Backpropagation Learning Routine:

Conversely as with perceptrons there the information in the network is stored in the weights than the learning problem comes down to the question as: how do we train the weights to best categorise the training examples. So whenever we hope that this representation provides a good way to categorise unseen examples. 

Hence in outline the backpropagation method is the as same for perceptrons as: 

1.  Whether we choose and fix our architecture for the network, which will contain input,  hiddedn and output units and all of that will contain sigmoid functions. 

2.  Hence we randomly assign the weights between all the nodes. So the assignments should be to small numbers, usually between -0.5 and 0.5. 

3.  Its means that each training example is used and one after another just to re-train the weights in the network. Hence in this way is done is given in detail below. 

4.  So now after each epoch as run through all the training examples and a termination condition is checked like also detailed below. So Notice that for this method whether we are not guaranteed to find weights that give  the network the global minimum error that is perfectly correct categorisation of the training examples. Thus the termination condition may have to be in terms of a as possibly small number of mis-categorisations. So we see later that this might not be such a good idea that is though.