Improvement of learning efficiency of the neural networks, intended for recognition of graphic images in systems of biometric authentication
L Tereykovskaya1, I Tereykovskiy2, E Aytkhozhaeva3, S Tynymbayev3, A Imanbayev3
COMPUTER MODELLING & NEW TECHNOLOGIES 2017 21(2) 54-57
1Institute of Solid State Physics, National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”, 14-a Polytechnichna str., 03056, Kyiv, Ukraine
2Kyiv National University of Construction and Architecture, 31,Povitroflotsky Avenue, Kyiv-037, 03680 Ukraine
3Kazakh National Research Technical University, Satpayev Str. 22, Almaty, Kazakhstan
Article is devoted to a problem of use of neural network technologies in the field of biometric authentication of users. It is shown that one of important the shortcomings of application of neural networks technology on the basis of a multi-layer perceptron for recognition graphic images in systems of biometric authentication of users is insufficient quality of processing of statistical data which are used when forming parameters of educational examples. It is offered to increase quality of educational examples due to use of the procedure of neural network coding of value of the expected output signal of educational examples which allows consider closeness of standards of the recognized classes in this signal. The coding procedure of the expected output signal providing use of a probable neural network is developed. The appropriate mathematical devices are created. As a result of numerical experiments it is shown that application of the developed procedure allows reduce the number of the computing iterations necessary for achievement of the given error of training by 30-50%. It specifies prospects of use of the proposed solutions for improvement of learning efficiency of the neural networks, intended for recognition of graphic images in systems of biometric authentication.