T Cell–Inspired Algorithms - Immunological Computation: Theory and Applications - page 80

(Ji and Dasgupta, 2004a and 2004b); or (4) optimization with aftermath adjust-
ment (Dasgupta et al., 2004). h ese algorithms are described in the following text.
In RNS, a detector is defi ned by an n -dimensional vector that corresponds to
the center and by a real value that represents its radius; therefore, a detector can be
seen as a hypersphere in R n . h e detector-antigen matching rule is expressed by the
“membership function” of the detector, which is a function of the detector-antigen
Euclidean distance and the radius of the detector. h is approach is similar to the NS
greedy algorithm (D'haeseleer, 1995b), but in a real-valued representation space.
h e input to the algorithm is a set of self-samples represented by n -dimen-
sional points (vectors). h e algorithm tries to evolve another set of points (called
detectors) that cover the nonself space. h is is accomplished by an iterative process
that updates the position of the detector driven by the following two goals:

Move the detector away from self-points
Keep the detectors separated to maximize the covering of nonself space

h e logical steps of the algorithm are shown in Figure 4.6, which are described in
the pseudocode (NS Algorithm 4).

For each detector 'd

Yes

No

Does 'd match
any self point?

Yes

No

Move 'd away
from other
detectors

'd.age > 't ?

'd.age ++
Move 'd away
from self

'd.age = 0

Discard 'd

Figure 4.6 Shows an iteration of the RNS algorithm. This approach is similar
to greedy algorithm but uses real-valued space. (From D'haeseleer P. An
immunological approach to change detection: Theoretical results. Proceedings
of the 9th IEEE Computer Security Foundations Workshop , pp. 18-26,
Los Alamitos, CA, June 1996; D'haeseleer, P. Further Effi cient Algorithms for
Generating Antibody Strings , Technical Report CS95-3, The University of
New Mexico, Albuquerque, NM, 1995a.)