The actual target classification of each training case is stored with each hidden node; the weighted value coming out of a hidden node is fed only to the Summation layer that corresponds to the hidden nodes’s classification. The Summation layer add the values for the classification they represent.