Computational Intelligence

184 Chapter Five—Neural Network Concepts and Paradigms

to extend the two-layer er ror correction adaptation algorithm to multiple layers, but

the same problem kept arising: For how much of an output-layer PE error is each

hidden-layer PE responsible? Using the three-layer neural network in Figure 5.10 to

illustrate, the problem of multilayer adaptation (in this case, three-layer adaptation)

is calculating the amount of error each hidden-layer PE, y

j

, should be assigned for

an output-layer PE’s error. Note that the output layer of PEs has activation values

z

1

, z

2

,...,z

q

, and that the weight matrix from the input layer to the hidden layer is

denoted