Effect of influence of quantity of network's neurons on parameters of its educability

Dependence of education errors, forecast errors and rates of education of neural network on the number of neurons is analyzed. The effect of sharp deterioration of parameters of education and forecast at some values of the number of neurons is found out and can be connected with the influence of initial conditions bringing a casual component to start configuration of neural network's weight factors' values. Observed unmonotonicity of the parameters of educability has no common tendency.