Using software support for quantum superposition effects to speed up the solution of the targeted enumeration issue in biometric data when extracting knowledge from a neural network

Abstract

Background. The continuum/discrete processing of information by natural neurons and artificial neurons is considered. The purpose of the work is to demonstrate the possibilities of unlimitedly long-term support of the effects of quantum superposition by artificial neurons. Materials and methods. As an example, a network of 256 artificial binary neurons (perceptrons) is used. When simplifying the computational complexity of entropy estimates, the state standard GOST R 52633.3–2011 is used. When duplicating data, GOST R 52633.2–2010 is used. The inversion of matrices of neural network functionals with dimensions of 416×256 is performed iteratively according to the criterion of reducing the entropy of the output codes of the neural network. Results and conclusions. It is shown that the support of the effects of quantum superposition with a duration of about 20 minutes on a conventional computer makes it possible to solve the inverse problem of neural network biometrics. It is possible to extract knowledge from the neural network with a confidence probability of 0.97 about the cryptographic key of the user “Svoy” and about the parameters of the biometric image of the user “Svoy”.

Keywords

• natural neurons
• artificial neurons
• quantum superposition
• fast entropy estimation algorithms
• directed enumeration acceleration