent on the accuracy of spatial orientation of the video camera placed on the MRS. However, when video cameras are placed on the MRS, the level of errors of their spatial orientation increases sharply, caused by wind and seismic vibrations, movement of the MRS over rough terrain, etc. In this connection, the paper considers a general solution to the problem of stochastic estimation of spatial orientation parameters of video cameras in conditions of both random mast vibrations and arbitrary character of MRS movement. Since the methods of solving this problem on the basis of satellite measurements at high intensity of natural and artificial radio interference (the methods of formation of which are constantly being improved) are not able to provide the required accuracy of the solution, the proposed approach is based on the use of autonomous means of measurement — inertial and non-inertial. But when using them, the problem of building and stochastic estimation of the general model of video camera motion arises, the complexity of which is determined by arbitrary motion of the video camera, random mast oscillations, measurement disturbances, etc. The problem of stochastic estimation of the general model of video camera motion arises. Due to the unsolved nature of this problem, the paper considers the synthesis of both the video camera motion model in the most general case and the stochastic estimation of its state parameters. The developed algorithm for joint estimation of the spatial orientation parameters of the video camera placed on the mast of the MRS is invariant to the nature of motion of the mast, the video camera, and the MRS itself, providing stability and the required accuracy of estimation under the most general assumptions about the nature of interference of the sensitive elements of the autonomous measuring complex used. The results of the numerical experiment allow us to conclude that the proposed approach can be practically applied to solve the problem of the current spatial orientation of MRS and video cameras placed on them using inexpensive autonomous measuring devices.

We examine a phenomenological algorithm for generating morphology of astrocytes, a major class of glial brain cells, based on morphometric data of rat brain protoplasmic astrocytes and observations of general cell development trends in vivo, based on current literature. We adapted the Space Colonization Algorithm (SCA) for procedural generation of astrocytic morphology from scratch. Attractor points used in generation were spatially distributed in the model volume according to the synapse distribution density in the rat hippocampus tissue during the first week of postnatal brain development. We analyzed and compared astrocytic morphology reconstructions at different brain development stages using morphometry estimation techniques such as Sholl analysis, number of bifurcations, number of terminals, total tree length, and maximum branching order. Using morphometric data from protoplasmic astrocytes of rats at different ages, we selected the necessary generation parameters to obtain the most realistic three-dimensional cell morphology models. We demonstrate that our proposed algorithm allows not only to obtain individual cell geometry but also recreate the phenomenon of tiling domain organization in the cell populations. In our algorithm tiling emerges due to the cell competition for territory and the assignment of unique attractor points to their processes, which then become unavailable to other cells and their processes. We further extend the original algorithm by splitting morphology generation in two phases, thereby simulating astrocyte tree structure development during the first and third-fourth weeks of rat postnatal brain development: rapid space exploration at the first stage and extensive branching at the second stage. To this end, we introduce two attractor types to separate two different growth strategies in time. We hypothesize that the extended algorithm with dynamic attractor generation can explain the formation process of fine astrocyte cell structures and maturation of astrocytic arborizations.

Charge transfer in DNA is simulated by a discrete Holstein model «quantum particle + classical site chain + interaction». Thermostat temperature is taken into account as stochastic force, which acts on classical sites (Langevin equation). Thus dynamics of charge migration along the chain is described by ODE system with stochastic right-hand side. To integrate the system numerically, algorithms of order 1 or 2 are usually applied. We developed «mixed» algorithm having 4th order of accuracy for fast «quantum» variables (note that in quantum subsystem the condition «sum of probabilities of charge being on site is time-constant» must be held), and 2nd order for slow classical variables, which are affecting by stochastic force. The algorithm allows us to calculate trajectories on longer time intervals as compared to standard algorithms. Model calculations of polaron disruption in homogeneous chain caused by temperature fluctuations are given as an example.

The program for automatic revealing of threshold values for characterizing physiological state of vessels and detection of early stages of retina pathology is offered. The algorithm is based on checking character of crossing sites of vessel images with the "mask" consisting of concentric circumferences (the first circumference is imposed directly on the sclera capsules of an optic nerve disk). The new method allows revealing of a network of blood vessels and flanking zones and detection of initial stage of pathological changes in a retina by digital images.

In this article the experience of reading “CUDA and OpenCL programming” course during high perfomance computing summer school MIPT-2010 is analyzed. Content of lectures and practical tasks, as well as manner of presenting of the material are regarded. Performance issues of different algorithms implemented by students at practical training session are dicussed.

Langevin Dynamics, Monte Carlo, and all-atom Molecular Dynamics simulations in implicit solvent require a reliable source of pseudorandom numbers generated at each step of calculation. We present the two main approaches for implementation of pseudorandom number generators on a GPU. In the first approach, inherent in CPU-based calculations, one PRNG produces a stream of pseudorandom numbers in each thread of execution, whereas the second approach builds on the ability of different threads to communicate, thus, sharing random seeds across the entire device. We exemplify the use of these approaches through the development of Ran2, Hybrid Taus, and Lagged Fibonacci algorithms. As an application-based test of randomness, we carry out LD simulations of N independent harmonic oscillators coupled to a stochastic thermostat. This model allows us to assess statistical quality of pseudorandom numbers. We also profile performance of these generators in terms of the computational time, memory usage, and the speedup factor (CPU/GPU time).