Databases Reference
In-Depth Information
FIGURE 3.6: Maximum bias array size analysis for a GN implementation as
a function of pattern size. The results for several numbers of different pattern
elements are shown.
1. If the neuron is at the edge, σ
max,e
= n
row
.
2. For a non-edge neuron, σ
max,ne
= n
row
.
For a pattern of size, S = a, the total maximum bias array capacity for all
neurons in the network, max σ, can be estimated using the following equation:
max σ = n
row
×(σ
max,ne
×(a−2) + 2σ
max,e
)
= n
row
×
n
row
×(a−2) + 2n
row
(3.3)
= n
row
×(n
row
×(a−2) + 2)
The total maximum bias array capacity of a one-dimensional GN network is
significantly affected by the number of different elements in the input patterns.
However, the size of the pattern only moderately influences the maximum
capacity. In this context, large-scale patterns with minimum variation between
elements will have a lower impact on the bias array capacity than large-scale
patterns with high variation between elements. Figure 3.6 shows the growth
of the total bias array size for a GN network as a function of the number of
different pattern elements and pattern size.
The total maximum bias array size grows linearly with pattern size. In this
regard, the GN network has proven to offer scalability for large-scale patterns.
An increase in the dimension of the patterns also affects the total size of the
bias array. This is due to an increase in the number of possible combinations
