The problem to be solved

• It is straightforward to adjust the weights to the output layer, using the Perceptron rule

• But how can we adjust the weights to the hidden layer?