Summary: | Background. Obtaining numerical estimates of the corrective ability of classical
codes with high redundancy and neural network corrective structures by the example of
controlling 416 biometric parameters of the handwritten password word “Penza”. Materials
and methods. It is proposed to use the error corrector setting for a single code state consisting only of “0” states. Automatic training of the neural network corrector is carried out using
the standard algorithm State Standart R 52633.5-2011. Results. By the example of real
data, it is shown that the corrective ability of neural network structures made it possible to
reduce the flow of errors by half when using a data-enriching network of 416 neurons with
four inputs. When using neurons with 8 inputs, it is possible to additionally reduce the
number of errors by a further two times. Conclusions. Preliminary neural network enrichment
of data before their folding with redundant self-correcting code greatly reduces the
requirements for the corrective ability of the code.
|