Fig. 3

Example of determination of the possible new neural network (NN) architectures that can grow from a given NN (a) that emerged in the course of the repeated structuring and learning procedure (RS&LP). The new architecture will consist of the currently existing NN plus one additional neuron. The first attempt to create a new architecture consists of adding the extra neuron to the first hidden layer, this architecture is possible (b). The second attempt consists of adding an extra neuron to the second hidden layer, this architecture is not permitted because it would give the second hidden layer more neurons than the first hidden layer (c). The third attempt consists of adding the extra neuron to the third hidden layer, this architecture is possible (d). The fourth attempt consists of creating a new hidden layer with the extra neuron, this architecture is not permitted because the number of layers is limited to three (e). Hence, out of four attempts, two are successful (b, d) and will be evaluated in the next learning step