Perceptrons:

However the weights in any ANN are usually just real numbers and the learning problem boils down to choosing the best value for each weight in the network. Because there are two important decisions to make rather than we train a artificial neural network as: (i)overall architecture of the system in which how input nodes represent given examples and how many hidden units or hidden layers to have and how the output information will give us an answer and other (ii) that how the units calculate their real value output from the weighted sum of real valued inputs. 

Here answer to (i) is always found  through experimentation by respect to the learning problem at hand as: different architectures are tried  but they evaluated on the learning problem until the best one emerges. In generally perceptrons in which we have no hidden layer or the architecture problem boils down to just specifying how the input units represent the examples given to the network.