Information Technology Reference
In-Depth Information
a genealogical tree describing dependency relationship between the models. Our
approach would be the closest to the
Hunt & Cooke
branch which started from
the model proposed in 1996 [5].
In [3] a general-purpose framework of AIS is proposed where the three key
layers are distinguished: immune algorithms, anity measures and representa-
tion. The following section describes the first and the last layer of the framework.
The proposed idiotypic network model is presented, i.e. the main loop of the pro-
cess and the rules of anity and enmity which control the concentration of the
molecules are discussed. The formula of evaluation of new concentration levels
of the molecules is given. In Section 3 a description of the second layer of the
framework including anity measures can be found. A set of measures is pre-
sented as well as a novel transformation operator for binary strings. Section 4
includes results of the first group of experiments where the model was tuned
and average life span of antibodies was observed. Section 5 presents the last
group of experiments where a set of five antigens was cyclically presented to the
system. The paper is concluded with a summary of the current work and plans
for further research.
2
The Idiotypic Network Model
The model represents a network which consists of a set of antibodies and rules
of relationship between them. In our specification of the model different types of
antibodies are represented as objects in a binary shape space. Each of them is
equipped with a 32-bit paratope and a 32-bit epitope, and with two numerical
attributes: a concentration and a lifetime. It is assumed that each antibody have
just one binding site therefore every object represents all the antibodies with the
same patterns of the paratope and the epitope. The quantity of a set represented
by an object is defined by the concentration attribute. It should not be allowed to
exist two or more objects with the same paratope and epitope in the population
of objects at the same time
1
. The remaining two components of the object,
the two numerical attributes do not participate in binding rules. The second
attribute, the object's lifetime allows us to observe robustness of each of the sets
of antibodies. When a new object is added to the population its lifetime is set to
zero and then it is increased at the end of every iteration of the process as far as
the object exists in the population. The concentration of a newly created object
is set to an initial value which was equal to 1 in the experiments presented below.
During the lifetime of the system the objects stimulate each other to increase or
decrease the concentration of the antibodies which they represent.
The proposed specification allows also to introduce antigens into the system.
Different types of antigens are represented by objects equipped with epitope
and two numerical attributes, i.e. the concentration and the lifetime. The ob-
jects representing antigens interact with the objects representing antibodies in
1
It is hardly likely that such a situation will take place because the total number of
possible patterns is 2
64
. Therefore our software application did not have any special
procedures for elimination of redundant objects.
