B.V. Khakimov, I.M. Mikheev
A new neuron’s model is described, which “synapses“ are spline-transformators of entrance signals, “soma” calculates a weighed summary of transformated signals and “axon” executes a multipurpose spline-transformation. The main points’ abscissas of splines are set in equiprobable intervals and ordinates are calculated with an iterational algorithm of the steepest descent adopted to splines.
The analogue of the neuron’s spline-model is a multidimensional orthogonal spline, which forming and directing surfaces are one-dimensional splines providing for clear interpretation of modeled causal relationships.
The neuron’s spline-model works with incomplete input data, provides for high exactitude of the model operation, its multiresolution is equivalent to three-layer network of known neurons’ models.