M.A. Abrosimov – Post-graduate Student, 

Department of Applied Information Technologies, Yuri Gagarin State Technical University of Saratov E-mail: destinywatcher@gmail.com

A.V. Brovko – Dr. Sci. (Phys.-Math.), Professor, 

Department of Applied Information Technologies, Yuri Gagarin State Technical University of Saratov

Email: brovkoav@gmail.com

This paper describes the issue of error level fluctuations due to training set shrinking in RBF-networks.

The aim of this paper is in development of artificial neural network (ANN) architecture that has a range of applications similar to RBF-networks as well as not sharing training set reduction sensitivity that is common for RBF-network. That aim requires developing both an ANN architecture and a training algorithm for it.

As a result, this paper presents an ANN architecture based on RBF-network with a learning algorithm to train it. The presented architecture is multi-layer, unlike original RBF-network thus has a potential in deep learning. Numeric results lead to a conclusion about error level fluctuations being significantly lower for the presented architecture compared to RBF-network in case of training set shrinking. This displays a greater generalization ability of the presented architecture.

The paper contains an application of ANN to the task of restoring the dielectric parameters for subject placed in waveguide [1,2]. The presented ANN architecture can be generally applied to function approximation tasks.

Abrosimov M.A., Brovko A.V. A high generalization capability artificial neural network architecture based on RBF-network. Neurocomputers. 2020. V. 22. № 1. P. 39–45. DOI: 10.18127/j19998554-202001-04.

1. Brovko A.V., Murphy E.K., Yakovlev V.V. Waveguide microwave imaging: neural network reconstruction of functional 2-D permittivity profiles. IEEE Transactions on Microwave Theory and Techniques. 2009. V. 57. № 2. P. 406-414.
2. Brovko A.V. Metod opredeleniya triyohmernogo profilya dielektricheskoy pronitsaemosti obraztsa s ispolzovaniem neyrosetevoi matematicheskoi modeli. Radiotekhnika. 2015. № 10. S. 115-120. (in Russian).
3. Aksenov S.V., Novoseltsev V.B. Organizatsia i ispolzovanie neyronnih setei. Tomsk. Izdatel’stvo NTL. 2006. (in Russian).
4. Haykin S. Neural Networks: A Comprehensive Foundation. Second Edition. New Jersey: Prentice-Hall. 1998.
5. Abrosimov M.A., Brovko A.V. Kriteri opredeleniya minimalnogo nabora bazisov dlya iskusstevennoy neyronnoy seti s radialnimi bazinimi funktsiyami. Neyrocomputeri: razrabotka, primenenie. 2018. №3. S. 50-54. (in Russian).
6. Bianchini M., Frasconi P., Gori M. Learning without local minima in radial basis function networks. IEEE Transactions on Neural Networks. 1995. V. 6. P. 749-756.
7. Craddock R.J., Warwick K. Multi-layer radial basis function networks. An extension to the radial basis function // IEEE Proceedings of International Conference on Neural Networks (ICNN'96). 1996. P. 700-705.
8. Chao J., Hoshino M., Kitamura T., Masuda T. A multilayer RBF network and its supervised learning. IEEE IJCNN'01. International Joint Conference on Neural Networks. Proceedings (Cat. No.01CH37222). 2001. P. 1995-2000.
9. Krisilov V.A., Kondratyuk A.V. Preobrazovanie vhodnih dannih neyronnoi seti s tseliu uluchsheniya ih razlichaemosti [Digital resource in Russian] access link: http://neuroschool.narod.ru/pub/krikon.pdf 
10. Fausett L.V. Fundamentals of Neural Networks: Architectures Algorithms and Applications. New Jersey: Prentice-Hall. 1994. 
11. Dolinina O.N., Kuzmin A.K. Metod postroeniya testov dlya otladki neyrosetei na osnove tehnicheskoi diagnostiki. Sbornik trudov mezhdunarodnoi nauchnoi konferencii. Matematicheskie metody v tekhnike i tekhnologiyakh. MMTT-23. v 12 T. T.6. Sekcii 6, 7. Saratov: Sarat. gos.tekh. universitet. 2010. S.38-41. (in Russian).
12. Ivjenko S.P. Intellektualni analyz dannih na osnove kart Kohkhonena. Trudy XXI Vseros. nauchno-metod. Konf. «Telematika’2014» SPb.: Telematika. 2014. S.250. (in Russian).