Real-World Applications - Immunological Computation: Theory and Applications - page 197

Noisy data . Is there a need of mechanisms to deal with noise and outliners, or
any other “bad” self-samples? In real applications, it seems necessary in most
scenarios.
Recognizing each abnormal data or reacting only when the same anomaly
happens multiple times. h e latter is more reasonable in applications such as
network intrusion detection or fault detection.
Completeness of the self-samples . If all self-patterns are represented in the train-
ing data, we call it “complete.” It is usually not the case for real-valued repre-
sentation, but it is true for many applications using string representation.
Dynamic data . h is refers to the training data that are not stationary (self-set
changes over time).
Distributed data . In this case, the self-set is distributed or too large to observe
completely.
Forrest et al. (1994) mentioned that the method relies on the fact that the data is
not corrupted when the detectors are generated. h is refl ects the idea that the self-
samples are at least considered correct regardless of whether they are complete or
not. Even if the self-samples are complete as well as correct, NS algorithms are still
probabilistic in most methods, implying that they may not achieve perfect cover-
age. h e goal is to have a small number of detectors that are capable of detecting a
relatively large portion of nonself space. In brief, the algorithms should depend on
the data properties as little as possible, but the common assumptions are important
for a plausible comparison.
7.12 R e v i e w Q u e s t i o n s
1. Order and explain the steps necessary to apply an immunological model to
solve a particular problem.
a. Identify the elements involved in the problem
b. Identify how elements can be modeled as entities
c. Choose a representation for each one of the elements involved in the
problem
d. Defi ne appropriate a nity measures
e. Decide what immune algorithms will be used
2. Develop an immunit y-based multiagent system to play robot soccer. Describe
the original solution based on the steps shown in Figure 7.1.
3. Explain the advantages of using AINEs over other techniques in each of the
following applications:
a. Machine learning
b. GAs
c. Robot behavior and control
d. Fault diagnosis