Latest Immune Models and Hybrid Approaches - Immunological Computation: Theory and Applications - page 166

Self-tolerance Primary response Secondary response

Normal samples New samples

New samples

Negative
selection

Afinity

Maturation

Artificial
anomalies

SOM
training algorithm
labeling

Normal Known

anomaly

Unknown
anomaly

Normal

Unknown
anomaly

Normal
Unknown anomaly

Known anomaly

Figure 6.7 NS-SOM model structure. The model consists of three phases: self-
tolerization, primary response (affi nity maturation), and secondary response.
The squared arrangement of nodes corresponds to an SOM, where black, gray,
and white labels represent normal, unknown anomaly, and known anomaly
respectively. (From González, F. J., Galeano, A. Veloza and A. Rojas. Pro-
ceedings of the 2005 Conference on Genetic and Evolutionary Computation
(GECCO'05), ACM Press, New York, 2005.)

Particularly, this approach uses the positive (normal) samples to generate nega-
tive samples. h en samples from both the classes are used (as training data) for a
neural network, particularly, for a self-organizing map (SOM). Figure 6.7 shows the
algorithmic steps of this hybrid NS-SOM approach. An SOM, composed of nodes
or neurons (that are able to recognize diff erent types of inputs), is a type of artifi cial
immune network that is trained to produce a low-dimensional representation of the
input space or self-/nonself feature space of the training samples called “map.”
González et al. (2005) extended their earlier works of combining NS algo-
rithm and the SOM for producing the visual representation of the self-/nonself
feature space. h is representation provides the understanding of the structure of
self-/nonself space by producing a visual discrimination of the normal, known
abnormal, and unknown abnormal regions. h is model produces a network that
can discriminate normal samples from abnormal samples and can learn from
the encounters with antigens to improve specifi city of response. h is NS-SOM
anomaly model has three phases: self-tolerization, primary response, and second-
ary response. h ese three phases are illustrated in Figure 6.7.
h e fi rst phase, self-tolerization, uses NS algorithm to produce the artifi cial
anomalies and then, these anomalies are used to produce an anomaly classifi er by
using an SOM training algorithm instead of using classifi er training algorithm.