
26 High Performance Programming for Soft Computing
1.6.4 The Back-propagation Algorithm
The back-propagation algorithm considers two cases:
Case 1: The artifi cial neuron j is an output node. Here the error is calculated
using Eq. (1.6.3).
Case 2: The artifi cial neuron j is a hidden node. At this node we cannot
know the desired response, so the back-propagation algorithm ,
using error signals, propagate the error through the nodes making
possible to estimate the error named
F
, at each hidden node; in short,
the formula deduction of the error at the hidden nodes is given by
(1.6.7), where the index k identifi es neuron k, which is connected
to the ...