In our neural network, the first neuron in the output layer is trained to decide whether a given symbol is the uppercase English letter A. Assuming that the first three neurons in the hidden layer fire, we would expect the first neuron in the output layer to fire and the remaining 25 neurons not to fire. Our MLP would then classify this symbol as the letter A!
Note that our training architecture employed only a single hidden layer, which would only be able to learn very simple patterns. By adding more hidden layers, an ANN can learn more complicated patterns at the cost of computational complexity, resources, and training runtime. However, with the advent of distributed storage and processing technologies, ...
