Structure optimization of multilayer neural networks with cross connections

Problems of choosing the structure (the number of layers and the number of neurons in a layer) for multilayer neural networks with cross connections and consisting of neurons with two lattices are considered for the solution of pattern recognition problems. Consideration is given to multilayer neural networks with complete cross connections where the attribute set of each layer consists of initial space attributes and output signals of the first, second, and (j-1)th layers. An attempt is made to formalize the structural determination of these networks and their structural optimization according to various criteria.<
>