[Cover] [Contents] [Index] |
Page 124
Figure 3.11 The weight adjustment rule for counter-propagation. Only weights connecting to the winning neurone (shaded) are adjusted.
maximum weighted sum Sv. The weights connecting the input layer neurones to the winning neurone v are then updated in terms of following rule:
(3.21) |
where α, 0 α≤1, is the learning rate, and denotes the weight state at time n. The other weight sets, wz, for , retain their original values. The weight set wv is then again subjected to the following normalisation process to maintain its normalisation state:
(3.22) |
For the weight set , connecting the winning neurone v in the hidden layer to the output layer, hv is updated in terms of:
(3.23) |
where β is another learning constant, oj is the desired output for output neurone j, and aj is the network output derived from:
(3.24) |
[Cover] [Contents] [Index] |