B Cell–Inspired Algorithms - Immunological Computation: Theory and Applications - page 120

Table 5.1 Clonal Selection Algorithms versus Evolutionary Algorithms

Clonal Selection
Algorithm Features Evolutionary Algorithm

Search space Set of chromosomes Set of antibodies

Candidate solution,
individuals

Chromosome Antibody

Individual representation Any (strings, real vectors,

etc.)

Any (strings, real vectors,
etc.)

Population size Fixed Fixed

Fitness function
(performance measure)

Fitness based on the
function

Affi nity

Operators Chromosome selection Clone selection

Mutation Hypermutation

Crossover

In Table 5.1, the generic clonal selection algorithm is compared with a basic
evolutionary algorithm. It is to be noted that diff erences primarily lie in the
terminology used. In clonal selection algorithms, the mechanism to select the fi ttest
antibodies is based on their a nities with the antigens. h erefore, traditional selection
mechanisms used in evolutionary computation, such as proportionate selection or
tournament selection are easily adapted to be used in clonal selection algorithms. In
evolutionary algorithms, typically, the probability that an individual be selected is
determined by its fi tness. h e case is the same for the antibody or antigen a nities.
Several versions of clonal selection algorithms (ClonAlg) are proposed by De
Castro and Von Zuben (2000) and De Castro (2003) and are used to perform
pattern recognition and multimodal function optimization tasks. h e algorithm
is depicted in Figure 5.3.
h ere are diff erent versions of clonal selection algorithm; the version of clonAlg
applied to pattern recognition problems assumes a set of patterns to be recognized
as input, whereas the version used for optimization assumes an objective function
to be optimized.

5.2 Immune Network Models

h e basic IN models (Anderson et al., 1973; Neuman, 1992; Vertosick and Kelly,
1989) tried to model network properties of immune cells in the absence of foreign anti-
gens. h ese immune networks are mostly considered as idiotypic networks (Burnet,
1957; De Boer and Hogeweg, 1989). Generally, an antibody could be represented as
a pair ( p, e ), where p is the antibody's collection of paratopes and e the set of epitopes .
Each antibody has two paratopes and two epitopes, which are the specialized parts of
the antibody that identify and are identifi ed by other molecules, respectively.