Article 3318

Title of the article

A FAST LEARNING ALGORITHM OF LARGE NETWORKS OF ARTIFICIAL NEURONS OF THE SQUARED GEOMETRIC MEAN OF DENSITIES OF MULTIDIMENSIONAL VALUE DISTRIBUTION OF BIOMETRIC DATA 

Authors

Volchikhin Vladimir Ivanovich, Doctor of engineering sciences, professor, president of Penza State University (40 Krasnaya street, Penza, Russia), E-mail: president@pnzgu.ru
Ivanov Aleksandr Ivanovich, Doctor of engineering sciences, associate professor, head of the laboratory of biometric and neural network technologies, Penza Research Institute of Electrical Engineering (9 Sovetskaya street, Penza, Russia), E-mail: ivan@pniei.penza.ru
Perfilov Konstantin Aleksandrovich, Postgraduate student, Penza State University (40 Krasnaya street, Penza, Russia), E-mail: perfilov58@gmail.com
Malygina Elena Aleksandrovna, Candidate of engineering sciences, researcher, interdisciplinary laboratory of biometric devices and technologies testing, Penza State University (40 Krasnaya street, Penza, Russia), E-mail: mal890@yandex.ru
Serikova Yuliya Igorevna, Postgraduate student, Penza State University (40 Krasnaya street, Penza, Russia), E-mail: julia-ska@yandex.ru 

Index UDK

519.24; 53; 57.0171 

DOI

10.21685/2072-3059-2018-3-3 

Abstract

Background. The aim of the presented work is to describe artificial neurons, built as analogues of the statistical criterion of the square geometric mean of densities of "Friend" multidimensional values distribution of biometric data and multidimensional density distribution of values against the biometric data. The article solves the problem of transferring from one-dimensional to multidimensional statistical analysis of biometric data by creating a collection of neurons of the geometric mean.
Materials and methods. the method of simulation was to solve the task.
Results. Two options were proposed for the realization of neurons. The first option focused on computing equipment with high bitness, the software of which was able to accurately calculate the integral of the density distribution of values to compare works. The second option was based on the application of low-bit 2D logarithmic tables of pre-computed values of the squared geometric mean of probability densities. The difference between expected and observed mathematical expectations of compared biometric data was used as a table variable. The ratio of standard deviations of the two samples compared was used as another variable.
Conclusions. It is proved that the power generated by quadratic neurons is much higher than the power of classical quadratic radial basis of neurons. At the same time, the crucial property of linear computational complexity of quadratic neurons teaching is saved, allowing to quickly teach anyhow large artificial neural networks of the geometric mean for small training samples. 

Key words

Neural code-biometrics converter, biometric data, big data dimensionality, low bitness calculations using logarithmic tables of probability density 

