The paper considers the problem of regularization of automatic stable training of networks of artificial Mahalanobis neurons that simultaneously perform recognition of the “Own” image.