B.V. Khakimov

The author describes the algorithm of identification of the neuron-s spline model as a multidimensional orthogonal standardized spline. This spline is intended for modeling of non-linear correlative dependences, patterns recognition and other purporses. Easy and velocity calculation is assured by the correlative splines - identification with coordinates of branch points and with general rule of their join. The branch points - abscissas are determined with the equiprobable intervals of analytic points, ordinates are calculated with the iterational gradient algorithm of quickest descent by analytic points - set. General error of position of the model is distributed among spline-components in direct proportion to the value of arguments. There are two branch points of the interval to be set in each spline for fixing of the value of corresponding argument during function calculation. The algorithm tunes the spline-model by incomplete basic data, when parameters can be precised by means of complementary branch points - application.