Journal of Applied Economic Research
ISSN 2712-7435
Correct Formulation of the Inverse Problems of Reconstruction Of Multidimensional Functions Using Neural Network Model in Budget Administration
Biryukov A.N., Gluschenko O.I.
Abstract
Abstract. The object of consideration of this work is very deep penetration of requirements for effective neural network learning algorithms in preprocessing processing. A method has been developed for assessing the adequacy of neural network models in the absence of any a priori information about the distribution law of noise in the data. This is the scientific novelty of this article, as this method provides for interlinked control of the quality of financial data preprocessing and their quality of approximation in the neural network for the budgetary authorities. It is proposed tht the class of problems in budget administration should be considered for which the noise data is quite large, so the need for regularization of hypersurface restoration tasks is a prerequisite of the effectiveness of hybrid models. In particular, the authors show that it is impossible to ignore the incorrect formulation of the problem. To overcome it, there are two ways: introduction of an inverse problem in the class of proper (conditionally correct according to Tikhonov) through additional information about the desired solution, or the control of classical algorithms for solving ill-posed problems. Both ways are based on the achievements of mostly Russian scientists. Existing regularizing algorithms use the original database as some external immutable «Givens». The article uses a meaningful practical example of the proposed systematic approach to the problem of ensuring the stability of neural network mapping when restoring a hypersurface with strong noise data based on the theory of the regularization by A. N. Tikhonov and the Bayesian approach. The essence of this method is that to ensure the viability of the algorithm for the regularization of inverse problems by A. N. Tikhonov base source data is not used as some «over-Stiva» category, and is subjected to pre-processing (structuring) using system-wide laws of Cybernetics (the law of entropy balance of an open system, incomplete suppression dysfunctions of a structured system, redundancy). Thus, the aim of the study is achieved – the essence of the proposed approach is detailed and implemented in the practical concept of «regularization» in this study through the operations of the algorithm in constructing the neural network model.
Keywords
Key words: budget revenues; budget beneficiaries; neural network model; interpretation; tax and nontax revenues; network regularization.
About Authors
Biryukov Aleksandr Nikolaevich – Doctor of Economics, Associate Professor, Professor of the Department of Economic Theory and Analysis, Sterlitamak Branch of Bashkir State University, Sterlitamak, Russia (453103, Republic of Bashkortostan, Sterlitamak, Lenin Avenue, 49); e-mail: biryukov_str@mail.ru.
Gluschenko Olga Ivanovna – Candidate of Economic Sciences, Associate Professor, Doctoral Candidate, East Economic-Legal Humanitarian Academy, Ufa, Russia, (450054, Bashkortostan, Ufa, prospect October street, 71/3); e-mail: Olga.glushhenko@mail.ru.
DOI: http://dx.doi.org/10.15826/vestnik.2017.16.2.012
Download full text article:
~674 KB, *.pdf
(Uploaded
07.04.2017)
Created / Updated: 2 September 2015 / 20 September 2021
© Federal State Autonomous Educational Institution of Higher Education «Ural Federal University named after the first President of Russia B.N.Yeltsin»
Remarks?
select the text and press:
Ctrl + Enter
Portal design: Artsofte
Contact us
Rector's Office
Rector, Dr. Victor Koksharov
Tel. +7 (343) 375-45-03, e-mail: rector@urfu.ru
Vice-Rector for International Relations, Dr. Maxim Khomyakov
Tel. +7 (343) 375-46-27, e-mail: Maksim.Khomyakov@urfu.ru