It is noted, once again, that there is no need to solve for the
value of Wt at every iteration t. Both [Wn]t and [Wd]t can be
used to accumulate the intermediate results as long as there
is no sign of overflow either in the numerator or in the
denominator term.
At this point, we have already obtained the weight
update equation that supports a rational number representation.
In addition, the equation is also capable of
processing the rational form of the learning kernel. In the
following section, the algorithmic details of our proposed
rational-based weight update equation are given.