Search academic texts

Information × Registration Number 0221U106646, 0121U111617 , R & D reports Title Analytical methods and machine learning in control theory and decision-making in conditions of conflict and uncertainty popup.stage_title Head Norkin Volodymyr I., д.ф.-м.н. Registration Date 16-12-2021 Organization V.M. Glushkov Institute of Cybernetics of National Academy of Sciences of Ukraine popup.description2 ABSTRACT: Research report: 118 pages, 11 figures, 98 references. Keywords: nonsmooth global optimization, smoothing method, finite difference methods, stochastic programming, stochastic processes and fields, empirical mean method, dependent observations, large deviations, conflict controlled process, collision avoidance, resolving functions method. Object of research: Problems and methods for nonsmooth constrained optimization, stochastic programming, Markov fields with local interaction, conflict-driven dynamic processes. The purpose of the work. 1. To study conditions and rate of convergence of stochastic finite-difference methods for optimization of nonsmooth functions under constraints. 2. Extend the empirical means method to continuous models with a two-parameter random field and develop algorithms for finding estimates of unknown parameters for models with discontinuous regression functions. 3. Establish conditions to avoid clashes with a group of controlled objects with different awareness of the phase state and control of the enemy. Methods: variational analysis, theory of random processes and fields, differential games, methods of optimization. Results and novelty. Convergence conditions and the convergence rate of the stochastic smoothing method on different classes of nonconvex optimization problems under constraints are studied. A software "Stochastic sequential smoothing method" has been developed. Asymptotic properties of empirical estimates of solutions of stochastic optimization problems in conditions of dependent nonstationary observations are investigated. Conditions of consistency and exponential speed of convergence of the empirical estimates are given. Methods of avoiding collisions of one object with a group of controlled objects have been developed, sufficient conditions have been established to ensure a guaranteed result. Algorithms for optimizing interaction of groups and sufficient conditions for collision avoidance have been developed. Product Description popup.authors Bila Halyna D Bogdanov Oleksandr V Bulovatski Maxym Sergiovych Guro Dmytro Anatolievych Dunaievskyi Maksym S Knopov Pavlo S Kozyrev Anton Yurievych Norkin Volodymyr I Samosyenok Olexandr S Chikrii Arkadii O popup.nrat_date 2021-12-16 Close