Amelin K.S., Baklanovsii M.V., Granichin O.N. et al. (SPbSU) Adaptive multi-agent real-time operating system
Amelina N.O., Amelin K.S. (SPbSU), Vergados D.J. (NTNU) Scheduling in stochastic wireless multihop networks
Ivanskiy Y.V. (SPbSU) 3D-image recovery from compressed ultrasonography
Senov A. A. (SPbSU) On exact confidence sets for linear regression parameter under arbitrary noise
Khlebnikov M.V., Shcherbakov P.S. (IPU RAS) An LMI-approach to the design of stabilizing bounded control for linear systems
Khokhulina V.A., Chirkov M.K. (SPbSU) On abstract analysis of fuzzy minimax automata models
Shats V.N. (SPb) Two-level metrics and a new concept of machine learning
Stoynov P.T. (
Adaptive Multi-Agent Real-Time Operating System
K. S Amelin, M. V. Baklanovskiy, O. N. Granichin, Y. V. Ivanskiy, A. D. Kornivets, N. V. Mal’kovskiy, D. G. Naidanov, R. Y. Shein
Key words: adaptive systems, multi-agent systems, real-time operating system.
The development of computer technology is moving from rising the clock frequency and the performance of a single processor towards multi-core parallelism and integration of the individual processors into a complex. However the approach based on combining the computing nodes into a single system has its own drawbacks. Such systems have a rigid hierarchy which limits the scope of their efficient use, since the particular configuration of the computer system is targeted at solving a specific range of tasks. Next, failure of a small number of nodes at the upper levels of the hierarchy can affect the functioning of the entire system. In addition, when planning the load of computing nodes, serious difficulties arise due to the large number of nodes and complex switching rules. To overcome such difficulties in the construction of large-scale systems, the ideology of multi-agent systems becomes popular. Such systems do not assume the existence of a priori structured links between the elements (agents), and each element has certain autonomy and is capable of forming new links with other agents in the process of solving a problem.
The paper describes the main features of the new project by the authors which is aimed at the development of the prototype of a new real-time operating system (OS) targeted at properly scheduling the work of complexes consisting of a large number of computing nodes which are built on the principles of the multi-agent technology. Such decentralized OS should ensure the efficient load of computing devices and have the ability to change the topology of the network caused both by on-line network re-structuring and possible losses and/or acquisition of nodes.
Bibliogr.: 8 refs.
Scheduling in Stochastic Wireless Multihop Networks
N. O. Amelina, K. S. Amelin
D. J. Vergados
(natalia_amelina, konstantinamelin)@mail.ru, email@example.com
Key words: consensus problem, wireless multihop networks, scheduling, load balancing.
One of the challenges in wireless multihop networks is the problem of scheduling transmissions in an efficient and fair manner. The performance of a scheduling algorithm is closely related to its ability to adapt to the changing traffic conditions. Although theoretical results have been obtained regarding the capacity of wireless multihop networks, analytic results on the interaction of load balancing and scheduling algorithms have yet to be derived. In this paper we consider a stochastic wireless multihop network of nonlinear nodes with switching topology, noisy and delayed measurements. The problem of wireless scheduling was modeled as a load balancing problem and the consensus protocol was suggested to solve it. Conditions for an approximate consensus that gives an almost optimal behavior of the system were provided. Through analysis and simulation, we evaluate the performance of various scheduling algorithms. We show that load balancing improves the delay and fairness of the system.
Bibliogr.: 23 refs.
3D-Image Recovery from Compressed Ultrasonography Data
Y. V. Ivanskiy,
Key words: compressive sensing, compressive sampling, randomized measurements, ℓ1-optimization, ultrasonography.
The paper considers the applicability of compressed sensing to the reduction of the amount of channels connecting ultrasound transducers and the processing unit. The conceptual simulation model of ultrasound scanner was designed and implemented. The model provides compression, transmission, and recovery of signals generated from the ultrasonography data. A number of experiments with different parameters were performed. The results show the capabilities of the compressed sensing approach in the problem under consideration and the quality of ultrasonography data reconstruction at the processing unit.
Bibliogr.: 20 refs.
Determination of the Coordinates of Flying Vehicles from the Range-Difference Measurements Subject to Individual Systematic Errors
Y. K. Kisin,
Key words: measurements, range difference, systematic errors, algorithm, trajectory.
