An Immunological Filter for Spam - Artificial Immune Systems - page 450

Fig. 2. The use of a convergence window to evaluate the stopping criterion
Once defined the size ( s ) of the convergence window, the approach proposed
here evaluates the performance of networks with increasing numbers of antibod-
ies but restricted to the convergence window. If any of the topologies within this
window presents better performance than a given reference network , the refer-
ence network will then be replaced by the better performance network, as can be
seen in Fig. 3. If none of the topologies within the convergence window presents
better performance than the reference network, the convergence criterion is sat-
isfied and the learning procedure halts, finishing the topology adaptation. In this
case the resultant topology will be the reference network.
Fig. 3. Dynamics of the convergence window and reference network updating