B Cell–Inspired Algorithms - Immunological Computation: Theory and Applications - page 147

where f a nity is the measure of the similarity and complementarity between
elements in the shape-space and is given as f a nity : B

∪ × ∪

→ R .
Now, the amount of stimulation produced by an antigen with a given a nity
with the B cell is g : R → R.
h en, the 2.2 step calculates the B cell interaction by calculating the stimula-
tion and suppression eff ects between them. h ese functions are given as follows:

A B A

f stimulation

B

:B B →ℜ

and

B

→ℜ

f suppression :B B

h en, the 2.2.1 step calculates the stimulation and suppression of B cell/B cell
in the similar manner of antigen/B cell stimulation. h en, the total stimulation
F: B → R is calculated by adding all the eff ects caused by the antigen and network

connection, which is F( b ) (given in step 2.3.1).
Now, some of the B cells are selected and f cloning ( b ) copies of each selected B cell,
b are created. Now, mutation is done on these selected B cells. h is mutation is var-
ied from AIN model to model. After this, some of the B cells are deleted and some
are again created randomly in the network and the connection or links between
them are again created. Finally, this algorithm stops when the stopping criterion is
met and returns the current network.

5.4 Summary
h is chapter describes immune algorithms primarily based on clonal selection
principle and idiotypic INs, particularly, the ClonAlg algorithm and its varia-
tions, which are based on the clonal selection and a nity maturation principles.
h e ClonAlg is similar to mutation-based evolutionary algorithms and has several
interesting features: (i) dynamically adjustable population size, (ii) exploitation and
exploration of the search space, (iii) location of multiple optima, (iv) capability of
maintaining local optima solutions, and (v) defi ned stopping criterion.
h e IN theory and continuous and discrete models are described. h e derived
computational algorithms are discussed in detail. Particularly, Hunt and Cooke
(1996) proposed a supervised machine-learning algorithm based on IN model to
classify DNA sequences as either promoter or nonpromoter classes. Timmis and
Neal (2001) introduced another algorithm similar to it, but domain-independent,
called AINE. h is network constitutes a reduced version of the original data that
can be used for data clustering or compression.
A major drawback of AINE is the explosion in B cell population. h us, an
enhanced algorithm called RAIN was developed (Timmis et al., 2000). h e
main diff erence between AINE and RAIN is that the basic element of the RAIN