PERSPECTIVES OF USING ARTIFICIAL NEURAL NETWORKS WITH MULTILAYER QUANTIZER IN TECHNOLOGY OF BIOMETRIC- NEURAL-NETWORK AUTHENTICANTION 

Volchikhin Vladimir Ivanovich, Doctor of engineering sciences, professor, president of Penza State University
(40 Krasnaya street, Penza, Russia), 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), ivan_pniei@penza.ru
Funtikov Vyacheslav Aleksandrovich, Candidate of engineering sciences, general director of Penza Research Institute of Electrical Engineering (9 Sovetskaya street, Penza, Russia),pniei@penza.ru
Malygina Elena Aleksandrovna, Postgraduate student, Penza State University (40 Krasnaya street, Penza, Russia), mal890@yandex.ru 

612.087.1; 519.7; 519.66 

Background. At the present time the problems of personality authentication using biometric data are becoming topical. The advantage of artificial neural networks of large size over classical codes with error detection and correction lies in the fact that in the moment of learning the networks are capable of taking into account real distributions of multidimensional probabilities of biometric data, whereas all the classical codes of error detection and correction are based on the hypothesis of probable distribution of errors. The article is aimed at changing the paradigm of neural network processing; the authors suggest to switch from binary neurons (perceptron) to using neurons with multilayer quantizers.
Materials and methods. The comparison is conducted using a complex code quality index - entropy (proximity to "the white noise"). For codes of about 20 bits in length the entropy may be calculated according to Shannon. For longer codes the resources modern machines are insufficient. It is suggested to analyze only the initial part of the code sequence of the increasing length. After that it is necessary to build an extrapolating polynomial and predict the expected long code’s entropy value.
Results. The resulting value of 256 bit entropy of codes of the neural network converter turned to be higher than 51-vit entropy of codes of "the fuzzy extractor". The gain is conditioned by the length of the bio-code despite the fact that long codes have a higher level of correlation of their positions. The transition from binary neurons to neurons with multilevel quantizers increases the gain to up to million times.
Conclusions. In the course of transition from binary neurons to ternary neurons the length of the output code increases two times, and their entropy increases appoximately 1,5 times. The gain relating to the increase of biocode entropy increases with the number of quantization levels in each neuron. At the same time the problems of neural network learning also become complicated. It is necessary to modify the standard algorithm of learning ГОСТ Р 52633.5–2011 for networks consisting of the combination of regular binary neurons and ternary neurons.

artificial neural networks, transition of biometry into code, binary quantizers, multilevel quantizers, neurons with high number of quantum conditions.

1. Juels A. A, Wattenberg M. Proc. ACM Conf. Computer and Communications Security, 1999, pp. 28–36
2. Monrose F., Reiter M., Li Q., Wetzel S. In Proc. IEEE Symp. on Security and Privacy, 2001.
3. Juels A. A, Sudan M. IEEE International Symposi-um on Information Theory, 2002.
4. Dodis Y., Reyzin L., Smith A. In EUROCRYPT, Data April 13, 2004, pp. 523–540.
5. Yang S., Verbauwhede I. Proc. IEEE ICASSP, 2005, pp. 609–612.
6. Cauchie S., Brouard T., Cardot H. On the Move to Meaningful Internet Systems 2006: OTM 2006 Workshops, Springer, 2006, pp. 489–498,
7. Ramírez-Ruiz J., Pfeiffer C., Nolazco-Flores J. Advances in Artificial Intelligence – IBERAMIA-SBIA, 2006, pp. 178–187.
8. Arakala A., Jeffers J., Horadam K. J. Advances in Biometrics (LNCS 4642), Springer, 2007, pp. 760–769.
9. Lee Y. J., Bae K., Lee S. J., Park K. R., Kim J. Proceedings of 2nd International Conference on Biometrics, Seoul, South Korea, August, 2007, pp. 800–808.
10. Nandakumar K., Jain A. K., Pankanti S. IEEE Transactions on Information Forensics and Security, 2007, vol. 2(4), pp. 744–757.
11. Balakirsky V. B., Ghazaryan A. R., A. J. Han Vinck Advances in Biometrics (LNCS 5558), Springer, 2009, pp. 889–898.
12. Kanade S., Petrovska-Delacretaz D., Dorizzi B. Proceedings of the 3rd IEEE interna- tional conference on Biometrics: Theory, applications and systems, 2009, pp. 333–339.
13. Ivanov A. I. Biometricheskaya identifikatsiya lichnosti po dinamike podsoznatel'nykh dvizheniy: monogr. [Biometric identification of personality by dynamics of subconscious movements: monograph]. Penza: Izd-vo PenzGU, 2000, 156 p.
14. Volchikhin V. I., Ivanov A. I., Funtikov V. A. Bystrye algoritmy obucheniya neyrosetevykh mekhanizmov biometriko-kriptograficheskoy zashchity informatsii: monogr. [Fast algorithms of learning of neural network mechanisms in biometric-cryptographic data security: monograph]. Penza: Izd-vo PenzGU, 2005, 273 p.
15. Yazov Yu. K., Volchikhin V. I., Ivanov A. I., Funtikov V. A., Nazarov I. G. Neyrosetevaya zashchita personal'nykh biometricheskikh dannykh [Neural network security of personal biometric data]. Moscow: Radiotekhnika, 2012, 157 p.
16. Akhmetov B. S., Volchikhin V. I., Ivanov A. I., Malygin A. Yu. Algoritmy testirovaniya biometriko-neyrosetevykh mekhanizmov zashchity informatsii: monogr. [Testing algorithms of biometric-neural-network mechanisms of information security: monograph]. Almaty, Respublika Kazakhstan: Izd-vo KazNTU im. Satpaeva, 2013, 152 p.
17. Akhmetov B., Doszhanova A., Ivanov A., Kartbaev T. and Malygin A. World Academy of Science, Engineering and Technology. Singapore. 2013, July, issue 79, pp. 129–138.