Real-World Applications - Immunological Computation: Theory and Applications - page 196

network theory to create and sustain a diverse set of possible solutions in the search
space and combine it with traditional GAs.
Endoh et al. (1998) used immune system ideas to handle constraints in GAs.
Traditionally, the way to implement the constraints true to real life has been the
use of penalty functions to the fi tness evaluation process so that the solutions are
guided away from infeasible regions of the solution landscape. However, good pen-
alty factors are di cult to defi ne. h is technique is based on the NS algorithm. It
generates a random population and denotes the feasible individuals as the antigens
and the infeasible individuals as the antibodies. h e idea is that antibodies learn
from antigens to be closer to them or the infeasible individuals are motivated to
become feasible by their exposure to the feasible part of the population. h e evolu-
tion is carried out using standard GA operators.
Mori et al. (1998) looks at applying immune system metaphors to extended
GAs for search optimization problems. Here, the authors propose an extension to
a standard GA by the inclusion of immune system metaphors of B cells by using
a combination of memory and suppression cells. h is variation of the algorithm
creates a memory of the best cases for searching, allowing the reinforcement of
good solutions within the search space, and to use those good solutions for further
exploration. h is work was then extended by Coello and Cortes (2002).
7.11 S u m m a r y
IC emerged in the 1990s as a new paradigm in artifi cial intelligence (AI), and has
earned its position on the map of soft computing. h is chapter summarizes the appli-
cations of artifi cial immune systems in various science and engineering domains.
A survey of some of the applications in this emerging fi eld of artifi cial immune sys-
tems has been reported, which include computer security, anomaly detection, pattern
recognition, data mining, adaptive control, fault detection, and many others.
To apply an immune model to solve a particular problem from a specifi c domain,
one should select the immune algorithm according to the type of problem that is
being solved. h en, identify the elements involved in the problem and how they can
be modeled as entities in the particular immune model. To model such entities, a
representation for each of these elements should be chosen, specifi cally, a string rep-
resentation: integer, real-valued vector representation, or a hybrid representation.
Subsequently, appropriate a nity (distance) measure to determine corresponding
matching rules should be defi ned. h en the immune algorithm that will be used to
generate a set of suitable entities providing a good solution to the problem at hand
should be selected.
h e following issues concerning the type of the problems and the property of
the training data are important in any analysis of NS algorithms:
Frequency-refl ecting data . h e distribution probability of data is crucial to
evaluate the success of the learning algorithm.