 Download PDF
References

1. Kobzar' A. I. Prikladnaya matematicheskaya statistika dlya inzhenerov i nauchnykh rabotnikov [Applied mathematical statistics for engineers and researchers]. Moscow: Fizmatlit, 2006, 816 p.
2. GOST R 50.1.037–2002. Prikladnaya statistika. Pravila proverki soglasiya opytnogo raspredeleniya s teoreticheskim. Ch. 1. Kriterii tipa khi-kvadrat [Applied statistics. Rules of verifying agreement between testing and theoretical distribution. Part 1. Chisquare criteria]. Moscow: Gosstandart Rossii, 2001, 140 p.
3. GOST R 50.1.037–2002. Prikladnaya statistika. Pravila proverki soglasiya opytnogo raspredeleniya s teoreticheskim. Ch. 2. Neparametricheskie kriterii [Applied statistics. Rules of verifying agreement between testing and theoretical distribution. Part 2. Nonparametric criteria]. Moscow: Gosstandart Rossii, 2002, 123 p.
4. Ivanov A. I. Mnogomernaya neyrosetevaya obrabotka biometricheskikh dannykh s programmnym vosproizvedeniem effektov kvantovoy superpozitsii: monografiya [Multidimensional neural network processing of biometric data with software reproduction of quantum superposition effects: monograph]. Penza: Izd-vo AO «PNIEI», 2016, 133 p.
5. Volchikhin V. I., Ivanov A. I., Serikova N. I., Funtikova Yu. V. Izvestiya vysshikh uchebnykh zavedeniy. Povolzhskiy region. Tekhnicheskie nauki [University proceedings. Volga region. Engineering sciences]. 2015, no. 1 (33), pp. 50–59.
6. Malygin A. Yu., Volchikhin V. I., Ivanov A. I., Funtikov V. A. Bystrye algoritmy testirovaniya neyrosetevykh mekhanizmov biometriko-kriptograficheskoy zashchity informatsii [Fast algorithms of testing neural network mechanisms of biometric cryptographic
protection of information]. Penza: Izd-vo PGU, 2006, 161 p.
7. Serikova N. I., Ivanov A. I., Serikova Yu. I. Voprosy radioelektroniki [Issues of radio electronics]. 2015, no. 1 (1), pp. 85–94.
8. Ivanov A. I., Perfilov K. A. Sovremennye okhrannye tekhnologii i sred-stva obespecheniya kompleksnoy bezopasnosti ob"ektov: XI Vseros. nauch.-prakt. konf. [Modern security technologies and onject’s complex security means: XI All-Russian scientific and practical conference]. Penza; Zarechnyy, 2016, pp. 223–229.
9. GOST R 52633.0–2006. Zashchita informatsii. Tekhnika zashchity informatsii. Trebovaniya k sredstvam vysokonadezhnoy biometricheskoy autentifikatsii [Information protection. Information protection technology. Requirements to means of highreliability biometric authentification]. Moscow: Standartinform, 2007, 27 p.
10. GOST R 52633.5–2011. Zashchita informatsii. Tekhnika zashchity informatsii. Avtomaticheskoe obuchenie neyrosetevykh preobrazovateley biometriya-kod dostupa [Information protection. Information protection technology. Automatic learning of neural
network biometrics-access code converters]. Moscow: Standartinform, 2012, 16 p.
11. GOST R 52633.3–2011. Zashchita informatsii. Tekhnika zashchity informatsii. Testirovanie stoykosti sredstv vysokonadezhnoy biometricheskoy zashchity k atakam podbora [Information protection. Information protection technology. Testing of highreliability biometric protection means’ resistance to picking attacks]. Moscow: Standartinform, 2012, 16 p.
12. Saymon Kh. Neyronnye seti: polnyy kurs [Neural networks: complete course]. Moscow: Vil'yams, 2006, 1104 p.
13. Akhmetov B. B., Ivanov A. I., Bezyaev A. V., Funtikova Yu. V. Vestnik Natsional'noy akademii nauk Respubliki Kazakhstan [Bulletin of the National Academy of Sciences of the Republic of Kazakhstan]. 2015, no. 1, pp. 5–11.
14. Ivanov A. I., Gazin A. I., Perfilov K. A., Vyatchanin S. E. Problemy informatsionnoy bezopasnosti. Komp'yuternye sistemy [Problems of information security. Computer systems]. 2016, no. 2, pp. 19–25.
15. Akhmetov B., Ivanov A. Mnogomernye statistiki sushchestvenno zavisimykh biometricheskikh dannykh, porozhdaemye neyrosetevymi emulyatorami kvadratichnykh form: monografiya [Multidimensional statistics of significantly dependent biometric data
created by neural network emulators of quadratic forms: monograph]. Almaty, Kazakhstan: LEM, 2016, 86 p.
16. Ivanov A. I., Lozhnikov P. S., Sulavko A. E., Serikova Yu. I. Infokommunikatsionnye tekhnologii [Infocommunicational technologies]. 2017, no. 15 (2), pp. 186–193.
17. Ivanov A. I., Lozhnikov P. S., Kachaykin E. I. Voprosy zashchity informatsii [Issues of information protection]. 2015, no. 2, pp. 28–34.
18. Ivanov A. I., Lozhnikov P. S., Kachaykin E. I., Sulavko A. E. Voprosy zashchity informatsii [Issues of information protection]. 2015, no. 3, pp. 48–54.
19. Ivanov A. I., Zakharov O. S. Sreda modelirovaniya «BioNeyroAvtograf» [Simulation environment “BioNeuroAvutograph”]. [Pro-grammnyy produkt sozdan laboratoriey biometricheskikh i neyrosetevykh tekhnologiy]. Available at: http://pniei.rf/activity/science/noc.htm (accessed Aug. 10, 2017).
20. Ivanov A. I., Perfilov K. A., Malygina E. A. Vestnik Sibirskogo gosudarstvennogo aerokosmicheskogo universiteta imeni akademika M. F. Reshetnikova [Bulletin of Siberia State Aerospace University named after academician M.F. Reshetnikov]. 2016, no. 4 (17), pp. 864–871.

 

Дата создания: 19.04.2019 14:00
Дата обновления: 22.04.2019 08:05