Volchikhin Vladimir Ivanovich, doctor of technical sciences, professor, president of Penza State University (440026, 40 Krasnaya street, Penza, Russia), email@example.com
Bezyaev Aleksandr Viktorovich, candidate of technical sciences, leading specialist of STC «Atlas» Penza branch (440026, 9 Sovetskaya street, Penza, Russia), Bezyaev_Alex@mail.ru
Ivanova Nadezhda Aleksandrovna, analyst, BioCrypt LLC (440031, 111 Okruzhnaya street, Penza, Russia), firstname.lastname@example.org
Serikova Julia Igorevna, master degree student, Penza State University (440026, 40 Krasnaya street, Penza, Russia), email@example.com
Background. The aim of the paper is to estimate the gain from using a quantum oracle when testing the quality of learning the network of artificial neurons. The urgency of the work is due to the need to test the neural network after each of its training or aftertraining.
Materials and methods. The algorithm for testing a neural network for a small sample according to GOST R 52633.3 is considered from the positions of quantum cybernetics and from the standpoint of classical statistics. In the Hamming distance space, the quantum superposition of the output states of a neural network is well described by the normal law of distribution of values.
Results. It is shown that a quantum oracle predicting the probability of occurrence of rare events, random guessing of «Alien» code «Svoy», gives the acceleration of testing in proportion to the reciprocal of the probability of errors of the second kind. The more reliable the means of biometricneuronet authentication, the greater the gain from using a quantum oracle.
Conclusions. The abandonment of classical testing, built on the expectation of rare events, allows reducing the size of the test database from 1,000,000 images of «Alien» to 32 images, which is equivalent to reducing the cost of collecting and preparing data by about five orders of magnitude.