Computational Intelligence Techniques - Computational Intelligence in Biomedical Engineering - page 76

been used, for example, least squares regression and support vector
regression (SVR). Regression is particularly useful for problems in
statistics and engineering. Classification and regression is also known
as supervised learning because a supervisor has already provided the
data with the necessary labels or values.

c. Time series prediction/forecasting . In this problem, one is presented

with a series of data generated by some unknown

distribution or function. The challenge is to predict the value of the
next sample y t +1 . This problem often occurs in weather prediction and
stock market or financial predictions.

{ y 1 ,y 2 , ..., y t }

d. Clustering . This is a slight variation of the classification problem and

more dicult to solve. One is provided with unlabelled data and the
task is to find similarities between the individual data and group data.
This type of learning is known as unsupervised learning or clustering
and covers methods such as independent component analysis (ICA),
blind source separation (BSS) (Girolami, 1999), and principal compo-
nent analysis (PCA).

e. Optimization . In optimization, the aim is to solve a minimization or

maximization problem subject to a set of constraints. The problem
is usually modeled mathematically as an objective function and the
constraint set can be composed of equalities or inequalities, which must
be satisfied by the variables. The optimal solution is the set of variables
that solve the problem subject to the constraints. Many CI techniques
are formulated as optimization problems, which can be solved using
techniques such as gradient descent, Newton's method, and conjugate
gradients (Chong and Stanislaw, 2001; Fletcher, 1981; Gill et al., 1981).

f. Content-addressable memory . In the standard Von Neumann architec-

ture, the memory address is calculated and then accessed. Addresses
in memory have no physical meaning and if the incorrect address is
calculated, the correct information will not be retrieved. Associative
memory is memory, which can be addressed by the content and is
more robust to errors or partial addresses than the memory in the Von
Neumann architecture. Associative memory can be mimicked by, for
example, ANN and is important for applications such as multimedia
databases (Jain et al., 1996).

g. Control . Control is essential in the design of engineering systems. One

has an input-output relationship

where u ( t ) is the input
signal, which must be constrained to follow (controlled) the desired
output signal y ( t ) for some time t . Examples of control problems are
motor idle speed and water-level control in dams.

{ u ( t ) ,y ( t ) }

In the following sections, we describe the main CI techniques, which are
currently more applicable to problems in biomedical engineering.