Artificial molecules assembled from artificial neurons that reproduce the work of classical statistical criteria

The purpose of the work is to show the possibility of a neural network generalization of many classical statistical criteria for decision making on small samples of real data. It is shown that to solve the problem it is necessary to use its preliminary symmetrization, which allows you to remove the problem of modeling long random codes with dependent (linked) bits. The simplicity of simulated symmetrized data makes it possible to take into account the correlation between bits of random codes and observe the restrictions imposed by codes with the detection and correction of errors.