Databases Reference
In-Depth Information
FIGURE 3.2: GN network activation from input pattern “ABBAB.”
3.1.1 Pattern Representation
The GN pattern representation follows the representation of patterns in
other graph-matching based algorithms described in the previous subsection.
Each neuron in the network holds an information pair, (value, position), which
contains information about the elements that constitute the pattern. In corre-
spondence with the graph-based structure, each neuron acts as a vertex that
holds pattern element information in the form of a value or identification (ID).
The adjacency communication between two or more neurons is represented by
the edge of a graph. Message communications in a GN network are restricted
to adjacent neurons (of the array). As described in Khan et al. [31], if the num-
ber of neurons in the network increases, there is not a corresponding increase
in the communication. Figure 3.2 shows a two-dimensional GN graph-based
structure for a given input pattern. Note that only GNs that have a matched
pattern element and position will be activated and perform communication
with its adjacent neurons. This self-organization creates links between neurons
and builds up pattern information in the network.
Each neuron in the same row holds a similar ID or value, but their po-
sitions (column position) differ within the network, as shown in Figure 3.2.
These value assignments uniquely mark the position of each neuron with its
column number in the network. This type of arrangement is just one of the
possible structural neuron arrangements of the network. Neuron arrangement
will be discussed further in later chapters. To determine the positions of other
neurons, each neuron within the network must be able to obtain information
pertaining to the size of the network. For recognition processes, this informa-
tion is important in identifying adjacent neurons.
An input pattern for a GN network might be a signal spike, or stimulus
resulting from user activation or information derived from an executable pro-
gram or sensory device. In addition, it might represent bit elements of an
image [61] or a stimulus/signal spike produced in a network intrusion detec-
tion application [34]. Each neuron is able to identify its ID from the pattern
that has been introduced. For instance, a GN that holds the value “B” will
