РУС
ENG
ORTHOGONALIZATION OF NEURAL NETWORK WEIGHT COEFFICIENTS
Goals and sphere
Founder and publisher
Editorial Board
Licensing conditions
Confidentiality
Attitude towards plagiarism
Publication ethics
Archiving Policy
Subscription
Instructions for authors
The review process
Copyright
Agreement on the transfer of rights
Editorial fees
All issues
Search
Contacts
Zuev, Vladimir Mikhailovich
Head of the Department
IRTS of the Federal State Budgetary Scientific Institution "Institute of Artificial Intelligence Problems".
Research interests: intelligent robotic systems.
UDC 004.93
DOI 10.24412/2413-7383-135-140
Language: Russian
Annotation:
The article considers the regularization of a convolutional neural network using the orthogonalization method. The proposed method can be used to reduce the number of optimized weight coefficients. A preliminary filter adjustment stage is introduced, which allows choosing the optimal number of convolution layers and speeding up training due to the optimal filter adjustment at the beginning of training. The analysis shows that the method we propose is quite efficient.
Keywords: neural network, regularization, orthogonalization.
Список литературы:
1. Verzhbitskiy V.M. Numerical methods M. Higher School of Economics, 2001
2. Saad Yu. Iterative methods for sparse linear systems: Study guide. – In 2 volumes. Volume 1 / Translated from English by H.D. Ikramov; Preface by V.A. Sadovnichy, Moscow: Moscow University Press, 2013– 344 p.
3. https://habr.com/ru/companies/smartengines/articles/495816/4.8K
4. Shilov G. E. Finite-dimensional linear spaces. Moscow: Nauka Publ., 1969
5. Beklemishev D.V. Additional chapters of linear algebra. Moscow: Nauka, 1983.
6. Timofeeva, N. V. Linear algebra. Modern algebra: a textbook / N. V. Timofeeva; Yaroslav State University named after P. G. Demidov. Yaroslavl: YarGU, 2012. 114 p.
Release: 1(36)'2025
Chapter: ROBOTS, MECHATRONICS AND ROBOTIC SYSTEMS
How to quote:
Zuev V.M. ORTHOGONALIZATION OF NEURAL NETWORK WEIGHT COEFFICIENTS // Problems of artificial intelligence. 2025. №1.
