Databases Reference
In-Depth Information
FIGURE 3.8: Crosstalk phenomenon in GN pattern recognition.
The example is extended by analyzing the bias array of the GN network
for the example given previously. Figure 3.8 shows an illustration of the bias
array analysis of the GN network for the crosstalk example. Note that the
recall made for pattern “uvwxy” is perceived to be true by all of the activated
neurons because all the subpatterns are found. However, the actual recall is
inaccurate because the test pattern as a whole does not match the stored
patterns. This phenomenon is known as intersection or crosstalk problem.
In Figure 3.8, the bias arrays for patterns “uvwxz” and “zvwxy” are stored.
When the “uvwxy” pattern is introduced, all of the bias entries of the two
original patterns are recalled, and thus a false recall is created.
The inability of the GN algorithm to obtain an overview of the entire pattern
leads to false recalls. A mechanism to eliminate this problem needs to be
devised. Nasution and Khan [3] suggested a hierarchical GN implementation.
In the next chapter, we will discuss the algorithmic design and implementation
of the hierarchical GN model. We will also analyze the complexity of the model
and the recognition accuracy of the pattern classification.
