   |
  |
||
|
|||
 |
|||
“Informatics and Applications” scientific journalVolume 19, Issue 4, 2025Content Abstract and Keywords About Authors FILTERING OF SPECIAL MARKOV JUMP PROCESSES BY OBSERVATIONS WITH MULTIPLICATIVE NOISE
Abstract: The paper is devoted to the problem of optimal state filtering for a class of special Markov jump processes.
Keywords: special Markov jump process; observations with multiplicative noise; conditional probability density function; Kushner-Stratonovich equation A PRACTICAL STUDY OF THE EXTENDED KALMAN FILTER INSTABILITY
Abstract: The paper examines the variants of unstable operation of the extended Kalman filter (EKF). The set of experiments was performed with a typical model of a stochastic observation system. The motion of an autonomous object with a constant average velocity was modeled under conditions of uncontrolled velocity perturbations forming a chaotic trajectory with a regular target direction. Observations of two independent complexes consist of measurements of bearing angles (azimuth and elevation angle) and range. The estimation of the object's position is performed by the basic EKF and its modification using the method of linear pseudoobservations. The basic EKF turns out to be unstable in the initial model. The EKF uses the method of pseudomeasurements to provide a stable assessment of the position with high accuracy. The purpose of the experiments is to show which changes in the monitoring system model led to unstable operation of this EKF modification. For this purpose, 4 scenarios have been proposed, calculated, and analyzed: (i) inaccurate detection of the initial position; (ii) inability to identify the speed parameters in advance; (iii) movement with an abrupt change in speed parameters while maintaining the direction of the target; and (iv) inaccurate setting of statistical characteristics (covariance) of measurement errors. In each of the scenarios, the EKF turns out to be unstable, forming an estimate of the object's position with unacceptable accuracy. At the same time, the nature of instability and the behavior of the EKF estimates are different as demonstrated by numerical and graphical calculation results. Keywords: stochastic filtering; discrete stochastic observation system; extended Kalman filter (EKF); EKF by the method of linear pseudomeasurement ON LATENCY ANALYSIS IN INTEGRATED ACCESS AND BACKHAUL DEPLOYMENTS WITH LINEAR TOPOLOGY
Abstract: Integrated Access and Backhaul (IAB) technology standardized by the 3GPP provides the possibility to significantly reduce the cost of 5G New Radio deployments. The authors study the transmission delay in dense urban areas operating in the millimeter frequency range. The proposed system model allows for buffering analysis at transit IAB-nodes using the queuing theory and includes parameterization of the radio channel using the methods of stochastic geometry. The numerical experiment showed that the system performance in terms of packet delay and resource utilization is determined mainly by road traffic conditions and not by the coverage area of IAB-nodes. Keywords: 5G New Radio; latency; queue HEURISTIC ONLINE LOAD BALANCING IN TWO-PHASE TANDEM QUEUES WITH DELAYS
Abstract: A single flow of customers arrives to the two-phase tandem queueing system. The first phase is the infinite-server queue which models the individual customer's delay. The second phase consists of N identical single server infinite capacity queues running in parallel. Upon arrival of a customer, the dispatcher must immediately decide which queue of the second phase will serve it. The dispatcher has certain a priori static information about the system and the incoming flow but the dynamic information about the queues arrives with a random delay.
Keywords: parallel service systems; dispatching; load balancing; delayed information; redundancy REGULAR REPRESENTATIVE ELEMENTARY CLASSIFIERS OVER THE PRODUCT OF PARTIAL PRODUCTS
Abstract: The authors consider the issues of creating algorithmic support for supervised classification problem which is the one of the central tasks of machine learning. Original procedures of logical analysis and classification of integer data represented as a set of elements of Cartesian product of finite partially ordered sets (product of partial orders) are constructed and investigated. At the training stage of the proposed procedures, the search for so-called regular representative elementary classifiers (special fragments in feature descriptions of precedents that distinguish objects belonging to different classes) is performed. An asymptotically optimal algorithm for enumerating the required elementary classifiers over a product of antichains is constructed and the results of its testing on real-world tasks are presented. Theoretical and experimental justifications for the efficiency of the new classification procedures are provided for the case when linear orders on sets of feature values are defined.
Keywords: supervised classification; correct logical classifier; regular representative elementary classifier; partially ordered data; Cartesian product of partial orders; metric (quantitative) properties of the set of elementary classifiers COLOR IMAGE RESTORATION VIA THE LATTICE BOLTZMANN METHOD FOR ANISOTROPIC NONLINEAR DIFFUSION
Abstract: The work proposes a method for restoring damaged regions of color three-channel images (the inpainting problem) based on the equation of nonlinear anisotropic diffusion. As the numerical solution algorithm, the lattice Boltzmann equation with five discrete velocities and multiple relaxation times is employed. The direction and intensity of the smoothing are determined using the structure matrix. A parallel implementation of the algorithm has been developed using MPI (Message Passing Interface) technology with image domain decomposition in a Cartesian topology. The application of the proposed method to images with defects of various shapes and sizes is examined. The results demonstrate the correctness of structural and color information restoration in the damaged regions. The accuracy of the method is evaluated on a test set of 10,000 images, and the execution times of sequential and parallel versions of the algorithm are compared. Keywords: image restoration; inpainting; lattice Boltzmann equations; anisotropic diffusion OPTIMIZATION ACCORDING TO THE QUANTILE CRITERION OF THE TEST TAKER POSITION STRATEGY IN THE DYNAMIC MODEL OF PASSING THE TIME-LIMITED TEST
Abstract: The problem of building optimal program and positional strategy in dynamic model of passing time- limited test is considered. The tester sequentially solves the test tasks, gaining a certain number of points for each task in case of the correct solution. The correctness of the test of each task is modeled by a random variable with a Bernoulli distribution. The time spent on solving each task is also considered to be random. The positional strategy is a function of the number of points scored after solving the next task and the total time spent on solving previous test tasks. The function takes the value one if the tester solves the next task and zero if misses. The criterion is the number of points scored for the test, the excess of which, while simultaneously fulfilling the limit on the test execution time, is guaranteed with a predetermined level of confidence which acts as a task parameter. To solve the problems under consideration, the equivalence property is used between the problem with the quantile criterion and the problem of maximizing the corresponding probability function. After that, a modification of the algorithm for solving a similar problem with a probabilistic quality criterion proposed earlier by the authors is used. Keywords: time-limited test; dynamic model; positional strategy; quantile criterion INDIRECT PROPERTIES IN CLASSIFICATION OF LARGE DATA WITH THE HELP OF CAUSE-AND-EFFECT RELATIONSHIPS
Abstract: The usage of cause-and-effect relationships to classify small data sets of high dimension can generate conflicts due to the fact that a significant part of the data does not play a significant role in the classification task and can be considered as random data. In this case, in terms of cause-and-effect relationships, random data can generate pieces ofinformation that interfere with correct classification or generate classification errors. Additional information is needed to neutralize characteristics errors. In the present paper, such additional information was also found using causal relationships. The authors define indirect characteristics that can be used to resolve conflicts and to refine the classification. Using the task ofclassifying ofthree informative classes as an example, it is shown how to obtain and how to use indirect characteristics to resolve conflict situations during the classification process and error prevention process. Keywords: classification; cause-and-effect relationships; indirect characteristics of correct classification
|
 | |
 Phone of the Center: +7 (499) 135-62-60E-mail of the Center: ipiran@ipiran.ru | 
RUS
|