Synthesis of recognition systems with variable structure by three layer neural networks

The authors consider the problem of the synthesis of recognition systems (RSs) with variable structure into two classes of patterns chosen during adjustment in accordance with a given probability of correct recognition. The RS structure is chosen from the class of three-layer networks, in particular from the linear threshold elements (LTEs). Synthesis of the RS is based on sequential teaching of RS layers. The teaching algorithms for different layers of the RS are different. It is proved that a three-layer network with only LTEs in the third layer is sufficient for this task solving