Backpropagation in equations I

• If j is a node in an output layer, the error ?j is:

  ?j = (tj - aj) aj(aj -1)
  

• where aj is the activation of node j

• tj is its target activation value, and

• ?j its error value