The paper addresses the problem of determining the coordinates of an flying vehicles from the range-difference measurements corrupted by systematic errors in every range difference. The results are illustrated via simulation examples.
Bibliogr.: 9 refs.
On Exact Confidence Regions under External Arbitrary Noise Within the Context of Linear Recommender System
A. A. Senov,
Key words: linear regression, confidence intervals, sign-perturbed sums, randomization.
The paper presents a new finite sample method for identifying non-asymptotic confidence sets for multiple linear regression parameters, referred to as Modified Sign-Perturbed Sums (MSPS), for almost arbitrary noise. The MSPS method is a modification of the Sign-Perturbed Sums (SPS) method (proposed earlier by Csaji B.C., Campi M.C., Weyer E.), with generalization to the case of arbitrary noise, typical to data in recommendation systems. The SPS and MSPS methods are compared on both simulated and real data. Statistical properties of confidence intervals are discussed. An analytical expression of confidence sets is given. The advantages of the MSPS method are illustrated.
Bibliogr.: 25 refs.
An Effective Construction of Genetic Maps from Completely Sequenced Genome Sections
S. S. Sysoev,
Key words: genetic linkage, linkage maps, sequenation, algorithm, pheno-type mapping.
A new approach is proposed to the construction of genetic maps from completely sequenced portions of data (markers). An algorithmic implementation of the approach using the Python environment is provided. Preliminary testing results from the genetic data for a set of cats with 35 markers are presented.
Bibliogr.: 8 refs.
An LMI-approach to the Design of Stabilizing Bounded Control for Linear Systems1
M. V. Khlebnikov,
P. S. Shcherbakov
Key words: linear control systems, linear matrix inequalities bounded control, linear quadratic regulator.
Using the linear matrix inequalities technique, a stabilizing linear static state feedback is designed for linear systems under the constraint on the magnitude of the control signal. A quadratic functional is then constructed such that it is minimized with the designed control.
Bibliogr.: 12 refs.
On Abstract Analysis of Fuzzy Minimax Automata Models
V. A. Khokhulina,
M. K. Chirkov
Key words: fuzzy automata models, fuzzy minimax automata, fuzzy languages, equivalence, special method of analysis.
In the paper, a special method of abstract analysis of fuzzy minimax automata is presented, by which means decisions can be made on general and special analysis tasks of this type of automata. The method is based on the equivalence of languages represented by different types of fuzzy automata and on the possibility of their decomposition on fuzziness degrees. This makes it possible to reduce abstract analysis of fuzzy minimax automata models to that of maximin automata, for which there exist solutions procedures. Examples are given.
Bibliogr.: 10 refs.
Optimal Synthesis of Nonstationary Nondeterministic
M. K. Chirkov,
Key words: nondeterministic automata, nonstationary automata, optimal synthesis of automata, optimal behavior, fuzzy functioning conditions.
The paper deals with the design of nonstationary nondeterministic finite state automaton that maximizes the performance index under fuzzy conditions; the additional optimality criterion is the number of states of the automaton. A general design procedure is proposed and an example is discussed.
Bibliogr.: 7 refs.
Two-Level Metrics and a New Concept of Machine Learning
V. N. Shats,
Key words: machine learning; self-organizing system; metric; supervised learning; learning without a teacher.
We develop a new concept of machine learning associated with a multi-step mechanism of processing external information in self-organizing systems. This concept is based on a two-level conformity metrics between the object and a class of plants. In accordance with this metrics, for each of the individual attributes and the overall set of them, we evaluate the probability that an object belongs to a given class; to determine the class, the maximum likelihood method is used. The metrics is implemented as a module having simple calculation algorithm. Essentially, the problems of learning with or without a teacher reduce to the calling to the module. The efficiency of the concept is shown via applications to specific problems.
Bibliogr.: 13 refs.
P. T. Stoynov,
Key words: surplus processes, reflected surplus processes, budget restriction.
In this paper we
consider a risk model based on reflected surplus processes. This kind of processes
can represent the surplus of companies with steady outflows and sporadic
inflows. We consider the case where the company can adjust its expense rate by
reducing it if no inflow is made by a random time with a specific distribution
following the last inflow.
Bibliogr.: 7 refs.