Reducing the Size of a Sample Sufficient for Learning Due to the Symmetrization of Correlation Relationships Between Biometric Data

This paper shows that correlation coefficients obtained from small test samples for biometric data involve considerable uncertainty. This interferes with using them for machine training (setting) of classical quadratic forms and Bayesian networks. A method for symmetrizing correlation relationships is proposed. The requirement on the volume of biometric data is proved to be reduced considerably in this case. As a consequence, the setting (teaching) of quadratic forms and setting of maximum likelihood Bayesian networks become much more stable problems. This enables many-fold reduction in the requirement on the size of the training sample for an “own” biometric image.