Digital Signal Processing Reference
In-Depth Information
Bio-Inspire
d Optimization Methods
The ensemble of methods and techniques discussed all through this topic are,
after all, closely related to solving an optimization problem. The proposed
solutions have been currently obtained by the minimization or maximization
of a given criterion or cost function. In this chapter, we discuss a different
paradigm, the foundations of which come from the study of auto-regulation
processes observed in nature. The so-called bio-inspired optimization meth-
ods are generally characterized by their global search potential, and do
not require significant a priori information about the problem to be solved.
Such characteristics encourage their application in the nonlinear and/or
unsupervised problems we are dealing with.
The field of bio-inspired techniques for optimization is certainly a broad
one. In this chapter, and taking into account the applications in mind, we
have decided to concentrate our attention on three classes of tools:
genetic
algorithms
(GAs),
artificial immune systems
(AISs), and
particle swarm
meth-
ods. The first class is certainly a most emblematic bio-inspired optimization
approach. On the other hand, the two are currently the object of attention
of many researchers due to some desirable features that, in our opinion, are
valid to expose.
In order to better expose our purposes,
Section 8.1
provides a general dis-
cussion about some points that could motivate the use of bio-inspired tools
in some signal processing problems. Then the rest of the chapter is organized
as follows:
•
based on elements of the modern synthesis of evolution theory. They
are also an interesting starting point for our discussion, given their
historical importance and their widespread use in many practical
domains.
•
In
Section 8.3
,
we analyze another class of bio-inspired methods that
can also be considered as an
evolutionary technique
: that of AISs. The
analysis is focused in the so-called opt-aiNet, an interesting opti-
mization tool that presents some points of contact with the GAs [86].
•
In
Section 8.4
,
we present another branch of techniques, the inspira-
tion of which comes from the collective behavior observed in the
nature. To illustrate this branch, we choose the approach known
as
particle swarm optimization
(PSO), which is particularly suited
to the continuous-valued problems that characterize most filtering
applications.
253
