Digital Signal Processing Reference
In-Depth Information
and The weights are effectively the coordinates (
i
,
j)
of the
location of the neuron in the SOFM. The output of the winning neuron is
obtained from the weight
of the output neuron.
Figure 8-2.
CMAC Memory Architecture.
As in the Kohonen's neural networks framework, the practice in CMAC
learning is a competitive learning process and follows the well-known
SOFM learning rules. With a simple caveat, since the weights of the cluster
of neurons that represent indices to the CMAC memory are fixed, learning
occurs only at the output neuron. To implement this, the learning rule for
CMAC has been based on the well-known Grossberg competitive learning
rule and it is applied only at the output layer. Furthermore, no competitive
Kohonen learning rule is applied to the input layer. Therefore, the CMAC
learning rule can be represented by [2]:
where
= Learning parameter,
= Plant input at step
k,
x(kT)
