Databases Reference
In-Depth Information
FIGURE 4.12: A 1-bit distortion occurring in the overall input pattern 'A'
stays encapsulated within the left composition. (With kind permission from
John Wiley & Sons, Inc.: Mobile Intelligence, “An Online Scheme for Threat
Detection Within Mobile Ad Hoc Networks,” pp. 380-411, 2010, Khan, A.
I. and Muhamad Amin, A. H. and Raja Mahmood, R. A., Figure 17.23,
http://dx.doi.org/10.1002/9780470579398.ch17.)
analyzed the distorted part of the pattern; the remaining subnets are not
affected by the distortion.
The HGN subnets are able to provide higher recall accuracy owing to this
encapsulation effect. The downside to this effect is that if the distortion occurs
in the larger subnet, the recall accuracy might be adversely affected. This
problem can be easily resolved by using compositions of similar size. The
uniform approach implements equal-sized compositions.
4.4.2 Uniform Distribution
The uniform model of the distributed HGN is introduced to delimit the
effects of distortion location experienced in the non-uniform model. In this
pattern recognition simulation, five HGN subnets for 7-bit subpatterns were
implemented to analyze 35-bit binary character patterns. Figure 4.13 shows
the structure of the compositions.
The uniform approach was developed to enable the distribution of the HGN
algorithm for networks comprising small devices and/or limited processing and
storage capabilities. With the relatively smaller sized subnets, each processing
node/neuron is able to store smaller pattern segments, and thus requires less
processing capability for the pattern recognition process. Having similar sized
compositions also removes the problem of a single composition affecting the
accuracy of the results.
It was reported in [64], that the uniform model's recall values are signif-
icantly higher than those of the HGN. The increase in the recall accuracy
is owing to the encapsulation effect. The distortions are generally compart-
mentalized in a specific composition(s), and thus do not affect the findings of
